blob: cc93b23ca3520c9ad7bf2919e77f61cf400c2e4c [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Chris Mason8f18cf12008-04-25 16:53:30 -040019#include <linux/kernel.h>
Chris Mason065631f2008-02-20 12:07:25 -050020#include <linux/bio.h>
Chris Mason39279cc2007-06-12 06:35:45 -040021#include <linux/buffer_head.h>
Sage Weilf2eb0a22008-05-02 14:43:14 -040022#include <linux/file.h>
Chris Mason39279cc2007-06-12 06:35:45 -040023#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
Chris Mason39279cc2007-06-12 06:35:45 -040029#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
Chris Mason9ebefb182007-06-15 13:50:00 -040035#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050036#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040037#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040038#include <linux/falloc.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090039#include <linux/slab.h>
David Sterba7a36dde2011-05-06 15:33:15 +020040#include <linux/ratelimit.h>
Josef Bacik22c44fe2011-11-30 10:45:38 -050041#include <linux/mount.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050042#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040043#include "ctree.h"
44#include "disk-io.h"
45#include "transaction.h"
46#include "btrfs_inode.h"
47#include "ioctl.h"
48#include "print-tree.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040049#include "ordered-data.h"
Christoph Hellwig95819c02008-08-28 06:21:17 -040050#include "xattr.h"
Chris Masone02119d2008-09-05 16:13:11 -040051#include "tree-log.h"
Jan Schmidt4a54c8c2011-07-22 15:41:52 +020052#include "volumes.h"
Chris Masonc8b97812008-10-29 14:49:59 -040053#include "compression.h"
Chris Masonb4ce94d2009-02-04 09:25:08 -050054#include "locking.h"
Josef Bacikdc89e982011-01-28 17:05:48 -050055#include "free-space-cache.h"
Li Zefan581bb052011-04-20 10:06:11 +080056#include "inode-map.h"
Chris Mason39279cc2007-06-12 06:35:45 -040057
58struct btrfs_iget_args {
59 u64 ino;
60 struct btrfs_root *root;
61};
62
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070063static const struct inode_operations btrfs_dir_inode_operations;
64static const struct inode_operations btrfs_symlink_inode_operations;
65static const struct inode_operations btrfs_dir_ro_inode_operations;
66static const struct inode_operations btrfs_special_inode_operations;
67static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070068static const struct address_space_operations btrfs_aops;
69static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070070static const struct file_operations btrfs_dir_file_operations;
Chris Masond1310b22008-01-24 16:13:08 -050071static struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040072
73static struct kmem_cache *btrfs_inode_cachep;
Miao Xie8ccf6f192012-10-25 09:28:04 +000074static struct kmem_cache *btrfs_delalloc_work_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040075struct kmem_cache *btrfs_trans_handle_cachep;
76struct kmem_cache *btrfs_transaction_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040077struct kmem_cache *btrfs_path_cachep;
Josef Bacikdc89e982011-01-28 17:05:48 -050078struct kmem_cache *btrfs_free_space_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040079
80#define S_SHIFT 12
81static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
89};
90
Eric Sandeen3972f262013-01-12 02:57:22 +000091static int btrfs_setsize(struct inode *inode, struct iattr *attr);
Josef Bacika41ad392011-01-31 15:30:16 -050092static int btrfs_truncate(struct inode *inode);
Josef Bacik5fd02042012-05-02 14:00:54 -040093static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
Chris Mason771ed682008-11-06 22:02:51 -050094static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, int *page_started,
97 unsigned long *nr_written, int unlock);
Josef Bacik70c8a912012-10-11 16:54:30 -040098static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
99 u64 len, u64 orig_start,
100 u64 block_start, u64 block_len,
101 u64 orig_block_len, int type);
Josef Bacik7b128762008-07-24 12:17:14 -0400102
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000103static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
Eric Paris2a7dba32011-02-01 11:05:39 -0500104 struct inode *inode, struct inode *dir,
105 const struct qstr *qstr)
Jim Owens0279b4c2009-02-04 09:29:13 -0500106{
107 int err;
108
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000109 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500110 if (!err)
Eric Paris2a7dba32011-02-01 11:05:39 -0500111 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
Jim Owens0279b4c2009-02-04 09:29:13 -0500112 return err;
113}
114
Chris Masond352ac62008-09-29 15:18:18 -0400115/*
Chris Masonc8b97812008-10-29 14:49:59 -0400116 * this does all the hard work for inserting an inline extent into
117 * the btree. The caller should have done a btrfs_drop_extents so that
118 * no overlapping inline items exist in the btree
119 */
Chris Masond3977122009-01-05 21:25:51 -0500120static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400121 struct btrfs_root *root, struct inode *inode,
122 u64 start, size_t size, size_t compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000123 int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400124 struct page **compressed_pages)
125{
126 struct btrfs_key key;
127 struct btrfs_path *path;
128 struct extent_buffer *leaf;
129 struct page *page = NULL;
130 char *kaddr;
131 unsigned long ptr;
132 struct btrfs_file_extent_item *ei;
133 int err = 0;
134 int ret;
135 size_t cur_size = size;
136 size_t datasize;
137 unsigned long offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400138
Li Zefanfe3f5662011-03-28 08:30:38 +0000139 if (compressed_size && compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400140 cur_size = compressed_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400141
Chris Masond3977122009-01-05 21:25:51 -0500142 path = btrfs_alloc_path();
143 if (!path)
Chris Masonc8b97812008-10-29 14:49:59 -0400144 return -ENOMEM;
145
Chris Masonb9473432009-03-13 11:00:37 -0400146 path->leave_spinning = 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400147
Li Zefan33345d012011-04-20 10:31:50 +0800148 key.objectid = btrfs_ino(inode);
Chris Masonc8b97812008-10-29 14:49:59 -0400149 key.offset = start;
150 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
Chris Masonc8b97812008-10-29 14:49:59 -0400151 datasize = btrfs_file_extent_calc_inline_size(cur_size);
152
153 inode_add_bytes(inode, size);
154 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 datasize);
Chris Masonc8b97812008-10-29 14:49:59 -0400156 if (ret) {
157 err = ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400158 goto fail;
159 }
160 leaf = path->nodes[0];
161 ei = btrfs_item_ptr(leaf, path->slots[0],
162 struct btrfs_file_extent_item);
163 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
164 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
165 btrfs_set_file_extent_encryption(leaf, ei, 0);
166 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
167 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
168 ptr = btrfs_file_extent_inline_start(ei);
169
Li Zefan261507a02010-12-17 14:21:50 +0800170 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400171 struct page *cpage;
172 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500173 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400174 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500175 cur_size = min_t(unsigned long, compressed_size,
Chris Masonc8b97812008-10-29 14:49:59 -0400176 PAGE_CACHE_SIZE);
177
Cong Wang7ac687d2011-11-25 23:14:28 +0800178 kaddr = kmap_atomic(cpage);
Chris Masonc8b97812008-10-29 14:49:59 -0400179 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800180 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400181
182 i++;
183 ptr += cur_size;
184 compressed_size -= cur_size;
185 }
186 btrfs_set_file_extent_compression(leaf, ei,
Li Zefan261507a02010-12-17 14:21:50 +0800187 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400188 } else {
189 page = find_get_page(inode->i_mapping,
190 start >> PAGE_CACHE_SHIFT);
191 btrfs_set_file_extent_compression(leaf, ei, 0);
Cong Wang7ac687d2011-11-25 23:14:28 +0800192 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400193 offset = start & (PAGE_CACHE_SIZE - 1);
194 write_extent_buffer(leaf, kaddr + offset, ptr, size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800195 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400196 page_cache_release(page);
197 }
198 btrfs_mark_buffer_dirty(leaf);
199 btrfs_free_path(path);
200
Yan, Zhengc2167752009-11-12 09:34:21 +0000201 /*
202 * we're an inline extent, so nobody can
203 * extend the file past i_size without locking
204 * a page we already have locked.
205 *
206 * We must do any isize and inode updates
207 * before we unlock the pages. Otherwise we
208 * could end up racing with unlink.
209 */
Chris Masonc8b97812008-10-29 14:49:59 -0400210 BTRFS_I(inode)->disk_i_size = inode->i_size;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100211 ret = btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000212
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100213 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400214fail:
215 btrfs_free_path(path);
216 return err;
217}
218
219
220/*
221 * conditionally insert an inline extent into the file. This
222 * does the checks required to make sure the data is small enough
223 * to fit as an inline extent.
224 */
Chris Mason7f366cf2009-03-12 20:12:45 -0400225static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400226 struct btrfs_root *root,
227 struct inode *inode, u64 start, u64 end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000228 size_t compressed_size, int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400229 struct page **compressed_pages)
230{
231 u64 isize = i_size_read(inode);
232 u64 actual_end = min(end + 1, isize);
233 u64 inline_len = actual_end - start;
234 u64 aligned_end = (end + root->sectorsize - 1) &
235 ~((u64)root->sectorsize - 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400236 u64 data_len = inline_len;
237 int ret;
238
239 if (compressed_size)
240 data_len = compressed_size;
241
242 if (start > 0 ||
Chris Mason70b99e62008-10-31 12:46:39 -0400243 actual_end >= PAGE_CACHE_SIZE ||
Chris Masonc8b97812008-10-29 14:49:59 -0400244 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
245 (!compressed_size &&
246 (actual_end & (root->sectorsize - 1)) == 0) ||
247 end + 1 < isize ||
248 data_len > root->fs_info->max_inline) {
249 return 1;
250 }
251
Josef Bacik26714852012-08-29 12:24:27 -0400252 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100253 if (ret)
254 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400255
256 if (isize > actual_end)
257 inline_len = min_t(u64, isize, actual_end);
258 ret = insert_inline_extent(trans, root, inode, start,
259 inline_len, compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000260 compress_type, compressed_pages);
Josef Bacik2adcac12012-05-23 16:10:14 -0400261 if (ret && ret != -ENOSPC) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100262 btrfs_abort_transaction(trans, root, ret);
263 return ret;
Josef Bacik2adcac12012-05-23 16:10:14 -0400264 } else if (ret == -ENOSPC) {
265 return 1;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100266 }
Josef Bacik2adcac12012-05-23 16:10:14 -0400267
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400268 btrfs_delalloc_release_metadata(inode, end + 1 - start);
Chris Masona1ed835e12009-09-11 12:27:37 -0400269 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400270 return 0;
271}
272
Chris Mason771ed682008-11-06 22:02:51 -0500273struct async_extent {
274 u64 start;
275 u64 ram_size;
276 u64 compressed_size;
277 struct page **pages;
278 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800279 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500280 struct list_head list;
281};
282
283struct async_cow {
284 struct inode *inode;
285 struct btrfs_root *root;
286 struct page *locked_page;
287 u64 start;
288 u64 end;
289 struct list_head extents;
290 struct btrfs_work work;
291};
292
293static noinline int add_async_extent(struct async_cow *cow,
294 u64 start, u64 ram_size,
295 u64 compressed_size,
296 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800297 unsigned long nr_pages,
298 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500299{
300 struct async_extent *async_extent;
301
302 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100303 BUG_ON(!async_extent); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500304 async_extent->start = start;
305 async_extent->ram_size = ram_size;
306 async_extent->compressed_size = compressed_size;
307 async_extent->pages = pages;
308 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800309 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500310 list_add_tail(&async_extent->list, &cow->extents);
311 return 0;
312}
313
Chris Masonc8b97812008-10-29 14:49:59 -0400314/*
Chris Mason771ed682008-11-06 22:02:51 -0500315 * we create compressed extents in two phases. The first
316 * phase compresses a range of pages that have already been
317 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400318 *
Chris Mason771ed682008-11-06 22:02:51 -0500319 * This is done inside an ordered work queue, and the compression
320 * is spread across many cpus. The actual IO submission is step
321 * two, and the ordered work queue takes care of making sure that
322 * happens in the same order things were put onto the queue by
323 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400324 *
Chris Mason771ed682008-11-06 22:02:51 -0500325 * If this code finds it can't get good compression, it puts an
326 * entry onto the work queue to write the uncompressed bytes. This
327 * makes sure that both compressed inodes and uncompressed inodes
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300328 * are written in the same order that the flusher thread sent them
329 * down.
Chris Masond352ac62008-09-29 15:18:18 -0400330 */
Chris Mason771ed682008-11-06 22:02:51 -0500331static noinline int compress_file_range(struct inode *inode,
332 struct page *locked_page,
333 u64 start, u64 end,
334 struct async_cow *async_cow,
335 int *num_added)
Chris Masonb888db2b2007-08-27 16:49:44 -0400336{
337 struct btrfs_root *root = BTRFS_I(inode)->root;
338 struct btrfs_trans_handle *trans;
Chris Masondb945352007-10-15 16:15:53 -0400339 u64 num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400340 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400341 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500342 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400343 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400344 struct page **pages = NULL;
345 unsigned long nr_pages;
346 unsigned long nr_pages_ret = 0;
347 unsigned long total_compressed = 0;
348 unsigned long total_in = 0;
349 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500350 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400351 int i;
352 int will_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800353 int compress_type = root->fs_info->compress_type;
Chris Masonb888db2b2007-08-27 16:49:44 -0400354
Liu Bo4cb13e52012-03-29 09:57:45 -0400355 /* if this is a small write inside eof, kick off a defrag */
356 if ((end - start + 1) < 16 * 1024 &&
357 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Chris Mason4cb53002011-05-24 15:35:30 -0400358 btrfs_add_inode_defrag(NULL, inode);
359
Chris Mason42dc7ba2008-12-15 11:44:56 -0500360 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400361again:
362 will_compress = 0;
363 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
364 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
365
Chris Masonf03d9301f2009-02-04 09:31:06 -0500366 /*
367 * we don't want to send crud past the end of i_size through
368 * compression, that's just a waste of CPU time. So, if the
369 * end of the file is before the start of our current
370 * requested range of bytes, we bail out to the uncompressed
371 * cleanup code that can deal with all of this.
372 *
373 * It isn't really the fastest way to fix things, but this is a
374 * very uncommon corner.
375 */
376 if (actual_end <= start)
377 goto cleanup_and_bail_uncompressed;
378
Chris Masonc8b97812008-10-29 14:49:59 -0400379 total_compressed = actual_end - start;
380
381 /* we want to make sure that amount of ram required to uncompress
382 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500383 * of a compressed extent to 128k. This is a crucial number
384 * because it also controls how easily we can spread reads across
385 * cpus for decompression.
386 *
387 * We also want to make sure the amount of IO required to do
388 * a random read is reasonably small, so we limit the size of
389 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400390 */
391 total_compressed = min(total_compressed, max_uncompressed);
Chris Masondb945352007-10-15 16:15:53 -0400392 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
Chris Masonbe20aa92007-12-17 20:14:01 -0500393 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400394 total_in = 0;
395 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400396
Chris Mason771ed682008-11-06 22:02:51 -0500397 /*
398 * we do compression for mount -o compress and when the
399 * inode has not been flagged as nocompress. This flag can
400 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400401 */
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200402 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
Chris Mason1e701a32010-03-11 09:42:04 -0500403 (btrfs_test_opt(root, COMPRESS) ||
Liu Bo75e7cb72011-03-22 10:12:20 +0000404 (BTRFS_I(inode)->force_compress) ||
405 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
Chris Masonc8b97812008-10-29 14:49:59 -0400406 WARN_ON(pages);
Chris Masoncfbc2462008-10-30 13:22:14 -0400407 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
Li Zefan560f7d72011-09-08 10:22:01 +0800408 if (!pages) {
409 /* just bail out to the uncompressed code */
410 goto cont;
411 }
Chris Mason179e29e2007-11-01 11:28:41 -0400412
Li Zefan261507a02010-12-17 14:21:50 +0800413 if (BTRFS_I(inode)->force_compress)
414 compress_type = BTRFS_I(inode)->force_compress;
415
416 ret = btrfs_compress_pages(compress_type,
417 inode->i_mapping, start,
418 total_compressed, pages,
419 nr_pages, &nr_pages_ret,
420 &total_in,
421 &total_compressed,
422 max_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400423
424 if (!ret) {
425 unsigned long offset = total_compressed &
426 (PAGE_CACHE_SIZE - 1);
427 struct page *page = pages[nr_pages_ret - 1];
428 char *kaddr;
429
430 /* zero the tail end of the last page, we might be
431 * sending it down to disk
432 */
433 if (offset) {
Cong Wang7ac687d2011-11-25 23:14:28 +0800434 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400435 memset(kaddr + offset, 0,
436 PAGE_CACHE_SIZE - offset);
Cong Wang7ac687d2011-11-25 23:14:28 +0800437 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400438 }
439 will_compress = 1;
440 }
441 }
Li Zefan560f7d72011-09-08 10:22:01 +0800442cont:
Chris Masonc8b97812008-10-29 14:49:59 -0400443 if (start == 0) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400444 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100445 if (IS_ERR(trans)) {
446 ret = PTR_ERR(trans);
447 trans = NULL;
448 goto cleanup_and_out;
449 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400450 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500451
Chris Masonc8b97812008-10-29 14:49:59 -0400452 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500453 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400454 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500455 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400456 */
457 ret = cow_file_range_inline(trans, root, inode,
Li Zefanfe3f5662011-03-28 08:30:38 +0000458 start, end, 0, 0, NULL);
Chris Masonc8b97812008-10-29 14:49:59 -0400459 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500460 /* try making a compressed inline extent */
Chris Masonc8b97812008-10-29 14:49:59 -0400461 ret = cow_file_range_inline(trans, root, inode,
462 start, end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000463 total_compressed,
464 compress_type, pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400465 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100466 if (ret <= 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500467 /*
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100468 * inline extent creation worked or returned error,
469 * we don't need to create any more async work items.
470 * Unlock and free up our temp pages.
Chris Mason771ed682008-11-06 22:02:51 -0500471 */
Chris Masonc8b97812008-10-29 14:49:59 -0400472 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400473 &BTRFS_I(inode)->io_tree,
474 start, end, NULL,
475 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
Chris Masona3429ab2009-10-08 12:30:20 -0400476 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400477 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000478
479 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400480 goto free_pages_out;
481 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000482 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400483 }
484
485 if (will_compress) {
486 /*
487 * we aren't doing an inline extent round the compressed size
488 * up to a block size boundary so the allocator does sane
489 * things
490 */
491 total_compressed = (total_compressed + blocksize - 1) &
492 ~(blocksize - 1);
493
494 /*
495 * one last check to make sure the compression is really a
496 * win, compare the page count read with the blocks on disk
497 */
498 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
499 ~(PAGE_CACHE_SIZE - 1);
500 if (total_compressed >= total_in) {
501 will_compress = 0;
502 } else {
Chris Masonc8b97812008-10-29 14:49:59 -0400503 num_bytes = total_in;
504 }
505 }
506 if (!will_compress && pages) {
507 /*
508 * the compression code ran but failed to make things smaller,
509 * free any pages it allocated and our page pointer array
510 */
511 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400512 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400513 page_cache_release(pages[i]);
514 }
515 kfree(pages);
516 pages = NULL;
517 total_compressed = 0;
518 nr_pages_ret = 0;
519
520 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500521 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
522 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500523 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500524 }
Chris Masonc8b97812008-10-29 14:49:59 -0400525 }
Chris Mason771ed682008-11-06 22:02:51 -0500526 if (will_compress) {
527 *num_added += 1;
528
529 /* the async work queues will take care of doing actual
530 * allocation on disk for these compressed pages,
531 * and will submit them to the elevator.
532 */
533 add_async_extent(async_cow, start, num_bytes,
Li Zefan261507a02010-12-17 14:21:50 +0800534 total_compressed, pages, nr_pages_ret,
535 compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500536
Yan, Zheng24ae6362010-12-06 07:02:36 +0000537 if (start + num_bytes < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500538 start += num_bytes;
539 pages = NULL;
540 cond_resched();
541 goto again;
542 }
543 } else {
Chris Masonf03d9301f2009-02-04 09:31:06 -0500544cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500545 /*
546 * No compression, but we still need to write the pages in
547 * the file we've been given so far. redirty the locked
548 * page if it corresponds to our extent and set things up
549 * for the async work queue to run cow_file_range to do
550 * the normal delalloc dance
551 */
552 if (page_offset(locked_page) >= start &&
553 page_offset(locked_page) <= end) {
554 __set_page_dirty_nobuffers(locked_page);
555 /* unlocked later on in the async handlers */
556 }
Li Zefan261507a02010-12-17 14:21:50 +0800557 add_async_extent(async_cow, start, end - start + 1,
558 0, NULL, 0, BTRFS_COMPRESS_NONE);
Chris Mason771ed682008-11-06 22:02:51 -0500559 *num_added += 1;
560 }
561
562out:
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100563 return ret;
Chris Mason771ed682008-11-06 22:02:51 -0500564
565free_pages_out:
566 for (i = 0; i < nr_pages_ret; i++) {
567 WARN_ON(pages[i]->mapping);
568 page_cache_release(pages[i]);
569 }
Chris Masond3977122009-01-05 21:25:51 -0500570 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500571
572 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100573
574cleanup_and_out:
575 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
576 start, end, NULL,
577 EXTENT_CLEAR_UNLOCK_PAGE |
578 EXTENT_CLEAR_DIRTY |
579 EXTENT_CLEAR_DELALLOC |
580 EXTENT_SET_WRITEBACK |
581 EXTENT_END_WRITEBACK);
582 if (!trans || IS_ERR(trans))
583 btrfs_error(root->fs_info, ret, "Failed to join transaction");
584 else
585 btrfs_abort_transaction(trans, root, ret);
586 goto free_pages_out;
Chris Mason771ed682008-11-06 22:02:51 -0500587}
588
589/*
590 * phase two of compressed writeback. This is the ordered portion
591 * of the code, which only gets called in the order the work was
592 * queued. We walk all the async extents created by compress_file_range
593 * and send them down to the disk.
594 */
595static noinline int submit_compressed_extents(struct inode *inode,
596 struct async_cow *async_cow)
597{
598 struct async_extent *async_extent;
599 u64 alloc_hint = 0;
600 struct btrfs_trans_handle *trans;
601 struct btrfs_key ins;
602 struct extent_map *em;
603 struct btrfs_root *root = BTRFS_I(inode)->root;
604 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
605 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500606 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500607
608 if (list_empty(&async_cow->extents))
609 return 0;
610
Chris Mason771ed682008-11-06 22:02:51 -0500611
Chris Masond3977122009-01-05 21:25:51 -0500612 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500613 async_extent = list_entry(async_cow->extents.next,
614 struct async_extent, list);
615 list_del(&async_extent->list);
616
617 io_tree = &BTRFS_I(inode)->io_tree;
618
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500619retry:
Chris Mason771ed682008-11-06 22:02:51 -0500620 /* did the compression code fall back to uncompressed IO? */
621 if (!async_extent->pages) {
622 int page_started = 0;
623 unsigned long nr_written = 0;
624
625 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000626 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100627 async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500628
629 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500630 ret = cow_file_range(inode, async_cow->locked_page,
631 async_extent->start,
632 async_extent->start +
633 async_extent->ram_size - 1,
634 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500635
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100636 /* JDM XXX */
637
Chris Mason771ed682008-11-06 22:02:51 -0500638 /*
639 * if page_started, cow_file_range inserted an
640 * inline extent and took care of all the unlocking
641 * and IO for us. Otherwise, we need to submit
642 * all those pages down to the drive.
643 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500644 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500645 extent_write_locked_range(io_tree,
646 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500647 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500648 async_extent->ram_size - 1,
649 btrfs_get_extent,
650 WB_SYNC_ALL);
651 kfree(async_extent);
652 cond_resched();
653 continue;
654 }
655
656 lock_extent(io_tree, async_extent->start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100657 async_extent->start + async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500658
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400659 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100660 if (IS_ERR(trans)) {
661 ret = PTR_ERR(trans);
662 } else {
663 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
664 ret = btrfs_reserve_extent(trans, root,
Chris Mason771ed682008-11-06 22:02:51 -0500665 async_extent->compressed_size,
666 async_extent->compressed_size,
Josef Bacik81c9ad22012-01-18 10:56:06 -0500667 0, alloc_hint, &ins, 1);
Miao Xie962197b2012-09-11 23:27:35 -0600668 if (ret && ret != -ENOSPC)
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100669 btrfs_abort_transaction(trans, root, ret);
670 btrfs_end_transaction(trans, root);
671 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000672
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500673 if (ret) {
674 int i;
675 for (i = 0; i < async_extent->nr_pages; i++) {
676 WARN_ON(async_extent->pages[i]->mapping);
677 page_cache_release(async_extent->pages[i]);
678 }
679 kfree(async_extent->pages);
680 async_extent->nr_pages = 0;
681 async_extent->pages = NULL;
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100684 async_extent->ram_size - 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100685 if (ret == -ENOSPC)
686 goto retry;
687 goto out_free; /* JDM: Requeue? */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500688 }
689
Yan, Zhengc2167752009-11-12 09:34:21 +0000690 /*
691 * here we're doing allocation and writeback of the
692 * compressed pages
693 */
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
697
David Sterba172ddd62011-04-21 00:48:27 +0200698 em = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100699 BUG_ON(!em); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500700 em->start = async_extent->start;
701 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500702 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500703
704 em->block_start = ins.objectid;
705 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500706 em->orig_block_len = ins.offset;
Chris Mason771ed682008-11-06 22:02:51 -0500707 em->bdev = root->fs_info->fs_devices->latest_bdev;
Li Zefan261507a02010-12-17 14:21:50 +0800708 em->compress_type = async_extent->compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500709 set_bit(EXTENT_FLAG_PINNED, &em->flags);
710 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400711 em->generation = -1;
Chris Mason771ed682008-11-06 22:02:51 -0500712
Chris Masond3977122009-01-05 21:25:51 -0500713 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400714 write_lock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500715 ret = add_extent_mapping(em_tree, em);
Josef Bacik70c8a912012-10-11 16:54:30 -0400716 if (!ret)
717 list_move(&em->list,
718 &em_tree->modified_extents);
Chris Mason890871b2009-09-02 16:24:52 -0400719 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500720 if (ret != -EEXIST) {
721 free_extent_map(em);
722 break;
723 }
724 btrfs_drop_extent_cache(inode, async_extent->start,
725 async_extent->start +
726 async_extent->ram_size - 1, 0);
727 }
728
Li Zefan261507a02010-12-17 14:21:50 +0800729 ret = btrfs_add_ordered_extent_compress(inode,
730 async_extent->start,
731 ins.objectid,
732 async_extent->ram_size,
733 ins.offset,
734 BTRFS_ORDERED_COMPRESSED,
735 async_extent->compress_type);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100736 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500737
Chris Mason771ed682008-11-06 22:02:51 -0500738 /*
739 * clear dirty, set writeback and unlock the pages.
740 */
741 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400742 &BTRFS_I(inode)->io_tree,
743 async_extent->start,
744 async_extent->start +
745 async_extent->ram_size - 1,
746 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
747 EXTENT_CLEAR_UNLOCK |
Chris Masona3429ab2009-10-08 12:30:20 -0400748 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400749 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500750
751 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500752 async_extent->start,
753 async_extent->ram_size,
754 ins.objectid,
755 ins.offset, async_extent->pages,
756 async_extent->nr_pages);
Chris Mason771ed682008-11-06 22:02:51 -0500757
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100758 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500759 alloc_hint = ins.objectid + ins.offset;
760 kfree(async_extent);
761 cond_resched();
762 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100763 ret = 0;
764out:
765 return ret;
766out_free:
767 kfree(async_extent);
768 goto out;
Chris Mason771ed682008-11-06 22:02:51 -0500769}
770
Josef Bacik4b46fce2010-05-23 11:00:55 -0400771static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
772 u64 num_bytes)
773{
774 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
775 struct extent_map *em;
776 u64 alloc_hint = 0;
777
778 read_lock(&em_tree->lock);
779 em = search_extent_mapping(em_tree, start, num_bytes);
780 if (em) {
781 /*
782 * if block start isn't an actual block number then find the
783 * first block in this inode and use that as a hint. If that
784 * block is also bogus then just don't worry about it.
785 */
786 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
787 free_extent_map(em);
788 em = search_extent_mapping(em_tree, 0, 0);
789 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
790 alloc_hint = em->block_start;
791 if (em)
792 free_extent_map(em);
793 } else {
794 alloc_hint = em->block_start;
795 free_extent_map(em);
796 }
797 }
798 read_unlock(&em_tree->lock);
799
800 return alloc_hint;
801}
802
Chris Mason771ed682008-11-06 22:02:51 -0500803/*
804 * when extent_io.c finds a delayed allocation range in the file,
805 * the call backs end up in this code. The basic idea is to
806 * allocate extents on disk for the range, and create ordered data structs
807 * in ram to track those extents.
808 *
809 * locked_page is the page that writepage had locked already. We use
810 * it to make sure we don't do extra locks or unlocks.
811 *
812 * *page_started is set to one if we unlock locked_page and do everything
813 * required to start IO on it. It may be clean and already done with
814 * IO when we return.
815 */
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000816static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
817 struct inode *inode,
818 struct btrfs_root *root,
819 struct page *locked_page,
820 u64 start, u64 end, int *page_started,
821 unsigned long *nr_written,
822 int unlock)
Chris Mason771ed682008-11-06 22:02:51 -0500823{
Chris Mason771ed682008-11-06 22:02:51 -0500824 u64 alloc_hint = 0;
825 u64 num_bytes;
826 unsigned long ram_size;
827 u64 disk_num_bytes;
828 u64 cur_alloc_size;
829 u64 blocksize = root->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500830 struct btrfs_key ins;
831 struct extent_map *em;
832 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
833 int ret = 0;
834
Liu Bo83eea1f2012-07-10 05:28:39 -0600835 BUG_ON(btrfs_is_free_space_inode(inode));
Chris Mason771ed682008-11-06 22:02:51 -0500836
Chris Mason771ed682008-11-06 22:02:51 -0500837 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
838 num_bytes = max(blocksize, num_bytes);
839 disk_num_bytes = num_bytes;
Chris Mason771ed682008-11-06 22:02:51 -0500840
Chris Mason4cb53002011-05-24 15:35:30 -0400841 /* if this is a small write inside eof, kick off defrag */
Liu Bo4cb13e52012-03-29 09:57:45 -0400842 if (num_bytes < 64 * 1024 &&
843 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Chris Mason4cb53002011-05-24 15:35:30 -0400844 btrfs_add_inode_defrag(trans, inode);
845
Chris Mason771ed682008-11-06 22:02:51 -0500846 if (start == 0) {
847 /* lets try to make an inline extent */
848 ret = cow_file_range_inline(trans, root, inode,
Li Zefanfe3f5662011-03-28 08:30:38 +0000849 start, end, 0, 0, NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500850 if (ret == 0) {
851 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400852 &BTRFS_I(inode)->io_tree,
853 start, end, NULL,
854 EXTENT_CLEAR_UNLOCK_PAGE |
855 EXTENT_CLEAR_UNLOCK |
856 EXTENT_CLEAR_DELALLOC |
857 EXTENT_CLEAR_DIRTY |
858 EXTENT_SET_WRITEBACK |
859 EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000860
Chris Mason771ed682008-11-06 22:02:51 -0500861 *nr_written = *nr_written +
862 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
863 *page_started = 1;
Chris Mason771ed682008-11-06 22:02:51 -0500864 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100865 } else if (ret < 0) {
866 btrfs_abort_transaction(trans, root, ret);
867 goto out_unlock;
Chris Mason771ed682008-11-06 22:02:51 -0500868 }
869 }
Chris Masonc8b97812008-10-29 14:49:59 -0400870
871 BUG_ON(disk_num_bytes >
David Sterba6c417612011-04-13 15:41:04 +0200872 btrfs_super_total_bytes(root->fs_info->super_copy));
Chris Masonc8b97812008-10-29 14:49:59 -0400873
Josef Bacik4b46fce2010-05-23 11:00:55 -0400874 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400875 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400876
Chris Masond3977122009-01-05 21:25:51 -0500877 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400878 unsigned long op;
879
Josef Bacik287a0ab2010-03-19 18:07:23 +0000880 cur_alloc_size = disk_num_bytes;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400881 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500882 root->sectorsize, 0, alloc_hint,
Josef Bacik81c9ad22012-01-18 10:56:06 -0500883 &ins, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100884 if (ret < 0) {
885 btrfs_abort_transaction(trans, root, ret);
886 goto out_unlock;
887 }
Chris Masond3977122009-01-05 21:25:51 -0500888
David Sterba172ddd62011-04-21 00:48:27 +0200889 em = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100890 BUG_ON(!em); /* -ENOMEM */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400891 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500892 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500893 ram_size = ins.offset;
894 em->len = ins.offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400895
Chris Masone6dcd2d2008-07-17 12:53:50 -0400896 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400897 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500898 em->orig_block_len = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400899 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400900 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400901 em->generation = -1;
Chris Masonc8b97812008-10-29 14:49:59 -0400902
Chris Masond3977122009-01-05 21:25:51 -0500903 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400904 write_lock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400905 ret = add_extent_mapping(em_tree, em);
Josef Bacik70c8a912012-10-11 16:54:30 -0400906 if (!ret)
907 list_move(&em->list,
908 &em_tree->modified_extents);
Chris Mason890871b2009-09-02 16:24:52 -0400909 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400910 if (ret != -EEXIST) {
911 free_extent_map(em);
912 break;
913 }
914 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400915 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400916 }
917
Chris Mason98d20f62008-04-14 09:46:10 -0400918 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400919 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -0500920 ram_size, cur_alloc_size, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100921 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400922
Yan Zheng17d217f2008-12-12 10:03:38 -0500923 if (root->root_key.objectid ==
924 BTRFS_DATA_RELOC_TREE_OBJECTID) {
925 ret = btrfs_reloc_clone_csums(inode, start,
926 cur_alloc_size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100927 if (ret) {
928 btrfs_abort_transaction(trans, root, ret);
929 goto out_unlock;
930 }
Yan Zheng17d217f2008-12-12 10:03:38 -0500931 }
932
Chris Masond3977122009-01-05 21:25:51 -0500933 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -0400934 break;
Chris Masond3977122009-01-05 21:25:51 -0500935
Chris Masonc8b97812008-10-29 14:49:59 -0400936 /* we're not doing compressed IO, don't unlock the first
937 * page (which the caller expects to stay locked), don't
938 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -0400939 *
940 * Do set the Private2 bit so we know this page was properly
941 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -0400942 */
Chris Masona791e352009-10-08 11:27:10 -0400943 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
944 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
945 EXTENT_SET_PRIVATE2;
946
Chris Masonc8b97812008-10-29 14:49:59 -0400947 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
948 start, start + ram_size - 1,
Chris Masona791e352009-10-08 11:27:10 -0400949 locked_page, op);
Chris Masonc8b97812008-10-29 14:49:59 -0400950 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -0500951 num_bytes -= cur_alloc_size;
952 alloc_hint = ins.objectid + ins.offset;
953 start += cur_alloc_size;
Chris Masonb888db2b2007-08-27 16:49:44 -0400954 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100955out:
Chris Masonbe20aa92007-12-17 20:14:01 -0500956 return ret;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000957
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100958out_unlock:
959 extent_clear_unlock_delalloc(inode,
960 &BTRFS_I(inode)->io_tree,
Josef Bacikbeb42dd2012-05-30 15:35:17 -0400961 start, end, locked_page,
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100962 EXTENT_CLEAR_UNLOCK_PAGE |
963 EXTENT_CLEAR_UNLOCK |
964 EXTENT_CLEAR_DELALLOC |
965 EXTENT_CLEAR_DIRTY |
966 EXTENT_SET_WRITEBACK |
967 EXTENT_END_WRITEBACK);
968
969 goto out;
Chris Mason771ed682008-11-06 22:02:51 -0500970}
Chris Masonc8b97812008-10-29 14:49:59 -0400971
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000972static noinline int cow_file_range(struct inode *inode,
973 struct page *locked_page,
974 u64 start, u64 end, int *page_started,
975 unsigned long *nr_written,
976 int unlock)
977{
978 struct btrfs_trans_handle *trans;
979 struct btrfs_root *root = BTRFS_I(inode)->root;
980 int ret;
981
982 trans = btrfs_join_transaction(root);
983 if (IS_ERR(trans)) {
984 extent_clear_unlock_delalloc(inode,
985 &BTRFS_I(inode)->io_tree,
986 start, end, locked_page,
987 EXTENT_CLEAR_UNLOCK_PAGE |
988 EXTENT_CLEAR_UNLOCK |
989 EXTENT_CLEAR_DELALLOC |
990 EXTENT_CLEAR_DIRTY |
991 EXTENT_SET_WRITEBACK |
992 EXTENT_END_WRITEBACK);
993 return PTR_ERR(trans);
994 }
995 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
996
997 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
998 page_started, nr_written, unlock);
999
1000 btrfs_end_transaction(trans, root);
1001
1002 return ret;
1003}
1004
Chris Mason771ed682008-11-06 22:02:51 -05001005/*
1006 * work queue call back to started compression on a file and pages
1007 */
1008static noinline void async_cow_start(struct btrfs_work *work)
1009{
1010 struct async_cow *async_cow;
1011 int num_added = 0;
1012 async_cow = container_of(work, struct async_cow, work);
1013
1014 compress_file_range(async_cow->inode, async_cow->locked_page,
1015 async_cow->start, async_cow->end, async_cow,
1016 &num_added);
Josef Bacik8180ef82012-06-08 15:16:12 -04001017 if (num_added == 0) {
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001018 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001019 async_cow->inode = NULL;
Josef Bacik8180ef82012-06-08 15:16:12 -04001020 }
Chris Mason771ed682008-11-06 22:02:51 -05001021}
1022
1023/*
1024 * work queue call back to submit previously compressed pages
1025 */
1026static noinline void async_cow_submit(struct btrfs_work *work)
1027{
1028 struct async_cow *async_cow;
1029 struct btrfs_root *root;
1030 unsigned long nr_pages;
1031
1032 async_cow = container_of(work, struct async_cow, work);
1033
1034 root = async_cow->root;
1035 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1036 PAGE_CACHE_SHIFT;
1037
Josef Bacik66657b32012-08-01 15:36:24 -04001038 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
Liu Bo287082b2012-06-28 04:02:24 -06001039 5 * 1024 * 1024 &&
Chris Mason771ed682008-11-06 22:02:51 -05001040 waitqueue_active(&root->fs_info->async_submit_wait))
1041 wake_up(&root->fs_info->async_submit_wait);
1042
Chris Masond3977122009-01-05 21:25:51 -05001043 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -05001044 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -05001045}
Chris Masonc8b97812008-10-29 14:49:59 -04001046
Chris Mason771ed682008-11-06 22:02:51 -05001047static noinline void async_cow_free(struct btrfs_work *work)
1048{
1049 struct async_cow *async_cow;
1050 async_cow = container_of(work, struct async_cow, work);
Josef Bacik8180ef82012-06-08 15:16:12 -04001051 if (async_cow->inode)
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001052 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001053 kfree(async_cow);
1054}
1055
1056static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1057 u64 start, u64 end, int *page_started,
1058 unsigned long *nr_written)
1059{
1060 struct async_cow *async_cow;
1061 struct btrfs_root *root = BTRFS_I(inode)->root;
1062 unsigned long nr_pages;
1063 u64 cur_end;
Liu Bo287082b2012-06-28 04:02:24 -06001064 int limit = 10 * 1024 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -05001065
Chris Masona3429ab2009-10-08 12:30:20 -04001066 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1067 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -05001068 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -05001069 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001070 BUG_ON(!async_cow); /* -ENOMEM */
Josef Bacik8180ef82012-06-08 15:16:12 -04001071 async_cow->inode = igrab(inode);
Chris Mason771ed682008-11-06 22:02:51 -05001072 async_cow->root = root;
1073 async_cow->locked_page = locked_page;
1074 async_cow->start = start;
1075
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001076 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
Chris Mason771ed682008-11-06 22:02:51 -05001077 cur_end = end;
1078 else
1079 cur_end = min(end, start + 512 * 1024 - 1);
1080
1081 async_cow->end = cur_end;
1082 INIT_LIST_HEAD(&async_cow->extents);
1083
1084 async_cow->work.func = async_cow_start;
1085 async_cow->work.ordered_func = async_cow_submit;
1086 async_cow->work.ordered_free = async_cow_free;
1087 async_cow->work.flags = 0;
1088
Chris Mason771ed682008-11-06 22:02:51 -05001089 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1090 PAGE_CACHE_SHIFT;
1091 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1092
1093 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1094 &async_cow->work);
1095
1096 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1097 wait_event(root->fs_info->async_submit_wait,
1098 (atomic_read(&root->fs_info->async_delalloc_pages) <
1099 limit));
1100 }
1101
Chris Masond3977122009-01-05 21:25:51 -05001102 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -05001103 atomic_read(&root->fs_info->async_delalloc_pages)) {
1104 wait_event(root->fs_info->async_submit_wait,
1105 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1106 0));
1107 }
1108
1109 *nr_written += nr_pages;
1110 start = cur_end + 1;
1111 }
1112 *page_started = 1;
1113 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001114}
1115
Chris Masond3977122009-01-05 21:25:51 -05001116static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -05001117 u64 bytenr, u64 num_bytes)
1118{
1119 int ret;
1120 struct btrfs_ordered_sum *sums;
1121 LIST_HEAD(list);
1122
Yan Zheng07d400a2009-01-06 11:42:00 -05001123 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01001124 bytenr + num_bytes - 1, &list, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001125 if (ret == 0 && list_empty(&list))
1126 return 0;
1127
1128 while (!list_empty(&list)) {
1129 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1130 list_del(&sums->list);
1131 kfree(sums);
1132 }
1133 return 1;
1134}
1135
Chris Masond352ac62008-09-29 15:18:18 -04001136/*
1137 * when nowcow writeback call back. This checks for snapshots or COW copies
1138 * of the extents that exist in the file, and COWs the file as required.
1139 *
1140 * If no cow copies or snapshots exist, we write directly to the existing
1141 * blocks on disk
1142 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001143static noinline int run_delalloc_nocow(struct inode *inode,
1144 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001145 u64 start, u64 end, int *page_started, int force,
1146 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001147{
Chris Masonbe20aa92007-12-17 20:14:01 -05001148 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001149 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001150 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001151 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001152 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001153 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001154 u64 cow_start;
1155 u64 cur_offset;
1156 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001157 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001158 u64 disk_bytenr;
1159 u64 num_bytes;
Josef Bacikb4939682012-12-03 10:31:19 -05001160 u64 disk_num_bytes;
Yan Zheng80ff3852008-10-30 14:20:02 -04001161 int extent_type;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001162 int ret, err;
Yan Zhengd899e052008-10-30 14:25:28 -04001163 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001164 int nocow;
1165 int check_prev = 1;
Li Zefan82d59022011-04-20 10:33:24 +08001166 bool nolock;
Li Zefan33345d012011-04-20 10:31:50 +08001167 u64 ino = btrfs_ino(inode);
Chris Masonbe20aa92007-12-17 20:14:01 -05001168
1169 path = btrfs_alloc_path();
Josef Bacik17ca04a2012-05-31 15:58:55 -04001170 if (!path) {
1171 extent_clear_unlock_delalloc(inode,
1172 &BTRFS_I(inode)->io_tree,
1173 start, end, locked_page,
1174 EXTENT_CLEAR_UNLOCK_PAGE |
1175 EXTENT_CLEAR_UNLOCK |
1176 EXTENT_CLEAR_DELALLOC |
1177 EXTENT_CLEAR_DIRTY |
1178 EXTENT_SET_WRITEBACK |
1179 EXTENT_END_WRITEBACK);
Mark Fashehd8926bb2011-07-13 10:38:47 -07001180 return -ENOMEM;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001181 }
Li Zefan82d59022011-04-20 10:33:24 +08001182
Liu Bo83eea1f2012-07-10 05:28:39 -06001183 nolock = btrfs_is_free_space_inode(inode);
Li Zefan82d59022011-04-20 10:33:24 +08001184
1185 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001186 trans = btrfs_join_transaction_nolock(root);
Li Zefan82d59022011-04-20 10:33:24 +08001187 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001188 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04001189
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001190 if (IS_ERR(trans)) {
Josef Bacik17ca04a2012-05-31 15:58:55 -04001191 extent_clear_unlock_delalloc(inode,
1192 &BTRFS_I(inode)->io_tree,
1193 start, end, locked_page,
1194 EXTENT_CLEAR_UNLOCK_PAGE |
1195 EXTENT_CLEAR_UNLOCK |
1196 EXTENT_CLEAR_DELALLOC |
1197 EXTENT_CLEAR_DIRTY |
1198 EXTENT_SET_WRITEBACK |
1199 EXTENT_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001200 btrfs_free_path(path);
1201 return PTR_ERR(trans);
1202 }
1203
Josef Bacik74b21072011-04-13 12:02:53 -04001204 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Masonbe20aa92007-12-17 20:14:01 -05001205
Yan Zheng80ff3852008-10-30 14:20:02 -04001206 cow_start = (u64)-1;
1207 cur_offset = start;
1208 while (1) {
Li Zefan33345d012011-04-20 10:31:50 +08001209 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Yan Zheng80ff3852008-10-30 14:20:02 -04001210 cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001211 if (ret < 0) {
1212 btrfs_abort_transaction(trans, root, ret);
1213 goto error;
1214 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001215 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1216 leaf = path->nodes[0];
1217 btrfs_item_key_to_cpu(leaf, &found_key,
1218 path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +08001219 if (found_key.objectid == ino &&
Yan Zheng80ff3852008-10-30 14:20:02 -04001220 found_key.type == BTRFS_EXTENT_DATA_KEY)
1221 path->slots[0]--;
1222 }
1223 check_prev = 0;
1224next_slot:
1225 leaf = path->nodes[0];
1226 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1227 ret = btrfs_next_leaf(root, path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001228 if (ret < 0) {
1229 btrfs_abort_transaction(trans, root, ret);
1230 goto error;
1231 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001232 if (ret > 0)
1233 break;
1234 leaf = path->nodes[0];
1235 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001236
Yan Zheng80ff3852008-10-30 14:20:02 -04001237 nocow = 0;
1238 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001239 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001240 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001241
Li Zefan33345d012011-04-20 10:31:50 +08001242 if (found_key.objectid > ino ||
Yan Zheng80ff3852008-10-30 14:20:02 -04001243 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1244 found_key.offset > end)
1245 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001246
Yan Zheng80ff3852008-10-30 14:20:02 -04001247 if (found_key.offset > cur_offset) {
1248 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001249 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001250 goto out_check;
1251 }
Chris Masonc31f8832008-01-08 15:46:31 -05001252
Yan Zheng80ff3852008-10-30 14:20:02 -04001253 fi = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_file_extent_item);
1255 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001256
Yan Zhengd899e052008-10-30 14:25:28 -04001257 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1258 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001259 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001260 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001261 extent_end = found_key.offset +
1262 btrfs_file_extent_num_bytes(leaf, fi);
Josef Bacikb4939682012-12-03 10:31:19 -05001263 disk_num_bytes =
1264 btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001265 if (extent_end <= start) {
1266 path->slots[0]++;
1267 goto next_slot;
1268 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001269 if (disk_bytenr == 0)
1270 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001271 if (btrfs_file_extent_compression(leaf, fi) ||
1272 btrfs_file_extent_encryption(leaf, fi) ||
1273 btrfs_file_extent_other_encoding(leaf, fi))
1274 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001275 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1276 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001277 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001278 goto out_check;
Li Zefan33345d012011-04-20 10:31:50 +08001279 if (btrfs_cross_ref_exist(trans, root, ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001280 found_key.offset -
1281 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001282 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001283 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001284 disk_bytenr += cur_offset - found_key.offset;
1285 num_bytes = min(end + 1, extent_end) - cur_offset;
1286 /*
1287 * force cow if csum exists in the range.
1288 * this ensure that csum for a given extent are
1289 * either valid or do not exist.
1290 */
1291 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1292 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001293 nocow = 1;
1294 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1295 extent_end = found_key.offset +
1296 btrfs_file_extent_inline_len(leaf, fi);
1297 extent_end = ALIGN(extent_end, root->sectorsize);
1298 } else {
1299 BUG_ON(1);
1300 }
1301out_check:
1302 if (extent_end <= start) {
1303 path->slots[0]++;
1304 goto next_slot;
1305 }
1306 if (!nocow) {
1307 if (cow_start == (u64)-1)
1308 cow_start = cur_offset;
1309 cur_offset = extent_end;
1310 if (cur_offset > end)
1311 break;
1312 path->slots[0]++;
1313 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001314 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001315
David Sterbab3b4aa72011-04-21 01:20:15 +02001316 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001317 if (cow_start != (u64)-1) {
Miao Xieb7d5b0a2012-11-01 07:32:18 +00001318 ret = __cow_file_range(trans, inode, root, locked_page,
1319 cow_start, found_key.offset - 1,
1320 page_started, nr_written, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001321 if (ret) {
1322 btrfs_abort_transaction(trans, root, ret);
1323 goto error;
1324 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001325 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001326 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001327
Yan Zhengd899e052008-10-30 14:25:28 -04001328 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1329 struct extent_map *em;
1330 struct extent_map_tree *em_tree;
1331 em_tree = &BTRFS_I(inode)->extent_tree;
David Sterba172ddd62011-04-21 00:48:27 +02001332 em = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001333 BUG_ON(!em); /* -ENOMEM */
Yan Zhengd899e052008-10-30 14:25:28 -04001334 em->start = cur_offset;
Josef Bacik70c8a912012-10-11 16:54:30 -04001335 em->orig_start = found_key.offset - extent_offset;
Yan Zhengd899e052008-10-30 14:25:28 -04001336 em->len = num_bytes;
1337 em->block_len = num_bytes;
1338 em->block_start = disk_bytenr;
Josef Bacikb4939682012-12-03 10:31:19 -05001339 em->orig_block_len = disk_num_bytes;
Yan Zhengd899e052008-10-30 14:25:28 -04001340 em->bdev = root->fs_info->fs_devices->latest_bdev;
1341 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacikb11e2342012-12-03 10:58:15 -05001342 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -04001343 em->generation = -1;
Yan Zhengd899e052008-10-30 14:25:28 -04001344 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001345 write_lock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001346 ret = add_extent_mapping(em_tree, em);
Josef Bacik70c8a912012-10-11 16:54:30 -04001347 if (!ret)
1348 list_move(&em->list,
1349 &em_tree->modified_extents);
Chris Mason890871b2009-09-02 16:24:52 -04001350 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001351 if (ret != -EEXIST) {
1352 free_extent_map(em);
1353 break;
1354 }
1355 btrfs_drop_extent_cache(inode, em->start,
1356 em->start + em->len - 1, 0);
1357 }
1358 type = BTRFS_ORDERED_PREALLOC;
1359 } else {
1360 type = BTRFS_ORDERED_NOCOW;
1361 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001362
1363 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001364 num_bytes, num_bytes, type);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001365 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -05001366
Yan, Zhengefa56462010-05-16 10:49:59 -04001367 if (root->root_key.objectid ==
1368 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1369 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1370 num_bytes);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001371 if (ret) {
1372 btrfs_abort_transaction(trans, root, ret);
1373 goto error;
1374 }
Yan, Zhengefa56462010-05-16 10:49:59 -04001375 }
1376
Yan Zhengd899e052008-10-30 14:25:28 -04001377 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
Chris Masona791e352009-10-08 11:27:10 -04001378 cur_offset, cur_offset + num_bytes - 1,
1379 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1380 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1381 EXTENT_SET_PRIVATE2);
Yan Zheng80ff3852008-10-30 14:20:02 -04001382 cur_offset = extent_end;
1383 if (cur_offset > end)
1384 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001385 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001386 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001387
Josef Bacik17ca04a2012-05-31 15:58:55 -04001388 if (cur_offset <= end && cow_start == (u64)-1) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001389 cow_start = cur_offset;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001390 cur_offset = end;
1391 }
1392
Yan Zheng80ff3852008-10-30 14:20:02 -04001393 if (cow_start != (u64)-1) {
Miao Xieb7d5b0a2012-11-01 07:32:18 +00001394 ret = __cow_file_range(trans, inode, root, locked_page,
1395 cow_start, end,
1396 page_started, nr_written, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001397 if (ret) {
1398 btrfs_abort_transaction(trans, root, ret);
1399 goto error;
1400 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001401 }
1402
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001403error:
Miao Xiea698d0752012-09-20 01:51:59 -06001404 err = btrfs_end_transaction(trans, root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001405 if (!ret)
1406 ret = err;
1407
Josef Bacik17ca04a2012-05-31 15:58:55 -04001408 if (ret && cur_offset < end)
1409 extent_clear_unlock_delalloc(inode,
1410 &BTRFS_I(inode)->io_tree,
1411 cur_offset, end, locked_page,
1412 EXTENT_CLEAR_UNLOCK_PAGE |
1413 EXTENT_CLEAR_UNLOCK |
1414 EXTENT_CLEAR_DELALLOC |
1415 EXTENT_CLEAR_DIRTY |
1416 EXTENT_SET_WRITEBACK |
1417 EXTENT_END_WRITEBACK);
1418
Yan Zheng7ea394f2008-08-05 13:05:02 -04001419 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001420 return ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05001421}
1422
Chris Masond352ac62008-09-29 15:18:18 -04001423/*
1424 * extent_io.c call back to do delayed allocation processing
1425 */
Chris Masonc8b97812008-10-29 14:49:59 -04001426static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001427 u64 start, u64 end, int *page_started,
1428 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001429{
Chris Masonbe20aa92007-12-17 20:14:01 -05001430 int ret;
Chris Mason7f366cf2009-03-12 20:12:45 -04001431 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masona2135012008-06-25 16:01:30 -04001432
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001433 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
Chris Masonc8b97812008-10-29 14:49:59 -04001434 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001435 page_started, 1, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001436 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
Yan Zhengd899e052008-10-30 14:25:28 -04001437 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001438 page_started, 0, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001439 } else if (!btrfs_test_opt(root, COMPRESS) &&
1440 !(BTRFS_I(inode)->force_compress) &&
1441 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
Chris Mason7f366cf2009-03-12 20:12:45 -04001442 ret = cow_file_range(inode, locked_page, start, end,
1443 page_started, nr_written, 1);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001444 } else {
1445 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1446 &BTRFS_I(inode)->runtime_flags);
Chris Mason771ed682008-11-06 22:02:51 -05001447 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001448 page_started, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001449 }
Chris Masonb888db2b2007-08-27 16:49:44 -04001450 return ret;
1451}
1452
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001453static void btrfs_split_extent_hook(struct inode *inode,
1454 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001455{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001456 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001457 if (!(orig->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001458 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001459
Josef Bacik9e0baf62011-07-15 15:16:44 +00001460 spin_lock(&BTRFS_I(inode)->lock);
1461 BTRFS_I(inode)->outstanding_extents++;
1462 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001463}
1464
1465/*
1466 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1467 * extents so we can keep track of new extents that are just merged onto old
1468 * extents, such as when we are doing sequential writes, so we can properly
1469 * account for the metadata space we'll need.
1470 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001471static void btrfs_merge_extent_hook(struct inode *inode,
1472 struct extent_state *new,
1473 struct extent_state *other)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001474{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001475 /* not delalloc, ignore it */
1476 if (!(other->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001477 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001478
Josef Bacik9e0baf62011-07-15 15:16:44 +00001479 spin_lock(&BTRFS_I(inode)->lock);
1480 BTRFS_I(inode)->outstanding_extents--;
1481 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001482}
1483
Chris Masond352ac62008-09-29 15:18:18 -04001484/*
1485 * extent_io.c set_bit_hook, used to track delayed allocation
1486 * bytes in this file, and to maintain the list of inodes that
1487 * have pending delalloc work to be done.
1488 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001489static void btrfs_set_bit_hook(struct inode *inode,
1490 struct extent_state *state, int *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001491{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001492
Chris Mason75eff682008-12-15 15:54:40 -05001493 /*
1494 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001495 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001496 * bit, which is only set or cleared with irqs on
1497 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001498 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001499 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001500 u64 len = state->end + 1 - state->start;
Liu Bo83eea1f2012-07-10 05:28:39 -06001501 bool do_list = !btrfs_is_free_space_inode(inode);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001502
Josef Bacik9e0baf62011-07-15 15:16:44 +00001503 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001504 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001505 } else {
1506 spin_lock(&BTRFS_I(inode)->lock);
1507 BTRFS_I(inode)->outstanding_extents++;
1508 spin_unlock(&BTRFS_I(inode)->lock);
1509 }
Josef Bacik287a0ab2010-03-19 18:07:23 +00001510
Chris Mason75eff682008-12-15 15:54:40 -05001511 spin_lock(&root->fs_info->delalloc_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001512 BTRFS_I(inode)->delalloc_bytes += len;
1513 root->fs_info->delalloc_bytes += len;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001514 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
Chris Masonea8c2812008-08-04 23:17:27 -04001515 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1516 &root->fs_info->delalloc_inodes);
1517 }
Chris Mason75eff682008-12-15 15:54:40 -05001518 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001519 }
Chris Mason291d6732008-01-29 15:55:23 -05001520}
1521
Chris Masond352ac62008-09-29 15:18:18 -04001522/*
1523 * extent_io.c clear_bit_hook, see set_bit_hook for why
1524 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001525static void btrfs_clear_bit_hook(struct inode *inode,
1526 struct extent_state *state, int *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001527{
Chris Mason75eff682008-12-15 15:54:40 -05001528 /*
1529 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001530 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001531 * bit, which is only set or cleared with irqs on
1532 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001533 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001534 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001535 u64 len = state->end + 1 - state->start;
Liu Bo83eea1f2012-07-10 05:28:39 -06001536 bool do_list = !btrfs_is_free_space_inode(inode);
Chris Masonbcbfce82008-04-22 13:26:47 -04001537
Josef Bacik9e0baf62011-07-15 15:16:44 +00001538 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001539 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001540 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541 spin_lock(&BTRFS_I(inode)->lock);
1542 BTRFS_I(inode)->outstanding_extents--;
1543 spin_unlock(&BTRFS_I(inode)->lock);
1544 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001545
1546 if (*bits & EXTENT_DO_ACCOUNTING)
1547 btrfs_delalloc_release_metadata(inode, len);
1548
Josef Bacik0cb59c92010-07-02 12:14:14 -04001549 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1550 && do_list)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001551 btrfs_free_reserved_data_space(inode, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001552
Chris Mason75eff682008-12-15 15:54:40 -05001553 spin_lock(&root->fs_info->delalloc_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001554 root->fs_info->delalloc_bytes -= len;
1555 BTRFS_I(inode)->delalloc_bytes -= len;
1556
Josef Bacik0cb59c92010-07-02 12:14:14 -04001557 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
Chris Masonea8c2812008-08-04 23:17:27 -04001558 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1559 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1560 }
Chris Mason75eff682008-12-15 15:54:40 -05001561 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001562 }
Chris Mason291d6732008-01-29 15:55:23 -05001563}
1564
Chris Masond352ac62008-09-29 15:18:18 -04001565/*
1566 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1567 * we don't create bios that span stripes or chunks
1568 */
Chris Mason239b14b2008-03-24 15:02:07 -04001569int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001570 size_t size, struct bio *bio,
1571 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001572{
1573 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
Chris Masona62b9402008-10-03 16:31:08 -04001574 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001575 u64 length = 0;
1576 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001577 int ret;
1578
Chris Mason771ed682008-11-06 22:02:51 -05001579 if (bio_flags & EXTENT_BIO_COMPRESSED)
1580 return 0;
1581
Chris Masonf2d8d742008-04-21 10:03:05 -04001582 length = bio->bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001583 map_length = length;
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001584 ret = btrfs_map_block(root->fs_info, READ, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001585 &map_length, NULL, 0);
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001586 /* Will always return 0 with map_multi == NULL */
Jeff Mahoney3444a972011-10-03 23:23:13 -04001587 BUG_ON(ret < 0);
Chris Masond3977122009-01-05 21:25:51 -05001588 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001589 return 1;
Jeff Mahoney3444a972011-10-03 23:23:13 -04001590 return 0;
Chris Mason239b14b2008-03-24 15:02:07 -04001591}
1592
Chris Masond352ac62008-09-29 15:18:18 -04001593/*
1594 * in order to insert checksums into the metadata in large chunks,
1595 * we wait until bio submission time. All the pages in the bio are
1596 * checksummed and sums are attached onto the ordered extent record.
1597 *
1598 * At IO completion time the cums attached on the ordered extent record
1599 * are inserted into the btree
1600 */
Chris Masond3977122009-01-05 21:25:51 -05001601static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1602 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001603 unsigned long bio_flags,
1604 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001605{
Chris Mason065631f2008-02-20 12:07:25 -05001606 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001607 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001608
Chris Masond20f7042008-12-08 16:58:54 -05001609 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001610 BUG_ON(ret); /* -ENOMEM */
Chris Mason4a69a412008-11-06 22:03:00 -05001611 return 0;
1612}
Chris Masone0156402008-04-16 11:15:20 -04001613
Chris Mason4a69a412008-11-06 22:03:00 -05001614/*
1615 * in order to insert checksums into the metadata in large chunks,
1616 * we wait until bio submission time. All the pages in the bio are
1617 * checksummed and sums are attached onto the ordered extent record.
1618 *
1619 * At IO completion time the cums attached on the ordered extent record
1620 * are inserted into the btree
1621 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001622static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001623 int mirror_num, unsigned long bio_flags,
1624 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001625{
1626 struct btrfs_root *root = BTRFS_I(inode)->root;
Stefan Behrens61891922012-11-05 18:51:52 +01001627 int ret;
1628
1629 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1630 if (ret)
1631 bio_endio(bio, ret);
1632 return ret;
Chris Mason44b8bd72008-04-16 11:14:51 -04001633}
1634
Chris Masond352ac62008-09-29 15:18:18 -04001635/*
Chris Masoncad321a2008-12-17 14:51:42 -05001636 * extent_io.c submission hook. This does the right thing for csum calculation
1637 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001638 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001639static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001640 int mirror_num, unsigned long bio_flags,
1641 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001642{
1643 struct btrfs_root *root = BTRFS_I(inode)->root;
1644 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001645 int skip_sum;
Jeff Mahoney04173412011-10-03 23:23:12 -04001646 int metadata = 0;
Josef Bacikb812ce22012-11-16 13:56:32 -05001647 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
Chris Mason44b8bd72008-04-16 11:14:51 -04001648
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001649 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001650
Liu Bo83eea1f2012-07-10 05:28:39 -06001651 if (btrfs_is_free_space_inode(inode))
Jeff Mahoney04173412011-10-03 23:23:12 -04001652 metadata = 2;
1653
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001654 if (!(rw & REQ_WRITE)) {
Josef Bacik5fd02042012-05-02 14:00:54 -04001655 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1656 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001657 goto out;
Josef Bacik5fd02042012-05-02 14:00:54 -04001658
Chris Masond20f7042008-12-08 16:58:54 -05001659 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Stefan Behrens61891922012-11-05 18:51:52 +01001660 ret = btrfs_submit_compressed_read(inode, bio,
1661 mirror_num,
1662 bio_flags);
1663 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001664 } else if (!skip_sum) {
1665 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1666 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001667 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001668 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001669 goto mapit;
Josef Bacikb812ce22012-11-16 13:56:32 -05001670 } else if (async && !skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001671 /* csum items have already been cloned */
1672 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1673 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001674 /* we're doing a write, do the async checksumming */
Stefan Behrens61891922012-11-05 18:51:52 +01001675 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001676 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001677 bio_flags, bio_offset,
1678 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001679 __btrfs_submit_bio_done);
Stefan Behrens61891922012-11-05 18:51:52 +01001680 goto out;
Josef Bacikb812ce22012-11-16 13:56:32 -05001681 } else if (!skip_sum) {
1682 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1683 if (ret)
1684 goto out;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001685 }
1686
Chris Mason0b86a832008-03-24 15:01:56 -04001687mapit:
Stefan Behrens61891922012-11-05 18:51:52 +01001688 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1689
1690out:
1691 if (ret < 0)
1692 bio_endio(bio, ret);
1693 return ret;
Chris Mason065631f2008-02-20 12:07:25 -05001694}
Chris Mason6885f302008-02-20 16:11:05 -05001695
Chris Masond352ac62008-09-29 15:18:18 -04001696/*
1697 * given a list of ordered sums record them in the inode. This happens
1698 * at IO completion time based on sums calculated at bio submission time.
1699 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001700static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001701 struct inode *inode, u64 file_offset,
1702 struct list_head *list)
1703{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001704 struct btrfs_ordered_sum *sum;
1705
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001706 list_for_each_entry(sum, list, list) {
Chris Masond20f7042008-12-08 16:58:54 -05001707 btrfs_csum_file_blocks(trans,
1708 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001709 }
1710 return 0;
1711}
1712
Josef Bacik2ac55d42010-02-03 19:33:23 +00001713int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1714 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001715{
Julia Lawall6c1500f2012-11-03 20:30:18 +00001716 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
Chris Masonea8c2812008-08-04 23:17:27 -04001717 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001718 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001719}
1720
Chris Masond352ac62008-09-29 15:18:18 -04001721/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001722struct btrfs_writepage_fixup {
1723 struct page *page;
1724 struct btrfs_work work;
1725};
1726
Christoph Hellwigb2950862008-12-02 09:54:17 -05001727static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001728{
1729 struct btrfs_writepage_fixup *fixup;
1730 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001731 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001732 struct page *page;
1733 struct inode *inode;
1734 u64 page_start;
1735 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01001736 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04001737
1738 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1739 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001740again:
Chris Mason247e7432008-07-17 12:53:51 -04001741 lock_page(page);
1742 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1743 ClearPageChecked(page);
1744 goto out_page;
1745 }
1746
1747 inode = page->mapping->host;
1748 page_start = page_offset(page);
1749 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1750
Josef Bacik2ac55d42010-02-03 19:33:23 +00001751 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001752 &cached_state);
Chris Mason4a096752008-07-21 10:29:44 -04001753
1754 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001755 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001756 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001757
1758 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1759 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001760 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1761 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001762 unlock_page(page);
1763 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001764 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04001765 goto again;
1766 }
Chris Mason247e7432008-07-17 12:53:51 -04001767
Jeff Mahoney87826df2012-02-15 16:23:57 +01001768 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1769 if (ret) {
1770 mapping_set_error(page->mapping, ret);
1771 end_extent_writepage(page, ret, page_start, page_end);
1772 ClearPageChecked(page);
1773 goto out;
1774 }
1775
Josef Bacik2ac55d42010-02-03 19:33:23 +00001776 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001777 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001778 set_page_dirty(page);
Chris Mason247e7432008-07-17 12:53:51 -04001779out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001780 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1781 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001782out_page:
1783 unlock_page(page);
1784 page_cache_release(page);
Miao Xieb897abe2011-01-26 16:19:22 +08001785 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04001786}
1787
1788/*
1789 * There are a few paths in the higher layers of the kernel that directly
1790 * set the page dirty bit without asking the filesystem if it is a
1791 * good idea. This causes problems because we want to make sure COW
1792 * properly happens and the data=ordered rules are followed.
1793 *
Chris Masonc8b97812008-10-29 14:49:59 -04001794 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001795 * hasn't been properly setup for IO. We kick off an async process
1796 * to fix it up. The async helper will wait for ordered extents, set
1797 * the delalloc bit and make it safe to write the page.
1798 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001799static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001800{
1801 struct inode *inode = page->mapping->host;
1802 struct btrfs_writepage_fixup *fixup;
1803 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001804
Chris Mason8b62b722009-09-02 16:53:46 -04001805 /* this page is properly in the ordered list */
1806 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001807 return 0;
1808
1809 if (PageChecked(page))
1810 return -EAGAIN;
1811
1812 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1813 if (!fixup)
1814 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001815
Chris Mason247e7432008-07-17 12:53:51 -04001816 SetPageChecked(page);
1817 page_cache_get(page);
1818 fixup->work.func = btrfs_writepage_fixup_worker;
1819 fixup->page = page;
1820 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001821 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04001822}
1823
Yan Zhengd899e052008-10-30 14:25:28 -04001824static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1825 struct inode *inode, u64 file_pos,
1826 u64 disk_bytenr, u64 disk_num_bytes,
1827 u64 num_bytes, u64 ram_bytes,
1828 u8 compression, u8 encryption,
1829 u16 other_encoding, int extent_type)
1830{
1831 struct btrfs_root *root = BTRFS_I(inode)->root;
1832 struct btrfs_file_extent_item *fi;
1833 struct btrfs_path *path;
1834 struct extent_buffer *leaf;
1835 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04001836 int ret;
1837
1838 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001839 if (!path)
1840 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04001841
Chris Masonb9473432009-03-13 11:00:37 -04001842 path->leave_spinning = 1;
Chris Masona1ed835e12009-09-11 12:27:37 -04001843
1844 /*
1845 * we may be replacing one extent in the tree with another.
1846 * The new extent is pinned in the extent map, and we don't want
1847 * to drop it from the cache until it is completely in the btree.
1848 *
1849 * So, tell btrfs_drop_extents to leave this extent in the cache.
1850 * the caller is expected to unpin it and allow it to be merged
1851 * with the others.
1852 */
Josef Bacik5dc562c2012-08-17 13:14:17 -04001853 ret = btrfs_drop_extents(trans, root, inode, file_pos,
Josef Bacik26714852012-08-29 12:24:27 -04001854 file_pos + num_bytes, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001855 if (ret)
1856 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04001857
Li Zefan33345d012011-04-20 10:31:50 +08001858 ins.objectid = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04001859 ins.offset = file_pos;
1860 ins.type = BTRFS_EXTENT_DATA_KEY;
1861 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001862 if (ret)
1863 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04001864 leaf = path->nodes[0];
1865 fi = btrfs_item_ptr(leaf, path->slots[0],
1866 struct btrfs_file_extent_item);
1867 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1868 btrfs_set_file_extent_type(leaf, fi, extent_type);
1869 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1870 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1871 btrfs_set_file_extent_offset(leaf, fi, 0);
1872 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1873 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1874 btrfs_set_file_extent_compression(leaf, fi, compression);
1875 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1876 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001877
Yan Zhengd899e052008-10-30 14:25:28 -04001878 btrfs_mark_buffer_dirty(leaf);
Josef Bacikce195332012-09-25 15:26:16 -04001879 btrfs_release_path(path);
Yan Zhengd899e052008-10-30 14:25:28 -04001880
1881 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001882
1883 ins.objectid = disk_bytenr;
1884 ins.offset = disk_num_bytes;
1885 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001886 ret = btrfs_alloc_reserved_file_extent(trans, root,
1887 root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001888 btrfs_ino(inode), file_pos, &ins);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001889out:
Yan Zhengd899e052008-10-30 14:25:28 -04001890 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001891
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001892 return ret;
Yan Zhengd899e052008-10-30 14:25:28 -04001893}
1894
Chris Mason5d13a982009-03-13 11:41:46 -04001895/*
1896 * helper function for btrfs_finish_ordered_io, this
1897 * just reads in some of the csum leaves to prime them into ram
1898 * before we start the transaction. It limits the amount of btree
1899 * reads required while inside the transaction.
1900 */
Chris Masond352ac62008-09-29 15:18:18 -04001901/* as ordered data IO finishes, this gets called so we can finish
1902 * an ordered extent if the range of bytes in the file it covers are
1903 * fully written.
1904 */
Josef Bacik5fd02042012-05-02 14:00:54 -04001905static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001906{
Josef Bacik5fd02042012-05-02 14:00:54 -04001907 struct inode *inode = ordered_extent->inode;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001908 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001909 struct btrfs_trans_handle *trans = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001910 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001911 struct extent_state *cached_state = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08001912 int compress_type = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001913 int ret;
Li Zefan82d59022011-04-20 10:33:24 +08001914 bool nolock;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001915
Liu Bo83eea1f2012-07-10 05:28:39 -06001916 nolock = btrfs_is_free_space_inode(inode);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001917
Josef Bacik5fd02042012-05-02 14:00:54 -04001918 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1919 ret = -EIO;
1920 goto out;
1921 }
1922
Yan, Zhengc2167752009-11-12 09:34:21 +00001923 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001924 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
Josef Bacik6c760c02012-11-09 10:53:21 -05001925 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1926 if (nolock)
1927 trans = btrfs_join_transaction_nolock(root);
1928 else
1929 trans = btrfs_join_transaction(root);
1930 if (IS_ERR(trans)) {
1931 ret = PTR_ERR(trans);
1932 trans = NULL;
1933 goto out;
Yan, Zhengc2167752009-11-12 09:34:21 +00001934 }
Josef Bacik6c760c02012-11-09 10:53:21 -05001935 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1936 ret = btrfs_update_inode_fallback(trans, root, inode);
1937 if (ret) /* -ENOMEM or corruption */
1938 btrfs_abort_transaction(trans, root, ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00001939 goto out;
1940 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001941
Josef Bacik2ac55d42010-02-03 19:33:23 +00001942 lock_extent_bits(io_tree, ordered_extent->file_offset,
1943 ordered_extent->file_offset + ordered_extent->len - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001944 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001945
Josef Bacik0cb59c92010-07-02 12:14:14 -04001946 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001947 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001948 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001949 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001950 if (IS_ERR(trans)) {
1951 ret = PTR_ERR(trans);
1952 trans = NULL;
1953 goto out_unlock;
1954 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001955 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00001956
Chris Masonc8b97812008-10-29 14:49:59 -04001957 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08001958 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04001959 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08001960 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001961 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04001962 ordered_extent->file_offset,
1963 ordered_extent->file_offset +
1964 ordered_extent->len);
Yan Zhengd899e052008-10-30 14:25:28 -04001965 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04001966 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04001967 ret = insert_reserved_file_extent(trans, inode,
1968 ordered_extent->file_offset,
1969 ordered_extent->start,
1970 ordered_extent->disk_len,
1971 ordered_extent->len,
1972 ordered_extent->len,
Li Zefan261507a02010-12-17 14:21:50 +08001973 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04001974 BTRFS_FILE_EXTENT_REG);
Yan Zhengd899e052008-10-30 14:25:28 -04001975 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04001976 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1977 ordered_extent->file_offset, ordered_extent->len,
1978 trans->transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001979 if (ret < 0) {
1980 btrfs_abort_transaction(trans, root, ret);
Josef Bacik5fd02042012-05-02 14:00:54 -04001981 goto out_unlock;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001982 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00001983
Chris Masone6dcd2d2008-07-17 12:53:50 -04001984 add_pending_csums(trans, inode, ordered_extent->file_offset,
1985 &ordered_extent->list);
1986
Josef Bacik6c760c02012-11-09 10:53:21 -05001987 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1988 ret = btrfs_update_inode_fallback(trans, root, inode);
1989 if (ret) { /* -ENOMEM or corruption */
1990 btrfs_abort_transaction(trans, root, ret);
1991 goto out_unlock;
Josef Bacik1ef30be2011-04-05 19:25:36 -04001992 }
1993 ret = 0;
Josef Bacik5fd02042012-05-02 14:00:54 -04001994out_unlock:
1995 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1996 ordered_extent->file_offset +
1997 ordered_extent->len - 1, &cached_state, GFP_NOFS);
Yan, Zhengc2167752009-11-12 09:34:21 +00001998out:
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001999 if (root != root->fs_info->tree_root)
Josef Bacik0cb59c92010-07-02 12:14:14 -04002000 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Miao Xiea698d0752012-09-20 01:51:59 -06002001 if (trans)
2002 btrfs_end_transaction(trans, root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002003
Josef Bacik5fd02042012-05-02 14:00:54 -04002004 if (ret)
2005 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2006 ordered_extent->file_offset +
2007 ordered_extent->len - 1, NULL, GFP_NOFS);
2008
2009 /*
Liu Bo8bad3c02012-06-18 12:14:23 +08002010 * This needs to be done to make sure anybody waiting knows we are done
2011 * updating everything for this ordered extent.
Josef Bacik5fd02042012-05-02 14:00:54 -04002012 */
2013 btrfs_remove_ordered_extent(inode, ordered_extent);
2014
Chris Masone6dcd2d2008-07-17 12:53:50 -04002015 /* once for us */
2016 btrfs_put_ordered_extent(ordered_extent);
2017 /* once for the tree */
2018 btrfs_put_ordered_extent(ordered_extent);
2019
Josef Bacik5fd02042012-05-02 14:00:54 -04002020 return ret;
2021}
2022
2023static void finish_ordered_fn(struct btrfs_work *work)
2024{
2025 struct btrfs_ordered_extent *ordered_extent;
2026 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2027 btrfs_finish_ordered_io(ordered_extent);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002028}
2029
Christoph Hellwigb2950862008-12-02 09:54:17 -05002030static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04002031 struct extent_state *state, int uptodate)
2032{
Josef Bacik5fd02042012-05-02 14:00:54 -04002033 struct inode *inode = page->mapping->host;
2034 struct btrfs_root *root = BTRFS_I(inode)->root;
2035 struct btrfs_ordered_extent *ordered_extent = NULL;
2036 struct btrfs_workers *workers;
2037
liubo1abe9b82011-03-24 11:18:59 +00002038 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2039
Chris Mason8b62b722009-09-02 16:53:46 -04002040 ClearPagePrivate2(page);
Josef Bacik5fd02042012-05-02 14:00:54 -04002041 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2042 end - start + 1, uptodate))
2043 return 0;
2044
2045 ordered_extent->work.func = finish_ordered_fn;
2046 ordered_extent->work.flags = 0;
2047
Liu Bo83eea1f2012-07-10 05:28:39 -06002048 if (btrfs_is_free_space_inode(inode))
Josef Bacik5fd02042012-05-02 14:00:54 -04002049 workers = &root->fs_info->endio_freespace_worker;
2050 else
2051 workers = &root->fs_info->endio_write_workers;
2052 btrfs_queue_worker(workers, &ordered_extent->work);
2053
2054 return 0;
Chris Mason211f90e2008-07-18 11:56:15 -04002055}
2056
Chris Masond352ac62008-09-29 15:18:18 -04002057/*
Chris Masond352ac62008-09-29 15:18:18 -04002058 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02002059 * if there's a match, we allow the bio to finish. If not, the code in
2060 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04002061 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05002062static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
Josef Bacik5cf1ab52012-04-16 09:42:26 -04002063 struct extent_state *state, int mirror)
Chris Mason07157aa2007-08-30 08:50:51 -04002064{
Chris Mason35ebb932007-10-30 16:56:53 -04002065 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
Chris Mason07157aa2007-08-30 08:50:51 -04002066 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05002067 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04002068 char *kaddr;
Chris Masonaadfeb62008-01-29 09:10:27 -05002069 u64 private = ~(u32)0;
Chris Mason07157aa2007-08-30 08:50:51 -04002070 int ret;
Chris Masonff79f812007-10-15 16:22:25 -04002071 struct btrfs_root *root = BTRFS_I(inode)->root;
2072 u32 csum = ~(u32)0;
Chris Masond1310b22008-01-24 16:13:08 -05002073
Chris Masond20f7042008-12-08 16:58:54 -05002074 if (PageChecked(page)) {
2075 ClearPageChecked(page);
2076 goto good;
2077 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002078
2079 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Jan Schmidt08d2f342011-05-04 16:18:50 +02002080 goto good;
Chris Masond20f7042008-12-08 16:58:54 -05002081
Yan Zheng17d217f2008-12-12 10:03:38 -05002082 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04002083 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05002084 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2085 GFP_NOFS);
2086 return 0;
2087 }
2088
Yanc2e639f2008-02-04 08:57:25 -05002089 if (state && state->start == start) {
Chris Mason70dec802008-01-29 09:59:12 -05002090 private = state->private;
2091 ret = 0;
2092 } else {
2093 ret = get_state_private(io_tree, start, &private);
2094 }
Cong Wang7ac687d2011-11-25 23:14:28 +08002095 kaddr = kmap_atomic(page);
Chris Masond3977122009-01-05 21:25:51 -05002096 if (ret)
Chris Mason07157aa2007-08-30 08:50:51 -04002097 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05002098
Chris Masonff79f812007-10-15 16:22:25 -04002099 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2100 btrfs_csum_final(csum, (char *)&csum);
Chris Masond3977122009-01-05 21:25:51 -05002101 if (csum != private)
Chris Mason07157aa2007-08-30 08:50:51 -04002102 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05002103
Cong Wang7ac687d2011-11-25 23:14:28 +08002104 kunmap_atomic(kaddr);
Chris Masond20f7042008-12-08 16:58:54 -05002105good:
Chris Mason07157aa2007-08-30 08:50:51 -04002106 return 0;
2107
2108zeroit:
Chris Mason945d8962011-05-22 12:33:42 -04002109 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
Li Zefan33345d012011-04-20 10:31:50 +08002110 "private %llu\n",
2111 (unsigned long long)btrfs_ino(page->mapping->host),
Chris Mason193f2842009-04-27 07:29:05 -04002112 (unsigned long long)start, csum,
2113 (unsigned long long)private);
Chris Masondb945352007-10-15 16:15:53 -04002114 memset(kaddr + offset, 1, end - start + 1);
2115 flush_dcache_page(page);
Cong Wang7ac687d2011-11-25 23:14:28 +08002116 kunmap_atomic(kaddr);
Chris Mason3b951512008-04-17 11:29:12 -04002117 if (private == 0)
2118 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04002119 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04002120}
Chris Masonb888db2b2007-08-27 16:49:44 -04002121
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002122struct delayed_iput {
2123 struct list_head list;
2124 struct inode *inode;
2125};
2126
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002127/* JDM: If this is fs-wide, why can't we add a pointer to
2128 * btrfs_inode instead and avoid the allocation? */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002129void btrfs_add_delayed_iput(struct inode *inode)
2130{
2131 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2132 struct delayed_iput *delayed;
2133
2134 if (atomic_add_unless(&inode->i_count, -1, 1))
2135 return;
2136
2137 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2138 delayed->inode = inode;
2139
2140 spin_lock(&fs_info->delayed_iput_lock);
2141 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2142 spin_unlock(&fs_info->delayed_iput_lock);
2143}
2144
2145void btrfs_run_delayed_iputs(struct btrfs_root *root)
2146{
2147 LIST_HEAD(list);
2148 struct btrfs_fs_info *fs_info = root->fs_info;
2149 struct delayed_iput *delayed;
2150 int empty;
2151
2152 spin_lock(&fs_info->delayed_iput_lock);
2153 empty = list_empty(&fs_info->delayed_iputs);
2154 spin_unlock(&fs_info->delayed_iput_lock);
2155 if (empty)
2156 return;
2157
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002158 spin_lock(&fs_info->delayed_iput_lock);
2159 list_splice_init(&fs_info->delayed_iputs, &list);
2160 spin_unlock(&fs_info->delayed_iput_lock);
2161
2162 while (!list_empty(&list)) {
2163 delayed = list_entry(list.next, struct delayed_iput, list);
2164 list_del(&delayed->list);
2165 iput(delayed->inode);
2166 kfree(delayed);
2167 }
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002168}
2169
Yan, Zhengd68fc572010-05-16 10:49:58 -04002170enum btrfs_orphan_cleanup_state {
2171 ORPHAN_CLEANUP_STARTED = 1,
2172 ORPHAN_CLEANUP_DONE = 2,
2173};
2174
2175/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002176 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04002177 * files in the subvolume, it removes orphan item and frees block_rsv
2178 * structure.
2179 */
2180void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *root)
2182{
Josef Bacik90290e12011-12-02 15:44:12 -05002183 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002184 int ret;
2185
Josef Bacik8a35d952012-05-23 14:26:42 -04002186 if (atomic_read(&root->orphan_inodes) ||
Yan, Zhengd68fc572010-05-16 10:49:58 -04002187 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2188 return;
2189
Josef Bacik90290e12011-12-02 15:44:12 -05002190 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04002191 if (atomic_read(&root->orphan_inodes)) {
Josef Bacik90290e12011-12-02 15:44:12 -05002192 spin_unlock(&root->orphan_lock);
2193 return;
2194 }
2195
2196 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2197 spin_unlock(&root->orphan_lock);
2198 return;
2199 }
2200
2201 block_rsv = root->orphan_block_rsv;
2202 root->orphan_block_rsv = NULL;
2203 spin_unlock(&root->orphan_lock);
2204
Yan, Zhengd68fc572010-05-16 10:49:58 -04002205 if (root->orphan_item_inserted &&
2206 btrfs_root_refs(&root->root_item) > 0) {
2207 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2208 root->root_key.objectid);
2209 BUG_ON(ret);
2210 root->orphan_item_inserted = 0;
2211 }
2212
Josef Bacik90290e12011-12-02 15:44:12 -05002213 if (block_rsv) {
2214 WARN_ON(block_rsv->size > 0);
2215 btrfs_free_block_rsv(root, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002216 }
2217}
2218
2219/*
Josef Bacik7b128762008-07-24 12:17:14 -04002220 * This creates an orphan entry for the given inode in case something goes
2221 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04002222 *
2223 * NOTE: caller of this function should reserve 5 units of metadata for
2224 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04002225 */
2226int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2227{
2228 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002229 struct btrfs_block_rsv *block_rsv = NULL;
2230 int reserve = 0;
2231 int insert = 0;
2232 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002233
Yan, Zhengd68fc572010-05-16 10:49:58 -04002234 if (!root->orphan_block_rsv) {
Miao Xie66d8f3d2012-09-06 04:02:28 -06002235 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002236 if (!block_rsv)
2237 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04002238 }
2239
Yan, Zhengd68fc572010-05-16 10:49:58 -04002240 spin_lock(&root->orphan_lock);
2241 if (!root->orphan_block_rsv) {
2242 root->orphan_block_rsv = block_rsv;
2243 } else if (block_rsv) {
2244 btrfs_free_block_rsv(root, block_rsv);
2245 block_rsv = NULL;
2246 }
Josef Bacik7b128762008-07-24 12:17:14 -04002247
Josef Bacik8a35d952012-05-23 14:26:42 -04002248 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249 &BTRFS_I(inode)->runtime_flags)) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04002250#if 0
2251 /*
2252 * For proper ENOSPC handling, we should do orphan
2253 * cleanup when mounting. But this introduces backward
2254 * compatibility issue.
2255 */
2256 if (!xchg(&root->orphan_item_inserted, 1))
2257 insert = 2;
2258 else
2259 insert = 1;
2260#endif
2261 insert = 1;
Miao Xie321f0e72012-08-28 22:13:02 -06002262 atomic_inc(&root->orphan_inodes);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002263 }
Josef Bacik7b128762008-07-24 12:17:14 -04002264
Josef Bacik72ac3c02012-05-23 14:13:11 -04002265 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2266 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04002267 reserve = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002268 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002269
Yan, Zhengd68fc572010-05-16 10:49:58 -04002270 /* grab metadata reservation from transaction handle */
2271 if (reserve) {
2272 ret = btrfs_orphan_reserve_metadata(trans, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002273 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
Yan, Zhengd68fc572010-05-16 10:49:58 -04002274 }
2275
2276 /* insert an orphan item to track this unlinked/truncated file */
2277 if (insert >= 1) {
Li Zefan33345d012011-04-20 10:31:50 +08002278 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002279 if (ret && ret != -EEXIST) {
Josef Bacik8a35d952012-05-23 14:26:42 -04002280 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281 &BTRFS_I(inode)->runtime_flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002282 btrfs_abort_transaction(trans, root, ret);
2283 return ret;
2284 }
2285 ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002286 }
2287
2288 /* insert an orphan item to track subvolume contains orphan files */
2289 if (insert >= 2) {
2290 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2291 root->root_key.objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002292 if (ret && ret != -EEXIST) {
2293 btrfs_abort_transaction(trans, root, ret);
2294 return ret;
2295 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04002296 }
2297 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002298}
2299
2300/*
2301 * We have done the truncate/delete so we can go ahead and remove the orphan
2302 * item for this particular inode.
2303 */
2304int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2305{
2306 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002307 int delete_item = 0;
2308 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002309 int ret = 0;
2310
Yan, Zhengd68fc572010-05-16 10:49:58 -04002311 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04002312 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2313 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04002314 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04002315
Josef Bacik72ac3c02012-05-23 14:13:11 -04002316 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2317 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04002318 release_rsv = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002319 spin_unlock(&root->orphan_lock);
2320
2321 if (trans && delete_item) {
Li Zefan33345d012011-04-20 10:31:50 +08002322 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002323 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
Josef Bacik7b128762008-07-24 12:17:14 -04002324 }
2325
Josef Bacik8a35d952012-05-23 14:26:42 -04002326 if (release_rsv) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04002327 btrfs_orphan_release_metadata(inode);
Josef Bacik8a35d952012-05-23 14:26:42 -04002328 atomic_dec(&root->orphan_inodes);
2329 }
Josef Bacik7b128762008-07-24 12:17:14 -04002330
Yan, Zhengd68fc572010-05-16 10:49:58 -04002331 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002332}
2333
2334/*
2335 * this cleans up any orphans that may be left on the list from the last use
2336 * of this root.
2337 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002338int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04002339{
2340 struct btrfs_path *path;
2341 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04002342 struct btrfs_key key, found_key;
2343 struct btrfs_trans_handle *trans;
2344 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04002345 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002346 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2347
Yan, Zhengd68fc572010-05-16 10:49:58 -04002348 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002349 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00002350
2351 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002352 if (!path) {
2353 ret = -ENOMEM;
2354 goto out;
2355 }
Josef Bacik7b128762008-07-24 12:17:14 -04002356 path->reada = -1;
2357
2358 key.objectid = BTRFS_ORPHAN_OBJECTID;
2359 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2360 key.offset = (u64)-1;
2361
Josef Bacik7b128762008-07-24 12:17:14 -04002362 while (1) {
2363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002364 if (ret < 0)
2365 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002366
2367 /*
2368 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002369 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04002370 * find the key and see if we have stuff that matches
2371 */
2372 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002373 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002374 if (path->slots[0] == 0)
2375 break;
2376 path->slots[0]--;
2377 }
2378
2379 /* pull out the item */
2380 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04002381 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2382
2383 /* make sure the item matches what we want */
2384 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2385 break;
2386 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2387 break;
2388
2389 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02002390 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04002391
2392 /*
2393 * this is where we are basically btrfs_lookup, without the
2394 * crossing root thing. we store the inode number in the
2395 * offset of the orphan item.
2396 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04002397
2398 if (found_key.offset == last_objectid) {
2399 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2400 "stopping orphan cleanup\n");
2401 ret = -EINVAL;
2402 goto out;
2403 }
2404
2405 last_objectid = found_key.offset;
2406
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002407 found_key.objectid = found_key.offset;
2408 found_key.type = BTRFS_INODE_ITEM_KEY;
2409 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00002410 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Josef Bacika8c9e572011-09-21 16:55:59 -04002411 ret = PTR_RET(inode);
2412 if (ret && ret != -ESTALE)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002413 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04002414
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05002415 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2416 struct btrfs_root *dead_root;
2417 struct btrfs_fs_info *fs_info = root->fs_info;
2418 int is_dead_root = 0;
2419
2420 /*
2421 * this is an orphan in the tree root. Currently these
2422 * could come from 2 sources:
2423 * a) a snapshot deletion in progress
2424 * b) a free space cache inode
2425 * We need to distinguish those two, as the snapshot
2426 * orphan must not get deleted.
2427 * find_dead_roots already ran before us, so if this
2428 * is a snapshot deletion, we should find the root
2429 * in the dead_roots list
2430 */
2431 spin_lock(&fs_info->trans_lock);
2432 list_for_each_entry(dead_root, &fs_info->dead_roots,
2433 root_list) {
2434 if (dead_root->root_key.objectid ==
2435 found_key.objectid) {
2436 is_dead_root = 1;
2437 break;
2438 }
2439 }
2440 spin_unlock(&fs_info->trans_lock);
2441 if (is_dead_root) {
2442 /* prevent this orphan from being found again */
2443 key.offset = found_key.objectid - 1;
2444 continue;
2445 }
2446 }
Josef Bacika8c9e572011-09-21 16:55:59 -04002447 /*
2448 * Inode is already gone but the orphan item is still there,
2449 * kill the orphan item.
2450 */
2451 if (ret == -ESTALE) {
2452 trans = btrfs_start_transaction(root, 1);
2453 if (IS_ERR(trans)) {
2454 ret = PTR_ERR(trans);
2455 goto out;
2456 }
Josef Bacik8a35d952012-05-23 14:26:42 -04002457 printk(KERN_ERR "auto deleting %Lu\n",
2458 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04002459 ret = btrfs_del_orphan_item(trans, root,
2460 found_key.objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002461 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
Josef Bacika8c9e572011-09-21 16:55:59 -04002462 btrfs_end_transaction(trans, root);
2463 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002464 }
Josef Bacik7b128762008-07-24 12:17:14 -04002465
Josef Bacik7b128762008-07-24 12:17:14 -04002466 /*
2467 * add this inode to the orphan list so btrfs_orphan_del does
2468 * the proper thing when we hit it
2469 */
Josef Bacik8a35d952012-05-23 14:26:42 -04002470 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2471 &BTRFS_I(inode)->runtime_flags);
Josef Bacik7b128762008-07-24 12:17:14 -04002472
Josef Bacik7b128762008-07-24 12:17:14 -04002473 /* if we have links, this was a truncate, lets do that */
2474 if (inode->i_nlink) {
Josef Bacika41ad392011-01-31 15:30:16 -05002475 if (!S_ISREG(inode->i_mode)) {
2476 WARN_ON(1);
2477 iput(inode);
2478 continue;
2479 }
Josef Bacik7b128762008-07-24 12:17:14 -04002480 nr_truncate++;
Josef Bacikf3fe8202013-01-07 17:03:21 -05002481
2482 /* 1 for the orphan item deletion. */
2483 trans = btrfs_start_transaction(root, 1);
2484 if (IS_ERR(trans)) {
2485 ret = PTR_ERR(trans);
2486 goto out;
2487 }
2488 ret = btrfs_orphan_add(trans, inode);
2489 btrfs_end_transaction(trans, root);
2490 if (ret)
2491 goto out;
2492
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002493 ret = btrfs_truncate(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04002494 } else {
2495 nr_unlink++;
2496 }
2497
2498 /* this will do delete_inode and everything for us */
2499 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002500 if (ret)
2501 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002502 }
Miao Xie3254c872011-11-10 20:45:05 -05002503 /* release the path since we're done with it */
2504 btrfs_release_path(path);
2505
Yan, Zhengd68fc572010-05-16 10:49:58 -04002506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2507
2508 if (root->orphan_block_rsv)
2509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2510 (u64)-1);
2511
2512 if (root->orphan_block_rsv || root->orphan_item_inserted) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002513 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002514 if (!IS_ERR(trans))
2515 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002516 }
Josef Bacik7b128762008-07-24 12:17:14 -04002517
2518 if (nr_unlink)
2519 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2520 if (nr_truncate)
2521 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002522
2523out:
2524 if (ret)
2525 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2526 btrfs_free_path(path);
2527 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002528}
2529
Chris Masond352ac62008-09-29 15:18:18 -04002530/*
Chris Mason46a53cc2009-04-27 11:47:50 -04002531 * very simple check to peek ahead in the leaf looking for xattrs. If we
2532 * don't find any xattrs, we know there can't be any acls.
2533 *
2534 * slot is the slot the inode is in, objectid is the objectid of the inode
2535 */
2536static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2537 int slot, u64 objectid)
2538{
2539 u32 nritems = btrfs_header_nritems(leaf);
2540 struct btrfs_key found_key;
2541 int scanned = 0;
2542
2543 slot++;
2544 while (slot < nritems) {
2545 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2546
2547 /* we found a different objectid, there must not be acls */
2548 if (found_key.objectid != objectid)
2549 return 0;
2550
2551 /* we found an xattr, assume we've got an acl */
2552 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2553 return 1;
2554
2555 /*
2556 * we found a key greater than an xattr key, there can't
2557 * be any acls later on
2558 */
2559 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2560 return 0;
2561
2562 slot++;
2563 scanned++;
2564
2565 /*
2566 * it goes inode, inode backrefs, xattrs, extents,
2567 * so if there are a ton of hard links to an inode there can
2568 * be a lot of backrefs. Don't waste time searching too hard,
2569 * this is just an optimization
2570 */
2571 if (scanned >= 8)
2572 break;
2573 }
2574 /* we hit the end of the leaf before we found an xattr or
2575 * something larger than an xattr. We have to assume the inode
2576 * has acls
2577 */
2578 return 1;
2579}
2580
2581/*
Chris Masond352ac62008-09-29 15:18:18 -04002582 * read an inode from the btree into the in-memory inode
2583 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002584static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002585{
2586 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002587 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002588 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04002589 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04002590 struct btrfs_root *root = BTRFS_I(inode)->root;
2591 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04002592 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04002593 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04002594 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00002595 bool filled = false;
2596
2597 ret = btrfs_fill_inode(inode, &rdev);
2598 if (!ret)
2599 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04002600
2601 path = btrfs_alloc_path();
Mark Fasheh1748f842011-07-12 11:25:31 -07002602 if (!path)
2603 goto make_bad;
2604
Josef Bacikd90c7322011-05-17 09:50:54 -04002605 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002606 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05002607
Chris Mason39279cc2007-06-12 06:35:45 -04002608 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002609 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002610 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04002611
Chris Mason5f39d392007-10-15 16:14:19 -04002612 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00002613
2614 if (filled)
2615 goto cache_acl;
2616
Chris Mason5f39d392007-10-15 16:14:19 -04002617 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2618 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002619 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002620 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Eric W. Biederman2f2f43d32012-02-10 11:05:07 -08002621 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2622 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
Chris Masondbe674a2008-07-17 12:54:05 -04002623 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002624
2625 tspec = btrfs_inode_atime(inode_item);
2626 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2627 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2628
2629 tspec = btrfs_inode_mtime(inode_item);
2630 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2631 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2632
2633 tspec = btrfs_inode_ctime(inode_item);
2634 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2635 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2636
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002637 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04002638 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Josef Bacik5dc562c2012-08-17 13:14:17 -04002639 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2640
2641 /*
2642 * If we were modified in the current generation and evicted from memory
2643 * and then re-read we need to do a full sync since we don't have any
2644 * idea about which extents were modified before we were evicted from
2645 * cache.
2646 */
2647 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2648 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2649 &BTRFS_I(inode)->runtime_flags);
2650
Josef Bacik0c4d2d92012-04-05 15:03:02 -04002651 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04002652 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04002653 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002654 rdev = btrfs_inode_rdev(leaf, inode_item);
2655
Josef Bacikaec74772008-07-24 12:12:38 -04002656 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002657 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie2f7e33d2011-06-23 07:27:13 +00002658cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04002659 /*
2660 * try to precache a NULL acl entry for files that don't have
2661 * any xattrs or acls
2662 */
Li Zefan33345d012011-04-20 10:31:50 +08002663 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2664 btrfs_ino(inode));
Al Viro72c04902009-06-24 16:58:48 -04002665 if (!maybe_acls)
2666 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04002667
Chris Mason39279cc2007-06-12 06:35:45 -04002668 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002669
Chris Mason39279cc2007-06-12 06:35:45 -04002670 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04002671 case S_IFREG:
2672 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04002673 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05002674 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04002675 inode->i_fop = &btrfs_file_operations;
2676 inode->i_op = &btrfs_file_inode_operations;
2677 break;
2678 case S_IFDIR:
2679 inode->i_fop = &btrfs_dir_file_operations;
2680 if (root == root->fs_info->tree_root)
2681 inode->i_op = &btrfs_dir_ro_inode_operations;
2682 else
2683 inode->i_op = &btrfs_dir_inode_operations;
2684 break;
2685 case S_IFLNK:
2686 inode->i_op = &btrfs_symlink_inode_operations;
2687 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04002688 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04002689 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04002690 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05002691 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04002692 init_special_inode(inode, inode->i_mode, rdev);
2693 break;
Chris Mason39279cc2007-06-12 06:35:45 -04002694 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002695
2696 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002697 return;
2698
2699make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04002700 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002701 make_bad_inode(inode);
2702}
2703
Chris Masond352ac62008-09-29 15:18:18 -04002704/*
2705 * given a leaf and an inode, copy the inode fields into the leaf
2706 */
Chris Masone02119d2008-09-05 16:13:11 -04002707static void fill_inode_item(struct btrfs_trans_handle *trans,
2708 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04002709 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04002710 struct inode *inode)
2711{
Eric W. Biederman2f2f43d32012-02-10 11:05:07 -08002712 btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2713 btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
Chris Masondbe674a2008-07-17 12:54:05 -04002714 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
Chris Mason5f39d392007-10-15 16:14:19 -04002715 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2716 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2717
2718 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2719 inode->i_atime.tv_sec);
2720 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2721 inode->i_atime.tv_nsec);
2722
2723 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2724 inode->i_mtime.tv_sec);
2725 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2726 inode->i_mtime.tv_nsec);
2727
2728 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2729 inode->i_ctime.tv_sec);
2730 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2731 inode->i_ctime.tv_nsec);
2732
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002733 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
Chris Masone02119d2008-09-05 16:13:11 -04002734 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04002735 btrfs_set_inode_sequence(leaf, item, inode->i_version);
Chris Masone02119d2008-09-05 16:13:11 -04002736 btrfs_set_inode_transid(leaf, item, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002737 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Yanb98b6762008-01-08 15:54:37 -05002738 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04002739 btrfs_set_inode_block_group(leaf, item, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002740}
2741
Chris Masond352ac62008-09-29 15:18:18 -04002742/*
2743 * copy everything in the in-memory inode into the btree.
2744 */
Chris Mason21151332011-11-10 20:39:08 -05002745static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05002746 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002747{
2748 struct btrfs_inode_item *inode_item;
2749 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002750 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002751 int ret;
2752
2753 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08002754 if (!path)
2755 return -ENOMEM;
2756
Chris Masonb9473432009-03-13 11:00:37 -04002757 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08002758 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2759 1);
Chris Mason39279cc2007-06-12 06:35:45 -04002760 if (ret) {
2761 if (ret > 0)
2762 ret = -ENOENT;
2763 goto failed;
2764 }
2765
Chris Masonb4ce94d2009-02-04 09:25:08 -05002766 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002767 leaf = path->nodes[0];
2768 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08002769 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04002770
Chris Masone02119d2008-09-05 16:13:11 -04002771 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04002772 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002773 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002774 ret = 0;
2775failed:
Chris Mason39279cc2007-06-12 06:35:45 -04002776 btrfs_free_path(path);
2777 return ret;
2778}
2779
Chris Masond352ac62008-09-29 15:18:18 -04002780/*
Chris Mason21151332011-11-10 20:39:08 -05002781 * copy everything in the in-memory inode into the btree.
2782 */
2783noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root, struct inode *inode)
2785{
2786 int ret;
2787
2788 /*
2789 * If the inode is a free space inode, we can deadlock during commit
2790 * if we put it into the delayed code.
2791 *
2792 * The data relocation inode should also be directly updated
2793 * without delay
2794 */
Liu Bo83eea1f2012-07-10 05:28:39 -06002795 if (!btrfs_is_free_space_inode(inode)
Chris Mason21151332011-11-10 20:39:08 -05002796 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
Alexander Block8ea05e32012-07-25 17:35:53 +02002797 btrfs_update_root_times(trans, root);
2798
Chris Mason21151332011-11-10 20:39:08 -05002799 ret = btrfs_delayed_update_inode(trans, root, inode);
2800 if (!ret)
2801 btrfs_set_inode_last_trans(trans, inode);
2802 return ret;
2803 }
2804
2805 return btrfs_update_inode_item(trans, root, inode);
2806}
2807
Josef Bacikbe6aef62012-10-22 15:43:12 -04002808noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2809 struct btrfs_root *root,
2810 struct inode *inode)
Chris Mason21151332011-11-10 20:39:08 -05002811{
2812 int ret;
2813
2814 ret = btrfs_update_inode(trans, root, inode);
2815 if (ret == -ENOSPC)
2816 return btrfs_update_inode_item(trans, root, inode);
2817 return ret;
2818}
2819
2820/*
Chris Masond352ac62008-09-29 15:18:18 -04002821 * unlink helper that gets used here in inode.c and in the tree logging
2822 * recovery code. It remove a link in a directory with a given name, and
2823 * also drops the back refs in the inode to the directory
2824 */
Al Viro92986792011-03-04 17:14:37 +00002825static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2826 struct btrfs_root *root,
2827 struct inode *dir, struct inode *inode,
2828 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04002829{
2830 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04002831 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002832 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002833 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04002834 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04002835 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08002836 u64 ino = btrfs_ino(inode);
2837 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04002838
2839 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002840 if (!path) {
2841 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00002842 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04002843 }
2844
Chris Masonb9473432009-03-13 11:00:37 -04002845 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08002846 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04002847 name, name_len, -1);
2848 if (IS_ERR(di)) {
2849 ret = PTR_ERR(di);
2850 goto err;
2851 }
2852 if (!di) {
2853 ret = -ENOENT;
2854 goto err;
2855 }
Chris Mason5f39d392007-10-15 16:14:19 -04002856 leaf = path->nodes[0];
2857 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04002858 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04002859 if (ret)
2860 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02002861 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002862
Li Zefan33345d012011-04-20 10:31:50 +08002863 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2864 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04002865 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05002866 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
Li Zefan33345d012011-04-20 10:31:50 +08002867 "inode %llu parent %llu\n", name_len, name,
2868 (unsigned long long)ino, (unsigned long long)dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002869 btrfs_abort_transaction(trans, root, ret);
Josef Bacikaec74772008-07-24 12:12:38 -04002870 goto err;
2871 }
2872
Miao Xie16cdcec2011-04-22 18:12:22 +08002873 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002874 if (ret) {
2875 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04002876 goto err;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002877 }
Chris Mason39279cc2007-06-12 06:35:45 -04002878
Chris Masone02119d2008-09-05 16:13:11 -04002879 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
Li Zefan33345d012011-04-20 10:31:50 +08002880 inode, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002881 if (ret != 0 && ret != -ENOENT) {
2882 btrfs_abort_transaction(trans, root, ret);
2883 goto err;
2884 }
Chris Masone02119d2008-09-05 16:13:11 -04002885
2886 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2887 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04002888 if (ret == -ENOENT)
2889 ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002890err:
2891 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002892 if (ret)
2893 goto out;
2894
2895 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04002896 inode_inc_iversion(inode);
2897 inode_inc_iversion(dir);
Chris Masone02119d2008-09-05 16:13:11 -04002898 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
Tsutomu Itohb9959292012-06-25 21:25:22 -06002899 ret = btrfs_update_inode(trans, root, dir);
Chris Masone02119d2008-09-05 16:13:11 -04002900out:
Chris Mason39279cc2007-06-12 06:35:45 -04002901 return ret;
2902}
2903
Al Viro92986792011-03-04 17:14:37 +00002904int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2905 struct btrfs_root *root,
2906 struct inode *dir, struct inode *inode,
2907 const char *name, int name_len)
2908{
2909 int ret;
2910 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2911 if (!ret) {
2912 btrfs_drop_nlink(inode);
2913 ret = btrfs_update_inode(trans, root, inode);
2914 }
2915 return ret;
2916}
2917
2918
Yan, Zhenga22285a2010-05-16 10:48:46 -04002919/* helper to check if there is any shared block in the path */
2920static int check_path_shared(struct btrfs_root *root,
2921 struct btrfs_path *path)
2922{
2923 struct extent_buffer *eb;
2924 int level;
Dan Carpenter0e4dcbef2010-06-01 08:23:11 +00002925 u64 refs = 1;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002926
2927 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
Josef Bacikdedefd72011-01-24 21:43:18 +00002928 int ret;
2929
Yan, Zhenga22285a2010-05-16 10:48:46 -04002930 if (!path->nodes[level])
2931 break;
2932 eb = path->nodes[level];
2933 if (!btrfs_block_can_be_shared(root, eb))
2934 continue;
2935 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2936 &refs, NULL);
2937 if (refs > 1)
2938 return 1;
2939 }
Josef Bacikdedefd72011-01-24 21:43:18 +00002940 return 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002941}
2942
2943/*
2944 * helper to start transaction for unlink and rmdir.
2945 *
2946 * unlink and rmdir are special in btrfs, they do not always free space.
2947 * so in enospc case, we should make sure they will free space before
2948 * allowing them to use the global metadata reservation.
2949 */
2950static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2951 struct dentry *dentry)
2952{
2953 struct btrfs_trans_handle *trans;
2954 struct btrfs_root *root = BTRFS_I(dir)->root;
2955 struct btrfs_path *path;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002956 struct btrfs_dir_item *di;
2957 struct inode *inode = dentry->d_inode;
2958 u64 index;
2959 int check_link = 1;
2960 int err = -ENOSPC;
2961 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08002962 u64 ino = btrfs_ino(inode);
2963 u64 dir_ino = btrfs_ino(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002964
Josef Bacike70bea52011-10-11 14:18:24 -04002965 /*
2966 * 1 for the possible orphan item
2967 * 1 for the dir item
2968 * 1 for the dir index
2969 * 1 for the inode ref
2970 * 1 for the inode ref in the tree log
2971 * 2 for the dir entries in the log
2972 * 1 for the inode
2973 */
2974 trans = btrfs_start_transaction(root, 8);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002975 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2976 return trans;
2977
Li Zefan33345d012011-04-20 10:31:50 +08002978 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhenga22285a2010-05-16 10:48:46 -04002979 return ERR_PTR(-ENOSPC);
2980
2981 /* check if there is someone else holds reference */
2982 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2983 return ERR_PTR(-ENOSPC);
2984
2985 if (atomic_read(&inode->i_count) > 2)
2986 return ERR_PTR(-ENOSPC);
2987
2988 if (xchg(&root->fs_info->enospc_unlink, 1))
2989 return ERR_PTR(-ENOSPC);
2990
2991 path = btrfs_alloc_path();
2992 if (!path) {
2993 root->fs_info->enospc_unlink = 0;
2994 return ERR_PTR(-ENOMEM);
2995 }
2996
Josef Bacik3880a1b2011-10-14 14:46:51 -04002997 /* 1 for the orphan item */
2998 trans = btrfs_start_transaction(root, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002999 if (IS_ERR(trans)) {
3000 btrfs_free_path(path);
3001 root->fs_info->enospc_unlink = 0;
3002 return trans;
3003 }
3004
3005 path->skip_locking = 1;
3006 path->search_commit_root = 1;
3007
3008 ret = btrfs_lookup_inode(trans, root, path,
3009 &BTRFS_I(dir)->location, 0);
3010 if (ret < 0) {
3011 err = ret;
3012 goto out;
3013 }
3014 if (ret == 0) {
3015 if (check_path_shared(root, path))
3016 goto out;
3017 } else {
3018 check_link = 0;
3019 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003020 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003021
3022 ret = btrfs_lookup_inode(trans, root, path,
3023 &BTRFS_I(inode)->location, 0);
3024 if (ret < 0) {
3025 err = ret;
3026 goto out;
3027 }
3028 if (ret == 0) {
3029 if (check_path_shared(root, path))
3030 goto out;
3031 } else {
3032 check_link = 0;
3033 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003034 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003035
3036 if (ret == 0 && S_ISREG(inode->i_mode)) {
3037 ret = btrfs_lookup_file_extent(trans, root, path,
Li Zefan33345d012011-04-20 10:31:50 +08003038 ino, (u64)-1, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003039 if (ret < 0) {
3040 err = ret;
3041 goto out;
3042 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003043 BUG_ON(ret == 0); /* Corruption */
Yan, Zhenga22285a2010-05-16 10:48:46 -04003044 if (check_path_shared(root, path))
3045 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +02003046 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003047 }
3048
3049 if (!check_link) {
3050 err = 0;
3051 goto out;
3052 }
3053
Li Zefan33345d012011-04-20 10:31:50 +08003054 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zhenga22285a2010-05-16 10:48:46 -04003055 dentry->d_name.name, dentry->d_name.len, 0);
3056 if (IS_ERR(di)) {
3057 err = PTR_ERR(di);
3058 goto out;
3059 }
3060 if (di) {
3061 if (check_path_shared(root, path))
3062 goto out;
3063 } else {
3064 err = 0;
3065 goto out;
3066 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003067 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003068
Mark Fashehf1863732012-08-08 11:32:27 -07003069 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3070 dentry->d_name.len, ino, dir_ino, 0,
3071 &index);
3072 if (ret) {
3073 err = ret;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003074 goto out;
3075 }
Mark Fashehf1863732012-08-08 11:32:27 -07003076
Yan, Zhenga22285a2010-05-16 10:48:46 -04003077 if (check_path_shared(root, path))
3078 goto out;
Mark Fashehf1863732012-08-08 11:32:27 -07003079
David Sterbab3b4aa72011-04-21 01:20:15 +02003080 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003081
Miao Xie16cdcec2011-04-22 18:12:22 +08003082 /*
3083 * This is a commit root search, if we can lookup inode item and other
3084 * relative items in the commit root, it means the transaction of
3085 * dir/file creation has been committed, and the dir index item that we
3086 * delay to insert has also been inserted into the commit root. So
3087 * we needn't worry about the delayed insertion of the dir index item
3088 * here.
3089 */
Li Zefan33345d012011-04-20 10:31:50 +08003090 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04003091 dentry->d_name.name, dentry->d_name.len, 0);
3092 if (IS_ERR(di)) {
3093 err = PTR_ERR(di);
3094 goto out;
3095 }
3096 BUG_ON(ret == -ENOENT);
3097 if (check_path_shared(root, path))
3098 goto out;
3099
3100 err = 0;
3101out:
3102 btrfs_free_path(path);
Josef Bacik3880a1b2011-10-14 14:46:51 -04003103 /* Migrate the orphan reservation over */
3104 if (!err)
3105 err = btrfs_block_rsv_migrate(trans->block_rsv,
3106 &root->fs_info->global_block_rsv,
Josef Bacik5a77d762011-11-01 14:32:23 -04003107 trans->bytes_reserved);
Josef Bacik3880a1b2011-10-14 14:46:51 -04003108
Yan, Zhenga22285a2010-05-16 10:48:46 -04003109 if (err) {
3110 btrfs_end_transaction(trans, root);
3111 root->fs_info->enospc_unlink = 0;
3112 return ERR_PTR(err);
3113 }
3114
3115 trans->block_rsv = &root->fs_info->global_block_rsv;
3116 return trans;
3117}
3118
3119static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3120 struct btrfs_root *root)
3121{
Miao Xie66d8f3d2012-09-06 04:02:28 -06003122 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
Josef Bacik5a77d762011-11-01 14:32:23 -04003123 btrfs_block_rsv_release(root, trans->block_rsv,
3124 trans->bytes_reserved);
3125 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003126 BUG_ON(!root->fs_info->enospc_unlink);
3127 root->fs_info->enospc_unlink = 0;
3128 }
Josef Bacik7ad85bb2012-01-12 19:10:12 -05003129 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003130}
3131
Chris Mason39279cc2007-06-12 06:35:45 -04003132static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3133{
Yan, Zhenga22285a2010-05-16 10:48:46 -04003134 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003135 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04003136 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003137 int ret;
3138
Yan, Zhenga22285a2010-05-16 10:48:46 -04003139 trans = __unlink_start_trans(dir, dentry);
3140 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003141 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04003142
Chris Mason12fcfd22009-03-24 10:24:20 -04003143 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3144
Chris Masone02119d2008-09-05 16:13:11 -04003145 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3146 dentry->d_name.name, dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003147 if (ret)
3148 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003149
Yan, Zhenga22285a2010-05-16 10:48:46 -04003150 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04003151 ret = btrfs_orphan_add(trans, inode);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003152 if (ret)
3153 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003154 }
Josef Bacik7b128762008-07-24 12:17:14 -04003155
Tsutomu Itohb5324022011-07-19 07:27:20 +00003156out:
Yan, Zhenga22285a2010-05-16 10:48:46 -04003157 __unlink_end_trans(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003158 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04003159 return ret;
3160}
3161
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003162int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root,
3164 struct inode *dir, u64 objectid,
3165 const char *name, int name_len)
3166{
3167 struct btrfs_path *path;
3168 struct extent_buffer *leaf;
3169 struct btrfs_dir_item *di;
3170 struct btrfs_key key;
3171 u64 index;
3172 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08003173 u64 dir_ino = btrfs_ino(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003174
3175 path = btrfs_alloc_path();
3176 if (!path)
3177 return -ENOMEM;
3178
Li Zefan33345d012011-04-20 10:31:50 +08003179 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003180 name, name_len, -1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003181 if (IS_ERR_OR_NULL(di)) {
3182 if (!di)
3183 ret = -ENOENT;
3184 else
3185 ret = PTR_ERR(di);
3186 goto out;
3187 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003188
3189 leaf = path->nodes[0];
3190 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3191 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3192 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003193 if (ret) {
3194 btrfs_abort_transaction(trans, root, ret);
3195 goto out;
3196 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003197 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003198
3199 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3200 objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08003201 dir_ino, &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003202 if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003203 if (ret != -ENOENT) {
3204 btrfs_abort_transaction(trans, root, ret);
3205 goto out;
3206 }
Li Zefan33345d012011-04-20 10:31:50 +08003207 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003208 name, name_len);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003209 if (IS_ERR_OR_NULL(di)) {
3210 if (!di)
3211 ret = -ENOENT;
3212 else
3213 ret = PTR_ERR(di);
3214 btrfs_abort_transaction(trans, root, ret);
3215 goto out;
3216 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003217
3218 leaf = path->nodes[0];
3219 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003220 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003221 index = key.offset;
3222 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003223 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003224
Miao Xie16cdcec2011-04-22 18:12:22 +08003225 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003226 if (ret) {
3227 btrfs_abort_transaction(trans, root, ret);
3228 goto out;
3229 }
Miao Xie16cdcec2011-04-22 18:12:22 +08003230
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003231 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003232 inode_inc_iversion(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003233 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
Josef Bacik5a24e84c2012-08-08 10:12:59 -06003234 ret = btrfs_update_inode_fallback(trans, root, dir);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003235 if (ret)
3236 btrfs_abort_transaction(trans, root, ret);
3237out:
Josef Bacik71d7aed2011-06-14 14:24:32 -04003238 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003239 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003240}
3241
Chris Mason39279cc2007-06-12 06:35:45 -04003242static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3243{
3244 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05003245 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003246 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003247 struct btrfs_trans_handle *trans;
Chris Mason39279cc2007-06-12 06:35:45 -04003248
David Sterbab3ae2442012-09-13 16:04:34 -06003249 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Yan134d4512007-10-25 15:49:25 -04003250 return -ENOTEMPTY;
David Sterbab3ae2442012-09-13 16:04:34 -06003251 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3252 return -EPERM;
Yan134d4512007-10-25 15:49:25 -04003253
Yan, Zhenga22285a2010-05-16 10:48:46 -04003254 trans = __unlink_start_trans(dir, dentry);
3255 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003256 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003257
Li Zefan33345d012011-04-20 10:31:50 +08003258 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003259 err = btrfs_unlink_subvol(trans, root, dir,
3260 BTRFS_I(inode)->location.objectid,
3261 dentry->d_name.name,
3262 dentry->d_name.len);
3263 goto out;
3264 }
3265
Josef Bacik7b128762008-07-24 12:17:14 -04003266 err = btrfs_orphan_add(trans, inode);
3267 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003268 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003269
Chris Mason39279cc2007-06-12 06:35:45 -04003270 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04003271 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3272 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05003273 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04003274 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003275out:
Yan, Zhenga22285a2010-05-16 10:48:46 -04003276 __unlink_end_trans(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003277 btrfs_btree_balance_dirty(root);
Chris Mason39544012007-12-12 14:38:19 -05003278
Chris Mason39279cc2007-06-12 06:35:45 -04003279 return err;
3280}
3281
Chris Mason323ac952008-10-01 19:05:46 -04003282/*
Chris Mason39279cc2007-06-12 06:35:45 -04003283 * this can truncate away extent items, csum items and directory items.
3284 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04003285 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04003286 *
3287 * csum items that cross the new i_size are truncated to the new size
3288 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04003289 *
3290 * min_type is the minimum key type to truncate down to. If set to 0, this
3291 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04003292 */
Yan, Zheng80825102009-11-12 09:35:36 +00003293int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3294 struct btrfs_root *root,
3295 struct inode *inode,
3296 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003297{
Chris Mason39279cc2007-06-12 06:35:45 -04003298 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003299 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003300 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00003301 struct btrfs_key key;
3302 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003303 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04003304 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003305 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003306 u64 item_end = 0;
Yan, Zheng80825102009-11-12 09:35:36 +00003307 u64 mask = root->sectorsize - 1;
3308 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04003309 int found_extent;
3310 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05003311 int pending_del_nr = 0;
3312 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04003313 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00003314 int ret;
3315 int err = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003316 u64 ino = btrfs_ino(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003317
3318 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04003319
Mark Fasheh0eb0e192011-07-12 16:44:10 -07003320 path = btrfs_alloc_path();
3321 if (!path)
3322 return -ENOMEM;
3323 path->reada = -1;
3324
Josef Bacik5dc562c2012-08-17 13:14:17 -04003325 /*
3326 * We want to drop from the next block forward in case this new size is
3327 * not block aligned since we will be keeping the last block of the
3328 * extent just the way it is.
3329 */
Josef Bacik0af3d002010-06-21 14:48:16 -04003330 if (root->ref_cows || root == root->fs_info->tree_root)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003331 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003332
Miao Xie16cdcec2011-04-22 18:12:22 +08003333 /*
3334 * This function is also used to drop the items in the log tree before
3335 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3336 * it is used to drop the loged items. So we shouldn't kill the delayed
3337 * items.
3338 */
3339 if (min_type == 0 && root == BTRFS_I(inode)->root)
3340 btrfs_kill_delayed_inode_items(inode);
3341
Li Zefan33345d012011-04-20 10:31:50 +08003342 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04003343 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04003344 key.type = (u8)-1;
3345
Chris Mason85e21ba2008-01-29 15:11:36 -05003346search_again:
Chris Masonb9473432009-03-13 11:00:37 -04003347 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05003348 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00003349 if (ret < 0) {
3350 err = ret;
3351 goto out;
3352 }
Chris Masond3977122009-01-05 21:25:51 -05003353
Chris Mason85e21ba2008-01-29 15:11:36 -05003354 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003355 /* there are no items in the tree for us to truncate, we're
3356 * done
3357 */
Yan, Zheng80825102009-11-12 09:35:36 +00003358 if (path->slots[0] == 0)
3359 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05003360 path->slots[0]--;
3361 }
3362
Chris Masond3977122009-01-05 21:25:51 -05003363 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003364 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04003365 leaf = path->nodes[0];
3366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3367 found_type = btrfs_key_type(&found_key);
Chris Mason39279cc2007-06-12 06:35:45 -04003368
Li Zefan33345d012011-04-20 10:31:50 +08003369 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04003370 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003371
Chris Mason85e21ba2008-01-29 15:11:36 -05003372 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003373 break;
3374
Chris Mason5f39d392007-10-15 16:14:19 -04003375 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04003376 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04003377 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04003378 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04003379 extent_type = btrfs_file_extent_type(leaf, fi);
3380 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04003381 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04003382 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04003383 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04003384 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04003385 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003386 }
Yan008630c2007-11-07 13:31:09 -05003387 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04003388 }
Yan, Zheng80825102009-11-12 09:35:36 +00003389 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04003390 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003391 } else {
3392 if (item_end < new_size)
3393 break;
3394 if (found_key.offset >= new_size)
3395 del_item = 1;
3396 else
3397 del_item = 0;
3398 }
Chris Mason39279cc2007-06-12 06:35:45 -04003399 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003400 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04003401 if (found_type != BTRFS_EXTENT_DATA_KEY)
3402 goto delete;
3403
3404 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04003405 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04003406 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a6b2012-01-12 19:10:12 -05003407 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04003408 u64 orig_num_bytes =
3409 btrfs_file_extent_num_bytes(leaf, fi);
Chris Masone02119d2008-09-05 16:13:11 -04003410 extent_num_bytes = new_size -
Chris Mason5f39d392007-10-15 16:14:19 -04003411 found_key.offset + root->sectorsize - 1;
Yanb1632b12008-01-30 11:54:04 -05003412 extent_num_bytes = extent_num_bytes &
3413 ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -04003414 btrfs_set_file_extent_num_bytes(leaf, fi,
3415 extent_num_bytes);
3416 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05003417 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04003418 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003419 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04003420 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003421 } else {
Chris Masondb945352007-10-15 16:15:53 -04003422 extent_num_bytes =
3423 btrfs_file_extent_disk_num_bytes(leaf,
3424 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003425 extent_offset = found_key.offset -
3426 btrfs_file_extent_offset(leaf, fi);
3427
Chris Mason39279cc2007-06-12 06:35:45 -04003428 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05003429 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003430 if (extent_start != 0) {
3431 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04003432 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003433 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04003434 }
3435 }
Chris Mason90692182008-02-08 13:49:28 -05003436 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04003437 /*
3438 * we can't truncate inline items that have had
3439 * special encodings
3440 */
3441 if (!del_item &&
3442 btrfs_file_extent_compression(leaf, fi) == 0 &&
3443 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3444 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003445 u32 size = new_size - found_key.offset;
3446
3447 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003448 inode_sub_bytes(inode, item_end + 1 -
3449 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04003450 }
3451 size =
3452 btrfs_file_extent_calc_inline_size(size);
Jeff Mahoney143bede2012-03-01 14:56:26 +01003453 btrfs_truncate_item(trans, root, path,
3454 size, 1);
Chris Masone02119d2008-09-05 16:13:11 -04003455 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003456 inode_sub_bytes(inode, item_end + 1 -
3457 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05003458 }
Chris Mason39279cc2007-06-12 06:35:45 -04003459 }
Chris Mason179e29e2007-11-01 11:28:41 -04003460delete:
Chris Mason39279cc2007-06-12 06:35:45 -04003461 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05003462 if (!pending_del_nr) {
3463 /* no pending yet, add ourselves */
3464 pending_del_slot = path->slots[0];
3465 pending_del_nr = 1;
3466 } else if (pending_del_nr &&
3467 path->slots[0] + 1 == pending_del_slot) {
3468 /* hop on the pending chunk */
3469 pending_del_nr++;
3470 pending_del_slot = path->slots[0];
3471 } else {
Chris Masond3977122009-01-05 21:25:51 -05003472 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05003473 }
Chris Mason39279cc2007-06-12 06:35:45 -04003474 } else {
3475 break;
3476 }
Josef Bacik0af3d002010-06-21 14:48:16 -04003477 if (found_extent && (root->ref_cows ||
3478 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04003479 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003480 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003481 extent_num_bytes, 0,
3482 btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003483 ino, extent_offset, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04003484 BUG_ON(ret);
3485 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003486
Yan, Zheng80825102009-11-12 09:35:36 +00003487 if (found_type == BTRFS_INODE_ITEM_KEY)
3488 break;
3489
3490 if (path->slots[0] == 0 ||
3491 path->slots[0] != pending_del_slot) {
Yan, Zheng80825102009-11-12 09:35:36 +00003492 if (pending_del_nr) {
3493 ret = btrfs_del_items(trans, root, path,
3494 pending_del_slot,
3495 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003496 if (ret) {
3497 btrfs_abort_transaction(trans,
3498 root, ret);
3499 goto error;
3500 }
Yan, Zheng80825102009-11-12 09:35:36 +00003501 pending_del_nr = 0;
3502 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003503 btrfs_release_path(path);
Chris Mason85e21ba2008-01-29 15:11:36 -05003504 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00003505 } else {
3506 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05003507 }
Chris Mason39279cc2007-06-12 06:35:45 -04003508 }
Yan, Zheng80825102009-11-12 09:35:36 +00003509out:
Chris Mason85e21ba2008-01-29 15:11:36 -05003510 if (pending_del_nr) {
3511 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3512 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003513 if (ret)
3514 btrfs_abort_transaction(trans, root, ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05003515 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003516error:
Chris Mason39279cc2007-06-12 06:35:45 -04003517 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00003518 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003519}
3520
Chris Masona52d9a82007-08-27 16:49:44 -04003521/*
Josef Bacik2aaa6652012-08-29 14:27:18 -04003522 * btrfs_truncate_page - read, zero a chunk and write a page
3523 * @inode - inode that we're zeroing
3524 * @from - the offset to start zeroing
3525 * @len - the length to zero, 0 to zero the entire range respective to the
3526 * offset
3527 * @front - zero up to the offset instead of from the offset on
3528 *
3529 * This will find the page for the "from" offset and cow the page and zero the
3530 * part we want to zero. This is used with truncate and hole punching.
Chris Masona52d9a82007-08-27 16:49:44 -04003531 */
Josef Bacik2aaa6652012-08-29 14:27:18 -04003532int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3533 int front)
Chris Masona52d9a82007-08-27 16:49:44 -04003534{
Josef Bacik2aaa6652012-08-29 14:27:18 -04003535 struct address_space *mapping = inode->i_mapping;
Chris Masondb945352007-10-15 16:15:53 -04003536 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003537 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3538 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003539 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003540 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04003541 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04003542 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3543 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3544 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003545 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04003546 int ret = 0;
3547 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003548 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04003549
Josef Bacik2aaa6652012-08-29 14:27:18 -04003550 if ((offset & (blocksize - 1)) == 0 &&
3551 (!len || ((len & (blocksize - 1)) == 0)))
Chris Masona52d9a82007-08-27 16:49:44 -04003552 goto out;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003553 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003554 if (ret)
3555 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04003556
Chris Mason211c17f2008-05-15 09:13:45 -04003557again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003558 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003559 if (!page) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003560 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Miao Xieac6a2b32012-12-05 10:56:13 +00003561 ret = -ENOMEM;
Chris Masona52d9a82007-08-27 16:49:44 -04003562 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04003563 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04003564
3565 page_start = page_offset(page);
3566 page_end = page_start + PAGE_CACHE_SIZE - 1;
3567
Chris Masona52d9a82007-08-27 16:49:44 -04003568 if (!PageUptodate(page)) {
3569 ret = btrfs_readpage(NULL, page);
3570 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04003571 if (page->mapping != mapping) {
3572 unlock_page(page);
3573 page_cache_release(page);
3574 goto again;
3575 }
Chris Masona52d9a82007-08-27 16:49:44 -04003576 if (!PageUptodate(page)) {
3577 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04003578 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04003579 }
3580 }
Chris Mason211c17f2008-05-15 09:13:45 -04003581 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003582
Jeff Mahoneyd0082372012-03-01 14:57:19 +01003583 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003584 set_page_extent_mapped(page);
3585
3586 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3587 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003588 unlock_extent_cached(io_tree, page_start, page_end,
3589 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003590 unlock_page(page);
3591 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04003592 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003593 btrfs_put_ordered_extent(ordered);
3594 goto again;
3595 }
3596
Josef Bacik2ac55d42010-02-03 19:33:23 +00003597 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06003598 EXTENT_DIRTY | EXTENT_DELALLOC |
3599 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00003600 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003601
Josef Bacik2ac55d42010-02-03 19:33:23 +00003602 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3603 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003604 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003605 unlock_extent_cached(io_tree, page_start, page_end,
3606 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003607 goto out_unlock;
3608 }
3609
Chris Masone6dcd2d2008-07-17 12:53:50 -04003610 if (offset != PAGE_CACHE_SIZE) {
Josef Bacik2aaa6652012-08-29 14:27:18 -04003611 if (!len)
3612 len = PAGE_CACHE_SIZE - offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003613 kaddr = kmap(page);
Josef Bacik2aaa6652012-08-29 14:27:18 -04003614 if (front)
3615 memset(kaddr, 0, offset);
3616 else
3617 memset(kaddr + offset, 0, len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003618 flush_dcache_page(page);
3619 kunmap(page);
3620 }
Chris Mason247e7432008-07-17 12:53:51 -04003621 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003622 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003623 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3624 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04003625
Chris Mason89642222008-07-24 09:41:53 -04003626out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04003627 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003628 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason39279cc2007-06-12 06:35:45 -04003629 unlock_page(page);
3630 page_cache_release(page);
3631out:
3632 return ret;
3633}
3634
Josef Bacik695a0d02011-03-04 15:46:53 -05003635/*
3636 * This function puts in dummy file extents for the area we're creating a hole
3637 * for. So if we are truncating this file to a larger size we need to insert
3638 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3639 * the range between oldsize and size
3640 */
Josef Bacika41ad392011-01-31 15:30:16 -05003641int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04003642{
3643 struct btrfs_trans_handle *trans;
3644 struct btrfs_root *root = BTRFS_I(inode)->root;
3645 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003646 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003647 struct extent_state *cached_state = NULL;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003648 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Yan Zheng9036c102008-10-30 14:19:41 -04003649 u64 mask = root->sectorsize - 1;
Josef Bacika41ad392011-01-31 15:30:16 -05003650 u64 hole_start = (oldsize + mask) & ~mask;
Yan Zheng9036c102008-10-30 14:19:41 -04003651 u64 block_end = (size + mask) & ~mask;
3652 u64 last_byte;
3653 u64 cur_offset;
3654 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003655 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003656
3657 if (size <= hole_start)
3658 return 0;
3659
Yan Zheng9036c102008-10-30 14:19:41 -04003660 while (1) {
3661 struct btrfs_ordered_extent *ordered;
3662 btrfs_wait_ordered_range(inode, hole_start,
3663 block_end - hole_start);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003664 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01003665 &cached_state);
Yan Zheng9036c102008-10-30 14:19:41 -04003666 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3667 if (!ordered)
3668 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003669 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3670 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003671 btrfs_put_ordered_extent(ordered);
3672 }
3673
Yan Zheng9036c102008-10-30 14:19:41 -04003674 cur_offset = hole_start;
3675 while (1) {
3676 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3677 block_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003678 if (IS_ERR(em)) {
3679 err = PTR_ERR(em);
Zach Brownf2767952013-01-08 19:37:58 +00003680 em = NULL;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003681 break;
3682 }
Yan Zheng9036c102008-10-30 14:19:41 -04003683 last_byte = min(extent_map_end(em), block_end);
3684 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng80825102009-11-12 09:35:36 +00003685 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Josef Bacik5dc562c2012-08-17 13:14:17 -04003686 struct extent_map *hole_em;
Yan Zheng9036c102008-10-30 14:19:41 -04003687 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00003688
Miao Xie36423202011-12-14 20:12:02 -05003689 trans = btrfs_start_transaction(root, 3);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003690 if (IS_ERR(trans)) {
3691 err = PTR_ERR(trans);
Chris Mason771ed682008-11-06 22:02:51 -05003692 break;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003693 }
Yan, Zheng80825102009-11-12 09:35:36 +00003694
Josef Bacik5dc562c2012-08-17 13:14:17 -04003695 err = btrfs_drop_extents(trans, root, inode,
3696 cur_offset,
Josef Bacik26714852012-08-29 12:24:27 -04003697 cur_offset + hole_size, 1);
Miao Xie5b397372011-09-11 10:52:24 -04003698 if (err) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003699 btrfs_abort_transaction(trans, root, err);
Miao Xie5b397372011-09-11 10:52:24 -04003700 btrfs_end_transaction(trans, root);
Josef Bacik3893e332011-01-31 16:03:11 -05003701 break;
Miao Xie5b397372011-09-11 10:52:24 -04003702 }
Josef Bacik9ed74f22009-09-11 16:12:44 -04003703
Yan Zheng9036c102008-10-30 14:19:41 -04003704 err = btrfs_insert_file_extent(trans, root,
Li Zefan33345d012011-04-20 10:31:50 +08003705 btrfs_ino(inode), cur_offset, 0,
Yan Zheng9036c102008-10-30 14:19:41 -04003706 0, hole_size, 0, hole_size,
3707 0, 0, 0);
Miao Xie5b397372011-09-11 10:52:24 -04003708 if (err) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003709 btrfs_abort_transaction(trans, root, err);
Miao Xie5b397372011-09-11 10:52:24 -04003710 btrfs_end_transaction(trans, root);
Josef Bacik3893e332011-01-31 16:03:11 -05003711 break;
Miao Xie5b397372011-09-11 10:52:24 -04003712 }
Yan, Zheng80825102009-11-12 09:35:36 +00003713
Josef Bacik5dc562c2012-08-17 13:14:17 -04003714 btrfs_drop_extent_cache(inode, cur_offset,
3715 cur_offset + hole_size - 1, 0);
3716 hole_em = alloc_extent_map();
3717 if (!hole_em) {
3718 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3719 &BTRFS_I(inode)->runtime_flags);
3720 goto next;
3721 }
3722 hole_em->start = cur_offset;
3723 hole_em->len = hole_size;
3724 hole_em->orig_start = cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00003725
Josef Bacik5dc562c2012-08-17 13:14:17 -04003726 hole_em->block_start = EXTENT_MAP_HOLE;
3727 hole_em->block_len = 0;
Josef Bacikb4939682012-12-03 10:31:19 -05003728 hole_em->orig_block_len = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003729 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3730 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3731 hole_em->generation = trans->transid;
3732
3733 while (1) {
3734 write_lock(&em_tree->lock);
3735 err = add_extent_mapping(em_tree, hole_em);
3736 if (!err)
3737 list_move(&hole_em->list,
3738 &em_tree->modified_extents);
3739 write_unlock(&em_tree->lock);
3740 if (err != -EEXIST)
3741 break;
3742 btrfs_drop_extent_cache(inode, cur_offset,
3743 cur_offset +
3744 hole_size - 1, 0);
3745 }
3746 free_extent_map(hole_em);
3747next:
Miao Xie36423202011-12-14 20:12:02 -05003748 btrfs_update_inode(trans, root, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003749 btrfs_end_transaction(trans, root);
Yan Zheng9036c102008-10-30 14:19:41 -04003750 }
3751 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003752 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003753 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00003754 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04003755 break;
3756 }
3757
Yan, Zhenga22285a2010-05-16 10:48:46 -04003758 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003759 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3760 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003761 return err;
3762}
3763
Eric Sandeen3972f262013-01-12 02:57:22 +00003764static int btrfs_setsize(struct inode *inode, struct iattr *attr)
Yan, Zheng80825102009-11-12 09:35:36 +00003765{
Miao Xief4a2f4c2011-12-14 20:12:01 -05003766 struct btrfs_root *root = BTRFS_I(inode)->root;
3767 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05003768 loff_t oldsize = i_size_read(inode);
Eric Sandeen3972f262013-01-12 02:57:22 +00003769 loff_t newsize = attr->ia_size;
3770 int mask = attr->ia_valid;
Yan, Zheng80825102009-11-12 09:35:36 +00003771 int ret;
3772
Josef Bacika41ad392011-01-31 15:30:16 -05003773 if (newsize == oldsize)
Yan, Zheng80825102009-11-12 09:35:36 +00003774 return 0;
3775
Eric Sandeen3972f262013-01-12 02:57:22 +00003776 /*
3777 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
3778 * special case where we need to update the times despite not having
3779 * these flags set. For all other operations the VFS set these flags
3780 * explicitly if it wants a timestamp update.
3781 */
3782 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
3783 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
3784
Josef Bacika41ad392011-01-31 15:30:16 -05003785 if (newsize > oldsize) {
Josef Bacika41ad392011-01-31 15:30:16 -05003786 truncate_pagecache(inode, oldsize, newsize);
3787 ret = btrfs_cont_expand(inode, oldsize, newsize);
Miao Xief4a2f4c2011-12-14 20:12:01 -05003788 if (ret)
Yan, Zheng80825102009-11-12 09:35:36 +00003789 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00003790
Miao Xief4a2f4c2011-12-14 20:12:01 -05003791 trans = btrfs_start_transaction(root, 1);
3792 if (IS_ERR(trans))
3793 return PTR_ERR(trans);
3794
3795 i_size_write(inode, newsize);
3796 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3797 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05003798 btrfs_end_transaction(trans, root);
Josef Bacika41ad392011-01-31 15:30:16 -05003799 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00003800
Josef Bacika41ad392011-01-31 15:30:16 -05003801 /*
3802 * We're truncating a file that used to have good data down to
3803 * zero. Make sure it gets into the ordered flush list so that
3804 * any new writes get down to disk quickly.
3805 */
3806 if (newsize == 0)
Josef Bacik72ac3c02012-05-23 14:13:11 -04003807 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3808 &BTRFS_I(inode)->runtime_flags);
Yan, Zheng80825102009-11-12 09:35:36 +00003809
Josef Bacikf3fe8202013-01-07 17:03:21 -05003810 /*
3811 * 1 for the orphan item we're going to add
3812 * 1 for the orphan item deletion.
3813 */
3814 trans = btrfs_start_transaction(root, 2);
3815 if (IS_ERR(trans))
3816 return PTR_ERR(trans);
3817
3818 /*
3819 * We need to do this in case we fail at _any_ point during the
3820 * actual truncate. Once we do the truncate_setsize we could
3821 * invalidate pages which forces any outstanding ordered io to
3822 * be instantly completed which will give us extents that need
3823 * to be truncated. If we fail to get an orphan inode down we
3824 * could have left over extents that were never meant to live,
3825 * so we need to garuntee from this point on that everything
3826 * will be consistent.
3827 */
3828 ret = btrfs_orphan_add(trans, inode);
3829 btrfs_end_transaction(trans, root);
3830 if (ret)
3831 return ret;
3832
Josef Bacika41ad392011-01-31 15:30:16 -05003833 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3834 truncate_setsize(inode, newsize);
3835 ret = btrfs_truncate(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003836 if (ret && inode->i_nlink)
3837 btrfs_orphan_del(NULL, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003838 }
3839
Josef Bacika41ad392011-01-31 15:30:16 -05003840 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00003841}
3842
Chris Mason39279cc2007-06-12 06:35:45 -04003843static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3844{
3845 struct inode *inode = dentry->d_inode;
Li Zefanb83cc962010-12-20 16:04:08 +08003846 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003847 int err;
3848
Li Zefanb83cc962010-12-20 16:04:08 +08003849 if (btrfs_root_readonly(root))
3850 return -EROFS;
3851
Chris Mason39279cc2007-06-12 06:35:45 -04003852 err = inode_change_ok(inode, attr);
3853 if (err)
3854 return err;
3855
Chris Mason5a3f23d2009-03-31 13:27:11 -04003856 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Eric Sandeen3972f262013-01-12 02:57:22 +00003857 err = btrfs_setsize(inode, attr);
Yan, Zheng80825102009-11-12 09:35:36 +00003858 if (err)
3859 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003860 }
Yan Zheng9036c102008-10-30 14:19:41 -04003861
Christoph Hellwig10257742010-06-04 11:30:02 +02003862 if (attr->ia_valid) {
3863 setattr_copy(inode, attr);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003864 inode_inc_iversion(inode);
Josef Bacik22c44fe2011-11-30 10:45:38 -05003865 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04003866
Josef Bacik22c44fe2011-11-30 10:45:38 -05003867 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig10257742010-06-04 11:30:02 +02003868 err = btrfs_acl_chmod(inode);
3869 }
3870
Chris Mason39279cc2007-06-12 06:35:45 -04003871 return err;
3872}
Chris Mason61295eb2008-01-14 16:24:38 -05003873
Al Virobd555972010-06-07 11:35:40 -04003874void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003875{
3876 struct btrfs_trans_handle *trans;
3877 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04003878 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik07127182011-08-19 10:29:59 -04003879 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003880 int ret;
3881
liubo1abe9b82011-03-24 11:18:59 +00003882 trace_btrfs_inode_evict(inode);
3883
Chris Mason39279cc2007-06-12 06:35:45 -04003884 truncate_inode_pages(&inode->i_data, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -04003885 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
Liu Bo83eea1f2012-07-10 05:28:39 -06003886 btrfs_is_free_space_inode(inode)))
Al Virobd555972010-06-07 11:35:40 -04003887 goto no_delete;
3888
Chris Mason39279cc2007-06-12 06:35:45 -04003889 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04003890 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003891 goto no_delete;
3892 }
Al Virobd555972010-06-07 11:35:40 -04003893 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Chris Mason4a096752008-07-21 10:29:44 -04003894 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04003895
Yan, Zhengc71bf092009-11-12 09:34:40 +00003896 if (root->fs_info->log_root_recovering) {
Liu Bo6bf02312012-06-25 21:59:09 -06003897 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Josef Bacik8a35d952012-05-23 14:26:42 -04003898 &BTRFS_I(inode)->runtime_flags));
Yan, Zhengc71bf092009-11-12 09:34:40 +00003899 goto no_delete;
3900 }
3901
Yan, Zheng76dda932009-09-21 16:00:26 -04003902 if (inode->i_nlink > 0) {
3903 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3904 goto no_delete;
3905 }
3906
Miao Xie66d8f3d2012-09-06 04:02:28 -06003907 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacik4289a662011-08-05 13:22:24 -04003908 if (!rsv) {
3909 btrfs_orphan_del(NULL, inode);
3910 goto no_delete;
3911 }
Josef Bacik4a338542011-08-29 11:01:31 -04003912 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04003913 rsv->failfast = 1;
Josef Bacik726c35f2011-09-26 15:46:06 -04003914 global_rsv = &root->fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04003915
Chris Masondbe674a2008-07-17 12:54:05 -04003916 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003917
Josef Bacik4289a662011-08-05 13:22:24 -04003918 /*
Miao Xie8407aa42012-09-07 01:43:32 -06003919 * This is a bit simpler than btrfs_truncate since we've already
3920 * reserved our space for our orphan item in the unlink, so we just
3921 * need to reserve some slack space in case we add bytes and update
3922 * inode item when doing the truncate.
Josef Bacik4289a662011-08-05 13:22:24 -04003923 */
Yan, Zheng80825102009-11-12 09:35:36 +00003924 while (1) {
Miao Xie08e007d2012-10-16 11:33:38 +00003925 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3926 BTRFS_RESERVE_FLUSH_LIMIT);
Yan, Zheng80825102009-11-12 09:35:36 +00003927
Josef Bacik726c35f2011-09-26 15:46:06 -04003928 /*
3929 * Try and steal from the global reserve since we will
3930 * likely not use this space anyway, we want to try as
3931 * hard as possible to get this to work.
3932 */
3933 if (ret)
3934 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3935
Yan, Zhengd68fc572010-05-16 10:49:58 -04003936 if (ret) {
Josef Bacik4289a662011-08-05 13:22:24 -04003937 printk(KERN_WARNING "Could not get space for a "
Josef Bacik482e6dc2011-08-19 10:31:56 -04003938 "delete, will truncate on mount %d\n", ret);
Josef Bacik4289a662011-08-05 13:22:24 -04003939 btrfs_orphan_del(NULL, inode);
3940 btrfs_free_block_rsv(root, rsv);
3941 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003942 }
3943
Miao Xie08e007d2012-10-16 11:33:38 +00003944 trans = btrfs_start_transaction_lflush(root, 1);
Josef Bacik4289a662011-08-05 13:22:24 -04003945 if (IS_ERR(trans)) {
3946 btrfs_orphan_del(NULL, inode);
3947 btrfs_free_block_rsv(root, rsv);
3948 goto no_delete;
3949 }
3950
3951 trans->block_rsv = rsv;
3952
Yan, Zhengd68fc572010-05-16 10:49:58 -04003953 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Josef Bacikca7e70f2012-08-27 17:48:15 -04003954 if (ret != -ENOSPC)
Yan, Zheng80825102009-11-12 09:35:36 +00003955 break;
3956
Miao Xie8407aa42012-09-07 01:43:32 -06003957 trans->block_rsv = &root->fs_info->trans_block_rsv;
3958 ret = btrfs_update_inode(trans, root, inode);
3959 BUG_ON(ret);
3960
Yan, Zheng80825102009-11-12 09:35:36 +00003961 btrfs_end_transaction(trans, root);
3962 trans = NULL;
Liu Bob53d3f52012-11-14 14:34:34 +00003963 btrfs_btree_balance_dirty(root);
Josef Bacik7b128762008-07-24 12:17:14 -04003964 }
3965
Josef Bacik4289a662011-08-05 13:22:24 -04003966 btrfs_free_block_rsv(root, rsv);
3967
Yan, Zheng80825102009-11-12 09:35:36 +00003968 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04003969 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00003970 ret = btrfs_orphan_del(trans, inode);
3971 BUG_ON(ret);
3972 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003973
Josef Bacik4289a662011-08-05 13:22:24 -04003974 trans->block_rsv = &root->fs_info->trans_block_rsv;
Li Zefan581bb052011-04-20 10:06:11 +08003975 if (!(root == root->fs_info->tree_root ||
3976 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Li Zefan33345d012011-04-20 10:31:50 +08003977 btrfs_return_ino(root, btrfs_ino(inode));
Li Zefan581bb052011-04-20 10:06:11 +08003978
Chris Mason54aa1f42007-06-22 14:16:25 -04003979 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003980 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04003981no_delete:
Jan Karadbd57682012-05-03 14:48:02 +02003982 clear_inode(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003983 return;
Chris Mason39279cc2007-06-12 06:35:45 -04003984}
3985
3986/*
3987 * this returns the key found in the dir entry in the location pointer.
3988 * If no dir entries were found, location->objectid is 0.
3989 */
3990static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3991 struct btrfs_key *location)
3992{
3993 const char *name = dentry->d_name.name;
3994 int namelen = dentry->d_name.len;
3995 struct btrfs_dir_item *di;
3996 struct btrfs_path *path;
3997 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04003998 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003999
4000 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07004001 if (!path)
4002 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05004003
Li Zefan33345d012011-04-20 10:31:50 +08004004 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
Chris Mason39279cc2007-06-12 06:35:45 -04004005 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04004006 if (IS_ERR(di))
4007 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05004008
David Sterbac7040052011-04-19 18:00:01 +02004009 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05004010 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05004011
Chris Mason5f39d392007-10-15 16:14:19 -04004012 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04004013out:
Chris Mason39279cc2007-06-12 06:35:45 -04004014 btrfs_free_path(path);
4015 return ret;
Chris Mason39544012007-12-12 14:38:19 -05004016out_err:
4017 location->objectid = 0;
4018 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04004019}
4020
4021/*
4022 * when we hit a tree root in a directory, the btrfs part of the inode
4023 * needs to be changed to reflect the root directory of the tree root. This
4024 * is kind of like crossing a mount point.
4025 */
4026static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004027 struct inode *dir,
4028 struct dentry *dentry,
4029 struct btrfs_key *location,
4030 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04004031{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004032 struct btrfs_path *path;
4033 struct btrfs_root *new_root;
4034 struct btrfs_root_ref *ref;
4035 struct extent_buffer *leaf;
4036 int ret;
4037 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004038
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004039 path = btrfs_alloc_path();
4040 if (!path) {
4041 err = -ENOMEM;
4042 goto out;
4043 }
Chris Mason39279cc2007-06-12 06:35:45 -04004044
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004045 err = -ENOENT;
4046 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4047 BTRFS_I(dir)->root->root_key.objectid,
4048 location->objectid);
4049 if (ret) {
4050 if (ret < 0)
4051 err = ret;
4052 goto out;
4053 }
Chris Mason39279cc2007-06-12 06:35:45 -04004054
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004055 leaf = path->nodes[0];
4056 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Li Zefan33345d012011-04-20 10:31:50 +08004057 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004058 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4059 goto out;
4060
4061 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4062 (unsigned long)(ref + 1),
4063 dentry->d_name.len);
4064 if (ret)
4065 goto out;
4066
David Sterbab3b4aa72011-04-21 01:20:15 +02004067 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004068
4069 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4070 if (IS_ERR(new_root)) {
4071 err = PTR_ERR(new_root);
4072 goto out;
4073 }
4074
4075 if (btrfs_root_refs(&new_root->root_item) == 0) {
4076 err = -ENOENT;
4077 goto out;
4078 }
4079
4080 *sub_root = new_root;
4081 location->objectid = btrfs_root_dirid(&new_root->root_item);
4082 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04004083 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004084 err = 0;
4085out:
4086 btrfs_free_path(path);
4087 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004088}
4089
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004090static void inode_tree_add(struct inode *inode)
4091{
4092 struct btrfs_root *root = BTRFS_I(inode)->root;
4093 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004094 struct rb_node **p;
4095 struct rb_node *parent;
Li Zefan33345d012011-04-20 10:31:50 +08004096 u64 ino = btrfs_ino(inode);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004097again:
4098 p = &root->inode_tree.rb_node;
4099 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004100
Al Viro1d3382cb2010-10-23 15:19:20 -04004101 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04004102 return;
4103
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004104 spin_lock(&root->inode_lock);
4105 while (*p) {
4106 parent = *p;
4107 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4108
Li Zefan33345d012011-04-20 10:31:50 +08004109 if (ino < btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004110 p = &parent->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08004111 else if (ino > btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004112 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004113 else {
4114 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04004115 (I_WILL_FREE | I_FREEING)));
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004116 rb_erase(parent, &root->inode_tree);
4117 RB_CLEAR_NODE(parent);
4118 spin_unlock(&root->inode_lock);
4119 goto again;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004120 }
4121 }
4122 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4123 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4124 spin_unlock(&root->inode_lock);
4125}
4126
4127static void inode_tree_del(struct inode *inode)
4128{
4129 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04004130 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004131
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004132 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004133 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004134 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004135 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04004136 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004137 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004138 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04004139
Josef Bacik0af3d002010-06-21 14:48:16 -04004140 /*
4141 * Free space cache has inodes in the tree root, but the tree root has a
4142 * root_refs of 0, so this could end up dropping the tree root as a
4143 * snapshot, so we need the extra !root->fs_info->tree_root check to
4144 * make sure we don't drop it.
4145 */
4146 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4147 root != root->fs_info->tree_root) {
Yan, Zheng76dda932009-09-21 16:00:26 -04004148 synchronize_srcu(&root->fs_info->subvol_srcu);
4149 spin_lock(&root->inode_lock);
4150 empty = RB_EMPTY_ROOT(&root->inode_tree);
4151 spin_unlock(&root->inode_lock);
4152 if (empty)
4153 btrfs_add_dead_root(root);
4154 }
4155}
4156
Jeff Mahoney143bede2012-03-01 14:56:26 +01004157void btrfs_invalidate_inodes(struct btrfs_root *root)
Yan, Zheng76dda932009-09-21 16:00:26 -04004158{
4159 struct rb_node *node;
4160 struct rb_node *prev;
4161 struct btrfs_inode *entry;
4162 struct inode *inode;
4163 u64 objectid = 0;
4164
4165 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4166
4167 spin_lock(&root->inode_lock);
4168again:
4169 node = root->inode_tree.rb_node;
4170 prev = NULL;
4171 while (node) {
4172 prev = node;
4173 entry = rb_entry(node, struct btrfs_inode, rb_node);
4174
Li Zefan33345d012011-04-20 10:31:50 +08004175 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04004176 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08004177 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04004178 node = node->rb_right;
4179 else
4180 break;
4181 }
4182 if (!node) {
4183 while (prev) {
4184 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08004185 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04004186 node = prev;
4187 break;
4188 }
4189 prev = rb_next(prev);
4190 }
4191 }
4192 while (node) {
4193 entry = rb_entry(node, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08004194 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04004195 inode = igrab(&entry->vfs_inode);
4196 if (inode) {
4197 spin_unlock(&root->inode_lock);
4198 if (atomic_read(&inode->i_count) > 1)
4199 d_prune_aliases(inode);
4200 /*
Al Viro45321ac2010-06-07 13:43:19 -04004201 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04004202 * the inode cache when its usage count
4203 * hits zero.
4204 */
4205 iput(inode);
4206 cond_resched();
4207 spin_lock(&root->inode_lock);
4208 goto again;
4209 }
4210
4211 if (cond_resched_lock(&root->inode_lock))
4212 goto again;
4213
4214 node = rb_next(node);
4215 }
4216 spin_unlock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004217}
4218
Chris Masone02119d2008-09-05 16:13:11 -04004219static int btrfs_init_locked_inode(struct inode *inode, void *p)
4220{
4221 struct btrfs_iget_args *args = p;
4222 inode->i_ino = args->ino;
Chris Masone02119d2008-09-05 16:13:11 -04004223 BTRFS_I(inode)->root = args->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004224 return 0;
4225}
4226
4227static int btrfs_find_actor(struct inode *inode, void *opaque)
4228{
4229 struct btrfs_iget_args *args = opaque;
Li Zefan33345d012011-04-20 10:31:50 +08004230 return args->ino == btrfs_ino(inode) &&
Chris Masond3977122009-01-05 21:25:51 -05004231 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004232}
4233
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004234static struct inode *btrfs_iget_locked(struct super_block *s,
4235 u64 objectid,
4236 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04004237{
4238 struct inode *inode;
4239 struct btrfs_iget_args args;
4240 args.ino = objectid;
4241 args.root = root;
4242
4243 inode = iget5_locked(s, objectid, btrfs_find_actor,
4244 btrfs_init_locked_inode,
4245 (void *)&args);
4246 return inode;
4247}
4248
Balaji Rao1a54ef82008-07-21 02:01:04 +05304249/* Get an inode object given its location and corresponding root.
4250 * Returns in *is_new if the inode was read from disk
4251 */
4252struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00004253 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05304254{
4255 struct inode *inode;
4256
4257 inode = btrfs_iget_locked(s, location->objectid, root);
4258 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004259 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05304260
4261 if (inode->i_state & I_NEW) {
4262 BTRFS_I(inode)->root = root;
4263 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4264 btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07004265 if (!is_bad_inode(inode)) {
4266 inode_tree_add(inode);
4267 unlock_new_inode(inode);
4268 if (new)
4269 *new = 1;
4270 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04004271 unlock_new_inode(inode);
4272 iput(inode);
4273 inode = ERR_PTR(-ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07004274 }
4275 }
4276
Balaji Rao1a54ef82008-07-21 02:01:04 +05304277 return inode;
4278}
4279
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004280static struct inode *new_simple_dir(struct super_block *s,
4281 struct btrfs_key *key,
4282 struct btrfs_root *root)
4283{
4284 struct inode *inode = new_inode(s);
4285
4286 if (!inode)
4287 return ERR_PTR(-ENOMEM);
4288
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004289 BTRFS_I(inode)->root = root;
4290 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
Josef Bacik72ac3c02012-05-23 14:13:11 -04004291 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004292
4293 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
Li Zefan848cce02012-02-21 17:04:28 +08004294 inode->i_op = &btrfs_dir_ro_inode_operations;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004295 inode->i_fop = &simple_dir_operations;
4296 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4297 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4298
4299 return inode;
4300}
4301
Chris Mason3de45862008-11-17 21:02:50 -05004302struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04004303{
Chris Masond3977122009-01-05 21:25:51 -05004304 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004305 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004306 struct btrfs_root *sub_root = root;
4307 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04004308 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004309 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004310
4311 if (dentry->d_name.len > BTRFS_NAME_LEN)
4312 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04004313
Jeff Layton39e3c952012-11-28 11:30:53 -05004314 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason39279cc2007-06-12 06:35:45 -04004315 if (ret < 0)
4316 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04004317
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004318 if (location.objectid == 0)
4319 return NULL;
4320
4321 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00004322 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004323 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04004324 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004325
4326 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4327
Yan, Zheng76dda932009-09-21 16:00:26 -04004328 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004329 ret = fixup_tree_root_location(root, dir, dentry,
4330 &location, &sub_root);
4331 if (ret < 0) {
4332 if (ret != -ENOENT)
4333 inode = ERR_PTR(ret);
4334 else
4335 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4336 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00004337 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004338 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004339 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4340
Julia Lawall34d19ba2011-01-24 19:55:19 +00004341 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00004342 down_read(&root->fs_info->cleanup_work_sem);
4343 if (!(inode->i_sb->s_flags & MS_RDONLY))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05004344 ret = btrfs_orphan_cleanup(sub_root);
Yan, Zhengc71bf092009-11-12 09:34:40 +00004345 up_read(&root->fs_info->cleanup_work_sem);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05004346 if (ret)
4347 inode = ERR_PTR(ret);
Yan, Zhengc71bf092009-11-12 09:34:40 +00004348 }
4349
Chris Mason3de45862008-11-17 21:02:50 -05004350 return inode;
4351}
4352
Nick Pigginfe15ce42011-01-07 17:49:23 +11004353static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04004354{
4355 struct btrfs_root *root;
Li Zefan848cce02012-02-21 17:04:28 +08004356 struct inode *inode = dentry->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04004357
Li Zefan848cce02012-02-21 17:04:28 +08004358 if (!inode && !IS_ROOT(dentry))
4359 inode = dentry->d_parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04004360
Li Zefan848cce02012-02-21 17:04:28 +08004361 if (inode) {
4362 root = BTRFS_I(inode)->root;
Yan, Zhengefefb142009-10-09 09:25:16 -04004363 if (btrfs_root_refs(&root->root_item) == 0)
4364 return 1;
Li Zefan848cce02012-02-21 17:04:28 +08004365
4366 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4367 return 1;
Yan, Zhengefefb142009-10-09 09:25:16 -04004368 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004369 return 0;
4370}
4371
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004372static void btrfs_dentry_release(struct dentry *dentry)
4373{
4374 if (dentry->d_fsdata)
4375 kfree(dentry->d_fsdata);
4376}
4377
Chris Mason3de45862008-11-17 21:02:50 -05004378static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -04004379 unsigned int flags)
Chris Mason3de45862008-11-17 21:02:50 -05004380{
Josef Bacika66e7cc2011-09-18 10:34:03 -04004381 struct dentry *ret;
4382
4383 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
Josef Bacika66e7cc2011-09-18 10:34:03 -04004384 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004385}
4386
Miao Xie16cdcec2011-04-22 18:12:22 +08004387unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04004388 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4389};
4390
David Woodhousecbdf5a22008-08-06 19:42:33 +01004391static int btrfs_real_readdir(struct file *filp, void *dirent,
4392 filldir_t filldir)
Chris Mason39279cc2007-06-12 06:35:45 -04004393{
Chris Mason6da6aba2007-12-18 16:15:09 -05004394 struct inode *inode = filp->f_dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04004395 struct btrfs_root *root = BTRFS_I(inode)->root;
4396 struct btrfs_item *item;
4397 struct btrfs_dir_item *di;
4398 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04004399 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004400 struct btrfs_path *path;
Miao Xie16cdcec2011-04-22 18:12:22 +08004401 struct list_head ins_list;
4402 struct list_head del_list;
Chris Mason39279cc2007-06-12 06:35:45 -04004403 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04004404 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004405 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -04004406 unsigned char d_type;
4407 int over = 0;
4408 u32 di_cur;
4409 u32 di_total;
4410 u32 di_len;
4411 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004412 char tmp_name[32];
4413 char *name_ptr;
4414 int name_len;
Miao Xie16cdcec2011-04-22 18:12:22 +08004415 int is_curr = 0; /* filp->f_pos points to the current index? */
Chris Mason39279cc2007-06-12 06:35:45 -04004416
4417 /* FIXME, use a real flag for deciding about the key type */
4418 if (root->fs_info->tree_root == root)
4419 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004420
Chris Mason39544012007-12-12 14:38:19 -05004421 /* special case for "." */
4422 if (filp->f_pos == 0) {
Hidetoshi Seto3765fef2011-09-18 10:20:46 -04004423 over = filldir(dirent, ".", 1,
4424 filp->f_pos, btrfs_ino(inode), DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004425 if (over)
4426 return 0;
4427 filp->f_pos = 1;
4428 }
Chris Mason39544012007-12-12 14:38:19 -05004429 /* special case for .., just use the back ref */
4430 if (filp->f_pos == 1) {
David Woodhouse5ecc7e52008-08-17 15:14:48 +01004431 u64 pino = parent_ino(filp->f_path.dentry);
Chris Mason39544012007-12-12 14:38:19 -05004432 over = filldir(dirent, "..", 2,
Hidetoshi Seto3765fef2011-09-18 10:20:46 -04004433 filp->f_pos, pino, DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004434 if (over)
David Woodhouse49593bf2008-08-17 17:08:36 +01004435 return 0;
Chris Mason39544012007-12-12 14:38:19 -05004436 filp->f_pos = 2;
4437 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004438 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08004439 if (!path)
4440 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04004441
Josef Bacik026fd312011-05-13 10:32:11 -04004442 path->reada = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01004443
Miao Xie16cdcec2011-04-22 18:12:22 +08004444 if (key_type == BTRFS_DIR_INDEX_KEY) {
4445 INIT_LIST_HEAD(&ins_list);
4446 INIT_LIST_HEAD(&del_list);
4447 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4448 }
4449
Chris Mason39279cc2007-06-12 06:35:45 -04004450 btrfs_set_key_type(&key, key_type);
4451 key.offset = filp->f_pos;
Li Zefan33345d012011-04-20 10:31:50 +08004452 key.objectid = btrfs_ino(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004453
Chris Mason39279cc2007-06-12 06:35:45 -04004454 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4455 if (ret < 0)
4456 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01004457
4458 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04004459 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04004460 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08004461 if (slot >= btrfs_header_nritems(leaf)) {
4462 ret = btrfs_next_leaf(root, path);
4463 if (ret < 0)
4464 goto err;
4465 else if (ret > 0)
4466 break;
4467 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04004468 }
Chris Mason3de45862008-11-17 21:02:50 -05004469
Chris Mason5f39d392007-10-15 16:14:19 -04004470 item = btrfs_item_nr(leaf, slot);
4471 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4472
4473 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04004474 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004475 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004476 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004477 if (found_key.offset < filp->f_pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08004478 goto next;
Miao Xie16cdcec2011-04-22 18:12:22 +08004479 if (key_type == BTRFS_DIR_INDEX_KEY &&
4480 btrfs_should_delete_dir_index(&del_list,
4481 found_key.offset))
4482 goto next;
Chris Mason5f39d392007-10-15 16:14:19 -04004483
4484 filp->f_pos = found_key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08004485 is_curr = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01004486
Chris Mason39279cc2007-06-12 06:35:45 -04004487 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4488 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04004489 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01004490
4491 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04004492 struct btrfs_key location;
4493
Josef Bacik22a94d42011-03-16 16:47:17 -04004494 if (verify_dir_item(root, leaf, di))
4495 break;
4496
Chris Mason5f39d392007-10-15 16:14:19 -04004497 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01004498 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04004499 name_ptr = tmp_name;
4500 } else {
4501 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01004502 if (!name_ptr) {
4503 ret = -ENOMEM;
4504 goto err;
4505 }
Chris Mason5f39d392007-10-15 16:14:19 -04004506 }
4507 read_extent_buffer(leaf, name_ptr,
4508 (unsigned long)(di + 1), name_len);
4509
4510 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4511 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05004512
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004513
Chris Mason3de45862008-11-17 21:02:50 -05004514 /* is this a reference to our own snapshot? If so
Arne Jansen8c9c2bf2012-02-25 09:09:30 +01004515 * skip it.
4516 *
4517 * In contrast to old kernels, we insert the snapshot's
4518 * dir item and dir index after it has been created, so
4519 * we won't find a reference to our own snapshot. We
4520 * still keep the following code for backward
4521 * compatibility.
Chris Mason3de45862008-11-17 21:02:50 -05004522 */
4523 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4524 location.objectid == root->root_key.objectid) {
4525 over = 0;
4526 goto skip;
4527 }
Chris Mason5f39d392007-10-15 16:14:19 -04004528 over = filldir(dirent, name_ptr, name_len,
David Woodhouse49593bf2008-08-17 17:08:36 +01004529 found_key.offset, location.objectid,
Chris Mason39279cc2007-06-12 06:35:45 -04004530 d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04004531
Chris Mason3de45862008-11-17 21:02:50 -05004532skip:
Chris Mason5f39d392007-10-15 16:14:19 -04004533 if (name_ptr != tmp_name)
4534 kfree(name_ptr);
4535
Chris Mason39279cc2007-06-12 06:35:45 -04004536 if (over)
4537 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05004538 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01004539 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04004540 di_cur += di_len;
4541 di = (struct btrfs_dir_item *)((char *)di + di_len);
4542 }
Li Zefanb9e03af2011-03-23 10:43:58 +08004543next:
4544 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04004545 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004546
Miao Xie16cdcec2011-04-22 18:12:22 +08004547 if (key_type == BTRFS_DIR_INDEX_KEY) {
4548 if (is_curr)
4549 filp->f_pos++;
4550 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4551 &ins_list);
4552 if (ret)
4553 goto nopos;
4554 }
4555
David Woodhouse49593bf2008-08-17 17:08:36 +01004556 /* Reached end of directory/root. Bump pos past the last item. */
Yan Zheng5e591a02008-02-19 11:41:02 -05004557 if (key_type == BTRFS_DIR_INDEX_KEY)
Jan Engelhardt406266a2009-12-09 22:00:38 +00004558 /*
4559 * 32-bit glibc will use getdents64, but then strtol -
4560 * so the last number we can serve is this.
4561 */
4562 filp->f_pos = 0x7fffffff;
Yan Zheng5e591a02008-02-19 11:41:02 -05004563 else
4564 filp->f_pos++;
Chris Mason39279cc2007-06-12 06:35:45 -04004565nopos:
4566 ret = 0;
4567err:
Miao Xie16cdcec2011-04-22 18:12:22 +08004568 if (key_type == BTRFS_DIR_INDEX_KEY)
4569 btrfs_put_delayed_items(&ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04004570 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004571 return ret;
4572}
4573
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004574int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04004575{
4576 struct btrfs_root *root = BTRFS_I(inode)->root;
4577 struct btrfs_trans_handle *trans;
4578 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04004579 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04004580
Josef Bacik72ac3c02012-05-23 14:13:11 -04004581 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Chris Mason4ca8b412008-08-05 13:30:48 -04004582 return 0;
4583
Liu Bo83eea1f2012-07-10 05:28:39 -06004584 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
Li Zefan82d59022011-04-20 10:33:24 +08004585 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04004586
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004587 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04004588 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004589 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04004590 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004591 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004592 if (IS_ERR(trans))
4593 return PTR_ERR(trans);
Miao Xiea698d0752012-09-20 01:51:59 -06004594 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004595 }
4596 return ret;
4597}
4598
4599/*
Chris Mason54aa1f42007-06-22 14:16:25 -04004600 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04004601 * inode changes. But, it is most likely to find the inode in cache.
4602 * FIXME, needs more benchmarking...there are no reasons other than performance
4603 * to keep or drop this code.
4604 */
Josef Bacik22c44fe2011-11-30 10:45:38 -05004605int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04004606{
4607 struct btrfs_root *root = BTRFS_I(inode)->root;
4608 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004609 int ret;
4610
Josef Bacik72ac3c02012-05-23 14:13:11 -04004611 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Josef Bacik22c44fe2011-11-30 10:45:38 -05004612 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004613
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004614 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004615 if (IS_ERR(trans))
4616 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004617
4618 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004619 if (ret && ret == -ENOSPC) {
4620 /* whoops, lets try again with the full transaction */
4621 btrfs_end_transaction(trans, root);
4622 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004623 if (IS_ERR(trans))
4624 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004625
Chris Mason94b60442010-05-26 11:02:00 -04004626 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004627 }
Chris Mason39279cc2007-06-12 06:35:45 -04004628 btrfs_end_transaction(trans, root);
Miao Xie16cdcec2011-04-22 18:12:22 +08004629 if (BTRFS_I(inode)->delayed_node)
4630 btrfs_balance_delayed_items(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004631
4632 return ret;
4633}
4634
4635/*
4636 * This is a copy of file_update_time. We need this so we can return error on
4637 * ENOSPC for updating the inode in the case of file write and mmap writes.
4638 */
Josef Bacike41f9412012-03-26 09:46:47 -04004639static int btrfs_update_time(struct inode *inode, struct timespec *now,
4640 int flags)
Josef Bacik22c44fe2011-11-30 10:45:38 -05004641{
Alexander Block2bc5565282012-06-15 09:49:33 +02004642 struct btrfs_root *root = BTRFS_I(inode)->root;
4643
4644 if (btrfs_root_readonly(root))
4645 return -EROFS;
4646
Josef Bacike41f9412012-03-26 09:46:47 -04004647 if (flags & S_VERSION)
Josef Bacik22c44fe2011-11-30 10:45:38 -05004648 inode_inc_iversion(inode);
Josef Bacike41f9412012-03-26 09:46:47 -04004649 if (flags & S_CTIME)
4650 inode->i_ctime = *now;
4651 if (flags & S_MTIME)
4652 inode->i_mtime = *now;
4653 if (flags & S_ATIME)
4654 inode->i_atime = *now;
4655 return btrfs_dirty_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004656}
4657
Chris Masond352ac62008-09-29 15:18:18 -04004658/*
4659 * find the highest existing sequence number in a directory
4660 * and then set the in-memory index_cnt variable to reflect
4661 * free sequence numbers
4662 */
Josef Bacikaec74772008-07-24 12:12:38 -04004663static int btrfs_set_inode_index_count(struct inode *inode)
4664{
4665 struct btrfs_root *root = BTRFS_I(inode)->root;
4666 struct btrfs_key key, found_key;
4667 struct btrfs_path *path;
4668 struct extent_buffer *leaf;
4669 int ret;
4670
Li Zefan33345d012011-04-20 10:31:50 +08004671 key.objectid = btrfs_ino(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004672 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4673 key.offset = (u64)-1;
4674
4675 path = btrfs_alloc_path();
4676 if (!path)
4677 return -ENOMEM;
4678
4679 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4680 if (ret < 0)
4681 goto out;
4682 /* FIXME: we should be able to handle this */
4683 if (ret == 0)
4684 goto out;
4685 ret = 0;
4686
4687 /*
4688 * MAGIC NUMBER EXPLANATION:
4689 * since we search a directory based on f_pos we have to start at 2
4690 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4691 * else has to start at 2
4692 */
4693 if (path->slots[0] == 0) {
4694 BTRFS_I(inode)->index_cnt = 2;
4695 goto out;
4696 }
4697
4698 path->slots[0]--;
4699
4700 leaf = path->nodes[0];
4701 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4702
Li Zefan33345d012011-04-20 10:31:50 +08004703 if (found_key.objectid != btrfs_ino(inode) ||
Josef Bacikaec74772008-07-24 12:12:38 -04004704 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4705 BTRFS_I(inode)->index_cnt = 2;
4706 goto out;
4707 }
4708
4709 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4710out:
4711 btrfs_free_path(path);
4712 return ret;
4713}
4714
Chris Masond352ac62008-09-29 15:18:18 -04004715/*
4716 * helper to find a free sequence number in a given directory. This current
4717 * code is very simple, later versions will do smarter things in the btree
4718 */
Chris Mason3de45862008-11-17 21:02:50 -05004719int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04004720{
4721 int ret = 0;
4722
4723 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
Miao Xie16cdcec2011-04-22 18:12:22 +08004724 ret = btrfs_inode_delayed_dir_index_count(dir);
4725 if (ret) {
4726 ret = btrfs_set_inode_index_count(dir);
4727 if (ret)
4728 return ret;
4729 }
Josef Bacikaec74772008-07-24 12:12:38 -04004730 }
4731
Chris Mason00e4e6b2008-08-05 11:18:09 -04004732 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04004733 BTRFS_I(dir)->index_cnt++;
4734
4735 return ret;
4736}
4737
Chris Mason39279cc2007-06-12 06:35:45 -04004738static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4739 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04004740 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05004741 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04004742 u64 ref_objectid, u64 objectid,
4743 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04004744{
4745 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004746 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04004747 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04004748 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05004749 struct btrfs_inode_ref *ref;
4750 struct btrfs_key key[2];
4751 u32 sizes[2];
4752 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04004753 int ret;
4754 int owner;
4755
Chris Mason5f39d392007-10-15 16:14:19 -04004756 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07004757 if (!path)
4758 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04004759
Chris Mason39279cc2007-06-12 06:35:45 -04004760 inode = new_inode(root->fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004761 if (!inode) {
4762 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004763 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004764 }
Chris Mason39279cc2007-06-12 06:35:45 -04004765
Li Zefan581bb052011-04-20 10:06:11 +08004766 /*
4767 * we have to initialize this early, so we can reclaim the inode
4768 * number if we fail afterwards in this function.
4769 */
4770 inode->i_ino = objectid;
4771
Josef Bacikaec74772008-07-24 12:12:38 -04004772 if (dir) {
liubo1abe9b82011-03-24 11:18:59 +00004773 trace_btrfs_inode_request(dir);
4774
Chris Mason3de45862008-11-17 21:02:50 -05004775 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04004776 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004777 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004778 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004779 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04004780 }
Josef Bacikaec74772008-07-24 12:12:38 -04004781 }
4782 /*
4783 * index_cnt is ignored for everything but a dir,
4784 * btrfs_get_inode_index_count has an explanation for the magic
4785 * number
4786 */
4787 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04004788 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04004789 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00004790 inode->i_generation = BTRFS_I(inode)->generation;
Chris Masonb888db2b2007-08-27 16:49:44 -04004791
Josef Bacik5dc562c2012-08-17 13:14:17 -04004792 /*
4793 * We could have gotten an inode number from somebody who was fsynced
4794 * and then removed in this same transaction, so let's just set full
4795 * sync since it will be a full sync anyway and this will blow away the
4796 * old info in the log.
4797 */
4798 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4799
Al Viro569254b2011-07-24 17:08:40 -04004800 if (S_ISDIR(mode))
Chris Mason39279cc2007-06-12 06:35:45 -04004801 owner = 0;
4802 else
4803 owner = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004804
4805 key[0].objectid = objectid;
4806 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4807 key[0].offset = 0;
4808
Mark Fashehf1863732012-08-08 11:32:27 -07004809 /*
4810 * Start new inodes with an inode_ref. This is slightly more
4811 * efficient for small numbers of hard links since they will
4812 * be packed into one item. Extended refs will kick in if we
4813 * add more hard links than can fit in the ref item.
4814 */
Chris Mason9c583092008-01-29 15:15:18 -05004815 key[1].objectid = objectid;
4816 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4817 key[1].offset = ref_objectid;
4818
4819 sizes[0] = sizeof(struct btrfs_inode_item);
4820 sizes[1] = name_len + sizeof(*ref);
4821
Chris Masonb9473432009-03-13 11:00:37 -04004822 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004823 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4824 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04004825 goto fail;
4826
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03004827 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004828 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004829 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04004830 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4831 struct btrfs_inode_item);
Li Zefan293f7e02012-07-10 00:58:58 -06004832 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4833 sizeof(*inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04004834 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05004835
4836 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4837 struct btrfs_inode_ref);
4838 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004839 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05004840 ptr = (unsigned long)(ref + 1);
4841 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4842
Chris Mason5f39d392007-10-15 16:14:19 -04004843 btrfs_mark_buffer_dirty(path->nodes[0]);
4844 btrfs_free_path(path);
4845
Chris Mason39279cc2007-06-12 06:35:45 -04004846 location = &BTRFS_I(inode)->location;
4847 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04004848 location->offset = 0;
4849 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4850
Christoph Hellwig6cbff002009-04-17 10:37:41 +02004851 btrfs_inherit_iflags(inode, dir);
4852
Al Viro569254b2011-07-24 17:08:40 -04004853 if (S_ISREG(mode)) {
Chris Mason94272162009-07-02 12:26:06 -04004854 if (btrfs_test_opt(root, NODATASUM))
4855 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Liu Bo213490b2012-09-11 08:33:50 -06004856 if (btrfs_test_opt(root, NODATACOW))
Chris Mason94272162009-07-02 12:26:06 -04004857 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4858 }
4859
Chris Mason39279cc2007-06-12 06:35:45 -04004860 insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004861 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00004862
4863 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04004864 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00004865
Alexander Block8ea05e32012-07-25 17:35:53 +02004866 btrfs_update_root_times(trans, root);
4867
Chris Mason39279cc2007-06-12 06:35:45 -04004868 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004869fail:
Josef Bacikaec74772008-07-24 12:12:38 -04004870 if (dir)
4871 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04004872 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004873 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004874 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004875}
4876
4877static inline u8 btrfs_inode_type(struct inode *inode)
4878{
4879 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4880}
4881
Chris Masond352ac62008-09-29 15:18:18 -04004882/*
4883 * utility function to add 'inode' into 'parent_inode' with
4884 * a give name and a given sequence number.
4885 * if 'add_backref' is true, also insert a backref from the
4886 * inode to the parent directory.
4887 */
Chris Masone02119d2008-09-05 16:13:11 -04004888int btrfs_add_link(struct btrfs_trans_handle *trans,
4889 struct inode *parent_inode, struct inode *inode,
4890 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004891{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004892 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004893 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04004894 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Li Zefan33345d012011-04-20 10:31:50 +08004895 u64 ino = btrfs_ino(inode);
4896 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004897
Li Zefan33345d012011-04-20 10:31:50 +08004898 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004899 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4900 } else {
Li Zefan33345d012011-04-20 10:31:50 +08004901 key.objectid = ino;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004902 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4903 key.offset = 0;
4904 }
Chris Mason39279cc2007-06-12 06:35:45 -04004905
Li Zefan33345d012011-04-20 10:31:50 +08004906 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004907 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4908 key.objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08004909 parent_ino, index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004910 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08004911 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4912 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004913 }
4914
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004915 /* Nothing to clean up yet */
4916 if (ret)
4917 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004918
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004919 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4920 parent_inode, &key,
4921 btrfs_inode_type(inode), index);
Chris Mason9c520572012-12-17 14:26:57 -05004922 if (ret == -EEXIST || ret == -EOVERFLOW)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004923 goto fail_dir_item;
4924 else if (ret) {
4925 btrfs_abort_transaction(trans, root, ret);
4926 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004927 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004928
4929 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4930 name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004931 inode_inc_iversion(parent_inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004932 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4933 ret = btrfs_update_inode(trans, root, parent_inode);
4934 if (ret)
4935 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004936 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05004937
4938fail_dir_item:
4939 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4940 u64 local_index;
4941 int err;
4942 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4943 key.objectid, root->root_key.objectid,
4944 parent_ino, &local_index, name, name_len);
4945
4946 } else if (add_backref) {
4947 u64 local_index;
4948 int err;
4949
4950 err = btrfs_del_inode_ref(trans, root, name, name_len,
4951 ino, parent_ino, &local_index);
4952 }
4953 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004954}
4955
4956static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00004957 struct inode *dir, struct dentry *dentry,
4958 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004959{
Josef Bacika1b075d2010-11-19 20:36:11 +00004960 int err = btrfs_add_link(trans, dir, inode,
4961 dentry->d_name.name, dentry->d_name.len,
4962 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004963 if (err > 0)
4964 err = -EEXIST;
4965 return err;
4966}
4967
Josef Bacik618e21d2007-07-11 10:18:17 -04004968static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04004969 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04004970{
4971 struct btrfs_trans_handle *trans;
4972 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004973 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04004974 int err;
4975 int drop_inode = 0;
4976 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004977 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04004978
4979 if (!new_valid_dev(rdev))
4980 return -EINVAL;
4981
Josef Bacik9ed74f22009-09-11 16:12:44 -04004982 /*
4983 * 2 for inode item and ref
4984 * 2 for dir items
4985 * 1 for xattr if selinux is on
4986 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004987 trans = btrfs_start_transaction(root, 5);
4988 if (IS_ERR(trans))
4989 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05004990
Li Zefan581bb052011-04-20 10:06:11 +08004991 err = btrfs_find_free_ino(root, &objectid);
4992 if (err)
4993 goto out_unlock;
4994
Josef Bacikaec74772008-07-24 12:12:38 -04004995 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004996 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04004997 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004998 if (IS_ERR(inode)) {
4999 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04005000 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005001 }
Josef Bacik618e21d2007-07-11 10:18:17 -04005002
Eric Paris2a7dba32011-02-01 11:05:39 -05005003 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04005004 if (err) {
5005 drop_inode = 1;
5006 goto out_unlock;
5007 }
5008
Filipe Brandenburger2794ed02012-11-30 03:40:08 +00005009 err = btrfs_update_inode(trans, root, inode);
5010 if (err) {
5011 drop_inode = 1;
5012 goto out_unlock;
5013 }
5014
Casey Schauflerad19db72011-12-15 10:09:07 -05005015 /*
5016 * If the active LSM wants to access the inode during
5017 * d_instantiate it needs these. Smack checks to see
5018 * if the filesystem supports xattrs by looking at the
5019 * ops vector.
5020 */
5021
5022 inode->i_op = &btrfs_special_inode_operations;
Josef Bacika1b075d2010-11-19 20:36:11 +00005023 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04005024 if (err)
5025 drop_inode = 1;
5026 else {
Josef Bacik618e21d2007-07-11 10:18:17 -04005027 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04005028 btrfs_update_inode(trans, root, inode);
Al Viro08c422c2011-12-23 07:58:13 -05005029 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04005030 }
Josef Bacik618e21d2007-07-11 10:18:17 -04005031out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005032 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00005033 btrfs_btree_balance_dirty(root);
Josef Bacik618e21d2007-07-11 10:18:17 -04005034 if (drop_inode) {
5035 inode_dec_link_count(inode);
5036 iput(inode);
5037 }
Josef Bacik618e21d2007-07-11 10:18:17 -04005038 return err;
5039}
5040
Chris Mason39279cc2007-06-12 06:35:45 -04005041static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -04005042 umode_t mode, bool excl)
Chris Mason39279cc2007-06-12 06:35:45 -04005043{
5044 struct btrfs_trans_handle *trans;
5045 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05005046 struct inode *inode = NULL;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005047 int drop_inode_on_err = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04005048 int err;
Chris Mason39279cc2007-06-12 06:35:45 -04005049 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005050 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005051
Josef Bacik9ed74f22009-09-11 16:12:44 -04005052 /*
5053 * 2 for inode item and ref
5054 * 2 for dir items
5055 * 1 for xattr if selinux is on
5056 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04005057 trans = btrfs_start_transaction(root, 5);
5058 if (IS_ERR(trans))
5059 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005060
Li Zefan581bb052011-04-20 10:06:11 +08005061 err = btrfs_find_free_ino(root, &objectid);
5062 if (err)
5063 goto out_unlock;
5064
Josef Bacikaec74772008-07-24 12:12:38 -04005065 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08005066 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04005067 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005068 if (IS_ERR(inode)) {
5069 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005070 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005071 }
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005072 drop_inode_on_err = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04005073
Eric Paris2a7dba32011-02-01 11:05:39 -05005074 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005075 if (err)
Josef Bacik33268ea2008-07-24 12:16:36 -04005076 goto out_unlock;
Josef Bacik33268ea2008-07-24 12:16:36 -04005077
Filipe Brandenburger9185aa52012-11-30 03:40:08 +00005078 err = btrfs_update_inode(trans, root, inode);
5079 if (err)
5080 goto out_unlock;
5081
Casey Schauflerad19db72011-12-15 10:09:07 -05005082 /*
5083 * If the active LSM wants to access the inode during
5084 * d_instantiate it needs these. Smack checks to see
5085 * if the filesystem supports xattrs by looking at the
5086 * ops vector.
5087 */
5088 inode->i_fop = &btrfs_file_operations;
5089 inode->i_op = &btrfs_file_inode_operations;
5090
Josef Bacika1b075d2010-11-19 20:36:11 +00005091 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005092 if (err)
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005093 goto out_unlock;
5094
5095 inode->i_mapping->a_ops = &btrfs_aops;
5096 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5097 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5098 d_instantiate(dentry, inode);
5099
Chris Mason39279cc2007-06-12 06:35:45 -04005100out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005101 btrfs_end_transaction(trans, root);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005102 if (err && drop_inode_on_err) {
Chris Mason39279cc2007-06-12 06:35:45 -04005103 inode_dec_link_count(inode);
5104 iput(inode);
5105 }
Liu Bob53d3f52012-11-14 14:34:34 +00005106 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005107 return err;
5108}
5109
5110static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5111 struct dentry *dentry)
5112{
5113 struct btrfs_trans_handle *trans;
5114 struct btrfs_root *root = BTRFS_I(dir)->root;
5115 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005116 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04005117 int err;
5118 int drop_inode = 0;
5119
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00005120 /* do not allow sys_link's with other subvols of the same device */
5121 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00005122 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00005123
Mark Fashehf1863732012-08-08 11:32:27 -07005124 if (inode->i_nlink >= BTRFS_LINK_MAX)
Al Viroc055e992011-03-04 17:15:18 +00005125 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04005126
Chris Mason3de45862008-11-17 21:02:50 -05005127 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04005128 if (err)
5129 goto fail;
5130
Yan, Zhenga22285a2010-05-16 10:48:46 -04005131 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00005132 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04005133 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00005134 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04005135 */
Miao Xie7e6b6462011-02-18 09:21:17 +00005136 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04005137 if (IS_ERR(trans)) {
5138 err = PTR_ERR(trans);
5139 goto fail;
5140 }
Chris Mason5f39d392007-10-15 16:14:19 -04005141
Miao Xie31534952011-04-13 13:19:21 +08005142 btrfs_inc_nlink(inode);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04005143 inode_inc_iversion(inode);
Miao Xie31534952011-04-13 13:19:21 +08005144 inode->i_ctime = CURRENT_TIME;
Al Viro7de9c6ee2010-10-23 11:11:40 -04005145 ihold(inode);
Josef Bacike9976152012-10-11 15:53:56 -04005146 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
Josef Bacikaec74772008-07-24 12:12:38 -04005147
Josef Bacika1b075d2010-11-19 20:36:11 +00005148 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04005149
Yan, Zhenga5719522009-09-24 09:17:31 -04005150 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04005151 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04005152 } else {
Al Viro10d9f302011-07-16 23:09:10 -04005153 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04005154 err = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005155 if (err)
5156 goto fail;
Al Viro08c422c2011-12-23 07:58:13 -05005157 d_instantiate(dentry, inode);
Josef Bacik6a912212010-11-20 09:48:00 +00005158 btrfs_log_new_name(trans, inode, NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04005159 }
Chris Mason39279cc2007-06-12 06:35:45 -04005160
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005161 btrfs_end_transaction(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05005162fail:
Chris Mason39279cc2007-06-12 06:35:45 -04005163 if (drop_inode) {
5164 inode_dec_link_count(inode);
5165 iput(inode);
5166 }
Liu Bob53d3f52012-11-14 14:34:34 +00005167 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005168 return err;
5169}
5170
Al Viro18bb1db2011-07-26 01:41:39 -04005171static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04005172{
Chris Masonb9d86662008-05-02 16:13:49 -04005173 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04005174 struct btrfs_trans_handle *trans;
5175 struct btrfs_root *root = BTRFS_I(dir)->root;
5176 int err = 0;
5177 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04005178 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005179 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005180
Josef Bacik9ed74f22009-09-11 16:12:44 -04005181 /*
5182 * 2 items for inode and ref
5183 * 2 items for dir items
5184 * 1 for xattr if selinux is on
5185 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04005186 trans = btrfs_start_transaction(root, 5);
5187 if (IS_ERR(trans))
5188 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04005189
Li Zefan581bb052011-04-20 10:06:11 +08005190 err = btrfs_find_free_ino(root, &objectid);
5191 if (err)
5192 goto out_fail;
5193
Josef Bacikaec74772008-07-24 12:12:38 -04005194 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08005195 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04005196 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04005197 if (IS_ERR(inode)) {
5198 err = PTR_ERR(inode);
5199 goto out_fail;
5200 }
Chris Mason5f39d392007-10-15 16:14:19 -04005201
Chris Mason39279cc2007-06-12 06:35:45 -04005202 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04005203
Eric Paris2a7dba32011-02-01 11:05:39 -05005204 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04005205 if (err)
5206 goto out_fail;
5207
Chris Mason39279cc2007-06-12 06:35:45 -04005208 inode->i_op = &btrfs_dir_inode_operations;
5209 inode->i_fop = &btrfs_dir_file_operations;
Chris Mason39279cc2007-06-12 06:35:45 -04005210
Chris Masondbe674a2008-07-17 12:54:05 -04005211 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04005212 err = btrfs_update_inode(trans, root, inode);
5213 if (err)
5214 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04005215
Josef Bacika1b075d2010-11-19 20:36:11 +00005216 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5217 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005218 if (err)
5219 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04005220
Chris Mason39279cc2007-06-12 06:35:45 -04005221 d_instantiate(dentry, inode);
5222 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005223
5224out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005225 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04005226 if (drop_on_err)
5227 iput(inode);
Liu Bob53d3f52012-11-14 14:34:34 +00005228 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005229 return err;
5230}
5231
Chris Masond352ac62008-09-29 15:18:18 -04005232/* helper for btfs_get_extent. Given an existing extent in the tree,
5233 * and an extent that you want to insert, deal with overlap and insert
5234 * the new extent into the tree.
5235 */
Chris Mason3b951512008-04-17 11:29:12 -04005236static int merge_extent_mapping(struct extent_map_tree *em_tree,
5237 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005238 struct extent_map *em,
5239 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04005240{
5241 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04005242
Chris Masone6dcd2d2008-07-17 12:53:50 -04005243 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5244 start_diff = map_start - em->start;
5245 em->start = map_start;
5246 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04005247 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5248 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04005249 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04005250 em->block_len -= start_diff;
5251 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005252 return add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04005253}
5254
Chris Masonc8b97812008-10-29 14:49:59 -04005255static noinline int uncompress_inline(struct btrfs_path *path,
5256 struct inode *inode, struct page *page,
5257 size_t pg_offset, u64 extent_offset,
5258 struct btrfs_file_extent_item *item)
5259{
5260 int ret;
5261 struct extent_buffer *leaf = path->nodes[0];
5262 char *tmp;
5263 size_t max_size;
5264 unsigned long inline_size;
5265 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08005266 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04005267
5268 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08005269 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04005270 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5271 inline_size = btrfs_file_extent_inline_item_len(leaf,
5272 btrfs_item_nr(leaf, path->slots[0]));
5273 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04005274 if (!tmp)
5275 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04005276 ptr = btrfs_file_extent_inline_start(item);
5277
5278 read_extent_buffer(leaf, tmp, ptr, inline_size);
5279
Chris Mason5b050f02008-11-11 09:34:41 -05005280 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08005281 ret = btrfs_decompress(compress_type, tmp, page,
5282 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04005283 if (ret) {
Cong Wang7ac687d2011-11-25 23:14:28 +08005284 char *kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -04005285 unsigned long copy_size = min_t(u64,
5286 PAGE_CACHE_SIZE - pg_offset,
5287 max_size - extent_offset);
5288 memset(kaddr + pg_offset, 0, copy_size);
Cong Wang7ac687d2011-11-25 23:14:28 +08005289 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -04005290 }
5291 kfree(tmp);
5292 return 0;
5293}
5294
Chris Masond352ac62008-09-29 15:18:18 -04005295/*
5296 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05005297 * the ugly parts come from merging extents from the disk with the in-ram
5298 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04005299 * where the in-ram extents might be locked pending data=ordered completion.
5300 *
5301 * This also copies inline extents directly into the page.
5302 */
Chris Masond3977122009-01-05 21:25:51 -05005303
Chris Masona52d9a82007-08-27 16:49:44 -04005304struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05005305 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04005306 int create)
5307{
5308 int ret;
5309 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04005310 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04005311 u64 extent_start = 0;
5312 u64 extent_end = 0;
Li Zefan33345d012011-04-20 10:31:50 +08005313 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04005314 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04005315 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04005316 struct btrfs_root *root = BTRFS_I(inode)->root;
5317 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04005318 struct extent_buffer *leaf;
5319 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04005320 struct extent_map *em = NULL;
5321 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05005322 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04005323 struct btrfs_trans_handle *trans = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08005324 int compress_type;
Chris Masona52d9a82007-08-27 16:49:44 -04005325
Chris Masona52d9a82007-08-27 16:49:44 -04005326again:
Chris Mason890871b2009-09-02 16:24:52 -04005327 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05005328 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04005329 if (em)
5330 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04005331 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05005332
Chris Masona52d9a82007-08-27 16:49:44 -04005333 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04005334 if (em->start > start || em->start + em->len <= start)
5335 free_extent_map(em);
5336 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05005337 free_extent_map(em);
5338 else
5339 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005340 }
David Sterba172ddd62011-04-21 00:48:27 +02005341 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04005342 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05005343 err = -ENOMEM;
5344 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005345 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005346 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05005347 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05005348 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05005349 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04005350 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04005351
5352 if (!path) {
5353 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04005354 if (!path) {
5355 err = -ENOMEM;
5356 goto out;
5357 }
5358 /*
5359 * Chances are we'll be called again, so go ahead and do
5360 * readahead
5361 */
5362 path->reada = 1;
Chris Masonf4219502008-07-22 11:18:09 -04005363 }
5364
Chris Mason179e29e2007-11-01 11:28:41 -04005365 ret = btrfs_lookup_file_extent(trans, root, path,
5366 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04005367 if (ret < 0) {
5368 err = ret;
5369 goto out;
5370 }
5371
5372 if (ret != 0) {
5373 if (path->slots[0] == 0)
5374 goto not_found;
5375 path->slots[0]--;
5376 }
5377
Chris Mason5f39d392007-10-15 16:14:19 -04005378 leaf = path->nodes[0];
5379 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04005380 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04005381 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04005382 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5383 found_type = btrfs_key_type(&found_key);
5384 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04005385 found_type != BTRFS_EXTENT_DATA_KEY) {
5386 goto not_found;
5387 }
5388
Chris Mason5f39d392007-10-15 16:14:19 -04005389 found_type = btrfs_file_extent_type(leaf, item);
5390 extent_start = found_key.offset;
Li Zefan261507a02010-12-17 14:21:50 +08005391 compress_type = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04005392 if (found_type == BTRFS_FILE_EXTENT_REG ||
5393 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04005394 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04005395 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04005396 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5397 size_t size;
5398 size = btrfs_file_extent_inline_len(leaf, item);
5399 extent_end = (extent_start + size + root->sectorsize - 1) &
5400 ~((u64)root->sectorsize - 1);
5401 }
5402
5403 if (start >= extent_end) {
5404 path->slots[0]++;
5405 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5406 ret = btrfs_next_leaf(root, path);
5407 if (ret < 0) {
5408 err = ret;
5409 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005410 }
Yan Zheng9036c102008-10-30 14:19:41 -04005411 if (ret > 0)
5412 goto not_found;
5413 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04005414 }
Yan Zheng9036c102008-10-30 14:19:41 -04005415 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5416 if (found_key.objectid != objectid ||
5417 found_key.type != BTRFS_EXTENT_DATA_KEY)
5418 goto not_found;
5419 if (start + len <= found_key.offset)
5420 goto not_found;
5421 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04005422 em->orig_start = start;
Yan Zheng9036c102008-10-30 14:19:41 -04005423 em->len = found_key.offset - start;
5424 goto not_found_em;
5425 }
5426
Yan Zhengd899e052008-10-30 14:25:28 -04005427 if (found_type == BTRFS_FILE_EXTENT_REG ||
5428 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04005429 em->start = extent_start;
5430 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005431 em->orig_start = extent_start -
5432 btrfs_file_extent_offset(leaf, item);
Josef Bacikb4939682012-12-03 10:31:19 -05005433 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5434 item);
Chris Masondb945352007-10-15 16:15:53 -04005435 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5436 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04005437 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04005438 goto insert;
5439 }
Li Zefan261507a02010-12-17 14:21:50 +08005440 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005441 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005442 em->compress_type = compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04005443 em->block_start = bytenr;
Josef Bacikb4939682012-12-03 10:31:19 -05005444 em->block_len = em->orig_block_len;
Chris Masonc8b97812008-10-29 14:49:59 -04005445 } else {
5446 bytenr += btrfs_file_extent_offset(leaf, item);
5447 em->block_start = bytenr;
5448 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04005449 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5450 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04005451 }
Chris Masona52d9a82007-08-27 16:49:44 -04005452 goto insert;
5453 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04005454 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04005455 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04005456 size_t size;
5457 size_t extent_offset;
5458 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04005459
Chris Masona52d9a82007-08-27 16:49:44 -04005460 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04005461 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04005462 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04005463 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04005464 goto out;
5465 }
5466
Yan Zheng9036c102008-10-30 14:19:41 -04005467 size = btrfs_file_extent_inline_len(leaf, item);
5468 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05005469 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04005470 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04005471 em->start = extent_start + extent_offset;
Chris Mason70dec802008-01-29 09:59:12 -05005472 em->len = (copy_size + root->sectorsize - 1) &
5473 ~((u64)root->sectorsize - 1);
Josef Bacikb4939682012-12-03 10:31:19 -05005474 em->orig_block_len = em->len;
Josef Bacik70c8a912012-10-11 16:54:30 -04005475 em->orig_start = em->start;
Li Zefan261507a02010-12-17 14:21:50 +08005476 if (compress_type) {
Chris Masonc8b97812008-10-29 14:49:59 -04005477 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005478 em->compress_type = compress_type;
5479 }
Yan689f9342007-10-29 11:41:07 -04005480 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04005481 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08005482 if (btrfs_file_extent_compression(leaf, item) !=
5483 BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005484 ret = uncompress_inline(path, inode, page,
5485 pg_offset,
5486 extent_offset, item);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005487 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -04005488 } else {
5489 map = kmap(page);
5490 read_extent_buffer(leaf, map + pg_offset, ptr,
5491 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04005492 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5493 memset(map + pg_offset + copy_size, 0,
5494 PAGE_CACHE_SIZE - pg_offset -
5495 copy_size);
5496 }
Chris Masonc8b97812008-10-29 14:49:59 -04005497 kunmap(page);
5498 }
Chris Mason179e29e2007-11-01 11:28:41 -04005499 flush_dcache_page(page);
5500 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e0802011-12-01 14:35:19 +01005501 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04005502 if (!trans) {
5503 kunmap(page);
5504 free_extent_map(em);
5505 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04005506
David Sterbab3b4aa72011-04-21 01:20:15 +02005507 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005508 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04005509
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005510 if (IS_ERR(trans))
5511 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04005512 goto again;
5513 }
Chris Masonc8b97812008-10-29 14:49:59 -04005514 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05005515 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04005516 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04005517 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04005518 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04005519 }
Chris Masond1310b22008-01-24 16:13:08 -05005520 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00005521 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04005522 goto insert;
5523 } else {
Julia Lawall31b1a2b2012-11-03 10:58:34 +00005524 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04005525 }
5526not_found:
5527 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04005528 em->orig_start = start;
Chris Masond1310b22008-01-24 16:13:08 -05005529 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04005530not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04005531 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04005532 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04005533insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02005534 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05005535 if (em->start > start || extent_map_end(em) <= start) {
Chris Masond3977122009-01-05 21:25:51 -05005536 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5537 "[%llu %llu]\n", (unsigned long long)em->start,
5538 (unsigned long long)em->len,
5539 (unsigned long long)start,
5540 (unsigned long long)len);
Chris Masona52d9a82007-08-27 16:49:44 -04005541 err = -EIO;
5542 goto out;
5543 }
Chris Masond1310b22008-01-24 16:13:08 -05005544
5545 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04005546 write_lock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005547 ret = add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04005548 /* it is possible that someone inserted the extent into the tree
5549 * while we had the lock dropped. It is also possible that
5550 * an overlapping map exists in the tree
5551 */
Chris Masona52d9a82007-08-27 16:49:44 -04005552 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04005553 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005554
5555 ret = 0;
5556
Chris Mason3b951512008-04-17 11:29:12 -04005557 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04005558 if (existing && (existing->start > start ||
5559 existing->start + existing->len <= start)) {
5560 free_extent_map(existing);
5561 existing = NULL;
5562 }
Chris Mason3b951512008-04-17 11:29:12 -04005563 if (!existing) {
5564 existing = lookup_extent_mapping(em_tree, em->start,
5565 em->len);
5566 if (existing) {
5567 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005568 em, start,
5569 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04005570 free_extent_map(existing);
5571 if (err) {
5572 free_extent_map(em);
5573 em = NULL;
5574 }
5575 } else {
5576 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04005577 free_extent_map(em);
5578 em = NULL;
5579 }
5580 } else {
5581 free_extent_map(em);
5582 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005583 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04005584 }
Chris Masona52d9a82007-08-27 16:49:44 -04005585 }
Chris Mason890871b2009-09-02 16:24:52 -04005586 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005587out:
liubo1abe9b82011-03-24 11:18:59 +00005588
Tsutomu Itohf0bd95e2012-10-01 03:08:37 -06005589 if (em)
5590 trace_btrfs_get_extent(root, em);
liubo1abe9b82011-03-24 11:18:59 +00005591
Chris Masonf4219502008-07-22 11:18:09 -04005592 if (path)
5593 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04005594 if (trans) {
5595 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05005596 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04005597 err = ret;
5598 }
Chris Masona52d9a82007-08-27 16:49:44 -04005599 if (err) {
5600 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04005601 return ERR_PTR(err);
5602 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005603 BUG_ON(!em); /* Error is always set */
Chris Masona52d9a82007-08-27 16:49:44 -04005604 return em;
5605}
5606
Chris Masonec29ed52011-02-23 16:23:20 -05005607struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5608 size_t pg_offset, u64 start, u64 len,
5609 int create)
5610{
5611 struct extent_map *em;
5612 struct extent_map *hole_em = NULL;
5613 u64 range_start = start;
5614 u64 end;
5615 u64 found;
5616 u64 found_end;
5617 int err = 0;
5618
5619 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5620 if (IS_ERR(em))
5621 return em;
5622 if (em) {
5623 /*
Liu Bof9e4fb52013-01-07 10:10:12 +00005624 * if our em maps to
5625 * - a hole or
5626 * - a pre-alloc extent,
5627 * there might actually be delalloc bytes behind it.
Chris Masonec29ed52011-02-23 16:23:20 -05005628 */
Liu Bof9e4fb52013-01-07 10:10:12 +00005629 if (em->block_start != EXTENT_MAP_HOLE &&
5630 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
Chris Masonec29ed52011-02-23 16:23:20 -05005631 return em;
5632 else
5633 hole_em = em;
5634 }
5635
5636 /* check to see if we've wrapped (len == -1 or similar) */
5637 end = start + len;
5638 if (end < start)
5639 end = (u64)-1;
5640 else
5641 end -= 1;
5642
5643 em = NULL;
5644
5645 /* ok, we didn't find anything, lets look for delalloc */
5646 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5647 end, len, EXTENT_DELALLOC, 1);
5648 found_end = range_start + found;
5649 if (found_end < range_start)
5650 found_end = (u64)-1;
5651
5652 /*
5653 * we didn't find anything useful, return
5654 * the original results from get_extent()
5655 */
5656 if (range_start > end || found_end <= start) {
5657 em = hole_em;
5658 hole_em = NULL;
5659 goto out;
5660 }
5661
5662 /* adjust the range_start to make sure it doesn't
5663 * go backwards from the start they passed in
5664 */
5665 range_start = max(start,range_start);
5666 found = found_end - range_start;
5667
5668 if (found > 0) {
5669 u64 hole_start = start;
5670 u64 hole_len = len;
5671
David Sterba172ddd62011-04-21 00:48:27 +02005672 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05005673 if (!em) {
5674 err = -ENOMEM;
5675 goto out;
5676 }
5677 /*
5678 * when btrfs_get_extent can't find anything it
5679 * returns one huge hole
5680 *
5681 * make sure what it found really fits our range, and
5682 * adjust to make sure it is based on the start from
5683 * the caller
5684 */
5685 if (hole_em) {
5686 u64 calc_end = extent_map_end(hole_em);
5687
5688 if (calc_end <= start || (hole_em->start > end)) {
5689 free_extent_map(hole_em);
5690 hole_em = NULL;
5691 } else {
5692 hole_start = max(hole_em->start, start);
5693 hole_len = calc_end - hole_start;
5694 }
5695 }
5696 em->bdev = NULL;
5697 if (hole_em && range_start > hole_start) {
5698 /* our hole starts before our delalloc, so we
5699 * have to return just the parts of the hole
5700 * that go until the delalloc starts
5701 */
5702 em->len = min(hole_len,
5703 range_start - hole_start);
5704 em->start = hole_start;
5705 em->orig_start = hole_start;
5706 /*
5707 * don't adjust block start at all,
5708 * it is fixed at EXTENT_MAP_HOLE
5709 */
5710 em->block_start = hole_em->block_start;
5711 em->block_len = hole_len;
Liu Bof9e4fb52013-01-07 10:10:12 +00005712 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
5713 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonec29ed52011-02-23 16:23:20 -05005714 } else {
5715 em->start = range_start;
5716 em->len = found;
5717 em->orig_start = range_start;
5718 em->block_start = EXTENT_MAP_DELALLOC;
5719 em->block_len = found;
5720 }
5721 } else if (hole_em) {
5722 return hole_em;
5723 }
5724out:
5725
5726 free_extent_map(hole_em);
5727 if (err) {
5728 free_extent_map(em);
5729 return ERR_PTR(err);
5730 }
5731 return em;
5732}
5733
Josef Bacik4b46fce2010-05-23 11:00:55 -04005734static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5735 u64 start, u64 len)
5736{
5737 struct btrfs_root *root = BTRFS_I(inode)->root;
5738 struct btrfs_trans_handle *trans;
Josef Bacik70c8a912012-10-11 16:54:30 -04005739 struct extent_map *em;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005740 struct btrfs_key ins;
5741 u64 alloc_hint;
5742 int ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005743
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005744 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005745 if (IS_ERR(trans))
5746 return ERR_CAST(trans);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005747
5748 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5749
5750 alloc_hint = get_extent_allocation_hint(inode, start, len);
5751 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
Josef Bacik81c9ad22012-01-18 10:56:06 -05005752 alloc_hint, &ins, 1);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005753 if (ret) {
5754 em = ERR_PTR(ret);
5755 goto out;
5756 }
5757
Josef Bacik70c8a912012-10-11 16:54:30 -04005758 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5759 ins.offset, ins.offset, 0);
5760 if (IS_ERR(em))
5761 goto out;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005762
5763 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5764 ins.offset, ins.offset, 0);
5765 if (ret) {
5766 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5767 em = ERR_PTR(ret);
5768 }
5769out:
5770 btrfs_end_transaction(trans, root);
5771 return em;
5772}
5773
Chris Mason46bfbb52010-05-26 11:04:10 -04005774/*
5775 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5776 * block must be cow'd
5777 */
5778static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5779 struct inode *inode, u64 offset, u64 len)
5780{
5781 struct btrfs_path *path;
5782 int ret;
5783 struct extent_buffer *leaf;
5784 struct btrfs_root *root = BTRFS_I(inode)->root;
5785 struct btrfs_file_extent_item *fi;
5786 struct btrfs_key key;
5787 u64 disk_bytenr;
5788 u64 backref_offset;
5789 u64 extent_end;
5790 u64 num_bytes;
5791 int slot;
5792 int found_type;
5793
5794 path = btrfs_alloc_path();
5795 if (!path)
5796 return -ENOMEM;
5797
Li Zefan33345d012011-04-20 10:31:50 +08005798 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04005799 offset, 0);
5800 if (ret < 0)
5801 goto out;
5802
5803 slot = path->slots[0];
5804 if (ret == 1) {
5805 if (slot == 0) {
5806 /* can't find the item, must cow */
5807 ret = 0;
5808 goto out;
5809 }
5810 slot--;
5811 }
5812 ret = 0;
5813 leaf = path->nodes[0];
5814 btrfs_item_key_to_cpu(leaf, &key, slot);
Li Zefan33345d012011-04-20 10:31:50 +08005815 if (key.objectid != btrfs_ino(inode) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04005816 key.type != BTRFS_EXTENT_DATA_KEY) {
5817 /* not our file or wrong item type, must cow */
5818 goto out;
5819 }
5820
5821 if (key.offset > offset) {
5822 /* Wrong offset, must cow */
5823 goto out;
5824 }
5825
5826 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5827 found_type = btrfs_file_extent_type(leaf, fi);
5828 if (found_type != BTRFS_FILE_EXTENT_REG &&
5829 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5830 /* not a regular extent, must cow */
5831 goto out;
5832 }
5833 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5834 backref_offset = btrfs_file_extent_offset(leaf, fi);
5835
5836 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5837 if (extent_end < offset + len) {
5838 /* extent doesn't include our full range, must cow */
5839 goto out;
5840 }
5841
5842 if (btrfs_extent_readonly(root, disk_bytenr))
5843 goto out;
5844
5845 /*
5846 * look for other files referencing this extent, if we
5847 * find any we must cow
5848 */
Li Zefan33345d012011-04-20 10:31:50 +08005849 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04005850 key.offset - backref_offset, disk_bytenr))
5851 goto out;
5852
5853 /*
5854 * adjust disk_bytenr and num_bytes to cover just the bytes
5855 * in this extent we are about to write. If there
5856 * are any csums in that range we have to cow in order
5857 * to keep the csums correct
5858 */
5859 disk_bytenr += backref_offset;
5860 disk_bytenr += offset - key.offset;
5861 num_bytes = min(offset + len, extent_end) - offset;
5862 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5863 goto out;
5864 /*
5865 * all of the above have passed, it is safe to overwrite this extent
5866 * without cow
5867 */
5868 ret = 1;
5869out:
5870 btrfs_free_path(path);
5871 return ret;
5872}
5873
Josef Bacikeb838e72012-07-31 16:28:48 -04005874static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5875 struct extent_state **cached_state, int writing)
5876{
5877 struct btrfs_ordered_extent *ordered;
5878 int ret = 0;
5879
5880 while (1) {
5881 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5882 0, cached_state);
5883 /*
5884 * We're concerned with the entire range that we're going to be
5885 * doing DIO to, so we need to make sure theres no ordered
5886 * extents in this range.
5887 */
5888 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5889 lockend - lockstart + 1);
5890
5891 /*
5892 * We need to make sure there are no buffered pages in this
5893 * range either, we could have raced between the invalidate in
5894 * generic_file_direct_write and locking the extent. The
5895 * invalidate needs to happen so that reads after a write do not
5896 * get stale data.
5897 */
5898 if (!ordered && (!writing ||
5899 !test_range_bit(&BTRFS_I(inode)->io_tree,
5900 lockstart, lockend, EXTENT_UPTODATE, 0,
5901 *cached_state)))
5902 break;
5903
5904 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5905 cached_state, GFP_NOFS);
5906
5907 if (ordered) {
5908 btrfs_start_ordered_extent(inode, ordered, 1);
5909 btrfs_put_ordered_extent(ordered);
5910 } else {
5911 /* Screw you mmap */
5912 ret = filemap_write_and_wait_range(inode->i_mapping,
5913 lockstart,
5914 lockend);
5915 if (ret)
5916 break;
5917
5918 /*
5919 * If we found a page that couldn't be invalidated just
5920 * fall back to buffered.
5921 */
5922 ret = invalidate_inode_pages2_range(inode->i_mapping,
5923 lockstart >> PAGE_CACHE_SHIFT,
5924 lockend >> PAGE_CACHE_SHIFT);
5925 if (ret)
5926 break;
5927 }
5928
5929 cond_resched();
5930 }
5931
5932 return ret;
5933}
5934
Josef Bacik69ffb542012-09-11 15:40:07 -04005935static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5936 u64 len, u64 orig_start,
5937 u64 block_start, u64 block_len,
Josef Bacikb4939682012-12-03 10:31:19 -05005938 u64 orig_block_len, int type)
Josef Bacik69ffb542012-09-11 15:40:07 -04005939{
5940 struct extent_map_tree *em_tree;
5941 struct extent_map *em;
5942 struct btrfs_root *root = BTRFS_I(inode)->root;
5943 int ret;
5944
5945 em_tree = &BTRFS_I(inode)->extent_tree;
5946 em = alloc_extent_map();
5947 if (!em)
5948 return ERR_PTR(-ENOMEM);
5949
5950 em->start = start;
5951 em->orig_start = orig_start;
5952 em->len = len;
5953 em->block_len = block_len;
5954 em->block_start = block_start;
5955 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacikb4939682012-12-03 10:31:19 -05005956 em->orig_block_len = orig_block_len;
Josef Bacik70c8a912012-10-11 16:54:30 -04005957 em->generation = -1;
Josef Bacik69ffb542012-09-11 15:40:07 -04005958 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5959 if (type == BTRFS_ORDERED_PREALLOC)
Josef Bacikb11e2342012-12-03 10:58:15 -05005960 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik69ffb542012-09-11 15:40:07 -04005961
5962 do {
5963 btrfs_drop_extent_cache(inode, em->start,
5964 em->start + em->len - 1, 0);
5965 write_lock(&em_tree->lock);
5966 ret = add_extent_mapping(em_tree, em);
Josef Bacik70c8a912012-10-11 16:54:30 -04005967 if (!ret)
5968 list_move(&em->list,
5969 &em_tree->modified_extents);
Josef Bacik69ffb542012-09-11 15:40:07 -04005970 write_unlock(&em_tree->lock);
5971 } while (ret == -EEXIST);
5972
5973 if (ret) {
5974 free_extent_map(em);
5975 return ERR_PTR(ret);
5976 }
5977
5978 return em;
5979}
5980
5981
Josef Bacik4b46fce2010-05-23 11:00:55 -04005982static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5983 struct buffer_head *bh_result, int create)
5984{
5985 struct extent_map *em;
5986 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikeb838e72012-07-31 16:28:48 -04005987 struct extent_state *cached_state = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005988 u64 start = iblock << inode->i_blkbits;
Josef Bacikeb838e72012-07-31 16:28:48 -04005989 u64 lockstart, lockend;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005990 u64 len = bh_result->b_size;
Chris Mason46bfbb52010-05-26 11:04:10 -04005991 struct btrfs_trans_handle *trans;
Josef Bacikeb838e72012-07-31 16:28:48 -04005992 int unlock_bits = EXTENT_LOCKED;
5993 int ret;
5994
Josef Bacikeb838e72012-07-31 16:28:48 -04005995 if (create) {
5996 ret = btrfs_delalloc_reserve_space(inode, len);
5997 if (ret)
5998 return ret;
5999 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
Josef Bacikc3298612012-08-03 16:49:19 -04006000 } else {
6001 len = min_t(u64, len, root->sectorsize);
Josef Bacikeb838e72012-07-31 16:28:48 -04006002 }
6003
Josef Bacikc3298612012-08-03 16:49:19 -04006004 lockstart = start;
6005 lockend = start + len - 1;
6006
Josef Bacikeb838e72012-07-31 16:28:48 -04006007 /*
6008 * If this errors out it's because we couldn't invalidate pagecache for
6009 * this range and we need to fallback to buffered.
6010 */
6011 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6012 return -ENOTBLK;
6013
6014 if (create) {
6015 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6016 lockend, EXTENT_DELALLOC, NULL,
6017 &cached_state, GFP_NOFS);
6018 if (ret)
6019 goto unlock_err;
6020 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006021
6022 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
Josef Bacikeb838e72012-07-31 16:28:48 -04006023 if (IS_ERR(em)) {
6024 ret = PTR_ERR(em);
6025 goto unlock_err;
6026 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006027
6028 /*
6029 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6030 * io. INLINE is special, and we could probably kludge it in here, but
6031 * it's still buffered so for safety lets just fall back to the generic
6032 * buffered path.
6033 *
6034 * For COMPRESSED we _have_ to read the entire extent in so we can
6035 * decompress it, so there will be buffering required no matter what we
6036 * do, so go ahead and fallback to buffered.
6037 *
6038 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6039 * to buffered IO. Don't blame me, this is the price we pay for using
6040 * the generic code.
6041 */
6042 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6043 em->block_start == EXTENT_MAP_INLINE) {
6044 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04006045 ret = -ENOTBLK;
6046 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006047 }
6048
6049 /* Just a good old fashioned hole, return */
6050 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6051 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6052 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04006053 ret = 0;
6054 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006055 }
6056
6057 /*
6058 * We don't allocate a new extent in the following cases
6059 *
6060 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6061 * existing extent.
6062 * 2) The extent is marked as PREALLOC. We're good to go here and can
6063 * just use the extent.
6064 *
6065 */
Chris Mason46bfbb52010-05-26 11:04:10 -04006066 if (!create) {
Josef Bacikeb838e72012-07-31 16:28:48 -04006067 len = min(len, em->len - (start - em->start));
6068 lockstart = start + len;
6069 goto unlock;
Chris Mason46bfbb52010-05-26 11:04:10 -04006070 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006071
6072 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6073 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6074 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04006075 int type;
6076 int ret;
Chris Mason46bfbb52010-05-26 11:04:10 -04006077 u64 block_start;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006078
6079 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6080 type = BTRFS_ORDERED_PREALLOC;
6081 else
6082 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04006083 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04006084 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04006085
6086 /*
6087 * we're not going to log anything, but we do need
6088 * to make sure the current transaction stays open
6089 * while we look for nocow cross refs
6090 */
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006091 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00006092 if (IS_ERR(trans))
Chris Mason46bfbb52010-05-26 11:04:10 -04006093 goto must_cow;
6094
6095 if (can_nocow_odirect(trans, inode, start, len) == 1) {
Josef Bacik70c8a912012-10-11 16:54:30 -04006096 u64 orig_start = em->orig_start;
Josef Bacikb4939682012-12-03 10:31:19 -05006097 u64 orig_block_len = em->orig_block_len;
Josef Bacik69ffb542012-09-11 15:40:07 -04006098
6099 if (type == BTRFS_ORDERED_PREALLOC) {
6100 free_extent_map(em);
6101 em = create_pinned_em(inode, start, len,
6102 orig_start,
Josef Bacikb4939682012-12-03 10:31:19 -05006103 block_start, len,
6104 orig_block_len, type);
Josef Bacik69ffb542012-09-11 15:40:07 -04006105 if (IS_ERR(em)) {
6106 btrfs_end_transaction(trans, root);
6107 goto unlock_err;
6108 }
6109 }
6110
Chris Mason46bfbb52010-05-26 11:04:10 -04006111 ret = btrfs_add_ordered_extent_dio(inode, start,
6112 block_start, len, len, type);
6113 btrfs_end_transaction(trans, root);
6114 if (ret) {
6115 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04006116 goto unlock_err;
Chris Mason46bfbb52010-05-26 11:04:10 -04006117 }
6118 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006119 }
Chris Mason46bfbb52010-05-26 11:04:10 -04006120 btrfs_end_transaction(trans, root);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006121 }
Chris Mason46bfbb52010-05-26 11:04:10 -04006122must_cow:
6123 /*
6124 * this will cow the extent, reset the len in case we changed
6125 * it above
6126 */
6127 len = bh_result->b_size;
Josef Bacik70c8a912012-10-11 16:54:30 -04006128 free_extent_map(em);
6129 em = btrfs_new_extent_direct(inode, start, len);
Josef Bacikeb838e72012-07-31 16:28:48 -04006130 if (IS_ERR(em)) {
6131 ret = PTR_ERR(em);
6132 goto unlock_err;
6133 }
Chris Mason46bfbb52010-05-26 11:04:10 -04006134 len = min(len, em->len - (start - em->start));
6135unlock:
Josef Bacik4b46fce2010-05-23 11:00:55 -04006136 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6137 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04006138 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006139 bh_result->b_bdev = em->bdev;
6140 set_buffer_mapped(bh_result);
Josef Bacikc3473e82012-06-19 10:59:00 -04006141 if (create) {
6142 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6143 set_buffer_new(bh_result);
6144
6145 /*
6146 * Need to update the i_size under the extent lock so buffered
6147 * readers will get the updated i_size when we unlock.
6148 */
6149 if (start + len > i_size_read(inode))
6150 i_size_write(inode, start + len);
6151 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006152
Josef Bacikeb838e72012-07-31 16:28:48 -04006153 /*
6154 * In the case of write we need to clear and unlock the entire range,
6155 * in the case of read we need to unlock only the end area that we
6156 * aren't using if there is any left over space.
6157 */
Liu Bo24c03fa2012-08-22 20:10:38 -06006158 if (lockstart < lockend) {
6159 if (create && len < lockend - lockstart) {
6160 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
Liu Bo9e8a4a82012-09-05 19:10:51 -06006161 lockstart + len - 1,
6162 unlock_bits | EXTENT_DEFRAG, 1, 0,
Liu Bo24c03fa2012-08-22 20:10:38 -06006163 &cached_state, GFP_NOFS);
6164 /*
6165 * Beside unlock, we also need to cleanup reserved space
6166 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6167 */
6168 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6169 lockstart + len, lockend,
Liu Bo9e8a4a82012-09-05 19:10:51 -06006170 unlock_bits | EXTENT_DO_ACCOUNTING |
6171 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
Liu Bo24c03fa2012-08-22 20:10:38 -06006172 } else {
6173 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6174 lockend, unlock_bits, 1, 0,
6175 &cached_state, GFP_NOFS);
6176 }
6177 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04006178 free_extent_state(cached_state);
Liu Bo24c03fa2012-08-22 20:10:38 -06006179 }
Josef Bacikeb838e72012-07-31 16:28:48 -04006180
Josef Bacik4b46fce2010-05-23 11:00:55 -04006181 free_extent_map(em);
6182
6183 return 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04006184
6185unlock_err:
6186 if (create)
6187 unlock_bits |= EXTENT_DO_ACCOUNTING;
6188
6189 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6190 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6191 return ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006192}
6193
6194struct btrfs_dio_private {
6195 struct inode *inode;
6196 u64 logical_offset;
6197 u64 disk_bytenr;
6198 u64 bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006199 void *private;
Miao Xiee65e1532010-11-22 03:04:43 +00006200
6201 /* number of bios pending for this dio */
6202 atomic_t pending_bios;
6203
6204 /* IO errors */
6205 int errors;
6206
6207 struct bio *orig_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006208};
6209
6210static void btrfs_endio_direct_read(struct bio *bio, int err)
6211{
Miao Xiee65e1532010-11-22 03:04:43 +00006212 struct btrfs_dio_private *dip = bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006213 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6214 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006215 struct inode *inode = dip->inode;
6216 struct btrfs_root *root = BTRFS_I(inode)->root;
6217 u64 start;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006218
6219 start = dip->logical_offset;
6220 do {
6221 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6222 struct page *page = bvec->bv_page;
6223 char *kaddr;
6224 u32 csum = ~(u32)0;
Josef Bacikc3298612012-08-03 16:49:19 -04006225 u64 private = ~(u32)0;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006226 unsigned long flags;
6227
Josef Bacikc3298612012-08-03 16:49:19 -04006228 if (get_state_private(&BTRFS_I(inode)->io_tree,
6229 start, &private))
6230 goto failed;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006231 local_irq_save(flags);
Cong Wang7ac687d2011-11-25 23:14:28 +08006232 kaddr = kmap_atomic(page);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006233 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6234 csum, bvec->bv_len);
6235 btrfs_csum_final(csum, (char *)&csum);
Cong Wang7ac687d2011-11-25 23:14:28 +08006236 kunmap_atomic(kaddr);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006237 local_irq_restore(flags);
6238
6239 flush_dcache_page(bvec->bv_page);
Josef Bacikc3298612012-08-03 16:49:19 -04006240 if (csum != private) {
6241failed:
Li Zefan33345d012011-04-20 10:31:50 +08006242 printk(KERN_ERR "btrfs csum failed ino %llu off"
Josef Bacik4b46fce2010-05-23 11:00:55 -04006243 " %llu csum %u private %u\n",
Li Zefan33345d012011-04-20 10:31:50 +08006244 (unsigned long long)btrfs_ino(inode),
6245 (unsigned long long)start,
Josef Bacikc3298612012-08-03 16:49:19 -04006246 csum, (unsigned)private);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006247 err = -EIO;
6248 }
6249 }
6250
6251 start += bvec->bv_len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006252 bvec++;
6253 } while (bvec <= bvec_end);
6254
6255 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006256 dip->logical_offset + dip->bytes - 1);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006257 bio->bi_private = dip->private;
6258
Josef Bacik4b46fce2010-05-23 11:00:55 -04006259 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04006260
6261 /* If we had a csum failure make sure to clear the uptodate flag */
6262 if (err)
6263 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006264 dio_end_io(bio, err);
6265}
6266
6267static void btrfs_endio_direct_write(struct bio *bio, int err)
6268{
6269 struct btrfs_dio_private *dip = bio->bi_private;
6270 struct inode *inode = dip->inode;
6271 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006272 struct btrfs_ordered_extent *ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05006273 u64 ordered_offset = dip->logical_offset;
6274 u64 ordered_bytes = dip->bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006275 int ret;
6276
6277 if (err)
6278 goto out_done;
Chris Mason163cf092010-11-28 19:56:33 -05006279again:
6280 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6281 &ordered_offset,
Josef Bacik5fd02042012-05-02 14:00:54 -04006282 ordered_bytes, !err);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006283 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05006284 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006285
Josef Bacik5fd02042012-05-02 14:00:54 -04006286 ordered->work.func = finish_ordered_fn;
6287 ordered->work.flags = 0;
6288 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6289 &ordered->work);
Chris Mason163cf092010-11-28 19:56:33 -05006290out_test:
6291 /*
6292 * our bio might span multiple ordered extents. If we haven't
6293 * completed the accounting for the whole dio, go back and try again
6294 */
6295 if (ordered_offset < dip->logical_offset + dip->bytes) {
6296 ordered_bytes = dip->logical_offset + dip->bytes -
6297 ordered_offset;
Josef Bacik5fd02042012-05-02 14:00:54 -04006298 ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05006299 goto again;
6300 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006301out_done:
6302 bio->bi_private = dip->private;
6303
Josef Bacik4b46fce2010-05-23 11:00:55 -04006304 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04006305
6306 /* If we had an error make sure to clear the uptodate flag */
6307 if (err)
6308 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006309 dio_end_io(bio, err);
6310}
6311
Chris Masoneaf25d92010-05-25 09:48:28 -04006312static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6313 struct bio *bio, int mirror_num,
6314 unsigned long bio_flags, u64 offset)
6315{
6316 int ret;
6317 struct btrfs_root *root = BTRFS_I(inode)->root;
6318 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006319 BUG_ON(ret); /* -ENOMEM */
Chris Masoneaf25d92010-05-25 09:48:28 -04006320 return 0;
6321}
6322
Miao Xiee65e1532010-11-22 03:04:43 +00006323static void btrfs_end_dio_bio(struct bio *bio, int err)
6324{
6325 struct btrfs_dio_private *dip = bio->bi_private;
6326
6327 if (err) {
Li Zefan33345d012011-04-20 10:31:50 +08006328 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
Jan Beulich3dd14622010-12-07 14:54:09 +00006329 "sector %#Lx len %u err no %d\n",
Li Zefan33345d012011-04-20 10:31:50 +08006330 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
Jan Beulich3dd14622010-12-07 14:54:09 +00006331 (unsigned long long)bio->bi_sector, bio->bi_size, err);
Miao Xiee65e1532010-11-22 03:04:43 +00006332 dip->errors = 1;
6333
6334 /*
6335 * before atomic variable goto zero, we must make sure
6336 * dip->errors is perceived to be set.
6337 */
6338 smp_mb__before_atomic_dec();
6339 }
6340
6341 /* if there are more bios still pending for this dio, just exit */
6342 if (!atomic_dec_and_test(&dip->pending_bios))
6343 goto out;
6344
6345 if (dip->errors)
6346 bio_io_error(dip->orig_bio);
6347 else {
6348 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6349 bio_endio(dip->orig_bio, 0);
6350 }
6351out:
6352 bio_put(bio);
6353}
6354
6355static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6356 u64 first_sector, gfp_t gfp_flags)
6357{
6358 int nr_vecs = bio_get_nr_vecs(bdev);
6359 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6360}
6361
6362static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6363 int rw, u64 file_offset, int skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04006364 int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00006365{
6366 int write = rw & REQ_WRITE;
6367 struct btrfs_root *root = BTRFS_I(inode)->root;
6368 int ret;
6369
Josef Bacikb812ce22012-11-16 13:56:32 -05006370 if (async_submit)
6371 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6372
Miao Xiee65e1532010-11-22 03:04:43 +00006373 bio_get(bio);
Josef Bacik5fd02042012-05-02 14:00:54 -04006374
6375 if (!write) {
6376 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6377 if (ret)
6378 goto err;
6379 }
Miao Xiee65e1532010-11-22 03:04:43 +00006380
Josef Bacik1ae39932011-04-06 14:41:34 -04006381 if (skip_sum)
6382 goto map;
6383
6384 if (write && async_submit) {
Miao Xiee65e1532010-11-22 03:04:43 +00006385 ret = btrfs_wq_submit_bio(root->fs_info,
6386 inode, rw, bio, 0, 0,
6387 file_offset,
6388 __btrfs_submit_bio_start_direct_io,
6389 __btrfs_submit_bio_done);
6390 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04006391 } else if (write) {
6392 /*
6393 * If we aren't doing async submit, calculate the csum of the
6394 * bio now.
6395 */
6396 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6397 if (ret)
6398 goto err;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00006399 } else if (!skip_sum) {
Josef Bacikc3298612012-08-03 16:49:19 -04006400 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00006401 if (ret)
6402 goto err;
6403 }
Miao Xiee65e1532010-11-22 03:04:43 +00006404
Josef Bacik1ae39932011-04-06 14:41:34 -04006405map:
6406 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006407err:
6408 bio_put(bio);
6409 return ret;
6410}
6411
6412static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6413 int skip_sum)
6414{
6415 struct inode *inode = dip->inode;
6416 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xiee65e1532010-11-22 03:04:43 +00006417 struct bio *bio;
6418 struct bio *orig_bio = dip->orig_bio;
6419 struct bio_vec *bvec = orig_bio->bi_io_vec;
6420 u64 start_sector = orig_bio->bi_sector;
6421 u64 file_offset = dip->logical_offset;
6422 u64 submit_len = 0;
6423 u64 map_length;
6424 int nr_pages = 0;
Miao Xiee65e1532010-11-22 03:04:43 +00006425 int ret = 0;
Josef Bacik1ae39932011-04-06 14:41:34 -04006426 int async_submit = 0;
Miao Xiee65e1532010-11-22 03:04:43 +00006427
Miao Xiee65e1532010-11-22 03:04:43 +00006428 map_length = orig_bio->bi_size;
Stefan Behrens3ec706c2012-11-05 15:46:42 +01006429 ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00006430 &map_length, NULL, 0);
6431 if (ret) {
Josef Bacik64728bb2011-04-25 19:43:52 -04006432 bio_put(orig_bio);
Miao Xiee65e1532010-11-22 03:04:43 +00006433 return -EIO;
6434 }
6435
Josef Bacik02f57c72011-04-06 14:25:44 -04006436 if (map_length >= orig_bio->bi_size) {
6437 bio = orig_bio;
6438 goto submit;
6439 }
6440
Josef Bacik1ae39932011-04-06 14:41:34 -04006441 async_submit = 1;
Josef Bacik02f57c72011-04-06 14:25:44 -04006442 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6443 if (!bio)
6444 return -ENOMEM;
6445 bio->bi_private = dip;
6446 bio->bi_end_io = btrfs_end_dio_bio;
6447 atomic_inc(&dip->pending_bios);
6448
Miao Xiee65e1532010-11-22 03:04:43 +00006449 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6450 if (unlikely(map_length < submit_len + bvec->bv_len ||
6451 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6452 bvec->bv_offset) < bvec->bv_len)) {
6453 /*
6454 * inc the count before we submit the bio so
6455 * we know the end IO handler won't happen before
6456 * we inc the count. Otherwise, the dip might get freed
6457 * before we're done setting it up
6458 */
6459 atomic_inc(&dip->pending_bios);
6460 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6461 file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04006462 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006463 if (ret) {
6464 bio_put(bio);
6465 atomic_dec(&dip->pending_bios);
6466 goto out_err;
6467 }
6468
Miao Xiee65e1532010-11-22 03:04:43 +00006469 start_sector += submit_len >> 9;
6470 file_offset += submit_len;
6471
6472 submit_len = 0;
6473 nr_pages = 0;
6474
6475 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6476 start_sector, GFP_NOFS);
6477 if (!bio)
6478 goto out_err;
6479 bio->bi_private = dip;
6480 bio->bi_end_io = btrfs_end_dio_bio;
6481
6482 map_length = orig_bio->bi_size;
Stefan Behrens3ec706c2012-11-05 15:46:42 +01006483 ret = btrfs_map_block(root->fs_info, READ,
6484 start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00006485 &map_length, NULL, 0);
6486 if (ret) {
6487 bio_put(bio);
6488 goto out_err;
6489 }
6490 } else {
6491 submit_len += bvec->bv_len;
6492 nr_pages ++;
6493 bvec++;
6494 }
6495 }
6496
Josef Bacik02f57c72011-04-06 14:25:44 -04006497submit:
Miao Xiee65e1532010-11-22 03:04:43 +00006498 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04006499 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006500 if (!ret)
6501 return 0;
6502
6503 bio_put(bio);
6504out_err:
6505 dip->errors = 1;
6506 /*
6507 * before atomic variable goto zero, we must
6508 * make sure dip->errors is perceived to be set.
6509 */
6510 smp_mb__before_atomic_dec();
6511 if (atomic_dec_and_test(&dip->pending_bios))
6512 bio_io_error(dip->orig_bio);
6513
6514 /* bio_end_io() will handle error, so we needn't return it */
6515 return 0;
6516}
6517
Josef Bacik4b46fce2010-05-23 11:00:55 -04006518static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6519 loff_t file_offset)
6520{
6521 struct btrfs_root *root = BTRFS_I(inode)->root;
6522 struct btrfs_dio_private *dip;
6523 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006524 int skip_sum;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02006525 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006526 int ret = 0;
6527
6528 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6529
6530 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6531 if (!dip) {
6532 ret = -ENOMEM;
6533 goto free_ordered;
6534 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006535
6536 dip->private = bio->bi_private;
6537 dip->inode = inode;
6538 dip->logical_offset = file_offset;
6539
Josef Bacik4b46fce2010-05-23 11:00:55 -04006540 dip->bytes = 0;
6541 do {
6542 dip->bytes += bvec->bv_len;
6543 bvec++;
6544 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6545
Chris Mason46bfbb52010-05-26 11:04:10 -04006546 dip->disk_bytenr = (u64)bio->bi_sector << 9;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006547 bio->bi_private = dip;
Miao Xiee65e1532010-11-22 03:04:43 +00006548 dip->errors = 0;
6549 dip->orig_bio = bio;
6550 atomic_set(&dip->pending_bios, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006551
6552 if (write)
6553 bio->bi_end_io = btrfs_endio_direct_write;
6554 else
6555 bio->bi_end_io = btrfs_endio_direct_read;
6556
Miao Xiee65e1532010-11-22 03:04:43 +00006557 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6558 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04006559 return;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006560free_ordered:
6561 /*
6562 * If this is a write, we need to clean up the reserved space and kill
6563 * the ordered extent.
6564 */
6565 if (write) {
6566 struct btrfs_ordered_extent *ordered;
Josef Bacik955256f2010-11-19 09:41:10 -05006567 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006568 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6569 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6570 btrfs_free_reserved_extent(root, ordered->start,
6571 ordered->disk_len);
6572 btrfs_put_ordered_extent(ordered);
6573 btrfs_put_ordered_extent(ordered);
6574 }
6575 bio_endio(bio, ret);
6576}
6577
Chris Mason5a5f79b2010-05-26 21:33:37 -04006578static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6579 const struct iovec *iov, loff_t offset,
6580 unsigned long nr_segs)
6581{
6582 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00006583 int i;
Chris Mason5a5f79b2010-05-26 21:33:37 -04006584 size_t size;
6585 unsigned long addr;
6586 unsigned blocksize_mask = root->sectorsize - 1;
6587 ssize_t retval = -EINVAL;
6588 loff_t end = offset;
6589
6590 if (offset & blocksize_mask)
6591 goto out;
6592
6593 /* Check the memory alignment. Blocks cannot straddle pages */
6594 for (seg = 0; seg < nr_segs; seg++) {
6595 addr = (unsigned long)iov[seg].iov_base;
6596 size = iov[seg].iov_len;
6597 end += size;
Josef Bacika1b75f72011-04-08 15:51:18 +00006598 if ((addr & blocksize_mask) || (size & blocksize_mask))
Chris Mason5a5f79b2010-05-26 21:33:37 -04006599 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00006600
6601 /* If this is a write we don't need to check anymore */
6602 if (rw & WRITE)
6603 continue;
6604
6605 /*
6606 * Check to make sure we don't have duplicate iov_base's in this
6607 * iovec, if so return EINVAL, otherwise we'll get csum errors
6608 * when reading back.
6609 */
6610 for (i = seg + 1; i < nr_segs; i++) {
6611 if (iov[seg].iov_base == iov[i].iov_base)
6612 goto out;
6613 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04006614 }
6615 retval = 0;
6616out:
6617 return retval;
6618}
Josef Bacikeb838e72012-07-31 16:28:48 -04006619
Chris Mason16432982008-04-10 10:23:21 -04006620static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6621 const struct iovec *iov, loff_t offset,
6622 unsigned long nr_segs)
6623{
Josef Bacik4b46fce2010-05-23 11:00:55 -04006624 struct file *file = iocb->ki_filp;
6625 struct inode *inode = file->f_mapping->host;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006626
Chris Mason5a5f79b2010-05-26 21:33:37 -04006627 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
Josef Bacikeb838e72012-07-31 16:28:48 -04006628 offset, nr_segs))
Chris Mason5a5f79b2010-05-26 21:33:37 -04006629 return 0;
Chris Mason5a5f79b2010-05-26 21:33:37 -04006630
Josef Bacikeb838e72012-07-31 16:28:48 -04006631 return __blockdev_direct_IO(rw, iocb, inode,
Chris Mason5a5f79b2010-05-26 21:33:37 -04006632 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6633 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6634 btrfs_submit_direct, 0);
Chris Mason16432982008-04-10 10:23:21 -04006635}
6636
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00006637#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
6638
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006639static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6640 __u64 start, __u64 len)
6641{
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00006642 int ret;
6643
6644 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6645 if (ret)
6646 return ret;
6647
Chris Masonec29ed52011-02-23 16:23:20 -05006648 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006649}
6650
Chris Mason9ebefb182007-06-15 13:50:00 -04006651int btrfs_readpage(struct file *file, struct page *page)
6652{
Chris Masond1310b22008-01-24 16:13:08 -05006653 struct extent_io_tree *tree;
6654 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02006655 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006656}
Chris Mason1832a6d2007-12-21 16:27:21 -05006657
Chris Mason39279cc2007-06-12 06:35:45 -04006658static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6659{
Chris Masond1310b22008-01-24 16:13:08 -05006660 struct extent_io_tree *tree;
Chris Masonb888db2b2007-08-27 16:49:44 -04006661
6662
6663 if (current->flags & PF_MEMALLOC) {
6664 redirty_page_for_writepage(wbc, page);
6665 unlock_page(page);
6666 return 0;
6667 }
Chris Masond1310b22008-01-24 16:13:08 -05006668 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006669 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6670}
Chris Mason39279cc2007-06-12 06:35:45 -04006671
Chris Masonf4219502008-07-22 11:18:09 -04006672int btrfs_writepages(struct address_space *mapping,
6673 struct writeback_control *wbc)
Chris Masonb293f02e2007-11-01 19:45:34 -04006674{
Chris Masond1310b22008-01-24 16:13:08 -05006675 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05006676
Chris Masond1310b22008-01-24 16:13:08 -05006677 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f02e2007-11-01 19:45:34 -04006678 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6679}
6680
Chris Mason3ab2fb52007-11-08 10:59:22 -05006681static int
6682btrfs_readpages(struct file *file, struct address_space *mapping,
6683 struct list_head *pages, unsigned nr_pages)
6684{
Chris Masond1310b22008-01-24 16:13:08 -05006685 struct extent_io_tree *tree;
6686 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05006687 return extent_readpages(tree, mapping, pages, nr_pages,
6688 btrfs_get_extent);
6689}
Chris Masone6dcd2d2008-07-17 12:53:50 -04006690static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04006691{
Chris Masond1310b22008-01-24 16:13:08 -05006692 struct extent_io_tree *tree;
6693 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04006694 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006695
Chris Masond1310b22008-01-24 16:13:08 -05006696 tree = &BTRFS_I(page->mapping->host)->io_tree;
6697 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05006698 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006699 if (ret == 1) {
6700 ClearPagePrivate(page);
6701 set_page_private(page, 0);
6702 page_cache_release(page);
6703 }
6704 return ret;
6705}
Chris Mason39279cc2007-06-12 06:35:45 -04006706
Chris Masone6dcd2d2008-07-17 12:53:50 -04006707static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6708{
Chris Mason98509cf2008-09-11 15:51:43 -04006709 if (PageWriteback(page) || PageDirty(page))
6710 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05006711 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006712}
6713
Chris Masona52d9a82007-08-27 16:49:44 -04006714static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6715{
Josef Bacik5fd02042012-05-02 14:00:54 -04006716 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05006717 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006718 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006719 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006720 u64 page_start = page_offset(page);
6721 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006722
Chris Mason8b62b722009-09-02 16:53:46 -04006723 /*
6724 * we have the page locked, so new writeback can't start,
6725 * and the dirty bit won't be cleared while we are here.
6726 *
6727 * Wait for IO on this page so that we can safely clear
6728 * the PagePrivate2 bit and do ordered accounting
6729 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006730 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04006731
Josef Bacik5fd02042012-05-02 14:00:54 -04006732 tree = &BTRFS_I(inode)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006733 if (offset) {
6734 btrfs_releasepage(page, GFP_NOFS);
6735 return;
6736 }
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006737 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
Josef Bacik5fd02042012-05-02 14:00:54 -04006738 ordered = btrfs_lookup_ordered_extent(inode,
Chris Masone6dcd2d2008-07-17 12:53:50 -04006739 page_offset(page));
6740 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04006741 /*
6742 * IO on this page will never be started, so we need
6743 * to account for any ordered extents now
6744 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006745 clear_extent_bit(tree, page_start, page_end,
6746 EXTENT_DIRTY | EXTENT_DELALLOC |
Liu Bo9e8a4a82012-09-05 19:10:51 -06006747 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6748 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04006749 /*
6750 * whoever cleared the private bit is responsible
6751 * for the finish_ordered_io
6752 */
Josef Bacik5fd02042012-05-02 14:00:54 -04006753 if (TestClearPagePrivate2(page) &&
6754 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6755 PAGE_CACHE_SIZE, 1)) {
6756 btrfs_finish_ordered_io(ordered);
Chris Mason8b62b722009-09-02 16:53:46 -04006757 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006758 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00006759 cached_state = NULL;
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006760 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006761 }
6762 clear_extent_bit(tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006763 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
Liu Bo9e8a4a82012-09-05 19:10:51 -06006764 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6765 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006766 __btrfs_releasepage(page, GFP_NOFS);
6767
Chris Mason4a096752008-07-21 10:29:44 -04006768 ClearPageChecked(page);
Chris Mason9ad6b7bc2008-04-18 16:11:30 -04006769 if (PagePrivate(page)) {
Chris Mason9ad6b7bc2008-04-18 16:11:30 -04006770 ClearPagePrivate(page);
6771 set_page_private(page, 0);
6772 page_cache_release(page);
6773 }
Chris Mason39279cc2007-06-12 06:35:45 -04006774}
6775
Chris Mason9ebefb182007-06-15 13:50:00 -04006776/*
6777 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6778 * called from a page fault handler when a page is first dirtied. Hence we must
6779 * be careful to check for EOF conditions here. We set the page up correctly
6780 * for a written page which means we get ENOSPC checking when writing into
6781 * holes and correct delalloc and unwritten extent mapping on filesystems that
6782 * support these features.
6783 *
6784 * We are not allowed to take the i_mutex here so we have to play games to
6785 * protect against truncate races as the page could now be beyond EOF. Because
6786 * vmtruncate() writes the inode size before removing pages, once we have the
6787 * page lock we can determine safely if the page is beyond EOF. If it is not
6788 * beyond EOF, then the page is guaranteed safe against truncation until we
6789 * unlock the page.
6790 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07006791int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04006792{
Nick Pigginc2ec1752009-03-31 15:23:21 -07006793 struct page *page = vmf->page;
Chris Mason6da6aba2007-12-18 16:15:09 -05006794 struct inode *inode = fdentry(vma->vm_file)->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05006795 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006796 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6797 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006798 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006799 char *kaddr;
6800 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04006801 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05006802 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05006803 int reserved = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006804 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006805 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04006806
Jan Karab2b5ef52012-06-12 16:20:45 +02006807 sb_start_pagefault(inode->i_sb);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006808 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05006809 if (!ret) {
Josef Bacike41f9412012-03-26 09:46:47 -04006810 ret = file_update_time(vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05006811 reserved = 1;
6812 }
Nick Piggin56a76f82009-03-31 15:23:23 -07006813 if (ret) {
6814 if (ret == -ENOMEM)
6815 ret = VM_FAULT_OOM;
6816 else /* -ENOSPC, -EIO, etc */
6817 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05006818 if (reserved)
6819 goto out;
6820 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07006821 }
Chris Mason1832a6d2007-12-21 16:27:21 -05006822
Nick Piggin56a76f82009-03-31 15:23:23 -07006823 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006824again:
Chris Mason9ebefb182007-06-15 13:50:00 -04006825 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04006826 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006827 page_start = page_offset(page);
6828 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006829
Chris Mason9ebefb182007-06-15 13:50:00 -04006830 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04006831 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04006832 /* page got truncated out from underneath us */
6833 goto out_unlock;
6834 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006835 wait_on_page_writeback(page);
6836
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006837 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006838 set_page_extent_mapped(page);
6839
Chris Masoneb84ae02008-07-17 13:53:27 -04006840 /*
6841 * we can't set the delalloc bits if there are pending ordered
6842 * extents. Drop our locks and wait for them to finish
6843 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006844 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6845 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006846 unlock_extent_cached(io_tree, page_start, page_end,
6847 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006848 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04006849 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006850 btrfs_put_ordered_extent(ordered);
6851 goto again;
6852 }
6853
Josef Bacikfbf19082009-10-01 17:10:23 -04006854 /*
6855 * XXX - page_mkwrite gets called every time the page is dirtied, even
6856 * if it was already dirty, so for space accounting reasons we need to
6857 * clear any delalloc bits for the range we are fixing to save. There
6858 * is probably a better way to do this, but for now keep consistent with
6859 * prepare_pages in the normal write path.
6860 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00006861 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06006862 EXTENT_DIRTY | EXTENT_DELALLOC |
6863 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00006864 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04006865
Josef Bacik2ac55d42010-02-03 19:33:23 +00006866 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6867 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006868 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006869 unlock_extent_cached(io_tree, page_start, page_end,
6870 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006871 ret = VM_FAULT_SIGBUS;
6872 goto out_unlock;
6873 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006874 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04006875
6876 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04006877 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006878 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04006879 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04006880 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04006881
Chris Masone6dcd2d2008-07-17 12:53:50 -04006882 if (zero_start != PAGE_CACHE_SIZE) {
6883 kaddr = kmap(page);
6884 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6885 flush_dcache_page(page);
6886 kunmap(page);
6887 }
Chris Mason247e7432008-07-17 12:53:51 -04006888 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006889 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04006890 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006891
Chris Mason257c62e2009-10-13 13:21:08 -04006892 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6893 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
Liu Bo46d8bc32012-08-29 01:07:55 -06006894 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
Chris Mason257c62e2009-10-13 13:21:08 -04006895
Josef Bacik2ac55d42010-02-03 19:33:23 +00006896 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04006897
6898out_unlock:
Jan Karab2b5ef52012-06-12 16:20:45 +02006899 if (!ret) {
6900 sb_end_pagefault(inode->i_sb);
Chris Mason50a9b212009-09-11 12:33:12 -04006901 return VM_FAULT_LOCKED;
Jan Karab2b5ef52012-06-12 16:20:45 +02006902 }
Chris Mason9ebefb182007-06-15 13:50:00 -04006903 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05006904out:
Josef Bacikec39e182012-01-12 19:10:12 -05006905 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05006906out_noreserve:
Jan Karab2b5ef52012-06-12 16:20:45 +02006907 sb_end_pagefault(inode->i_sb);
Chris Mason9ebefb182007-06-15 13:50:00 -04006908 return ret;
6909}
6910
Josef Bacika41ad392011-01-31 15:30:16 -05006911static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04006912{
6913 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04006914 struct btrfs_block_rsv *rsv;
Chris Mason39279cc2007-06-12 06:35:45 -04006915 int ret;
Josef Bacik3893e332011-01-31 16:03:11 -05006916 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006917 struct btrfs_trans_handle *trans;
Chris Masondbe674a2008-07-17 12:54:05 -04006918 u64 mask = root->sectorsize - 1;
Josef Bacik07127182011-08-19 10:29:59 -04006919 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04006920
Josef Bacik2aaa6652012-08-29 14:27:18 -04006921 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
Josef Bacik5d5e1032009-10-13 16:46:49 -04006922 if (ret)
Josef Bacika41ad392011-01-31 15:30:16 -05006923 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00006924
Chris Mason4a096752008-07-21 10:29:44 -04006925 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
Yan, Zheng80825102009-11-12 09:35:36 +00006926 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006927
Josef Bacikfcb80c22011-05-03 10:40:22 -04006928 /*
6929 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6930 * 3 things going on here
6931 *
6932 * 1) We need to reserve space for our orphan item and the space to
6933 * delete our orphan item. Lord knows we don't want to have a dangling
6934 * orphan item because we didn't reserve space to remove it.
6935 *
6936 * 2) We need to reserve space to update our inode.
6937 *
6938 * 3) We need to have something to cache all the space that is going to
6939 * be free'd up by the truncate operation, but also have some slack
6940 * space reserved in case it uses space during the truncate (thank you
6941 * very much snapshotting).
6942 *
6943 * And we need these to all be seperate. The fact is we can use alot of
6944 * space doing the truncate, and we have no earthly idea how much space
6945 * we will use, so we need the truncate reservation to be seperate so it
6946 * doesn't end up using space reserved for updating the inode or
6947 * removing the orphan item. We also need to be able to stop the
6948 * transaction and start a new one, which means we need to be able to
6949 * update the inode several times, and we have no idea of knowing how
6950 * many times that will be, so we can't just reserve 1 item for the
6951 * entirety of the opration, so that has to be done seperately as well.
6952 * Then there is the orphan item, which does indeed need to be held on
6953 * to for the whole operation, and we need nobody to touch this reserved
6954 * space except the orphan code.
6955 *
6956 * So that leaves us with
6957 *
6958 * 1) root->orphan_block_rsv - for the orphan deletion.
6959 * 2) rsv - for the truncate reservation, which we will steal from the
6960 * transaction reservation.
6961 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6962 * updating the inode.
6963 */
Miao Xie66d8f3d2012-09-06 04:02:28 -06006964 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006965 if (!rsv)
6966 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04006967 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04006968 rsv->failfast = 1;
Josef Bacikf0cd8462011-03-04 14:37:08 -05006969
Josef Bacik907cbce2011-08-08 13:46:15 -04006970 /*
Josef Bacik07127182011-08-19 10:29:59 -04006971 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04006972 * 1 for updating the inode.
6973 */
Josef Bacikf3fe8202013-01-07 17:03:21 -05006974 trans = btrfs_start_transaction(root, 2);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006975 if (IS_ERR(trans)) {
6976 err = PTR_ERR(trans);
6977 goto out;
6978 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05006979
Josef Bacik907cbce2011-08-08 13:46:15 -04006980 /* Migrate the slack space for the truncate to our reserve */
6981 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6982 min_size);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006983 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05006984
Chris Mason5a3f23d2009-03-31 13:27:11 -04006985 /*
6986 * setattr is responsible for setting the ordered_data_close flag,
6987 * but that is only tested during the last file release. That
6988 * could happen well after the next commit, leaving a great big
6989 * window where new writes may get lost if someone chooses to write
6990 * to this file after truncating to zero
6991 *
6992 * The inode doesn't have any dirty data here, and so if we commit
6993 * this is a noop. If someone immediately starts writing to the inode
6994 * it is very likely we'll catch some of their writes in this
6995 * transaction, and the commit will find this file on the ordered
6996 * data list with good things to send down.
6997 *
6998 * This is a best effort solution, there is still a window where
6999 * using truncate to replace the contents of the file will
7000 * end up with a zero length file after a crash.
7001 */
Josef Bacik72ac3c02012-05-23 14:13:11 -04007002 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7003 &BTRFS_I(inode)->runtime_flags))
Chris Mason5a3f23d2009-03-31 13:27:11 -04007004 btrfs_add_ordered_operation(trans, root, inode);
7005
Josef Bacik5dc562c2012-08-17 13:14:17 -04007006 /*
7007 * So if we truncate and then write and fsync we normally would just
7008 * write the extents that changed, which is a problem if we need to
7009 * first truncate that entire inode. So set this flag so we write out
7010 * all of the extents in the inode to the sync log so we're completely
7011 * safe.
7012 */
7013 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Josef Bacikca7e70f2012-08-27 17:48:15 -04007014 trans->block_rsv = rsv;
Josef Bacik5dc562c2012-08-17 13:14:17 -04007015
Yan, Zheng80825102009-11-12 09:35:36 +00007016 while (1) {
7017 ret = btrfs_truncate_inode_items(trans, root, inode,
7018 inode->i_size,
7019 BTRFS_EXTENT_DATA_KEY);
Josef Bacikca7e70f2012-08-27 17:48:15 -04007020 if (ret != -ENOSPC) {
Josef Bacik3893e332011-01-31 16:03:11 -05007021 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00007022 break;
Josef Bacik3893e332011-01-31 16:03:11 -05007023 }
Chris Mason39279cc2007-06-12 06:35:45 -04007024
Josef Bacikfcb80c22011-05-03 10:40:22 -04007025 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00007026 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05007027 if (ret) {
7028 err = ret;
7029 break;
7030 }
Josef Bacikca7e70f2012-08-27 17:48:15 -04007031
Yan, Zheng80825102009-11-12 09:35:36 +00007032 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00007033 btrfs_btree_balance_dirty(root);
Josef Bacikca7e70f2012-08-27 17:48:15 -04007034
7035 trans = btrfs_start_transaction(root, 2);
7036 if (IS_ERR(trans)) {
7037 ret = err = PTR_ERR(trans);
7038 trans = NULL;
7039 break;
7040 }
7041
7042 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7043 rsv, min_size);
7044 BUG_ON(ret); /* shouldn't happen */
7045 trans->block_rsv = rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00007046 }
7047
7048 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04007049 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00007050 ret = btrfs_orphan_del(trans, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05007051 if (ret)
7052 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00007053 }
7054
Chris Mason917c16b2011-11-08 14:49:59 -05007055 if (trans) {
7056 trans->block_rsv = &root->fs_info->trans_block_rsv;
7057 ret = btrfs_update_inode(trans, root, inode);
7058 if (ret && !err)
7059 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04007060
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007061 ret = btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00007062 btrfs_btree_balance_dirty(root);
Chris Mason917c16b2011-11-08 14:49:59 -05007063 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04007064
7065out:
7066 btrfs_free_block_rsv(root, rsv);
7067
Josef Bacik3893e332011-01-31 16:03:11 -05007068 if (ret && !err)
7069 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05007070
Josef Bacik3893e332011-01-31 16:03:11 -05007071 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04007072}
7073
Sven Wegener3b963622008-06-09 21:57:42 -04007074/*
Chris Masond352ac62008-09-29 15:18:18 -04007075 * create a new subvolume directory/inode (helper for the ioctl).
7076 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05007077int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04007078 struct btrfs_root *new_root, u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04007079{
Chris Mason39279cc2007-06-12 06:35:45 -04007080 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04007081 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007082 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007083
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01007084 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7085 new_dirid, new_dirid,
7086 S_IFDIR | (~current_umask() & S_IRWXUGO),
7087 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04007088 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04007089 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007090 inode->i_op = &btrfs_dir_inode_operations;
7091 inode->i_fop = &btrfs_dir_file_operations;
7092
Miklos Szeredibfe86842011-10-28 14:13:29 +02007093 set_nlink(inode, 1);
Chris Masondbe674a2008-07-17 12:54:05 -04007094 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04007095
Yan, Zheng76dda932009-09-21 16:00:26 -04007096 err = btrfs_update_inode(trans, new_root, inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04007097
Yan, Zheng76dda932009-09-21 16:00:26 -04007098 iput(inode);
Mark Fashehce598972011-07-26 11:32:23 -07007099 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04007100}
7101
Chris Mason39279cc2007-06-12 06:35:45 -04007102struct inode *btrfs_alloc_inode(struct super_block *sb)
7103{
7104 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007105 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04007106
7107 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7108 if (!ei)
7109 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007110
7111 ei->root = NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007112 ei->generation = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04007113 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04007114 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04007115 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007116 ei->delalloc_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007117 ei->disk_i_size = 0;
7118 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04007119 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007120 ei->index_cnt = (u64)-1;
7121 ei->last_unlink_trans = 0;
Liu Bo46d8bc32012-08-29 01:07:55 -06007122 ei->last_log_commit = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007123
Josef Bacik9e0baf62011-07-15 15:16:44 +00007124 spin_lock_init(&ei->lock);
7125 ei->outstanding_extents = 0;
7126 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007127
Josef Bacik72ac3c02012-05-23 14:13:11 -04007128 ei->runtime_flags = 0;
Li Zefan261507a02010-12-17 14:21:50 +08007129 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007130
Miao Xie16cdcec2011-04-22 18:12:22 +08007131 ei->delayed_node = NULL;
7132
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007133 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02007134 extent_map_tree_init(&ei->extent_tree);
David Sterbaf993c882011-04-20 23:35:57 +02007135 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7136 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
Josef Bacik0b32f4b2012-03-13 09:38:00 -04007137 ei->io_tree.track_uptodate = 1;
7138 ei->io_failure_tree.track_uptodate = 1;
Josef Bacikb812ce22012-11-16 13:56:32 -05007139 atomic_set(&ei->sync_writers, 0);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007140 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05007141 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007142 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007143 INIT_LIST_HEAD(&ei->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007144 INIT_LIST_HEAD(&ei->ordered_operations);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007145 RB_CLEAR_NODE(&ei->rb_node);
7146
7147 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04007148}
7149
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11007150static void btrfs_i_callback(struct rcu_head *head)
7151{
7152 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11007153 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7154}
7155
Chris Mason39279cc2007-06-12 06:35:45 -04007156void btrfs_destroy_inode(struct inode *inode)
7157{
Chris Masone6dcd2d2008-07-17 12:53:50 -04007158 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04007159 struct btrfs_root *root = BTRFS_I(inode)->root;
7160
Al Virob3d9b7a2012-06-09 13:51:19 -04007161 WARN_ON(!hlist_empty(&inode->i_dentry));
Chris Mason39279cc2007-06-12 06:35:45 -04007162 WARN_ON(inode->i_data.nrpages);
Josef Bacik9e0baf62011-07-15 15:16:44 +00007163 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7164 WARN_ON(BTRFS_I(inode)->reserved_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04007165 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7166 WARN_ON(BTRFS_I(inode)->csum_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04007167
Chris Mason5a3f23d2009-03-31 13:27:11 -04007168 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05007169 * This can happen where we create an inode, but somebody else also
7170 * created the same inode and we need to destroy the one we already
7171 * created.
7172 */
7173 if (!root)
7174 goto free;
7175
7176 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04007177 * Make sure we're properly removed from the ordered operation
7178 * lists.
7179 */
7180 smp_mb();
7181 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7182 spin_lock(&root->fs_info->ordered_extent_lock);
7183 list_del_init(&BTRFS_I(inode)->ordered_operations);
7184 spin_unlock(&root->fs_info->ordered_extent_lock);
7185 }
7186
Josef Bacik8a35d952012-05-23 14:26:42 -04007187 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7188 &BTRFS_I(inode)->runtime_flags)) {
Li Zefan33345d012011-04-20 10:31:50 +08007189 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7190 (unsigned long long)btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04007191 atomic_dec(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04007192 }
Josef Bacik7b128762008-07-24 12:17:14 -04007193
Chris Masond3977122009-01-05 21:25:51 -05007194 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04007195 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7196 if (!ordered)
7197 break;
7198 else {
Chris Masond3977122009-01-05 21:25:51 -05007199 printk(KERN_ERR "btrfs found ordered "
7200 "extent %llu %llu on inode cleanup\n",
7201 (unsigned long long)ordered->file_offset,
7202 (unsigned long long)ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007203 btrfs_remove_ordered_extent(inode, ordered);
7204 btrfs_put_ordered_extent(ordered);
7205 btrfs_put_ordered_extent(ordered);
7206 }
7207 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007208 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007209 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05007210free:
Miao Xie16cdcec2011-04-22 18:12:22 +08007211 btrfs_remove_delayed_node(inode);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11007212 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04007213}
7214
Al Viro45321ac2010-06-07 13:43:19 -04007215int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04007216{
7217 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04007218
Josef Bacik0af3d002010-06-21 14:48:16 -04007219 if (btrfs_root_refs(&root->root_item) == 0 &&
Liu Bo83eea1f2012-07-10 05:28:39 -06007220 !btrfs_is_free_space_inode(inode))
Al Viro45321ac2010-06-07 13:43:19 -04007221 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04007222 else
Al Viro45321ac2010-06-07 13:43:19 -04007223 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04007224}
7225
Sven Wegener0ee0fda2008-07-30 16:54:26 -04007226static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04007227{
7228 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7229
7230 inode_init_once(&ei->vfs_inode);
7231}
7232
7233void btrfs_destroy_cachep(void)
7234{
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +10007235 /*
7236 * Make sure all delayed rcu free inodes are flushed before we
7237 * destroy cache.
7238 */
7239 rcu_barrier();
Chris Mason39279cc2007-06-12 06:35:45 -04007240 if (btrfs_inode_cachep)
7241 kmem_cache_destroy(btrfs_inode_cachep);
7242 if (btrfs_trans_handle_cachep)
7243 kmem_cache_destroy(btrfs_trans_handle_cachep);
7244 if (btrfs_transaction_cachep)
7245 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04007246 if (btrfs_path_cachep)
7247 kmem_cache_destroy(btrfs_path_cachep);
Josef Bacikdc89e982011-01-28 17:05:48 -05007248 if (btrfs_free_space_cachep)
7249 kmem_cache_destroy(btrfs_free_space_cachep);
Miao Xie8ccf6f192012-10-25 09:28:04 +00007250 if (btrfs_delalloc_work_cachep)
7251 kmem_cache_destroy(btrfs_delalloc_work_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04007252}
7253
7254int btrfs_init_cachep(void)
7255{
David Sterba837e1972012-09-07 03:00:48 -06007256 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007257 sizeof(struct btrfs_inode), 0,
7258 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04007259 if (!btrfs_inode_cachep)
7260 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007261
David Sterba837e1972012-09-07 03:00:48 -06007262 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007263 sizeof(struct btrfs_trans_handle), 0,
7264 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04007265 if (!btrfs_trans_handle_cachep)
7266 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007267
David Sterba837e1972012-09-07 03:00:48 -06007268 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007269 sizeof(struct btrfs_transaction), 0,
7270 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04007271 if (!btrfs_transaction_cachep)
7272 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007273
David Sterba837e1972012-09-07 03:00:48 -06007274 btrfs_path_cachep = kmem_cache_create("btrfs_path",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007275 sizeof(struct btrfs_path), 0,
7276 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04007277 if (!btrfs_path_cachep)
7278 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02007279
David Sterba837e1972012-09-07 03:00:48 -06007280 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
Josef Bacikdc89e982011-01-28 17:05:48 -05007281 sizeof(struct btrfs_free_space), 0,
7282 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7283 if (!btrfs_free_space_cachep)
7284 goto fail;
7285
Miao Xie8ccf6f192012-10-25 09:28:04 +00007286 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7287 sizeof(struct btrfs_delalloc_work), 0,
7288 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7289 NULL);
7290 if (!btrfs_delalloc_work_cachep)
7291 goto fail;
7292
Chris Mason39279cc2007-06-12 06:35:45 -04007293 return 0;
7294fail:
7295 btrfs_destroy_cachep();
7296 return -ENOMEM;
7297}
7298
7299static int btrfs_getattr(struct vfsmount *mnt,
7300 struct dentry *dentry, struct kstat *stat)
7301{
7302 struct inode *inode = dentry->d_inode;
David Sterbafadc0d82011-11-20 07:33:38 -05007303 u32 blocksize = inode->i_sb->s_blocksize;
7304
Chris Mason39279cc2007-06-12 06:35:45 -04007305 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04007306 stat->dev = BTRFS_I(inode)->root->anon_dev;
Chris Masond6667462008-01-03 14:51:00 -05007307 stat->blksize = PAGE_CACHE_SIZE;
David Sterbafadc0d82011-11-20 07:33:38 -05007308 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7309 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04007310 return 0;
7311}
7312
Liu Bo75e7cb72011-03-22 10:12:20 +00007313/*
7314 * If a file is moved, it will inherit the cow and compression flags of the new
7315 * directory.
7316 */
7317static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7318{
7319 struct btrfs_inode *b_dir = BTRFS_I(dir);
7320 struct btrfs_inode *b_inode = BTRFS_I(inode);
7321
7322 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7323 b_inode->flags |= BTRFS_INODE_NODATACOW;
7324 else
7325 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7326
Liu Bobc178232012-06-14 02:23:18 -06007327 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
Liu Bo75e7cb72011-03-22 10:12:20 +00007328 b_inode->flags |= BTRFS_INODE_COMPRESS;
Liu Bobc178232012-06-14 02:23:18 -06007329 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7330 } else {
7331 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7332 BTRFS_INODE_NOCOMPRESS);
7333 }
Liu Bo75e7cb72011-03-22 10:12:20 +00007334}
7335
Chris Masond3977122009-01-05 21:25:51 -05007336static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7337 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04007338{
7339 struct btrfs_trans_handle *trans;
7340 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007341 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04007342 struct inode *new_inode = new_dentry->d_inode;
7343 struct inode *old_inode = old_dentry->d_inode;
7344 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007345 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007346 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04007347 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08007348 u64 old_ino = btrfs_ino(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007349
Li Zefan33345d012011-04-20 10:31:50 +08007350 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04007351 return -EPERM;
7352
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007353 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08007354 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05007355 return -EXDEV;
7356
Li Zefan33345d012011-04-20 10:31:50 +08007357 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7358 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007359 return -ENOTEMPTY;
7360
Chris Mason39279cc2007-06-12 06:35:45 -04007361 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007362 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04007363 return -ENOTEMPTY;
Chris Mason9c520572012-12-17 14:26:57 -05007364
7365
7366 /* check for collisions, even if the name isn't there */
7367 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7368 new_dentry->d_name.name,
7369 new_dentry->d_name.len);
7370
7371 if (ret) {
7372 if (ret == -EEXIST) {
7373 /* we shouldn't get
7374 * eexist without a new_inode */
7375 if (!new_inode) {
7376 WARN_ON(1);
7377 return ret;
7378 }
7379 } else {
7380 /* maybe -EOVERFLOW */
7381 return ret;
7382 }
7383 }
7384 ret = 0;
7385
Chris Mason5a3f23d2009-03-31 13:27:11 -04007386 /*
7387 * we're using rename to replace one file with another.
7388 * and the replacement file is large. Start IO on it now so
7389 * we don't add too much work to the end of the transaction
7390 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04007391 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04007392 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7393 filemap_flush(old_inode->i_mapping);
7394
Yan, Zheng76dda932009-09-21 16:00:26 -04007395 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08007396 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04007397 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007398 /*
7399 * We want to reserve the absolute worst case amount of items. So if
7400 * both inodes are subvols and we need to unlink them then that would
7401 * require 4 item modifications, but if they are both normal inodes it
7402 * would require 5 item modifications, so we'll assume their normal
7403 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7404 * should cover the worst case number of items we'll modify.
7405 */
7406 trans = btrfs_start_transaction(root, 20);
Johann Lombardib44c59a2011-03-31 13:23:47 +00007407 if (IS_ERR(trans)) {
7408 ret = PTR_ERR(trans);
7409 goto out_notrans;
7410 }
Chris Mason5f39d392007-10-15 16:14:19 -04007411
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007412 if (dest != root)
7413 btrfs_record_root_in_trans(trans, dest);
7414
Yan, Zhenga5719522009-09-24 09:17:31 -04007415 ret = btrfs_set_inode_index(new_dir, &index);
7416 if (ret)
7417 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04007418
Li Zefan33345d012011-04-20 10:31:50 +08007419 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007420 /* force full log commit if subvolume involved. */
7421 root->fs_info->last_trans_log_full_commit = trans->transid;
7422 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04007423 ret = btrfs_insert_inode_ref(trans, dest,
7424 new_dentry->d_name.name,
7425 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08007426 old_ino,
7427 btrfs_ino(new_dir), index);
Yan, Zhenga5719522009-09-24 09:17:31 -04007428 if (ret)
7429 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007430 /*
7431 * this is an ugly little race, but the rename is required
7432 * to make sure that if we crash, the inode is either at the
7433 * old name or the new one. pinning the log transaction lets
7434 * us make sure we don't allow a log commit to come in after
7435 * we unlink the name but before we add the new name back in.
7436 */
7437 btrfs_pin_log_trans(root);
7438 }
Chris Mason12fcfd22009-03-24 10:24:20 -04007439 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04007440 * make sure the inode gets flushed if it is replacing
7441 * something.
7442 */
Li Zefan33345d012011-04-20 10:31:50 +08007443 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
Chris Mason5a3f23d2009-03-31 13:27:11 -04007444 btrfs_add_ordered_operation(trans, root, old_inode);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007445
Josef Bacik0c4d2d92012-04-05 15:03:02 -04007446 inode_inc_iversion(old_dir);
7447 inode_inc_iversion(new_dir);
7448 inode_inc_iversion(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007449 old_dir->i_ctime = old_dir->i_mtime = ctime;
7450 new_dir->i_ctime = new_dir->i_mtime = ctime;
7451 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04007452
Chris Mason12fcfd22009-03-24 10:24:20 -04007453 if (old_dentry->d_parent != new_dentry->d_parent)
7454 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7455
Li Zefan33345d012011-04-20 10:31:50 +08007456 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007457 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7458 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7459 old_dentry->d_name.name,
7460 old_dentry->d_name.len);
7461 } else {
Al Viro92986792011-03-04 17:14:37 +00007462 ret = __btrfs_unlink_inode(trans, root, old_dir,
7463 old_dentry->d_inode,
7464 old_dentry->d_name.name,
7465 old_dentry->d_name.len);
7466 if (!ret)
7467 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007468 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007469 if (ret) {
7470 btrfs_abort_transaction(trans, root, ret);
7471 goto out_fail;
7472 }
Chris Mason39279cc2007-06-12 06:35:45 -04007473
7474 if (new_inode) {
Josef Bacik0c4d2d92012-04-05 15:03:02 -04007475 inode_inc_iversion(new_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007476 new_inode->i_ctime = CURRENT_TIME;
Li Zefan33345d012011-04-20 10:31:50 +08007477 if (unlikely(btrfs_ino(new_inode) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007478 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7479 root_objectid = BTRFS_I(new_inode)->location.objectid;
7480 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7481 root_objectid,
7482 new_dentry->d_name.name,
7483 new_dentry->d_name.len);
7484 BUG_ON(new_inode->i_nlink == 0);
7485 } else {
7486 ret = btrfs_unlink_inode(trans, dest, new_dir,
7487 new_dentry->d_inode,
7488 new_dentry->d_name.name,
7489 new_dentry->d_name.len);
7490 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007491 if (!ret && new_inode->i_nlink == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04007492 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007493 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04007494 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007495 if (ret) {
7496 btrfs_abort_transaction(trans, root, ret);
7497 goto out_fail;
7498 }
Chris Mason39279cc2007-06-12 06:35:45 -04007499 }
Josef Bacikaec74772008-07-24 12:12:38 -04007500
Liu Bo75e7cb72011-03-22 10:12:20 +00007501 fixup_inode_flags(new_dir, old_inode);
7502
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007503 ret = btrfs_add_link(trans, new_dir, old_inode,
7504 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04007505 new_dentry->d_name.len, 0, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007506 if (ret) {
7507 btrfs_abort_transaction(trans, root, ret);
7508 goto out_fail;
7509 }
Chris Mason39279cc2007-06-12 06:35:45 -04007510
Li Zefan33345d012011-04-20 10:31:50 +08007511 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
Al Viro10d9f302011-07-16 23:09:10 -04007512 struct dentry *parent = new_dentry->d_parent;
Josef Bacik6a912212010-11-20 09:48:00 +00007513 btrfs_log_new_name(trans, old_inode, old_dir, parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007514 btrfs_end_log_trans(root);
7515 }
Chris Mason39279cc2007-06-12 06:35:45 -04007516out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007517 btrfs_end_transaction(trans, root);
Johann Lombardib44c59a2011-03-31 13:23:47 +00007518out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +08007519 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04007520 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04007521
Chris Mason39279cc2007-06-12 06:35:45 -04007522 return ret;
7523}
7524
Miao Xie8ccf6f192012-10-25 09:28:04 +00007525static void btrfs_run_delalloc_work(struct btrfs_work *work)
7526{
7527 struct btrfs_delalloc_work *delalloc_work;
7528
7529 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7530 work);
7531 if (delalloc_work->wait)
7532 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7533 else
7534 filemap_flush(delalloc_work->inode->i_mapping);
7535
7536 if (delalloc_work->delay_iput)
7537 btrfs_add_delayed_iput(delalloc_work->inode);
7538 else
7539 iput(delalloc_work->inode);
7540 complete(&delalloc_work->completion);
7541}
7542
7543struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7544 int wait, int delay_iput)
7545{
7546 struct btrfs_delalloc_work *work;
7547
7548 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7549 if (!work)
7550 return NULL;
7551
7552 init_completion(&work->completion);
7553 INIT_LIST_HEAD(&work->list);
7554 work->inode = inode;
7555 work->wait = wait;
7556 work->delay_iput = delay_iput;
7557 work->work.func = btrfs_run_delalloc_work;
7558
7559 return work;
7560}
7561
7562void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7563{
7564 wait_for_completion(&work->completion);
7565 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7566}
7567
Chris Masond352ac62008-09-29 15:18:18 -04007568/*
7569 * some fairly slow code that needs optimization. This walks the list
7570 * of all the inodes with pending delalloc and forces them to disk.
7571 */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00007572int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
Chris Masonea8c2812008-08-04 23:17:27 -04007573{
Chris Masonea8c2812008-08-04 23:17:27 -04007574 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007575 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00007576 struct btrfs_delalloc_work *work, *next;
7577 struct list_head works;
Miao Xie1eafa6c2013-01-22 10:49:00 +00007578 struct list_head splice;
Miao Xie8ccf6f192012-10-25 09:28:04 +00007579 int ret = 0;
Chris Masonea8c2812008-08-04 23:17:27 -04007580
Yan Zhengc146afa2008-11-12 14:34:12 -05007581 if (root->fs_info->sb->s_flags & MS_RDONLY)
7582 return -EROFS;
7583
Miao Xie8ccf6f192012-10-25 09:28:04 +00007584 INIT_LIST_HEAD(&works);
Miao Xie1eafa6c2013-01-22 10:49:00 +00007585 INIT_LIST_HEAD(&splice);
7586again:
Chris Mason75eff682008-12-15 15:54:40 -05007587 spin_lock(&root->fs_info->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00007588 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
7589 while (!list_empty(&splice)) {
7590 binode = list_entry(splice.next, struct btrfs_inode,
Chris Masonea8c2812008-08-04 23:17:27 -04007591 delalloc_inodes);
Miao Xie1eafa6c2013-01-22 10:49:00 +00007592
7593 list_del_init(&binode->delalloc_inodes);
7594
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007595 inode = igrab(&binode->vfs_inode);
7596 if (!inode)
Miao Xie1eafa6c2013-01-22 10:49:00 +00007597 continue;
7598
7599 list_add_tail(&binode->delalloc_inodes,
7600 &root->fs_info->delalloc_inodes);
Chris Mason75eff682008-12-15 15:54:40 -05007601 spin_unlock(&root->fs_info->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00007602
7603 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7604 if (unlikely(!work)) {
7605 ret = -ENOMEM;
7606 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007607 }
Miao Xie1eafa6c2013-01-22 10:49:00 +00007608 list_add_tail(&work->list, &works);
7609 btrfs_queue_worker(&root->fs_info->flush_workers,
7610 &work->work);
7611
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007612 cond_resched();
Chris Mason75eff682008-12-15 15:54:40 -05007613 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04007614 }
Chris Mason75eff682008-12-15 15:54:40 -05007615 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04007616
Miao Xie1eafa6c2013-01-22 10:49:00 +00007617 list_for_each_entry_safe(work, next, &works, list) {
7618 list_del_init(&work->list);
7619 btrfs_wait_and_free_delalloc_work(work);
7620 }
7621
7622 spin_lock(&root->fs_info->delalloc_lock);
7623 if (!list_empty(&root->fs_info->delalloc_inodes)) {
7624 spin_unlock(&root->fs_info->delalloc_lock);
7625 goto again;
7626 }
7627 spin_unlock(&root->fs_info->delalloc_lock);
7628
Chris Mason8c8bee12008-09-29 11:19:10 -04007629 /* the filemap_flush will queue IO into the worker threads, but
7630 * we have to make sure the IO is actually started and that
7631 * ordered extents get created before we return
7632 */
7633 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05007634 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05007635 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04007636 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05007637 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7638 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04007639 }
7640 atomic_dec(&root->fs_info->async_submit_draining);
Miao Xie1eafa6c2013-01-22 10:49:00 +00007641 return 0;
Miao Xie8ccf6f192012-10-25 09:28:04 +00007642out:
7643 list_for_each_entry_safe(work, next, &works, list) {
7644 list_del_init(&work->list);
7645 btrfs_wait_and_free_delalloc_work(work);
7646 }
Miao Xie1eafa6c2013-01-22 10:49:00 +00007647
7648 if (!list_empty_careful(&splice)) {
7649 spin_lock(&root->fs_info->delalloc_lock);
7650 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
7651 spin_unlock(&root->fs_info->delalloc_lock);
7652 }
Miao Xie8ccf6f192012-10-25 09:28:04 +00007653 return ret;
Chris Masonea8c2812008-08-04 23:17:27 -04007654}
7655
Chris Mason39279cc2007-06-12 06:35:45 -04007656static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7657 const char *symname)
7658{
7659 struct btrfs_trans_handle *trans;
7660 struct btrfs_root *root = BTRFS_I(dir)->root;
7661 struct btrfs_path *path;
7662 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05007663 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04007664 int err;
7665 int drop_inode = 0;
7666 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007667 u64 index = 0 ;
Chris Mason39279cc2007-06-12 06:35:45 -04007668 int name_len;
7669 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04007670 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04007671 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04007672 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04007673
7674 name_len = strlen(symname) + 1;
7675 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7676 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05007677
Josef Bacik9ed74f22009-09-11 16:12:44 -04007678 /*
7679 * 2 items for inode item and ref
7680 * 2 items for dir items
7681 * 1 item for xattr if selinux is on
7682 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04007683 trans = btrfs_start_transaction(root, 5);
7684 if (IS_ERR(trans))
7685 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05007686
Li Zefan581bb052011-04-20 10:06:11 +08007687 err = btrfs_find_free_ino(root, &objectid);
7688 if (err)
7689 goto out_unlock;
7690
Josef Bacikaec74772008-07-24 12:12:38 -04007691 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08007692 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04007693 S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04007694 if (IS_ERR(inode)) {
7695 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007696 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04007697 }
Chris Mason39279cc2007-06-12 06:35:45 -04007698
Eric Paris2a7dba32011-02-01 11:05:39 -05007699 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04007700 if (err) {
7701 drop_inode = 1;
7702 goto out_unlock;
7703 }
7704
Casey Schauflerad19db72011-12-15 10:09:07 -05007705 /*
7706 * If the active LSM wants to access the inode during
7707 * d_instantiate it needs these. Smack checks to see
7708 * if the filesystem supports xattrs by looking at the
7709 * ops vector.
7710 */
7711 inode->i_fop = &btrfs_file_operations;
7712 inode->i_op = &btrfs_file_inode_operations;
7713
Josef Bacika1b075d2010-11-19 20:36:11 +00007714 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04007715 if (err)
7716 drop_inode = 1;
7717 else {
7718 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04007719 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05007720 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04007721 }
Chris Mason39279cc2007-06-12 06:35:45 -04007722 if (drop_inode)
7723 goto out_unlock;
7724
7725 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07007726 if (!path) {
7727 err = -ENOMEM;
7728 drop_inode = 1;
7729 goto out_unlock;
7730 }
Li Zefan33345d012011-04-20 10:31:50 +08007731 key.objectid = btrfs_ino(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007732 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007733 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7734 datasize = btrfs_file_extent_calc_inline_size(name_len);
7735 err = btrfs_insert_empty_item(trans, root, path, &key,
7736 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04007737 if (err) {
7738 drop_inode = 1;
Julia Lawallb0839162011-05-14 07:10:51 +00007739 btrfs_free_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04007740 goto out_unlock;
7741 }
Chris Mason5f39d392007-10-15 16:14:19 -04007742 leaf = path->nodes[0];
7743 ei = btrfs_item_ptr(leaf, path->slots[0],
7744 struct btrfs_file_extent_item);
7745 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7746 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04007747 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04007748 btrfs_set_file_extent_encryption(leaf, ei, 0);
7749 btrfs_set_file_extent_compression(leaf, ei, 0);
7750 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7751 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7752
Chris Mason39279cc2007-06-12 06:35:45 -04007753 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04007754 write_extent_buffer(leaf, symname, ptr, name_len);
7755 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04007756 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04007757
Chris Mason39279cc2007-06-12 06:35:45 -04007758 inode->i_op = &btrfs_symlink_inode_operations;
7759 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04007760 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04007761 inode_set_bytes(inode, name_len);
Chris Masondbe674a2008-07-17 12:54:05 -04007762 btrfs_i_size_write(inode, name_len - 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04007763 err = btrfs_update_inode(trans, root, inode);
7764 if (err)
7765 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04007766
7767out_unlock:
Al Viro08c422c2011-12-23 07:58:13 -05007768 if (!err)
7769 d_instantiate(dentry, inode);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007770 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04007771 if (drop_inode) {
7772 inode_dec_link_count(inode);
7773 iput(inode);
7774 }
Liu Bob53d3f52012-11-14 14:34:34 +00007775 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04007776 return err;
7777}
Chris Mason16432982008-04-10 10:23:21 -04007778
Josef Bacik0af3d002010-06-21 14:48:16 -04007779static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7780 u64 start, u64 num_bytes, u64 min_size,
7781 loff_t actual_len, u64 *alloc_hint,
7782 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04007783{
Josef Bacik5dc562c2012-08-17 13:14:17 -04007784 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7785 struct extent_map *em;
Yan Zhengd899e052008-10-30 14:25:28 -04007786 struct btrfs_root *root = BTRFS_I(inode)->root;
7787 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04007788 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00007789 u64 i_size;
Yan Zhengd899e052008-10-30 14:25:28 -04007790 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04007791 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04007792
Josef Bacik0af3d002010-06-21 14:48:16 -04007793 if (trans)
7794 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04007795 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007796 if (own_trans) {
7797 trans = btrfs_start_transaction(root, 3);
7798 if (IS_ERR(trans)) {
7799 ret = PTR_ERR(trans);
7800 break;
7801 }
Yan Zhengd899e052008-10-30 14:25:28 -04007802 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00007803
Yan, Zhengefa56462010-05-16 10:49:59 -04007804 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
Josef Bacik81c9ad22012-01-18 10:56:06 -05007805 0, *alloc_hint, &ins, 1);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007806 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007807 if (own_trans)
7808 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007809 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007810 }
7811
Yan Zhengd899e052008-10-30 14:25:28 -04007812 ret = insert_reserved_file_extent(trans, inode,
7813 cur_offset, ins.objectid,
7814 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00007815 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04007816 BTRFS_FILE_EXTENT_PREALLOC);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007817 if (ret) {
7818 btrfs_abort_transaction(trans, root, ret);
7819 if (own_trans)
7820 btrfs_end_transaction(trans, root);
7821 break;
7822 }
Chris Masona1ed835e12009-09-11 12:27:37 -04007823 btrfs_drop_extent_cache(inode, cur_offset,
7824 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007825
Josef Bacik5dc562c2012-08-17 13:14:17 -04007826 em = alloc_extent_map();
7827 if (!em) {
7828 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7829 &BTRFS_I(inode)->runtime_flags);
7830 goto next;
7831 }
7832
7833 em->start = cur_offset;
7834 em->orig_start = cur_offset;
7835 em->len = ins.offset;
7836 em->block_start = ins.objectid;
7837 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -05007838 em->orig_block_len = ins.offset;
Josef Bacik5dc562c2012-08-17 13:14:17 -04007839 em->bdev = root->fs_info->fs_devices->latest_bdev;
7840 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7841 em->generation = trans->transid;
7842
7843 while (1) {
7844 write_lock(&em_tree->lock);
7845 ret = add_extent_mapping(em_tree, em);
7846 if (!ret)
7847 list_move(&em->list,
7848 &em_tree->modified_extents);
7849 write_unlock(&em_tree->lock);
7850 if (ret != -EEXIST)
7851 break;
7852 btrfs_drop_extent_cache(inode, cur_offset,
7853 cur_offset + ins.offset - 1,
7854 0);
7855 }
7856 free_extent_map(em);
7857next:
Yan Zhengd899e052008-10-30 14:25:28 -04007858 num_bytes -= ins.offset;
7859 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04007860 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007861
Josef Bacik0c4d2d92012-04-05 15:03:02 -04007862 inode_inc_iversion(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04007863 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02007864 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04007865 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -04007866 (actual_len > inode->i_size) &&
7867 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007868 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +00007869 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007870 else
Josef Bacik55a61d12010-11-22 18:50:32 +00007871 i_size = cur_offset;
7872 i_size_write(inode, i_size);
7873 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007874 }
7875
Yan Zhengd899e052008-10-30 14:25:28 -04007876 ret = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007877
7878 if (ret) {
7879 btrfs_abort_transaction(trans, root, ret);
7880 if (own_trans)
7881 btrfs_end_transaction(trans, root);
7882 break;
7883 }
Yan Zhengd899e052008-10-30 14:25:28 -04007884
Josef Bacik0af3d002010-06-21 14:48:16 -04007885 if (own_trans)
7886 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007887 }
Yan Zhengd899e052008-10-30 14:25:28 -04007888 return ret;
7889}
7890
Josef Bacik0af3d002010-06-21 14:48:16 -04007891int btrfs_prealloc_file_range(struct inode *inode, int mode,
7892 u64 start, u64 num_bytes, u64 min_size,
7893 loff_t actual_len, u64 *alloc_hint)
7894{
7895 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7896 min_size, actual_len, alloc_hint,
7897 NULL);
7898}
7899
7900int btrfs_prealloc_file_range_trans(struct inode *inode,
7901 struct btrfs_trans_handle *trans, int mode,
7902 u64 start, u64 num_bytes, u64 min_size,
7903 loff_t actual_len, u64 *alloc_hint)
7904{
7905 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7906 min_size, actual_len, alloc_hint, trans);
7907}
7908
Chris Masone6dcd2d2008-07-17 12:53:50 -04007909static int btrfs_set_page_dirty(struct page *page)
7910{
Chris Masone6dcd2d2008-07-17 12:53:50 -04007911 return __set_page_dirty_nobuffers(page);
7912}
7913
Al Viro10556cb22011-06-20 19:28:19 -04007914static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05007915{
Li Zefanb83cc962010-12-20 16:04:08 +08007916 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00007917 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +08007918
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00007919 if (mask & MAY_WRITE &&
7920 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7921 if (btrfs_root_readonly(root))
7922 return -EROFS;
7923 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7924 return -EACCES;
7925 }
Al Viro2830ba72011-06-20 19:16:29 -04007926 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -05007927}
Chris Mason39279cc2007-06-12 06:35:45 -04007928
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007929static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05007930 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04007931 .lookup = btrfs_lookup,
7932 .create = btrfs_create,
7933 .unlink = btrfs_unlink,
7934 .link = btrfs_link,
7935 .mkdir = btrfs_mkdir,
7936 .rmdir = btrfs_rmdir,
7937 .rename = btrfs_rename,
7938 .symlink = btrfs_symlink,
7939 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04007940 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007941 .setxattr = btrfs_setxattr,
7942 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007943 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007944 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05007945 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007946 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007947};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007948static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007949 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05007950 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007951 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007952};
Yan, Zheng76dda932009-09-21 16:00:26 -04007953
Alexey Dobriyan828c0952009-10-01 15:43:56 -07007954static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007955 .llseek = generic_file_llseek,
7956 .read = generic_read_dir,
David Woodhousecbdf5a22008-08-06 19:42:33 +01007957 .readdir = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007958 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007959#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007960 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007961#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04007962 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04007963 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04007964};
7965
Chris Masond1310b22008-01-24 16:13:08 -05007966static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04007967 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05007968 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04007969 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007970 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007971 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04007972 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05007973 .set_bit_hook = btrfs_set_bit_hook,
7974 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04007975 .merge_extent_hook = btrfs_merge_extent_hook,
7976 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007977};
7978
Chris Mason35054392009-01-21 13:11:13 -05007979/*
7980 * btrfs doesn't support the bmap operation because swapfiles
7981 * use bmap to make a mapping of extents in the file. They assume
7982 * these extents won't change over the life of the file and they
7983 * use the bmap result to do IO directly to the drive.
7984 *
7985 * the btrfs bmap call would return logical addresses that aren't
7986 * suitable for IO and they also will change frequently as COW
7987 * operations happen. So, swapfile + btrfs == corruption.
7988 *
7989 * For now we're avoiding this by dropping bmap.
7990 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007991static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007992 .readpage = btrfs_readpage,
7993 .writepage = btrfs_writepage,
Chris Masonb293f02e2007-11-01 19:45:34 -04007994 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05007995 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -04007996 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04007997 .invalidatepage = btrfs_invalidatepage,
7998 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007999 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02008000 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04008001};
8002
Alexey Dobriyan7f094102009-09-21 17:01:10 -07008003static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04008004 .readpage = btrfs_readpage,
8005 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04008006 .invalidatepage = btrfs_invalidatepage,
8007 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04008008};
8009
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008010static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04008011 .getattr = btrfs_getattr,
8012 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008013 .setxattr = btrfs_setxattr,
8014 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05008015 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008016 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05008017 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008018 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008019 .get_acl = btrfs_get_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04008020 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04008021};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008022static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04008023 .getattr = btrfs_getattr,
8024 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05008025 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008026 .setxattr = btrfs_setxattr,
8027 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04008028 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008029 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008030 .get_acl = btrfs_get_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04008031 .update_time = btrfs_update_time,
Josef Bacik618e21d2007-07-11 10:18:17 -04008032};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008033static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04008034 .readlink = generic_readlink,
8035 .follow_link = page_follow_link_light,
8036 .put_link = page_put_link,
Li Zefanf2095612010-11-19 02:05:24 +00008037 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -05008038 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05008039 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05008040 .setxattr = btrfs_setxattr,
8041 .getxattr = btrfs_getxattr,
8042 .listxattr = btrfs_listxattr,
8043 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008044 .get_acl = btrfs_get_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04008045 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04008046};
Yan, Zheng76dda932009-09-21 16:00:26 -04008047
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04008048const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04008049 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -04008050 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -04008051};