blob: 1d65f6ef00ca8ba7a42eae4c3e87ff885201c43b [file] [log] [blame]
Thomas Gleixner328970d2019-05-24 12:04:05 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Masahiro Yamadafa60ce22021-05-06 18:06:44 -07002/*
Mark Fashehccd979b2005-12-15 14:31:24 -08003 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
Mark Fashehccd979b2005-12-15 14:31:24 -08004 */
5
6#include <linux/fs.h>
7#include <linux/slab.h>
8#include <linux/highmem.h>
9#include <linux/pagemap.h>
10#include <asm/byteorder.h>
Mark Fasheh9517bac2007-02-09 20:24:12 -080011#include <linux/swap.h>
Mark Fasheh628a24f2007-10-30 12:08:32 -070012#include <linux/mpage.h>
Jan Karaa90714c2008-10-09 19:38:40 +020013#include <linux/quotaops.h>
Joseph Qi24c40b32015-02-16 16:00:00 -080014#include <linux/blkdev.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080015#include <linux/uio.h>
Nikolay Borisovf86196e2019-01-03 15:29:02 -080016#include <linux/mm.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080017
Mark Fashehccd979b2005-12-15 14:31:24 -080018#include <cluster/masklog.h>
19
20#include "ocfs2.h"
21
22#include "alloc.h"
23#include "aops.h"
24#include "dlmglue.h"
25#include "extent_map.h"
26#include "file.h"
27#include "inode.h"
28#include "journal.h"
Mark Fasheh9517bac2007-02-09 20:24:12 -080029#include "suballoc.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080030#include "super.h"
31#include "symlink.h"
Tao Ma293b2f72009-08-25 08:02:48 +080032#include "refcounttree.h"
Tao Ma95581562011-02-22 21:33:59 +080033#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080034
35#include "buffer_head_io.h"
Joseph Qi24c40b32015-02-16 16:00:00 -080036#include "dir.h"
37#include "namei.h"
38#include "sysfile.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080039
40static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
41 struct buffer_head *bh_result, int create)
42{
43 int err = -EIO;
44 int status;
45 struct ocfs2_dinode *fe = NULL;
46 struct buffer_head *bh = NULL;
47 struct buffer_head *buffer_cache_bh = NULL;
48 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
49 void *kaddr;
50
Tao Ma95581562011-02-22 21:33:59 +080051 trace_ocfs2_symlink_get_block(
52 (unsigned long long)OCFS2_I(inode)->ip_blkno,
53 (unsigned long long)iblock, bh_result, create);
Mark Fashehccd979b2005-12-15 14:31:24 -080054
55 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
56
57 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
58 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
59 (unsigned long long)iblock);
60 goto bail;
61 }
62
Joel Beckerb657c952008-11-13 14:49:11 -080063 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -080064 if (status < 0) {
65 mlog_errno(status);
66 goto bail;
67 }
68 fe = (struct ocfs2_dinode *) bh->b_data;
69
Mark Fashehccd979b2005-12-15 14:31:24 -080070 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
71 le32_to_cpu(fe->i_clusters))) {
Rui Xiang7391a292013-11-12 15:06:54 -080072 err = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -080073 mlog(ML_ERROR, "block offset is outside the allocated size: "
74 "%llu\n", (unsigned long long)iblock);
75 goto bail;
76 }
77
78 /* We don't use the page cache to create symlink data, so if
79 * need be, copy it over from the buffer cache. */
80 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
81 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
82 iblock;
83 buffer_cache_bh = sb_getblk(osb->sb, blkno);
84 if (!buffer_cache_bh) {
Rui Xiang7391a292013-11-12 15:06:54 -080085 err = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -080086 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
87 goto bail;
88 }
89
90 /* we haven't locked out transactions, so a commit
91 * could've happened. Since we've got a reference on
92 * the bh, even if it commits while we're doing the
93 * copy, the data is still good. */
94 if (buffer_jbd(buffer_cache_bh)
95 && ocfs2_inode_is_new(inode)) {
Cong Wangc4bc8dc2011-11-25 23:14:34 +080096 kaddr = kmap_atomic(bh_result->b_page);
Mark Fashehccd979b2005-12-15 14:31:24 -080097 if (!kaddr) {
98 mlog(ML_ERROR, "couldn't kmap!\n");
99 goto bail;
100 }
101 memcpy(kaddr + (bh_result->b_size * iblock),
102 buffer_cache_bh->b_data,
103 bh_result->b_size);
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800104 kunmap_atomic(kaddr);
Mark Fashehccd979b2005-12-15 14:31:24 -0800105 set_buffer_uptodate(bh_result);
106 }
107 brelse(buffer_cache_bh);
108 }
109
110 map_bh(bh_result, inode->i_sb,
111 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
112
113 err = 0;
114
115bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700116 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800117
Mark Fashehccd979b2005-12-15 14:31:24 -0800118 return err;
119}
120
alex chen3e4c56d2017-11-15 17:31:44 -0800121static int ocfs2_lock_get_block(struct inode *inode, sector_t iblock,
122 struct buffer_head *bh_result, int create)
123{
124 int ret = 0;
125 struct ocfs2_inode_info *oi = OCFS2_I(inode);
126
127 down_read(&oi->ip_alloc_sem);
128 ret = ocfs2_get_block(inode, iblock, bh_result, create);
129 up_read(&oi->ip_alloc_sem);
130
131 return ret;
132}
133
Tao Ma6f70fa52009-08-25 08:05:12 +0800134int ocfs2_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh_result, int create)
Mark Fashehccd979b2005-12-15 14:31:24 -0800136{
137 int err = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800138 unsigned int ext_flags;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700139 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
140 u64 p_blkno, count, past_eof;
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800141 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800142
Tao Ma95581562011-02-22 21:33:59 +0800143 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
144 (unsigned long long)iblock, bh_result, create);
Mark Fashehccd979b2005-12-15 14:31:24 -0800145
146 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
147 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
148 inode, inode->i_ino);
149
150 if (S_ISLNK(inode->i_mode)) {
151 /* this always does I/O for some reason. */
152 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
153 goto bail;
154 }
155
Mark Fasheh628a24f2007-10-30 12:08:32 -0700156 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800157 &ext_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800158 if (err) {
159 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
Mark Fashehb06970532006-03-03 10:24:33 -0800160 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
161 (unsigned long long)p_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800162 goto bail;
163 }
164
Mark Fasheh628a24f2007-10-30 12:08:32 -0700165 if (max_blocks < count)
166 count = max_blocks;
167
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800168 /*
169 * ocfs2 never allocates in this function - the only time we
170 * need to use BH_New is when we're extending i_size on a file
171 * system which doesn't support holes, in which case BH_New
Christoph Hellwigebdec242010-10-06 10:47:23 +0200172 * allows __block_write_begin() to zero.
Coly Lic0420ad2008-06-30 18:45:45 +0800173 *
174 * If we see this on a sparse file system, then a truncate has
175 * raced us and removed the cluster. In this case, we clear
176 * the buffers dirty and uptodate bits and let the buffer code
177 * ignore it as a hole.
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800178 */
Coly Lic0420ad2008-06-30 18:45:45 +0800179 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
180 clear_buffer_dirty(bh_result);
181 clear_buffer_uptodate(bh_result);
182 goto bail;
183 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800184
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800185 /* Treat the unwritten extent as a hole for zeroing purposes. */
186 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800187 map_bh(bh_result, inode->i_sb, p_blkno);
188
Mark Fasheh628a24f2007-10-30 12:08:32 -0700189 bh_result->b_size = count << inode->i_blkbits;
190
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800191 if (!ocfs2_sparse_alloc(osb)) {
192 if (p_blkno == 0) {
193 err = -EIO;
194 mlog(ML_ERROR,
195 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
196 (unsigned long long)iblock,
197 (unsigned long long)p_blkno,
198 (unsigned long long)OCFS2_I(inode)->ip_blkno);
199 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
200 dump_stack();
Wengang Wang1f4cea32009-07-13 11:38:58 +0800201 goto bail;
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800202 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800203 }
204
Joel Becker56934862010-07-01 15:13:31 -0700205 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
Tao Ma95581562011-02-22 21:33:59 +0800206
207 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
208 (unsigned long long)past_eof);
Joel Becker56934862010-07-01 15:13:31 -0700209 if (create && (iblock >= past_eof))
210 set_buffer_new(bh_result);
211
Mark Fashehccd979b2005-12-15 14:31:24 -0800212bail:
213 if (err < 0)
214 err = -EIO;
215
Mark Fashehccd979b2005-12-15 14:31:24 -0800216 return err;
217}
218
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700219int ocfs2_read_inline_data(struct inode *inode, struct page *page,
220 struct buffer_head *di_bh)
Mark Fasheh6798d352007-09-07 14:05:51 -0700221{
222 void *kaddr;
Jan Karad2849fb2007-12-19 15:24:09 +0100223 loff_t size;
Mark Fasheh6798d352007-09-07 14:05:51 -0700224 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
225
226 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
Joe Perches7ecef142015-09-04 15:44:51 -0700227 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
Mark Fasheh6798d352007-09-07 14:05:51 -0700228 (unsigned long long)OCFS2_I(inode)->ip_blkno);
229 return -EROFS;
230 }
231
232 size = i_size_read(inode);
233
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300234 if (size > PAGE_SIZE ||
Tiger Yangd9ae49d2009-03-05 11:06:15 +0800235 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
Mark Fasheh6798d352007-09-07 14:05:51 -0700236 ocfs2_error(inode->i_sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700237 "Inode %llu has with inline data has bad size: %Lu\n",
Jan Karad2849fb2007-12-19 15:24:09 +0100238 (unsigned long long)OCFS2_I(inode)->ip_blkno,
239 (unsigned long long)size);
Mark Fasheh6798d352007-09-07 14:05:51 -0700240 return -EROFS;
241 }
242
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800243 kaddr = kmap_atomic(page);
Mark Fasheh6798d352007-09-07 14:05:51 -0700244 if (size)
245 memcpy(kaddr, di->id2.i_data.id_data, size);
246 /* Clear the remaining part of the page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300247 memset(kaddr + size, 0, PAGE_SIZE - size);
Mark Fasheh6798d352007-09-07 14:05:51 -0700248 flush_dcache_page(page);
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800249 kunmap_atomic(kaddr);
Mark Fasheh6798d352007-09-07 14:05:51 -0700250
251 SetPageUptodate(page);
252
253 return 0;
254}
255
256static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
257{
258 int ret;
259 struct buffer_head *di_bh = NULL;
Mark Fasheh6798d352007-09-07 14:05:51 -0700260
261 BUG_ON(!PageLocked(page));
Julia Lawall86c838b2008-02-26 21:45:56 +0100262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
Mark Fasheh6798d352007-09-07 14:05:51 -0700263
Joel Beckerb657c952008-11-13 14:49:11 -0800264 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh6798d352007-09-07 14:05:51 -0700265 if (ret) {
266 mlog_errno(ret);
267 goto out;
268 }
269
270 ret = ocfs2_read_inline_data(inode, page, di_bh);
271out:
272 unlock_page(page);
273
274 brelse(di_bh);
275 return ret;
276}
277
Matthew Wilcox (Oracle)bb9263f2022-04-29 11:12:16 -0400278static int ocfs2_read_folio(struct file *file, struct folio *folio)
Mark Fashehccd979b2005-12-15 14:31:24 -0800279{
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400280 struct inode *inode = folio->mapping->host;
Mark Fasheh6798d352007-09-07 14:05:51 -0700281 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400282 loff_t start = folio_pos(folio);
Mark Fashehccd979b2005-12-15 14:31:24 -0800283 int ret, unlock = 1;
284
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400285 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno, folio->index);
Mark Fashehccd979b2005-12-15 14:31:24 -0800286
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400287 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, &folio->page);
Mark Fashehccd979b2005-12-15 14:31:24 -0800288 if (ret != 0) {
289 if (ret == AOP_TRUNCATED_PAGE)
290 unlock = 0;
291 mlog_errno(ret);
292 goto out;
293 }
294
Mark Fasheh6798d352007-09-07 14:05:51 -0700295 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
Jan Karac7e25e62011-06-23 22:51:47 +0200296 /*
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400297 * Unlock the folio and cycle ip_alloc_sem so that we don't
Jan Karac7e25e62011-06-23 22:51:47 +0200298 * busyloop waiting for ip_alloc_sem to unlock
299 */
Mark Fashehe9dfc0b2007-05-14 11:38:51 -0700300 ret = AOP_TRUNCATED_PAGE;
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400301 folio_unlock(folio);
Jan Karac7e25e62011-06-23 22:51:47 +0200302 unlock = 0;
303 down_read(&oi->ip_alloc_sem);
304 up_read(&oi->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700305 goto out_inode_unlock;
Mark Fashehe9dfc0b2007-05-14 11:38:51 -0700306 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800307
308 /*
309 * i_size might have just been updated as we grabed the meta lock. We
310 * might now be discovering a truncate that hit on another node.
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -0400311 * block_read_full_folio->get_block freaks out if it is asked to read
Mark Fashehccd979b2005-12-15 14:31:24 -0800312 * beyond the end of a file, so we check here. Callers
Nick Piggin54cb8822007-07-19 01:46:59 -0700313 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400314 * and notice that the folio they just read isn't needed.
Mark Fashehccd979b2005-12-15 14:31:24 -0800315 *
316 * XXX sys_readahead() seems to get that wrong?
317 */
318 if (start >= i_size_read(inode)) {
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400319 folio_zero_segment(folio, 0, folio_size(folio));
320 folio_mark_uptodate(folio);
Mark Fashehccd979b2005-12-15 14:31:24 -0800321 ret = 0;
322 goto out_alloc;
323 }
324
Mark Fasheh6798d352007-09-07 14:05:51 -0700325 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400326 ret = ocfs2_readpage_inline(inode, &folio->page);
Mark Fasheh6798d352007-09-07 14:05:51 -0700327 else
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400328 ret = block_read_full_folio(folio, ocfs2_get_block);
Mark Fashehccd979b2005-12-15 14:31:24 -0800329 unlock = 0;
330
Mark Fashehccd979b2005-12-15 14:31:24 -0800331out_alloc:
piaojund324cd42018-04-05 16:18:37 -0700332 up_read(&oi->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700333out_inode_unlock:
334 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800335out:
336 if (unlock)
Matthew Wilcox (Oracle)9bb88982022-04-29 11:12:16 -0400337 folio_unlock(folio);
Mark Fashehccd979b2005-12-15 14:31:24 -0800338 return ret;
339}
340
Mark Fasheh628a24f2007-10-30 12:08:32 -0700341/*
342 * This is used only for read-ahead. Failures or difficult to handle
343 * situations are safe to ignore.
344 *
345 * Right now, we don't bother with BH_Boundary - in-inode extent lists
346 * are quite large (243 extents on 4k blocks), so most inodes don't
347 * grow out to a tree. If need be, detecting boundary extents could
348 * trivially be added in a future version of ocfs2_get_block().
349 */
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700350static void ocfs2_readahead(struct readahead_control *rac)
Mark Fasheh628a24f2007-10-30 12:08:32 -0700351{
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700352 int ret;
353 struct inode *inode = rac->mapping->host;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700354 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fasheh628a24f2007-10-30 12:08:32 -0700355
356 /*
357 * Use the nonblocking flag for the dlm code to avoid page
358 * lock inversion, but don't bother with retrying.
359 */
360 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
361 if (ret)
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700362 return;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700363
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700364 if (down_read_trylock(&oi->ip_alloc_sem) == 0)
365 goto out_unlock;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700366
367 /*
368 * Don't bother with inline-data. There isn't anything
369 * to read-ahead in that case anyway...
370 */
371 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700372 goto out_up;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700373
374 /*
375 * Check whether a remote node truncated this file - we just
376 * drop out in that case as it's not worth handling here.
377 */
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700378 if (readahead_pos(rac) >= i_size_read(inode))
379 goto out_up;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700380
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700381 mpage_readahead(rac, ocfs2_get_block);
Mark Fasheh628a24f2007-10-30 12:08:32 -0700382
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700383out_up:
Mark Fasheh628a24f2007-10-30 12:08:32 -0700384 up_read(&oi->ip_alloc_sem);
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700385out_unlock:
Mark Fasheh628a24f2007-10-30 12:08:32 -0700386 ocfs2_inode_unlock(inode, 0);
Mark Fasheh628a24f2007-10-30 12:08:32 -0700387}
388
Mark Fashehccd979b2005-12-15 14:31:24 -0800389/* Note: Because we don't support holes, our allocation has
390 * already happened (allocation writes zeros to the file data)
391 * so we don't have to worry about ordered writes in
392 * ocfs2_writepage.
393 *
394 * ->writepage is called during the process of invalidating the page cache
395 * during blocked lock processing. It can't block on any cluster locks
396 * to during block mapping. It's relying on the fact that the block
397 * mapping can't have disappeared under the dirty pages that it is
398 * being asked to write back.
399 */
400static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
401{
Tao Ma95581562011-02-22 21:33:59 +0800402 trace_ocfs2_writepage(
403 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
404 page->index);
Mark Fashehccd979b2005-12-15 14:31:24 -0800405
Tao Ma95581562011-02-22 21:33:59 +0800406 return block_write_full_page(page, ocfs2_get_block, wbc);
Mark Fashehccd979b2005-12-15 14:31:24 -0800407}
408
Mark Fashehccd979b2005-12-15 14:31:24 -0800409/* Taken from ext3. We don't necessarily need the full blown
410 * functionality yet, but IMHO it's better to cut and paste the whole
411 * thing so we can avoid introducing our own bugs (and easily pick up
412 * their fixes when they happen) --Mark */
Mark Fasheh60b11392007-02-16 11:46:50 -0800413int walk_page_buffers( handle_t *handle,
414 struct buffer_head *head,
415 unsigned from,
416 unsigned to,
417 int *partial,
418 int (*fn)( handle_t *handle,
419 struct buffer_head *bh))
Mark Fashehccd979b2005-12-15 14:31:24 -0800420{
421 struct buffer_head *bh;
422 unsigned block_start, block_end;
423 unsigned blocksize = head->b_size;
424 int err, ret = 0;
425 struct buffer_head *next;
426
427 for ( bh = head, block_start = 0;
428 ret == 0 && (bh != head || !block_start);
429 block_start = block_end, bh = next)
430 {
431 next = bh->b_this_page;
432 block_end = block_start + blocksize;
433 if (block_end <= from || block_start >= to) {
434 if (partial && !buffer_uptodate(bh))
435 *partial = 1;
436 continue;
437 }
438 err = (*fn)(handle, bh);
439 if (!ret)
440 ret = err;
441 }
442 return ret;
443}
444
Mark Fashehccd979b2005-12-15 14:31:24 -0800445static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
446{
447 sector_t status;
448 u64 p_blkno = 0;
449 int err = 0;
450 struct inode *inode = mapping->host;
451
Tao Ma95581562011-02-22 21:33:59 +0800452 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
453 (unsigned long long)block);
Mark Fashehccd979b2005-12-15 14:31:24 -0800454
Darrick J. Wong06a70302016-11-09 14:13:10 -0800455 /*
456 * The swap code (ab-)uses ->bmap to get a block mapping and then
457 * bypasseѕ the file system for actual I/O. We really can't allow
458 * that on refcounted inodes, so we have to skip out here. And yes,
459 * 0 is the magic code for a bmap error..
460 */
461 if (ocfs2_is_refcount_inode(inode))
462 return 0;
463
Mark Fashehccd979b2005-12-15 14:31:24 -0800464 /* We don't need to lock journal system files, since they aren't
465 * accessed concurrently from multiple nodes.
466 */
467 if (!INODE_JOURNAL(inode)) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700468 err = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800469 if (err) {
470 if (err != -ENOENT)
471 mlog_errno(err);
472 goto bail;
473 }
474 down_read(&OCFS2_I(inode)->ip_alloc_sem);
475 }
476
Mark Fasheh6798d352007-09-07 14:05:51 -0700477 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
478 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
479 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800480
481 if (!INODE_JOURNAL(inode)) {
482 up_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700483 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800484 }
485
486 if (err) {
487 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
488 (unsigned long long)block);
489 mlog_errno(err);
490 goto bail;
491 }
492
Mark Fashehccd979b2005-12-15 14:31:24 -0800493bail:
494 status = err ? 0 : p_blkno;
495
Mark Fashehccd979b2005-12-15 14:31:24 -0800496 return status;
497}
498
Matthew Wilcox (Oracle)eca66382022-05-01 00:10:21 -0400499static bool ocfs2_release_folio(struct folio *folio, gfp_t wait)
Joel Becker03f981c2007-01-04 14:54:41 -0800500{
Matthew Wilcox (Oracle)eca66382022-05-01 00:10:21 -0400501 if (!folio_buffers(folio))
502 return false;
Matthew Wilcox (Oracle)68189fe2022-05-01 01:08:08 -0400503 return try_to_free_buffers(folio);
Joel Becker03f981c2007-01-04 14:54:41 -0800504}
505
Mark Fasheh9517bac2007-02-09 20:24:12 -0800506static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
507 u32 cpos,
508 unsigned int *start,
509 unsigned int *end)
510{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300511 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800512
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300513 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800514 unsigned int cpp;
515
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300516 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800517
518 cluster_start = cpos % cpp;
519 cluster_start = cluster_start << osb->s_clustersize_bits;
520
521 cluster_end = cluster_start + osb->s_clustersize;
522 }
523
524 BUG_ON(cluster_start > PAGE_SIZE);
525 BUG_ON(cluster_end > PAGE_SIZE);
526
527 if (start)
528 *start = cluster_start;
529 if (end)
530 *end = cluster_end;
531}
532
533/*
534 * 'from' and 'to' are the region in the page to avoid zeroing.
535 *
536 * If pagesize > clustersize, this function will avoid zeroing outside
537 * of the cluster boundary.
538 *
539 * from == to == 0 is code for "zero the entire cluster region"
540 */
541static void ocfs2_clear_page_regions(struct page *page,
542 struct ocfs2_super *osb, u32 cpos,
543 unsigned from, unsigned to)
544{
545 void *kaddr;
546 unsigned int cluster_start, cluster_end;
547
548 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
549
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800550 kaddr = kmap_atomic(page);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800551
552 if (from || to) {
553 if (from > cluster_start)
554 memset(kaddr + cluster_start, 0, from - cluster_start);
555 if (to < cluster_end)
556 memset(kaddr + to, 0, cluster_end - to);
557 } else {
558 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
559 }
560
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800561 kunmap_atomic(kaddr);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800562}
563
564/*
Mark Fasheh4e9563f2007-11-01 11:37:48 -0700565 * Nonsparse file systems fully allocate before we get to the write
566 * code. This prevents ocfs2_write() from tagging the write as an
567 * allocating one, which means ocfs2_map_page_blocks() might try to
568 * read-in the blocks at the tail of our file. Avoid reading them by
569 * testing i_size against each block offset.
570 */
571static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
572 unsigned int block_start)
573{
574 u64 offset = page_offset(page) + block_start;
575
576 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
577 return 1;
578
579 if (i_size_read(inode) > offset)
580 return 1;
581
582 return 0;
583}
584
585/*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200586 * Some of this taken from __block_write_begin(). We already have our
Mark Fasheh9517bac2007-02-09 20:24:12 -0800587 * mapping by now though, and the entire write will be allocating or
588 * it won't, so not much need to use BH_New.
589 *
590 * This will also skip zeroing, which is handled externally.
591 */
Mark Fasheh60b11392007-02-16 11:46:50 -0800592int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
593 struct inode *inode, unsigned int from,
594 unsigned int to, int new)
Mark Fasheh9517bac2007-02-09 20:24:12 -0800595{
596 int ret = 0;
597 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
598 unsigned int block_end, block_start;
Fabian Frederick93407472017-02-27 14:28:32 -0800599 unsigned int bsize = i_blocksize(inode);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800600
601 if (!page_has_buffers(page))
602 create_empty_buffers(page, bsize, 0);
603
604 head = page_buffers(page);
605 for (bh = head, block_start = 0; bh != head || !block_start;
606 bh = bh->b_this_page, block_start += bsize) {
607 block_end = block_start + bsize;
608
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700609 clear_buffer_new(bh);
610
Mark Fasheh9517bac2007-02-09 20:24:12 -0800611 /*
612 * Ignore blocks outside of our i/o range -
613 * they may belong to unallocated clusters.
614 */
Mark Fasheh60b11392007-02-16 11:46:50 -0800615 if (block_start >= to || block_end <= from) {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800616 if (PageUptodate(page))
617 set_buffer_uptodate(bh);
618 continue;
619 }
620
621 /*
622 * For an allocating write with cluster size >= page
623 * size, we always write the entire page.
624 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700625 if (new)
626 set_buffer_new(bh);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800627
628 if (!buffer_mapped(bh)) {
629 map_bh(bh, inode->i_sb, *p_blkno);
Jan Karae64855c2016-11-04 18:08:15 +0100630 clean_bdev_bh_alias(bh);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800631 }
632
633 if (PageUptodate(page)) {
Wan Jiabing01f01392021-06-28 19:34:08 -0700634 set_buffer_uptodate(bh);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800635 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
Mark Fashehbce99762007-06-18 11:12:36 -0700636 !buffer_new(bh) &&
Mark Fasheh4e9563f2007-11-01 11:37:48 -0700637 ocfs2_should_read_blk(inode, page, block_start) &&
Mark Fashehbce99762007-06-18 11:12:36 -0700638 (block_start < from || block_end > to)) {
Zhang Yi54d91712022-09-01 21:34:59 +0800639 bh_read_nowait(bh, 0);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800640 *wait_bh++=bh;
641 }
642
643 *p_blkno = *p_blkno + 1;
644 }
645
646 /*
647 * If we issued read requests - let them complete.
648 */
649 while(wait_bh > wait) {
650 wait_on_buffer(*--wait_bh);
651 if (!buffer_uptodate(*wait_bh))
652 ret = -EIO;
653 }
654
655 if (ret == 0 || !new)
656 return ret;
657
658 /*
659 * If we get -EIO above, zero out any newly allocated blocks
660 * to avoid exposing stale data.
661 */
662 bh = head;
663 block_start = 0;
664 do {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800665 block_end = block_start + bsize;
666 if (block_end <= from)
667 goto next_bh;
668 if (block_start >= to)
669 break;
670
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800671 zero_user(page, block_start, bh->b_size);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800672 set_buffer_uptodate(bh);
673 mark_buffer_dirty(bh);
674
675next_bh:
676 block_start = block_end;
677 bh = bh->b_this_page;
678 } while (bh != head);
679
680 return ret;
681}
682
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300683#if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700684#define OCFS2_MAX_CTXT_PAGES 1
685#else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300686#define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700687#endif
Mark Fasheh6af67d82007-03-06 17:24:46 -0800688
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300689#define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
Mark Fasheh6af67d82007-03-06 17:24:46 -0800690
Ryan Ding4506cfb2016-03-25 14:21:06 -0700691struct ocfs2_unwritten_extent {
692 struct list_head ue_node;
693 struct list_head ue_ip_node;
694 u32 ue_cpos;
695 u32 ue_phys;
696};
697
Mark Fasheh6af67d82007-03-06 17:24:46 -0800698/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700699 * Describe the state of a single cluster to be written to.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800700 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700701struct ocfs2_write_cluster_desc {
702 u32 c_cpos;
703 u32 c_phys;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800704 /*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700705 * Give this a unique field because c_phys eventually gets
706 * filled.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800707 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700708 unsigned c_new;
Ryan Dingb46637d2016-03-25 14:20:55 -0700709 unsigned c_clear_unwritten;
Sunil Mushrane7432672009-08-06 16:12:58 -0700710 unsigned c_needs_zero;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700711};
Mark Fasheh9517bac2007-02-09 20:24:12 -0800712
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700713struct ocfs2_write_ctxt {
714 /* Logical cluster position / len of write */
715 u32 w_cpos;
716 u32 w_clen;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800717
Sunil Mushrane7432672009-08-06 16:12:58 -0700718 /* First cluster allocated in a nonsparse extend */
719 u32 w_first_new_cpos;
720
Ryan Dingc1ad1e32016-03-25 14:20:52 -0700721 /* Type of caller. Must be one of buffer, mmap, direct. */
722 ocfs2_write_type_t w_type;
723
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700724 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
Mark Fasheh9517bac2007-02-09 20:24:12 -0800725
726 /*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700727 * This is true if page_size > cluster_size.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800728 *
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700729 * It triggers a set of special cases during write which might
730 * have to deal with allocating writes to partial pages.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800731 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700732 unsigned int w_large_pages;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800733
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700734 /*
735 * Pages involved in this write.
736 *
737 * w_target_page is the page being written to by the user.
738 *
739 * w_pages is an array of pages which always contains
740 * w_target_page, and in the case of an allocating write with
741 * page_size < cluster size, it will contain zero'd and mapped
742 * pages adjacent to w_target_page which need to be written
743 * out in so that future reads from that region will get
744 * zero's.
745 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700746 unsigned int w_num_pages;
Goldwyn Rodrigues83fd9c72010-06-10 17:21:36 -0500747 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700748 struct page *w_target_page;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800749
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700750 /*
Wengang Wang5cffff92011-07-24 10:36:54 -0700751 * w_target_locked is used for page_mkwrite path indicating no unlocking
752 * against w_target_page in ocfs2_write_end_nolock.
753 */
754 unsigned int w_target_locked:1;
755
756 /*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700757 * ocfs2_write_end() uses this to know what the real range to
758 * write in the target should be.
759 */
760 unsigned int w_target_from;
761 unsigned int w_target_to;
762
763 /*
764 * We could use journal_current_handle() but this is cleaner,
765 * IMHO -Mark
766 */
767 handle_t *w_handle;
768
769 struct buffer_head *w_di_bh;
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700770
771 struct ocfs2_cached_dealloc_ctxt w_dealloc;
Ryan Ding4506cfb2016-03-25 14:21:06 -0700772
773 struct list_head w_unwritten_list;
Changwei Ge63de8bd2018-01-31 16:15:02 -0800774 unsigned int w_unwritten_count;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700775};
776
Mark Fasheh1d410a62007-09-07 14:20:45 -0700777void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700778{
779 int i;
780
Mark Fasheh1d410a62007-09-07 14:20:45 -0700781 for(i = 0; i < num_pages; i++) {
782 if (pages[i]) {
783 unlock_page(pages[i]);
784 mark_page_accessed(pages[i]);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300785 put_page(pages[i]);
Mark Fasheh1d410a62007-09-07 14:20:45 -0700786 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700787 }
Mark Fasheh1d410a62007-09-07 14:20:45 -0700788}
789
Junxiao Bi136f49b2014-12-18 16:17:37 -0800790static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
Mark Fasheh1d410a62007-09-07 14:20:45 -0700791{
Wengang Wang5cffff92011-07-24 10:36:54 -0700792 int i;
793
794 /*
795 * w_target_locked is only set to true in the page_mkwrite() case.
796 * The intent is to allow us to lock the target page from write_begin()
797 * to write_end(). The caller must hold a ref on w_target_page.
798 */
799 if (wc->w_target_locked) {
800 BUG_ON(!wc->w_target_page);
801 for (i = 0; i < wc->w_num_pages; i++) {
802 if (wc->w_target_page == wc->w_pages[i]) {
803 wc->w_pages[i] = NULL;
804 break;
805 }
806 }
807 mark_page_accessed(wc->w_target_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300808 put_page(wc->w_target_page);
Wengang Wang5cffff92011-07-24 10:36:54 -0700809 }
Mark Fasheh1d410a62007-09-07 14:20:45 -0700810 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
Junxiao Bi136f49b2014-12-18 16:17:37 -0800811}
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700812
Ryan Ding4506cfb2016-03-25 14:21:06 -0700813static void ocfs2_free_unwritten_list(struct inode *inode,
814 struct list_head *head)
Junxiao Bi136f49b2014-12-18 16:17:37 -0800815{
Ryan Ding4506cfb2016-03-25 14:21:06 -0700816 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Ryan Dingc15471f2016-03-25 14:21:09 -0700817 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
Ryan Ding4506cfb2016-03-25 14:21:06 -0700818
Ryan Dingc15471f2016-03-25 14:21:09 -0700819 list_for_each_entry_safe(ue, tmp, head, ue_node) {
820 list_del(&ue->ue_node);
Ryan Ding4506cfb2016-03-25 14:21:06 -0700821 spin_lock(&oi->ip_lock);
Ryan Dingc15471f2016-03-25 14:21:09 -0700822 list_del(&ue->ue_ip_node);
Ryan Ding4506cfb2016-03-25 14:21:06 -0700823 spin_unlock(&oi->ip_lock);
Ryan Dingc15471f2016-03-25 14:21:09 -0700824 kfree(ue);
Ryan Ding4506cfb2016-03-25 14:21:06 -0700825 }
826}
827
828static void ocfs2_free_write_ctxt(struct inode *inode,
829 struct ocfs2_write_ctxt *wc)
830{
831 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
Junxiao Bi136f49b2014-12-18 16:17:37 -0800832 ocfs2_unlock_pages(wc);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700833 brelse(wc->w_di_bh);
834 kfree(wc);
835}
836
837static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
838 struct ocfs2_super *osb, loff_t pos,
Ryan Dingc1ad1e32016-03-25 14:20:52 -0700839 unsigned len, ocfs2_write_type_t type,
840 struct buffer_head *di_bh)
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700841{
tao.ma@oracle.com30b85482007-09-06 08:02:25 +0800842 u32 cend;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700843 struct ocfs2_write_ctxt *wc;
844
845 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
846 if (!wc)
847 return -ENOMEM;
848
849 wc->w_cpos = pos >> osb->s_clustersize_bits;
Sunil Mushrane7432672009-08-06 16:12:58 -0700850 wc->w_first_new_cpos = UINT_MAX;
tao.ma@oracle.com30b85482007-09-06 08:02:25 +0800851 cend = (pos + len - 1) >> osb->s_clustersize_bits;
852 wc->w_clen = cend - wc->w_cpos + 1;
Mark Fasheh607d44a2007-05-09 15:14:45 -0700853 get_bh(di_bh);
854 wc->w_di_bh = di_bh;
Ryan Dingc1ad1e32016-03-25 14:20:52 -0700855 wc->w_type = type;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700856
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300857 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700858 wc->w_large_pages = 1;
859 else
860 wc->w_large_pages = 0;
861
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700862 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
Ryan Ding4506cfb2016-03-25 14:21:06 -0700863 INIT_LIST_HEAD(&wc->w_unwritten_list);
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700864
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700865 *wcp = wc;
866
867 return 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800868}
869
870/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700871 * If a page has any new buffers, zero them out here, and mark them uptodate
872 * and dirty so they'll be written out (in order to prevent uninitialised
873 * block data from leaking). And clear the new bit.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800874 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700875static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
Mark Fasheh9517bac2007-02-09 20:24:12 -0800876{
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700877 unsigned int block_start, block_end;
878 struct buffer_head *head, *bh;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800879
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700880 BUG_ON(!PageLocked(page));
881 if (!page_has_buffers(page))
882 return;
883
884 bh = head = page_buffers(page);
885 block_start = 0;
886 do {
887 block_end = block_start + bh->b_size;
888
889 if (buffer_new(bh)) {
890 if (block_end > from && block_start < to) {
891 if (!PageUptodate(page)) {
892 unsigned start, end;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700893
894 start = max(from, block_start);
895 end = min(to, block_end);
896
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800897 zero_user_segment(page, start, end);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700898 set_buffer_uptodate(bh);
899 }
900
901 clear_buffer_new(bh);
902 mark_buffer_dirty(bh);
903 }
904 }
905
906 block_start = block_end;
907 bh = bh->b_this_page;
908 } while (bh != head);
909}
910
911/*
912 * Only called when we have a failure during allocating write to write
913 * zero's to the newly allocated region.
914 */
915static void ocfs2_write_failure(struct inode *inode,
916 struct ocfs2_write_ctxt *wc,
917 loff_t user_pos, unsigned user_len)
918{
919 int i;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300920 unsigned from = user_pos & (PAGE_SIZE - 1),
Mark Fasheh5c26a7b2007-09-18 17:49:29 -0700921 to = user_pos + user_len;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700922 struct page *tmppage;
923
Ryan Ding65c4db82016-03-25 14:20:58 -0700924 if (wc->w_target_page)
925 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700926
927 for(i = 0; i < wc->w_num_pages; i++) {
928 tmppage = wc->w_pages[i];
929
Ryan Ding65c4db82016-03-25 14:20:58 -0700930 if (tmppage && page_has_buffers(tmppage)) {
Mark Fasheh53ef99c2008-11-18 16:53:43 -0800931 if (ocfs2_should_order_data(inode))
Joseph Qibbd0f322019-09-23 15:33:08 -0700932 ocfs2_jbd2_inode_add_write(wc->w_handle, inode,
933 user_pos, user_len);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700934
Sunil Mushran961cecb2008-07-16 17:22:22 -0700935 block_commit_write(tmppage, from, to);
936 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700937 }
938}
939
940static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
941 struct ocfs2_write_ctxt *wc,
942 struct page *page, u32 cpos,
943 loff_t user_pos, unsigned user_len,
944 int new)
945{
946 int ret;
947 unsigned int map_from = 0, map_to = 0;
948 unsigned int cluster_start, cluster_end;
949 unsigned int user_data_from = 0, user_data_to = 0;
950
951 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800952 &cluster_start, &cluster_end);
953
Goldwyn Rodrigues272b62c2011-02-17 09:44:40 -0600954 /* treat the write as new if the a hole/lseek spanned across
955 * the page boundary.
956 */
957 new = new | ((i_size_read(inode) <= page_offset(page)) &&
958 (page_offset(page) <= user_pos));
959
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700960 if (page == wc->w_target_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300961 map_from = user_pos & (PAGE_SIZE - 1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700962 map_to = map_from + user_len;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800963
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700964 if (new)
965 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
966 cluster_start, cluster_end,
967 new);
968 else
969 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
970 map_from, map_to, new);
971 if (ret) {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800972 mlog_errno(ret);
973 goto out;
974 }
975
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700976 user_data_from = map_from;
977 user_data_to = map_to;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800978 if (new) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700979 map_from = cluster_start;
980 map_to = cluster_end;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800981 }
982 } else {
983 /*
984 * If we haven't allocated the new page yet, we
985 * shouldn't be writing it out without copying user
986 * data. This is likely a math error from the caller.
987 */
988 BUG_ON(!new);
989
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700990 map_from = cluster_start;
991 map_to = cluster_end;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800992
993 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700994 cluster_start, cluster_end, new);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800995 if (ret) {
996 mlog_errno(ret);
997 goto out;
998 }
999 }
1000
1001 /*
1002 * Parts of newly allocated pages need to be zero'd.
1003 *
1004 * Above, we have also rewritten 'to' and 'from' - as far as
1005 * the rest of the function is concerned, the entire cluster
1006 * range inside of a page needs to be written.
1007 *
1008 * We can skip this if the page is up to date - it's already
1009 * been zero'd from being read in as a hole.
1010 */
1011 if (new && !PageUptodate(page))
1012 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001013 cpos, user_data_from, user_data_to);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001014
1015 flush_dcache_page(page);
1016
Mark Fasheh9517bac2007-02-09 20:24:12 -08001017out:
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001018 return ret;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001019}
1020
1021/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001022 * This function will only grab one clusters worth of pages.
Mark Fasheh9517bac2007-02-09 20:24:12 -08001023 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001024static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1025 struct ocfs2_write_ctxt *wc,
Joel Becker693c2412010-07-02 17:20:27 -07001026 u32 cpos, loff_t user_pos,
1027 unsigned user_len, int new,
Mark Fasheh7307de82007-05-09 15:16:19 -07001028 struct page *mmap_page)
Mark Fasheh9517bac2007-02-09 20:24:12 -08001029{
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001030 int ret = 0, i;
Joel Becker693c2412010-07-02 17:20:27 -07001031 unsigned long start, target_index, end_index, index;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001032 struct inode *inode = mapping->host;
Joel Becker693c2412010-07-02 17:20:27 -07001033 loff_t last_byte;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001034
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001035 target_index = user_pos >> PAGE_SHIFT;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001036
1037 /*
1038 * Figure out how many pages we'll be manipulating here. For
Mark Fasheh60b11392007-02-16 11:46:50 -08001039 * non allocating write, we just change the one
Joel Becker693c2412010-07-02 17:20:27 -07001040 * page. Otherwise, we'll need a whole clusters worth. If we're
1041 * writing past i_size, we only need enough pages to cover the
1042 * last page of the write.
Mark Fasheh9517bac2007-02-09 20:24:12 -08001043 */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001044 if (new) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001045 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1046 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
Joel Becker693c2412010-07-02 17:20:27 -07001047 /*
1048 * We need the index *past* the last page we could possibly
1049 * touch. This is the page past the end of the write or
1050 * i_size, whichever is greater.
1051 */
1052 last_byte = max(user_pos + user_len, i_size_read(inode));
1053 BUG_ON(last_byte < 1);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001054 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
Joel Becker693c2412010-07-02 17:20:27 -07001055 if ((start + wc->w_num_pages) > end_index)
1056 wc->w_num_pages = end_index - start;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001057 } else {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001058 wc->w_num_pages = 1;
1059 start = target_index;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001060 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001061 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001062
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001063 for(i = 0; i < wc->w_num_pages; i++) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001064 index = start + i;
1065
Ryan Ding65c4db82016-03-25 14:20:58 -07001066 if (index >= target_index && index <= end_index &&
1067 wc->w_type == OCFS2_WRITE_MMAP) {
Mark Fasheh7307de82007-05-09 15:16:19 -07001068 /*
1069 * ocfs2_pagemkwrite() is a little different
1070 * and wants us to directly use the page
1071 * passed in.
1072 */
1073 lock_page(mmap_page);
1074
Wengang Wang5cffff92011-07-24 10:36:54 -07001075 /* Exit and let the caller retry */
Mark Fasheh7307de82007-05-09 15:16:19 -07001076 if (mmap_page->mapping != mapping) {
Wengang Wang5cffff92011-07-24 10:36:54 -07001077 WARN_ON(mmap_page->mapping);
Mark Fasheh7307de82007-05-09 15:16:19 -07001078 unlock_page(mmap_page);
Wengang Wang5cffff92011-07-24 10:36:54 -07001079 ret = -EAGAIN;
Mark Fasheh7307de82007-05-09 15:16:19 -07001080 goto out;
1081 }
1082
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001083 get_page(mmap_page);
Mark Fasheh7307de82007-05-09 15:16:19 -07001084 wc->w_pages[i] = mmap_page;
Wengang Wang5cffff92011-07-24 10:36:54 -07001085 wc->w_target_locked = true;
Ryan Ding65c4db82016-03-25 14:20:58 -07001086 } else if (index >= target_index && index <= end_index &&
1087 wc->w_type == OCFS2_WRITE_DIRECT) {
1088 /* Direct write has no mapping page. */
1089 wc->w_pages[i] = NULL;
1090 continue;
Mark Fasheh7307de82007-05-09 15:16:19 -07001091 } else {
1092 wc->w_pages[i] = find_or_create_page(mapping, index,
1093 GFP_NOFS);
1094 if (!wc->w_pages[i]) {
1095 ret = -ENOMEM;
1096 mlog_errno(ret);
1097 goto out;
1098 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001099 }
Jan Kara12695292013-02-21 16:42:57 -08001100 wait_for_stable_page(wc->w_pages[i]);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001101
1102 if (index == target_index)
1103 wc->w_target_page = wc->w_pages[i];
Mark Fasheh9517bac2007-02-09 20:24:12 -08001104 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001105out:
Wengang Wang5cffff92011-07-24 10:36:54 -07001106 if (ret)
1107 wc->w_target_locked = false;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001108 return ret;
1109}
1110
1111/*
1112 * Prepare a single cluster for write one cluster into the file.
1113 */
1114static int ocfs2_write_cluster(struct address_space *mapping,
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001115 u32 *phys, unsigned int new,
Ryan Dingb46637d2016-03-25 14:20:55 -07001116 unsigned int clear_unwritten,
Sunil Mushrane7432672009-08-06 16:12:58 -07001117 unsigned int should_zero,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001118 struct ocfs2_alloc_context *data_ac,
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001119 struct ocfs2_alloc_context *meta_ac,
1120 struct ocfs2_write_ctxt *wc, u32 cpos,
1121 loff_t user_pos, unsigned user_len)
1122{
Ryan Dingb46637d2016-03-25 14:20:55 -07001123 int ret, i;
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001124 u64 p_blkno;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001125 struct inode *inode = mapping->host;
Joel Beckerf99b9b72008-08-20 19:36:33 -07001126 struct ocfs2_extent_tree et;
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001127 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001128
Mark Fasheh9517bac2007-02-09 20:24:12 -08001129 if (new) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001130 u32 tmp_pos;
1131
Mark Fasheh9517bac2007-02-09 20:24:12 -08001132 /*
1133 * This is safe to call with the page locks - it won't take
1134 * any additional semaphores or cluster locks.
1135 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001136 tmp_pos = cpos;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08001137 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
Ryan Dingb46637d2016-03-25 14:20:55 -07001138 &tmp_pos, 1, !clear_unwritten,
1139 wc->w_di_bh, wc->w_handle,
1140 data_ac, meta_ac, NULL);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001141 /*
1142 * This shouldn't happen because we must have already
1143 * calculated the correct meta data allocation required. The
1144 * internal tree allocation code should know how to increase
1145 * transaction credits itself.
1146 *
1147 * If need be, we could handle -EAGAIN for a
1148 * RESTART_TRANS here.
1149 */
1150 mlog_bug_on_msg(ret == -EAGAIN,
1151 "Inode %llu: EAGAIN return during allocation.\n",
1152 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1153 if (ret < 0) {
1154 mlog_errno(ret);
1155 goto out;
1156 }
Ryan Dingb46637d2016-03-25 14:20:55 -07001157 } else if (clear_unwritten) {
Joel Becker5e404e92009-02-13 03:54:22 -08001158 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1159 wc->w_di_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -07001160 ret = ocfs2_mark_extent_written(inode, &et,
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001161 wc->w_handle, cpos, 1, *phys,
Joel Beckerf99b9b72008-08-20 19:36:33 -07001162 meta_ac, &wc->w_dealloc);
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001163 if (ret < 0) {
1164 mlog_errno(ret);
1165 goto out;
1166 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001167 }
1168
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001169 /*
1170 * The only reason this should fail is due to an inability to
1171 * find the extent added.
1172 */
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001173 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001174 if (ret < 0) {
jiangyiwen61fb9ea2014-12-10 15:42:02 -08001175 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001176 "at logical cluster %u",
1177 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001178 goto out;
1179 }
1180
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001181 BUG_ON(*phys == 0);
1182
1183 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1184 if (!should_zero)
1185 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001186
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001187 for(i = 0; i < wc->w_num_pages; i++) {
1188 int tmpret;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001189
Ryan Ding65c4db82016-03-25 14:20:58 -07001190 /* This is the direct io target page. */
1191 if (wc->w_pages[i] == NULL) {
1192 p_blkno++;
1193 continue;
1194 }
1195
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001196 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1197 wc->w_pages[i], cpos,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001198 user_pos, user_len,
1199 should_zero);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001200 if (tmpret) {
1201 mlog_errno(tmpret);
1202 if (ret == 0)
Wengang Wangcbfa9632009-07-13 11:38:23 +08001203 ret = tmpret;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001204 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001205 }
1206
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001207 /*
1208 * We only have cleanup to do in case of allocating write.
1209 */
1210 if (ret && new)
1211 ocfs2_write_failure(inode, wc, user_pos, user_len);
1212
Mark Fasheh9517bac2007-02-09 20:24:12 -08001213out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08001214
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001215 return ret;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001216}
1217
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001218static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1219 struct ocfs2_alloc_context *data_ac,
1220 struct ocfs2_alloc_context *meta_ac,
1221 struct ocfs2_write_ctxt *wc,
1222 loff_t pos, unsigned len)
1223{
1224 int ret, i;
Mark Fashehdb562462007-09-17 09:06:29 -07001225 loff_t cluster_off;
1226 unsigned int local_len = len;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001227 struct ocfs2_write_cluster_desc *desc;
Mark Fashehdb562462007-09-17 09:06:29 -07001228 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001229
1230 for (i = 0; i < wc->w_clen; i++) {
1231 desc = &wc->w_desc[i];
1232
Mark Fashehdb562462007-09-17 09:06:29 -07001233 /*
1234 * We have to make sure that the total write passed in
1235 * doesn't extend past a single cluster.
1236 */
1237 local_len = len;
1238 cluster_off = pos & (osb->s_clustersize - 1);
1239 if ((cluster_off + local_len) > osb->s_clustersize)
1240 local_len = osb->s_clustersize - cluster_off;
1241
Ryan Ding2de6a3c2016-03-25 14:21:03 -07001242 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
Ryan Dingb46637d2016-03-25 14:20:55 -07001243 desc->c_new,
1244 desc->c_clear_unwritten,
Sunil Mushrane7432672009-08-06 16:12:58 -07001245 desc->c_needs_zero,
1246 data_ac, meta_ac,
Mark Fashehdb562462007-09-17 09:06:29 -07001247 wc, desc->c_cpos, pos, local_len);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001248 if (ret) {
1249 mlog_errno(ret);
1250 goto out;
1251 }
Mark Fashehdb562462007-09-17 09:06:29 -07001252
1253 len -= local_len;
1254 pos += local_len;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001255 }
1256
1257 ret = 0;
1258out:
1259 return ret;
1260}
1261
Mark Fasheh9517bac2007-02-09 20:24:12 -08001262/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001263 * ocfs2_write_end() wants to know which parts of the target page it
1264 * should complete the write on. It's easiest to compute them ahead of
1265 * time when a more complete view of the write is available.
Mark Fasheh9517bac2007-02-09 20:24:12 -08001266 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001267static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1268 struct ocfs2_write_ctxt *wc,
1269 loff_t pos, unsigned len, int alloc)
Mark Fasheh9517bac2007-02-09 20:24:12 -08001270{
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001271 struct ocfs2_write_cluster_desc *desc;
1272
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001273 wc->w_target_from = pos & (PAGE_SIZE - 1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001274 wc->w_target_to = wc->w_target_from + len;
1275
1276 if (alloc == 0)
1277 return;
1278
1279 /*
1280 * Allocating write - we may have different boundaries based
1281 * on page size and cluster size.
1282 *
1283 * NOTE: We can no longer compute one value from the other as
1284 * the actual write length and user provided length may be
1285 * different.
1286 */
1287
1288 if (wc->w_large_pages) {
1289 /*
1290 * We only care about the 1st and last cluster within
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001291 * our range and whether they should be zero'd or not. Either
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001292 * value may be extended out to the start/end of a
1293 * newly allocated cluster.
1294 */
1295 desc = &wc->w_desc[0];
Sunil Mushrane7432672009-08-06 16:12:58 -07001296 if (desc->c_needs_zero)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001297 ocfs2_figure_cluster_boundaries(osb,
1298 desc->c_cpos,
1299 &wc->w_target_from,
1300 NULL);
1301
1302 desc = &wc->w_desc[wc->w_clen - 1];
Sunil Mushrane7432672009-08-06 16:12:58 -07001303 if (desc->c_needs_zero)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001304 ocfs2_figure_cluster_boundaries(osb,
1305 desc->c_cpos,
1306 NULL,
1307 &wc->w_target_to);
1308 } else {
1309 wc->w_target_from = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001310 wc->w_target_to = PAGE_SIZE;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001311 }
1312}
1313
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001314/*
Ryan Ding4506cfb2016-03-25 14:21:06 -07001315 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to
1316 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1317 * by the direct io procedure.
1318 * If this is a new extent that allocated by direct io, we should mark it in
1319 * the ip_unwritten_list.
1320 */
1321static int ocfs2_unwritten_check(struct inode *inode,
1322 struct ocfs2_write_ctxt *wc,
1323 struct ocfs2_write_cluster_desc *desc)
1324{
1325 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Ryan Dingc15471f2016-03-25 14:21:09 -07001326 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
Ryan Ding4506cfb2016-03-25 14:21:06 -07001327 int ret = 0;
1328
1329 if (!desc->c_needs_zero)
1330 return 0;
1331
1332retry:
1333 spin_lock(&oi->ip_lock);
1334 /* Needs not to zero no metter buffer or direct. The one who is zero
1335 * the cluster is doing zero. And he will clear unwritten after all
1336 * cluster io finished. */
Ryan Dingc15471f2016-03-25 14:21:09 -07001337 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1338 if (desc->c_cpos == ue->ue_cpos) {
Ryan Ding4506cfb2016-03-25 14:21:06 -07001339 BUG_ON(desc->c_new);
1340 desc->c_needs_zero = 0;
1341 desc->c_clear_unwritten = 0;
1342 goto unlock;
1343 }
1344 }
1345
1346 if (wc->w_type != OCFS2_WRITE_DIRECT)
1347 goto unlock;
1348
1349 if (new == NULL) {
1350 spin_unlock(&oi->ip_lock);
1351 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1352 GFP_NOFS);
1353 if (new == NULL) {
1354 ret = -ENOMEM;
1355 goto out;
1356 }
1357 goto retry;
1358 }
1359 /* This direct write will doing zero. */
1360 new->ue_cpos = desc->c_cpos;
1361 new->ue_phys = desc->c_phys;
1362 desc->c_clear_unwritten = 0;
1363 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1364 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
Changwei Ge63de8bd2018-01-31 16:15:02 -08001365 wc->w_unwritten_count++;
Ryan Ding4506cfb2016-03-25 14:21:06 -07001366 new = NULL;
1367unlock:
1368 spin_unlock(&oi->ip_lock);
1369out:
Ding Xiang0ae1c2d2018-10-26 15:02:48 -07001370 kfree(new);
Ryan Ding4506cfb2016-03-25 14:21:06 -07001371 return ret;
1372}
1373
1374/*
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001375 * Populate each single-cluster write descriptor in the write context
1376 * with information about the i/o to be done.
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001377 *
1378 * Returns the number of clusters that will have to be allocated, as
1379 * well as a worst case estimate of the number of extent records that
1380 * would have to be created during a write to an unwritten region.
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001381 */
1382static int ocfs2_populate_write_desc(struct inode *inode,
1383 struct ocfs2_write_ctxt *wc,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001384 unsigned int *clusters_to_alloc,
1385 unsigned int *extents_to_split)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001386{
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001387 int ret;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001388 struct ocfs2_write_cluster_desc *desc;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001389 unsigned int num_clusters = 0;
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001390 unsigned int ext_flags = 0;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001391 u32 phys = 0;
1392 int i;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001393
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001394 *clusters_to_alloc = 0;
1395 *extents_to_split = 0;
1396
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001397 for (i = 0; i < wc->w_clen; i++) {
1398 desc = &wc->w_desc[i];
1399 desc->c_cpos = wc->w_cpos + i;
1400
1401 if (num_clusters == 0) {
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001402 /*
1403 * Need to look up the next extent record.
1404 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001405 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001406 &num_clusters, &ext_flags);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001407 if (ret) {
1408 mlog_errno(ret);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001409 goto out;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001410 }
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001411
Tao Ma293b2f72009-08-25 08:02:48 +08001412 /* We should already CoW the refcountd extent. */
1413 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1414
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001415 /*
1416 * Assume worst case - that we're writing in
1417 * the middle of the extent.
1418 *
1419 * We can assume that the write proceeds from
1420 * left to right, in which case the extent
1421 * insert code is smart enough to coalesce the
1422 * next splits into the previous records created.
1423 */
1424 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1425 *extents_to_split = *extents_to_split + 2;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001426 } else if (phys) {
1427 /*
1428 * Only increment phys if it doesn't describe
1429 * a hole.
1430 */
1431 phys++;
1432 }
1433
Sunil Mushrane7432672009-08-06 16:12:58 -07001434 /*
1435 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1436 * file that got extended. w_first_new_cpos tells us
1437 * where the newly allocated clusters are so we can
1438 * zero them.
1439 */
1440 if (desc->c_cpos >= wc->w_first_new_cpos) {
1441 BUG_ON(phys == 0);
1442 desc->c_needs_zero = 1;
1443 }
1444
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001445 desc->c_phys = phys;
1446 if (phys == 0) {
1447 desc->c_new = 1;
Sunil Mushrane7432672009-08-06 16:12:58 -07001448 desc->c_needs_zero = 1;
Ryan Dingb46637d2016-03-25 14:20:55 -07001449 desc->c_clear_unwritten = 1;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001450 *clusters_to_alloc = *clusters_to_alloc + 1;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001451 }
Sunil Mushrane7432672009-08-06 16:12:58 -07001452
1453 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
Ryan Dingb46637d2016-03-25 14:20:55 -07001454 desc->c_clear_unwritten = 1;
Sunil Mushrane7432672009-08-06 16:12:58 -07001455 desc->c_needs_zero = 1;
1456 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001457
Ryan Ding4506cfb2016-03-25 14:21:06 -07001458 ret = ocfs2_unwritten_check(inode, wc, desc);
1459 if (ret) {
1460 mlog_errno(ret);
1461 goto out;
1462 }
1463
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001464 num_clusters--;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001465 }
1466
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001467 ret = 0;
1468out:
1469 return ret;
1470}
1471
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001472static int ocfs2_write_begin_inline(struct address_space *mapping,
1473 struct inode *inode,
1474 struct ocfs2_write_ctxt *wc)
1475{
1476 int ret;
1477 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1478 struct page *page;
1479 handle_t *handle;
1480 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1481
Junxiao Bif775da22014-10-09 15:25:15 -07001482 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1483 if (IS_ERR(handle)) {
1484 ret = PTR_ERR(handle);
1485 mlog_errno(ret);
1486 goto out;
1487 }
1488
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001489 page = find_or_create_page(mapping, 0, GFP_NOFS);
1490 if (!page) {
Junxiao Bif775da22014-10-09 15:25:15 -07001491 ocfs2_commit_trans(osb, handle);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001492 ret = -ENOMEM;
1493 mlog_errno(ret);
1494 goto out;
1495 }
1496 /*
1497 * If we don't set w_num_pages then this page won't get unlocked
1498 * and freed on cleanup of the write context.
1499 */
1500 wc->w_pages[0] = wc->w_target_page = page;
1501 wc->w_num_pages = 1;
1502
Joel Becker0cf2f762009-02-12 16:41:25 -08001503 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001504 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001505 if (ret) {
1506 ocfs2_commit_trans(osb, handle);
1507
1508 mlog_errno(ret);
1509 goto out;
1510 }
1511
1512 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1513 ocfs2_set_inode_data_inline(inode, di);
1514
1515 if (!PageUptodate(page)) {
1516 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1517 if (ret) {
1518 ocfs2_commit_trans(osb, handle);
1519
1520 goto out;
1521 }
1522 }
1523
1524 wc->w_handle = handle;
1525out:
1526 return ret;
1527}
1528
1529int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1530{
1531 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1532
Mark Fasheh0d8a4e02007-11-20 11:48:41 -08001533 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001534 return 1;
1535 return 0;
1536}
1537
1538static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1539 struct inode *inode, loff_t pos,
1540 unsigned len, struct page *mmap_page,
1541 struct ocfs2_write_ctxt *wc)
1542{
1543 int ret, written = 0;
1544 loff_t end = pos + len;
1545 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Tiger Yangd9ae49d2009-03-05 11:06:15 +08001546 struct ocfs2_dinode *di = NULL;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001547
Tao Ma95581562011-02-22 21:33:59 +08001548 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1549 len, (unsigned long long)pos,
1550 oi->ip_dyn_features);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001551
1552 /*
1553 * Handle inodes which already have inline data 1st.
1554 */
1555 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1556 if (mmap_page == NULL &&
1557 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1558 goto do_inline_write;
1559
1560 /*
1561 * The write won't fit - we have to give this inode an
1562 * inline extent list now.
1563 */
1564 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1565 if (ret)
1566 mlog_errno(ret);
1567 goto out;
1568 }
1569
1570 /*
1571 * Check whether the inode can accept inline data.
1572 */
1573 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1574 return 0;
1575
1576 /*
1577 * Check whether the write can fit.
1578 */
Tiger Yangd9ae49d2009-03-05 11:06:15 +08001579 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1580 if (mmap_page ||
1581 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001582 return 0;
1583
1584do_inline_write:
1585 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1586 if (ret) {
1587 mlog_errno(ret);
1588 goto out;
1589 }
1590
1591 /*
1592 * This signals to the caller that the data can be written
1593 * inline.
1594 */
1595 written = 1;
1596out:
1597 return written ? written : ret;
1598}
1599
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001600/*
1601 * This function only does anything for file systems which can't
1602 * handle sparse files.
1603 *
1604 * What we want to do here is fill in any hole between the current end
1605 * of allocation and the end of our write. That way the rest of the
1606 * write path can treat it as an non-allocating write, which has no
1607 * special case code for sparse/nonsparse files.
1608 */
Joel Becker56934862010-07-01 15:13:31 -07001609static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1610 struct buffer_head *di_bh,
1611 loff_t pos, unsigned len,
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001612 struct ocfs2_write_ctxt *wc)
1613{
1614 int ret;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001615 loff_t newsize = pos + len;
1616
Joel Becker56934862010-07-01 15:13:31 -07001617 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001618
1619 if (newsize <= i_size_read(inode))
1620 return 0;
1621
Joel Becker56934862010-07-01 15:13:31 -07001622 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001623 if (ret)
1624 mlog_errno(ret);
1625
Ryan Ding46e62552016-03-25 14:21:01 -07001626 /* There is no wc if this is call from direct. */
1627 if (wc)
1628 wc->w_first_new_cpos =
1629 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
Sunil Mushrane7432672009-08-06 16:12:58 -07001630
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001631 return ret;
1632}
1633
Joel Becker56934862010-07-01 15:13:31 -07001634static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1635 loff_t pos)
1636{
1637 int ret = 0;
1638
1639 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1640 if (pos > i_size_read(inode))
1641 ret = ocfs2_zero_extend(inode, di_bh, pos);
1642
1643 return ret;
1644}
1645
Ryan Dingc1ad1e32016-03-25 14:20:52 -07001646int ocfs2_write_begin_nolock(struct address_space *mapping,
1647 loff_t pos, unsigned len, ocfs2_write_type_t type,
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001648 struct page **pagep, void **fsdata,
1649 struct buffer_head *di_bh, struct page *mmap_page)
1650{
Sunil Mushrane7432672009-08-06 16:12:58 -07001651 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
Tao Ma50308d82010-11-04 15:14:11 +08001652 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001653 struct ocfs2_write_ctxt *wc;
1654 struct inode *inode = mapping->host;
1655 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1656 struct ocfs2_dinode *di;
1657 struct ocfs2_alloc_context *data_ac = NULL;
1658 struct ocfs2_alloc_context *meta_ac = NULL;
1659 handle_t *handle;
Joel Beckerf99b9b72008-08-20 19:36:33 -07001660 struct ocfs2_extent_tree et;
Tao Ma50308d82010-11-04 15:14:11 +08001661 int try_free = 1, ret1;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001662
Tao Ma50308d82010-11-04 15:14:11 +08001663try_again:
Ryan Dingc1ad1e32016-03-25 14:20:52 -07001664 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001665 if (ret) {
1666 mlog_errno(ret);
1667 return ret;
1668 }
1669
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001670 if (ocfs2_supports_inline_data(osb)) {
1671 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1672 mmap_page, wc);
1673 if (ret == 1) {
1674 ret = 0;
1675 goto success;
1676 }
1677 if (ret < 0) {
1678 mlog_errno(ret);
1679 goto out;
1680 }
1681 }
1682
Ryan Ding46e62552016-03-25 14:21:01 -07001683 /* Direct io change i_size late, should not zero tail here. */
1684 if (type != OCFS2_WRITE_DIRECT) {
1685 if (ocfs2_sparse_alloc(osb))
1686 ret = ocfs2_zero_tail(inode, di_bh, pos);
1687 else
1688 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1689 len, wc);
1690 if (ret) {
1691 mlog_errno(ret);
1692 goto out;
1693 }
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001694 }
1695
Tao Ma293b2f72009-08-25 08:02:48 +08001696 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1697 if (ret < 0) {
1698 mlog_errno(ret);
1699 goto out;
1700 } else if (ret == 1) {
Tao Ma50308d82010-11-04 15:14:11 +08001701 clusters_need = wc->w_clen;
Tiger Yangc7dd3392013-08-13 16:00:58 -07001702 ret = ocfs2_refcount_cow(inode, di_bh,
Tao Ma37f8a2b2009-08-26 09:47:28 +08001703 wc->w_cpos, wc->w_clen, UINT_MAX);
Tao Ma293b2f72009-08-25 08:02:48 +08001704 if (ret) {
1705 mlog_errno(ret);
1706 goto out;
1707 }
1708 }
1709
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001710 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1711 &extents_to_split);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001712 if (ret) {
1713 mlog_errno(ret);
1714 goto out;
1715 }
Tao Ma50308d82010-11-04 15:14:11 +08001716 clusters_need += clusters_to_alloc;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001717
1718 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1719
Tao Ma95581562011-02-22 21:33:59 +08001720 trace_ocfs2_write_begin_nolock(
1721 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1722 (long long)i_size_read(inode),
1723 le32_to_cpu(di->i_clusters),
Ryan Dingc1ad1e32016-03-25 14:20:52 -07001724 pos, len, type, mmap_page,
Tao Ma95581562011-02-22 21:33:59 +08001725 clusters_to_alloc, extents_to_split);
1726
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001727 /*
1728 * We set w_target_from, w_target_to here so that
1729 * ocfs2_write_end() knows which range in the target page to
1730 * write out. An allocation requires that we write the entire
1731 * cluster range.
1732 */
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001733 if (clusters_to_alloc || extents_to_split) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001734 /*
1735 * XXX: We are stretching the limits of
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001736 * ocfs2_lock_allocators(). It greatly over-estimates
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001737 * the work to be done.
1738 */
Joel Becker5e404e92009-02-13 03:54:22 -08001739 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1740 wc->w_di_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -07001741 ret = ocfs2_lock_allocators(inode, &et,
Tao Ma231b87d2008-08-18 17:38:42 +08001742 clusters_to_alloc, extents_to_split,
Joel Beckerf99b9b72008-08-20 19:36:33 -07001743 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001744 if (ret) {
1745 mlog_errno(ret);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001746 goto out;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001747 }
1748
Mark Fasheh4fe370a2009-12-07 13:15:40 -08001749 if (data_ac)
1750 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1751
Tao Ma811f9332008-08-18 17:38:43 +08001752 credits = ocfs2_calc_extend_credits(inode->i_sb,
Goldwyn Rodrigues06f9da62013-11-12 15:06:52 -08001753 &di->id2.i_list);
Ryan Ding46e62552016-03-25 14:21:01 -07001754 } else if (type == OCFS2_WRITE_DIRECT)
1755 /* direct write needs not to start trans if no extents alloc. */
1756 goto success;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001757
Sunil Mushrane7432672009-08-06 16:12:58 -07001758 /*
1759 * We have to zero sparse allocated clusters, unwritten extent clusters,
1760 * and non-sparse clusters we just extended. For non-sparse writes,
1761 * we know zeros will only be needed in the first and/or last cluster.
1762 */
Ryan Ding4506cfb2016-03-25 14:21:06 -07001763 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1764 wc->w_desc[wc->w_clen - 1].c_needs_zero))
Sunil Mushrane7432672009-08-06 16:12:58 -07001765 cluster_of_pages = 1;
1766 else
1767 cluster_of_pages = 0;
1768
1769 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001770
Mark Fasheh9517bac2007-02-09 20:24:12 -08001771 handle = ocfs2_start_trans(osb, credits);
1772 if (IS_ERR(handle)) {
1773 ret = PTR_ERR(handle);
1774 mlog_errno(ret);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001775 goto out;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001776 }
1777
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001778 wc->w_handle = handle;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001779
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001780 if (clusters_to_alloc) {
1781 ret = dquot_alloc_space_nodirty(inode,
1782 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1783 if (ret)
1784 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02001785 }
yangwenfang7f27ec92015-09-04 15:44:45 -07001786
Joel Becker0cf2f762009-02-12 16:41:25 -08001787 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001788 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001789 if (ret) {
1790 mlog_errno(ret);
Jan Karaa90714c2008-10-09 19:38:40 +02001791 goto out_quota;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001792 }
1793
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001794 /*
1795 * Fill our page array first. That way we've grabbed enough so
1796 * that we can zero and flush if we error after adding the
1797 * extent.
1798 */
Joel Becker693c2412010-07-02 17:20:27 -07001799 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
Sunil Mushrane7432672009-08-06 16:12:58 -07001800 cluster_of_pages, mmap_page);
Joseph Qie07bf002022-01-14 14:03:38 -08001801 if (ret) {
1802 /*
1803 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
1804 * the target page. In this case, we exit with no error and no target
1805 * page. This will trigger the caller, page_mkwrite(), to re-try
1806 * the operation.
1807 */
1808 if (type == OCFS2_WRITE_MMAP && ret == -EAGAIN) {
1809 BUG_ON(wc->w_target_page);
1810 ret = 0;
1811 goto out_quota;
1812 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001813
Joseph Qie07bf002022-01-14 14:03:38 -08001814 mlog_errno(ret);
Wengang Wang5cffff92011-07-24 10:36:54 -07001815 goto out_quota;
1816 }
1817
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001818 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1819 len);
1820 if (ret) {
1821 mlog_errno(ret);
Jan Karaa90714c2008-10-09 19:38:40 +02001822 goto out_quota;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001823 }
1824
1825 if (data_ac)
1826 ocfs2_free_alloc_context(data_ac);
1827 if (meta_ac)
1828 ocfs2_free_alloc_context(meta_ac);
1829
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001830success:
Ryan Ding65c4db82016-03-25 14:20:58 -07001831 if (pagep)
1832 *pagep = wc->w_target_page;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001833 *fsdata = wc;
1834 return 0;
Jan Karaa90714c2008-10-09 19:38:40 +02001835out_quota:
1836 if (clusters_to_alloc)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001837 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02001838 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
Mark Fasheh9517bac2007-02-09 20:24:12 -08001839out_commit:
1840 ocfs2_commit_trans(osb, handle);
1841
Mark Fasheh9517bac2007-02-09 20:24:12 -08001842out:
Eric Renc33f0782016-09-30 15:11:32 -07001843 /*
1844 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1845 * even in case of error here like ENOSPC and ENOMEM. So, we need
1846 * to unlock the target page manually to prevent deadlocks when
1847 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1848 * to VM code.
1849 */
1850 if (wc->w_target_locked)
1851 unlock_page(mmap_page);
1852
Ryan Ding4506cfb2016-03-25 14:21:06 -07001853 ocfs2_free_write_ctxt(inode, wc);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001854
Xue jiufeib1214e42013-11-12 15:07:06 -08001855 if (data_ac) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001856 ocfs2_free_alloc_context(data_ac);
Xue jiufeib1214e42013-11-12 15:07:06 -08001857 data_ac = NULL;
1858 }
1859 if (meta_ac) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001860 ocfs2_free_alloc_context(meta_ac);
Xue jiufeib1214e42013-11-12 15:07:06 -08001861 meta_ac = NULL;
1862 }
Tao Ma50308d82010-11-04 15:14:11 +08001863
1864 if (ret == -ENOSPC && try_free) {
1865 /*
1866 * Try to free some truncate log so that we can have enough
1867 * clusters to allocate.
1868 */
1869 try_free = 0;
1870
1871 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1872 if (ret1 == 1)
1873 goto try_again;
1874
1875 if (ret1 < 0)
1876 mlog_errno(ret1);
1877 }
1878
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001879 return ret;
1880}
Mark Fasheh9517bac2007-02-09 20:24:12 -08001881
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001882static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
Matthew Wilcox (Oracle)9d6b0cd2022-02-22 14:31:43 -05001883 loff_t pos, unsigned len,
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001884 struct page **pagep, void **fsdata)
Mark Fasheh607d44a2007-05-09 15:14:45 -07001885{
1886 int ret;
1887 struct buffer_head *di_bh = NULL;
1888 struct inode *inode = mapping->host;
1889
Mark Fashehe63aecb62007-10-18 15:30:42 -07001890 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001891 if (ret) {
1892 mlog_errno(ret);
1893 return ret;
1894 }
1895
1896 /*
1897 * Take alloc sem here to prevent concurrent lookups. That way
1898 * the mapping, zeroing and tree manipulation within
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -04001899 * ocfs2_write() will be safe against ->read_folio(). This
Mark Fasheh607d44a2007-05-09 15:14:45 -07001900 * should also serve to lock out allocation from a shared
1901 * writeable region.
1902 */
1903 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1904
Ryan Dingc1ad1e32016-03-25 14:20:52 -07001905 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1906 pagep, fsdata, di_bh, NULL);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001907 if (ret) {
1908 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001909 goto out_fail;
Mark Fasheh607d44a2007-05-09 15:14:45 -07001910 }
1911
1912 brelse(di_bh);
1913
1914 return 0;
1915
Mark Fasheh607d44a2007-05-09 15:14:45 -07001916out_fail:
1917 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1918
1919 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001920 ocfs2_inode_unlock(inode, 1);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001921
1922 return ret;
1923}
1924
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001925static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1926 unsigned len, unsigned *copied,
1927 struct ocfs2_dinode *di,
1928 struct ocfs2_write_ctxt *wc)
1929{
1930 void *kaddr;
1931
1932 if (unlikely(*copied < len)) {
1933 if (!PageUptodate(wc->w_target_page)) {
1934 *copied = 0;
1935 return;
1936 }
1937 }
1938
Cong Wangc4bc8dc2011-11-25 23:14:34 +08001939 kaddr = kmap_atomic(wc->w_target_page);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001940 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
Cong Wangc4bc8dc2011-11-25 23:14:34 +08001941 kunmap_atomic(kaddr);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001942
Tao Ma95581562011-02-22 21:33:59 +08001943 trace_ocfs2_write_end_inline(
1944 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001945 (unsigned long long)pos, *copied,
1946 le16_to_cpu(di->id2.i_data.id_count),
1947 le16_to_cpu(di->i_dyn_features));
1948}
1949
Mark Fasheh7307de82007-05-09 15:16:19 -07001950int ocfs2_write_end_nolock(struct address_space *mapping,
piaojun07f38d92016-12-12 16:41:17 -08001951 loff_t pos, unsigned len, unsigned copied, void *fsdata)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001952{
yangwenfang7f27ec92015-09-04 15:44:45 -07001953 int i, ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001954 unsigned from, to, start = pos & (PAGE_SIZE - 1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001955 struct inode *inode = mapping->host;
1956 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1957 struct ocfs2_write_ctxt *wc = fsdata;
1958 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1959 handle_t *handle = wc->w_handle;
1960 struct page *tmppage;
1961
Ryan Ding4506cfb2016-03-25 14:21:06 -07001962 BUG_ON(!list_empty(&wc->w_unwritten_list));
1963
Ryan Ding46e62552016-03-25 14:21:01 -07001964 if (handle) {
1965 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1966 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1967 if (ret) {
1968 copied = ret;
1969 mlog_errno(ret);
1970 goto out;
1971 }
yangwenfang7f27ec92015-09-04 15:44:45 -07001972 }
1973
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001974 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1975 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1976 goto out_write_size;
1977 }
1978
Ryan Ding65c4db82016-03-25 14:20:58 -07001979 if (unlikely(copied < len) && wc->w_target_page) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001980 if (!PageUptodate(wc->w_target_page))
1981 copied = 0;
1982
1983 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1984 start+len);
1985 }
Ryan Ding65c4db82016-03-25 14:20:58 -07001986 if (wc->w_target_page)
1987 flush_dcache_page(wc->w_target_page);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001988
1989 for(i = 0; i < wc->w_num_pages; i++) {
1990 tmppage = wc->w_pages[i];
1991
Ryan Ding65c4db82016-03-25 14:20:58 -07001992 /* This is the direct io target page. */
1993 if (tmppage == NULL)
1994 continue;
1995
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001996 if (tmppage == wc->w_target_page) {
1997 from = wc->w_target_from;
1998 to = wc->w_target_to;
1999
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002000 BUG_ON(from > PAGE_SIZE ||
2001 to > PAGE_SIZE ||
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002002 to < from);
2003 } else {
2004 /*
2005 * Pages adjacent to the target (if any) imply
2006 * a hole-filling write in which case we want
2007 * to flush their entire range.
2008 */
2009 from = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002010 to = PAGE_SIZE;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002011 }
2012
Sunil Mushran961cecb2008-07-16 17:22:22 -07002013 if (page_has_buffers(tmppage)) {
Joseph Qibbd0f322019-09-23 15:33:08 -07002014 if (handle && ocfs2_should_order_data(inode)) {
2015 loff_t start_byte =
2016 ((loff_t)tmppage->index << PAGE_SHIFT) +
2017 from;
2018 loff_t length = to - from;
2019 ocfs2_jbd2_inode_add_write(handle, inode,
2020 start_byte, length);
2021 }
Sunil Mushran961cecb2008-07-16 17:22:22 -07002022 block_commit_write(tmppage, from, to);
2023 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002024 }
2025
Mark Fasheh1afc32b2007-09-07 14:46:51 -07002026out_write_size:
Ryan Ding46e62552016-03-25 14:21:01 -07002027 /* Direct io do not update i_size here. */
2028 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2029 pos += copied;
2030 if (pos > i_size_read(inode)) {
2031 i_size_write(inode, pos);
2032 mark_inode_dirty(inode);
2033 }
2034 inode->i_blocks = ocfs2_inode_sector_count(inode);
2035 di->i_size = cpu_to_le64((u64)i_size_read(inode));
Deepa Dinamani078cd822016-09-14 07:48:04 -07002036 inode->i_mtime = inode->i_ctime = current_time(inode);
Ryan Ding46e62552016-03-25 14:21:01 -07002037 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2038 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
Jia-Ju Bai583fee32019-10-06 17:57:54 -07002039 if (handle)
2040 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002041 }
Ryan Ding46e62552016-03-25 14:21:01 -07002042 if (handle)
2043 ocfs2_journal_dirty(handle, wc->w_di_bh);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002044
yangwenfang7f27ec92015-09-04 15:44:45 -07002045out:
Junxiao Bi136f49b2014-12-18 16:17:37 -08002046 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2047 * lock, or it will cause a deadlock since journal commit threads holds
2048 * this lock and will ask for the page lock when flushing the data.
2049 * put it here to preserve the unlock order.
2050 */
2051 ocfs2_unlock_pages(wc);
2052
Ryan Ding46e62552016-03-25 14:21:01 -07002053 if (handle)
2054 ocfs2_commit_trans(osb, handle);
Mark Fasheh59a5e412007-06-22 15:52:36 -07002055
Mark Fashehb27b7cb2007-06-18 11:22:56 -07002056 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2057
Junxiao Bi136f49b2014-12-18 16:17:37 -08002058 brelse(wc->w_di_bh);
2059 kfree(wc);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002060
2061 return copied;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002062}
2063
Nick Pigginb6af1bc2007-10-16 01:25:24 -07002064static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2065 loff_t pos, unsigned len, unsigned copied,
2066 struct page *page, void *fsdata)
Mark Fasheh607d44a2007-05-09 15:14:45 -07002067{
2068 int ret;
2069 struct inode *inode = mapping->host;
2070
piaojun07f38d92016-12-12 16:41:17 -08002071 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
Mark Fasheh607d44a2007-05-09 15:14:45 -07002072
Mark Fasheh607d44a2007-05-09 15:14:45 -07002073 up_write(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002074 ocfs2_inode_unlock(inode, 1);
Mark Fasheh607d44a2007-05-09 15:14:45 -07002075
2076 return ret;
2077}
2078
Ryan Dingc15471f2016-03-25 14:21:09 -07002079struct ocfs2_dio_write_ctxt {
2080 struct list_head dw_zero_list;
2081 unsigned dw_zero_count;
2082 int dw_orphaned;
2083 pid_t dw_writer_pid;
2084};
2085
2086static struct ocfs2_dio_write_ctxt *
2087ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2088{
2089 struct ocfs2_dio_write_ctxt *dwc = NULL;
2090
2091 if (bh->b_private)
2092 return bh->b_private;
2093
2094 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2095 if (dwc == NULL)
2096 return NULL;
2097 INIT_LIST_HEAD(&dwc->dw_zero_list);
2098 dwc->dw_zero_count = 0;
2099 dwc->dw_orphaned = 0;
2100 dwc->dw_writer_pid = task_pid_nr(current);
2101 bh->b_private = dwc;
2102 *alloc = 1;
2103
2104 return dwc;
2105}
2106
2107static void ocfs2_dio_free_write_ctx(struct inode *inode,
2108 struct ocfs2_dio_write_ctxt *dwc)
2109{
2110 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2111 kfree(dwc);
2112}
2113
2114/*
2115 * TODO: Make this into a generic get_blocks function.
2116 *
2117 * From do_direct_io in direct-io.c:
2118 * "So what we do is to permit the ->get_blocks function to populate
2119 * bh.b_size with the size of IO which is permitted at this offset and
2120 * this i_blkbits."
2121 *
2122 * This function is called directly from get_more_blocks in direct-io.c.
2123 *
2124 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2125 * fs_count, map_bh, dio->rw == WRITE);
2126 */
alex chen3e4c56d2017-11-15 17:31:44 -08002127static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock,
Ryan Dingc15471f2016-03-25 14:21:09 -07002128 struct buffer_head *bh_result, int create)
2129{
2130 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Ryan Dinga86a72a2016-03-25 14:21:18 -07002131 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Ryan Dingc15471f2016-03-25 14:21:09 -07002132 struct ocfs2_write_ctxt *wc;
2133 struct ocfs2_write_cluster_desc *desc = NULL;
2134 struct ocfs2_dio_write_ctxt *dwc = NULL;
2135 struct buffer_head *di_bh = NULL;
2136 u64 p_blkno;
Jia Guo7a243c82019-10-06 17:57:47 -07002137 unsigned int i_blkbits = inode->i_sb->s_blocksize_bits;
2138 loff_t pos = iblock << i_blkbits;
2139 sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits;
Ryan Dingc15471f2016-03-25 14:21:09 -07002140 unsigned len, total_len = bh_result->b_size;
2141 int ret = 0, first_get_block = 0;
2142
2143 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2144 len = min(total_len, len);
2145
Jia Guo7a243c82019-10-06 17:57:47 -07002146 /*
2147 * bh_result->b_size is count in get_more_blocks according to write
2148 * "pos" and "end", we need map twice to return different buffer state:
2149 * 1. area in file size, not set NEW;
2150 * 2. area out file size, set NEW.
2151 *
2152 * iblock endblk
2153 * |--------|---------|---------|---------
2154 * |<-------area in file------->|
2155 */
2156
2157 if ((iblock <= endblk) &&
2158 ((iblock + ((len - 1) >> i_blkbits)) > endblk))
2159 len = (endblk - iblock + 1) << i_blkbits;
2160
Ryan Dingc15471f2016-03-25 14:21:09 -07002161 mlog(0, "get block of %lu at %llu:%u req %u\n",
2162 inode->i_ino, pos, len, total_len);
2163
Ryan Dingce170822016-03-25 14:21:20 -07002164 /*
2165 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2166 * we may need to add it to orphan dir. So can not fall to fast path
2167 * while file size will be changed.
2168 */
2169 if (pos + total_len <= i_size_read(inode)) {
alex chen3e4c56d2017-11-15 17:31:44 -08002170
Ryan Dingce170822016-03-25 14:21:20 -07002171 /* This is the fast path for re-write. */
alex chen3e4c56d2017-11-15 17:31:44 -08002172 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create);
Ryan Dingce170822016-03-25 14:21:20 -07002173 if (buffer_mapped(bh_result) &&
2174 !buffer_new(bh_result) &&
2175 ret == 0)
2176 goto out;
Ryan Dingc15471f2016-03-25 14:21:09 -07002177
Ryan Dingce170822016-03-25 14:21:20 -07002178 /* Clear state set by ocfs2_get_block. */
2179 bh_result->b_state = 0;
2180 }
Ryan Dingc15471f2016-03-25 14:21:09 -07002181
2182 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2183 if (unlikely(dwc == NULL)) {
2184 ret = -ENOMEM;
2185 mlog_errno(ret);
2186 goto out;
2187 }
2188
2189 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2190 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2191 !dwc->dw_orphaned) {
2192 /*
2193 * when we are going to alloc extents beyond file size, add the
2194 * inode to orphan dir, so we can recall those spaces when
2195 * system crashed during write.
2196 */
2197 ret = ocfs2_add_inode_to_orphan(osb, inode);
2198 if (ret < 0) {
2199 mlog_errno(ret);
2200 goto out;
2201 }
2202 dwc->dw_orphaned = 1;
2203 }
2204
2205 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2206 if (ret) {
2207 mlog_errno(ret);
2208 goto out;
2209 }
2210
Ryan Dinga86a72a2016-03-25 14:21:18 -07002211 down_write(&oi->ip_alloc_sem);
2212
Ryan Dingc15471f2016-03-25 14:21:09 -07002213 if (first_get_block) {
piaojun1119d3c2018-04-05 16:18:33 -07002214 if (ocfs2_sparse_alloc(osb))
Ryan Dingc15471f2016-03-25 14:21:09 -07002215 ret = ocfs2_zero_tail(inode, di_bh, pos);
2216 else
2217 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2218 total_len, NULL);
2219 if (ret < 0) {
2220 mlog_errno(ret);
2221 goto unlock;
2222 }
2223 }
2224
2225 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2226 OCFS2_WRITE_DIRECT, NULL,
2227 (void **)&wc, di_bh, NULL);
2228 if (ret) {
2229 mlog_errno(ret);
2230 goto unlock;
2231 }
2232
2233 desc = &wc->w_desc[0];
2234
2235 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2236 BUG_ON(p_blkno == 0);
2237 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2238
2239 map_bh(bh_result, inode->i_sb, p_blkno);
2240 bh_result->b_size = len;
2241 if (desc->c_needs_zero)
2242 set_buffer_new(bh_result);
2243
Jia Guo7a243c82019-10-06 17:57:47 -07002244 if (iblock > endblk)
2245 set_buffer_new(bh_result);
2246
Ryan Dingc15471f2016-03-25 14:21:09 -07002247 /* May sleep in end_io. It should not happen in a irq context. So defer
2248 * it to dio work queue. */
2249 set_buffer_defer_completion(bh_result);
2250
2251 if (!list_empty(&wc->w_unwritten_list)) {
2252 struct ocfs2_unwritten_extent *ue = NULL;
2253
2254 ue = list_first_entry(&wc->w_unwritten_list,
2255 struct ocfs2_unwritten_extent,
2256 ue_node);
2257 BUG_ON(ue->ue_cpos != desc->c_cpos);
2258 /* The physical address may be 0, fill it. */
2259 ue->ue_phys = desc->c_phys;
2260
2261 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
Changwei Ge63de8bd2018-01-31 16:15:02 -08002262 dwc->dw_zero_count += wc->w_unwritten_count;
Ryan Dingc15471f2016-03-25 14:21:09 -07002263 }
2264
piaojun07f38d92016-12-12 16:41:17 -08002265 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
Ryan Dingc15471f2016-03-25 14:21:09 -07002266 BUG_ON(ret != len);
2267 ret = 0;
2268unlock:
Ryan Dinga86a72a2016-03-25 14:21:18 -07002269 up_write(&oi->ip_alloc_sem);
Ryan Dingc15471f2016-03-25 14:21:09 -07002270 ocfs2_inode_unlock(inode, 1);
2271 brelse(di_bh);
2272out:
2273 if (ret < 0)
2274 ret = -EIO;
2275 return ret;
2276}
2277
Darrick J. Wong08554952016-11-09 14:42:49 -08002278static int ocfs2_dio_end_io_write(struct inode *inode,
2279 struct ocfs2_dio_write_ctxt *dwc,
2280 loff_t offset,
2281 ssize_t bytes)
Ryan Dingc15471f2016-03-25 14:21:09 -07002282{
2283 struct ocfs2_cached_dealloc_ctxt dealloc;
2284 struct ocfs2_extent_tree et;
2285 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Ryan Dinga86a72a2016-03-25 14:21:18 -07002286 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Ryan Dingc15471f2016-03-25 14:21:09 -07002287 struct ocfs2_unwritten_extent *ue = NULL;
2288 struct buffer_head *di_bh = NULL;
2289 struct ocfs2_dinode *di;
2290 struct ocfs2_alloc_context *data_ac = NULL;
2291 struct ocfs2_alloc_context *meta_ac = NULL;
2292 handle_t *handle = NULL;
2293 loff_t end = offset + bytes;
Wengang Wang90bd0702021-04-09 13:27:29 -07002294 int ret = 0, credits = 0;
Ryan Dingc15471f2016-03-25 14:21:09 -07002295
2296 ocfs2_init_dealloc_ctxt(&dealloc);
2297
2298 /* We do clear unwritten, delete orphan, change i_size here. If neither
2299 * of these happen, we can skip all this. */
2300 if (list_empty(&dwc->dw_zero_list) &&
2301 end <= i_size_read(inode) &&
2302 !dwc->dw_orphaned)
2303 goto out;
2304
Ryan Dinga86a72a2016-03-25 14:21:18 -07002305 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2306 if (ret < 0) {
2307 mlog_errno(ret);
2308 goto out;
2309 }
2310
2311 down_write(&oi->ip_alloc_sem);
2312
hongnanli137cebf2022-03-22 14:38:45 -07002313 /* Delete orphan before acquire i_rwsem. */
Ryan Dingc15471f2016-03-25 14:21:09 -07002314 if (dwc->dw_orphaned) {
2315 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2316
2317 end = end > i_size_read(inode) ? end : 0;
2318
2319 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2320 !!end, end);
2321 if (ret < 0)
2322 mlog_errno(ret);
2323 }
2324
Darrick J. Wongaef73a62016-12-09 16:10:15 -08002325 di = (struct ocfs2_dinode *)di_bh->b_data;
Ryan Dingc15471f2016-03-25 14:21:09 -07002326
2327 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2328
Changwei Ge71a36942018-01-31 16:15:06 -08002329 /* Attach dealloc with extent tree in case that we may reuse extents
2330 * which are already unlinked from current extent tree due to extent
2331 * rotation and merging.
2332 */
2333 et.et_dealloc = &dealloc;
2334
Ryan Dingc15471f2016-03-25 14:21:09 -07002335 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2336 &data_ac, &meta_ac);
Ryan Ding28888682016-03-25 14:21:23 -07002337 if (ret) {
2338 mlog_errno(ret);
2339 goto unlock;
2340 }
Ryan Dingc15471f2016-03-25 14:21:09 -07002341
2342 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2343
2344 handle = ocfs2_start_trans(osb, credits);
2345 if (IS_ERR(handle)) {
2346 ret = PTR_ERR(handle);
2347 mlog_errno(ret);
2348 goto unlock;
2349 }
2350 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2351 OCFS2_JOURNAL_ACCESS_WRITE);
2352 if (ret) {
2353 mlog_errno(ret);
2354 goto commit;
2355 }
2356
2357 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2358 ret = ocfs2_mark_extent_written(inode, &et, handle,
2359 ue->ue_cpos, 1,
2360 ue->ue_phys,
2361 meta_ac, &dealloc);
2362 if (ret < 0) {
2363 mlog_errno(ret);
2364 break;
2365 }
2366 }
2367
2368 if (end > i_size_read(inode)) {
2369 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2370 if (ret < 0)
2371 mlog_errno(ret);
2372 }
2373commit:
2374 ocfs2_commit_trans(osb, handle);
2375unlock:
Ryan Dinga86a72a2016-03-25 14:21:18 -07002376 up_write(&oi->ip_alloc_sem);
Ryan Dingc15471f2016-03-25 14:21:09 -07002377 ocfs2_inode_unlock(inode, 1);
2378 brelse(di_bh);
2379out:
Ryan Dingc15471f2016-03-25 14:21:09 -07002380 if (data_ac)
2381 ocfs2_free_alloc_context(data_ac);
2382 if (meta_ac)
2383 ocfs2_free_alloc_context(meta_ac);
Ryan Ding28888682016-03-25 14:21:23 -07002384 ocfs2_run_deallocs(osb, &dealloc);
Ryan Ding28888682016-03-25 14:21:23 -07002385 ocfs2_dio_free_write_ctx(inode, dwc);
Darrick J. Wong08554952016-11-09 14:42:49 -08002386
2387 return ret;
Ryan Dingc15471f2016-03-25 14:21:09 -07002388}
2389
2390/*
2391 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
2392 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2393 * to protect io on one node from truncation on another.
2394 */
2395static int ocfs2_dio_end_io(struct kiocb *iocb,
2396 loff_t offset,
2397 ssize_t bytes,
2398 void *private)
2399{
2400 struct inode *inode = file_inode(iocb->ki_filp);
2401 int level;
Darrick J. Wong08554952016-11-09 14:42:49 -08002402 int ret = 0;
Ryan Dingc15471f2016-03-25 14:21:09 -07002403
2404 /* this io's submitter should not have unlocked this before we could */
2405 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2406
Wengang Wang5040f8d2018-11-16 15:08:25 -08002407 if (bytes <= 0)
2408 mlog_ratelimited(ML_ERROR, "Direct IO failed, bytes = %lld",
2409 (long long)bytes);
2410 if (private) {
2411 if (bytes > 0)
2412 ret = ocfs2_dio_end_io_write(inode, private, offset,
2413 bytes);
2414 else
2415 ocfs2_dio_free_write_ctx(inode, private);
2416 }
Ryan Dingc15471f2016-03-25 14:21:09 -07002417
2418 ocfs2_iocb_clear_rw_locked(iocb);
2419
2420 level = ocfs2_iocb_rw_locked_level(iocb);
2421 ocfs2_rw_unlock(inode, level);
Darrick J. Wong08554952016-11-09 14:42:49 -08002422 return ret;
Ryan Dingc15471f2016-03-25 14:21:09 -07002423}
2424
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002425static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
Ryan Dingc15471f2016-03-25 14:21:09 -07002426{
2427 struct file *file = iocb->ki_filp;
Al Viro93c76a32015-12-04 23:45:44 -05002428 struct inode *inode = file->f_mapping->host;
Ryan Dingc15471f2016-03-25 14:21:09 -07002429 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Ryan Dingc15471f2016-03-25 14:21:09 -07002430 get_block_t *get_block;
2431
2432 /*
2433 * Fallback to buffered I/O if we see an inode without
2434 * extents.
2435 */
2436 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2437 return 0;
2438
2439 /* Fallback to buffered I/O if we do not support append dio. */
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002440 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2441 !ocfs2_supports_append_dio(osb))
Ryan Dingc15471f2016-03-25 14:21:09 -07002442 return 0;
2443
2444 if (iov_iter_rw(iter) == READ)
alex chen3e4c56d2017-11-15 17:31:44 -08002445 get_block = ocfs2_lock_get_block;
Ryan Dingc15471f2016-03-25 14:21:09 -07002446 else
alex chen3e4c56d2017-11-15 17:31:44 -08002447 get_block = ocfs2_dio_wr_get_block;
Ryan Dingc15471f2016-03-25 14:21:09 -07002448
2449 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002450 iter, get_block,
Ryan Dingc15471f2016-03-25 14:21:09 -07002451 ocfs2_dio_end_io, NULL, 0);
2452}
2453
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002454const struct address_space_operations ocfs2_aops = {
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +00002455 .dirty_folio = block_dirty_folio,
Matthew Wilcox (Oracle)bb9263f2022-04-29 11:12:16 -04002456 .read_folio = ocfs2_read_folio,
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -07002457 .readahead = ocfs2_readahead,
Hisashi Hifumi1fca3a02009-03-05 17:22:21 +09002458 .writepage = ocfs2_writepage,
2459 .write_begin = ocfs2_write_begin,
2460 .write_end = ocfs2_write_end,
2461 .bmap = ocfs2_bmap,
Hisashi Hifumi1fca3a02009-03-05 17:22:21 +09002462 .direct_IO = ocfs2_direct_IO,
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00002463 .invalidate_folio = block_invalidate_folio,
Matthew Wilcox (Oracle)eca66382022-05-01 00:10:21 -04002464 .release_folio = ocfs2_release_folio,
Matthew Wilcox (Oracle)67235182022-06-06 10:20:31 -04002465 .migrate_folio = buffer_migrate_folio,
Hisashi Hifumi1fca3a02009-03-05 17:22:21 +09002466 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002467 .error_remove_page = generic_error_remove_page,
Mark Fashehccd979b2005-12-15 14:31:24 -08002468};