blob: 4f8197caa487f59c1868d4e292e254587a46eab1 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * file.c
5 *
6 * File open, close, extend, truncate
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080026#include <linux/capability.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080027#include <linux/fs.h>
28#include <linux/types.h>
29#include <linux/slab.h>
30#include <linux/highmem.h>
31#include <linux/pagemap.h>
32#include <linux/uio.h>
Mark Fashehe2057c52006-10-03 17:53:05 -070033#include <linux/sched.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020034#include <linux/splice.h>
Tiger Yang7f1a37e2006-11-15 15:48:42 +080035#include <linux/mount.h>
Mark Fasheh9517bac2007-02-09 20:24:12 -080036#include <linux/writeback.h>
Mark Fasheh385820a2007-07-19 00:14:38 -070037#include <linux/falloc.h>
Jan Karaa90714c2008-10-09 19:38:40 +020038#include <linux/quotaops.h>
Jan Kara04eda1a2010-08-05 20:32:45 +020039#include <linux/blkdev.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080040
Mark Fashehccd979b2005-12-15 14:31:24 -080041#include <cluster/masklog.h>
42
43#include "ocfs2.h"
44
45#include "alloc.h"
46#include "aops.h"
47#include "dir.h"
48#include "dlmglue.h"
49#include "extent_map.h"
50#include "file.h"
51#include "sysfile.h"
52#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070053#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080054#include "journal.h"
Mark Fasheh53fc6222007-12-20 16:49:04 -080055#include "locks.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080056#include "mmap.h"
57#include "suballoc.h"
58#include "super.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080059#include "xattr.h"
Tiger Yang23fc2702008-11-14 11:17:18 +080060#include "acl.h"
Jan Karaa90714c2008-10-09 19:38:40 +020061#include "quota.h"
Tao Ma293b2f72009-08-25 08:02:48 +080062#include "refcounttree.h"
Tao Ma468eedd2011-02-22 22:14:41 +080063#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080064
65#include "buffer_head_io.h"
66
Mark Fasheh53fc6222007-12-20 16:49:04 -080067static int ocfs2_init_file_private(struct inode *inode, struct file *file)
68{
69 struct ocfs2_file_private *fp;
70
71 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
72 if (!fp)
73 return -ENOMEM;
74
75 fp->fp_file = file;
76 mutex_init(&fp->fp_mutex);
77 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
78 file->private_data = fp;
79
80 return 0;
81}
82
83static void ocfs2_free_file_private(struct inode *inode, struct file *file)
84{
85 struct ocfs2_file_private *fp = file->private_data;
86 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
87
88 if (fp) {
89 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
90 ocfs2_lock_res_free(&fp->fp_flock);
91 kfree(fp);
92 file->private_data = NULL;
93 }
94}
95
Mark Fashehccd979b2005-12-15 14:31:24 -080096static int ocfs2_file_open(struct inode *inode, struct file *file)
97{
98 int status;
99 int mode = file->f_flags;
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
101
Tao Ma468eedd2011-02-22 22:14:41 +0800102 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
103 (unsigned long long)OCFS2_I(inode)->ip_blkno,
104 file->f_path.dentry->d_name.len,
105 file->f_path.dentry->d_name.name, mode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800106
Christoph Hellwig907f4552010-03-03 09:05:06 -0500107 if (file->f_mode & FMODE_WRITE)
Christoph Hellwig871a2932010-03-03 09:05:07 -0500108 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500109
Mark Fashehccd979b2005-12-15 14:31:24 -0800110 spin_lock(&oi->ip_lock);
111
112 /* Check that the inode hasn't been wiped from disk by another
113 * node. If it hasn't then we're safe as long as we hold the
114 * spin lock until our increment of open count. */
115 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
116 spin_unlock(&oi->ip_lock);
117
118 status = -ENOENT;
119 goto leave;
120 }
121
122 if (mode & O_DIRECT)
123 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
124
125 oi->ip_open_count++;
126 spin_unlock(&oi->ip_lock);
Mark Fasheh53fc6222007-12-20 16:49:04 -0800127
128 status = ocfs2_init_file_private(inode, file);
129 if (status) {
130 /*
131 * We want to set open count back if we're failing the
132 * open.
133 */
134 spin_lock(&oi->ip_lock);
135 oi->ip_open_count--;
136 spin_unlock(&oi->ip_lock);
137 }
138
Mark Fashehccd979b2005-12-15 14:31:24 -0800139leave:
Mark Fashehccd979b2005-12-15 14:31:24 -0800140 return status;
141}
142
143static int ocfs2_file_release(struct inode *inode, struct file *file)
144{
145 struct ocfs2_inode_info *oi = OCFS2_I(inode);
146
Mark Fashehccd979b2005-12-15 14:31:24 -0800147 spin_lock(&oi->ip_lock);
148 if (!--oi->ip_open_count)
149 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
Tao Ma468eedd2011-02-22 22:14:41 +0800150
151 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
152 oi->ip_blkno,
153 file->f_path.dentry->d_name.len,
154 file->f_path.dentry->d_name.name,
155 oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -0800156 spin_unlock(&oi->ip_lock);
157
Mark Fasheh53fc6222007-12-20 16:49:04 -0800158 ocfs2_free_file_private(inode, file);
159
Mark Fashehccd979b2005-12-15 14:31:24 -0800160 return 0;
161}
162
Mark Fasheh53fc6222007-12-20 16:49:04 -0800163static int ocfs2_dir_open(struct inode *inode, struct file *file)
164{
165 return ocfs2_init_file_private(inode, file);
166}
167
168static int ocfs2_dir_release(struct inode *inode, struct file *file)
169{
170 ocfs2_free_file_private(inode, file);
171 return 0;
172}
173
Josef Bacik02c24a82011-07-16 20:44:56 -0400174static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
175 int datasync)
Mark Fashehccd979b2005-12-15 14:31:24 -0800176{
177 int err = 0;
178 journal_t *journal;
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200179 struct inode *inode = file->f_mapping->host;
Mark Fashehccd979b2005-12-15 14:31:24 -0800180 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
181
Tao Ma468eedd2011-02-22 22:14:41 +0800182 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
183 OCFS2_I(inode)->ip_blkno,
184 file->f_path.dentry->d_name.len,
185 file->f_path.dentry->d_name.name,
186 (unsigned long long)datasync);
Mark Fashehccd979b2005-12-15 14:31:24 -0800187
Josef Bacik02c24a82011-07-16 20:44:56 -0400188 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
189 if (err)
190 return err;
191
192 /*
193 * Probably don't need the i_mutex at all in here, just putting it here
194 * to be consistent with how fsync used to be called, someone more
195 * familiar with the fs could possibly remove it.
196 */
197 mutex_lock(&inode->i_mutex);
Jan Kara04eda1a2010-08-05 20:32:45 +0200198 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
199 /*
200 * We still have to flush drive's caches to get data to the
201 * platter
202 */
203 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
Linus Torvaldsf8cae0f2010-10-22 19:30:38 -0700204 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
Hisashi Hifumie04cc152009-06-09 16:47:45 +0900205 goto bail;
Jan Kara04eda1a2010-08-05 20:32:45 +0200206 }
Hisashi Hifumie04cc152009-06-09 16:47:45 +0900207
Mark Fashehccd979b2005-12-15 14:31:24 -0800208 journal = osb->journal->j_journal;
Joel Becker2b4e30f2008-09-03 20:03:41 -0700209 err = jbd2_journal_force_commit(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800210
211bail:
Tao Mac1e8d352011-03-07 16:43:21 +0800212 if (err)
213 mlog_errno(err);
Josef Bacik02c24a82011-07-16 20:44:56 -0400214 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800215
216 return (err < 0) ? -EIO : 0;
217}
218
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800219int ocfs2_should_update_atime(struct inode *inode,
220 struct vfsmount *vfsmnt)
221{
222 struct timespec now;
223 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
224
225 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
226 return 0;
227
228 if ((inode->i_flags & S_NOATIME) ||
229 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
230 return 0;
231
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800232 /*
233 * We can be called with no vfsmnt structure - NFSD will
234 * sometimes do this.
235 *
236 * Note that our action here is different than touch_atime() -
237 * if we can't tell whether this is a noatime mount, then we
238 * don't know whether to trust the value of s_atime_quantum.
239 */
240 if (vfsmnt == NULL)
241 return 0;
242
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800243 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
244 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
245 return 0;
246
Mark Fasheh7e913c52006-12-13 00:34:35 -0800247 if (vfsmnt->mnt_flags & MNT_RELATIME) {
248 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
249 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
250 return 1;
251
252 return 0;
253 }
254
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800255 now = CURRENT_TIME;
256 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
257 return 0;
258 else
259 return 1;
260}
261
262int ocfs2_update_inode_atime(struct inode *inode,
263 struct buffer_head *bh)
264{
265 int ret;
266 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
267 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700268 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800269
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800270 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200271 if (IS_ERR(handle)) {
272 ret = PTR_ERR(handle);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800273 mlog_errno(ret);
274 goto out;
275 }
276
Joel Becker0cf2f762009-02-12 16:41:25 -0800277 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -0700278 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700279 if (ret) {
280 mlog_errno(ret);
281 goto out_commit;
282 }
283
284 /*
285 * Don't use ocfs2_mark_inode_dirty() here as we don't always
286 * have i_mutex to guard against concurrent changes to other
287 * inode fields.
288 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800289 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700290 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
291 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
Joel Beckerec20cec2010-03-19 14:13:52 -0700292 ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800293
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700294out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800295 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
296out:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800297 return ret;
298}
299
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700300static int ocfs2_set_inode_size(handle_t *handle,
301 struct inode *inode,
302 struct buffer_head *fe_bh,
303 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800304{
305 int status;
306
Mark Fashehccd979b2005-12-15 14:31:24 -0800307 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700308 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800309 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
310
311 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
312 if (status < 0) {
313 mlog_errno(status);
314 goto bail;
315 }
316
317bail:
Mark Fashehccd979b2005-12-15 14:31:24 -0800318 return status;
319}
320
Jan Kara9e33d692008-08-25 19:56:50 +0200321int ocfs2_simple_size_update(struct inode *inode,
322 struct buffer_head *di_bh,
323 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800324{
325 int ret;
326 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700327 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800328
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700329 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200330 if (IS_ERR(handle)) {
331 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800332 mlog_errno(ret);
333 goto out;
334 }
335
336 ret = ocfs2_set_inode_size(handle, inode, di_bh,
337 new_i_size);
338 if (ret < 0)
339 mlog_errno(ret);
340
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700341 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800342out:
343 return ret;
344}
345
Tao Ma37f8a2b2009-08-26 09:47:28 +0800346static int ocfs2_cow_file_pos(struct inode *inode,
347 struct buffer_head *fe_bh,
348 u64 offset)
349{
350 int status;
351 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
352 unsigned int num_clusters = 0;
353 unsigned int ext_flags = 0;
354
355 /*
356 * If the new offset is aligned to the range of the cluster, there is
357 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
358 * CoW either.
359 */
360 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
361 return 0;
362
363 status = ocfs2_get_clusters(inode, cpos, &phys,
364 &num_clusters, &ext_flags);
365 if (status) {
366 mlog_errno(status);
367 goto out;
368 }
369
370 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
371 goto out;
372
Tiger Yangc7dd3392013-08-13 16:00:58 -0700373 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
Tao Ma37f8a2b2009-08-26 09:47:28 +0800374
375out:
376 return status;
377}
378
Mark Fashehccd979b2005-12-15 14:31:24 -0800379static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
380 struct inode *inode,
381 struct buffer_head *fe_bh,
382 u64 new_i_size)
383{
384 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700385 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800386 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700387 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800388
Tao Ma37f8a2b2009-08-26 09:47:28 +0800389 /*
390 * We need to CoW the cluster contains the offset if it is reflinked
391 * since we will call ocfs2_zero_range_for_truncate later which will
392 * write "0" from offset to the end of the cluster.
393 */
394 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
395 if (status) {
396 mlog_errno(status);
397 return status;
398 }
399
Mark Fashehccd979b2005-12-15 14:31:24 -0800400 /* TODO: This needs to actually orphan the inode in this
401 * transaction. */
402
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700403 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800404 if (IS_ERR(handle)) {
405 status = PTR_ERR(handle);
406 mlog_errno(status);
407 goto out;
408 }
409
Joel Becker0cf2f762009-02-12 16:41:25 -0800410 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700411 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800412 if (status < 0) {
413 mlog_errno(status);
414 goto out_commit;
415 }
416
417 /*
418 * Do this before setting i_size.
419 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700420 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
421 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
422 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800423 if (status) {
424 mlog_errno(status);
425 goto out_commit;
426 }
427
428 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800429 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
430
431 di = (struct ocfs2_dinode *) fe_bh->b_data;
432 di->i_size = cpu_to_le64(new_i_size);
433 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
434 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
435
Joel Beckerec20cec2010-03-19 14:13:52 -0700436 ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800437
Mark Fasheh60b11392007-02-16 11:46:50 -0800438out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700439 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800440out:
Mark Fashehccd979b2005-12-15 14:31:24 -0800441 return status;
442}
443
444static int ocfs2_truncate_file(struct inode *inode,
445 struct buffer_head *di_bh,
446 u64 new_i_size)
447{
448 int status = 0;
449 struct ocfs2_dinode *fe = NULL;
450 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800451
Joel Beckerb657c952008-11-13 14:49:11 -0800452 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
453 * already validated it */
Mark Fashehccd979b2005-12-15 14:31:24 -0800454 fe = (struct ocfs2_dinode *) di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800455
Tao Ma468eedd2011-02-22 22:14:41 +0800456 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
457 (unsigned long long)le64_to_cpu(fe->i_size),
458 (unsigned long long)new_i_size);
459
Mark Fashehccd979b2005-12-15 14:31:24 -0800460 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800461 "Inode %llu, inode i_size = %lld != di "
462 "i_size = %llu, i_flags = 0x%x\n",
463 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800464 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800465 (unsigned long long)le64_to_cpu(fe->i_size),
466 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800467
468 if (new_i_size > le64_to_cpu(fe->i_size)) {
Tao Ma468eedd2011-02-22 22:14:41 +0800469 trace_ocfs2_truncate_file_error(
470 (unsigned long long)le64_to_cpu(fe->i_size),
471 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800472 status = -EINVAL;
473 mlog_errno(status);
474 goto bail;
475 }
476
Mark Fashehccd979b2005-12-15 14:31:24 -0800477 /* lets handle the simple truncate cases before doing any more
478 * cluster locking. */
479 if (new_i_size == le64_to_cpu(fe->i_size))
480 goto bail;
481
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700482 down_write(&OCFS2_I(inode)->ip_alloc_sem);
483
Mark Fasheh4fe370a2009-12-07 13:15:40 -0800484 ocfs2_resv_discard(&osb->osb_la_resmap,
485 &OCFS2_I(inode)->ip_la_data_resv);
486
Mark Fashehc934a922007-10-18 15:23:46 -0700487 /*
488 * The inode lock forced other nodes to sync and drop their
489 * pages, which (correctly) happens even if we have a truncate
490 * without allocation change - ocfs2 cluster sizes can be much
491 * greater than page size, so we have to truncate them
492 * anyway.
493 */
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700494 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
495 truncate_inode_pages(inode->i_mapping, new_i_size);
496
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700497 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
498 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
Mark Fashehb1967d02007-11-20 11:56:39 -0800499 i_size_read(inode), 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700500 if (status)
501 mlog_errno(status);
502
Mark Fashehc934a922007-10-18 15:23:46 -0700503 goto bail_unlock_sem;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700504 }
505
Mark Fashehccd979b2005-12-15 14:31:24 -0800506 /* alright, we're going to need to do a full blown alloc size
507 * change. Orphan the inode so that recovery can complete the
508 * truncate if necessary. This does the task of marking
509 * i_size. */
510 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
511 if (status < 0) {
512 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700513 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800514 }
515
Tristan Ye78f94672010-05-11 17:54:42 +0800516 status = ocfs2_commit_truncate(osb, inode, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800517 if (status < 0) {
518 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700519 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800520 }
521
522 /* TODO: orphan dir cleanup here. */
Mark Fashehc934a922007-10-18 15:23:46 -0700523bail_unlock_sem:
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700524 up_write(&OCFS2_I(inode)->ip_alloc_sem);
525
Mark Fashehccd979b2005-12-15 14:31:24 -0800526bail:
Tao Ma8b2c0db2009-08-18 11:43:49 +0800527 if (!status && OCFS2_I(inode)->ip_clusters == 0)
528 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800529
Mark Fashehccd979b2005-12-15 14:31:24 -0800530 return status;
531}
532
533/*
Tao Ma0eb8d47e2008-08-18 17:38:45 +0800534 * extend file allocation only here.
Mark Fashehccd979b2005-12-15 14:31:24 -0800535 * we'll update all the disk stuff, and oip->alloc_size
536 *
537 * expect stuff to be locked, a transaction started and enough data /
538 * metadata reservations in the contexts.
539 *
540 * Will return -EAGAIN, and a reason if a restart is needed.
541 * If passed in, *reason will always be set, even in error.
542 */
Tao Ma0eb8d47e2008-08-18 17:38:45 +0800543int ocfs2_add_inode_data(struct ocfs2_super *osb,
544 struct inode *inode,
545 u32 *logical_offset,
546 u32 clusters_to_add,
547 int mark_unwritten,
548 struct buffer_head *fe_bh,
549 handle_t *handle,
550 struct ocfs2_alloc_context *data_ac,
551 struct ocfs2_alloc_context *meta_ac,
552 enum ocfs2_alloc_restarted *reason_ret)
Mark Fashehccd979b2005-12-15 14:31:24 -0800553{
Joel Beckerf99b9b72008-08-20 19:36:33 -0700554 int ret;
555 struct ocfs2_extent_tree et;
Mark Fashehccd979b2005-12-15 14:31:24 -0800556
Joel Becker5e404e92009-02-13 03:54:22 -0800557 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
Joel Beckercbee7e12009-02-13 03:34:15 -0800558 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
559 clusters_to_add, mark_unwritten,
560 data_ac, meta_ac, reason_ret);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700561
562 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800563}
564
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800565static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
566 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800567{
568 int status = 0;
569 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800570 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800571 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800572 struct buffer_head *bh = NULL;
573 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700574 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800575 struct ocfs2_alloc_context *data_ac = NULL;
576 struct ocfs2_alloc_context *meta_ac = NULL;
577 enum ocfs2_alloc_restarted why;
578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700579 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +0200580 int did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800581
Mark Fashehdcd05382007-01-16 11:32:23 -0800582 /*
583 * This function only exists for file systems which don't
584 * support holes.
585 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800586 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800587
Joel Beckerb657c952008-11-13 14:49:11 -0800588 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800589 if (status < 0) {
590 mlog_errno(status);
591 goto leave;
592 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800593 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800594
595restart_all:
596 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
597
Joel Becker5e404e92009-02-13 03:54:22 -0800598 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700599 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
600 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800601 if (status) {
602 mlog_errno(status);
603 goto leave;
604 }
605
Tao Ma811f9332008-08-18 17:38:43 +0800606 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
607 clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700608 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800609 if (IS_ERR(handle)) {
610 status = PTR_ERR(handle);
611 handle = NULL;
612 mlog_errno(status);
613 goto leave;
614 }
615
616restarted_transaction:
Tao Ma468eedd2011-02-22 22:14:41 +0800617 trace_ocfs2_extend_allocation(
618 (unsigned long long)OCFS2_I(inode)->ip_blkno,
619 (unsigned long long)i_size_read(inode),
620 le32_to_cpu(fe->i_clusters), clusters_to_add,
621 why, restart_func);
622
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500623 status = dquot_alloc_space_nodirty(inode,
624 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
625 if (status)
Jan Karaa90714c2008-10-09 19:38:40 +0200626 goto leave;
Jan Karaa90714c2008-10-09 19:38:40 +0200627 did_quota = 1;
628
Mark Fashehccd979b2005-12-15 14:31:24 -0800629 /* reserve a write to the file entry early on - that we if we
630 * run out of credits in the allocation path, we can still
631 * update i_size. */
Joel Becker0cf2f762009-02-12 16:41:25 -0800632 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -0700633 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800634 if (status < 0) {
635 mlog_errno(status);
636 goto leave;
637 }
638
639 prev_clusters = OCFS2_I(inode)->ip_clusters;
640
Tao Ma0eb8d47e2008-08-18 17:38:45 +0800641 status = ocfs2_add_inode_data(osb,
642 inode,
643 &logical_start,
644 clusters_to_add,
645 mark_unwritten,
646 bh,
647 handle,
648 data_ac,
649 meta_ac,
650 &why);
Mark Fashehccd979b2005-12-15 14:31:24 -0800651 if ((status < 0) && (status != -EAGAIN)) {
652 if (status != -ENOSPC)
653 mlog_errno(status);
654 goto leave;
655 }
656
Joel Beckerec20cec2010-03-19 14:13:52 -0700657 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800658
659 spin_lock(&OCFS2_I(inode)->ip_lock);
660 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
661 spin_unlock(&OCFS2_I(inode)->ip_lock);
Jan Karaa90714c2008-10-09 19:38:40 +0200662 /* Release unused quota reservation */
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500663 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +0200664 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
665 did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800666
667 if (why != RESTART_NONE && clusters_to_add) {
668 if (why == RESTART_META) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800669 restart_func = 1;
Tao Ma79681842010-04-16 13:59:25 +0800670 status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800671 } else {
672 BUG_ON(why != RESTART_TRANS);
673
Younger Liu2b1e55c2013-09-11 14:19:44 -0700674 status = ocfs2_allocate_extend_trans(handle, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800675 if (status < 0) {
676 /* handle still has to be committed at
677 * this point. */
678 status = -ENOMEM;
679 mlog_errno(status);
680 goto leave;
681 }
682 goto restarted_transaction;
683 }
684 }
685
Tao Ma468eedd2011-02-22 22:14:41 +0800686 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700687 le32_to_cpu(fe->i_clusters),
Tao Ma468eedd2011-02-22 22:14:41 +0800688 (unsigned long long)le64_to_cpu(fe->i_size),
689 OCFS2_I(inode)->ip_clusters,
690 (unsigned long long)i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800691
692leave:
Jan Karaa90714c2008-10-09 19:38:40 +0200693 if (status < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500694 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +0200695 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
Mark Fashehccd979b2005-12-15 14:31:24 -0800696 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700697 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800698 handle = NULL;
699 }
700 if (data_ac) {
701 ocfs2_free_alloc_context(data_ac);
702 data_ac = NULL;
703 }
704 if (meta_ac) {
705 ocfs2_free_alloc_context(meta_ac);
706 meta_ac = NULL;
707 }
708 if ((!status) && restart_func) {
709 restart_func = 0;
710 goto restart_all;
711 }
Mark Fasheha81cb882008-10-07 14:25:16 -0700712 brelse(bh);
713 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800714
Mark Fashehccd979b2005-12-15 14:31:24 -0800715 return status;
716}
717
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700718/*
719 * While a write will already be ordering the data, a truncate will not.
720 * Thus, we need to explicitly order the zeroed pages.
721 */
722static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
723{
724 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
725 handle_t *handle = NULL;
726 int ret = 0;
727
728 if (!ocfs2_should_order_data(inode))
729 goto out;
730
731 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
732 if (IS_ERR(handle)) {
733 ret = -ENOMEM;
734 mlog_errno(ret);
735 goto out;
736 }
737
738 ret = ocfs2_jbd2_file_inode(handle, inode);
739 if (ret < 0)
740 mlog_errno(ret);
741
742out:
743 if (ret) {
744 if (!IS_ERR(handle))
745 ocfs2_commit_trans(osb, handle);
746 handle = ERR_PTR(ret);
747 }
748 return handle;
749}
750
Mark Fashehccd979b2005-12-15 14:31:24 -0800751/* Some parts of this taken from generic_cont_expand, which turned out
752 * to be too fragile to do exactly what we need without us having to
Nick Piggin4e02ed42008-10-29 14:00:55 -0700753 * worry about recursive locking in ->write_begin() and ->write_end(). */
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700754static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
755 u64 abs_to)
Mark Fashehccd979b2005-12-15 14:31:24 -0800756{
757 struct address_space *mapping = inode->i_mapping;
758 struct page *page;
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700759 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700760 handle_t *handle = NULL;
Joel Becker54532582010-07-16 13:32:33 -0700761 int ret = 0;
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700762 unsigned zero_from, zero_to, block_start, block_end;
Mark Fashehccd979b2005-12-15 14:31:24 -0800763
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700764 BUG_ON(abs_from >= abs_to);
765 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
766 BUG_ON(abs_from & (inode->i_blkbits - 1));
Mark Fashehccd979b2005-12-15 14:31:24 -0800767
Jan Kara9b4c0ff2010-08-24 14:28:03 +0200768 page = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800769 if (!page) {
770 ret = -ENOMEM;
771 mlog_errno(ret);
772 goto out;
773 }
774
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700775 /* Get the offsets within the page that we want to zero */
776 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
777 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
778 if (!zero_to)
779 zero_to = PAGE_CACHE_SIZE;
Mark Fashehccd979b2005-12-15 14:31:24 -0800780
Tao Ma468eedd2011-02-22 22:14:41 +0800781 trace_ocfs2_write_zero_page(
782 (unsigned long long)OCFS2_I(inode)->ip_blkno,
783 (unsigned long long)abs_from,
784 (unsigned long long)abs_to,
785 index, zero_from, zero_to);
Joel Becker56934862010-07-01 15:13:31 -0700786
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700787 /* We know that zero_from is block aligned */
788 for (block_start = zero_from; block_start < zero_to;
789 block_start = block_end) {
790 block_end = block_start + (1 << inode->i_blkbits);
791
792 /*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200793 * block_start is block-aligned. Bump it by one to force
794 * __block_write_begin and block_commit_write to zero the
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700795 * whole block.
796 */
Christoph Hellwigebdec242010-10-06 10:47:23 +0200797 ret = __block_write_begin(page, block_start + 1, 0,
798 ocfs2_get_block);
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700799 if (ret < 0) {
800 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800801 goto out_unlock;
802 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800803
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700804 if (!handle) {
805 handle = ocfs2_zero_start_ordered_transaction(inode);
806 if (IS_ERR(handle)) {
807 ret = PTR_ERR(handle);
808 handle = NULL;
809 break;
810 }
811 }
812
813 /* must not update i_size! */
814 ret = block_commit_write(page, block_start + 1,
815 block_start + 1);
816 if (ret < 0)
817 mlog_errno(ret);
818 else
819 ret = 0;
820 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800821
822 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700823 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700824
Mark Fashehccd979b2005-12-15 14:31:24 -0800825out_unlock:
826 unlock_page(page);
827 page_cache_release(page);
828out:
829 return ret;
830}
831
Joel Becker56934862010-07-01 15:13:31 -0700832/*
833 * Find the next range to zero. We do this in terms of bytes because
834 * that's what ocfs2_zero_extend() wants, and it is dealing with the
835 * pagecache. We may return multiple extents.
836 *
837 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
838 * needs to be zeroed. range_start and range_end return the next zeroing
839 * range. A subsequent call should pass the previous range_end as its
840 * zero_start. If range_end is 0, there's nothing to do.
841 *
842 * Unwritten extents are skipped over. Refcounted extents are CoWd.
843 */
844static int ocfs2_zero_extend_get_range(struct inode *inode,
845 struct buffer_head *di_bh,
846 u64 zero_start, u64 zero_end,
847 u64 *range_start, u64 *range_end)
Mark Fashehccd979b2005-12-15 14:31:24 -0800848{
Joel Becker56934862010-07-01 15:13:31 -0700849 int rc = 0, needs_cow = 0;
850 u32 p_cpos, zero_clusters = 0;
851 u32 zero_cpos =
852 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
853 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
854 unsigned int num_clusters = 0;
855 unsigned int ext_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800856
Joel Becker56934862010-07-01 15:13:31 -0700857 while (zero_cpos < last_cpos) {
858 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
859 &num_clusters, &ext_flags);
860 if (rc) {
861 mlog_errno(rc);
Mark Fashehccd979b2005-12-15 14:31:24 -0800862 goto out;
863 }
864
Joel Becker56934862010-07-01 15:13:31 -0700865 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
866 zero_clusters = num_clusters;
867 if (ext_flags & OCFS2_EXT_REFCOUNTED)
868 needs_cow = 1;
869 break;
870 }
871
872 zero_cpos += num_clusters;
873 }
874 if (!zero_clusters) {
875 *range_end = 0;
876 goto out;
877 }
878
879 while ((zero_cpos + zero_clusters) < last_cpos) {
880 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
881 &p_cpos, &num_clusters,
882 &ext_flags);
883 if (rc) {
884 mlog_errno(rc);
885 goto out;
886 }
887
888 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
889 break;
890 if (ext_flags & OCFS2_EXT_REFCOUNTED)
891 needs_cow = 1;
892 zero_clusters += num_clusters;
893 }
894 if ((zero_cpos + zero_clusters) > last_cpos)
895 zero_clusters = last_cpos - zero_cpos;
896
897 if (needs_cow) {
Tiger Yangc7dd3392013-08-13 16:00:58 -0700898 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
Tao Ma15502712010-08-12 10:36:38 +0800899 zero_clusters, UINT_MAX);
Joel Becker56934862010-07-01 15:13:31 -0700900 if (rc) {
901 mlog_errno(rc);
902 goto out;
903 }
904 }
905
906 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
907 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
908 zero_cpos + zero_clusters);
909
910out:
911 return rc;
912}
913
914/*
915 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
916 * has made sure that the entire range needs zeroing.
917 */
918static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
919 u64 range_end)
920{
921 int rc = 0;
922 u64 next_pos;
923 u64 zero_pos = range_start;
924
Tao Ma468eedd2011-02-22 22:14:41 +0800925 trace_ocfs2_zero_extend_range(
926 (unsigned long long)OCFS2_I(inode)->ip_blkno,
927 (unsigned long long)range_start,
928 (unsigned long long)range_end);
Joel Becker56934862010-07-01 15:13:31 -0700929 BUG_ON(range_start >= range_end);
930
931 while (zero_pos < range_end) {
932 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
933 if (next_pos > range_end)
934 next_pos = range_end;
935 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
936 if (rc < 0) {
937 mlog_errno(rc);
938 break;
939 }
940 zero_pos = next_pos;
Mark Fashehe2057c52006-10-03 17:53:05 -0700941
942 /*
943 * Very large extends have the potential to lock up
944 * the cpu for extended periods of time.
945 */
946 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800947 }
948
Joel Becker56934862010-07-01 15:13:31 -0700949 return rc;
950}
951
952int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
953 loff_t zero_to_size)
954{
955 int ret = 0;
956 u64 zero_start, range_start = 0, range_end = 0;
957 struct super_block *sb = inode->i_sb;
958
959 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
Tao Ma468eedd2011-02-22 22:14:41 +0800960 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
961 (unsigned long long)zero_start,
962 (unsigned long long)i_size_read(inode));
Joel Becker56934862010-07-01 15:13:31 -0700963 while (zero_start < zero_to_size) {
964 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
965 zero_to_size,
966 &range_start,
967 &range_end);
968 if (ret) {
969 mlog_errno(ret);
970 break;
971 }
972 if (!range_end)
973 break;
974 /* Trim the ends */
975 if (range_start < zero_start)
976 range_start = zero_start;
977 if (range_end > zero_to_size)
978 range_end = zero_to_size;
979
980 ret = ocfs2_zero_extend_range(inode, range_start,
981 range_end);
982 if (ret) {
983 mlog_errno(ret);
984 break;
985 }
986 zero_start = range_end;
987 }
988
Mark Fashehccd979b2005-12-15 14:31:24 -0800989 return ret;
990}
991
Joel Becker56934862010-07-01 15:13:31 -0700992int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
993 u64 new_i_size, u64 zero_to)
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700994{
995 int ret;
996 u32 clusters_to_add;
997 struct ocfs2_inode_info *oi = OCFS2_I(inode);
998
Joel Becker56934862010-07-01 15:13:31 -0700999 /*
1000 * Only quota files call this without a bh, and they can't be
1001 * refcounted.
1002 */
1003 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1004 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1005
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001006 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1007 if (clusters_to_add < oi->ip_clusters)
1008 clusters_to_add = 0;
1009 else
1010 clusters_to_add -= oi->ip_clusters;
1011
1012 if (clusters_to_add) {
1013 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1014 clusters_to_add, 0);
1015 if (ret) {
1016 mlog_errno(ret);
1017 goto out;
1018 }
1019 }
1020
1021 /*
1022 * Call this even if we don't add any clusters to the tree. We
1023 * still need to zero the area between the old i_size and the
1024 * new i_size.
1025 */
Joel Becker56934862010-07-01 15:13:31 -07001026 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001027 if (ret < 0)
1028 mlog_errno(ret);
1029
1030out:
1031 return ret;
1032}
1033
Mark Fashehccd979b2005-12-15 14:31:24 -08001034static int ocfs2_extend_file(struct inode *inode,
1035 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001036 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -08001037{
Mark Fashehc934a922007-10-18 15:23:46 -07001038 int ret = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001039 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001040
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001041 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -07001042
Mark Fashehccd979b2005-12-15 14:31:24 -08001043 /* setattr sometimes calls us like this. */
1044 if (new_i_size == 0)
1045 goto out;
1046
1047 if (i_size_read(inode) == new_i_size)
Joel Becker56934862010-07-01 15:13:31 -07001048 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001049 BUG_ON(new_i_size < i_size_read(inode));
1050
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001051 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001052 * The alloc sem blocks people in read/write from reading our
1053 * allocation until we're done changing it. We depend on
1054 * i_mutex to block other extend/truncate calls while we're
Joel Becker56934862010-07-01 15:13:31 -07001055 * here. We even have to hold it for sparse files because there
1056 * might be some tail zeroing.
Mark Fasheh0effef72006-10-03 17:44:42 -07001057 */
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001058 down_write(&oi->ip_alloc_sem);
1059
1060 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1061 /*
1062 * We can optimize small extends by keeping the inodes
1063 * inline data.
1064 */
1065 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1066 up_write(&oi->ip_alloc_sem);
1067 goto out_update_size;
1068 }
1069
1070 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1071 if (ret) {
1072 up_write(&oi->ip_alloc_sem);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001073 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001074 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001075 }
1076 }
1077
Joel Becker56934862010-07-01 15:13:31 -07001078 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1079 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1080 else
1081 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1082 new_i_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001083
1084 up_write(&oi->ip_alloc_sem);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001085
Mark Fasheh0effef72006-10-03 17:44:42 -07001086 if (ret < 0) {
1087 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001088 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001089 }
1090
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001091out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001092 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1093 if (ret < 0)
1094 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -07001095
Mark Fashehccd979b2005-12-15 14:31:24 -08001096out:
1097 return ret;
1098}
1099
1100int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1101{
1102 int status = 0, size_change;
1103 struct inode *inode = dentry->d_inode;
1104 struct super_block *sb = inode->i_sb;
1105 struct ocfs2_super *osb = OCFS2_SB(sb);
1106 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001107 handle_t *handle = NULL;
Jan Kara65bac572009-06-02 14:24:01 +02001108 struct dquot *transfer_to[MAXQUOTAS] = { };
Jan Kara52a9ee22010-05-13 20:18:45 +02001109 int qtype;
Mark Fashehccd979b2005-12-15 14:31:24 -08001110
Tao Ma468eedd2011-02-22 22:14:41 +08001111 trace_ocfs2_setattr(inode, dentry,
1112 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1113 dentry->d_name.len, dentry->d_name.name,
1114 attr->ia_valid, attr->ia_mode,
Eric W. Biedermanba613562013-01-31 17:33:53 -08001115 from_kuid(&init_user_ns, attr->ia_uid),
1116 from_kgid(&init_user_ns, attr->ia_gid));
Mark Fashehccd979b2005-12-15 14:31:24 -08001117
Sunil Mushranbc535802008-04-18 10:23:53 -07001118 /* ensuring we don't even attempt to truncate a symlink */
1119 if (S_ISLNK(inode->i_mode))
1120 attr->ia_valid &= ~ATTR_SIZE;
1121
Mark Fashehccd979b2005-12-15 14:31:24 -08001122#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1123 | ATTR_GID | ATTR_UID | ATTR_MODE)
Tao Ma468eedd2011-02-22 22:14:41 +08001124 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
Mark Fashehccd979b2005-12-15 14:31:24 -08001125 return 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001126
1127 status = inode_change_ok(inode, attr);
1128 if (status)
1129 return status;
1130
Dmitry Monakhov12755622010-04-08 22:04:20 +04001131 if (is_quota_modification(inode, attr))
1132 dquot_initialize(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001133 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1134 if (size_change) {
1135 status = ocfs2_rw_lock(inode, 1);
1136 if (status < 0) {
1137 mlog_errno(status);
1138 goto bail;
1139 }
1140 }
1141
Mark Fashehe63aecb62007-10-18 15:30:42 -07001142 status = ocfs2_inode_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001143 if (status < 0) {
1144 if (status != -ENOENT)
1145 mlog_errno(status);
1146 goto bail_unlock_rw;
1147 }
1148
1149 if (size_change && attr->ia_size != i_size_read(inode)) {
Wengang Wang5051f762010-02-26 18:18:25 +08001150 status = inode_newsize_ok(inode, attr->ia_size);
1151 if (status)
Mark Fashehce76fd32007-07-20 12:02:14 -07001152 goto bail_unlock;
Mark Fashehce76fd32007-07-20 12:02:14 -07001153
Christoph Hellwig562c72aa52011-06-24 14:29:45 -04001154 inode_dio_wait(inode);
1155
Joel Becker2b4e30f2008-09-03 20:03:41 -07001156 if (i_size_read(inode) > attr->ia_size) {
1157 if (ocfs2_should_order_data(inode)) {
1158 status = ocfs2_begin_ordered_truncate(inode,
1159 attr->ia_size);
1160 if (status)
1161 goto bail_unlock;
1162 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001163 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
Joel Becker2b4e30f2008-09-03 20:03:41 -07001164 } else
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001165 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001166 if (status < 0) {
1167 if (status != -ENOSPC)
1168 mlog_errno(status);
1169 status = -ENOSPC;
1170 goto bail_unlock;
1171 }
1172 }
1173
Eric W. Biederman488c8ef2013-01-31 17:35:29 -08001174 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1175 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Jan Kara65bac572009-06-02 14:24:01 +02001176 /*
1177 * Gather pointers to quota structures so that allocation /
1178 * freeing of quota structures happens here and not inside
Christoph Hellwigb43fa822010-03-03 09:05:03 -05001179 * dquot_transfer() where we have problems with lock ordering
Jan Kara65bac572009-06-02 14:24:01 +02001180 */
Eric W. Biederman488c8ef2013-01-31 17:35:29 -08001181 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
Jan Karaa90714c2008-10-09 19:38:40 +02001182 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1183 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
Eric W. Biedermanaca645a2012-09-16 03:11:50 -07001184 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
Jan Kara52a9ee22010-05-13 20:18:45 +02001185 if (!transfer_to[USRQUOTA]) {
Jan Kara65bac572009-06-02 14:24:01 +02001186 status = -ESRCH;
Jan Karaa90714c2008-10-09 19:38:40 +02001187 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001188 }
Jan Karaa90714c2008-10-09 19:38:40 +02001189 }
Eric W. Biederman488c8ef2013-01-31 17:35:29 -08001190 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
Jan Karaa90714c2008-10-09 19:38:40 +02001191 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1192 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
Eric W. Biedermanaca645a2012-09-16 03:11:50 -07001193 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
Jan Kara52a9ee22010-05-13 20:18:45 +02001194 if (!transfer_to[GRPQUOTA]) {
Jan Kara65bac572009-06-02 14:24:01 +02001195 status = -ESRCH;
Jan Karaa90714c2008-10-09 19:38:40 +02001196 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001197 }
Jan Karaa90714c2008-10-09 19:38:40 +02001198 }
Jan Kara65bac572009-06-02 14:24:01 +02001199 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1200 2 * ocfs2_quota_trans_credits(sb));
Jan Karaa90714c2008-10-09 19:38:40 +02001201 if (IS_ERR(handle)) {
1202 status = PTR_ERR(handle);
1203 mlog_errno(status);
1204 goto bail_unlock;
1205 }
Jan Kara52a9ee22010-05-13 20:18:45 +02001206 status = __dquot_transfer(inode, transfer_to);
Jan Karaa90714c2008-10-09 19:38:40 +02001207 if (status < 0)
1208 goto bail_commit;
1209 } else {
1210 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1211 if (IS_ERR(handle)) {
1212 status = PTR_ERR(handle);
1213 mlog_errno(status);
1214 goto bail_unlock;
1215 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001216 }
1217
Christoph Hellwig10257742010-06-04 11:30:02 +02001218 setattr_copy(inode, attr);
1219 mark_inode_dirty(inode);
1220
Mark Fashehccd979b2005-12-15 14:31:24 -08001221 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1222 if (status < 0)
1223 mlog_errno(status);
1224
1225bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001226 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001227bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001228 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001229bail_unlock_rw:
1230 if (size_change)
1231 ocfs2_rw_unlock(inode, 1);
1232bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001233 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001234
Jan Kara65bac572009-06-02 14:24:01 +02001235 /* Release quota pointers in case we acquired them */
Jan Kara52a9ee22010-05-13 20:18:45 +02001236 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
Jan Kara65bac572009-06-02 14:24:01 +02001237 dqput(transfer_to[qtype]);
Jan Kara65bac572009-06-02 14:24:01 +02001238
Tiger Yang060bc662008-11-14 11:17:29 +08001239 if (!status && attr->ia_valid & ATTR_MODE) {
1240 status = ocfs2_acl_chmod(inode);
1241 if (status < 0)
1242 mlog_errno(status);
1243 }
1244
Mark Fashehccd979b2005-12-15 14:31:24 -08001245 return status;
1246}
1247
1248int ocfs2_getattr(struct vfsmount *mnt,
1249 struct dentry *dentry,
1250 struct kstat *stat)
1251{
1252 struct inode *inode = dentry->d_inode;
1253 struct super_block *sb = dentry->d_inode->i_sb;
1254 struct ocfs2_super *osb = sb->s_fs_info;
1255 int err;
1256
Mark Fashehccd979b2005-12-15 14:31:24 -08001257 err = ocfs2_inode_revalidate(dentry);
1258 if (err) {
1259 if (err != -ENOENT)
1260 mlog_errno(err);
1261 goto bail;
1262 }
1263
1264 generic_fillattr(inode, stat);
1265
1266 /* We set the blksize from the cluster size for performance */
1267 stat->blksize = osb->s_clustersize;
1268
1269bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001270 return err;
1271}
1272
Al Viro10556cb22011-06-20 19:28:19 -04001273int ocfs2_permission(struct inode *inode, int mask)
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001274{
1275 int ret;
1276
Al Viro10556cb22011-06-20 19:28:19 -04001277 if (mask & MAY_NOT_BLOCK)
Nick Pigginb74c79e2011-01-07 17:49:58 +11001278 return -ECHILD;
1279
Mark Fashehe63aecb62007-10-18 15:30:42 -07001280 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001281 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001282 if (ret != -ENOENT)
1283 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001284 goto out;
1285 }
1286
Al Viro2830ba72011-06-20 19:16:29 -04001287 ret = generic_permission(inode, mask);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001288
Mark Fashehe63aecb62007-10-18 15:30:42 -07001289 ocfs2_inode_unlock(inode, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001290out:
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001291 return ret;
1292}
1293
Mark Fashehb2580102007-03-09 16:53:21 -08001294static int __ocfs2_write_remove_suid(struct inode *inode,
1295 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001296{
1297 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001298 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001299 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1300 struct ocfs2_dinode *di;
1301
Tao Ma468eedd2011-02-22 22:14:41 +08001302 trace_ocfs2_write_remove_suid(
1303 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1304 inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001305
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001306 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001307 if (IS_ERR(handle)) {
1308 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001309 mlog_errno(ret);
1310 goto out;
1311 }
1312
Joel Becker0cf2f762009-02-12 16:41:25 -08001313 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -07001314 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001315 if (ret < 0) {
1316 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001317 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001318 }
1319
1320 inode->i_mode &= ~S_ISUID;
1321 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1322 inode->i_mode &= ~S_ISGID;
1323
1324 di = (struct ocfs2_dinode *) bh->b_data;
1325 di->i_mode = cpu_to_le16(inode->i_mode);
1326
Joel Beckerec20cec2010-03-19 14:13:52 -07001327 ocfs2_journal_dirty(handle, bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001328
Mark Fashehccd979b2005-12-15 14:31:24 -08001329out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001330 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001331out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001332 return ret;
1333}
1334
Mark Fasheh9517bac2007-02-09 20:24:12 -08001335/*
1336 * Will look for holes and unwritten extents in the range starting at
1337 * pos for count bytes (inclusive).
1338 */
1339static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1340 size_t count)
1341{
1342 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001343 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001344 u32 cpos, clusters, extent_len, phys_cpos;
1345 struct super_block *sb = inode->i_sb;
1346
1347 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1348 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1349
1350 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001351 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1352 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001353 if (ret < 0) {
1354 mlog_errno(ret);
1355 goto out;
1356 }
1357
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001358 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001359 ret = 1;
1360 break;
1361 }
1362
1363 if (extent_len > clusters)
1364 extent_len = clusters;
1365
1366 clusters -= extent_len;
1367 cpos += extent_len;
1368 }
1369out:
1370 return ret;
1371}
1372
Mark Fashehb2580102007-03-09 16:53:21 -08001373static int ocfs2_write_remove_suid(struct inode *inode)
1374{
1375 int ret;
1376 struct buffer_head *bh = NULL;
Mark Fashehb2580102007-03-09 16:53:21 -08001377
Joel Beckerb657c952008-11-13 14:49:11 -08001378 ret = ocfs2_read_inode_block(inode, &bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001379 if (ret < 0) {
1380 mlog_errno(ret);
1381 goto out;
1382 }
1383
1384 ret = __ocfs2_write_remove_suid(inode, bh);
1385out:
1386 brelse(bh);
1387 return ret;
1388}
1389
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001390/*
1391 * Allocate enough extents to cover the region starting at byte offset
1392 * start for len bytes. Existing extents are skipped, any extents
1393 * added are marked as "unwritten".
1394 */
1395static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1396 u64 start, u64 len)
1397{
1398 int ret;
1399 u32 cpos, phys_cpos, clusters, alloc_size;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001400 u64 end = start + len;
1401 struct buffer_head *di_bh = NULL;
1402
1403 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Joel Beckerb657c952008-11-13 14:49:11 -08001404 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001405 if (ret) {
1406 mlog_errno(ret);
1407 goto out;
1408 }
1409
1410 /*
1411 * Nothing to do if the requested reservation range
1412 * fits within the inode.
1413 */
1414 if (ocfs2_size_fits_inline_data(di_bh, end))
1415 goto out;
1416
1417 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1418 if (ret) {
1419 mlog_errno(ret);
1420 goto out;
1421 }
1422 }
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001423
1424 /*
1425 * We consider both start and len to be inclusive.
1426 */
1427 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1428 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1429 clusters -= cpos;
1430
1431 while (clusters) {
1432 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1433 &alloc_size, NULL);
1434 if (ret) {
1435 mlog_errno(ret);
1436 goto out;
1437 }
1438
1439 /*
1440 * Hole or existing extent len can be arbitrary, so
1441 * cap it to our own allocation request.
1442 */
1443 if (alloc_size > clusters)
1444 alloc_size = clusters;
1445
1446 if (phys_cpos) {
1447 /*
1448 * We already have an allocation at this
1449 * region so we can safely skip it.
1450 */
1451 goto next;
1452 }
1453
1454 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1455 if (ret) {
1456 if (ret != -ENOSPC)
1457 mlog_errno(ret);
1458 goto out;
1459 }
1460
1461next:
1462 cpos += alloc_size;
1463 clusters -= alloc_size;
1464 }
1465
1466 ret = 0;
1467out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001468
1469 brelse(di_bh);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001470 return ret;
1471}
1472
Mark Fasheh063c4562007-07-03 13:34:11 -07001473/*
1474 * Truncate a byte range, avoiding pages within partial clusters. This
1475 * preserves those pages for the zeroing code to write to.
1476 */
1477static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1478 u64 byte_len)
1479{
1480 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1481 loff_t start, end;
1482 struct address_space *mapping = inode->i_mapping;
1483
1484 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1485 end = byte_start + byte_len;
1486 end = end & ~(osb->s_clustersize - 1);
1487
1488 if (start < end) {
1489 unmap_mapping_range(mapping, start, end - start, 0);
1490 truncate_inode_pages_range(mapping, start, end - 1);
1491 }
1492}
1493
1494static int ocfs2_zero_partial_clusters(struct inode *inode,
1495 u64 start, u64 len)
1496{
1497 int ret = 0;
1498 u64 tmpend, end = start + len;
1499 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1500 unsigned int csize = osb->s_clustersize;
1501 handle_t *handle;
1502
1503 /*
1504 * The "start" and "end" values are NOT necessarily part of
1505 * the range whose allocation is being deleted. Rather, this
1506 * is what the user passed in with the request. We must zero
1507 * partial clusters here. There's no need to worry about
1508 * physical allocation - the zeroing code knows to skip holes.
1509 */
Tao Ma468eedd2011-02-22 22:14:41 +08001510 trace_ocfs2_zero_partial_clusters(
1511 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1512 (unsigned long long)start, (unsigned long long)end);
Mark Fasheh063c4562007-07-03 13:34:11 -07001513
1514 /*
1515 * If both edges are on a cluster boundary then there's no
1516 * zeroing required as the region is part of the allocation to
1517 * be truncated.
1518 */
1519 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1520 goto out;
1521
1522 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001523 if (IS_ERR(handle)) {
1524 ret = PTR_ERR(handle);
Mark Fasheh063c4562007-07-03 13:34:11 -07001525 mlog_errno(ret);
1526 goto out;
1527 }
1528
1529 /*
1530 * We want to get the byte offset of the end of the 1st cluster.
1531 */
1532 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1533 if (tmpend > end)
1534 tmpend = end;
1535
Tao Ma468eedd2011-02-22 22:14:41 +08001536 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1537 (unsigned long long)tmpend);
Mark Fasheh063c4562007-07-03 13:34:11 -07001538
1539 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1540 if (ret)
1541 mlog_errno(ret);
1542
1543 if (tmpend < end) {
1544 /*
1545 * This may make start and end equal, but the zeroing
1546 * code will skip any work in that case so there's no
1547 * need to catch it up here.
1548 */
1549 start = end & ~(osb->s_clustersize - 1);
1550
Tao Ma468eedd2011-02-22 22:14:41 +08001551 trace_ocfs2_zero_partial_clusters_range2(
1552 (unsigned long long)start, (unsigned long long)end);
Mark Fasheh063c4562007-07-03 13:34:11 -07001553
1554 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1555 if (ret)
1556 mlog_errno(ret);
1557 }
1558
1559 ocfs2_commit_trans(osb, handle);
1560out:
1561 return ret;
1562}
1563
Tristan Yec1631d42010-05-11 17:54:45 +08001564static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1565{
1566 int i;
1567 struct ocfs2_extent_rec *rec = NULL;
1568
1569 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1570
1571 rec = &el->l_recs[i];
1572
1573 if (le32_to_cpu(rec->e_cpos) < pos)
1574 break;
1575 }
1576
1577 return i;
1578}
1579
1580/*
1581 * Helper to calculate the punching pos and length in one run, we handle the
1582 * following three cases in order:
1583 *
1584 * - remove the entire record
1585 * - remove a partial record
1586 * - no record needs to be removed (hole-punching completed)
1587*/
1588static void ocfs2_calc_trunc_pos(struct inode *inode,
1589 struct ocfs2_extent_list *el,
1590 struct ocfs2_extent_rec *rec,
1591 u32 trunc_start, u32 *trunc_cpos,
1592 u32 *trunc_len, u32 *trunc_end,
1593 u64 *blkno, int *done)
1594{
1595 int ret = 0;
1596 u32 coff, range;
1597
1598 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1599
1600 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
Tristan Ye9a790ba2011-05-12 20:47:07 +08001601 /*
1602 * remove an entire extent record.
1603 */
Tristan Yec1631d42010-05-11 17:54:45 +08001604 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1605 /*
1606 * Skip holes if any.
1607 */
1608 if (range < *trunc_end)
1609 *trunc_end = range;
1610 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1611 *blkno = le64_to_cpu(rec->e_blkno);
1612 *trunc_end = le32_to_cpu(rec->e_cpos);
1613 } else if (range > trunc_start) {
Tristan Ye9a790ba2011-05-12 20:47:07 +08001614 /*
1615 * remove a partial extent record, which means we're
1616 * removing the last extent record.
1617 */
Tristan Yec1631d42010-05-11 17:54:45 +08001618 *trunc_cpos = trunc_start;
Tristan Ye9a790ba2011-05-12 20:47:07 +08001619 /*
1620 * skip hole if any.
1621 */
1622 if (range < *trunc_end)
1623 *trunc_end = range;
Tristan Yec1631d42010-05-11 17:54:45 +08001624 *trunc_len = *trunc_end - trunc_start;
1625 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1626 *blkno = le64_to_cpu(rec->e_blkno) +
1627 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1628 *trunc_end = trunc_start;
1629 } else {
1630 /*
1631 * It may have two following possibilities:
1632 *
1633 * - last record has been removed
1634 * - trunc_start was within a hole
1635 *
1636 * both two cases mean the completion of hole punching.
1637 */
1638 ret = 1;
1639 }
1640
1641 *done = ret;
1642}
1643
Mark Fasheh063c4562007-07-03 13:34:11 -07001644static int ocfs2_remove_inode_range(struct inode *inode,
1645 struct buffer_head *di_bh, u64 byte_start,
1646 u64 byte_len)
1647{
Tristan Yec1631d42010-05-11 17:54:45 +08001648 int ret = 0, flags = 0, done = 0, i;
1649 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1650 u32 cluster_in_el;
Mark Fasheh063c4562007-07-03 13:34:11 -07001651 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1652 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehb1967d02007-11-20 11:56:39 -08001653 struct address_space *mapping = inode->i_mapping;
Mark Fashehfecc0112008-11-12 15:16:38 -08001654 struct ocfs2_extent_tree et;
Tristan Yec1631d42010-05-11 17:54:45 +08001655 struct ocfs2_path *path = NULL;
1656 struct ocfs2_extent_list *el = NULL;
1657 struct ocfs2_extent_rec *rec = NULL;
Tristan Yee8aec062010-05-11 17:54:43 +08001658 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Tristan Yec1631d42010-05-11 17:54:45 +08001659 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
Mark Fasheh063c4562007-07-03 13:34:11 -07001660
Joel Becker5e404e92009-02-13 03:54:22 -08001661 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Mark Fasheh063c4562007-07-03 13:34:11 -07001662 ocfs2_init_dealloc_ctxt(&dealloc);
1663
Tao Ma468eedd2011-02-22 22:14:41 +08001664 trace_ocfs2_remove_inode_range(
1665 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1666 (unsigned long long)byte_start,
1667 (unsigned long long)byte_len);
1668
Mark Fasheh063c4562007-07-03 13:34:11 -07001669 if (byte_len == 0)
1670 return 0;
1671
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001672 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1673 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
Mark Fashehb1967d02007-11-20 11:56:39 -08001674 byte_start + byte_len, 0);
1675 if (ret) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001676 mlog_errno(ret);
Mark Fashehb1967d02007-11-20 11:56:39 -08001677 goto out;
1678 }
1679 /*
1680 * There's no need to get fancy with the page cache
1681 * truncate of an inline-data inode. We're talking
1682 * about less than a page here, which will be cached
1683 * in the dinode buffer anyway.
1684 */
1685 unmap_mapping_range(mapping, 0, 0, 0);
1686 truncate_inode_pages(mapping, 0);
1687 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001688 }
1689
Tristan Yee8aec062010-05-11 17:54:43 +08001690 /*
1691 * For reflinks, we may need to CoW 2 clusters which might be
1692 * partially zero'd later, if hole's start and end offset were
1693 * within one cluster(means is not exactly aligned to clustersize).
1694 */
Mark Fasheh063c4562007-07-03 13:34:11 -07001695
Tristan Yee8aec062010-05-11 17:54:43 +08001696 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1697
1698 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1699 if (ret) {
1700 mlog_errno(ret);
1701 goto out;
1702 }
1703
1704 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1705 if (ret) {
1706 mlog_errno(ret);
1707 goto out;
1708 }
1709 }
1710
Mark Fasheh063c4562007-07-03 13:34:11 -07001711 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
Tristan Yec1631d42010-05-11 17:54:45 +08001712 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1713 cluster_in_el = trunc_end;
Mark Fasheh063c4562007-07-03 13:34:11 -07001714
Mark Fasheh063c4562007-07-03 13:34:11 -07001715 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1716 if (ret) {
1717 mlog_errno(ret);
1718 goto out;
1719 }
1720
Tristan Yec1631d42010-05-11 17:54:45 +08001721 path = ocfs2_new_path_from_et(&et);
1722 if (!path) {
1723 ret = -ENOMEM;
1724 mlog_errno(ret);
1725 goto out;
1726 }
1727
1728 while (trunc_end > trunc_start) {
1729
1730 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1731 cluster_in_el);
Mark Fasheh063c4562007-07-03 13:34:11 -07001732 if (ret) {
1733 mlog_errno(ret);
1734 goto out;
1735 }
1736
Tristan Yec1631d42010-05-11 17:54:45 +08001737 el = path_leaf_el(path);
Mark Fasheh063c4562007-07-03 13:34:11 -07001738
Tristan Yec1631d42010-05-11 17:54:45 +08001739 i = ocfs2_find_rec(el, trunc_end);
1740 /*
1741 * Need to go to previous extent block.
1742 */
1743 if (i < 0) {
1744 if (path->p_tree_depth == 0)
1745 break;
1746
1747 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1748 path,
1749 &cluster_in_el);
Mark Fasheh063c4562007-07-03 13:34:11 -07001750 if (ret) {
1751 mlog_errno(ret);
1752 goto out;
1753 }
Tristan Yec1631d42010-05-11 17:54:45 +08001754
1755 /*
1756 * We've reached the leftmost extent block,
1757 * it's safe to leave.
1758 */
1759 if (cluster_in_el == 0)
1760 break;
1761
1762 /*
1763 * The 'pos' searched for previous extent block is
1764 * always one cluster less than actual trunc_end.
1765 */
1766 trunc_end = cluster_in_el + 1;
1767
1768 ocfs2_reinit_path(path, 1);
1769
1770 continue;
1771
1772 } else
1773 rec = &el->l_recs[i];
1774
1775 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1776 &trunc_len, &trunc_end, &blkno, &done);
1777 if (done)
1778 break;
1779
1780 flags = rec->e_flags;
1781 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1782
1783 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1784 phys_cpos, trunc_len, flags,
1785 &dealloc, refcount_loc);
1786 if (ret < 0) {
1787 mlog_errno(ret);
1788 goto out;
Mark Fasheh063c4562007-07-03 13:34:11 -07001789 }
1790
Tristan Yec1631d42010-05-11 17:54:45 +08001791 cluster_in_el = trunc_end;
1792
1793 ocfs2_reinit_path(path, 1);
Mark Fasheh063c4562007-07-03 13:34:11 -07001794 }
1795
1796 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1797
1798out:
Younger Liu7aebff182013-09-11 14:19:59 -07001799 ocfs2_free_path(path);
Mark Fasheh063c4562007-07-03 13:34:11 -07001800 ocfs2_schedule_truncate_log_flush(osb, 1);
1801 ocfs2_run_deallocs(osb, &dealloc);
1802
1803 return ret;
1804}
1805
Mark Fashehb2580102007-03-09 16:53:21 -08001806/*
1807 * Parts of this function taken from xfs_change_file_space()
1808 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001809static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1810 loff_t f_pos, unsigned int cmd,
1811 struct ocfs2_space_resv *sr,
1812 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001813{
1814 int ret;
1815 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001816 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001817 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1818 struct buffer_head *di_bh = NULL;
1819 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001820 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001821
Mark Fashehb2580102007-03-09 16:53:21 -08001822 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1823 return -EROFS;
1824
1825 mutex_lock(&inode->i_mutex);
1826
1827 /*
1828 * This prevents concurrent writes on other nodes
1829 */
1830 ret = ocfs2_rw_lock(inode, 1);
1831 if (ret) {
1832 mlog_errno(ret);
1833 goto out;
1834 }
1835
Mark Fashehe63aecb62007-10-18 15:30:42 -07001836 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001837 if (ret) {
1838 mlog_errno(ret);
1839 goto out_rw_unlock;
1840 }
1841
1842 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1843 ret = -EPERM;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001844 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001845 }
1846
1847 switch (sr->l_whence) {
1848 case 0: /*SEEK_SET*/
1849 break;
1850 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001851 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001852 break;
1853 case 2: /*SEEK_END*/
1854 sr->l_start += i_size_read(inode);
1855 break;
1856 default:
1857 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001858 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001859 }
1860 sr->l_whence = 0;
1861
1862 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1863
1864 if (sr->l_start < 0
1865 || sr->l_start > max_off
1866 || (sr->l_start + llen) < 0
1867 || (sr->l_start + llen) > max_off) {
1868 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001869 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001870 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001871 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001872
1873 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1874 if (sr->l_len <= 0) {
1875 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001876 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001877 }
1878 }
1879
Mark Fasheh385820a2007-07-19 00:14:38 -07001880 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001881 ret = __ocfs2_write_remove_suid(inode, di_bh);
1882 if (ret) {
1883 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001884 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001885 }
1886 }
1887
1888 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1889 switch (cmd) {
1890 case OCFS2_IOC_RESVSP:
1891 case OCFS2_IOC_RESVSP64:
1892 /*
1893 * This takes unsigned offsets, but the signed ones we
1894 * pass have been checked against overflow above.
1895 */
1896 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1897 sr->l_len);
1898 break;
1899 case OCFS2_IOC_UNRESVSP:
1900 case OCFS2_IOC_UNRESVSP64:
1901 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1902 sr->l_len);
1903 break;
1904 default:
1905 ret = -EINVAL;
1906 }
1907 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1908 if (ret) {
1909 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001910 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001911 }
1912
1913 /*
1914 * We update c/mtime for these changes
1915 */
1916 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1917 if (IS_ERR(handle)) {
1918 ret = PTR_ERR(handle);
1919 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001920 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001921 }
1922
Mark Fasheh385820a2007-07-19 00:14:38 -07001923 if (change_size && i_size_read(inode) < size)
1924 i_size_write(inode, size);
1925
Mark Fashehb2580102007-03-09 16:53:21 -08001926 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1927 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1928 if (ret < 0)
1929 mlog_errno(ret);
1930
Luis Henriquesa4e08d02012-07-11 14:02:10 -07001931 if (file && (file->f_flags & O_SYNC))
Mark Fashehdf295d4a2011-11-16 12:03:10 -08001932 handle->h_sync = 1;
1933
Mark Fashehb2580102007-03-09 16:53:21 -08001934 ocfs2_commit_trans(osb, handle);
1935
Mark Fashehe63aecb62007-10-18 15:30:42 -07001936out_inode_unlock:
Mark Fashehb2580102007-03-09 16:53:21 -08001937 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001938 ocfs2_inode_unlock(inode, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001939out_rw_unlock:
1940 ocfs2_rw_unlock(inode, 1);
1941
Mark Fashehb2580102007-03-09 16:53:21 -08001942out:
Julia Lawallc259ae52008-07-21 09:59:15 +02001943 mutex_unlock(&inode->i_mutex);
Mark Fashehb2580102007-03-09 16:53:21 -08001944 return ret;
1945}
1946
Mark Fasheh385820a2007-07-19 00:14:38 -07001947int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1948 struct ocfs2_space_resv *sr)
1949{
Al Viro496ad9a2013-01-23 17:07:38 -05001950 struct inode *inode = file_inode(file);
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08001951 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Jan Karafef69252012-06-12 16:20:40 +02001952 int ret;
Mark Fasheh385820a2007-07-19 00:14:38 -07001953
1954 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1955 !ocfs2_writes_unwritten_extents(osb))
1956 return -ENOTTY;
1957 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1958 !ocfs2_sparse_alloc(osb))
1959 return -ENOTTY;
1960
1961 if (!S_ISREG(inode->i_mode))
1962 return -EINVAL;
1963
1964 if (!(file->f_mode & FMODE_WRITE))
1965 return -EBADF;
1966
Jan Karafef69252012-06-12 16:20:40 +02001967 ret = mnt_want_write_file(file);
1968 if (ret)
1969 return ret;
1970 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1971 mnt_drop_write_file(file);
1972 return ret;
Mark Fasheh385820a2007-07-19 00:14:38 -07001973}
1974
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001975static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
Mark Fasheh385820a2007-07-19 00:14:38 -07001976 loff_t len)
1977{
Al Viro496ad9a2013-01-23 17:07:38 -05001978 struct inode *inode = file_inode(file);
Mark Fasheh385820a2007-07-19 00:14:38 -07001979 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1980 struct ocfs2_space_resv sr;
1981 int change_size = 1;
Josef Bacikdb47fef2010-11-17 20:46:17 -05001982 int cmd = OCFS2_IOC_RESVSP64;
Mark Fasheh385820a2007-07-19 00:14:38 -07001983
Christoph Hellwig64c23e82011-01-14 13:07:30 +01001984 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1985 return -EOPNOTSUPP;
Mark Fasheh385820a2007-07-19 00:14:38 -07001986 if (!ocfs2_writes_unwritten_extents(osb))
1987 return -EOPNOTSUPP;
1988
Mark Fasheh385820a2007-07-19 00:14:38 -07001989 if (mode & FALLOC_FL_KEEP_SIZE)
1990 change_size = 0;
1991
Josef Bacikdb47fef2010-11-17 20:46:17 -05001992 if (mode & FALLOC_FL_PUNCH_HOLE)
1993 cmd = OCFS2_IOC_UNRESVSP64;
1994
Mark Fasheh385820a2007-07-19 00:14:38 -07001995 sr.l_whence = 0;
1996 sr.l_start = (s64)offset;
1997 sr.l_len = (s64)len;
1998
Josef Bacikdb47fef2010-11-17 20:46:17 -05001999 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2000 change_size);
Mark Fasheh385820a2007-07-19 00:14:38 -07002001}
2002
Tao Ma293b2f72009-08-25 08:02:48 +08002003int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2004 size_t count)
2005{
2006 int ret = 0;
2007 unsigned int extent_flags;
2008 u32 cpos, clusters, extent_len, phys_cpos;
2009 struct super_block *sb = inode->i_sb;
2010
2011 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
Tao Ma2f48d5932009-10-15 11:10:49 +08002012 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2013 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Tao Ma293b2f72009-08-25 08:02:48 +08002014 return 0;
2015
2016 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2017 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2018
2019 while (clusters) {
2020 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2021 &extent_flags);
2022 if (ret < 0) {
2023 mlog_errno(ret);
2024 goto out;
2025 }
2026
2027 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2028 ret = 1;
2029 break;
2030 }
2031
2032 if (extent_len > clusters)
2033 extent_len = clusters;
2034
2035 clusters -= extent_len;
2036 cpos += extent_len;
2037 }
2038out:
2039 return ret;
2040}
2041
Mark Fasheha11f7e62011-06-22 14:23:38 -07002042static void ocfs2_aiodio_wait(struct inode *inode)
2043{
2044 wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2045
2046 wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2047}
2048
2049static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2050{
2051 int blockmask = inode->i_sb->s_blocksize - 1;
2052 loff_t final_size = pos + count;
2053
2054 if ((pos & blockmask) || (final_size & blockmask))
2055 return 1;
2056 return 0;
2057}
2058
Tao Ma293b2f72009-08-25 08:02:48 +08002059static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
Tao Mab8908232010-08-12 10:27:14 +08002060 struct file *file,
Tao Ma293b2f72009-08-25 08:02:48 +08002061 loff_t pos, size_t count,
2062 int *meta_level)
2063{
2064 int ret;
2065 struct buffer_head *di_bh = NULL;
2066 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2067 u32 clusters =
2068 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2069
2070 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2071 if (ret) {
2072 mlog_errno(ret);
2073 goto out;
2074 }
2075
2076 *meta_level = 1;
2077
Tiger Yangc7dd3392013-08-13 16:00:58 -07002078 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
Tao Ma293b2f72009-08-25 08:02:48 +08002079 if (ret)
2080 mlog_errno(ret);
2081out:
2082 brelse(di_bh);
2083 return ret;
2084}
2085
Tao Mab8908232010-08-12 10:27:14 +08002086static int ocfs2_prepare_inode_for_write(struct file *file,
Tiger Yang8659ac22006-10-17 18:29:52 -07002087 loff_t *ppos,
2088 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08002089 int appending,
Tao Ma86470e92009-12-03 21:55:05 +08002090 int *direct_io,
2091 int *has_refcount)
Mark Fashehccd979b2005-12-15 14:31:24 -08002092{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002093 int ret = 0, meta_level = 0;
Tao Mab8908232010-08-12 10:27:14 +08002094 struct dentry *dentry = file->f_path.dentry;
Tiger Yang8659ac22006-10-17 18:29:52 -07002095 struct inode *inode = dentry->d_inode;
Tao Ma468eedd2011-02-22 22:14:41 +08002096 loff_t saved_pos = 0, end;
Mark Fashehccd979b2005-12-15 14:31:24 -08002097
Sunil Mushran2bd63212010-01-25 16:57:38 -08002098 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002099 * We start with a read level meta lock and only jump to an ex
2100 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08002101 */
Mark Fashehccd979b2005-12-15 14:31:24 -08002102 for(;;) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07002103 ret = ocfs2_inode_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002104 if (ret < 0) {
2105 meta_level = -1;
2106 mlog_errno(ret);
2107 goto out;
2108 }
2109
2110 /* Clear suid / sgid if necessary. We do this here
2111 * instead of later in the write path because
2112 * remove_suid() calls ->setattr without any hint that
2113 * we may have already done our cluster locking. Since
2114 * ocfs2_setattr() *must* take cluster locks to
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002115 * proceed, this will lead us to recursively lock the
Mark Fashehccd979b2005-12-15 14:31:24 -08002116 * inode. There's also the dinode i_size state which
2117 * can be lost via setattr during extending writes (we
2118 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07002119 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002120 if (meta_level == 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07002121 ocfs2_inode_unlock(inode, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002122 meta_level = 1;
2123 continue;
2124 }
2125
2126 ret = ocfs2_write_remove_suid(inode);
2127 if (ret < 0) {
2128 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07002129 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08002130 }
2131 }
2132
2133 /* work on a copy of ppos until we're sure that we won't have
2134 * to recalculate it due to relocking. */
Tao Ma468eedd2011-02-22 22:14:41 +08002135 if (appending)
Mark Fashehccd979b2005-12-15 14:31:24 -08002136 saved_pos = i_size_read(inode);
Tao Ma468eedd2011-02-22 22:14:41 +08002137 else
Tiger Yang8659ac22006-10-17 18:29:52 -07002138 saved_pos = *ppos;
Mark Fasheh3a0782d2007-01-17 12:53:31 -08002139
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002140 end = saved_pos + count;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002141
Tao Ma293b2f72009-08-25 08:02:48 +08002142 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2143 if (ret == 1) {
2144 ocfs2_inode_unlock(inode, meta_level);
2145 meta_level = -1;
2146
2147 ret = ocfs2_prepare_inode_for_refcount(inode,
Tao Mab8908232010-08-12 10:27:14 +08002148 file,
Tao Ma293b2f72009-08-25 08:02:48 +08002149 saved_pos,
2150 count,
2151 &meta_level);
Tao Ma86470e92009-12-03 21:55:05 +08002152 if (has_refcount)
2153 *has_refcount = 1;
Wengang Wang96a1cc72010-02-09 14:57:45 +08002154 if (direct_io)
2155 *direct_io = 0;
Tao Ma293b2f72009-08-25 08:02:48 +08002156 }
2157
2158 if (ret < 0) {
2159 mlog_errno(ret);
2160 goto out_unlock;
2161 }
2162
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002163 /*
2164 * Skip the O_DIRECT checks if we don't need
2165 * them.
2166 */
2167 if (!direct_io || !(*direct_io))
2168 break;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002169
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002170 /*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07002171 * There's no sane way to do direct writes to an inode
2172 * with inline data.
2173 */
2174 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2175 *direct_io = 0;
2176 break;
2177 }
2178
2179 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002180 * Allowing concurrent direct writes means
2181 * i_size changes wouldn't be synchronized, so
2182 * one node could wind up truncating another
2183 * nodes writes.
2184 */
2185 if (end > i_size_read(inode)) {
2186 *direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002187 break;
2188 }
2189
Mark Fasheh3a0782d2007-01-17 12:53:31 -08002190 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002191 * We don't fill holes during direct io, so
2192 * check for them here. If any are found, the
2193 * caller will have to retake some cluster
2194 * locks and initiate the io as buffered.
Mark Fasheh3a0782d2007-01-17 12:53:31 -08002195 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002196 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2197 if (ret == 1) {
2198 *direct_io = 0;
2199 ret = 0;
2200 } else if (ret < 0)
2201 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08002202 break;
2203 }
2204
Tiger Yang8659ac22006-10-17 18:29:52 -07002205 if (appending)
2206 *ppos = saved_pos;
2207
2208out_unlock:
Tao Ma468eedd2011-02-22 22:14:41 +08002209 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2210 saved_pos, appending, count,
2211 direct_io, has_refcount);
2212
Tao Ma293b2f72009-08-25 08:02:48 +08002213 if (meta_level >= 0)
2214 ocfs2_inode_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002215
2216out:
2217 return ret;
2218}
2219
2220static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2221 const struct iovec *iov,
2222 unsigned long nr_segs,
2223 loff_t pos)
2224{
Mark Fasheh9517bac2007-02-09 20:24:12 -08002225 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
Tao Ma86470e92009-12-03 21:55:05 +08002226 int can_do_direct, has_refcount = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002227 ssize_t written = 0;
2228 size_t ocount; /* original count */
2229 size_t count; /* after file limit checks */
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002230 loff_t old_size, *ppos = &iocb->ki_pos;
2231 u32 old_clusters;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002232 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -05002233 struct inode *inode = file_inode(file);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002234 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002235 int full_coherency = !(osb->s_mount_opt &
2236 OCFS2_MOUNT_COHERENCY_BUFFERED);
Mark Fasheha11f7e62011-06-22 14:23:38 -07002237 int unaligned_dio = 0;
Tiger Yang8659ac22006-10-17 18:29:52 -07002238
Tao Ma468eedd2011-02-22 22:14:41 +08002239 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2241 file->f_path.dentry->d_name.len,
2242 file->f_path.dentry->d_name.name,
2243 (unsigned int)nr_segs);
Tiger Yang8659ac22006-10-17 18:29:52 -07002244
Tiger Yang8659ac22006-10-17 18:29:52 -07002245 if (iocb->ki_left == 0)
2246 return 0;
2247
Mark Fasheh9517bac2007-02-09 20:24:12 -08002248 appending = file->f_flags & O_APPEND ? 1 : 0;
2249 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2250
Tiger Yang8659ac22006-10-17 18:29:52 -07002251 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002252
Tristan Ye39c99f12010-12-07 14:35:07 +08002253 ocfs2_iocb_clear_sem_locked(iocb);
2254
Mark Fasheh9517bac2007-02-09 20:24:12 -08002255relock:
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -04002256 /* to match setattr's i_mutex -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08002257 if (direct_io) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08002258 have_alloc_sem = 1;
Tristan Ye39c99f12010-12-07 14:35:07 +08002259 /* communicate with ocfs2_dio_end_io */
2260 ocfs2_iocb_set_sem_locked(iocb);
Tiger Yang8659ac22006-10-17 18:29:52 -07002261 }
2262
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002263 /*
2264 * Concurrent O_DIRECT writes are allowed with
2265 * mount_option "coherency=buffered".
2266 */
2267 rw_level = (!direct_io || full_coherency);
2268
Tiger Yang8659ac22006-10-17 18:29:52 -07002269 ret = ocfs2_rw_lock(inode, rw_level);
2270 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08002271 mlog_errno(ret);
2272 goto out_sems;
2273 }
2274
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002275 /*
2276 * O_DIRECT writes with "coherency=full" need to take EX cluster
2277 * inode_lock to guarantee coherency.
2278 */
2279 if (direct_io && full_coherency) {
2280 /*
2281 * We need to take and drop the inode lock to force
2282 * other nodes to drop their caches. Buffered I/O
2283 * already does this in write_begin().
2284 */
2285 ret = ocfs2_inode_lock(inode, NULL, 1);
2286 if (ret < 0) {
2287 mlog_errno(ret);
Joseph Qiafe1bb72013-05-24 15:55:12 -07002288 goto out;
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002289 }
2290
2291 ocfs2_inode_unlock(inode, 1);
2292 }
2293
Mark Fasheh9517bac2007-02-09 20:24:12 -08002294 can_do_direct = direct_io;
Tao Mab8908232010-08-12 10:27:14 +08002295 ret = ocfs2_prepare_inode_for_write(file, ppos,
Mark Fasheh9517bac2007-02-09 20:24:12 -08002296 iocb->ki_left, appending,
Tao Ma86470e92009-12-03 21:55:05 +08002297 &can_do_direct, &has_refcount);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002298 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07002299 mlog_errno(ret);
2300 goto out;
2301 }
2302
Mark Fasheha11f7e62011-06-22 14:23:38 -07002303 if (direct_io && !is_sync_kiocb(iocb))
2304 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_left,
2305 *ppos);
2306
Mark Fasheh9517bac2007-02-09 20:24:12 -08002307 /*
2308 * We can't complete the direct I/O as requested, fall back to
2309 * buffered I/O.
2310 */
2311 if (direct_io && !can_do_direct) {
2312 ocfs2_rw_unlock(inode, rw_level);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002313
2314 have_alloc_sem = 0;
2315 rw_level = -1;
2316
2317 direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002318 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07002319 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002320
Mark Fasheha11f7e62011-06-22 14:23:38 -07002321 if (unaligned_dio) {
2322 /*
2323 * Wait on previous unaligned aio to complete before
2324 * proceeding.
2325 */
2326 ocfs2_aiodio_wait(inode);
2327
2328 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2329 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2330 ocfs2_iocb_set_unaligned_aio(iocb);
2331 }
2332
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002333 /*
2334 * To later detect whether a journal commit for sync writes is
2335 * necessary, we sample i_size, and cluster count here.
2336 */
2337 old_size = i_size_read(inode);
2338 old_clusters = OCFS2_I(inode)->ip_clusters;
2339
Mark Fashehccd979b2005-12-15 14:31:24 -08002340 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002341 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002342
Li Dongyang6b933c82010-04-17 17:49:10 +08002343 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2344 VERIFY_READ);
2345 if (ret)
2346 goto out_dio;
2347
2348 count = ocount;
2349 ret = generic_write_checks(file, ppos, &count,
2350 S_ISBLK(inode->i_mode));
2351 if (ret)
2352 goto out_dio;
2353
Mark Fasheh9517bac2007-02-09 20:24:12 -08002354 if (direct_io) {
2355 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2356 ppos, count, ocount);
2357 if (written < 0) {
2358 ret = written;
2359 goto out_dio;
2360 }
2361 } else {
Li Dongyang6b933c82010-04-17 17:49:10 +08002362 current->backing_dev_info = file->f_mapping->backing_dev_info;
2363 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2364 ppos, count, 0);
2365 current->backing_dev_info = NULL;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002366 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002367
Mark Fasheh9517bac2007-02-09 20:24:12 -08002368out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08002369 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08002370 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08002371
Tao Ma60c48672010-02-03 09:56:04 +08002372 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
Tristan Ye81c8c822010-08-19 15:15:00 +08002373 ((file->f_flags & O_DIRECT) && !direct_io)) {
Jan Kara918941a2009-08-17 18:50:08 +02002374 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2375 pos + count - 1);
2376 if (ret < 0)
2377 written = ret;
2378
Coly Lia03ab782010-03-26 05:15:12 +08002379 if (!ret && ((old_size != i_size_read(inode)) ||
2380 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2381 has_refcount)) {
Joel Becker2b4e30f2008-09-03 20:03:41 -07002382 ret = jbd2_journal_force_commit(osb->journal->j_journal);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002383 if (ret < 0)
2384 written = ret;
2385 }
Jan Kara918941a2009-08-17 18:50:08 +02002386
2387 if (!ret)
2388 ret = filemap_fdatawait_range(file->f_mapping, pos,
2389 pos + count - 1);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002390 }
2391
Sunil Mushran2bd63212010-01-25 16:57:38 -08002392 /*
Mark Fashehccd979b2005-12-15 14:31:24 -08002393 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2394 * function pointer which is called when o_direct io completes so that
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -04002395 * it can unlock our rw lock.
Mark Fashehccd979b2005-12-15 14:31:24 -08002396 * Unfortunately there are error cases which call end_io and others
2397 * that don't. so we don't have to unlock the rw_lock if either an
2398 * async dio is going to do it in the future or an end_io after an
2399 * error has already done it.
2400 */
Coly Li66b116c2010-02-25 14:57:13 +08002401 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002402 rw_level = -1;
2403 have_alloc_sem = 0;
Mark Fasheha11f7e62011-06-22 14:23:38 -07002404 unaligned_dio = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002405 }
2406
Junxiao Bi3e5d3c32012-06-27 17:09:55 +08002407 if (unaligned_dio) {
2408 ocfs2_iocb_clear_unaligned_aio(iocb);
Mark Fasheha11f7e62011-06-22 14:23:38 -07002409 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
Junxiao Bi3e5d3c32012-06-27 17:09:55 +08002410 }
Mark Fasheha11f7e62011-06-22 14:23:38 -07002411
Mark Fashehccd979b2005-12-15 14:31:24 -08002412out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08002413 if (rw_level != -1)
2414 ocfs2_rw_unlock(inode, rw_level);
2415
2416out_sems:
Christoph Hellwigdf2d6f22011-06-24 14:29:46 -04002417 if (have_alloc_sem)
Tristan Ye39c99f12010-12-07 14:35:07 +08002418 ocfs2_iocb_clear_sem_locked(iocb);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002419
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002420 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002421
Wengang Wang812e7a62009-07-10 13:26:04 +08002422 if (written)
2423 ret = written;
Wengang Wang812e7a62009-07-10 13:26:04 +08002424 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002425}
2426
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002427static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2428 struct file *out,
2429 struct splice_desc *sd)
2430{
2431 int ret;
2432
Tao Mab8908232010-08-12 10:27:14 +08002433 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
Tao Ma86470e92009-12-03 21:55:05 +08002434 sd->total_len, 0, NULL, NULL);
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002435 if (ret < 0) {
2436 mlog_errno(ret);
2437 return ret;
2438 }
2439
2440 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2441}
2442
Tiger Yang8659ac22006-10-17 18:29:52 -07002443static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2444 struct file *out,
2445 loff_t *ppos,
2446 size_t len,
2447 unsigned int flags)
2448{
2449 int ret;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002450 struct address_space *mapping = out->f_mapping;
2451 struct inode *inode = mapping->host;
2452 struct splice_desc sd = {
2453 .total_len = len,
2454 .flags = flags,
2455 .pos = *ppos,
2456 .u.file = out,
2457 };
Tiger Yang8659ac22006-10-17 18:29:52 -07002458
Tao Ma468eedd2011-02-22 22:14:41 +08002459
2460 trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2461 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2462 out->f_path.dentry->d_name.len,
2463 out->f_path.dentry->d_name.name, len);
Tiger Yang8659ac22006-10-17 18:29:52 -07002464
Al Viro72b0d9a2013-03-21 02:32:24 -04002465 pipe_lock(pipe);
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002466
2467 splice_from_pipe_begin(&sd);
2468 do {
2469 ret = splice_from_pipe_next(pipe, &sd);
2470 if (ret <= 0)
2471 break;
2472
2473 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2474 ret = ocfs2_rw_lock(inode, 1);
2475 if (ret < 0)
2476 mlog_errno(ret);
2477 else {
2478 ret = ocfs2_splice_to_file(pipe, out, &sd);
2479 ocfs2_rw_unlock(inode, 1);
2480 }
2481 mutex_unlock(&inode->i_mutex);
2482 } while (ret > 0);
2483 splice_from_pipe_end(pipe, &sd);
2484
Al Viro72b0d9a2013-03-21 02:32:24 -04002485 pipe_unlock(pipe);
Tiger Yang8659ac22006-10-17 18:29:52 -07002486
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002487 if (sd.num_spliced)
2488 ret = sd.num_spliced;
2489
2490 if (ret > 0) {
Jan Karad23c9372009-08-18 18:24:31 +02002491 int err;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002492
Jan Karad23c9372009-08-18 18:24:31 +02002493 err = generic_write_sync(out, *ppos, ret);
2494 if (err)
2495 ret = err;
2496 else
2497 *ppos += ret;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002498
Namjae Jeond0e1d662012-12-11 16:00:21 -08002499 balance_dirty_pages_ratelimited(mapping);
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002500 }
Tiger Yang8659ac22006-10-17 18:29:52 -07002501
Tiger Yang8659ac22006-10-17 18:29:52 -07002502 return ret;
2503}
2504
2505static ssize_t ocfs2_file_splice_read(struct file *in,
2506 loff_t *ppos,
2507 struct pipe_inode_info *pipe,
2508 size_t len,
2509 unsigned int flags)
2510{
Tao Ma1962f392009-06-19 15:36:52 +08002511 int ret = 0, lock_level = 0;
Al Viro496ad9a2013-01-23 17:07:38 -05002512 struct inode *inode = file_inode(in);
Tiger Yang8659ac22006-10-17 18:29:52 -07002513
Tao Ma468eedd2011-02-22 22:14:41 +08002514 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2515 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2516 in->f_path.dentry->d_name.len,
2517 in->f_path.dentry->d_name.name, len);
Tiger Yang8659ac22006-10-17 18:29:52 -07002518
2519 /*
2520 * See the comment in ocfs2_file_aio_read()
2521 */
Al Viro182be682013-01-24 02:21:54 -05002522 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002523 if (ret < 0) {
2524 mlog_errno(ret);
2525 goto bail;
2526 }
Tao Ma1962f392009-06-19 15:36:52 +08002527 ocfs2_inode_unlock(inode, lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002528
2529 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2530
2531bail:
Tiger Yang8659ac22006-10-17 18:29:52 -07002532 return ret;
2533}
2534
Mark Fashehccd979b2005-12-15 14:31:24 -08002535static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07002536 const struct iovec *iov,
2537 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08002538 loff_t pos)
2539{
Tiger Yang25899de2006-11-15 15:49:02 +08002540 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002541 struct file *filp = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -05002542 struct inode *inode = file_inode(filp);
Mark Fashehccd979b2005-12-15 14:31:24 -08002543
Tao Ma468eedd2011-02-22 22:14:41 +08002544 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2545 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2546 filp->f_path.dentry->d_name.len,
2547 filp->f_path.dentry->d_name.name, nr_segs);
2548
Mark Fashehccd979b2005-12-15 14:31:24 -08002549
2550 if (!inode) {
2551 ret = -EINVAL;
2552 mlog_errno(ret);
2553 goto bail;
2554 }
2555
Tristan Ye39c99f12010-12-07 14:35:07 +08002556 ocfs2_iocb_clear_sem_locked(iocb);
2557
Sunil Mushran2bd63212010-01-25 16:57:38 -08002558 /*
Mark Fashehccd979b2005-12-15 14:31:24 -08002559 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2560 * need locks to protect pending reads from racing with truncate.
2561 */
2562 if (filp->f_flags & O_DIRECT) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002563 have_alloc_sem = 1;
Tristan Ye39c99f12010-12-07 14:35:07 +08002564 ocfs2_iocb_set_sem_locked(iocb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002565
2566 ret = ocfs2_rw_lock(inode, 0);
2567 if (ret < 0) {
2568 mlog_errno(ret);
2569 goto bail;
2570 }
2571 rw_level = 0;
2572 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002573 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002574 }
2575
Mark Fashehc4374f82006-05-05 19:04:35 -07002576 /*
2577 * We're fine letting folks race truncates and extending
2578 * writes with read across the cluster, just like they can
2579 * locally. Hence no rw_lock during read.
Sunil Mushran2bd63212010-01-25 16:57:38 -08002580 *
Mark Fashehc4374f82006-05-05 19:04:35 -07002581 * Take and drop the meta data lock to update inode fields
2582 * like i_size. This allows the checks down below
Sunil Mushran2bd63212010-01-25 16:57:38 -08002583 * generic_file_aio_read() a chance of actually working.
Mark Fashehc4374f82006-05-05 19:04:35 -07002584 */
Al Viro182be682013-01-24 02:21:54 -05002585 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002586 if (ret < 0) {
2587 mlog_errno(ret);
2588 goto bail;
2589 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07002590 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002591
Badari Pulavarty027445c2006-09-30 23:28:46 -07002592 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Tao Ma468eedd2011-02-22 22:14:41 +08002593 trace_generic_file_aio_read_ret(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08002594
2595 /* buffered aio wouldn't have proper lock coverage today */
2596 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2597
2598 /* see ocfs2_file_aio_write */
2599 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2600 rw_level = -1;
2601 have_alloc_sem = 0;
2602 }
2603
2604bail:
Christoph Hellwigdf2d6f22011-06-24 14:29:46 -04002605 if (have_alloc_sem)
Tristan Ye39c99f12010-12-07 14:35:07 +08002606 ocfs2_iocb_clear_sem_locked(iocb);
Christoph Hellwigdf2d6f22011-06-24 14:29:46 -04002607
Sunil Mushran2bd63212010-01-25 16:57:38 -08002608 if (rw_level != -1)
Mark Fashehccd979b2005-12-15 14:31:24 -08002609 ocfs2_rw_unlock(inode, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002610
2611 return ret;
2612}
2613
Sunil Mushran93862d52011-07-25 14:58:15 -07002614/* Refer generic_file_llseek_unlocked() */
Andrew Morton965c8e52012-12-17 15:59:39 -08002615static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
Sunil Mushran93862d52011-07-25 14:58:15 -07002616{
2617 struct inode *inode = file->f_mapping->host;
2618 int ret = 0;
2619
2620 mutex_lock(&inode->i_mutex);
2621
Andrew Morton965c8e52012-12-17 15:59:39 -08002622 switch (whence) {
Sunil Mushran93862d52011-07-25 14:58:15 -07002623 case SEEK_SET:
2624 break;
2625 case SEEK_END:
2626 offset += inode->i_size;
2627 break;
2628 case SEEK_CUR:
2629 if (offset == 0) {
2630 offset = file->f_pos;
2631 goto out;
2632 }
2633 offset += file->f_pos;
2634 break;
2635 case SEEK_DATA:
2636 case SEEK_HOLE:
Andrew Morton965c8e52012-12-17 15:59:39 -08002637 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
Sunil Mushran93862d52011-07-25 14:58:15 -07002638 if (ret)
2639 goto out;
2640 break;
2641 default:
2642 ret = -EINVAL;
2643 goto out;
2644 }
2645
Jie Liu46a1c2c72013-06-25 12:02:13 +08002646 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sunil Mushran93862d52011-07-25 14:58:15 -07002647
2648out:
2649 mutex_unlock(&inode->i_mutex);
2650 if (ret)
2651 return ret;
2652 return offset;
2653}
2654
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002655const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002656 .setattr = ocfs2_setattr,
2657 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002658 .permission = ocfs2_permission,
Tiger Yangcf1d6c72008-08-18 17:11:00 +08002659 .setxattr = generic_setxattr,
2660 .getxattr = generic_getxattr,
2661 .listxattr = ocfs2_listxattr,
2662 .removexattr = generic_removexattr,
Mark Fasheh00dc4172008-10-03 17:32:11 -04002663 .fiemap = ocfs2_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02002664 .get_acl = ocfs2_iop_get_acl,
Mark Fashehccd979b2005-12-15 14:31:24 -08002665};
2666
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002667const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002668 .setattr = ocfs2_setattr,
2669 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002670 .permission = ocfs2_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02002671 .get_acl = ocfs2_iop_get_acl,
Mark Fashehccd979b2005-12-15 14:31:24 -08002672};
2673
Mark Fasheh53da4932008-07-21 14:29:16 -07002674/*
2675 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2676 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2677 */
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002678const struct file_operations ocfs2_fops = {
Sunil Mushran93862d52011-07-25 14:58:15 -07002679 .llseek = ocfs2_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002680 .read = do_sync_read,
2681 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002682 .mmap = ocfs2_mmap,
2683 .fsync = ocfs2_sync_file,
2684 .release = ocfs2_file_release,
2685 .open = ocfs2_file_open,
2686 .aio_read = ocfs2_file_aio_read,
2687 .aio_write = ocfs2_file_aio_write,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002688 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002689#ifdef CONFIG_COMPAT
2690 .compat_ioctl = ocfs2_compat_ioctl,
2691#endif
Mark Fasheh53da4932008-07-21 14:29:16 -07002692 .lock = ocfs2_lock,
2693 .flock = ocfs2_flock,
2694 .splice_read = ocfs2_file_splice_read,
2695 .splice_write = ocfs2_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002696 .fallocate = ocfs2_fallocate,
Mark Fasheh53da4932008-07-21 14:29:16 -07002697};
2698
2699const struct file_operations ocfs2_dops = {
2700 .llseek = generic_file_llseek,
2701 .read = generic_read_dir,
Al Viro3704412b2013-05-22 21:06:00 -04002702 .iterate = ocfs2_readdir,
Mark Fasheh53da4932008-07-21 14:29:16 -07002703 .fsync = ocfs2_sync_file,
2704 .release = ocfs2_dir_release,
2705 .open = ocfs2_dir_open,
2706 .unlocked_ioctl = ocfs2_ioctl,
2707#ifdef CONFIG_COMPAT
2708 .compat_ioctl = ocfs2_compat_ioctl,
2709#endif
2710 .lock = ocfs2_lock,
2711 .flock = ocfs2_flock,
2712};
2713
2714/*
2715 * POSIX-lockless variants of our file_operations.
2716 *
2717 * These will be used if the underlying cluster stack does not support
2718 * posix file locking, if the user passes the "localflocks" mount
2719 * option, or if we have a local-only fs.
2720 *
2721 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2722 * so we still want it in the case of no stack support for
2723 * plocks. Internally, it will do the right thing when asked to ignore
2724 * the cluster.
2725 */
2726const struct file_operations ocfs2_fops_no_plocks = {
Sunil Mushran93862d52011-07-25 14:58:15 -07002727 .llseek = ocfs2_file_llseek,
Mark Fasheh53da4932008-07-21 14:29:16 -07002728 .read = do_sync_read,
2729 .write = do_sync_write,
2730 .mmap = ocfs2_mmap,
2731 .fsync = ocfs2_sync_file,
2732 .release = ocfs2_file_release,
2733 .open = ocfs2_file_open,
2734 .aio_read = ocfs2_file_aio_read,
2735 .aio_write = ocfs2_file_aio_write,
2736 .unlocked_ioctl = ocfs2_ioctl,
2737#ifdef CONFIG_COMPAT
2738 .compat_ioctl = ocfs2_compat_ioctl,
2739#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002740 .flock = ocfs2_flock,
Tiger Yang8659ac22006-10-17 18:29:52 -07002741 .splice_read = ocfs2_file_splice_read,
2742 .splice_write = ocfs2_file_splice_write,
Tristan Ye3d1c1822011-05-23 15:57:26 +08002743 .fallocate = ocfs2_fallocate,
Mark Fashehccd979b2005-12-15 14:31:24 -08002744};
2745
Mark Fasheh53da4932008-07-21 14:29:16 -07002746const struct file_operations ocfs2_dops_no_plocks = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002747 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002748 .read = generic_read_dir,
Al Viro3704412b2013-05-22 21:06:00 -04002749 .iterate = ocfs2_readdir,
Mark Fashehccd979b2005-12-15 14:31:24 -08002750 .fsync = ocfs2_sync_file,
Mark Fasheh53fc6222007-12-20 16:49:04 -08002751 .release = ocfs2_dir_release,
2752 .open = ocfs2_dir_open,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002753 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002754#ifdef CONFIG_COMPAT
2755 .compat_ioctl = ocfs2_compat_ioctl,
2756#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002757 .flock = ocfs2_flock,
Mark Fashehccd979b2005-12-15 14:31:24 -08002758};