blob: 4e7b0dc22450564cd86064061d7d84e8461f2bbf [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>
Tejun Heo66114ca2015-05-22 17:13:32 -040040#include <linux/backing-dev.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080041
Mark Fashehccd979b2005-12-15 14:31:24 -080042#include <cluster/masklog.h>
43
44#include "ocfs2.h"
45
46#include "alloc.h"
47#include "aops.h"
48#include "dir.h"
49#include "dlmglue.h"
50#include "extent_map.h"
51#include "file.h"
52#include "sysfile.h"
53#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070054#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080055#include "journal.h"
Mark Fasheh53fc6222007-12-20 16:49:04 -080056#include "locks.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080057#include "mmap.h"
58#include "suballoc.h"
59#include "super.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080060#include "xattr.h"
Tiger Yang23fc2702008-11-14 11:17:18 +080061#include "acl.h"
Jan Karaa90714c2008-10-09 19:38:40 +020062#include "quota.h"
Tao Ma293b2f72009-08-25 08:02:48 +080063#include "refcounttree.h"
Tao Ma468eedd2011-02-22 22:14:41 +080064#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080065
66#include "buffer_head_io.h"
67
Mark Fasheh53fc6222007-12-20 16:49:04 -080068static int ocfs2_init_file_private(struct inode *inode, struct file *file)
69{
70 struct ocfs2_file_private *fp;
71
72 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
73 if (!fp)
74 return -ENOMEM;
75
76 fp->fp_file = file;
77 mutex_init(&fp->fp_mutex);
78 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
79 file->private_data = fp;
80
81 return 0;
82}
83
84static void ocfs2_free_file_private(struct inode *inode, struct file *file)
85{
86 struct ocfs2_file_private *fp = file->private_data;
87 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
88
89 if (fp) {
90 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
91 ocfs2_lock_res_free(&fp->fp_flock);
92 kfree(fp);
93 file->private_data = NULL;
94 }
95}
96
Mark Fashehccd979b2005-12-15 14:31:24 -080097static int ocfs2_file_open(struct inode *inode, struct file *file)
98{
99 int status;
100 int mode = file->f_flags;
101 struct ocfs2_inode_info *oi = OCFS2_I(inode);
102
Tao Ma468eedd2011-02-22 22:14:41 +0800103 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
104 (unsigned long long)OCFS2_I(inode)->ip_blkno,
105 file->f_path.dentry->d_name.len,
106 file->f_path.dentry->d_name.name, mode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800107
Jan Kara9c89fe02015-07-14 13:36:02 +0200108 if (file->f_mode & FMODE_WRITE) {
109 status = dquot_initialize(inode);
110 if (status)
111 goto leave;
112 }
Christoph Hellwig907f4552010-03-03 09:05:06 -0500113
Mark Fashehccd979b2005-12-15 14:31:24 -0800114 spin_lock(&oi->ip_lock);
115
116 /* Check that the inode hasn't been wiped from disk by another
117 * node. If it hasn't then we're safe as long as we hold the
118 * spin lock until our increment of open count. */
119 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
120 spin_unlock(&oi->ip_lock);
121
122 status = -ENOENT;
123 goto leave;
124 }
125
126 if (mode & O_DIRECT)
127 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
128
129 oi->ip_open_count++;
130 spin_unlock(&oi->ip_lock);
Mark Fasheh53fc6222007-12-20 16:49:04 -0800131
132 status = ocfs2_init_file_private(inode, file);
133 if (status) {
134 /*
135 * We want to set open count back if we're failing the
136 * open.
137 */
138 spin_lock(&oi->ip_lock);
139 oi->ip_open_count--;
140 spin_unlock(&oi->ip_lock);
141 }
142
Mark Fashehccd979b2005-12-15 14:31:24 -0800143leave:
Mark Fashehccd979b2005-12-15 14:31:24 -0800144 return status;
145}
146
147static int ocfs2_file_release(struct inode *inode, struct file *file)
148{
149 struct ocfs2_inode_info *oi = OCFS2_I(inode);
150
Mark Fashehccd979b2005-12-15 14:31:24 -0800151 spin_lock(&oi->ip_lock);
152 if (!--oi->ip_open_count)
153 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
Tao Ma468eedd2011-02-22 22:14:41 +0800154
155 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
156 oi->ip_blkno,
157 file->f_path.dentry->d_name.len,
158 file->f_path.dentry->d_name.name,
159 oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -0800160 spin_unlock(&oi->ip_lock);
161
Mark Fasheh53fc6222007-12-20 16:49:04 -0800162 ocfs2_free_file_private(inode, file);
163
Mark Fashehccd979b2005-12-15 14:31:24 -0800164 return 0;
165}
166
Mark Fasheh53fc6222007-12-20 16:49:04 -0800167static int ocfs2_dir_open(struct inode *inode, struct file *file)
168{
169 return ocfs2_init_file_private(inode, file);
170}
171
172static int ocfs2_dir_release(struct inode *inode, struct file *file)
173{
174 ocfs2_free_file_private(inode, file);
175 return 0;
176}
177
Josef Bacik02c24a82011-07-16 20:44:56 -0400178static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
179 int datasync)
Mark Fashehccd979b2005-12-15 14:31:24 -0800180{
181 int err = 0;
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200182 struct inode *inode = file->f_mapping->host;
Mark Fashehccd979b2005-12-15 14:31:24 -0800183 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Darrick J. Wong2931cdc2014-04-03 14:46:48 -0700184 struct ocfs2_inode_info *oi = OCFS2_I(inode);
185 journal_t *journal = osb->journal->j_journal;
186 int ret;
187 tid_t commit_tid;
188 bool needs_barrier = false;
Mark Fashehccd979b2005-12-15 14:31:24 -0800189
Tao Ma468eedd2011-02-22 22:14:41 +0800190 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
191 OCFS2_I(inode)->ip_blkno,
192 file->f_path.dentry->d_name.len,
193 file->f_path.dentry->d_name.name,
194 (unsigned long long)datasync);
Mark Fashehccd979b2005-12-15 14:31:24 -0800195
Younger Liua987c7c2014-02-10 14:25:44 -0800196 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
197 return -EROFS;
198
Josef Bacik02c24a82011-07-16 20:44:56 -0400199 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
200 if (err)
201 return err;
202
Darrick J. Wong2931cdc2014-04-03 14:46:48 -0700203 commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
204 if (journal->j_flags & JBD2_BARRIER &&
205 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
206 needs_barrier = true;
207 err = jbd2_complete_transaction(journal, commit_tid);
208 if (needs_barrier) {
209 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
210 if (!err)
211 err = ret;
Jan Kara04eda1a2010-08-05 20:32:45 +0200212 }
Hisashi Hifumie04cc152009-06-09 16:47:45 +0900213
Tao Mac1e8d352011-03-07 16:43:21 +0800214 if (err)
215 mlog_errno(err);
Mark Fashehccd979b2005-12-15 14:31:24 -0800216
217 return (err < 0) ? -EIO : 0;
218}
219
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800220int ocfs2_should_update_atime(struct inode *inode,
221 struct vfsmount *vfsmnt)
222{
223 struct timespec now;
224 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
225
226 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
227 return 0;
228
229 if ((inode->i_flags & S_NOATIME) ||
230 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
231 return 0;
232
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800233 /*
234 * We can be called with no vfsmnt structure - NFSD will
235 * sometimes do this.
236 *
237 * Note that our action here is different than touch_atime() -
238 * if we can't tell whether this is a noatime mount, then we
239 * don't know whether to trust the value of s_atime_quantum.
240 */
241 if (vfsmnt == NULL)
242 return 0;
243
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800244 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
245 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
246 return 0;
247
Mark Fasheh7e913c52006-12-13 00:34:35 -0800248 if (vfsmnt->mnt_flags & MNT_RELATIME) {
249 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
250 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
251 return 1;
252
253 return 0;
254 }
255
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800256 now = CURRENT_TIME;
257 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
258 return 0;
259 else
260 return 1;
261}
262
263int ocfs2_update_inode_atime(struct inode *inode,
264 struct buffer_head *bh)
265{
266 int ret;
267 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
268 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700269 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800270
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800271 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200272 if (IS_ERR(handle)) {
273 ret = PTR_ERR(handle);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800274 mlog_errno(ret);
275 goto out;
276 }
277
Joel Becker0cf2f762009-02-12 16:41:25 -0800278 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -0700279 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700280 if (ret) {
281 mlog_errno(ret);
282 goto out_commit;
283 }
284
285 /*
286 * Don't use ocfs2_mark_inode_dirty() here as we don't always
287 * have i_mutex to guard against concurrent changes to other
288 * inode fields.
289 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800290 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700291 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
292 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -0700293 ocfs2_update_inode_fsync_trans(handle, inode, 0);
Joel Beckerec20cec2010-03-19 14:13:52 -0700294 ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800295
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700296out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800297 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
298out:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800299 return ret;
300}
301
Joseph Qi026749a2015-02-16 15:59:50 -0800302int ocfs2_set_inode_size(handle_t *handle,
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700303 struct inode *inode,
304 struct buffer_head *fe_bh,
305 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800306{
307 int status;
308
Mark Fashehccd979b2005-12-15 14:31:24 -0800309 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700310 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800311 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
312
313 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
314 if (status < 0) {
315 mlog_errno(status);
316 goto bail;
317 }
318
319bail:
Mark Fashehccd979b2005-12-15 14:31:24 -0800320 return status;
321}
322
Jan Kara9e33d692008-08-25 19:56:50 +0200323int ocfs2_simple_size_update(struct inode *inode,
324 struct buffer_head *di_bh,
325 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800326{
327 int ret;
328 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700329 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800330
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700331 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200332 if (IS_ERR(handle)) {
333 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800334 mlog_errno(ret);
335 goto out;
336 }
337
338 ret = ocfs2_set_inode_size(handle, inode, di_bh,
339 new_i_size);
340 if (ret < 0)
341 mlog_errno(ret);
342
Darrick J. Wong6fdb7022014-04-03 14:47:08 -0700343 ocfs2_update_inode_fsync_trans(handle, inode, 0);
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700344 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800345out:
346 return ret;
347}
348
Tao Ma37f8a2b2009-08-26 09:47:28 +0800349static int ocfs2_cow_file_pos(struct inode *inode,
350 struct buffer_head *fe_bh,
351 u64 offset)
352{
353 int status;
354 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
355 unsigned int num_clusters = 0;
356 unsigned int ext_flags = 0;
357
358 /*
359 * If the new offset is aligned to the range of the cluster, there is
360 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
361 * CoW either.
362 */
363 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
364 return 0;
365
366 status = ocfs2_get_clusters(inode, cpos, &phys,
367 &num_clusters, &ext_flags);
368 if (status) {
369 mlog_errno(status);
370 goto out;
371 }
372
373 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
374 goto out;
375
Tiger Yangc7dd3392013-08-13 16:00:58 -0700376 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
Tao Ma37f8a2b2009-08-26 09:47:28 +0800377
378out:
379 return status;
380}
381
Mark Fashehccd979b2005-12-15 14:31:24 -0800382static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
383 struct inode *inode,
384 struct buffer_head *fe_bh,
385 u64 new_i_size)
386{
387 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700388 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800389 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700390 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800391
Tao Ma37f8a2b2009-08-26 09:47:28 +0800392 /*
393 * We need to CoW the cluster contains the offset if it is reflinked
394 * since we will call ocfs2_zero_range_for_truncate later which will
395 * write "0" from offset to the end of the cluster.
396 */
397 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
398 if (status) {
399 mlog_errno(status);
400 return status;
401 }
402
Mark Fashehccd979b2005-12-15 14:31:24 -0800403 /* TODO: This needs to actually orphan the inode in this
404 * transaction. */
405
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700406 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800407 if (IS_ERR(handle)) {
408 status = PTR_ERR(handle);
409 mlog_errno(status);
410 goto out;
411 }
412
Joel Becker0cf2f762009-02-12 16:41:25 -0800413 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700414 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800415 if (status < 0) {
416 mlog_errno(status);
417 goto out_commit;
418 }
419
420 /*
421 * Do this before setting i_size.
422 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700423 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
424 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
425 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800426 if (status) {
427 mlog_errno(status);
428 goto out_commit;
429 }
430
431 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800432 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
433
434 di = (struct ocfs2_dinode *) fe_bh->b_data;
435 di->i_size = cpu_to_le64(new_i_size);
436 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
437 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -0700438 ocfs2_update_inode_fsync_trans(handle, inode, 0);
Mark Fasheh60b11392007-02-16 11:46:50 -0800439
Joel Beckerec20cec2010-03-19 14:13:52 -0700440 ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800441
Mark Fasheh60b11392007-02-16 11:46:50 -0800442out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700443 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800444out:
Mark Fashehccd979b2005-12-15 14:31:24 -0800445 return status;
446}
447
Joseph Qi026749a2015-02-16 15:59:50 -0800448int ocfs2_truncate_file(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -0800449 struct buffer_head *di_bh,
450 u64 new_i_size)
451{
452 int status = 0;
453 struct ocfs2_dinode *fe = NULL;
454 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800455
Joel Beckerb657c952008-11-13 14:49:11 -0800456 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
457 * already validated it */
Mark Fashehccd979b2005-12-15 14:31:24 -0800458 fe = (struct ocfs2_dinode *) di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800459
Tao Ma468eedd2011-02-22 22:14:41 +0800460 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
461 (unsigned long long)le64_to_cpu(fe->i_size),
462 (unsigned long long)new_i_size);
463
Mark Fashehccd979b2005-12-15 14:31:24 -0800464 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800465 "Inode %llu, inode i_size = %lld != di "
466 "i_size = %llu, i_flags = 0x%x\n",
467 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800468 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800469 (unsigned long long)le64_to_cpu(fe->i_size),
470 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800471
472 if (new_i_size > le64_to_cpu(fe->i_size)) {
Tao Ma468eedd2011-02-22 22:14:41 +0800473 trace_ocfs2_truncate_file_error(
474 (unsigned long long)le64_to_cpu(fe->i_size),
475 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800476 status = -EINVAL;
477 mlog_errno(status);
478 goto bail;
479 }
480
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700481 down_write(&OCFS2_I(inode)->ip_alloc_sem);
482
Mark Fasheh4fe370a2009-12-07 13:15:40 -0800483 ocfs2_resv_discard(&osb->osb_la_resmap,
484 &OCFS2_I(inode)->ip_la_data_resv);
485
Mark Fashehc934a922007-10-18 15:23:46 -0700486 /*
487 * The inode lock forced other nodes to sync and drop their
488 * pages, which (correctly) happens even if we have a truncate
489 * without allocation change - ocfs2 cluster sizes can be much
490 * greater than page size, so we have to truncate them
491 * anyway.
492 */
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700493 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
494 truncate_inode_pages(inode->i_mapping, new_i_size);
495
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700496 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
497 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
Mark Fashehb1967d02007-11-20 11:56:39 -0800498 i_size_read(inode), 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700499 if (status)
500 mlog_errno(status);
501
Mark Fashehc934a922007-10-18 15:23:46 -0700502 goto bail_unlock_sem;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700503 }
504
Mark Fashehccd979b2005-12-15 14:31:24 -0800505 /* alright, we're going to need to do a full blown alloc size
506 * change. Orphan the inode so that recovery can complete the
507 * truncate if necessary. This does the task of marking
508 * i_size. */
509 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
510 if (status < 0) {
511 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700512 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800513 }
514
Tristan Ye78f94672010-05-11 17:54:42 +0800515 status = ocfs2_commit_truncate(osb, inode, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800516 if (status < 0) {
517 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700518 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800519 }
520
521 /* TODO: orphan dir cleanup here. */
Mark Fashehc934a922007-10-18 15:23:46 -0700522bail_unlock_sem:
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700523 up_write(&OCFS2_I(inode)->ip_alloc_sem);
524
Mark Fashehccd979b2005-12-15 14:31:24 -0800525bail:
Tao Ma8b2c0db2009-08-18 11:43:49 +0800526 if (!status && OCFS2_I(inode)->ip_clusters == 0)
527 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800528
Mark Fashehccd979b2005-12-15 14:31:24 -0800529 return status;
530}
531
532/*
Tao Ma0eb8d472008-08-18 17:38:45 +0800533 * extend file allocation only here.
Mark Fashehccd979b2005-12-15 14:31:24 -0800534 * we'll update all the disk stuff, and oip->alloc_size
535 *
536 * expect stuff to be locked, a transaction started and enough data /
537 * metadata reservations in the contexts.
538 *
539 * Will return -EAGAIN, and a reason if a restart is needed.
540 * If passed in, *reason will always be set, even in error.
541 */
Tao Ma0eb8d472008-08-18 17:38:45 +0800542int ocfs2_add_inode_data(struct ocfs2_super *osb,
543 struct inode *inode,
544 u32 *logical_offset,
545 u32 clusters_to_add,
546 int mark_unwritten,
547 struct buffer_head *fe_bh,
548 handle_t *handle,
549 struct ocfs2_alloc_context *data_ac,
550 struct ocfs2_alloc_context *meta_ac,
551 enum ocfs2_alloc_restarted *reason_ret)
Mark Fashehccd979b2005-12-15 14:31:24 -0800552{
Joel Beckerf99b9b72008-08-20 19:36:33 -0700553 int ret;
554 struct ocfs2_extent_tree et;
Mark Fashehccd979b2005-12-15 14:31:24 -0800555
Joel Becker5e404e92009-02-13 03:54:22 -0800556 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
Joel Beckercbee7e12009-02-13 03:34:15 -0800557 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
558 clusters_to_add, mark_unwritten,
559 data_ac, meta_ac, reason_ret);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700560
561 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800562}
563
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800564static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
565 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800566{
567 int status = 0;
568 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800569 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800570 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800571 struct buffer_head *bh = NULL;
572 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700573 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800574 struct ocfs2_alloc_context *data_ac = NULL;
575 struct ocfs2_alloc_context *meta_ac = NULL;
Junxiao Bi696cdf72015-02-10 14:08:46 -0800576 enum ocfs2_alloc_restarted why = RESTART_NONE;
Mark Fashehccd979b2005-12-15 14:31:24 -0800577 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700578 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +0200579 int did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800580
Mark Fashehdcd05382007-01-16 11:32:23 -0800581 /*
Junxiao Bif0cb0f02013-11-12 15:06:53 -0800582 * Unwritten extent only exists for file systems which
Mark Fashehdcd05382007-01-16 11:32:23 -0800583 * support holes.
584 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800585 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800586
Joel Beckerb657c952008-11-13 14:49:11 -0800587 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800588 if (status < 0) {
589 mlog_errno(status);
590 goto leave;
591 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800592 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800593
594restart_all:
595 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
596
Joel Becker5e404e92009-02-13 03:54:22 -0800597 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700598 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
599 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800600 if (status) {
601 mlog_errno(status);
602 goto leave;
603 }
604
Goldwyn Rodrigues06f9da62013-11-12 15:06:52 -0800605 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700606 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800607 if (IS_ERR(handle)) {
608 status = PTR_ERR(handle);
609 handle = NULL;
610 mlog_errno(status);
611 goto leave;
612 }
613
614restarted_transaction:
Tao Ma468eedd2011-02-22 22:14:41 +0800615 trace_ocfs2_extend_allocation(
616 (unsigned long long)OCFS2_I(inode)->ip_blkno,
617 (unsigned long long)i_size_read(inode),
618 le32_to_cpu(fe->i_clusters), clusters_to_add,
619 why, restart_func);
620
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500621 status = dquot_alloc_space_nodirty(inode,
622 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
623 if (status)
Jan Karaa90714c2008-10-09 19:38:40 +0200624 goto leave;
Jan Karaa90714c2008-10-09 19:38:40 +0200625 did_quota = 1;
626
Mark Fashehccd979b2005-12-15 14:31:24 -0800627 /* reserve a write to the file entry early on - that we if we
628 * run out of credits in the allocation path, we can still
629 * update i_size. */
Joel Becker0cf2f762009-02-12 16:41:25 -0800630 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -0700631 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800632 if (status < 0) {
633 mlog_errno(status);
634 goto leave;
635 }
636
637 prev_clusters = OCFS2_I(inode)->ip_clusters;
638
Tao Ma0eb8d472008-08-18 17:38:45 +0800639 status = ocfs2_add_inode_data(osb,
640 inode,
641 &logical_start,
642 clusters_to_add,
643 mark_unwritten,
644 bh,
645 handle,
646 data_ac,
647 meta_ac,
648 &why);
Mark Fashehccd979b2005-12-15 14:31:24 -0800649 if ((status < 0) && (status != -EAGAIN)) {
650 if (status != -ENOSPC)
651 mlog_errno(status);
652 goto leave;
653 }
Darrick J. Wong2931cdc2014-04-03 14:46:48 -0700654 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Joel Beckerec20cec2010-03-19 14:13:52 -0700655 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800656
657 spin_lock(&OCFS2_I(inode)->ip_lock);
658 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
659 spin_unlock(&OCFS2_I(inode)->ip_lock);
Jan Karaa90714c2008-10-09 19:38:40 +0200660 /* Release unused quota reservation */
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500661 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +0200662 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
663 did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800664
665 if (why != RESTART_NONE && clusters_to_add) {
666 if (why == RESTART_META) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800667 restart_func = 1;
Tao Ma79681842010-04-16 13:59:25 +0800668 status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800669 } else {
670 BUG_ON(why != RESTART_TRANS);
671
Younger Liu2b1e55c2013-09-11 14:19:44 -0700672 status = ocfs2_allocate_extend_trans(handle, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800673 if (status < 0) {
674 /* handle still has to be committed at
675 * this point. */
676 status = -ENOMEM;
677 mlog_errno(status);
678 goto leave;
679 }
680 goto restarted_transaction;
681 }
682 }
683
Tao Ma468eedd2011-02-22 22:14:41 +0800684 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700685 le32_to_cpu(fe->i_clusters),
Tao Ma468eedd2011-02-22 22:14:41 +0800686 (unsigned long long)le64_to_cpu(fe->i_size),
687 OCFS2_I(inode)->ip_clusters,
688 (unsigned long long)i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800689
690leave:
Jan Karaa90714c2008-10-09 19:38:40 +0200691 if (status < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500692 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +0200693 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
Mark Fashehccd979b2005-12-15 14:31:24 -0800694 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700695 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800696 handle = NULL;
697 }
698 if (data_ac) {
699 ocfs2_free_alloc_context(data_ac);
700 data_ac = NULL;
701 }
702 if (meta_ac) {
703 ocfs2_free_alloc_context(meta_ac);
704 meta_ac = NULL;
705 }
706 if ((!status) && restart_func) {
707 restart_func = 0;
708 goto restart_all;
709 }
Mark Fasheha81cb882008-10-07 14:25:16 -0700710 brelse(bh);
711 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800712
Mark Fashehccd979b2005-12-15 14:31:24 -0800713 return status;
714}
715
Joseph Qi026749a2015-02-16 15:59:50 -0800716int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
717 u32 clusters_to_add, int mark_unwritten)
718{
719 return __ocfs2_extend_allocation(inode, logical_start,
720 clusters_to_add, mark_unwritten);
721}
722
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700723/*
724 * While a write will already be ordering the data, a truncate will not.
725 * Thus, we need to explicitly order the zeroed pages.
726 */
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800727static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
728 struct buffer_head *di_bh)
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700729{
730 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
731 handle_t *handle = NULL;
732 int ret = 0;
733
734 if (!ocfs2_should_order_data(inode))
735 goto out;
736
737 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
738 if (IS_ERR(handle)) {
739 ret = -ENOMEM;
740 mlog_errno(ret);
741 goto out;
742 }
743
744 ret = ocfs2_jbd2_file_inode(handle, inode);
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800745 if (ret < 0) {
746 mlog_errno(ret);
747 goto out;
748 }
749
750 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
751 OCFS2_JOURNAL_ACCESS_WRITE);
752 if (ret)
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700753 mlog_errno(ret);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -0700754 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700755
756out:
757 if (ret) {
758 if (!IS_ERR(handle))
759 ocfs2_commit_trans(osb, handle);
760 handle = ERR_PTR(ret);
761 }
762 return handle;
763}
764
Mark Fashehccd979b2005-12-15 14:31:24 -0800765/* Some parts of this taken from generic_cont_expand, which turned out
766 * to be too fragile to do exactly what we need without us having to
Nick Piggin4e02ed42008-10-29 14:00:55 -0700767 * worry about recursive locking in ->write_begin() and ->write_end(). */
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700768static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800769 u64 abs_to, struct buffer_head *di_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -0800770{
771 struct address_space *mapping = inode->i_mapping;
772 struct page *page;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300773 unsigned long index = abs_from >> PAGE_SHIFT;
Junxiao Bif775da22014-10-09 15:25:15 -0700774 handle_t *handle;
Joel Becker54532582010-07-16 13:32:33 -0700775 int ret = 0;
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700776 unsigned zero_from, zero_to, block_start, block_end;
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800777 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800778
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700779 BUG_ON(abs_from >= abs_to);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300780 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700781 BUG_ON(abs_from & (inode->i_blkbits - 1));
Mark Fashehccd979b2005-12-15 14:31:24 -0800782
Junxiao Bif775da22014-10-09 15:25:15 -0700783 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
784 if (IS_ERR(handle)) {
785 ret = PTR_ERR(handle);
786 goto out;
787 }
788
Jan Kara9b4c0ff2010-08-24 14:28:03 +0200789 page = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800790 if (!page) {
791 ret = -ENOMEM;
792 mlog_errno(ret);
Junxiao Bif775da22014-10-09 15:25:15 -0700793 goto out_commit_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -0800794 }
795
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700796 /* Get the offsets within the page that we want to zero */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300797 zero_from = abs_from & (PAGE_SIZE - 1);
798 zero_to = abs_to & (PAGE_SIZE - 1);
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700799 if (!zero_to)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300800 zero_to = PAGE_SIZE;
Mark Fashehccd979b2005-12-15 14:31:24 -0800801
Tao Ma468eedd2011-02-22 22:14:41 +0800802 trace_ocfs2_write_zero_page(
803 (unsigned long long)OCFS2_I(inode)->ip_blkno,
804 (unsigned long long)abs_from,
805 (unsigned long long)abs_to,
806 index, zero_from, zero_to);
Joel Becker56934862010-07-01 15:13:31 -0700807
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700808 /* We know that zero_from is block aligned */
809 for (block_start = zero_from; block_start < zero_to;
810 block_start = block_end) {
811 block_end = block_start + (1 << inode->i_blkbits);
812
813 /*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200814 * block_start is block-aligned. Bump it by one to force
815 * __block_write_begin and block_commit_write to zero the
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700816 * whole block.
817 */
Christoph Hellwigebdec242010-10-06 10:47:23 +0200818 ret = __block_write_begin(page, block_start + 1, 0,
819 ocfs2_get_block);
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700820 if (ret < 0) {
821 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800822 goto out_unlock;
823 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800824
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700825
826 /* must not update i_size! */
827 ret = block_commit_write(page, block_start + 1,
828 block_start + 1);
829 if (ret < 0)
830 mlog_errno(ret);
831 else
832 ret = 0;
833 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800834
Junxiao Bif775da22014-10-09 15:25:15 -0700835 /*
836 * fs-writeback will release the dirty pages without page lock
837 * whose offset are over inode size, the release happens at
838 * block_write_full_page().
839 */
840 i_size_write(inode, abs_to);
841 inode->i_blocks = ocfs2_inode_sector_count(inode);
842 di->i_size = cpu_to_le64((u64)i_size_read(inode));
843 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
844 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
845 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
846 di->i_mtime_nsec = di->i_ctime_nsec;
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800847 if (handle) {
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800848 ocfs2_journal_dirty(handle, di_bh);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -0700849 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800850 }
Joel Beckera4bfb4c2010-07-06 14:36:06 -0700851
Mark Fashehccd979b2005-12-15 14:31:24 -0800852out_unlock:
853 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300854 put_page(page);
Junxiao Bif775da22014-10-09 15:25:15 -0700855out_commit_trans:
856 if (handle)
857 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800858out:
859 return ret;
860}
861
Joel Becker56934862010-07-01 15:13:31 -0700862/*
863 * Find the next range to zero. We do this in terms of bytes because
864 * that's what ocfs2_zero_extend() wants, and it is dealing with the
865 * pagecache. We may return multiple extents.
866 *
867 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
868 * needs to be zeroed. range_start and range_end return the next zeroing
869 * range. A subsequent call should pass the previous range_end as its
870 * zero_start. If range_end is 0, there's nothing to do.
871 *
872 * Unwritten extents are skipped over. Refcounted extents are CoWd.
873 */
874static int ocfs2_zero_extend_get_range(struct inode *inode,
875 struct buffer_head *di_bh,
876 u64 zero_start, u64 zero_end,
877 u64 *range_start, u64 *range_end)
Mark Fashehccd979b2005-12-15 14:31:24 -0800878{
Joel Becker56934862010-07-01 15:13:31 -0700879 int rc = 0, needs_cow = 0;
880 u32 p_cpos, zero_clusters = 0;
881 u32 zero_cpos =
882 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
883 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
884 unsigned int num_clusters = 0;
885 unsigned int ext_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800886
Joel Becker56934862010-07-01 15:13:31 -0700887 while (zero_cpos < last_cpos) {
888 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
889 &num_clusters, &ext_flags);
890 if (rc) {
891 mlog_errno(rc);
Mark Fashehccd979b2005-12-15 14:31:24 -0800892 goto out;
893 }
894
Joel Becker56934862010-07-01 15:13:31 -0700895 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
896 zero_clusters = num_clusters;
897 if (ext_flags & OCFS2_EXT_REFCOUNTED)
898 needs_cow = 1;
899 break;
900 }
901
902 zero_cpos += num_clusters;
903 }
904 if (!zero_clusters) {
905 *range_end = 0;
906 goto out;
907 }
908
909 while ((zero_cpos + zero_clusters) < last_cpos) {
910 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
911 &p_cpos, &num_clusters,
912 &ext_flags);
913 if (rc) {
914 mlog_errno(rc);
915 goto out;
916 }
917
918 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
919 break;
920 if (ext_flags & OCFS2_EXT_REFCOUNTED)
921 needs_cow = 1;
922 zero_clusters += num_clusters;
923 }
924 if ((zero_cpos + zero_clusters) > last_cpos)
925 zero_clusters = last_cpos - zero_cpos;
926
927 if (needs_cow) {
Tiger Yangc7dd3392013-08-13 16:00:58 -0700928 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
Tao Ma15502712010-08-12 10:36:38 +0800929 zero_clusters, UINT_MAX);
Joel Becker56934862010-07-01 15:13:31 -0700930 if (rc) {
931 mlog_errno(rc);
932 goto out;
933 }
934 }
935
936 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
937 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
938 zero_cpos + zero_clusters);
939
940out:
941 return rc;
942}
943
944/*
945 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
946 * has made sure that the entire range needs zeroing.
947 */
948static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800949 u64 range_end, struct buffer_head *di_bh)
Joel Becker56934862010-07-01 15:13:31 -0700950{
951 int rc = 0;
952 u64 next_pos;
953 u64 zero_pos = range_start;
954
Tao Ma468eedd2011-02-22 22:14:41 +0800955 trace_ocfs2_zero_extend_range(
956 (unsigned long long)OCFS2_I(inode)->ip_blkno,
957 (unsigned long long)range_start,
958 (unsigned long long)range_end);
Joel Becker56934862010-07-01 15:13:31 -0700959 BUG_ON(range_start >= range_end);
960
961 while (zero_pos < range_end) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300962 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
Joel Becker56934862010-07-01 15:13:31 -0700963 if (next_pos > range_end)
964 next_pos = range_end;
Junxiao Bic7d2cbc2014-02-10 14:25:53 -0800965 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
Joel Becker56934862010-07-01 15:13:31 -0700966 if (rc < 0) {
967 mlog_errno(rc);
968 break;
969 }
970 zero_pos = next_pos;
Mark Fashehe2057c52006-10-03 17:53:05 -0700971
972 /*
973 * Very large extends have the potential to lock up
974 * the cpu for extended periods of time.
975 */
976 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800977 }
978
Joel Becker56934862010-07-01 15:13:31 -0700979 return rc;
980}
981
982int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
983 loff_t zero_to_size)
984{
985 int ret = 0;
986 u64 zero_start, range_start = 0, range_end = 0;
987 struct super_block *sb = inode->i_sb;
988
989 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
Tao Ma468eedd2011-02-22 22:14:41 +0800990 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
991 (unsigned long long)zero_start,
992 (unsigned long long)i_size_read(inode));
Joel Becker56934862010-07-01 15:13:31 -0700993 while (zero_start < zero_to_size) {
994 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
995 zero_to_size,
996 &range_start,
997 &range_end);
998 if (ret) {
999 mlog_errno(ret);
1000 break;
1001 }
1002 if (!range_end)
1003 break;
1004 /* Trim the ends */
1005 if (range_start < zero_start)
1006 range_start = zero_start;
1007 if (range_end > zero_to_size)
1008 range_end = zero_to_size;
1009
1010 ret = ocfs2_zero_extend_range(inode, range_start,
Junxiao Bic7d2cbc2014-02-10 14:25:53 -08001011 range_end, di_bh);
Joel Becker56934862010-07-01 15:13:31 -07001012 if (ret) {
1013 mlog_errno(ret);
1014 break;
1015 }
1016 zero_start = range_end;
1017 }
1018
Mark Fashehccd979b2005-12-15 14:31:24 -08001019 return ret;
1020}
1021
Joel Becker56934862010-07-01 15:13:31 -07001022int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1023 u64 new_i_size, u64 zero_to)
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001024{
1025 int ret;
1026 u32 clusters_to_add;
1027 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1028
Joel Becker56934862010-07-01 15:13:31 -07001029 /*
1030 * Only quota files call this without a bh, and they can't be
1031 * refcounted.
1032 */
1033 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1034 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1035
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001036 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1037 if (clusters_to_add < oi->ip_clusters)
1038 clusters_to_add = 0;
1039 else
1040 clusters_to_add -= oi->ip_clusters;
1041
1042 if (clusters_to_add) {
1043 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1044 clusters_to_add, 0);
1045 if (ret) {
1046 mlog_errno(ret);
1047 goto out;
1048 }
1049 }
1050
1051 /*
1052 * Call this even if we don't add any clusters to the tree. We
1053 * still need to zero the area between the old i_size and the
1054 * new i_size.
1055 */
Joel Becker56934862010-07-01 15:13:31 -07001056 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001057 if (ret < 0)
1058 mlog_errno(ret);
1059
1060out:
1061 return ret;
1062}
1063
Mark Fashehccd979b2005-12-15 14:31:24 -08001064static int ocfs2_extend_file(struct inode *inode,
1065 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001066 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -08001067{
Mark Fashehc934a922007-10-18 15:23:46 -07001068 int ret = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001069 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001070
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001071 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -07001072
Mark Fashehccd979b2005-12-15 14:31:24 -08001073 /* setattr sometimes calls us like this. */
1074 if (new_i_size == 0)
1075 goto out;
1076
1077 if (i_size_read(inode) == new_i_size)
Joel Becker56934862010-07-01 15:13:31 -07001078 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001079 BUG_ON(new_i_size < i_size_read(inode));
1080
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001081 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001082 * The alloc sem blocks people in read/write from reading our
1083 * allocation until we're done changing it. We depend on
1084 * i_mutex to block other extend/truncate calls while we're
Joel Becker56934862010-07-01 15:13:31 -07001085 * here. We even have to hold it for sparse files because there
1086 * might be some tail zeroing.
Mark Fasheh0effef72006-10-03 17:44:42 -07001087 */
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001088 down_write(&oi->ip_alloc_sem);
1089
1090 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1091 /*
1092 * We can optimize small extends by keeping the inodes
1093 * inline data.
1094 */
1095 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1096 up_write(&oi->ip_alloc_sem);
1097 goto out_update_size;
1098 }
1099
1100 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1101 if (ret) {
1102 up_write(&oi->ip_alloc_sem);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001103 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001104 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001105 }
1106 }
1107
Joel Becker56934862010-07-01 15:13:31 -07001108 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1109 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1110 else
1111 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1112 new_i_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001113
1114 up_write(&oi->ip_alloc_sem);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001115
Mark Fasheh0effef72006-10-03 17:44:42 -07001116 if (ret < 0) {
1117 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001118 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001119 }
1120
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001121out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001122 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1123 if (ret < 0)
1124 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -07001125
Mark Fashehccd979b2005-12-15 14:31:24 -08001126out:
1127 return ret;
1128}
1129
1130int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1131{
1132 int status = 0, size_change;
Tariq Saeed3d46a442015-09-04 15:44:31 -07001133 int inode_locked = 0;
David Howells2b0143b2015-03-17 22:25:59 +00001134 struct inode *inode = d_inode(dentry);
Mark Fashehccd979b2005-12-15 14:31:24 -08001135 struct super_block *sb = inode->i_sb;
1136 struct ocfs2_super *osb = OCFS2_SB(sb);
1137 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001138 handle_t *handle = NULL;
Jan Kara65bac572009-06-02 14:24:01 +02001139 struct dquot *transfer_to[MAXQUOTAS] = { };
Jan Kara52a9ee22010-05-13 20:18:45 +02001140 int qtype;
Mark Fashehccd979b2005-12-15 14:31:24 -08001141
Tao Ma468eedd2011-02-22 22:14:41 +08001142 trace_ocfs2_setattr(inode, dentry,
1143 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1144 dentry->d_name.len, dentry->d_name.name,
1145 attr->ia_valid, attr->ia_mode,
Eric W. Biedermanba613562013-01-31 17:33:53 -08001146 from_kuid(&init_user_ns, attr->ia_uid),
1147 from_kgid(&init_user_ns, attr->ia_gid));
Mark Fashehccd979b2005-12-15 14:31:24 -08001148
Sunil Mushranbc535802008-04-18 10:23:53 -07001149 /* ensuring we don't even attempt to truncate a symlink */
1150 if (S_ISLNK(inode->i_mode))
1151 attr->ia_valid &= ~ATTR_SIZE;
1152
Mark Fashehccd979b2005-12-15 14:31:24 -08001153#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1154 | ATTR_GID | ATTR_UID | ATTR_MODE)
Tao Ma468eedd2011-02-22 22:14:41 +08001155 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
Mark Fashehccd979b2005-12-15 14:31:24 -08001156 return 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001157
1158 status = inode_change_ok(inode, attr);
1159 if (status)
1160 return status;
1161
Jan Kara9c89fe02015-07-14 13:36:02 +02001162 if (is_quota_modification(inode, attr)) {
1163 status = dquot_initialize(inode);
1164 if (status)
1165 return status;
1166 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001167 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1168 if (size_change) {
1169 status = ocfs2_rw_lock(inode, 1);
1170 if (status < 0) {
1171 mlog_errno(status);
1172 goto bail;
1173 }
1174 }
1175
Mark Fashehe63aecb62007-10-18 15:30:42 -07001176 status = ocfs2_inode_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001177 if (status < 0) {
1178 if (status != -ENOENT)
1179 mlog_errno(status);
1180 goto bail_unlock_rw;
1181 }
Tariq Saeed3d46a442015-09-04 15:44:31 -07001182 inode_locked = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08001183
Younger Liud62e74b2014-02-10 14:25:51 -08001184 if (size_change) {
Wengang Wang5051f762010-02-26 18:18:25 +08001185 status = inode_newsize_ok(inode, attr->ia_size);
1186 if (status)
Mark Fashehce76fd32007-07-20 12:02:14 -07001187 goto bail_unlock;
Mark Fashehce76fd32007-07-20 12:02:14 -07001188
Christoph Hellwig562c72a2011-06-24 14:29:45 -04001189 inode_dio_wait(inode);
1190
Younger Liud62e74b2014-02-10 14:25:51 -08001191 if (i_size_read(inode) >= attr->ia_size) {
Joel Becker2b4e30f2008-09-03 20:03:41 -07001192 if (ocfs2_should_order_data(inode)) {
1193 status = ocfs2_begin_ordered_truncate(inode,
1194 attr->ia_size);
1195 if (status)
1196 goto bail_unlock;
1197 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001198 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
Joel Becker2b4e30f2008-09-03 20:03:41 -07001199 } else
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001200 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001201 if (status < 0) {
1202 if (status != -ENOSPC)
1203 mlog_errno(status);
1204 status = -ENOSPC;
1205 goto bail_unlock;
1206 }
1207 }
1208
Eric W. Biederman488c8ef2013-01-31 17:35:29 -08001209 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1210 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Jan Kara65bac572009-06-02 14:24:01 +02001211 /*
1212 * Gather pointers to quota structures so that allocation /
1213 * freeing of quota structures happens here and not inside
Christoph Hellwigb43fa822010-03-03 09:05:03 -05001214 * dquot_transfer() where we have problems with lock ordering
Jan Kara65bac572009-06-02 14:24:01 +02001215 */
Eric W. Biederman488c8ef2013-01-31 17:35:29 -08001216 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
Jan Karaa90714c2008-10-09 19:38:40 +02001217 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1218 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
Eric W. Biedermanaca645a2012-09-16 03:11:50 -07001219 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
Jan Kara6184fc02015-06-24 18:07:02 +02001220 if (IS_ERR(transfer_to[USRQUOTA])) {
1221 status = PTR_ERR(transfer_to[USRQUOTA]);
Jan Karaa90714c2008-10-09 19:38:40 +02001222 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001223 }
Jan Karaa90714c2008-10-09 19:38:40 +02001224 }
Eric W. Biederman488c8ef2013-01-31 17:35:29 -08001225 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
Jan Karaa90714c2008-10-09 19:38:40 +02001226 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1227 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
Eric W. Biedermanaca645a2012-09-16 03:11:50 -07001228 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
Jan Kara6184fc02015-06-24 18:07:02 +02001229 if (IS_ERR(transfer_to[GRPQUOTA])) {
1230 status = PTR_ERR(transfer_to[GRPQUOTA]);
Jan Karaa90714c2008-10-09 19:38:40 +02001231 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001232 }
Jan Karaa90714c2008-10-09 19:38:40 +02001233 }
Jan Kara65bac572009-06-02 14:24:01 +02001234 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1235 2 * ocfs2_quota_trans_credits(sb));
Jan Karaa90714c2008-10-09 19:38:40 +02001236 if (IS_ERR(handle)) {
1237 status = PTR_ERR(handle);
1238 mlog_errno(status);
1239 goto bail_unlock;
1240 }
Jan Kara52a9ee22010-05-13 20:18:45 +02001241 status = __dquot_transfer(inode, transfer_to);
Jan Karaa90714c2008-10-09 19:38:40 +02001242 if (status < 0)
1243 goto bail_commit;
1244 } else {
1245 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1246 if (IS_ERR(handle)) {
1247 status = PTR_ERR(handle);
1248 mlog_errno(status);
1249 goto bail_unlock;
1250 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001251 }
1252
Christoph Hellwig10257742010-06-04 11:30:02 +02001253 setattr_copy(inode, attr);
1254 mark_inode_dirty(inode);
1255
Mark Fashehccd979b2005-12-15 14:31:24 -08001256 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1257 if (status < 0)
1258 mlog_errno(status);
1259
1260bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001261 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001262bail_unlock:
Tariq Saeed3d46a442015-09-04 15:44:31 -07001263 if (status) {
1264 ocfs2_inode_unlock(inode, 1);
1265 inode_locked = 0;
1266 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001267bail_unlock_rw:
1268 if (size_change)
1269 ocfs2_rw_unlock(inode, 1);
1270bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001271
Jan Kara65bac572009-06-02 14:24:01 +02001272 /* Release quota pointers in case we acquired them */
Jan Kara52362812014-09-10 21:06:39 +02001273 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
Jan Kara65bac572009-06-02 14:24:01 +02001274 dqput(transfer_to[qtype]);
Jan Kara65bac572009-06-02 14:24:01 +02001275
Tiger Yang060bc662008-11-14 11:17:29 +08001276 if (!status && attr->ia_valid & ATTR_MODE) {
Junxiao Bi5ee0fbd2016-05-12 15:42:15 -07001277 status = ocfs2_acl_chmod(inode, bh);
Tiger Yang060bc662008-11-14 11:17:29 +08001278 if (status < 0)
1279 mlog_errno(status);
1280 }
Tariq Saeed3d46a442015-09-04 15:44:31 -07001281 if (inode_locked)
1282 ocfs2_inode_unlock(inode, 1);
Tiger Yang060bc662008-11-14 11:17:29 +08001283
Junxiao Bi5ee0fbd2016-05-12 15:42:15 -07001284 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001285 return status;
1286}
1287
1288int ocfs2_getattr(struct vfsmount *mnt,
1289 struct dentry *dentry,
1290 struct kstat *stat)
1291{
David Howells2b0143b2015-03-17 22:25:59 +00001292 struct inode *inode = d_inode(dentry);
Al Virofc640052016-04-10 01:33:30 -04001293 struct super_block *sb = dentry->d_sb;
Mark Fashehccd979b2005-12-15 14:31:24 -08001294 struct ocfs2_super *osb = sb->s_fs_info;
1295 int err;
1296
Mark Fashehccd979b2005-12-15 14:31:24 -08001297 err = ocfs2_inode_revalidate(dentry);
1298 if (err) {
1299 if (err != -ENOENT)
1300 mlog_errno(err);
1301 goto bail;
1302 }
1303
1304 generic_fillattr(inode, stat);
John Haxbyd6364622016-01-14 15:17:12 -08001305 /*
1306 * If there is inline data in the inode, the inode will normally not
1307 * have data blocks allocated (it may have an external xattr block).
1308 * Report at least one sector for such files, so tools like tar, rsync,
1309 * others don't incorrectly think the file is completely sparse.
1310 */
1311 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1312 stat->blocks += (stat->size + 511)>>9;
Mark Fashehccd979b2005-12-15 14:31:24 -08001313
1314 /* We set the blksize from the cluster size for performance */
1315 stat->blksize = osb->s_clustersize;
1316
1317bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001318 return err;
1319}
1320
Al Viro10556cb2011-06-20 19:28:19 -04001321int ocfs2_permission(struct inode *inode, int mask)
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001322{
1323 int ret;
1324
Al Viro10556cb2011-06-20 19:28:19 -04001325 if (mask & MAY_NOT_BLOCK)
Nick Pigginb74c79e2011-01-07 17:49:58 +11001326 return -ECHILD;
1327
Mark Fashehe63aecb62007-10-18 15:30:42 -07001328 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001329 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001330 if (ret != -ENOENT)
1331 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001332 goto out;
1333 }
1334
Al Viro2830ba72011-06-20 19:16:29 -04001335 ret = generic_permission(inode, mask);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001336
Mark Fashehe63aecb62007-10-18 15:30:42 -07001337 ocfs2_inode_unlock(inode, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001338out:
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001339 return ret;
1340}
1341
Mark Fashehb2580102007-03-09 16:53:21 -08001342static int __ocfs2_write_remove_suid(struct inode *inode,
1343 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001344{
1345 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001346 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001347 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1348 struct ocfs2_dinode *di;
1349
Tao Ma468eedd2011-02-22 22:14:41 +08001350 trace_ocfs2_write_remove_suid(
1351 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1352 inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001353
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001354 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001355 if (IS_ERR(handle)) {
1356 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001357 mlog_errno(ret);
1358 goto out;
1359 }
1360
Joel Becker0cf2f762009-02-12 16:41:25 -08001361 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -07001362 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001363 if (ret < 0) {
1364 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001365 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001366 }
1367
1368 inode->i_mode &= ~S_ISUID;
1369 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1370 inode->i_mode &= ~S_ISGID;
1371
1372 di = (struct ocfs2_dinode *) bh->b_data;
1373 di->i_mode = cpu_to_le16(inode->i_mode);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07001374 ocfs2_update_inode_fsync_trans(handle, inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001375
Joel Beckerec20cec2010-03-19 14:13:52 -07001376 ocfs2_journal_dirty(handle, bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001377
Mark Fashehccd979b2005-12-15 14:31:24 -08001378out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001379 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001380out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001381 return ret;
1382}
1383
Mark Fashehb2580102007-03-09 16:53:21 -08001384static int ocfs2_write_remove_suid(struct inode *inode)
1385{
1386 int ret;
1387 struct buffer_head *bh = NULL;
Mark Fashehb2580102007-03-09 16:53:21 -08001388
Joel Beckerb657c952008-11-13 14:49:11 -08001389 ret = ocfs2_read_inode_block(inode, &bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001390 if (ret < 0) {
1391 mlog_errno(ret);
1392 goto out;
1393 }
1394
1395 ret = __ocfs2_write_remove_suid(inode, bh);
1396out:
1397 brelse(bh);
1398 return ret;
1399}
1400
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001401/*
1402 * Allocate enough extents to cover the region starting at byte offset
1403 * start for len bytes. Existing extents are skipped, any extents
1404 * added are marked as "unwritten".
1405 */
1406static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1407 u64 start, u64 len)
1408{
1409 int ret;
1410 u32 cpos, phys_cpos, clusters, alloc_size;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001411 u64 end = start + len;
1412 struct buffer_head *di_bh = NULL;
1413
1414 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Joel Beckerb657c952008-11-13 14:49:11 -08001415 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001416 if (ret) {
1417 mlog_errno(ret);
1418 goto out;
1419 }
1420
1421 /*
1422 * Nothing to do if the requested reservation range
1423 * fits within the inode.
1424 */
1425 if (ocfs2_size_fits_inline_data(di_bh, end))
1426 goto out;
1427
1428 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1429 if (ret) {
1430 mlog_errno(ret);
1431 goto out;
1432 }
1433 }
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001434
1435 /*
1436 * We consider both start and len to be inclusive.
1437 */
1438 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1439 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1440 clusters -= cpos;
1441
1442 while (clusters) {
1443 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1444 &alloc_size, NULL);
1445 if (ret) {
1446 mlog_errno(ret);
1447 goto out;
1448 }
1449
1450 /*
1451 * Hole or existing extent len can be arbitrary, so
1452 * cap it to our own allocation request.
1453 */
1454 if (alloc_size > clusters)
1455 alloc_size = clusters;
1456
1457 if (phys_cpos) {
1458 /*
1459 * We already have an allocation at this
1460 * region so we can safely skip it.
1461 */
1462 goto next;
1463 }
1464
1465 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1466 if (ret) {
1467 if (ret != -ENOSPC)
1468 mlog_errno(ret);
1469 goto out;
1470 }
1471
1472next:
1473 cpos += alloc_size;
1474 clusters -= alloc_size;
1475 }
1476
1477 ret = 0;
1478out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001479
1480 brelse(di_bh);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001481 return ret;
1482}
1483
Mark Fasheh063c4562007-07-03 13:34:11 -07001484/*
1485 * Truncate a byte range, avoiding pages within partial clusters. This
1486 * preserves those pages for the zeroing code to write to.
1487 */
1488static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1489 u64 byte_len)
1490{
1491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1492 loff_t start, end;
1493 struct address_space *mapping = inode->i_mapping;
1494
1495 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1496 end = byte_start + byte_len;
1497 end = end & ~(osb->s_clustersize - 1);
1498
1499 if (start < end) {
1500 unmap_mapping_range(mapping, start, end - start, 0);
1501 truncate_inode_pages_range(mapping, start, end - 1);
1502 }
1503}
1504
1505static int ocfs2_zero_partial_clusters(struct inode *inode,
1506 u64 start, u64 len)
1507{
1508 int ret = 0;
1509 u64 tmpend, end = start + len;
1510 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1511 unsigned int csize = osb->s_clustersize;
1512 handle_t *handle;
1513
1514 /*
1515 * The "start" and "end" values are NOT necessarily part of
1516 * the range whose allocation is being deleted. Rather, this
1517 * is what the user passed in with the request. We must zero
1518 * partial clusters here. There's no need to worry about
1519 * physical allocation - the zeroing code knows to skip holes.
1520 */
Tao Ma468eedd2011-02-22 22:14:41 +08001521 trace_ocfs2_zero_partial_clusters(
1522 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1523 (unsigned long long)start, (unsigned long long)end);
Mark Fasheh063c4562007-07-03 13:34:11 -07001524
1525 /*
1526 * If both edges are on a cluster boundary then there's no
1527 * zeroing required as the region is part of the allocation to
1528 * be truncated.
1529 */
1530 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1531 goto out;
1532
1533 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001534 if (IS_ERR(handle)) {
1535 ret = PTR_ERR(handle);
Mark Fasheh063c4562007-07-03 13:34:11 -07001536 mlog_errno(ret);
1537 goto out;
1538 }
1539
1540 /*
1541 * We want to get the byte offset of the end of the 1st cluster.
1542 */
1543 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1544 if (tmpend > end)
1545 tmpend = end;
1546
Tao Ma468eedd2011-02-22 22:14:41 +08001547 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1548 (unsigned long long)tmpend);
Mark Fasheh063c4562007-07-03 13:34:11 -07001549
1550 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1551 if (ret)
1552 mlog_errno(ret);
1553
1554 if (tmpend < end) {
1555 /*
1556 * This may make start and end equal, but the zeroing
1557 * code will skip any work in that case so there's no
1558 * need to catch it up here.
1559 */
1560 start = end & ~(osb->s_clustersize - 1);
1561
Tao Ma468eedd2011-02-22 22:14:41 +08001562 trace_ocfs2_zero_partial_clusters_range2(
1563 (unsigned long long)start, (unsigned long long)end);
Mark Fasheh063c4562007-07-03 13:34:11 -07001564
1565 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1566 if (ret)
1567 mlog_errno(ret);
1568 }
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07001569 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh063c4562007-07-03 13:34:11 -07001570
1571 ocfs2_commit_trans(osb, handle);
1572out:
1573 return ret;
1574}
1575
Tristan Yec1631d42010-05-11 17:54:45 +08001576static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1577{
1578 int i;
1579 struct ocfs2_extent_rec *rec = NULL;
1580
1581 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1582
1583 rec = &el->l_recs[i];
1584
1585 if (le32_to_cpu(rec->e_cpos) < pos)
1586 break;
1587 }
1588
1589 return i;
1590}
1591
1592/*
1593 * Helper to calculate the punching pos and length in one run, we handle the
1594 * following three cases in order:
1595 *
1596 * - remove the entire record
1597 * - remove a partial record
1598 * - no record needs to be removed (hole-punching completed)
1599*/
1600static void ocfs2_calc_trunc_pos(struct inode *inode,
1601 struct ocfs2_extent_list *el,
1602 struct ocfs2_extent_rec *rec,
1603 u32 trunc_start, u32 *trunc_cpos,
1604 u32 *trunc_len, u32 *trunc_end,
1605 u64 *blkno, int *done)
1606{
1607 int ret = 0;
1608 u32 coff, range;
1609
1610 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1611
1612 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
Tristan Ye9a790ba2011-05-12 20:47:07 +08001613 /*
1614 * remove an entire extent record.
1615 */
Tristan Yec1631d42010-05-11 17:54:45 +08001616 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1617 /*
1618 * Skip holes if any.
1619 */
1620 if (range < *trunc_end)
1621 *trunc_end = range;
1622 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1623 *blkno = le64_to_cpu(rec->e_blkno);
1624 *trunc_end = le32_to_cpu(rec->e_cpos);
1625 } else if (range > trunc_start) {
Tristan Ye9a790ba2011-05-12 20:47:07 +08001626 /*
1627 * remove a partial extent record, which means we're
1628 * removing the last extent record.
1629 */
Tristan Yec1631d42010-05-11 17:54:45 +08001630 *trunc_cpos = trunc_start;
Tristan Ye9a790ba2011-05-12 20:47:07 +08001631 /*
1632 * skip hole if any.
1633 */
1634 if (range < *trunc_end)
1635 *trunc_end = range;
Tristan Yec1631d42010-05-11 17:54:45 +08001636 *trunc_len = *trunc_end - trunc_start;
1637 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1638 *blkno = le64_to_cpu(rec->e_blkno) +
1639 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1640 *trunc_end = trunc_start;
1641 } else {
1642 /*
1643 * It may have two following possibilities:
1644 *
1645 * - last record has been removed
1646 * - trunc_start was within a hole
1647 *
1648 * both two cases mean the completion of hole punching.
1649 */
1650 ret = 1;
1651 }
1652
1653 *done = ret;
1654}
1655
Mark Fasheh063c4562007-07-03 13:34:11 -07001656static int ocfs2_remove_inode_range(struct inode *inode,
1657 struct buffer_head *di_bh, u64 byte_start,
1658 u64 byte_len)
1659{
Tristan Yec1631d42010-05-11 17:54:45 +08001660 int ret = 0, flags = 0, done = 0, i;
1661 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1662 u32 cluster_in_el;
Mark Fasheh063c4562007-07-03 13:34:11 -07001663 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1664 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehb1967d02007-11-20 11:56:39 -08001665 struct address_space *mapping = inode->i_mapping;
Mark Fashehfecc0112008-11-12 15:16:38 -08001666 struct ocfs2_extent_tree et;
Tristan Yec1631d42010-05-11 17:54:45 +08001667 struct ocfs2_path *path = NULL;
1668 struct ocfs2_extent_list *el = NULL;
1669 struct ocfs2_extent_rec *rec = NULL;
Tristan Yee8aec062010-05-11 17:54:43 +08001670 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Tristan Yec1631d42010-05-11 17:54:45 +08001671 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
Mark Fasheh063c4562007-07-03 13:34:11 -07001672
Joel Becker5e404e92009-02-13 03:54:22 -08001673 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Mark Fasheh063c4562007-07-03 13:34:11 -07001674 ocfs2_init_dealloc_ctxt(&dealloc);
1675
Tao Ma468eedd2011-02-22 22:14:41 +08001676 trace_ocfs2_remove_inode_range(
1677 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1678 (unsigned long long)byte_start,
1679 (unsigned long long)byte_len);
1680
Mark Fasheh063c4562007-07-03 13:34:11 -07001681 if (byte_len == 0)
1682 return 0;
1683
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001684 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1685 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
Mark Fashehb1967d02007-11-20 11:56:39 -08001686 byte_start + byte_len, 0);
1687 if (ret) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001688 mlog_errno(ret);
Mark Fashehb1967d02007-11-20 11:56:39 -08001689 goto out;
1690 }
1691 /*
1692 * There's no need to get fancy with the page cache
1693 * truncate of an inline-data inode. We're talking
1694 * about less than a page here, which will be cached
1695 * in the dinode buffer anyway.
1696 */
1697 unmap_mapping_range(mapping, 0, 0, 0);
1698 truncate_inode_pages(mapping, 0);
1699 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001700 }
1701
Tristan Yee8aec062010-05-11 17:54:43 +08001702 /*
1703 * For reflinks, we may need to CoW 2 clusters which might be
1704 * partially zero'd later, if hole's start and end offset were
1705 * within one cluster(means is not exactly aligned to clustersize).
1706 */
Mark Fasheh063c4562007-07-03 13:34:11 -07001707
Tristan Yee8aec062010-05-11 17:54:43 +08001708 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1709
1710 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1711 if (ret) {
1712 mlog_errno(ret);
1713 goto out;
1714 }
1715
1716 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1717 if (ret) {
1718 mlog_errno(ret);
1719 goto out;
1720 }
1721 }
1722
Mark Fasheh063c4562007-07-03 13:34:11 -07001723 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
Tristan Yec1631d42010-05-11 17:54:45 +08001724 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1725 cluster_in_el = trunc_end;
Mark Fasheh063c4562007-07-03 13:34:11 -07001726
Mark Fasheh063c4562007-07-03 13:34:11 -07001727 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1728 if (ret) {
1729 mlog_errno(ret);
1730 goto out;
1731 }
1732
Tristan Yec1631d42010-05-11 17:54:45 +08001733 path = ocfs2_new_path_from_et(&et);
1734 if (!path) {
1735 ret = -ENOMEM;
1736 mlog_errno(ret);
1737 goto out;
1738 }
1739
1740 while (trunc_end > trunc_start) {
1741
1742 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1743 cluster_in_el);
Mark Fasheh063c4562007-07-03 13:34:11 -07001744 if (ret) {
1745 mlog_errno(ret);
1746 goto out;
1747 }
1748
Tristan Yec1631d42010-05-11 17:54:45 +08001749 el = path_leaf_el(path);
Mark Fasheh063c4562007-07-03 13:34:11 -07001750
Tristan Yec1631d42010-05-11 17:54:45 +08001751 i = ocfs2_find_rec(el, trunc_end);
1752 /*
1753 * Need to go to previous extent block.
1754 */
1755 if (i < 0) {
1756 if (path->p_tree_depth == 0)
1757 break;
1758
1759 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1760 path,
1761 &cluster_in_el);
Mark Fasheh063c4562007-07-03 13:34:11 -07001762 if (ret) {
1763 mlog_errno(ret);
1764 goto out;
1765 }
Tristan Yec1631d42010-05-11 17:54:45 +08001766
1767 /*
1768 * We've reached the leftmost extent block,
1769 * it's safe to leave.
1770 */
1771 if (cluster_in_el == 0)
1772 break;
1773
1774 /*
1775 * The 'pos' searched for previous extent block is
1776 * always one cluster less than actual trunc_end.
1777 */
1778 trunc_end = cluster_in_el + 1;
1779
1780 ocfs2_reinit_path(path, 1);
1781
1782 continue;
1783
1784 } else
1785 rec = &el->l_recs[i];
1786
1787 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1788 &trunc_len, &trunc_end, &blkno, &done);
1789 if (done)
1790 break;
1791
1792 flags = rec->e_flags;
1793 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1794
1795 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1796 phys_cpos, trunc_len, flags,
Junxiao Bif62f12b2014-12-18 16:17:32 -08001797 &dealloc, refcount_loc, false);
Tristan Yec1631d42010-05-11 17:54:45 +08001798 if (ret < 0) {
1799 mlog_errno(ret);
1800 goto out;
Mark Fasheh063c4562007-07-03 13:34:11 -07001801 }
1802
Tristan Yec1631d42010-05-11 17:54:45 +08001803 cluster_in_el = trunc_end;
1804
1805 ocfs2_reinit_path(path, 1);
Mark Fasheh063c4562007-07-03 13:34:11 -07001806 }
1807
1808 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1809
1810out:
Younger Liu7aebff182013-09-11 14:19:59 -07001811 ocfs2_free_path(path);
Mark Fasheh063c4562007-07-03 13:34:11 -07001812 ocfs2_schedule_truncate_log_flush(osb, 1);
1813 ocfs2_run_deallocs(osb, &dealloc);
1814
1815 return ret;
1816}
1817
Mark Fashehb2580102007-03-09 16:53:21 -08001818/*
1819 * Parts of this function taken from xfs_change_file_space()
1820 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001821static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1822 loff_t f_pos, unsigned int cmd,
1823 struct ocfs2_space_resv *sr,
1824 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001825{
1826 int ret;
1827 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001828 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001829 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1830 struct buffer_head *di_bh = NULL;
1831 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001832 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001833
Mark Fashehb2580102007-03-09 16:53:21 -08001834 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1835 return -EROFS;
1836
Al Viro59551022016-01-22 15:40:57 -05001837 inode_lock(inode);
Mark Fashehb2580102007-03-09 16:53:21 -08001838
1839 /*
1840 * This prevents concurrent writes on other nodes
1841 */
1842 ret = ocfs2_rw_lock(inode, 1);
1843 if (ret) {
1844 mlog_errno(ret);
1845 goto out;
1846 }
1847
Mark Fashehe63aecb62007-10-18 15:30:42 -07001848 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001849 if (ret) {
1850 mlog_errno(ret);
1851 goto out_rw_unlock;
1852 }
1853
1854 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1855 ret = -EPERM;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001856 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001857 }
1858
1859 switch (sr->l_whence) {
1860 case 0: /*SEEK_SET*/
1861 break;
1862 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001863 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001864 break;
1865 case 2: /*SEEK_END*/
1866 sr->l_start += i_size_read(inode);
1867 break;
1868 default:
1869 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001870 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001871 }
1872 sr->l_whence = 0;
1873
1874 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1875
1876 if (sr->l_start < 0
1877 || sr->l_start > max_off
1878 || (sr->l_start + llen) < 0
1879 || (sr->l_start + llen) > max_off) {
1880 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001881 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001882 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001883 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001884
Tariq Saeeda2a3b392014-01-21 15:48:38 -08001885 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1886 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
Mark Fashehb2580102007-03-09 16:53:21 -08001887 if (sr->l_len <= 0) {
1888 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001889 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001890 }
1891 }
1892
Mark Fasheh385820a2007-07-19 00:14:38 -07001893 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001894 ret = __ocfs2_write_remove_suid(inode, di_bh);
1895 if (ret) {
1896 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001897 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001898 }
1899 }
1900
1901 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1902 switch (cmd) {
1903 case OCFS2_IOC_RESVSP:
1904 case OCFS2_IOC_RESVSP64:
1905 /*
1906 * This takes unsigned offsets, but the signed ones we
1907 * pass have been checked against overflow above.
1908 */
1909 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1910 sr->l_len);
1911 break;
1912 case OCFS2_IOC_UNRESVSP:
1913 case OCFS2_IOC_UNRESVSP64:
1914 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1915 sr->l_len);
1916 break;
1917 default:
1918 ret = -EINVAL;
1919 }
1920 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1921 if (ret) {
1922 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001923 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001924 }
1925
1926 /*
1927 * We update c/mtime for these changes
1928 */
1929 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1930 if (IS_ERR(handle)) {
1931 ret = PTR_ERR(handle);
1932 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001933 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001934 }
1935
Mark Fasheh385820a2007-07-19 00:14:38 -07001936 if (change_size && i_size_read(inode) < size)
1937 i_size_write(inode, size);
1938
Mark Fashehb2580102007-03-09 16:53:21 -08001939 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1940 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1941 if (ret < 0)
1942 mlog_errno(ret);
1943
Luis Henriquesa4e08d02012-07-11 14:02:10 -07001944 if (file && (file->f_flags & O_SYNC))
Mark Fashehdf295d4a2011-11-16 12:03:10 -08001945 handle->h_sync = 1;
1946
Mark Fashehb2580102007-03-09 16:53:21 -08001947 ocfs2_commit_trans(osb, handle);
1948
Mark Fashehe63aecb62007-10-18 15:30:42 -07001949out_inode_unlock:
Mark Fashehb2580102007-03-09 16:53:21 -08001950 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001951 ocfs2_inode_unlock(inode, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001952out_rw_unlock:
1953 ocfs2_rw_unlock(inode, 1);
1954
Mark Fashehb2580102007-03-09 16:53:21 -08001955out:
Al Viro59551022016-01-22 15:40:57 -05001956 inode_unlock(inode);
Mark Fashehb2580102007-03-09 16:53:21 -08001957 return ret;
1958}
1959
Mark Fasheh385820a2007-07-19 00:14:38 -07001960int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1961 struct ocfs2_space_resv *sr)
1962{
Al Viro496ad9a2013-01-23 17:07:38 -05001963 struct inode *inode = file_inode(file);
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08001964 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Jan Karafef69252012-06-12 16:20:40 +02001965 int ret;
Mark Fasheh385820a2007-07-19 00:14:38 -07001966
1967 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1968 !ocfs2_writes_unwritten_extents(osb))
1969 return -ENOTTY;
1970 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1971 !ocfs2_sparse_alloc(osb))
1972 return -ENOTTY;
1973
1974 if (!S_ISREG(inode->i_mode))
1975 return -EINVAL;
1976
1977 if (!(file->f_mode & FMODE_WRITE))
1978 return -EBADF;
1979
Jan Karafef69252012-06-12 16:20:40 +02001980 ret = mnt_want_write_file(file);
1981 if (ret)
1982 return ret;
1983 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1984 mnt_drop_write_file(file);
1985 return ret;
Mark Fasheh385820a2007-07-19 00:14:38 -07001986}
1987
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001988static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
Mark Fasheh385820a2007-07-19 00:14:38 -07001989 loff_t len)
1990{
Al Viro496ad9a2013-01-23 17:07:38 -05001991 struct inode *inode = file_inode(file);
Mark Fasheh385820a2007-07-19 00:14:38 -07001992 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1993 struct ocfs2_space_resv sr;
1994 int change_size = 1;
Josef Bacikdb47fef2010-11-17 20:46:17 -05001995 int cmd = OCFS2_IOC_RESVSP64;
Mark Fasheh385820a2007-07-19 00:14:38 -07001996
Christoph Hellwig64c23e82011-01-14 13:07:30 +01001997 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1998 return -EOPNOTSUPP;
Mark Fasheh385820a2007-07-19 00:14:38 -07001999 if (!ocfs2_writes_unwritten_extents(osb))
2000 return -EOPNOTSUPP;
2001
Mark Fasheh385820a2007-07-19 00:14:38 -07002002 if (mode & FALLOC_FL_KEEP_SIZE)
2003 change_size = 0;
2004
Josef Bacikdb47fef2010-11-17 20:46:17 -05002005 if (mode & FALLOC_FL_PUNCH_HOLE)
2006 cmd = OCFS2_IOC_UNRESVSP64;
2007
Mark Fasheh385820a2007-07-19 00:14:38 -07002008 sr.l_whence = 0;
2009 sr.l_start = (s64)offset;
2010 sr.l_len = (s64)len;
2011
Josef Bacikdb47fef2010-11-17 20:46:17 -05002012 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2013 change_size);
Mark Fasheh385820a2007-07-19 00:14:38 -07002014}
2015
Tao Ma293b2f72009-08-25 08:02:48 +08002016int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2017 size_t count)
2018{
2019 int ret = 0;
2020 unsigned int extent_flags;
2021 u32 cpos, clusters, extent_len, phys_cpos;
2022 struct super_block *sb = inode->i_sb;
2023
2024 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
Tao Ma2f48d592009-10-15 11:10:49 +08002025 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2026 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Tao Ma293b2f72009-08-25 08:02:48 +08002027 return 0;
2028
2029 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2030 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2031
2032 while (clusters) {
2033 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2034 &extent_flags);
2035 if (ret < 0) {
2036 mlog_errno(ret);
2037 goto out;
2038 }
2039
2040 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2041 ret = 1;
2042 break;
2043 }
2044
2045 if (extent_len > clusters)
2046 extent_len = clusters;
2047
2048 clusters -= extent_len;
2049 cpos += extent_len;
2050 }
2051out:
2052 return ret;
2053}
2054
Mark Fasheha11f7e62011-06-22 14:23:38 -07002055static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2056{
2057 int blockmask = inode->i_sb->s_blocksize - 1;
2058 loff_t final_size = pos + count;
2059
2060 if ((pos & blockmask) || (final_size & blockmask))
2061 return 1;
2062 return 0;
2063}
2064
Tao Ma293b2f72009-08-25 08:02:48 +08002065static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
Tao Mab8908232010-08-12 10:27:14 +08002066 struct file *file,
Tao Ma293b2f72009-08-25 08:02:48 +08002067 loff_t pos, size_t count,
2068 int *meta_level)
2069{
2070 int ret;
2071 struct buffer_head *di_bh = NULL;
2072 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2073 u32 clusters =
2074 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2075
2076 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2077 if (ret) {
2078 mlog_errno(ret);
2079 goto out;
2080 }
2081
2082 *meta_level = 1;
2083
Tiger Yangc7dd3392013-08-13 16:00:58 -07002084 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
Tao Ma293b2f72009-08-25 08:02:48 +08002085 if (ret)
2086 mlog_errno(ret);
2087out:
2088 brelse(di_bh);
2089 return ret;
2090}
2091
Tao Mab8908232010-08-12 10:27:14 +08002092static int ocfs2_prepare_inode_for_write(struct file *file,
Al Viro90320252015-04-09 11:14:45 -04002093 loff_t pos,
Ryan Dingf1f973f2016-03-25 14:21:12 -07002094 size_t count)
Mark Fashehccd979b2005-12-15 14:31:24 -08002095{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002096 int ret = 0, meta_level = 0;
Tao Mab8908232010-08-12 10:27:14 +08002097 struct dentry *dentry = file->f_path.dentry;
David Howells2b0143b2015-03-17 22:25:59 +00002098 struct inode *inode = d_inode(dentry);
Al Viro90320252015-04-09 11:14:45 -04002099 loff_t end;
Mark Fashehccd979b2005-12-15 14:31:24 -08002100
Sunil Mushran2bd63212010-01-25 16:57:38 -08002101 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07002102 * We start with a read level meta lock and only jump to an ex
2103 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08002104 */
Mark Fashehccd979b2005-12-15 14:31:24 -08002105 for(;;) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07002106 ret = ocfs2_inode_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002107 if (ret < 0) {
2108 meta_level = -1;
2109 mlog_errno(ret);
2110 goto out;
2111 }
2112
2113 /* Clear suid / sgid if necessary. We do this here
2114 * instead of later in the write path because
2115 * remove_suid() calls ->setattr without any hint that
2116 * we may have already done our cluster locking. Since
2117 * ocfs2_setattr() *must* take cluster locks to
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002118 * proceed, this will lead us to recursively lock the
Mark Fashehccd979b2005-12-15 14:31:24 -08002119 * inode. There's also the dinode i_size state which
2120 * can be lost via setattr during extending writes (we
2121 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07002122 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002123 if (meta_level == 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07002124 ocfs2_inode_unlock(inode, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002125 meta_level = 1;
2126 continue;
2127 }
2128
2129 ret = ocfs2_write_remove_suid(inode);
2130 if (ret < 0) {
2131 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07002132 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08002133 }
2134 }
2135
Al Viro90320252015-04-09 11:14:45 -04002136 end = pos + count;
Mark Fasheh3a0782d2007-01-17 12:53:31 -08002137
Al Viro90320252015-04-09 11:14:45 -04002138 ret = ocfs2_check_range_for_refcount(inode, pos, count);
Tao Ma293b2f72009-08-25 08:02:48 +08002139 if (ret == 1) {
2140 ocfs2_inode_unlock(inode, meta_level);
2141 meta_level = -1;
2142
2143 ret = ocfs2_prepare_inode_for_refcount(inode,
Tao Mab8908232010-08-12 10:27:14 +08002144 file,
Al Viro90320252015-04-09 11:14:45 -04002145 pos,
Tao Ma293b2f72009-08-25 08:02:48 +08002146 count,
2147 &meta_level);
2148 }
2149
2150 if (ret < 0) {
2151 mlog_errno(ret);
2152 goto out_unlock;
2153 }
2154
Mark Fashehccd979b2005-12-15 14:31:24 -08002155 break;
2156 }
2157
Tiger Yang8659ac22006-10-17 18:29:52 -07002158out_unlock:
Tao Ma468eedd2011-02-22 22:14:41 +08002159 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
Ryan Dingf1f973f2016-03-25 14:21:12 -07002160 pos, count);
Tao Ma468eedd2011-02-22 22:14:41 +08002161
Tao Ma293b2f72009-08-25 08:02:48 +08002162 if (meta_level >= 0)
2163 ocfs2_inode_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002164
2165out:
2166 return ret;
2167}
2168
Al Viro3ef045c2014-04-03 14:25:22 -04002169static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2170 struct iov_iter *from)
Tiger Yang8659ac22006-10-17 18:29:52 -07002171{
Ryan Dingf1f973f2016-03-25 14:21:12 -07002172 int direct_io, rw_level;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002173 ssize_t written = 0;
Al Viro3309dd02015-04-09 12:55:47 -04002174 ssize_t ret;
Ryan Dingf1f973f2016-03-25 14:21:12 -07002175 size_t count = iov_iter_count(from);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002176 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -05002177 struct inode *inode = file_inode(file);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002178 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002179 int full_coherency = !(osb->s_mount_opt &
2180 OCFS2_MOUNT_COHERENCY_BUFFERED);
Ryan Dinge63890f2016-03-25 14:21:15 -07002181 void *saved_ki_complete = NULL;
Joseph Qifaaebf12015-09-04 15:43:40 -07002182 int append_write = ((iocb->ki_pos + count) >=
2183 i_size_read(inode) ? 1 : 0);
Tiger Yang8659ac22006-10-17 18:29:52 -07002184
Tao Ma468eedd2011-02-22 22:14:41 +08002185 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2186 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2187 file->f_path.dentry->d_name.len,
2188 file->f_path.dentry->d_name.name,
Al Viro3ef045c2014-04-03 14:25:22 -04002189 (unsigned int)from->nr_segs); /* GRRRRR */
Tiger Yang8659ac22006-10-17 18:29:52 -07002190
Christoph Hellwig66ee59a2015-02-11 19:56:46 +01002191 if (count == 0)
Tiger Yang8659ac22006-10-17 18:29:52 -07002192 return 0;
2193
Al Viro2ba48ce2015-04-09 13:52:01 -04002194 direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002195
Al Viro59551022016-01-22 15:40:57 -05002196 inode_lock(inode);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002197
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002198 /*
2199 * Concurrent O_DIRECT writes are allowed with
2200 * mount_option "coherency=buffered".
Joseph Qifaaebf12015-09-04 15:43:40 -07002201 * For append write, we must take rw EX.
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002202 */
Joseph Qifaaebf12015-09-04 15:43:40 -07002203 rw_level = (!direct_io || full_coherency || append_write);
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002204
Tiger Yang8659ac22006-10-17 18:29:52 -07002205 ret = ocfs2_rw_lock(inode, rw_level);
2206 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08002207 mlog_errno(ret);
WeiWei Wangfa5a0eb2015-06-24 16:55:15 -07002208 goto out_mutex;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002209 }
2210
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002211 /*
2212 * O_DIRECT writes with "coherency=full" need to take EX cluster
2213 * inode_lock to guarantee coherency.
2214 */
2215 if (direct_io && full_coherency) {
2216 /*
2217 * We need to take and drop the inode lock to force
2218 * other nodes to drop their caches. Buffered I/O
2219 * already does this in write_begin().
2220 */
2221 ret = ocfs2_inode_lock(inode, NULL, 1);
2222 if (ret < 0) {
2223 mlog_errno(ret);
Joseph Qiafe1bb72013-05-24 15:55:12 -07002224 goto out;
Tristan Ye7bdb0d12010-10-11 16:46:39 +08002225 }
2226
2227 ocfs2_inode_unlock(inode, 1);
2228 }
2229
Al Viro3309dd02015-04-09 12:55:47 -04002230 ret = generic_write_checks(iocb, from);
2231 if (ret <= 0) {
2232 if (ret)
2233 mlog_errno(ret);
Al Viro90320252015-04-09 11:14:45 -04002234 goto out;
2235 }
Al Viro3309dd02015-04-09 12:55:47 -04002236 count = ret;
Al Viro90320252015-04-09 11:14:45 -04002237
Ryan Dingf1f973f2016-03-25 14:21:12 -07002238 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002239 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07002240 mlog_errno(ret);
2241 goto out;
2242 }
2243
Ryan Dinge63890f2016-03-25 14:21:15 -07002244 if (direct_io && !is_sync_kiocb(iocb) &&
2245 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
Mark Fasheha11f7e62011-06-22 14:23:38 -07002246 /*
Ryan Dinge63890f2016-03-25 14:21:15 -07002247 * Make it a sync io if it's an unaligned aio.
Mark Fasheha11f7e62011-06-22 14:23:38 -07002248 */
Ryan Dinge63890f2016-03-25 14:21:15 -07002249 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
Mark Fasheha11f7e62011-06-22 14:23:38 -07002250 }
2251
Mark Fashehccd979b2005-12-15 14:31:24 -08002252 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002253 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002254
Al Viro7da839c2015-04-09 14:01:33 -04002255 written = __generic_file_write_iter(iocb, from);
Mark Fashehccd979b2005-12-15 14:31:24 -08002256 /* buffered aio wouldn't have proper lock coverage today */
Al Viro7da839c2015-04-09 14:01:33 -04002257 BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08002258
Ryan Dingaa1057b2015-09-04 15:42:36 -07002259 /*
2260 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2261 * function pointer which is called when o_direct io completes so that
2262 * it can unlock our rw lock.
2263 * Unfortunately there are error cases which call end_io and others
2264 * that don't. so we don't have to unlock the rw_lock if either an
2265 * async dio is going to do it in the future or an end_io after an
2266 * error has already done it.
2267 */
2268 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2269 rw_level = -1;
Ryan Dingaa1057b2015-09-04 15:42:36 -07002270 }
2271
Al Viro64b4e252015-04-08 17:00:32 -04002272 if (unlikely(written <= 0))
Ryan Dinge63890f2016-03-25 14:21:15 -07002273 goto out;
Al Viro64b4e252015-04-08 17:00:32 -04002274
Al Viro7da839c2015-04-09 14:01:33 -04002275 if (((file->f_flags & O_DSYNC) && !direct_io) ||
Ryan Dingf1f973f2016-03-25 14:21:12 -07002276 IS_SYNC(inode)) {
Al Viro64b4e252015-04-08 17:00:32 -04002277 ret = filemap_fdatawrite_range(file->f_mapping,
2278 iocb->ki_pos - written,
2279 iocb->ki_pos - 1);
Jan Kara918941a2009-08-17 18:50:08 +02002280 if (ret < 0)
2281 written = ret;
2282
Goldwyn Rodrigues86b9c6f2014-12-10 15:41:53 -08002283 if (!ret) {
Joel Becker2b4e30f2008-09-03 20:03:41 -07002284 ret = jbd2_journal_force_commit(osb->journal->j_journal);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002285 if (ret < 0)
2286 written = ret;
2287 }
Jan Kara918941a2009-08-17 18:50:08 +02002288
2289 if (!ret)
Al Viro64b4e252015-04-08 17:00:32 -04002290 ret = filemap_fdatawait_range(file->f_mapping,
2291 iocb->ki_pos - written,
2292 iocb->ki_pos - 1);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002293 }
2294
Mark Fashehccd979b2005-12-15 14:31:24 -08002295out:
Ryan Dinge63890f2016-03-25 14:21:15 -07002296 if (saved_ki_complete)
2297 xchg(&iocb->ki_complete, saved_ki_complete);
2298
Mark Fasheh9517bac2007-02-09 20:24:12 -08002299 if (rw_level != -1)
2300 ocfs2_rw_unlock(inode, rw_level);
2301
WeiWei Wangfa5a0eb2015-06-24 16:55:15 -07002302out_mutex:
Al Viro59551022016-01-22 15:40:57 -05002303 inode_unlock(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08002304
Wengang Wang812e7a62009-07-10 13:26:04 +08002305 if (written)
2306 ret = written;
Wengang Wang812e7a62009-07-10 13:26:04 +08002307 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002308}
2309
Tiger Yang8659ac22006-10-17 18:29:52 -07002310static ssize_t ocfs2_file_splice_read(struct file *in,
2311 loff_t *ppos,
2312 struct pipe_inode_info *pipe,
2313 size_t len,
2314 unsigned int flags)
2315{
Tao Ma1962f392009-06-19 15:36:52 +08002316 int ret = 0, lock_level = 0;
Al Viro496ad9a2013-01-23 17:07:38 -05002317 struct inode *inode = file_inode(in);
Tiger Yang8659ac22006-10-17 18:29:52 -07002318
Tao Ma468eedd2011-02-22 22:14:41 +08002319 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2320 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2321 in->f_path.dentry->d_name.len,
2322 in->f_path.dentry->d_name.name, len);
Tiger Yang8659ac22006-10-17 18:29:52 -07002323
2324 /*
Al Viro3cd9ad52014-04-02 14:44:18 -04002325 * See the comment in ocfs2_file_read_iter()
Tiger Yang8659ac22006-10-17 18:29:52 -07002326 */
Al Viro182be682013-01-24 02:21:54 -05002327 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002328 if (ret < 0) {
2329 mlog_errno(ret);
2330 goto bail;
2331 }
Tao Ma1962f392009-06-19 15:36:52 +08002332 ocfs2_inode_unlock(inode, lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002333
2334 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2335
2336bail:
Tiger Yang8659ac22006-10-17 18:29:52 -07002337 return ret;
2338}
2339
Al Viro3cd9ad52014-04-02 14:44:18 -04002340static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2341 struct iov_iter *to)
Mark Fashehccd979b2005-12-15 14:31:24 -08002342{
WeiWei Wangfa5a0eb2015-06-24 16:55:15 -07002343 int ret = 0, rw_level = -1, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002344 struct file *filp = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -05002345 struct inode *inode = file_inode(filp);
Mark Fashehccd979b2005-12-15 14:31:24 -08002346
Tao Ma468eedd2011-02-22 22:14:41 +08002347 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2348 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2349 filp->f_path.dentry->d_name.len,
Al Viro3cd9ad52014-04-02 14:44:18 -04002350 filp->f_path.dentry->d_name.name,
2351 to->nr_segs); /* GRRRRR */
Tao Ma468eedd2011-02-22 22:14:41 +08002352
Mark Fashehccd979b2005-12-15 14:31:24 -08002353
2354 if (!inode) {
2355 ret = -EINVAL;
2356 mlog_errno(ret);
2357 goto bail;
2358 }
2359
Sunil Mushran2bd63212010-01-25 16:57:38 -08002360 /*
Mark Fashehccd979b2005-12-15 14:31:24 -08002361 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2362 * need locks to protect pending reads from racing with truncate.
2363 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002364 if (iocb->ki_flags & IOCB_DIRECT) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002365 ret = ocfs2_rw_lock(inode, 0);
2366 if (ret < 0) {
2367 mlog_errno(ret);
2368 goto bail;
2369 }
2370 rw_level = 0;
2371 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002372 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002373 }
2374
Mark Fashehc4374f82006-05-05 19:04:35 -07002375 /*
2376 * We're fine letting folks race truncates and extending
2377 * writes with read across the cluster, just like they can
2378 * locally. Hence no rw_lock during read.
Sunil Mushran2bd63212010-01-25 16:57:38 -08002379 *
Mark Fashehc4374f82006-05-05 19:04:35 -07002380 * Take and drop the meta data lock to update inode fields
2381 * like i_size. This allows the checks down below
Sunil Mushran2bd63212010-01-25 16:57:38 -08002382 * generic_file_aio_read() a chance of actually working.
Mark Fashehc4374f82006-05-05 19:04:35 -07002383 */
Al Viro182be682013-01-24 02:21:54 -05002384 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002385 if (ret < 0) {
2386 mlog_errno(ret);
2387 goto bail;
2388 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07002389 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002390
Al Viro3cd9ad52014-04-02 14:44:18 -04002391 ret = generic_file_read_iter(iocb, to);
Tao Ma468eedd2011-02-22 22:14:41 +08002392 trace_generic_file_aio_read_ret(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08002393
2394 /* buffered aio wouldn't have proper lock coverage today */
Al Viro2ba48ce2015-04-09 13:52:01 -04002395 BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08002396
Al Viro3ef045c2014-04-03 14:25:22 -04002397 /* see ocfs2_file_write_iter */
Mark Fashehccd979b2005-12-15 14:31:24 -08002398 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2399 rw_level = -1;
Mark Fashehccd979b2005-12-15 14:31:24 -08002400 }
2401
2402bail:
Sunil Mushran2bd63212010-01-25 16:57:38 -08002403 if (rw_level != -1)
Mark Fashehccd979b2005-12-15 14:31:24 -08002404 ocfs2_rw_unlock(inode, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002405
2406 return ret;
2407}
2408
Sunil Mushran93862d52011-07-25 14:58:15 -07002409/* Refer generic_file_llseek_unlocked() */
Andrew Morton965c8e52012-12-17 15:59:39 -08002410static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
Sunil Mushran93862d52011-07-25 14:58:15 -07002411{
2412 struct inode *inode = file->f_mapping->host;
2413 int ret = 0;
2414
Al Viro59551022016-01-22 15:40:57 -05002415 inode_lock(inode);
Sunil Mushran93862d52011-07-25 14:58:15 -07002416
Andrew Morton965c8e52012-12-17 15:59:39 -08002417 switch (whence) {
Sunil Mushran93862d52011-07-25 14:58:15 -07002418 case SEEK_SET:
2419 break;
2420 case SEEK_END:
Jensenc8d888d2014-04-03 14:47:01 -07002421 /* SEEK_END requires the OCFS2 inode lock for the file
2422 * because it references the file's size.
2423 */
2424 ret = ocfs2_inode_lock(inode, NULL, 0);
2425 if (ret < 0) {
2426 mlog_errno(ret);
2427 goto out;
2428 }
2429 offset += i_size_read(inode);
2430 ocfs2_inode_unlock(inode, 0);
Sunil Mushran93862d52011-07-25 14:58:15 -07002431 break;
2432 case SEEK_CUR:
2433 if (offset == 0) {
2434 offset = file->f_pos;
2435 goto out;
2436 }
2437 offset += file->f_pos;
2438 break;
2439 case SEEK_DATA:
2440 case SEEK_HOLE:
Andrew Morton965c8e52012-12-17 15:59:39 -08002441 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
Sunil Mushran93862d52011-07-25 14:58:15 -07002442 if (ret)
2443 goto out;
2444 break;
2445 default:
2446 ret = -EINVAL;
2447 goto out;
2448 }
2449
Jie Liu46a1c2c2013-06-25 12:02:13 +08002450 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sunil Mushran93862d52011-07-25 14:58:15 -07002451
2452out:
Al Viro59551022016-01-22 15:40:57 -05002453 inode_unlock(inode);
Sunil Mushran93862d52011-07-25 14:58:15 -07002454 if (ret)
2455 return ret;
2456 return offset;
2457}
2458
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002459const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002460 .setattr = ocfs2_setattr,
2461 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002462 .permission = ocfs2_permission,
Tiger Yangcf1d6c72008-08-18 17:11:00 +08002463 .setxattr = generic_setxattr,
2464 .getxattr = generic_getxattr,
2465 .listxattr = ocfs2_listxattr,
2466 .removexattr = generic_removexattr,
Mark Fasheh00dc4172008-10-03 17:32:11 -04002467 .fiemap = ocfs2_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02002468 .get_acl = ocfs2_iop_get_acl,
Christoph Hellwig702e5bc2013-12-20 05:16:48 -08002469 .set_acl = ocfs2_iop_set_acl,
Mark Fashehccd979b2005-12-15 14:31:24 -08002470};
2471
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002472const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002473 .setattr = ocfs2_setattr,
2474 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002475 .permission = ocfs2_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02002476 .get_acl = ocfs2_iop_get_acl,
Christoph Hellwig702e5bc2013-12-20 05:16:48 -08002477 .set_acl = ocfs2_iop_set_acl,
Mark Fashehccd979b2005-12-15 14:31:24 -08002478};
2479
Mark Fasheh53da4932008-07-21 14:29:16 -07002480/*
2481 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2482 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2483 */
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002484const struct file_operations ocfs2_fops = {
Sunil Mushran93862d52011-07-25 14:58:15 -07002485 .llseek = ocfs2_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002486 .mmap = ocfs2_mmap,
2487 .fsync = ocfs2_sync_file,
2488 .release = ocfs2_file_release,
2489 .open = ocfs2_file_open,
Al Viro3cd9ad52014-04-02 14:44:18 -04002490 .read_iter = ocfs2_file_read_iter,
Al Viro3ef045c2014-04-03 14:25:22 -04002491 .write_iter = ocfs2_file_write_iter,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002492 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002493#ifdef CONFIG_COMPAT
2494 .compat_ioctl = ocfs2_compat_ioctl,
2495#endif
Mark Fasheh53da4932008-07-21 14:29:16 -07002496 .lock = ocfs2_lock,
2497 .flock = ocfs2_flock,
2498 .splice_read = ocfs2_file_splice_read,
Al Viro6dc8bc02014-04-05 04:34:23 -04002499 .splice_write = iter_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002500 .fallocate = ocfs2_fallocate,
Mark Fasheh53da4932008-07-21 14:29:16 -07002501};
2502
2503const struct file_operations ocfs2_dops = {
2504 .llseek = generic_file_llseek,
2505 .read = generic_read_dir,
Al Viro3704412b2013-05-22 21:06:00 -04002506 .iterate = ocfs2_readdir,
Mark Fasheh53da4932008-07-21 14:29:16 -07002507 .fsync = ocfs2_sync_file,
2508 .release = ocfs2_dir_release,
2509 .open = ocfs2_dir_open,
2510 .unlocked_ioctl = ocfs2_ioctl,
2511#ifdef CONFIG_COMPAT
2512 .compat_ioctl = ocfs2_compat_ioctl,
2513#endif
2514 .lock = ocfs2_lock,
2515 .flock = ocfs2_flock,
2516};
2517
2518/*
2519 * POSIX-lockless variants of our file_operations.
2520 *
2521 * These will be used if the underlying cluster stack does not support
2522 * posix file locking, if the user passes the "localflocks" mount
2523 * option, or if we have a local-only fs.
2524 *
2525 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2526 * so we still want it in the case of no stack support for
2527 * plocks. Internally, it will do the right thing when asked to ignore
2528 * the cluster.
2529 */
2530const struct file_operations ocfs2_fops_no_plocks = {
Sunil Mushran93862d52011-07-25 14:58:15 -07002531 .llseek = ocfs2_file_llseek,
Mark Fasheh53da4932008-07-21 14:29:16 -07002532 .mmap = ocfs2_mmap,
2533 .fsync = ocfs2_sync_file,
2534 .release = ocfs2_file_release,
2535 .open = ocfs2_file_open,
Al Viro3cd9ad52014-04-02 14:44:18 -04002536 .read_iter = ocfs2_file_read_iter,
Al Viro3ef045c2014-04-03 14:25:22 -04002537 .write_iter = ocfs2_file_write_iter,
Mark Fasheh53da4932008-07-21 14:29:16 -07002538 .unlocked_ioctl = ocfs2_ioctl,
2539#ifdef CONFIG_COMPAT
2540 .compat_ioctl = ocfs2_compat_ioctl,
2541#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002542 .flock = ocfs2_flock,
Tiger Yang8659ac22006-10-17 18:29:52 -07002543 .splice_read = ocfs2_file_splice_read,
Al Viro6dc8bc02014-04-05 04:34:23 -04002544 .splice_write = iter_file_splice_write,
Tristan Ye3d1c1822011-05-23 15:57:26 +08002545 .fallocate = ocfs2_fallocate,
Mark Fashehccd979b2005-12-15 14:31:24 -08002546};
2547
Mark Fasheh53da4932008-07-21 14:29:16 -07002548const struct file_operations ocfs2_dops_no_plocks = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002549 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002550 .read = generic_read_dir,
Al Viro3704412b2013-05-22 21:06:00 -04002551 .iterate = ocfs2_readdir,
Mark Fashehccd979b2005-12-15 14:31:24 -08002552 .fsync = ocfs2_sync_file,
Mark Fasheh53fc6222007-12-20 16:49:04 -08002553 .release = ocfs2_dir_release,
2554 .open = ocfs2_dir_open,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002555 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002556#ifdef CONFIG_COMPAT
2557 .compat_ioctl = ocfs2_compat_ioctl,
2558#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002559 .flock = ocfs2_flock,
Mark Fashehccd979b2005-12-15 14:31:24 -08002560};