blob: 6cf3d8d18369a3b3e7c13b9c763b522f9641a731 [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>
Mark Fashehccd979b2005-12-15 14:31:24 -080039
40#define MLOG_MASK_PREFIX ML_INODE
41#include <cluster/masklog.h>
42
43#include "ocfs2.h"
44
45#include "alloc.h"
46#include "aops.h"
47#include "dir.h"
48#include "dlmglue.h"
49#include "extent_map.h"
50#include "file.h"
51#include "sysfile.h"
52#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070053#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080054#include "journal.h"
Mark Fasheh53fc6222007-12-20 16:49:04 -080055#include "locks.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080056#include "mmap.h"
57#include "suballoc.h"
58#include "super.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080059#include "xattr.h"
Tiger Yang23fc2702008-11-14 11:17:18 +080060#include "acl.h"
Jan Karaa90714c2008-10-09 19:38:40 +020061#include "quota.h"
Tao Ma293b2f72009-08-25 08:02:48 +080062#include "refcounttree.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080063
64#include "buffer_head_io.h"
65
66static int ocfs2_sync_inode(struct inode *inode)
67{
68 filemap_fdatawrite(inode->i_mapping);
69 return sync_mapping_buffers(inode->i_mapping);
70}
71
Mark Fasheh53fc6222007-12-20 16:49:04 -080072static int ocfs2_init_file_private(struct inode *inode, struct file *file)
73{
74 struct ocfs2_file_private *fp;
75
76 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
77 if (!fp)
78 return -ENOMEM;
79
80 fp->fp_file = file;
81 mutex_init(&fp->fp_mutex);
82 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
83 file->private_data = fp;
84
85 return 0;
86}
87
88static void ocfs2_free_file_private(struct inode *inode, struct file *file)
89{
90 struct ocfs2_file_private *fp = file->private_data;
91 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
92
93 if (fp) {
94 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
95 ocfs2_lock_res_free(&fp->fp_flock);
96 kfree(fp);
97 file->private_data = NULL;
98 }
99}
100
Mark Fashehccd979b2005-12-15 14:31:24 -0800101static int ocfs2_file_open(struct inode *inode, struct file *file)
102{
103 int status;
104 int mode = file->f_flags;
105 struct ocfs2_inode_info *oi = OCFS2_I(inode);
106
107 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800108 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800109
110 spin_lock(&oi->ip_lock);
111
112 /* Check that the inode hasn't been wiped from disk by another
113 * node. If it hasn't then we're safe as long as we hold the
114 * spin lock until our increment of open count. */
115 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
116 spin_unlock(&oi->ip_lock);
117
118 status = -ENOENT;
119 goto leave;
120 }
121
122 if (mode & O_DIRECT)
123 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
124
125 oi->ip_open_count++;
126 spin_unlock(&oi->ip_lock);
Mark Fasheh53fc6222007-12-20 16:49:04 -0800127
128 status = ocfs2_init_file_private(inode, file);
129 if (status) {
130 /*
131 * We want to set open count back if we're failing the
132 * open.
133 */
134 spin_lock(&oi->ip_lock);
135 oi->ip_open_count--;
136 spin_unlock(&oi->ip_lock);
137 }
138
Mark Fashehccd979b2005-12-15 14:31:24 -0800139leave:
140 mlog_exit(status);
141 return status;
142}
143
144static int ocfs2_file_release(struct inode *inode, struct file *file)
145{
146 struct ocfs2_inode_info *oi = OCFS2_I(inode);
147
148 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800149 file->f_path.dentry->d_name.len,
150 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800151
152 spin_lock(&oi->ip_lock);
153 if (!--oi->ip_open_count)
154 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
155 spin_unlock(&oi->ip_lock);
156
Mark Fasheh53fc6222007-12-20 16:49:04 -0800157 ocfs2_free_file_private(inode, file);
158
Mark Fashehccd979b2005-12-15 14:31:24 -0800159 mlog_exit(0);
160
161 return 0;
162}
163
Mark Fasheh53fc6222007-12-20 16:49:04 -0800164static int ocfs2_dir_open(struct inode *inode, struct file *file)
165{
166 return ocfs2_init_file_private(inode, file);
167}
168
169static int ocfs2_dir_release(struct inode *inode, struct file *file)
170{
171 ocfs2_free_file_private(inode, file);
172 return 0;
173}
174
Mark Fashehccd979b2005-12-15 14:31:24 -0800175static int ocfs2_sync_file(struct file *file,
176 struct dentry *dentry,
177 int datasync)
178{
179 int err = 0;
180 journal_t *journal;
181 struct inode *inode = dentry->d_inode;
182 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
183
184 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
185 dentry->d_name.len, dentry->d_name.name);
186
187 err = ocfs2_sync_inode(dentry->d_inode);
188 if (err)
189 goto bail;
190
Hisashi Hifumie04cc152009-06-09 16:47:45 +0900191 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
192 goto bail;
193
Mark Fashehccd979b2005-12-15 14:31:24 -0800194 journal = osb->journal->j_journal;
Joel Becker2b4e30f2008-09-03 20:03:41 -0700195 err = jbd2_journal_force_commit(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800196
197bail:
198 mlog_exit(err);
199
200 return (err < 0) ? -EIO : 0;
201}
202
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800203int ocfs2_should_update_atime(struct inode *inode,
204 struct vfsmount *vfsmnt)
205{
206 struct timespec now;
207 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
208
209 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
210 return 0;
211
212 if ((inode->i_flags & S_NOATIME) ||
213 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
214 return 0;
215
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800216 /*
217 * We can be called with no vfsmnt structure - NFSD will
218 * sometimes do this.
219 *
220 * Note that our action here is different than touch_atime() -
221 * if we can't tell whether this is a noatime mount, then we
222 * don't know whether to trust the value of s_atime_quantum.
223 */
224 if (vfsmnt == NULL)
225 return 0;
226
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800227 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
228 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
229 return 0;
230
Mark Fasheh7e913c52006-12-13 00:34:35 -0800231 if (vfsmnt->mnt_flags & MNT_RELATIME) {
232 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
233 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
234 return 1;
235
236 return 0;
237 }
238
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800239 now = CURRENT_TIME;
240 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
241 return 0;
242 else
243 return 1;
244}
245
246int ocfs2_update_inode_atime(struct inode *inode,
247 struct buffer_head *bh)
248{
249 int ret;
250 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
251 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700252 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800253
254 mlog_entry_void();
255
256 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200257 if (IS_ERR(handle)) {
258 ret = PTR_ERR(handle);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800259 mlog_errno(ret);
260 goto out;
261 }
262
Joel Becker0cf2f762009-02-12 16:41:25 -0800263 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -0700264 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700265 if (ret) {
266 mlog_errno(ret);
267 goto out_commit;
268 }
269
270 /*
271 * Don't use ocfs2_mark_inode_dirty() here as we don't always
272 * have i_mutex to guard against concurrent changes to other
273 * inode fields.
274 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800275 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700276 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
277 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
278
279 ret = ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800280 if (ret < 0)
281 mlog_errno(ret);
282
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700283out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800284 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
285out:
286 mlog_exit(ret);
287 return ret;
288}
289
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700290static int ocfs2_set_inode_size(handle_t *handle,
291 struct inode *inode,
292 struct buffer_head *fe_bh,
293 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800294{
295 int status;
296
297 mlog_entry_void();
298 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700299 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800300 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
301
302 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
303 if (status < 0) {
304 mlog_errno(status);
305 goto bail;
306 }
307
308bail:
309 mlog_exit(status);
310 return status;
311}
312
Jan Kara9e33d692008-08-25 19:56:50 +0200313int ocfs2_simple_size_update(struct inode *inode,
314 struct buffer_head *di_bh,
315 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800316{
317 int ret;
318 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700319 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800320
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700321 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200322 if (IS_ERR(handle)) {
323 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800324 mlog_errno(ret);
325 goto out;
326 }
327
328 ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 new_i_size);
330 if (ret < 0)
331 mlog_errno(ret);
332
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700333 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800334out:
335 return ret;
336}
337
Tao Ma37f8a2b2009-08-26 09:47:28 +0800338static int ocfs2_cow_file_pos(struct inode *inode,
339 struct buffer_head *fe_bh,
340 u64 offset)
341{
342 int status;
343 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
344 unsigned int num_clusters = 0;
345 unsigned int ext_flags = 0;
346
347 /*
348 * If the new offset is aligned to the range of the cluster, there is
349 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
350 * CoW either.
351 */
352 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
353 return 0;
354
355 status = ocfs2_get_clusters(inode, cpos, &phys,
356 &num_clusters, &ext_flags);
357 if (status) {
358 mlog_errno(status);
359 goto out;
360 }
361
362 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
363 goto out;
364
365 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
366
367out:
368 return status;
369}
370
Mark Fashehccd979b2005-12-15 14:31:24 -0800371static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
372 struct inode *inode,
373 struct buffer_head *fe_bh,
374 u64 new_i_size)
375{
376 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700377 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800378 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700379 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800380
381 mlog_entry_void();
382
Tao Ma37f8a2b2009-08-26 09:47:28 +0800383 /*
384 * We need to CoW the cluster contains the offset if it is reflinked
385 * since we will call ocfs2_zero_range_for_truncate later which will
386 * write "0" from offset to the end of the cluster.
387 */
388 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
389 if (status) {
390 mlog_errno(status);
391 return status;
392 }
393
Mark Fashehccd979b2005-12-15 14:31:24 -0800394 /* TODO: This needs to actually orphan the inode in this
395 * transaction. */
396
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700397 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800398 if (IS_ERR(handle)) {
399 status = PTR_ERR(handle);
400 mlog_errno(status);
401 goto out;
402 }
403
Joel Becker0cf2f762009-02-12 16:41:25 -0800404 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700405 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800406 if (status < 0) {
407 mlog_errno(status);
408 goto out_commit;
409 }
410
411 /*
412 * Do this before setting i_size.
413 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700414 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
415 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
416 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800417 if (status) {
418 mlog_errno(status);
419 goto out_commit;
420 }
421
422 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800423 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
424
425 di = (struct ocfs2_dinode *) fe_bh->b_data;
426 di->i_size = cpu_to_le64(new_i_size);
427 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
428 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
429
430 status = ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800431 if (status < 0)
432 mlog_errno(status);
433
Mark Fasheh60b11392007-02-16 11:46:50 -0800434out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700435 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800436out:
Mark Fasheh60b11392007-02-16 11:46:50 -0800437
Mark Fashehccd979b2005-12-15 14:31:24 -0800438 mlog_exit(status);
439 return status;
440}
441
442static int ocfs2_truncate_file(struct inode *inode,
443 struct buffer_head *di_bh,
444 u64 new_i_size)
445{
446 int status = 0;
447 struct ocfs2_dinode *fe = NULL;
448 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
449 struct ocfs2_truncate_context *tc = NULL;
450
Mark Fashehb06970532006-03-03 10:24:33 -0800451 mlog_entry("(inode = %llu, new_i_size = %llu\n",
452 (unsigned long long)OCFS2_I(inode)->ip_blkno,
453 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800454
Joel Beckerb657c952008-11-13 14:49:11 -0800455 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
456 * already validated it */
Mark Fashehccd979b2005-12-15 14:31:24 -0800457 fe = (struct ocfs2_dinode *) di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800458
459 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800460 "Inode %llu, inode i_size = %lld != di "
461 "i_size = %llu, i_flags = 0x%x\n",
462 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800463 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800464 (unsigned long long)le64_to_cpu(fe->i_size),
465 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800466
467 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800468 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
469 (unsigned long long)le64_to_cpu(fe->i_size),
470 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800471 status = -EINVAL;
472 mlog_errno(status);
473 goto bail;
474 }
475
Mark Fashehb06970532006-03-03 10:24:33 -0800476 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
477 (unsigned long long)le64_to_cpu(fe->i_blkno),
478 (unsigned long long)le64_to_cpu(fe->i_size),
479 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800480
481 /* lets handle the simple truncate cases before doing any more
482 * cluster locking. */
483 if (new_i_size == le64_to_cpu(fe->i_size))
484 goto bail;
485
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700486 down_write(&OCFS2_I(inode)->ip_alloc_sem);
487
Mark Fashehc934a922007-10-18 15:23:46 -0700488 /*
489 * The inode lock forced other nodes to sync and drop their
490 * pages, which (correctly) happens even if we have a truncate
491 * without allocation change - ocfs2 cluster sizes can be much
492 * greater than page size, so we have to truncate them
493 * anyway.
494 */
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700495 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
496 truncate_inode_pages(inode->i_mapping, new_i_size);
497
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700498 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
499 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
Mark Fashehb1967d02007-11-20 11:56:39 -0800500 i_size_read(inode), 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700501 if (status)
502 mlog_errno(status);
503
Mark Fashehc934a922007-10-18 15:23:46 -0700504 goto bail_unlock_sem;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700505 }
506
Mark Fashehccd979b2005-12-15 14:31:24 -0800507 /* alright, we're going to need to do a full blown alloc size
508 * change. Orphan the inode so that recovery can complete the
509 * truncate if necessary. This does the task of marking
510 * i_size. */
511 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
512 if (status < 0) {
513 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700514 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800515 }
516
517 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
518 if (status < 0) {
519 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700520 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800521 }
522
523 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
524 if (status < 0) {
525 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700526 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800527 }
528
529 /* TODO: orphan dir cleanup here. */
Mark Fashehc934a922007-10-18 15:23:46 -0700530bail_unlock_sem:
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700531 up_write(&OCFS2_I(inode)->ip_alloc_sem);
532
Mark Fashehccd979b2005-12-15 14:31:24 -0800533bail:
Tao Ma8b2c0db2009-08-18 11:43:49 +0800534 if (!status && OCFS2_I(inode)->ip_clusters == 0)
535 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800536
537 mlog_exit(status);
538 return status;
539}
540
541/*
Tao Ma0eb8d47e2008-08-18 17:38:45 +0800542 * extend file allocation only here.
Mark Fashehccd979b2005-12-15 14:31:24 -0800543 * we'll update all the disk stuff, and oip->alloc_size
544 *
545 * expect stuff to be locked, a transaction started and enough data /
546 * metadata reservations in the contexts.
547 *
548 * Will return -EAGAIN, and a reason if a restart is needed.
549 * If passed in, *reason will always be set, even in error.
550 */
Tao Ma0eb8d47e2008-08-18 17:38:45 +0800551int ocfs2_add_inode_data(struct ocfs2_super *osb,
552 struct inode *inode,
553 u32 *logical_offset,
554 u32 clusters_to_add,
555 int mark_unwritten,
556 struct buffer_head *fe_bh,
557 handle_t *handle,
558 struct ocfs2_alloc_context *data_ac,
559 struct ocfs2_alloc_context *meta_ac,
560 enum ocfs2_alloc_restarted *reason_ret)
Mark Fashehccd979b2005-12-15 14:31:24 -0800561{
Joel Beckerf99b9b72008-08-20 19:36:33 -0700562 int ret;
563 struct ocfs2_extent_tree et;
Mark Fashehccd979b2005-12-15 14:31:24 -0800564
Joel Becker5e404e92009-02-13 03:54:22 -0800565 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
Joel Beckercbee7e12009-02-13 03:34:15 -0800566 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
567 clusters_to_add, mark_unwritten,
568 data_ac, meta_ac, reason_ret);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700569
570 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800571}
572
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800573static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
574 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800575{
576 int status = 0;
577 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800578 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800579 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800580 struct buffer_head *bh = NULL;
581 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700582 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800583 struct ocfs2_alloc_context *data_ac = NULL;
584 struct ocfs2_alloc_context *meta_ac = NULL;
585 enum ocfs2_alloc_restarted why;
586 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700587 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +0200588 int did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800589
590 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
591
Mark Fashehdcd05382007-01-16 11:32:23 -0800592 /*
593 * This function only exists for file systems which don't
594 * support holes.
595 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800596 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800597
Joel Beckerb657c952008-11-13 14:49:11 -0800598 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800599 if (status < 0) {
600 mlog_errno(status);
601 goto leave;
602 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800603 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800604
605restart_all:
606 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
607
Tao Mae7d4cb62008-08-18 17:38:44 +0800608 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
609 "clusters_to_add = %u\n",
610 (unsigned long long)OCFS2_I(inode)->ip_blkno,
611 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
612 clusters_to_add);
Joel Becker5e404e92009-02-13 03:54:22 -0800613 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700614 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
615 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800616 if (status) {
617 mlog_errno(status);
618 goto leave;
619 }
620
Tao Ma811f9332008-08-18 17:38:43 +0800621 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
622 clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700623 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800624 if (IS_ERR(handle)) {
625 status = PTR_ERR(handle);
626 handle = NULL;
627 mlog_errno(status);
628 goto leave;
629 }
630
631restarted_transaction:
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500632 status = dquot_alloc_space_nodirty(inode,
633 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
634 if (status)
Jan Karaa90714c2008-10-09 19:38:40 +0200635 goto leave;
Jan Karaa90714c2008-10-09 19:38:40 +0200636 did_quota = 1;
637
Mark Fashehccd979b2005-12-15 14:31:24 -0800638 /* reserve a write to the file entry early on - that we if we
639 * run out of credits in the allocation path, we can still
640 * update i_size. */
Joel Becker0cf2f762009-02-12 16:41:25 -0800641 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -0700642 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800643 if (status < 0) {
644 mlog_errno(status);
645 goto leave;
646 }
647
648 prev_clusters = OCFS2_I(inode)->ip_clusters;
649
Tao Ma0eb8d47e2008-08-18 17:38:45 +0800650 status = ocfs2_add_inode_data(osb,
651 inode,
652 &logical_start,
653 clusters_to_add,
654 mark_unwritten,
655 bh,
656 handle,
657 data_ac,
658 meta_ac,
659 &why);
Mark Fashehccd979b2005-12-15 14:31:24 -0800660 if ((status < 0) && (status != -EAGAIN)) {
661 if (status != -ENOSPC)
662 mlog_errno(status);
663 goto leave;
664 }
665
666 status = ocfs2_journal_dirty(handle, bh);
667 if (status < 0) {
668 mlog_errno(status);
669 goto leave;
670 }
671
672 spin_lock(&OCFS2_I(inode)->ip_lock);
673 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
674 spin_unlock(&OCFS2_I(inode)->ip_lock);
Jan Karaa90714c2008-10-09 19:38:40 +0200675 /* Release unused quota reservation */
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500676 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +0200677 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
678 did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800679
680 if (why != RESTART_NONE && clusters_to_add) {
681 if (why == RESTART_META) {
682 mlog(0, "restarting function.\n");
683 restart_func = 1;
684 } else {
685 BUG_ON(why != RESTART_TRANS);
686
687 mlog(0, "restarting transaction.\n");
688 /* TODO: This can be more intelligent. */
689 credits = ocfs2_calc_extend_credits(osb->sb,
Tao Ma811f9332008-08-18 17:38:43 +0800690 &fe->id2.i_list,
Mark Fashehccd979b2005-12-15 14:31:24 -0800691 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700692 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800693 if (status < 0) {
694 /* handle still has to be committed at
695 * this point. */
696 status = -ENOMEM;
697 mlog_errno(status);
698 goto leave;
699 }
700 goto restarted_transaction;
701 }
702 }
703
Mark Fashehb06970532006-03-03 10:24:33 -0800704 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700705 le32_to_cpu(fe->i_clusters),
706 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800707 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
Jan Kara634bf742007-12-19 15:25:42 +0100708 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800709
710leave:
Jan Karaa90714c2008-10-09 19:38:40 +0200711 if (status < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500712 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +0200713 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
Mark Fashehccd979b2005-12-15 14:31:24 -0800714 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700715 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800716 handle = NULL;
717 }
718 if (data_ac) {
719 ocfs2_free_alloc_context(data_ac);
720 data_ac = NULL;
721 }
722 if (meta_ac) {
723 ocfs2_free_alloc_context(meta_ac);
724 meta_ac = NULL;
725 }
726 if ((!status) && restart_func) {
727 restart_func = 0;
728 goto restart_all;
729 }
Mark Fasheha81cb882008-10-07 14:25:16 -0700730 brelse(bh);
731 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800732
733 mlog_exit(status);
734 return status;
735}
736
737/* Some parts of this taken from generic_cont_expand, which turned out
738 * to be too fragile to do exactly what we need without us having to
Nick Piggin4e02ed42008-10-29 14:00:55 -0700739 * worry about recursive locking in ->write_begin() and ->write_end(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800740static int ocfs2_write_zero_page(struct inode *inode,
741 u64 size)
742{
743 struct address_space *mapping = inode->i_mapping;
744 struct page *page;
745 unsigned long index;
746 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700747 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800748 int ret;
749
750 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
Sunil Mushran2bd63212010-01-25 16:57:38 -0800751 /* ugh. in prepare/commit_write, if from==to==start of block, we
Mark Fashehccd979b2005-12-15 14:31:24 -0800752 ** skip the prepare. make sure we never send an offset for the start
753 ** of a block
754 */
755 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
756 offset++;
757 }
758 index = size >> PAGE_CACHE_SHIFT;
759
760 page = grab_cache_page(mapping, index);
761 if (!page) {
762 ret = -ENOMEM;
763 mlog_errno(ret);
764 goto out;
765 }
766
Mark Fasheh53013cb2006-05-05 19:04:03 -0700767 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800768 if (ret < 0) {
769 mlog_errno(ret);
770 goto out_unlock;
771 }
772
773 if (ocfs2_should_order_data(inode)) {
774 handle = ocfs2_start_walk_page_trans(inode, page, offset,
775 offset);
776 if (IS_ERR(handle)) {
777 ret = PTR_ERR(handle);
778 handle = NULL;
779 goto out_unlock;
780 }
781 }
782
783 /* must not update i_size! */
784 ret = block_commit_write(page, offset, offset);
785 if (ret < 0)
786 mlog_errno(ret);
787 else
788 ret = 0;
789
790 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700791 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800792out_unlock:
793 unlock_page(page);
794 page_cache_release(page);
795out:
796 return ret;
797}
798
799static int ocfs2_zero_extend(struct inode *inode,
800 u64 zero_to_size)
801{
802 int ret = 0;
803 u64 start_off;
804 struct super_block *sb = inode->i_sb;
805
806 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
807 while (start_off < zero_to_size) {
808 ret = ocfs2_write_zero_page(inode, start_off);
809 if (ret < 0) {
810 mlog_errno(ret);
811 goto out;
812 }
813
814 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700815
816 /*
817 * Very large extends have the potential to lock up
818 * the cpu for extended periods of time.
819 */
820 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800821 }
822
823out:
824 return ret;
825}
826
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700827int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
828{
829 int ret;
830 u32 clusters_to_add;
831 struct ocfs2_inode_info *oi = OCFS2_I(inode);
832
833 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
834 if (clusters_to_add < oi->ip_clusters)
835 clusters_to_add = 0;
836 else
837 clusters_to_add -= oi->ip_clusters;
838
839 if (clusters_to_add) {
840 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
841 clusters_to_add, 0);
842 if (ret) {
843 mlog_errno(ret);
844 goto out;
845 }
846 }
847
848 /*
849 * Call this even if we don't add any clusters to the tree. We
850 * still need to zero the area between the old i_size and the
851 * new i_size.
852 */
853 ret = ocfs2_zero_extend(inode, zero_to);
854 if (ret < 0)
855 mlog_errno(ret);
856
857out:
858 return ret;
859}
860
Mark Fashehccd979b2005-12-15 14:31:24 -0800861static int ocfs2_extend_file(struct inode *inode,
862 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700863 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800864{
Mark Fashehc934a922007-10-18 15:23:46 -0700865 int ret = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700866 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800867
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700868 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700869
Mark Fashehccd979b2005-12-15 14:31:24 -0800870 /* setattr sometimes calls us like this. */
871 if (new_i_size == 0)
872 goto out;
873
874 if (i_size_read(inode) == new_i_size)
875 goto out;
876 BUG_ON(new_i_size < i_size_read(inode));
877
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700878 /*
879 * Fall through for converting inline data, even if the fs
880 * supports sparse files.
881 *
882 * The check for inline data here is legal - nobody can add
883 * the feature since we have i_mutex. We must check it again
884 * after acquiring ip_alloc_sem though, as paths like mmap
885 * might have raced us to converting the inode to extents.
886 */
887 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
888 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800889 goto out_update_size;
Mark Fashehccd979b2005-12-15 14:31:24 -0800890
Mark Fasheh0effef72006-10-03 17:44:42 -0700891 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700892 * The alloc sem blocks people in read/write from reading our
893 * allocation until we're done changing it. We depend on
894 * i_mutex to block other extend/truncate calls while we're
895 * here.
Mark Fasheh0effef72006-10-03 17:44:42 -0700896 */
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700897 down_write(&oi->ip_alloc_sem);
898
899 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
900 /*
901 * We can optimize small extends by keeping the inodes
902 * inline data.
903 */
904 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
905 up_write(&oi->ip_alloc_sem);
906 goto out_update_size;
907 }
908
909 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
910 if (ret) {
911 up_write(&oi->ip_alloc_sem);
912
913 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -0700914 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700915 }
916 }
917
918 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
919 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
920
921 up_write(&oi->ip_alloc_sem);
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700922
Mark Fasheh0effef72006-10-03 17:44:42 -0700923 if (ret < 0) {
924 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -0700925 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800926 }
927
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800928out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700929 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
930 if (ret < 0)
931 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700932
Mark Fashehccd979b2005-12-15 14:31:24 -0800933out:
934 return ret;
935}
936
937int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
938{
939 int status = 0, size_change;
940 struct inode *inode = dentry->d_inode;
941 struct super_block *sb = inode->i_sb;
942 struct ocfs2_super *osb = OCFS2_SB(sb);
943 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700944 handle_t *handle = NULL;
Jan Kara65bac572009-06-02 14:24:01 +0200945 int qtype;
946 struct dquot *transfer_from[MAXQUOTAS] = { };
947 struct dquot *transfer_to[MAXQUOTAS] = { };
Mark Fashehccd979b2005-12-15 14:31:24 -0800948
949 mlog_entry("(0x%p, '%.*s')\n", dentry,
950 dentry->d_name.len, dentry->d_name.name);
951
Sunil Mushranbc535802008-04-18 10:23:53 -0700952 /* ensuring we don't even attempt to truncate a symlink */
953 if (S_ISLNK(inode->i_mode))
954 attr->ia_valid &= ~ATTR_SIZE;
955
Mark Fashehccd979b2005-12-15 14:31:24 -0800956 if (attr->ia_valid & ATTR_MODE)
957 mlog(0, "mode change: %d\n", attr->ia_mode);
958 if (attr->ia_valid & ATTR_UID)
959 mlog(0, "uid change: %d\n", attr->ia_uid);
960 if (attr->ia_valid & ATTR_GID)
961 mlog(0, "gid change: %d\n", attr->ia_gid);
962 if (attr->ia_valid & ATTR_SIZE)
963 mlog(0, "size change...\n");
964 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
965 mlog(0, "time change...\n");
966
967#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
968 | ATTR_GID | ATTR_UID | ATTR_MODE)
969 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
970 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
971 return 0;
972 }
973
974 status = inode_change_ok(inode, attr);
975 if (status)
976 return status;
977
978 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
979 if (size_change) {
980 status = ocfs2_rw_lock(inode, 1);
981 if (status < 0) {
982 mlog_errno(status);
983 goto bail;
984 }
985 }
986
Mark Fashehe63aecb62007-10-18 15:30:42 -0700987 status = ocfs2_inode_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800988 if (status < 0) {
989 if (status != -ENOENT)
990 mlog_errno(status);
991 goto bail_unlock_rw;
992 }
993
994 if (size_change && attr->ia_size != i_size_read(inode)) {
Mark Fashehce76fd32007-07-20 12:02:14 -0700995 if (attr->ia_size > sb->s_maxbytes) {
996 status = -EFBIG;
997 goto bail_unlock;
998 }
999
Joel Becker2b4e30f2008-09-03 20:03:41 -07001000 if (i_size_read(inode) > attr->ia_size) {
1001 if (ocfs2_should_order_data(inode)) {
1002 status = ocfs2_begin_ordered_truncate(inode,
1003 attr->ia_size);
1004 if (status)
1005 goto bail_unlock;
1006 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001007 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
Joel Becker2b4e30f2008-09-03 20:03:41 -07001008 } else
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001009 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001010 if (status < 0) {
1011 if (status != -ENOSPC)
1012 mlog_errno(status);
1013 status = -ENOSPC;
1014 goto bail_unlock;
1015 }
1016 }
1017
Jan Karaa90714c2008-10-09 19:38:40 +02001018 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1019 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
Jan Kara65bac572009-06-02 14:24:01 +02001020 /*
1021 * Gather pointers to quota structures so that allocation /
1022 * freeing of quota structures happens here and not inside
1023 * vfs_dq_transfer() where we have problems with lock ordering
1024 */
Jan Karaa90714c2008-10-09 19:38:40 +02001025 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1026 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1027 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
Jan Kara65bac572009-06-02 14:24:01 +02001028 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1029 USRQUOTA);
1030 transfer_from[USRQUOTA] = dqget(sb, inode->i_uid,
1031 USRQUOTA);
1032 if (!transfer_to[USRQUOTA] || !transfer_from[USRQUOTA]) {
1033 status = -ESRCH;
Jan Karaa90714c2008-10-09 19:38:40 +02001034 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001035 }
Jan Karaa90714c2008-10-09 19:38:40 +02001036 }
1037 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1038 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1039 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
Jan Kara65bac572009-06-02 14:24:01 +02001040 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1041 GRPQUOTA);
1042 transfer_from[GRPQUOTA] = dqget(sb, inode->i_gid,
1043 GRPQUOTA);
1044 if (!transfer_to[GRPQUOTA] || !transfer_from[GRPQUOTA]) {
1045 status = -ESRCH;
Jan Karaa90714c2008-10-09 19:38:40 +02001046 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001047 }
Jan Karaa90714c2008-10-09 19:38:40 +02001048 }
Jan Kara65bac572009-06-02 14:24:01 +02001049 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1050 2 * ocfs2_quota_trans_credits(sb));
Jan Karaa90714c2008-10-09 19:38:40 +02001051 if (IS_ERR(handle)) {
1052 status = PTR_ERR(handle);
1053 mlog_errno(status);
1054 goto bail_unlock;
1055 }
1056 status = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
1057 if (status < 0)
1058 goto bail_commit;
1059 } else {
1060 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1061 if (IS_ERR(handle)) {
1062 status = PTR_ERR(handle);
1063 mlog_errno(status);
1064 goto bail_unlock;
1065 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001066 }
1067
Mark Fasheh7307de82007-05-09 15:16:19 -07001068 /*
1069 * This will intentionally not wind up calling vmtruncate(),
1070 * since all the work for a size change has been done above.
1071 * Otherwise, we could get into problems with truncate as
1072 * ip_alloc_sem is used there to protect against i_size
1073 * changes.
1074 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001075 status = inode_setattr(inode, attr);
1076 if (status < 0) {
1077 mlog_errno(status);
1078 goto bail_commit;
1079 }
1080
1081 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1082 if (status < 0)
1083 mlog_errno(status);
1084
1085bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001086 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001087bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001088 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001089bail_unlock_rw:
1090 if (size_change)
1091 ocfs2_rw_unlock(inode, 1);
1092bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001093 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001094
Jan Kara65bac572009-06-02 14:24:01 +02001095 /* Release quota pointers in case we acquired them */
1096 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
1097 dqput(transfer_to[qtype]);
1098 dqput(transfer_from[qtype]);
1099 }
1100
Tiger Yang060bc662008-11-14 11:17:29 +08001101 if (!status && attr->ia_valid & ATTR_MODE) {
1102 status = ocfs2_acl_chmod(inode);
1103 if (status < 0)
1104 mlog_errno(status);
1105 }
1106
Mark Fashehccd979b2005-12-15 14:31:24 -08001107 mlog_exit(status);
1108 return status;
1109}
1110
1111int ocfs2_getattr(struct vfsmount *mnt,
1112 struct dentry *dentry,
1113 struct kstat *stat)
1114{
1115 struct inode *inode = dentry->d_inode;
1116 struct super_block *sb = dentry->d_inode->i_sb;
1117 struct ocfs2_super *osb = sb->s_fs_info;
1118 int err;
1119
1120 mlog_entry_void();
1121
1122 err = ocfs2_inode_revalidate(dentry);
1123 if (err) {
1124 if (err != -ENOENT)
1125 mlog_errno(err);
1126 goto bail;
1127 }
1128
1129 generic_fillattr(inode, stat);
1130
1131 /* We set the blksize from the cluster size for performance */
1132 stat->blksize = osb->s_clustersize;
1133
1134bail:
1135 mlog_exit(err);
1136
1137 return err;
1138}
1139
Al Viroe6305c42008-07-15 21:03:57 -04001140int ocfs2_permission(struct inode *inode, int mask)
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001141{
1142 int ret;
1143
1144 mlog_entry_void();
1145
Mark Fashehe63aecb62007-10-18 15:30:42 -07001146 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001147 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001148 if (ret != -ENOENT)
1149 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001150 goto out;
1151 }
1152
Tiger Yang23fc2702008-11-14 11:17:18 +08001153 ret = generic_permission(inode, mask, ocfs2_check_acl);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001154
Mark Fashehe63aecb62007-10-18 15:30:42 -07001155 ocfs2_inode_unlock(inode, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001156out:
1157 mlog_exit(ret);
1158 return ret;
1159}
1160
Mark Fashehb2580102007-03-09 16:53:21 -08001161static int __ocfs2_write_remove_suid(struct inode *inode,
1162 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001163{
1164 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001165 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001166 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1167 struct ocfs2_dinode *di;
1168
Mark Fashehb06970532006-03-03 10:24:33 -08001169 mlog_entry("(Inode %llu, mode 0%o)\n",
Mark Fashehb2580102007-03-09 16:53:21 -08001170 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001171
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001172 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001173 if (IS_ERR(handle)) {
1174 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001175 mlog_errno(ret);
1176 goto out;
1177 }
1178
Joel Becker0cf2f762009-02-12 16:41:25 -08001179 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -07001180 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001181 if (ret < 0) {
1182 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001183 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001184 }
1185
1186 inode->i_mode &= ~S_ISUID;
1187 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1188 inode->i_mode &= ~S_ISGID;
1189
1190 di = (struct ocfs2_dinode *) bh->b_data;
1191 di->i_mode = cpu_to_le16(inode->i_mode);
1192
1193 ret = ocfs2_journal_dirty(handle, bh);
1194 if (ret < 0)
1195 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001196
Mark Fashehccd979b2005-12-15 14:31:24 -08001197out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001198 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001199out:
1200 mlog_exit(ret);
1201 return ret;
1202}
1203
Mark Fasheh9517bac2007-02-09 20:24:12 -08001204/*
1205 * Will look for holes and unwritten extents in the range starting at
1206 * pos for count bytes (inclusive).
1207 */
1208static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1209 size_t count)
1210{
1211 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001212 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001213 u32 cpos, clusters, extent_len, phys_cpos;
1214 struct super_block *sb = inode->i_sb;
1215
1216 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1217 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1218
1219 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001220 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1221 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001222 if (ret < 0) {
1223 mlog_errno(ret);
1224 goto out;
1225 }
1226
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001227 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001228 ret = 1;
1229 break;
1230 }
1231
1232 if (extent_len > clusters)
1233 extent_len = clusters;
1234
1235 clusters -= extent_len;
1236 cpos += extent_len;
1237 }
1238out:
1239 return ret;
1240}
1241
Mark Fashehb2580102007-03-09 16:53:21 -08001242static int ocfs2_write_remove_suid(struct inode *inode)
1243{
1244 int ret;
1245 struct buffer_head *bh = NULL;
Mark Fashehb2580102007-03-09 16:53:21 -08001246
Joel Beckerb657c952008-11-13 14:49:11 -08001247 ret = ocfs2_read_inode_block(inode, &bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001248 if (ret < 0) {
1249 mlog_errno(ret);
1250 goto out;
1251 }
1252
1253 ret = __ocfs2_write_remove_suid(inode, bh);
1254out:
1255 brelse(bh);
1256 return ret;
1257}
1258
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001259/*
1260 * Allocate enough extents to cover the region starting at byte offset
1261 * start for len bytes. Existing extents are skipped, any extents
1262 * added are marked as "unwritten".
1263 */
1264static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1265 u64 start, u64 len)
1266{
1267 int ret;
1268 u32 cpos, phys_cpos, clusters, alloc_size;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001269 u64 end = start + len;
1270 struct buffer_head *di_bh = NULL;
1271
1272 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Joel Beckerb657c952008-11-13 14:49:11 -08001273 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001274 if (ret) {
1275 mlog_errno(ret);
1276 goto out;
1277 }
1278
1279 /*
1280 * Nothing to do if the requested reservation range
1281 * fits within the inode.
1282 */
1283 if (ocfs2_size_fits_inline_data(di_bh, end))
1284 goto out;
1285
1286 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1287 if (ret) {
1288 mlog_errno(ret);
1289 goto out;
1290 }
1291 }
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001292
1293 /*
1294 * We consider both start and len to be inclusive.
1295 */
1296 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1297 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1298 clusters -= cpos;
1299
1300 while (clusters) {
1301 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1302 &alloc_size, NULL);
1303 if (ret) {
1304 mlog_errno(ret);
1305 goto out;
1306 }
1307
1308 /*
1309 * Hole or existing extent len can be arbitrary, so
1310 * cap it to our own allocation request.
1311 */
1312 if (alloc_size > clusters)
1313 alloc_size = clusters;
1314
1315 if (phys_cpos) {
1316 /*
1317 * We already have an allocation at this
1318 * region so we can safely skip it.
1319 */
1320 goto next;
1321 }
1322
1323 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1324 if (ret) {
1325 if (ret != -ENOSPC)
1326 mlog_errno(ret);
1327 goto out;
1328 }
1329
1330next:
1331 cpos += alloc_size;
1332 clusters -= alloc_size;
1333 }
1334
1335 ret = 0;
1336out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001337
1338 brelse(di_bh);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001339 return ret;
1340}
1341
Mark Fasheh063c4562007-07-03 13:34:11 -07001342/*
1343 * Truncate a byte range, avoiding pages within partial clusters. This
1344 * preserves those pages for the zeroing code to write to.
1345 */
1346static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1347 u64 byte_len)
1348{
1349 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1350 loff_t start, end;
1351 struct address_space *mapping = inode->i_mapping;
1352
1353 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1354 end = byte_start + byte_len;
1355 end = end & ~(osb->s_clustersize - 1);
1356
1357 if (start < end) {
1358 unmap_mapping_range(mapping, start, end - start, 0);
1359 truncate_inode_pages_range(mapping, start, end - 1);
1360 }
1361}
1362
1363static int ocfs2_zero_partial_clusters(struct inode *inode,
1364 u64 start, u64 len)
1365{
1366 int ret = 0;
1367 u64 tmpend, end = start + len;
1368 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1369 unsigned int csize = osb->s_clustersize;
1370 handle_t *handle;
1371
1372 /*
1373 * The "start" and "end" values are NOT necessarily part of
1374 * the range whose allocation is being deleted. Rather, this
1375 * is what the user passed in with the request. We must zero
1376 * partial clusters here. There's no need to worry about
1377 * physical allocation - the zeroing code knows to skip holes.
1378 */
1379 mlog(0, "byte start: %llu, end: %llu\n",
1380 (unsigned long long)start, (unsigned long long)end);
1381
1382 /*
1383 * If both edges are on a cluster boundary then there's no
1384 * zeroing required as the region is part of the allocation to
1385 * be truncated.
1386 */
1387 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1388 goto out;
1389
1390 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001391 if (IS_ERR(handle)) {
1392 ret = PTR_ERR(handle);
Mark Fasheh063c4562007-07-03 13:34:11 -07001393 mlog_errno(ret);
1394 goto out;
1395 }
1396
1397 /*
1398 * We want to get the byte offset of the end of the 1st cluster.
1399 */
1400 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1401 if (tmpend > end)
1402 tmpend = end;
1403
1404 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1405 (unsigned long long)start, (unsigned long long)tmpend);
1406
1407 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1408 if (ret)
1409 mlog_errno(ret);
1410
1411 if (tmpend < end) {
1412 /*
1413 * This may make start and end equal, but the zeroing
1414 * code will skip any work in that case so there's no
1415 * need to catch it up here.
1416 */
1417 start = end & ~(osb->s_clustersize - 1);
1418
1419 mlog(0, "2nd range: start: %llu, end: %llu\n",
1420 (unsigned long long)start, (unsigned long long)end);
1421
1422 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1423 if (ret)
1424 mlog_errno(ret);
1425 }
1426
1427 ocfs2_commit_trans(osb, handle);
1428out:
1429 return ret;
1430}
1431
1432static int ocfs2_remove_inode_range(struct inode *inode,
1433 struct buffer_head *di_bh, u64 byte_start,
1434 u64 byte_len)
1435{
1436 int ret = 0;
1437 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1438 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1439 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehb1967d02007-11-20 11:56:39 -08001440 struct address_space *mapping = inode->i_mapping;
Mark Fashehfecc0112008-11-12 15:16:38 -08001441 struct ocfs2_extent_tree et;
Mark Fasheh063c4562007-07-03 13:34:11 -07001442
Joel Becker5e404e92009-02-13 03:54:22 -08001443 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Mark Fasheh063c4562007-07-03 13:34:11 -07001444 ocfs2_init_dealloc_ctxt(&dealloc);
1445
1446 if (byte_len == 0)
1447 return 0;
1448
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001449 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1450 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
Mark Fashehb1967d02007-11-20 11:56:39 -08001451 byte_start + byte_len, 0);
1452 if (ret) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001453 mlog_errno(ret);
Mark Fashehb1967d02007-11-20 11:56:39 -08001454 goto out;
1455 }
1456 /*
1457 * There's no need to get fancy with the page cache
1458 * truncate of an inline-data inode. We're talking
1459 * about less than a page here, which will be cached
1460 * in the dinode buffer anyway.
1461 */
1462 unmap_mapping_range(mapping, 0, 0, 0);
1463 truncate_inode_pages(mapping, 0);
1464 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001465 }
1466
Mark Fasheh063c4562007-07-03 13:34:11 -07001467 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1468 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1469 if (trunc_len >= trunc_start)
1470 trunc_len -= trunc_start;
1471 else
1472 trunc_len = 0;
1473
1474 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1475 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1476 (unsigned long long)byte_start,
1477 (unsigned long long)byte_len, trunc_start, trunc_len);
1478
1479 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1480 if (ret) {
1481 mlog_errno(ret);
1482 goto out;
1483 }
1484
1485 cpos = trunc_start;
1486 while (trunc_len) {
1487 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1488 &alloc_size, NULL);
1489 if (ret) {
1490 mlog_errno(ret);
1491 goto out;
1492 }
1493
1494 if (alloc_size > trunc_len)
1495 alloc_size = trunc_len;
1496
1497 /* Only do work for non-holes */
1498 if (phys_cpos != 0) {
Mark Fashehfecc0112008-11-12 15:16:38 -08001499 ret = ocfs2_remove_btree_range(inode, &et, cpos,
1500 phys_cpos, alloc_size,
1501 &dealloc);
Mark Fasheh063c4562007-07-03 13:34:11 -07001502 if (ret) {
1503 mlog_errno(ret);
1504 goto out;
1505 }
1506 }
1507
1508 cpos += alloc_size;
1509 trunc_len -= alloc_size;
1510 }
1511
1512 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1513
1514out:
1515 ocfs2_schedule_truncate_log_flush(osb, 1);
1516 ocfs2_run_deallocs(osb, &dealloc);
1517
1518 return ret;
1519}
1520
Mark Fashehb2580102007-03-09 16:53:21 -08001521/*
1522 * Parts of this function taken from xfs_change_file_space()
1523 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001524static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1525 loff_t f_pos, unsigned int cmd,
1526 struct ocfs2_space_resv *sr,
1527 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001528{
1529 int ret;
1530 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001531 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001532 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1533 struct buffer_head *di_bh = NULL;
1534 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001535 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001536
Mark Fashehb2580102007-03-09 16:53:21 -08001537 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1538 return -EROFS;
1539
1540 mutex_lock(&inode->i_mutex);
1541
1542 /*
1543 * This prevents concurrent writes on other nodes
1544 */
1545 ret = ocfs2_rw_lock(inode, 1);
1546 if (ret) {
1547 mlog_errno(ret);
1548 goto out;
1549 }
1550
Mark Fashehe63aecb62007-10-18 15:30:42 -07001551 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001552 if (ret) {
1553 mlog_errno(ret);
1554 goto out_rw_unlock;
1555 }
1556
1557 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1558 ret = -EPERM;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001559 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001560 }
1561
1562 switch (sr->l_whence) {
1563 case 0: /*SEEK_SET*/
1564 break;
1565 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001566 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001567 break;
1568 case 2: /*SEEK_END*/
1569 sr->l_start += i_size_read(inode);
1570 break;
1571 default:
1572 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001573 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001574 }
1575 sr->l_whence = 0;
1576
1577 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1578
1579 if (sr->l_start < 0
1580 || sr->l_start > max_off
1581 || (sr->l_start + llen) < 0
1582 || (sr->l_start + llen) > max_off) {
1583 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001584 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001585 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001586 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001587
1588 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1589 if (sr->l_len <= 0) {
1590 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001591 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001592 }
1593 }
1594
Mark Fasheh385820a2007-07-19 00:14:38 -07001595 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001596 ret = __ocfs2_write_remove_suid(inode, di_bh);
1597 if (ret) {
1598 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001599 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001600 }
1601 }
1602
1603 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1604 switch (cmd) {
1605 case OCFS2_IOC_RESVSP:
1606 case OCFS2_IOC_RESVSP64:
1607 /*
1608 * This takes unsigned offsets, but the signed ones we
1609 * pass have been checked against overflow above.
1610 */
1611 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1612 sr->l_len);
1613 break;
1614 case OCFS2_IOC_UNRESVSP:
1615 case OCFS2_IOC_UNRESVSP64:
1616 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1617 sr->l_len);
1618 break;
1619 default:
1620 ret = -EINVAL;
1621 }
1622 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1623 if (ret) {
1624 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001625 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001626 }
1627
1628 /*
1629 * We update c/mtime for these changes
1630 */
1631 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1632 if (IS_ERR(handle)) {
1633 ret = PTR_ERR(handle);
1634 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001635 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001636 }
1637
Mark Fasheh385820a2007-07-19 00:14:38 -07001638 if (change_size && i_size_read(inode) < size)
1639 i_size_write(inode, size);
1640
Mark Fashehb2580102007-03-09 16:53:21 -08001641 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1642 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1643 if (ret < 0)
1644 mlog_errno(ret);
1645
1646 ocfs2_commit_trans(osb, handle);
1647
Mark Fashehe63aecb62007-10-18 15:30:42 -07001648out_inode_unlock:
Mark Fashehb2580102007-03-09 16:53:21 -08001649 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001650 ocfs2_inode_unlock(inode, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001651out_rw_unlock:
1652 ocfs2_rw_unlock(inode, 1);
1653
Mark Fashehb2580102007-03-09 16:53:21 -08001654out:
Julia Lawallc259ae52008-07-21 09:59:15 +02001655 mutex_unlock(&inode->i_mutex);
Mark Fashehb2580102007-03-09 16:53:21 -08001656 return ret;
1657}
1658
Mark Fasheh385820a2007-07-19 00:14:38 -07001659int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1660 struct ocfs2_space_resv *sr)
1661{
1662 struct inode *inode = file->f_path.dentry->d_inode;
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08001663 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh385820a2007-07-19 00:14:38 -07001664
1665 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1666 !ocfs2_writes_unwritten_extents(osb))
1667 return -ENOTTY;
1668 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1669 !ocfs2_sparse_alloc(osb))
1670 return -ENOTTY;
1671
1672 if (!S_ISREG(inode->i_mode))
1673 return -EINVAL;
1674
1675 if (!(file->f_mode & FMODE_WRITE))
1676 return -EBADF;
1677
1678 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1679}
1680
1681static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1682 loff_t len)
1683{
1684 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1685 struct ocfs2_space_resv sr;
1686 int change_size = 1;
1687
1688 if (!ocfs2_writes_unwritten_extents(osb))
1689 return -EOPNOTSUPP;
1690
1691 if (S_ISDIR(inode->i_mode))
1692 return -ENODEV;
1693
1694 if (mode & FALLOC_FL_KEEP_SIZE)
1695 change_size = 0;
1696
1697 sr.l_whence = 0;
1698 sr.l_start = (s64)offset;
1699 sr.l_len = (s64)len;
1700
1701 return __ocfs2_change_file_space(NULL, inode, offset,
1702 OCFS2_IOC_RESVSP64, &sr, change_size);
1703}
1704
Tao Ma293b2f72009-08-25 08:02:48 +08001705int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
1706 size_t count)
1707{
1708 int ret = 0;
1709 unsigned int extent_flags;
1710 u32 cpos, clusters, extent_len, phys_cpos;
1711 struct super_block *sb = inode->i_sb;
1712
1713 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
Tao Ma2f48d5932009-10-15 11:10:49 +08001714 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
1715 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Tao Ma293b2f72009-08-25 08:02:48 +08001716 return 0;
1717
1718 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1719 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1720
1721 while (clusters) {
1722 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1723 &extent_flags);
1724 if (ret < 0) {
1725 mlog_errno(ret);
1726 goto out;
1727 }
1728
1729 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
1730 ret = 1;
1731 break;
1732 }
1733
1734 if (extent_len > clusters)
1735 extent_len = clusters;
1736
1737 clusters -= extent_len;
1738 cpos += extent_len;
1739 }
1740out:
1741 return ret;
1742}
1743
1744static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
1745 loff_t pos, size_t count,
1746 int *meta_level)
1747{
1748 int ret;
1749 struct buffer_head *di_bh = NULL;
1750 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1751 u32 clusters =
1752 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
1753
1754 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1755 if (ret) {
1756 mlog_errno(ret);
1757 goto out;
1758 }
1759
1760 *meta_level = 1;
1761
Tao Ma37f8a2b2009-08-26 09:47:28 +08001762 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
Tao Ma293b2f72009-08-25 08:02:48 +08001763 if (ret)
1764 mlog_errno(ret);
1765out:
1766 brelse(di_bh);
1767 return ret;
1768}
1769
Tiger Yang8659ac22006-10-17 18:29:52 -07001770static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1771 loff_t *ppos,
1772 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001773 int appending,
Tao Ma86470e92009-12-03 21:55:05 +08001774 int *direct_io,
1775 int *has_refcount)
Mark Fashehccd979b2005-12-15 14:31:24 -08001776{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001777 int ret = 0, meta_level = 0;
Tiger Yang8659ac22006-10-17 18:29:52 -07001778 struct inode *inode = dentry->d_inode;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001779 loff_t saved_pos, end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001780
Sunil Mushran2bd63212010-01-25 16:57:38 -08001781 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001782 * We start with a read level meta lock and only jump to an ex
1783 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08001784 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001785 for(;;) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001786 ret = ocfs2_inode_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001787 if (ret < 0) {
1788 meta_level = -1;
1789 mlog_errno(ret);
1790 goto out;
1791 }
1792
1793 /* Clear suid / sgid if necessary. We do this here
1794 * instead of later in the write path because
1795 * remove_suid() calls ->setattr without any hint that
1796 * we may have already done our cluster locking. Since
1797 * ocfs2_setattr() *must* take cluster locks to
1798 * proceeed, this will lead us to recursively lock the
1799 * inode. There's also the dinode i_size state which
1800 * can be lost via setattr during extending writes (we
1801 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001802 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001803 if (meta_level == 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001804 ocfs2_inode_unlock(inode, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001805 meta_level = 1;
1806 continue;
1807 }
1808
1809 ret = ocfs2_write_remove_suid(inode);
1810 if (ret < 0) {
1811 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001812 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001813 }
1814 }
1815
1816 /* work on a copy of ppos until we're sure that we won't have
1817 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001818 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001819 saved_pos = i_size_read(inode);
1820 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1821 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001822 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001823 }
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001824
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001825 end = saved_pos + count;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001826
Tao Ma293b2f72009-08-25 08:02:48 +08001827 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
1828 if (ret == 1) {
1829 ocfs2_inode_unlock(inode, meta_level);
1830 meta_level = -1;
1831
1832 ret = ocfs2_prepare_inode_for_refcount(inode,
1833 saved_pos,
1834 count,
1835 &meta_level);
Tao Ma86470e92009-12-03 21:55:05 +08001836 if (has_refcount)
1837 *has_refcount = 1;
Tao Ma293b2f72009-08-25 08:02:48 +08001838 }
1839
1840 if (ret < 0) {
1841 mlog_errno(ret);
1842 goto out_unlock;
1843 }
1844
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001845 /*
1846 * Skip the O_DIRECT checks if we don't need
1847 * them.
1848 */
1849 if (!direct_io || !(*direct_io))
1850 break;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001851
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001852 /*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001853 * There's no sane way to do direct writes to an inode
1854 * with inline data.
1855 */
1856 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1857 *direct_io = 0;
1858 break;
1859 }
1860
Tao Ma86470e92009-12-03 21:55:05 +08001861 if (has_refcount && *has_refcount == 1) {
1862 *direct_io = 0;
1863 break;
1864 }
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001865 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001866 * Allowing concurrent direct writes means
1867 * i_size changes wouldn't be synchronized, so
1868 * one node could wind up truncating another
1869 * nodes writes.
1870 */
1871 if (end > i_size_read(inode)) {
1872 *direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001873 break;
1874 }
1875
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001876 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001877 * We don't fill holes during direct io, so
1878 * check for them here. If any are found, the
1879 * caller will have to retake some cluster
1880 * locks and initiate the io as buffered.
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001881 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001882 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1883 if (ret == 1) {
1884 *direct_io = 0;
1885 ret = 0;
1886 } else if (ret < 0)
1887 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001888 break;
1889 }
1890
Tiger Yang8659ac22006-10-17 18:29:52 -07001891 if (appending)
1892 *ppos = saved_pos;
1893
1894out_unlock:
Tao Ma293b2f72009-08-25 08:02:48 +08001895 if (meta_level >= 0)
1896 ocfs2_inode_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001897
1898out:
1899 return ret;
1900}
1901
1902static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1903 const struct iovec *iov,
1904 unsigned long nr_segs,
1905 loff_t pos)
1906{
Mark Fasheh9517bac2007-02-09 20:24:12 -08001907 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
Tao Ma86470e92009-12-03 21:55:05 +08001908 int can_do_direct, has_refcount = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001909 ssize_t written = 0;
1910 size_t ocount; /* original count */
1911 size_t count; /* after file limit checks */
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001912 loff_t old_size, *ppos = &iocb->ki_pos;
1913 u32 old_clusters;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001914 struct file *file = iocb->ki_filp;
1915 struct inode *inode = file->f_path.dentry->d_inode;
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001916 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang8659ac22006-10-17 18:29:52 -07001917
Mark Fasheh9517bac2007-02-09 20:24:12 -08001918 mlog_entry("(0x%p, %u, '%.*s')\n", file,
Tiger Yang8659ac22006-10-17 18:29:52 -07001919 (unsigned int)nr_segs,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001920 file->f_path.dentry->d_name.len,
1921 file->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001922
Tiger Yang8659ac22006-10-17 18:29:52 -07001923 if (iocb->ki_left == 0)
1924 return 0;
1925
Mark Fasheh9517bac2007-02-09 20:24:12 -08001926 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1927
1928 appending = file->f_flags & O_APPEND ? 1 : 0;
1929 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1930
Tiger Yang8659ac22006-10-17 18:29:52 -07001931 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001932
1933relock:
Tiger Yang8659ac22006-10-17 18:29:52 -07001934 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001935 if (direct_io) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001936 down_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001937 have_alloc_sem = 1;
Tiger Yang8659ac22006-10-17 18:29:52 -07001938 }
1939
1940 /* concurrent O_DIRECT writes are allowed */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001941 rw_level = !direct_io;
Tiger Yang8659ac22006-10-17 18:29:52 -07001942 ret = ocfs2_rw_lock(inode, rw_level);
1943 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001944 mlog_errno(ret);
1945 goto out_sems;
1946 }
1947
1948 can_do_direct = direct_io;
1949 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1950 iocb->ki_left, appending,
Tao Ma86470e92009-12-03 21:55:05 +08001951 &can_do_direct, &has_refcount);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001952 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001953 mlog_errno(ret);
1954 goto out;
1955 }
1956
Mark Fasheh9517bac2007-02-09 20:24:12 -08001957 /*
1958 * We can't complete the direct I/O as requested, fall back to
1959 * buffered I/O.
1960 */
1961 if (direct_io && !can_do_direct) {
1962 ocfs2_rw_unlock(inode, rw_level);
1963 up_read(&inode->i_alloc_sem);
1964
1965 have_alloc_sem = 0;
1966 rw_level = -1;
1967
1968 direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001969 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07001970 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001971
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001972 /*
1973 * To later detect whether a journal commit for sync writes is
1974 * necessary, we sample i_size, and cluster count here.
1975 */
1976 old_size = i_size_read(inode);
1977 old_clusters = OCFS2_I(inode)->ip_clusters;
1978
Mark Fashehccd979b2005-12-15 14:31:24 -08001979 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07001980 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001981
Mark Fasheh9517bac2007-02-09 20:24:12 -08001982 if (direct_io) {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001983 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1984 VERIFY_READ);
1985 if (ret)
1986 goto out_dio;
1987
Goldwyn Rodriguescefcb802009-07-11 10:57:27 -05001988 count = ocount;
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001989 ret = generic_write_checks(file, ppos, &count,
1990 S_ISBLK(inode->i_mode));
1991 if (ret)
1992 goto out_dio;
1993
Mark Fasheh9517bac2007-02-09 20:24:12 -08001994 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1995 ppos, count, ocount);
1996 if (written < 0) {
Dmitri Monakhovc4354002008-10-27 13:01:49 -07001997 /*
1998 * direct write may have instantiated a few
1999 * blocks outside i_size. Trim these off again.
2000 * Don't need i_size_read because we hold i_mutex.
2001 */
2002 if (*ppos + count > inode->i_size)
2003 vmtruncate(inode, inode->i_size);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002004 ret = written;
2005 goto out_dio;
2006 }
2007 } else {
Jan Kara918941a2009-08-17 18:50:08 +02002008 written = __generic_file_aio_write(iocb, iov, nr_segs, ppos);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002009 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002010
Mark Fasheh9517bac2007-02-09 20:24:12 -08002011out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08002012 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08002013 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08002014
Tao Ma60c48672010-02-03 09:56:04 +08002015 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2016 ((file->f_flags & O_DIRECT) && has_refcount)) {
Jan Kara918941a2009-08-17 18:50:08 +02002017 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2018 pos + count - 1);
2019 if (ret < 0)
2020 written = ret;
2021
2022 if (!ret && (old_size != i_size_read(inode) ||
Tao Ma86470e92009-12-03 21:55:05 +08002023 old_clusters != OCFS2_I(inode)->ip_clusters ||
2024 has_refcount)) {
Joel Becker2b4e30f2008-09-03 20:03:41 -07002025 ret = jbd2_journal_force_commit(osb->journal->j_journal);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002026 if (ret < 0)
2027 written = ret;
2028 }
Jan Kara918941a2009-08-17 18:50:08 +02002029
2030 if (!ret)
2031 ret = filemap_fdatawait_range(file->f_mapping, pos,
2032 pos + count - 1);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002033 }
2034
Sunil Mushran2bd63212010-01-25 16:57:38 -08002035 /*
Mark Fashehccd979b2005-12-15 14:31:24 -08002036 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2037 * function pointer which is called when o_direct io completes so that
2038 * it can unlock our rw lock. (it's the clustered equivalent of
2039 * i_alloc_sem; protects truncate from racing with pending ios).
2040 * Unfortunately there are error cases which call end_io and others
2041 * that don't. so we don't have to unlock the rw_lock if either an
2042 * async dio is going to do it in the future or an end_io after an
2043 * error has already done it.
2044 */
2045 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2046 rw_level = -1;
2047 have_alloc_sem = 0;
2048 }
2049
2050out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08002051 if (rw_level != -1)
2052 ocfs2_rw_unlock(inode, rw_level);
2053
2054out_sems:
Mark Fashehccd979b2005-12-15 14:31:24 -08002055 if (have_alloc_sem)
2056 up_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002057
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002058 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002059
Wengang Wang812e7a62009-07-10 13:26:04 +08002060 if (written)
2061 ret = written;
Mark Fashehccd979b2005-12-15 14:31:24 -08002062 mlog_exit(ret);
Wengang Wang812e7a62009-07-10 13:26:04 +08002063 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002064}
2065
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002066static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2067 struct file *out,
2068 struct splice_desc *sd)
2069{
2070 int ret;
2071
2072 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos,
Tao Ma86470e92009-12-03 21:55:05 +08002073 sd->total_len, 0, NULL, NULL);
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002074 if (ret < 0) {
2075 mlog_errno(ret);
2076 return ret;
2077 }
2078
2079 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2080}
2081
Tiger Yang8659ac22006-10-17 18:29:52 -07002082static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2083 struct file *out,
2084 loff_t *ppos,
2085 size_t len,
2086 unsigned int flags)
2087{
2088 int ret;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002089 struct address_space *mapping = out->f_mapping;
2090 struct inode *inode = mapping->host;
2091 struct splice_desc sd = {
2092 .total_len = len,
2093 .flags = flags,
2094 .pos = *ppos,
2095 .u.file = out,
2096 };
Tiger Yang8659ac22006-10-17 18:29:52 -07002097
2098 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2099 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002100 out->f_path.dentry->d_name.len,
2101 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002102
Miklos Szeredi7bfac9e2009-04-06 17:41:00 +02002103 if (pipe->inode)
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002104 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2105
2106 splice_from_pipe_begin(&sd);
2107 do {
2108 ret = splice_from_pipe_next(pipe, &sd);
2109 if (ret <= 0)
2110 break;
2111
2112 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2113 ret = ocfs2_rw_lock(inode, 1);
2114 if (ret < 0)
2115 mlog_errno(ret);
2116 else {
2117 ret = ocfs2_splice_to_file(pipe, out, &sd);
2118 ocfs2_rw_unlock(inode, 1);
2119 }
2120 mutex_unlock(&inode->i_mutex);
2121 } while (ret > 0);
2122 splice_from_pipe_end(pipe, &sd);
2123
Miklos Szeredi7bfac9e2009-04-06 17:41:00 +02002124 if (pipe->inode)
2125 mutex_unlock(&pipe->inode->i_mutex);
Tiger Yang8659ac22006-10-17 18:29:52 -07002126
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002127 if (sd.num_spliced)
2128 ret = sd.num_spliced;
2129
2130 if (ret > 0) {
2131 unsigned long nr_pages;
Jan Karad23c9372009-08-18 18:24:31 +02002132 int err;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002133
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002134 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2135
Jan Karad23c9372009-08-18 18:24:31 +02002136 err = generic_write_sync(out, *ppos, ret);
2137 if (err)
2138 ret = err;
2139 else
2140 *ppos += ret;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002141
Miklos Szeredi328eaab2009-04-14 19:48:39 +02002142 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2143 }
Tiger Yang8659ac22006-10-17 18:29:52 -07002144
2145 mlog_exit(ret);
2146 return ret;
2147}
2148
2149static ssize_t ocfs2_file_splice_read(struct file *in,
2150 loff_t *ppos,
2151 struct pipe_inode_info *pipe,
2152 size_t len,
2153 unsigned int flags)
2154{
Tao Ma1962f392009-06-19 15:36:52 +08002155 int ret = 0, lock_level = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08002156 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07002157
2158 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2159 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002160 in->f_path.dentry->d_name.len,
2161 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002162
2163 /*
2164 * See the comment in ocfs2_file_aio_read()
2165 */
Tao Ma1962f392009-06-19 15:36:52 +08002166 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002167 if (ret < 0) {
2168 mlog_errno(ret);
2169 goto bail;
2170 }
Tao Ma1962f392009-06-19 15:36:52 +08002171 ocfs2_inode_unlock(inode, lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002172
2173 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2174
2175bail:
2176 mlog_exit(ret);
2177 return ret;
2178}
2179
Mark Fashehccd979b2005-12-15 14:31:24 -08002180static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07002181 const struct iovec *iov,
2182 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08002183 loff_t pos)
2184{
Tiger Yang25899de2006-11-15 15:49:02 +08002185 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002186 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08002187 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08002188
Badari Pulavarty027445c2006-09-30 23:28:46 -07002189 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2190 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08002191 filp->f_path.dentry->d_name.len,
2192 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08002193
2194 if (!inode) {
2195 ret = -EINVAL;
2196 mlog_errno(ret);
2197 goto bail;
2198 }
2199
Sunil Mushran2bd63212010-01-25 16:57:38 -08002200 /*
Mark Fashehccd979b2005-12-15 14:31:24 -08002201 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2202 * need locks to protect pending reads from racing with truncate.
2203 */
2204 if (filp->f_flags & O_DIRECT) {
2205 down_read(&inode->i_alloc_sem);
2206 have_alloc_sem = 1;
2207
2208 ret = ocfs2_rw_lock(inode, 0);
2209 if (ret < 0) {
2210 mlog_errno(ret);
2211 goto bail;
2212 }
2213 rw_level = 0;
2214 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002215 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002216 }
2217
Mark Fashehc4374f82006-05-05 19:04:35 -07002218 /*
2219 * We're fine letting folks race truncates and extending
2220 * writes with read across the cluster, just like they can
2221 * locally. Hence no rw_lock during read.
Sunil Mushran2bd63212010-01-25 16:57:38 -08002222 *
Mark Fashehc4374f82006-05-05 19:04:35 -07002223 * Take and drop the meta data lock to update inode fields
2224 * like i_size. This allows the checks down below
Sunil Mushran2bd63212010-01-25 16:57:38 -08002225 * generic_file_aio_read() a chance of actually working.
Mark Fashehc4374f82006-05-05 19:04:35 -07002226 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002227 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002228 if (ret < 0) {
2229 mlog_errno(ret);
2230 goto bail;
2231 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07002232 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002233
Badari Pulavarty027445c2006-09-30 23:28:46 -07002234 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08002235 if (ret == -EINVAL)
Sunil Mushran56753bd2008-06-09 11:24:41 -07002236 mlog(0, "generic_file_aio_read returned -EINVAL\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08002237
2238 /* buffered aio wouldn't have proper lock coverage today */
2239 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2240
2241 /* see ocfs2_file_aio_write */
2242 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2243 rw_level = -1;
2244 have_alloc_sem = 0;
2245 }
2246
2247bail:
2248 if (have_alloc_sem)
2249 up_read(&inode->i_alloc_sem);
Sunil Mushran2bd63212010-01-25 16:57:38 -08002250 if (rw_level != -1)
Mark Fashehccd979b2005-12-15 14:31:24 -08002251 ocfs2_rw_unlock(inode, rw_level);
2252 mlog_exit(ret);
2253
2254 return ret;
2255}
2256
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002257const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002258 .setattr = ocfs2_setattr,
2259 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002260 .permission = ocfs2_permission,
Tiger Yangcf1d6c72008-08-18 17:11:00 +08002261 .setxattr = generic_setxattr,
2262 .getxattr = generic_getxattr,
2263 .listxattr = ocfs2_listxattr,
2264 .removexattr = generic_removexattr,
Mark Fasheh385820a2007-07-19 00:14:38 -07002265 .fallocate = ocfs2_fallocate,
Mark Fasheh00dc4172008-10-03 17:32:11 -04002266 .fiemap = ocfs2_fiemap,
Mark Fashehccd979b2005-12-15 14:31:24 -08002267};
2268
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002269const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002270 .setattr = ocfs2_setattr,
2271 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002272 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08002273};
2274
Mark Fasheh53da4932008-07-21 14:29:16 -07002275/*
2276 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2277 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2278 */
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002279const struct file_operations ocfs2_fops = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002280 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002281 .read = do_sync_read,
2282 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002283 .mmap = ocfs2_mmap,
2284 .fsync = ocfs2_sync_file,
2285 .release = ocfs2_file_release,
2286 .open = ocfs2_file_open,
2287 .aio_read = ocfs2_file_aio_read,
2288 .aio_write = ocfs2_file_aio_write,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002289 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002290#ifdef CONFIG_COMPAT
2291 .compat_ioctl = ocfs2_compat_ioctl,
2292#endif
Mark Fasheh53da4932008-07-21 14:29:16 -07002293 .lock = ocfs2_lock,
2294 .flock = ocfs2_flock,
2295 .splice_read = ocfs2_file_splice_read,
2296 .splice_write = ocfs2_file_splice_write,
2297};
2298
2299const struct file_operations ocfs2_dops = {
2300 .llseek = generic_file_llseek,
2301 .read = generic_read_dir,
2302 .readdir = ocfs2_readdir,
2303 .fsync = ocfs2_sync_file,
2304 .release = ocfs2_dir_release,
2305 .open = ocfs2_dir_open,
2306 .unlocked_ioctl = ocfs2_ioctl,
2307#ifdef CONFIG_COMPAT
2308 .compat_ioctl = ocfs2_compat_ioctl,
2309#endif
2310 .lock = ocfs2_lock,
2311 .flock = ocfs2_flock,
2312};
2313
2314/*
2315 * POSIX-lockless variants of our file_operations.
2316 *
2317 * These will be used if the underlying cluster stack does not support
2318 * posix file locking, if the user passes the "localflocks" mount
2319 * option, or if we have a local-only fs.
2320 *
2321 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2322 * so we still want it in the case of no stack support for
2323 * plocks. Internally, it will do the right thing when asked to ignore
2324 * the cluster.
2325 */
2326const struct file_operations ocfs2_fops_no_plocks = {
2327 .llseek = generic_file_llseek,
2328 .read = do_sync_read,
2329 .write = do_sync_write,
2330 .mmap = ocfs2_mmap,
2331 .fsync = ocfs2_sync_file,
2332 .release = ocfs2_file_release,
2333 .open = ocfs2_file_open,
2334 .aio_read = ocfs2_file_aio_read,
2335 .aio_write = ocfs2_file_aio_write,
2336 .unlocked_ioctl = ocfs2_ioctl,
2337#ifdef CONFIG_COMPAT
2338 .compat_ioctl = ocfs2_compat_ioctl,
2339#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002340 .flock = ocfs2_flock,
Tiger Yang8659ac22006-10-17 18:29:52 -07002341 .splice_read = ocfs2_file_splice_read,
2342 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002343};
2344
Mark Fasheh53da4932008-07-21 14:29:16 -07002345const struct file_operations ocfs2_dops_no_plocks = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002346 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002347 .read = generic_read_dir,
2348 .readdir = ocfs2_readdir,
2349 .fsync = ocfs2_sync_file,
Mark Fasheh53fc6222007-12-20 16:49:04 -08002350 .release = ocfs2_dir_release,
2351 .open = ocfs2_dir_open,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002352 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002353#ifdef CONFIG_COMPAT
2354 .compat_ioctl = ocfs2_compat_ioctl,
2355#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002356 .flock = ocfs2_flock,
Mark Fashehccd979b2005-12-15 14:31:24 -08002357};