]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/ocfs2/file.c
ocfs2: Sync ocfs2_fs.h with ocfs2-tools
[net-next-2.6.git] / fs / ocfs2 / file.c
CommitLineData
ccd979bd
MF
1/* -*- 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
16f7e0fe 26#include <linux/capability.h>
ccd979bd
MF
27#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>
e2057c5a 33#include <linux/sched.h>
d6b29d7c 34#include <linux/splice.h>
7f1a37e3 35#include <linux/mount.h>
9517bac6 36#include <linux/writeback.h>
385820a3 37#include <linux/falloc.h>
ccd979bd
MF
38
39#define MLOG_MASK_PREFIX ML_INODE
40#include <cluster/masklog.h>
41
42#include "ocfs2.h"
43
44#include "alloc.h"
45#include "aops.h"
46#include "dir.h"
47#include "dlmglue.h"
48#include "extent_map.h"
49#include "file.h"
50#include "sysfile.h"
51#include "inode.h"
ca4d147e 52#include "ioctl.h"
ccd979bd
MF
53#include "journal.h"
54#include "mmap.h"
55#include "suballoc.h"
56#include "super.h"
57
58#include "buffer_head_io.h"
59
60static int ocfs2_sync_inode(struct inode *inode)
61{
62 filemap_fdatawrite(inode->i_mapping);
63 return sync_mapping_buffers(inode->i_mapping);
64}
65
66static int ocfs2_file_open(struct inode *inode, struct file *file)
67{
68 int status;
69 int mode = file->f_flags;
70 struct ocfs2_inode_info *oi = OCFS2_I(inode);
71
72 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
d28c9174 73 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
ccd979bd
MF
74
75 spin_lock(&oi->ip_lock);
76
77 /* Check that the inode hasn't been wiped from disk by another
78 * node. If it hasn't then we're safe as long as we hold the
79 * spin lock until our increment of open count. */
80 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
81 spin_unlock(&oi->ip_lock);
82
83 status = -ENOENT;
84 goto leave;
85 }
86
87 if (mode & O_DIRECT)
88 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
89
90 oi->ip_open_count++;
91 spin_unlock(&oi->ip_lock);
92 status = 0;
93leave:
94 mlog_exit(status);
95 return status;
96}
97
98static int ocfs2_file_release(struct inode *inode, struct file *file)
99{
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
101
102 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
d28c9174
JS
103 file->f_path.dentry->d_name.len,
104 file->f_path.dentry->d_name.name);
ccd979bd
MF
105
106 spin_lock(&oi->ip_lock);
107 if (!--oi->ip_open_count)
108 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
109 spin_unlock(&oi->ip_lock);
110
111 mlog_exit(0);
112
113 return 0;
114}
115
116static int ocfs2_sync_file(struct file *file,
117 struct dentry *dentry,
118 int datasync)
119{
120 int err = 0;
121 journal_t *journal;
122 struct inode *inode = dentry->d_inode;
123 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
124
125 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
126 dentry->d_name.len, dentry->d_name.name);
127
128 err = ocfs2_sync_inode(dentry->d_inode);
129 if (err)
130 goto bail;
131
132 journal = osb->journal->j_journal;
133 err = journal_force_commit(journal);
134
135bail:
136 mlog_exit(err);
137
138 return (err < 0) ? -EIO : 0;
139}
140
7f1a37e3
TY
141int ocfs2_should_update_atime(struct inode *inode,
142 struct vfsmount *vfsmnt)
143{
144 struct timespec now;
145 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
146
147 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
148 return 0;
149
150 if ((inode->i_flags & S_NOATIME) ||
151 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
152 return 0;
153
6c2aad05
MF
154 /*
155 * We can be called with no vfsmnt structure - NFSD will
156 * sometimes do this.
157 *
158 * Note that our action here is different than touch_atime() -
159 * if we can't tell whether this is a noatime mount, then we
160 * don't know whether to trust the value of s_atime_quantum.
161 */
162 if (vfsmnt == NULL)
163 return 0;
164
7f1a37e3
TY
165 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
166 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
167 return 0;
168
7e913c53
MF
169 if (vfsmnt->mnt_flags & MNT_RELATIME) {
170 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
171 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
172 return 1;
173
174 return 0;
175 }
176
7f1a37e3
TY
177 now = CURRENT_TIME;
178 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
179 return 0;
180 else
181 return 1;
182}
183
184int ocfs2_update_inode_atime(struct inode *inode,
185 struct buffer_head *bh)
186{
187 int ret;
188 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
189 handle_t *handle;
c11e9faf 190 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
7f1a37e3
TY
191
192 mlog_entry_void();
193
194 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
195 if (handle == NULL) {
196 ret = -ENOMEM;
197 mlog_errno(ret);
198 goto out;
199 }
200
c11e9faf
MF
201 ret = ocfs2_journal_access(handle, inode, bh,
202 OCFS2_JOURNAL_ACCESS_WRITE);
203 if (ret) {
204 mlog_errno(ret);
205 goto out_commit;
206 }
207
208 /*
209 * Don't use ocfs2_mark_inode_dirty() here as we don't always
210 * have i_mutex to guard against concurrent changes to other
211 * inode fields.
212 */
7f1a37e3 213 inode->i_atime = CURRENT_TIME;
c11e9faf
MF
214 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
215 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
216
217 ret = ocfs2_journal_dirty(handle, bh);
7f1a37e3
TY
218 if (ret < 0)
219 mlog_errno(ret);
220
c11e9faf 221out_commit:
7f1a37e3
TY
222 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
223out:
224 mlog_exit(ret);
225 return ret;
226}
227
6cb129f5
AB
228static int ocfs2_set_inode_size(handle_t *handle,
229 struct inode *inode,
230 struct buffer_head *fe_bh,
231 u64 new_i_size)
ccd979bd
MF
232{
233 int status;
234
235 mlog_entry_void();
236 i_size_write(inode, new_i_size);
8110b073 237 inode->i_blocks = ocfs2_inode_sector_count(inode);
ccd979bd
MF
238 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
239
240 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
241 if (status < 0) {
242 mlog_errno(status);
243 goto bail;
244 }
245
246bail:
247 mlog_exit(status);
248 return status;
249}
250
251static int ocfs2_simple_size_update(struct inode *inode,
252 struct buffer_head *di_bh,
253 u64 new_i_size)
254{
255 int ret;
256 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1fabe148 257 handle_t *handle = NULL;
ccd979bd 258
65eff9cc 259 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
ccd979bd
MF
260 if (handle == NULL) {
261 ret = -ENOMEM;
262 mlog_errno(ret);
263 goto out;
264 }
265
266 ret = ocfs2_set_inode_size(handle, inode, di_bh,
267 new_i_size);
268 if (ret < 0)
269 mlog_errno(ret);
270
02dc1af4 271 ocfs2_commit_trans(osb, handle);
ccd979bd
MF
272out:
273 return ret;
274}
275
276static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
277 struct inode *inode,
278 struct buffer_head *fe_bh,
279 u64 new_i_size)
280{
281 int status;
1fabe148 282 handle_t *handle;
60b11392 283 struct ocfs2_dinode *di;
35edec1d 284 u64 cluster_bytes;
ccd979bd
MF
285
286 mlog_entry_void();
287
288 /* TODO: This needs to actually orphan the inode in this
289 * transaction. */
290
65eff9cc 291 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
ccd979bd
MF
292 if (IS_ERR(handle)) {
293 status = PTR_ERR(handle);
294 mlog_errno(status);
295 goto out;
296 }
297
60b11392
MF
298 status = ocfs2_journal_access(handle, inode, fe_bh,
299 OCFS2_JOURNAL_ACCESS_WRITE);
300 if (status < 0) {
301 mlog_errno(status);
302 goto out_commit;
303 }
304
305 /*
306 * Do this before setting i_size.
307 */
35edec1d
MF
308 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
309 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
310 cluster_bytes);
60b11392
MF
311 if (status) {
312 mlog_errno(status);
313 goto out_commit;
314 }
315
316 i_size_write(inode, new_i_size);
60b11392
MF
317 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
318
319 di = (struct ocfs2_dinode *) fe_bh->b_data;
320 di->i_size = cpu_to_le64(new_i_size);
321 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
322 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
323
324 status = ocfs2_journal_dirty(handle, fe_bh);
ccd979bd
MF
325 if (status < 0)
326 mlog_errno(status);
327
60b11392 328out_commit:
02dc1af4 329 ocfs2_commit_trans(osb, handle);
ccd979bd 330out:
60b11392 331
ccd979bd
MF
332 mlog_exit(status);
333 return status;
334}
335
336static int ocfs2_truncate_file(struct inode *inode,
337 struct buffer_head *di_bh,
338 u64 new_i_size)
339{
340 int status = 0;
341 struct ocfs2_dinode *fe = NULL;
342 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
343 struct ocfs2_truncate_context *tc = NULL;
344
b0697053
MF
345 mlog_entry("(inode = %llu, new_i_size = %llu\n",
346 (unsigned long long)OCFS2_I(inode)->ip_blkno,
347 (unsigned long long)new_i_size);
ccd979bd 348
ccd979bd
MF
349 fe = (struct ocfs2_dinode *) di_bh->b_data;
350 if (!OCFS2_IS_VALID_DINODE(fe)) {
351 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
352 status = -EIO;
353 goto bail;
354 }
355
356 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
b0697053
MF
357 "Inode %llu, inode i_size = %lld != di "
358 "i_size = %llu, i_flags = 0x%x\n",
359 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd 360 i_size_read(inode),
b0697053
MF
361 (unsigned long long)le64_to_cpu(fe->i_size),
362 le32_to_cpu(fe->i_flags));
ccd979bd
MF
363
364 if (new_i_size > le64_to_cpu(fe->i_size)) {
b0697053
MF
365 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
366 (unsigned long long)le64_to_cpu(fe->i_size),
367 (unsigned long long)new_i_size);
ccd979bd
MF
368 status = -EINVAL;
369 mlog_errno(status);
370 goto bail;
371 }
372
b0697053
MF
373 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
374 (unsigned long long)le64_to_cpu(fe->i_blkno),
375 (unsigned long long)le64_to_cpu(fe->i_size),
376 (unsigned long long)new_i_size);
ccd979bd
MF
377
378 /* lets handle the simple truncate cases before doing any more
379 * cluster locking. */
380 if (new_i_size == le64_to_cpu(fe->i_size))
381 goto bail;
382
2e89b2e4
MF
383 down_write(&OCFS2_I(inode)->ip_alloc_sem);
384
ab0920ce
MF
385 /* This forces other nodes to sync and drop their pages. Do
386 * this even if we have a truncate without allocation change -
387 * ocfs2 cluster sizes can be much greater than page size, so
388 * we have to truncate them anyway. */
389 status = ocfs2_data_lock(inode, 1);
390 if (status < 0) {
2e89b2e4
MF
391 up_write(&OCFS2_I(inode)->ip_alloc_sem);
392
ab0920ce
MF
393 mlog_errno(status);
394 goto bail;
395 }
ab0920ce 396
2e89b2e4
MF
397 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
398 truncate_inode_pages(inode->i_mapping, new_i_size);
399
ccd979bd
MF
400 /* alright, we're going to need to do a full blown alloc size
401 * change. Orphan the inode so that recovery can complete the
402 * truncate if necessary. This does the task of marking
403 * i_size. */
404 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
405 if (status < 0) {
406 mlog_errno(status);
60b11392 407 goto bail_unlock_data;
ccd979bd
MF
408 }
409
410 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
411 if (status < 0) {
412 mlog_errno(status);
60b11392 413 goto bail_unlock_data;
ccd979bd
MF
414 }
415
416 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
417 if (status < 0) {
418 mlog_errno(status);
60b11392 419 goto bail_unlock_data;
ccd979bd
MF
420 }
421
422 /* TODO: orphan dir cleanup here. */
60b11392
MF
423bail_unlock_data:
424 ocfs2_data_unlock(inode, 1);
425
2e89b2e4
MF
426 up_write(&OCFS2_I(inode)->ip_alloc_sem);
427
ccd979bd
MF
428bail:
429
430 mlog_exit(status);
431 return status;
432}
433
434/*
435 * extend allocation only here.
436 * we'll update all the disk stuff, and oip->alloc_size
437 *
438 * expect stuff to be locked, a transaction started and enough data /
439 * metadata reservations in the contexts.
440 *
441 * Will return -EAGAIN, and a reason if a restart is needed.
442 * If passed in, *reason will always be set, even in error.
443 */
444int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
445 struct inode *inode,
dcd0538f 446 u32 *logical_offset,
ccd979bd 447 u32 clusters_to_add,
2ae99a60 448 int mark_unwritten,
ccd979bd 449 struct buffer_head *fe_bh,
1fabe148 450 handle_t *handle,
ccd979bd
MF
451 struct ocfs2_alloc_context *data_ac,
452 struct ocfs2_alloc_context *meta_ac,
453 enum ocfs2_alloc_restarted *reason_ret)
454{
455 int status = 0;
456 int free_extents;
457 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
458 enum ocfs2_alloc_restarted reason = RESTART_NONE;
459 u32 bit_off, num_bits;
460 u64 block;
2ae99a60 461 u8 flags = 0;
ccd979bd
MF
462
463 BUG_ON(!clusters_to_add);
464
2ae99a60
MF
465 if (mark_unwritten)
466 flags = OCFS2_EXT_UNWRITTEN;
467
ccd979bd
MF
468 free_extents = ocfs2_num_free_extents(osb, inode, fe);
469 if (free_extents < 0) {
470 status = free_extents;
471 mlog_errno(status);
472 goto leave;
473 }
474
475 /* there are two cases which could cause us to EAGAIN in the
476 * we-need-more-metadata case:
477 * 1) we haven't reserved *any*
478 * 2) we are so fragmented, we've needed to add metadata too
479 * many times. */
480 if (!free_extents && !meta_ac) {
481 mlog(0, "we haven't reserved any metadata!\n");
482 status = -EAGAIN;
483 reason = RESTART_META;
484 goto leave;
485 } else if ((!free_extents)
486 && (ocfs2_alloc_context_bits_left(meta_ac)
487 < ocfs2_extend_meta_needed(fe))) {
488 mlog(0, "filesystem is really fragmented...\n");
489 status = -EAGAIN;
490 reason = RESTART_META;
491 goto leave;
492 }
493
415cb800
MF
494 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
495 clusters_to_add, &bit_off, &num_bits);
ccd979bd
MF
496 if (status < 0) {
497 if (status != -ENOSPC)
498 mlog_errno(status);
499 goto leave;
500 }
501
502 BUG_ON(num_bits > clusters_to_add);
503
504 /* reserve our write early -- insert_extent may update the inode */
505 status = ocfs2_journal_access(handle, inode, fe_bh,
506 OCFS2_JOURNAL_ACCESS_WRITE);
507 if (status < 0) {
508 mlog_errno(status);
509 goto leave;
510 }
511
512 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
b0697053
MF
513 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
514 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
dcd0538f
MF
515 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
516 *logical_offset, block, num_bits,
2ae99a60 517 flags, meta_ac);
ccd979bd
MF
518 if (status < 0) {
519 mlog_errno(status);
520 goto leave;
521 }
522
ccd979bd
MF
523 status = ocfs2_journal_dirty(handle, fe_bh);
524 if (status < 0) {
525 mlog_errno(status);
526 goto leave;
527 }
528
529 clusters_to_add -= num_bits;
dcd0538f 530 *logical_offset += num_bits;
ccd979bd
MF
531
532 if (clusters_to_add) {
533 mlog(0, "need to alloc once more, clusters = %u, wanted = "
534 "%u\n", fe->i_clusters, clusters_to_add);
535 status = -EAGAIN;
536 reason = RESTART_TRANS;
537 }
538
539leave:
540 mlog_exit(status);
541 if (reason_ret)
542 *reason_ret = reason;
543 return status;
544}
545
abf8b156
MF
546/*
547 * For a given allocation, determine which allocators will need to be
548 * accessed, and lock them, reserving the appropriate number of bits.
549 *
2ae99a60
MF
550 * Sparse file systems call this from ocfs2_write_begin_nolock()
551 * and ocfs2_allocate_unwritten_extents().
552 *
553 * File systems which don't support holes call this from
554 * ocfs2_extend_allocation().
abf8b156 555 */
9517bac6 556int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
b27b7cbc 557 u32 clusters_to_add, u32 extents_to_split,
9517bac6
MF
558 struct ocfs2_alloc_context **data_ac,
559 struct ocfs2_alloc_context **meta_ac)
abf8b156 560{
063c4561 561 int ret = 0, num_free_extents;
b27b7cbc 562 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
abf8b156
MF
563 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
564
565 *meta_ac = NULL;
063c4561
MF
566 if (data_ac)
567 *data_ac = NULL;
568
569 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
abf8b156
MF
570
571 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
b27b7cbc 572 "clusters_to_add = %u, extents_to_split = %u\n",
abf8b156 573 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
b27b7cbc 574 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
abf8b156
MF
575
576 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
577 if (num_free_extents < 0) {
578 ret = num_free_extents;
579 mlog_errno(ret);
580 goto out;
581 }
582
583 /*
584 * Sparse allocation file systems need to be more conservative
585 * with reserving room for expansion - the actual allocation
586 * happens while we've got a journal handle open so re-taking
587 * a cluster lock (because we ran out of room for another
588 * extent) will violate ordering rules.
589 *
9517bac6 590 * Most of the time we'll only be seeing this 1 cluster at a time
abf8b156 591 * anyway.
b27b7cbc
MF
592 *
593 * Always lock for any unwritten extents - we might want to
594 * add blocks during a split.
abf8b156
MF
595 */
596 if (!num_free_extents ||
b27b7cbc 597 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
abf8b156
MF
598 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
599 if (ret < 0) {
600 if (ret != -ENOSPC)
601 mlog_errno(ret);
602 goto out;
603 }
604 }
605
063c4561
MF
606 if (clusters_to_add == 0)
607 goto out;
608
abf8b156
MF
609 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
610 if (ret < 0) {
611 if (ret != -ENOSPC)
612 mlog_errno(ret);
613 goto out;
614 }
615
616out:
617 if (ret) {
618 if (*meta_ac) {
619 ocfs2_free_alloc_context(*meta_ac);
620 *meta_ac = NULL;
621 }
622
623 /*
624 * We cannot have an error and a non null *data_ac.
625 */
626 }
627
628 return ret;
629}
630
2ae99a60
MF
631static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
632 u32 clusters_to_add, int mark_unwritten)
ccd979bd
MF
633{
634 int status = 0;
635 int restart_func = 0;
abf8b156 636 int credits;
2ae99a60 637 u32 prev_clusters;
ccd979bd
MF
638 struct buffer_head *bh = NULL;
639 struct ocfs2_dinode *fe = NULL;
1fabe148 640 handle_t *handle = NULL;
ccd979bd
MF
641 struct ocfs2_alloc_context *data_ac = NULL;
642 struct ocfs2_alloc_context *meta_ac = NULL;
643 enum ocfs2_alloc_restarted why;
644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
645
646 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
647
dcd0538f
MF
648 /*
649 * This function only exists for file systems which don't
650 * support holes.
651 */
2ae99a60 652 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
dcd0538f 653
ccd979bd
MF
654 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
655 OCFS2_BH_CACHED, inode);
656 if (status < 0) {
657 mlog_errno(status);
658 goto leave;
659 }
660
661 fe = (struct ocfs2_dinode *) bh->b_data;
662 if (!OCFS2_IS_VALID_DINODE(fe)) {
663 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
664 status = -EIO;
665 goto leave;
666 }
667
668restart_all:
669 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
670
b27b7cbc 671 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
9517bac6
MF
672 &meta_ac);
673 if (status) {
674 mlog_errno(status);
675 goto leave;
676 }
677
ccd979bd 678 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
65eff9cc 679 handle = ocfs2_start_trans(osb, credits);
ccd979bd
MF
680 if (IS_ERR(handle)) {
681 status = PTR_ERR(handle);
682 handle = NULL;
683 mlog_errno(status);
684 goto leave;
685 }
686
687restarted_transaction:
688 /* reserve a write to the file entry early on - that we if we
689 * run out of credits in the allocation path, we can still
690 * update i_size. */
691 status = ocfs2_journal_access(handle, inode, bh,
692 OCFS2_JOURNAL_ACCESS_WRITE);
693 if (status < 0) {
694 mlog_errno(status);
695 goto leave;
696 }
697
698 prev_clusters = OCFS2_I(inode)->ip_clusters;
699
700 status = ocfs2_do_extend_allocation(osb,
701 inode,
dcd0538f 702 &logical_start,
ccd979bd 703 clusters_to_add,
2ae99a60 704 mark_unwritten,
ccd979bd
MF
705 bh,
706 handle,
707 data_ac,
708 meta_ac,
709 &why);
710 if ((status < 0) && (status != -EAGAIN)) {
711 if (status != -ENOSPC)
712 mlog_errno(status);
713 goto leave;
714 }
715
716 status = ocfs2_journal_dirty(handle, bh);
717 if (status < 0) {
718 mlog_errno(status);
719 goto leave;
720 }
721
722 spin_lock(&OCFS2_I(inode)->ip_lock);
723 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
724 spin_unlock(&OCFS2_I(inode)->ip_lock);
725
726 if (why != RESTART_NONE && clusters_to_add) {
727 if (why == RESTART_META) {
728 mlog(0, "restarting function.\n");
729 restart_func = 1;
730 } else {
731 BUG_ON(why != RESTART_TRANS);
732
733 mlog(0, "restarting transaction.\n");
734 /* TODO: This can be more intelligent. */
735 credits = ocfs2_calc_extend_credits(osb->sb,
736 fe,
737 clusters_to_add);
1fabe148 738 status = ocfs2_extend_trans(handle, credits);
ccd979bd
MF
739 if (status < 0) {
740 /* handle still has to be committed at
741 * this point. */
742 status = -ENOMEM;
743 mlog_errno(status);
744 goto leave;
745 }
746 goto restarted_transaction;
747 }
748 }
749
b0697053 750 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
1ca1a111
MF
751 le32_to_cpu(fe->i_clusters),
752 (unsigned long long)le64_to_cpu(fe->i_size));
ccd979bd
MF
753 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
754 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
755
756leave:
ccd979bd 757 if (handle) {
02dc1af4 758 ocfs2_commit_trans(osb, handle);
ccd979bd
MF
759 handle = NULL;
760 }
761 if (data_ac) {
762 ocfs2_free_alloc_context(data_ac);
763 data_ac = NULL;
764 }
765 if (meta_ac) {
766 ocfs2_free_alloc_context(meta_ac);
767 meta_ac = NULL;
768 }
769 if ((!status) && restart_func) {
770 restart_func = 0;
771 goto restart_all;
772 }
773 if (bh) {
774 brelse(bh);
775 bh = NULL;
776 }
777
778 mlog_exit(status);
779 return status;
780}
781
2ae99a60
MF
782static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
783 u32 clusters_to_add, int mark_unwritten)
784{
785 int ret;
786
787 /*
788 * The alloc sem blocks peope in read/write from reading our
789 * allocation until we're done changing it. We depend on
790 * i_mutex to block other extend/truncate calls while we're
791 * here.
792 */
793 down_write(&OCFS2_I(inode)->ip_alloc_sem);
794 ret = __ocfs2_extend_allocation(inode, logical_start, clusters_to_add,
795 mark_unwritten);
796 up_write(&OCFS2_I(inode)->ip_alloc_sem);
797
798 return ret;
799}
800
ccd979bd
MF
801/* Some parts of this taken from generic_cont_expand, which turned out
802 * to be too fragile to do exactly what we need without us having to
53013cba
MF
803 * worry about recursive locking in ->prepare_write() and
804 * ->commit_write(). */
ccd979bd
MF
805static int ocfs2_write_zero_page(struct inode *inode,
806 u64 size)
807{
808 struct address_space *mapping = inode->i_mapping;
809 struct page *page;
810 unsigned long index;
811 unsigned int offset;
1fabe148 812 handle_t *handle = NULL;
ccd979bd
MF
813 int ret;
814
815 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
816 /* ugh. in prepare/commit_write, if from==to==start of block, we
817 ** skip the prepare. make sure we never send an offset for the start
818 ** of a block
819 */
820 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
821 offset++;
822 }
823 index = size >> PAGE_CACHE_SHIFT;
824
825 page = grab_cache_page(mapping, index);
826 if (!page) {
827 ret = -ENOMEM;
828 mlog_errno(ret);
829 goto out;
830 }
831
53013cba 832 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
ccd979bd
MF
833 if (ret < 0) {
834 mlog_errno(ret);
835 goto out_unlock;
836 }
837
838 if (ocfs2_should_order_data(inode)) {
839 handle = ocfs2_start_walk_page_trans(inode, page, offset,
840 offset);
841 if (IS_ERR(handle)) {
842 ret = PTR_ERR(handle);
843 handle = NULL;
844 goto out_unlock;
845 }
846 }
847
848 /* must not update i_size! */
849 ret = block_commit_write(page, offset, offset);
850 if (ret < 0)
851 mlog_errno(ret);
852 else
853 ret = 0;
854
855 if (handle)
02dc1af4 856 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
ccd979bd
MF
857out_unlock:
858 unlock_page(page);
859 page_cache_release(page);
860out:
861 return ret;
862}
863
864static int ocfs2_zero_extend(struct inode *inode,
865 u64 zero_to_size)
866{
867 int ret = 0;
868 u64 start_off;
869 struct super_block *sb = inode->i_sb;
870
871 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
872 while (start_off < zero_to_size) {
873 ret = ocfs2_write_zero_page(inode, start_off);
874 if (ret < 0) {
875 mlog_errno(ret);
876 goto out;
877 }
878
879 start_off += sb->s_blocksize;
e2057c5a
MF
880
881 /*
882 * Very large extends have the potential to lock up
883 * the cpu for extended periods of time.
884 */
885 cond_resched();
ccd979bd
MF
886 }
887
888out:
889 return ret;
890}
891
53013cba
MF
892/*
893 * A tail_to_skip value > 0 indicates that we're being called from
894 * ocfs2_file_aio_write(). This has the following implications:
895 *
896 * - we don't want to update i_size
897 * - di_bh will be NULL, which is fine because it's only used in the
898 * case where we want to update i_size.
899 * - ocfs2_zero_extend() will then only be filling the hole created
900 * between i_size and the start of the write.
901 */
ccd979bd
MF
902static int ocfs2_extend_file(struct inode *inode,
903 struct buffer_head *di_bh,
53013cba
MF
904 u64 new_i_size,
905 size_t tail_to_skip)
ccd979bd
MF
906{
907 int ret = 0;
3a0782d0 908 u32 clusters_to_add = 0;
ccd979bd 909
53013cba
MF
910 BUG_ON(!tail_to_skip && !di_bh);
911
ccd979bd
MF
912 /* setattr sometimes calls us like this. */
913 if (new_i_size == 0)
914 goto out;
915
916 if (i_size_read(inode) == new_i_size)
917 goto out;
918 BUG_ON(new_i_size < i_size_read(inode));
919
3a0782d0
MF
920 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
921 BUG_ON(tail_to_skip != 0);
922 goto out_update_size;
923 }
924
ccd979bd
MF
925 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
926 OCFS2_I(inode)->ip_clusters;
927
0effef77
MF
928 /*
929 * protect the pages that ocfs2_zero_extend is going to be
930 * pulling into the page cache.. we do this before the
931 * metadata extend so that we don't get into the situation
932 * where we've extended the metadata but can't get the data
933 * lock to zero.
934 */
935 ret = ocfs2_data_lock(inode, 1);
936 if (ret < 0) {
937 mlog_errno(ret);
938 goto out;
939 }
ccd979bd 940
0effef77 941 if (clusters_to_add) {
2ae99a60
MF
942 ret = ocfs2_extend_allocation(inode,
943 OCFS2_I(inode)->ip_clusters,
944 clusters_to_add, 0);
ccd979bd
MF
945 if (ret < 0) {
946 mlog_errno(ret);
53013cba 947 goto out_unlock;
ccd979bd 948 }
0effef77 949 }
ccd979bd 950
0effef77
MF
951 /*
952 * Call this even if we don't add any clusters to the tree. We
953 * still need to zero the area between the old i_size and the
954 * new i_size.
955 */
956 ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
957 if (ret < 0) {
958 mlog_errno(ret);
959 goto out_unlock;
53013cba
MF
960 }
961
3a0782d0 962out_update_size:
53013cba
MF
963 if (!tail_to_skip) {
964 /* We're being called from ocfs2_setattr() which wants
965 * us to update i_size */
966 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
967 if (ret < 0)
968 mlog_errno(ret);
ccd979bd
MF
969 }
970
53013cba 971out_unlock:
3a0782d0
MF
972 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
973 ocfs2_data_unlock(inode, 1);
53013cba 974
ccd979bd
MF
975out:
976 return ret;
977}
978
979int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
980{
981 int status = 0, size_change;
982 struct inode *inode = dentry->d_inode;
983 struct super_block *sb = inode->i_sb;
984 struct ocfs2_super *osb = OCFS2_SB(sb);
985 struct buffer_head *bh = NULL;
1fabe148 986 handle_t *handle = NULL;
ccd979bd
MF
987
988 mlog_entry("(0x%p, '%.*s')\n", dentry,
989 dentry->d_name.len, dentry->d_name.name);
990
991 if (attr->ia_valid & ATTR_MODE)
992 mlog(0, "mode change: %d\n", attr->ia_mode);
993 if (attr->ia_valid & ATTR_UID)
994 mlog(0, "uid change: %d\n", attr->ia_uid);
995 if (attr->ia_valid & ATTR_GID)
996 mlog(0, "gid change: %d\n", attr->ia_gid);
997 if (attr->ia_valid & ATTR_SIZE)
998 mlog(0, "size change...\n");
999 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1000 mlog(0, "time change...\n");
1001
1002#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1003 | ATTR_GID | ATTR_UID | ATTR_MODE)
1004 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1005 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1006 return 0;
1007 }
1008
1009 status = inode_change_ok(inode, attr);
1010 if (status)
1011 return status;
1012
1013 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1014 if (size_change) {
1015 status = ocfs2_rw_lock(inode, 1);
1016 if (status < 0) {
1017 mlog_errno(status);
1018 goto bail;
1019 }
1020 }
1021
4bcec184 1022 status = ocfs2_meta_lock(inode, &bh, 1);
ccd979bd
MF
1023 if (status < 0) {
1024 if (status != -ENOENT)
1025 mlog_errno(status);
1026 goto bail_unlock_rw;
1027 }
1028
1029 if (size_change && attr->ia_size != i_size_read(inode)) {
ce76fd30
MF
1030 if (attr->ia_size > sb->s_maxbytes) {
1031 status = -EFBIG;
1032 goto bail_unlock;
1033 }
1034
ccd979bd
MF
1035 if (i_size_read(inode) > attr->ia_size)
1036 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1037 else
53013cba 1038 status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
ccd979bd
MF
1039 if (status < 0) {
1040 if (status != -ENOSPC)
1041 mlog_errno(status);
1042 status = -ENOSPC;
1043 goto bail_unlock;
1044 }
1045 }
1046
65eff9cc 1047 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
ccd979bd
MF
1048 if (IS_ERR(handle)) {
1049 status = PTR_ERR(handle);
1050 mlog_errno(status);
1051 goto bail_unlock;
1052 }
1053
7307de80
MF
1054 /*
1055 * This will intentionally not wind up calling vmtruncate(),
1056 * since all the work for a size change has been done above.
1057 * Otherwise, we could get into problems with truncate as
1058 * ip_alloc_sem is used there to protect against i_size
1059 * changes.
1060 */
ccd979bd
MF
1061 status = inode_setattr(inode, attr);
1062 if (status < 0) {
1063 mlog_errno(status);
1064 goto bail_commit;
1065 }
1066
1067 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1068 if (status < 0)
1069 mlog_errno(status);
1070
1071bail_commit:
02dc1af4 1072 ocfs2_commit_trans(osb, handle);
ccd979bd
MF
1073bail_unlock:
1074 ocfs2_meta_unlock(inode, 1);
1075bail_unlock_rw:
1076 if (size_change)
1077 ocfs2_rw_unlock(inode, 1);
1078bail:
1079 if (bh)
1080 brelse(bh);
1081
1082 mlog_exit(status);
1083 return status;
1084}
1085
1086int ocfs2_getattr(struct vfsmount *mnt,
1087 struct dentry *dentry,
1088 struct kstat *stat)
1089{
1090 struct inode *inode = dentry->d_inode;
1091 struct super_block *sb = dentry->d_inode->i_sb;
1092 struct ocfs2_super *osb = sb->s_fs_info;
1093 int err;
1094
1095 mlog_entry_void();
1096
1097 err = ocfs2_inode_revalidate(dentry);
1098 if (err) {
1099 if (err != -ENOENT)
1100 mlog_errno(err);
1101 goto bail;
1102 }
1103
1104 generic_fillattr(inode, stat);
1105
1106 /* We set the blksize from the cluster size for performance */
1107 stat->blksize = osb->s_clustersize;
1108
1109bail:
1110 mlog_exit(err);
1111
1112 return err;
1113}
1114
d38eb8db
TY
1115int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1116{
1117 int ret;
1118
1119 mlog_entry_void();
1120
1121 ret = ocfs2_meta_lock(inode, NULL, 0);
1122 if (ret) {
a9f5f707
MF
1123 if (ret != -ENOENT)
1124 mlog_errno(ret);
d38eb8db
TY
1125 goto out;
1126 }
1127
1128 ret = generic_permission(inode, mask, NULL);
d38eb8db
TY
1129
1130 ocfs2_meta_unlock(inode, 0);
1131out:
1132 mlog_exit(ret);
1133 return ret;
1134}
1135
b2580103
MF
1136static int __ocfs2_write_remove_suid(struct inode *inode,
1137 struct buffer_head *bh)
ccd979bd
MF
1138{
1139 int ret;
1fabe148 1140 handle_t *handle;
ccd979bd
MF
1141 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1142 struct ocfs2_dinode *di;
1143
b0697053 1144 mlog_entry("(Inode %llu, mode 0%o)\n",
b2580103 1145 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
ccd979bd 1146
65eff9cc 1147 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
ccd979bd
MF
1148 if (handle == NULL) {
1149 ret = -ENOMEM;
1150 mlog_errno(ret);
1151 goto out;
1152 }
1153
ccd979bd
MF
1154 ret = ocfs2_journal_access(handle, inode, bh,
1155 OCFS2_JOURNAL_ACCESS_WRITE);
1156 if (ret < 0) {
1157 mlog_errno(ret);
b2580103 1158 goto out_trans;
ccd979bd
MF
1159 }
1160
1161 inode->i_mode &= ~S_ISUID;
1162 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1163 inode->i_mode &= ~S_ISGID;
1164
1165 di = (struct ocfs2_dinode *) bh->b_data;
1166 di->i_mode = cpu_to_le16(inode->i_mode);
1167
1168 ret = ocfs2_journal_dirty(handle, bh);
1169 if (ret < 0)
1170 mlog_errno(ret);
b2580103 1171
ccd979bd 1172out_trans:
02dc1af4 1173 ocfs2_commit_trans(osb, handle);
ccd979bd
MF
1174out:
1175 mlog_exit(ret);
1176 return ret;
1177}
1178
9517bac6
MF
1179/*
1180 * Will look for holes and unwritten extents in the range starting at
1181 * pos for count bytes (inclusive).
1182 */
1183static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1184 size_t count)
1185{
1186 int ret = 0;
49cb8d2d 1187 unsigned int extent_flags;
9517bac6
MF
1188 u32 cpos, clusters, extent_len, phys_cpos;
1189 struct super_block *sb = inode->i_sb;
1190
1191 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1192 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1193
1194 while (clusters) {
49cb8d2d
MF
1195 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1196 &extent_flags);
9517bac6
MF
1197 if (ret < 0) {
1198 mlog_errno(ret);
1199 goto out;
1200 }
1201
49cb8d2d 1202 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
9517bac6
MF
1203 ret = 1;
1204 break;
1205 }
1206
1207 if (extent_len > clusters)
1208 extent_len = clusters;
1209
1210 clusters -= extent_len;
1211 cpos += extent_len;
1212 }
1213out:
1214 return ret;
1215}
1216
b2580103
MF
1217static int ocfs2_write_remove_suid(struct inode *inode)
1218{
1219 int ret;
1220 struct buffer_head *bh = NULL;
1221 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1222
1223 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1224 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1225 if (ret < 0) {
1226 mlog_errno(ret);
1227 goto out;
1228 }
1229
1230 ret = __ocfs2_write_remove_suid(inode, bh);
1231out:
1232 brelse(bh);
1233 return ret;
1234}
1235
2ae99a60
MF
1236/*
1237 * Allocate enough extents to cover the region starting at byte offset
1238 * start for len bytes. Existing extents are skipped, any extents
1239 * added are marked as "unwritten".
1240 */
1241static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1242 u64 start, u64 len)
1243{
1244 int ret;
1245 u32 cpos, phys_cpos, clusters, alloc_size;
1246
1247 /*
1248 * We consider both start and len to be inclusive.
1249 */
1250 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1251 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1252 clusters -= cpos;
1253
1254 while (clusters) {
1255 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1256 &alloc_size, NULL);
1257 if (ret) {
1258 mlog_errno(ret);
1259 goto out;
1260 }
1261
1262 /*
1263 * Hole or existing extent len can be arbitrary, so
1264 * cap it to our own allocation request.
1265 */
1266 if (alloc_size > clusters)
1267 alloc_size = clusters;
1268
1269 if (phys_cpos) {
1270 /*
1271 * We already have an allocation at this
1272 * region so we can safely skip it.
1273 */
1274 goto next;
1275 }
1276
1277 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1278 if (ret) {
1279 if (ret != -ENOSPC)
1280 mlog_errno(ret);
1281 goto out;
1282 }
1283
1284next:
1285 cpos += alloc_size;
1286 clusters -= alloc_size;
1287 }
1288
1289 ret = 0;
1290out:
1291 return ret;
1292}
1293
063c4561
MF
1294static int __ocfs2_remove_inode_range(struct inode *inode,
1295 struct buffer_head *di_bh,
1296 u32 cpos, u32 phys_cpos, u32 len,
1297 struct ocfs2_cached_dealloc_ctxt *dealloc)
1298{
1299 int ret;
1300 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1301 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1302 struct inode *tl_inode = osb->osb_tl_inode;
1303 handle_t *handle;
1304 struct ocfs2_alloc_context *meta_ac = NULL;
1305 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1306
1307 ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1308 if (ret) {
1309 mlog_errno(ret);
1310 return ret;
1311 }
1312
1313 mutex_lock(&tl_inode->i_mutex);
1314
1315 if (ocfs2_truncate_log_needs_flush(osb)) {
1316 ret = __ocfs2_flush_truncate_log(osb);
1317 if (ret < 0) {
1318 mlog_errno(ret);
1319 goto out;
1320 }
1321 }
1322
1323 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1324 if (handle == NULL) {
1325 ret = -ENOMEM;
1326 mlog_errno(ret);
1327 goto out;
1328 }
1329
1330 ret = ocfs2_journal_access(handle, inode, di_bh,
1331 OCFS2_JOURNAL_ACCESS_WRITE);
1332 if (ret) {
1333 mlog_errno(ret);
1334 goto out;
1335 }
1336
1337 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1338 dealloc);
1339 if (ret) {
1340 mlog_errno(ret);
1341 goto out_commit;
1342 }
1343
1344 OCFS2_I(inode)->ip_clusters -= len;
1345 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1346
1347 ret = ocfs2_journal_dirty(handle, di_bh);
1348 if (ret) {
1349 mlog_errno(ret);
1350 goto out_commit;
1351 }
1352
1353 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1354 if (ret)
1355 mlog_errno(ret);
1356
1357out_commit:
1358 ocfs2_commit_trans(osb, handle);
1359out:
1360 mutex_unlock(&tl_inode->i_mutex);
1361
1362 if (meta_ac)
1363 ocfs2_free_alloc_context(meta_ac);
1364
1365 return ret;
1366}
1367
1368/*
1369 * Truncate a byte range, avoiding pages within partial clusters. This
1370 * preserves those pages for the zeroing code to write to.
1371 */
1372static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1373 u64 byte_len)
1374{
1375 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1376 loff_t start, end;
1377 struct address_space *mapping = inode->i_mapping;
1378
1379 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1380 end = byte_start + byte_len;
1381 end = end & ~(osb->s_clustersize - 1);
1382
1383 if (start < end) {
1384 unmap_mapping_range(mapping, start, end - start, 0);
1385 truncate_inode_pages_range(mapping, start, end - 1);
1386 }
1387}
1388
1389static int ocfs2_zero_partial_clusters(struct inode *inode,
1390 u64 start, u64 len)
1391{
1392 int ret = 0;
1393 u64 tmpend, end = start + len;
1394 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1395 unsigned int csize = osb->s_clustersize;
1396 handle_t *handle;
1397
1398 /*
1399 * The "start" and "end" values are NOT necessarily part of
1400 * the range whose allocation is being deleted. Rather, this
1401 * is what the user passed in with the request. We must zero
1402 * partial clusters here. There's no need to worry about
1403 * physical allocation - the zeroing code knows to skip holes.
1404 */
1405 mlog(0, "byte start: %llu, end: %llu\n",
1406 (unsigned long long)start, (unsigned long long)end);
1407
1408 /*
1409 * If both edges are on a cluster boundary then there's no
1410 * zeroing required as the region is part of the allocation to
1411 * be truncated.
1412 */
1413 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1414 goto out;
1415
1416 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1417 if (handle == NULL) {
1418 ret = -ENOMEM;
1419 mlog_errno(ret);
1420 goto out;
1421 }
1422
1423 /*
1424 * We want to get the byte offset of the end of the 1st cluster.
1425 */
1426 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1427 if (tmpend > end)
1428 tmpend = end;
1429
1430 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1431 (unsigned long long)start, (unsigned long long)tmpend);
1432
1433 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1434 if (ret)
1435 mlog_errno(ret);
1436
1437 if (tmpend < end) {
1438 /*
1439 * This may make start and end equal, but the zeroing
1440 * code will skip any work in that case so there's no
1441 * need to catch it up here.
1442 */
1443 start = end & ~(osb->s_clustersize - 1);
1444
1445 mlog(0, "2nd range: start: %llu, end: %llu\n",
1446 (unsigned long long)start, (unsigned long long)end);
1447
1448 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1449 if (ret)
1450 mlog_errno(ret);
1451 }
1452
1453 ocfs2_commit_trans(osb, handle);
1454out:
1455 return ret;
1456}
1457
1458static int ocfs2_remove_inode_range(struct inode *inode,
1459 struct buffer_head *di_bh, u64 byte_start,
1460 u64 byte_len)
1461{
1462 int ret = 0;
1463 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1464 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1465 struct ocfs2_cached_dealloc_ctxt dealloc;
1466
1467 ocfs2_init_dealloc_ctxt(&dealloc);
1468
1469 if (byte_len == 0)
1470 return 0;
1471
1472 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1473 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1474 if (trunc_len >= trunc_start)
1475 trunc_len -= trunc_start;
1476 else
1477 trunc_len = 0;
1478
1479 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1480 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1481 (unsigned long long)byte_start,
1482 (unsigned long long)byte_len, trunc_start, trunc_len);
1483
1484 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1485 if (ret) {
1486 mlog_errno(ret);
1487 goto out;
1488 }
1489
1490 cpos = trunc_start;
1491 while (trunc_len) {
1492 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1493 &alloc_size, NULL);
1494 if (ret) {
1495 mlog_errno(ret);
1496 goto out;
1497 }
1498
1499 if (alloc_size > trunc_len)
1500 alloc_size = trunc_len;
1501
1502 /* Only do work for non-holes */
1503 if (phys_cpos != 0) {
1504 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1505 phys_cpos, alloc_size,
1506 &dealloc);
1507 if (ret) {
1508 mlog_errno(ret);
1509 goto out;
1510 }
1511 }
1512
1513 cpos += alloc_size;
1514 trunc_len -= alloc_size;
1515 }
1516
1517 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1518
1519out:
1520 ocfs2_schedule_truncate_log_flush(osb, 1);
1521 ocfs2_run_deallocs(osb, &dealloc);
1522
1523 return ret;
1524}
1525
b2580103
MF
1526/*
1527 * Parts of this function taken from xfs_change_file_space()
1528 */
385820a3
MF
1529static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1530 loff_t f_pos, unsigned int cmd,
1531 struct ocfs2_space_resv *sr,
1532 int change_size)
b2580103
MF
1533{
1534 int ret;
1535 s64 llen;
385820a3 1536 loff_t size;
b2580103
MF
1537 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1538 struct buffer_head *di_bh = NULL;
1539 handle_t *handle;
a00cce35 1540 unsigned long long max_off = inode->i_sb->s_maxbytes;
b2580103 1541
b2580103
MF
1542 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1543 return -EROFS;
1544
1545 mutex_lock(&inode->i_mutex);
1546
1547 /*
1548 * This prevents concurrent writes on other nodes
1549 */
1550 ret = ocfs2_rw_lock(inode, 1);
1551 if (ret) {
1552 mlog_errno(ret);
1553 goto out;
1554 }
1555
1556 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1557 if (ret) {
1558 mlog_errno(ret);
1559 goto out_rw_unlock;
1560 }
1561
1562 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1563 ret = -EPERM;
1564 goto out_meta_unlock;
1565 }
1566
1567 switch (sr->l_whence) {
1568 case 0: /*SEEK_SET*/
1569 break;
1570 case 1: /*SEEK_CUR*/
385820a3 1571 sr->l_start += f_pos;
b2580103
MF
1572 break;
1573 case 2: /*SEEK_END*/
1574 sr->l_start += i_size_read(inode);
1575 break;
1576 default:
1577 ret = -EINVAL;
1578 goto out_meta_unlock;
1579 }
1580 sr->l_whence = 0;
1581
1582 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1583
1584 if (sr->l_start < 0
1585 || sr->l_start > max_off
1586 || (sr->l_start + llen) < 0
1587 || (sr->l_start + llen) > max_off) {
1588 ret = -EINVAL;
1589 goto out_meta_unlock;
1590 }
385820a3 1591 size = sr->l_start + sr->l_len;
b2580103
MF
1592
1593 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1594 if (sr->l_len <= 0) {
1595 ret = -EINVAL;
1596 goto out_meta_unlock;
1597 }
1598 }
1599
385820a3 1600 if (file && should_remove_suid(file->f_path.dentry)) {
b2580103
MF
1601 ret = __ocfs2_write_remove_suid(inode, di_bh);
1602 if (ret) {
1603 mlog_errno(ret);
1604 goto out_meta_unlock;
1605 }
1606 }
1607
1608 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1609 switch (cmd) {
1610 case OCFS2_IOC_RESVSP:
1611 case OCFS2_IOC_RESVSP64:
1612 /*
1613 * This takes unsigned offsets, but the signed ones we
1614 * pass have been checked against overflow above.
1615 */
1616 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1617 sr->l_len);
1618 break;
1619 case OCFS2_IOC_UNRESVSP:
1620 case OCFS2_IOC_UNRESVSP64:
1621 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1622 sr->l_len);
1623 break;
1624 default:
1625 ret = -EINVAL;
1626 }
1627 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1628 if (ret) {
1629 mlog_errno(ret);
1630 goto out_meta_unlock;
1631 }
1632
1633 /*
1634 * We update c/mtime for these changes
1635 */
1636 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1637 if (IS_ERR(handle)) {
1638 ret = PTR_ERR(handle);
1639 mlog_errno(ret);
1640 goto out_meta_unlock;
1641 }
1642
385820a3
MF
1643 if (change_size && i_size_read(inode) < size)
1644 i_size_write(inode, size);
1645
b2580103
MF
1646 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1647 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1648 if (ret < 0)
1649 mlog_errno(ret);
1650
1651 ocfs2_commit_trans(osb, handle);
1652
1653out_meta_unlock:
1654 brelse(di_bh);
1655 ocfs2_meta_unlock(inode, 1);
1656out_rw_unlock:
1657 ocfs2_rw_unlock(inode, 1);
1658
1659 mutex_unlock(&inode->i_mutex);
1660out:
1661 return ret;
1662}
1663
385820a3
MF
1664int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1665 struct ocfs2_space_resv *sr)
1666{
1667 struct inode *inode = file->f_path.dentry->d_inode;
1668 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1669
1670 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1671 !ocfs2_writes_unwritten_extents(osb))
1672 return -ENOTTY;
1673 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1674 !ocfs2_sparse_alloc(osb))
1675 return -ENOTTY;
1676
1677 if (!S_ISREG(inode->i_mode))
1678 return -EINVAL;
1679
1680 if (!(file->f_mode & FMODE_WRITE))
1681 return -EBADF;
1682
1683 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1684}
1685
1686static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1687 loff_t len)
1688{
1689 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1690 struct ocfs2_space_resv sr;
1691 int change_size = 1;
1692
1693 if (!ocfs2_writes_unwritten_extents(osb))
1694 return -EOPNOTSUPP;
1695
1696 if (S_ISDIR(inode->i_mode))
1697 return -ENODEV;
1698
1699 if (mode & FALLOC_FL_KEEP_SIZE)
1700 change_size = 0;
1701
1702 sr.l_whence = 0;
1703 sr.l_start = (s64)offset;
1704 sr.l_len = (s64)len;
1705
1706 return __ocfs2_change_file_space(NULL, inode, offset,
1707 OCFS2_IOC_RESVSP64, &sr, change_size);
1708}
1709
8659ac25
TY
1710static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1711 loff_t *ppos,
1712 size_t count,
9517bac6
MF
1713 int appending,
1714 int *direct_io)
ccd979bd 1715{
8659ac25
TY
1716 int ret = 0, meta_level = appending;
1717 struct inode *inode = dentry->d_inode;
ccd979bd 1718 u32 clusters;
ccd979bd 1719 loff_t newsize, saved_pos;
ccd979bd 1720
ccd979bd
MF
1721 /*
1722 * We sample i_size under a read level meta lock to see if our write
1723 * is extending the file, if it is we back off and get a write level
1724 * meta lock.
1725 */
ccd979bd 1726 for(;;) {
4bcec184 1727 ret = ocfs2_meta_lock(inode, NULL, meta_level);
ccd979bd
MF
1728 if (ret < 0) {
1729 meta_level = -1;
1730 mlog_errno(ret);
1731 goto out;
1732 }
1733
1734 /* Clear suid / sgid if necessary. We do this here
1735 * instead of later in the write path because
1736 * remove_suid() calls ->setattr without any hint that
1737 * we may have already done our cluster locking. Since
1738 * ocfs2_setattr() *must* take cluster locks to
1739 * proceeed, this will lead us to recursively lock the
1740 * inode. There's also the dinode i_size state which
1741 * can be lost via setattr during extending writes (we
1742 * set inode->i_size at the end of a write. */
8659ac25 1743 if (should_remove_suid(dentry)) {
ccd979bd
MF
1744 if (meta_level == 0) {
1745 ocfs2_meta_unlock(inode, meta_level);
1746 meta_level = 1;
1747 continue;
1748 }
1749
1750 ret = ocfs2_write_remove_suid(inode);
1751 if (ret < 0) {
1752 mlog_errno(ret);
8659ac25 1753 goto out_unlock;
ccd979bd
MF
1754 }
1755 }
1756
1757 /* work on a copy of ppos until we're sure that we won't have
1758 * to recalculate it due to relocking. */
8659ac25 1759 if (appending) {
ccd979bd
MF
1760 saved_pos = i_size_read(inode);
1761 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1762 } else {
8659ac25 1763 saved_pos = *ppos;
ccd979bd 1764 }
3a0782d0 1765
9517bac6
MF
1766 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
1767 loff_t end = saved_pos + count;
1768
1769 /*
1770 * Skip the O_DIRECT checks if we don't need
1771 * them.
1772 */
1773 if (!direct_io || !(*direct_io))
1774 break;
1775
1776 /*
1777 * Allowing concurrent direct writes means
1778 * i_size changes wouldn't be synchronized, so
1779 * one node could wind up truncating another
1780 * nodes writes.
1781 */
1782 if (end > i_size_read(inode)) {
1783 *direct_io = 0;
1784 break;
1785 }
1786
1787 /*
1788 * We don't fill holes during direct io, so
1789 * check for them here. If any are found, the
1790 * caller will have to retake some cluster
1791 * locks and initiate the io as buffered.
1792 */
1793 ret = ocfs2_check_range_for_holes(inode, saved_pos,
1794 count);
1795 if (ret == 1) {
1796 *direct_io = 0;
1797 ret = 0;
1798 } else if (ret < 0)
1799 mlog_errno(ret);
1800 break;
1801 }
1802
3a0782d0
MF
1803 /*
1804 * The rest of this loop is concerned with legacy file
1805 * systems which don't support sparse files.
1806 */
3a0782d0 1807
8659ac25 1808 newsize = count + saved_pos;
ccd979bd 1809
215c7f9f
MF
1810 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1811 (long long) saved_pos, (long long) newsize,
1812 (long long) i_size_read(inode));
ccd979bd
MF
1813
1814 /* No need for a higher level metadata lock if we're
1815 * never going past i_size. */
1816 if (newsize <= i_size_read(inode))
1817 break;
1818
1819 if (meta_level == 0) {
1820 ocfs2_meta_unlock(inode, meta_level);
1821 meta_level = 1;
1822 continue;
1823 }
1824
1825 spin_lock(&OCFS2_I(inode)->ip_lock);
1826 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1827 OCFS2_I(inode)->ip_clusters;
1828 spin_unlock(&OCFS2_I(inode)->ip_lock);
1829
1830 mlog(0, "Writing at EOF, may need more allocation: "
215c7f9f
MF
1831 "i_size = %lld, newsize = %lld, need %u clusters\n",
1832 (long long) i_size_read(inode), (long long) newsize,
1833 clusters);
ccd979bd
MF
1834
1835 /* We only want to continue the rest of this loop if
1836 * our extend will actually require more
1837 * allocation. */
1838 if (!clusters)
1839 break;
1840
8659ac25 1841 ret = ocfs2_extend_file(inode, NULL, newsize, count);
ccd979bd
MF
1842 if (ret < 0) {
1843 if (ret != -ENOSPC)
1844 mlog_errno(ret);
8659ac25 1845 goto out_unlock;
ccd979bd 1846 }
ccd979bd
MF
1847 break;
1848 }
1849
8659ac25
TY
1850 if (appending)
1851 *ppos = saved_pos;
1852
1853out_unlock:
ccd979bd 1854 ocfs2_meta_unlock(inode, meta_level);
8659ac25
TY
1855
1856out:
1857 return ret;
1858}
1859
9517bac6
MF
1860static inline void
1861ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1862{
1863 const struct iovec *iov = *iovp;
1864 size_t base = *basep;
1865
1866 do {
1867 int copy = min(bytes, iov->iov_len - base);
1868
1869 bytes -= copy;
1870 base += copy;
1871 if (iov->iov_len == base) {
1872 iov++;
1873 base = 0;
1874 }
1875 } while (bytes);
1876 *iovp = iov;
1877 *basep = base;
1878}
1879
3a307ffc 1880static struct page * ocfs2_get_write_source(char **ret_src_buf,
9517bac6
MF
1881 const struct iovec *cur_iov,
1882 size_t iov_offset)
1883{
1884 int ret;
3a307ffc 1885 char *buf = cur_iov->iov_base + iov_offset;
9517bac6 1886 struct page *src_page = NULL;
3a307ffc 1887 unsigned long off;
9517bac6 1888
3a307ffc 1889 off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
9517bac6
MF
1890
1891 if (!segment_eq(get_fs(), KERNEL_DS)) {
1892 /*
1893 * Pull in the user page. We want to do this outside
1894 * of the meta data locks in order to preserve locking
1895 * order in case of page fault.
1896 */
1897 ret = get_user_pages(current, current->mm,
1898 (unsigned long)buf & PAGE_CACHE_MASK, 1,
1899 0, 0, &src_page, NULL);
1900 if (ret == 1)
3a307ffc 1901 *ret_src_buf = kmap(src_page) + off;
9517bac6
MF
1902 else
1903 src_page = ERR_PTR(-EFAULT);
1904 } else {
3a307ffc 1905 *ret_src_buf = buf;
9517bac6
MF
1906 }
1907
1908 return src_page;
1909}
1910
3a307ffc 1911static void ocfs2_put_write_source(struct page *page)
9517bac6
MF
1912{
1913 if (page) {
1914 kunmap(page);
1915 page_cache_release(page);
1916 }
1917}
1918
1919static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
1920 const struct iovec *iov,
1921 unsigned long nr_segs,
1922 size_t count,
1923 ssize_t o_direct_written)
1924{
1925 int ret = 0;
1926 ssize_t copied, total = 0;
3a307ffc
MF
1927 size_t iov_offset = 0, bytes;
1928 loff_t pos;
9517bac6 1929 const struct iovec *cur_iov = iov;
3a307ffc 1930 struct page *user_page, *page;
8e1c091c
JG
1931 char * uninitialized_var(buf);
1932 char *dst;
3a307ffc 1933 void *fsdata;
9517bac6
MF
1934
1935 /*
1936 * handle partial DIO write. Adjust cur_iov if needed.
1937 */
1938 ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
1939
1940 do {
3a307ffc 1941 pos = *ppos;
9517bac6 1942
3a307ffc
MF
1943 user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
1944 if (IS_ERR(user_page)) {
1945 ret = PTR_ERR(user_page);
9517bac6
MF
1946 goto out;
1947 }
1948
3a307ffc
MF
1949 /* Stay within our page boundaries */
1950 bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
1951 (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
1952 /* Stay within the vector boundary */
1953 bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
1954 /* Stay within count */
1955 bytes = min(bytes, count);
1956
1957 page = NULL;
1958 ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
1959 &page, &fsdata);
1960 if (ret) {
1961 mlog_errno(ret);
1962 goto out;
1963 }
9517bac6 1964
3a307ffc 1965 dst = kmap_atomic(page, KM_USER0);
7c08d70c 1966 memcpy(dst + (pos & (loff_t)(PAGE_CACHE_SIZE - 1)), buf, bytes);
3a307ffc
MF
1967 kunmap_atomic(dst, KM_USER0);
1968 flush_dcache_page(page);
1969 ocfs2_put_write_source(user_page);
9517bac6 1970
3a307ffc
MF
1971 copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
1972 bytes, page, fsdata);
9517bac6
MF
1973 if (copied < 0) {
1974 mlog_errno(copied);
1975 ret = copied;
1976 goto out;
1977 }
1978
1979 total += copied;
3a307ffc 1980 *ppos = pos + copied;
9517bac6
MF
1981 count -= copied;
1982
1983 ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
1984 } while(count);
1985
1986out:
1987 return total ? total : ret;
1988}
1989
8659ac25
TY
1990static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1991 const struct iovec *iov,
1992 unsigned long nr_segs,
1993 loff_t pos)
1994{
9517bac6
MF
1995 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1996 int can_do_direct, sync = 0;
1997 ssize_t written = 0;
1998 size_t ocount; /* original count */
1999 size_t count; /* after file limit checks */
2000 loff_t *ppos = &iocb->ki_pos;
2001 struct file *file = iocb->ki_filp;
2002 struct inode *inode = file->f_path.dentry->d_inode;
2003
2004 mlog_entry("(0x%p, %u, '%.*s')\n", file,
8659ac25 2005 (unsigned int)nr_segs,
9517bac6
MF
2006 file->f_path.dentry->d_name.len,
2007 file->f_path.dentry->d_name.name);
8659ac25 2008
8659ac25
TY
2009 if (iocb->ki_left == 0)
2010 return 0;
2011
d9b08b9e 2012 ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
9517bac6
MF
2013 if (ret)
2014 return ret;
2015
2016 count = ocount;
2017
2018 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2019
2020 appending = file->f_flags & O_APPEND ? 1 : 0;
2021 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2022
8659ac25 2023 mutex_lock(&inode->i_mutex);
9517bac6
MF
2024
2025relock:
8659ac25 2026 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
9517bac6 2027 if (direct_io) {
8659ac25 2028 down_read(&inode->i_alloc_sem);
9517bac6 2029 have_alloc_sem = 1;
8659ac25
TY
2030 }
2031
2032 /* concurrent O_DIRECT writes are allowed */
9517bac6 2033 rw_level = !direct_io;
8659ac25
TY
2034 ret = ocfs2_rw_lock(inode, rw_level);
2035 if (ret < 0) {
8659ac25 2036 mlog_errno(ret);
9517bac6 2037 goto out_sems;
8659ac25
TY
2038 }
2039
9517bac6
MF
2040 can_do_direct = direct_io;
2041 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
2042 iocb->ki_left, appending,
2043 &can_do_direct);
8659ac25
TY
2044 if (ret < 0) {
2045 mlog_errno(ret);
2046 goto out;
2047 }
ccd979bd 2048
9517bac6
MF
2049 /*
2050 * We can't complete the direct I/O as requested, fall back to
2051 * buffered I/O.
2052 */
2053 if (direct_io && !can_do_direct) {
2054 ocfs2_rw_unlock(inode, rw_level);
2055 up_read(&inode->i_alloc_sem);
2056
2057 have_alloc_sem = 0;
2058 rw_level = -1;
2059
2060 direct_io = 0;
2061 sync = 1;
2062 goto relock;
2063 }
2064
2065 if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
2066 sync = 1;
2067
2068 /*
2069 * XXX: Is it ok to execute these checks a second time?
2070 */
2071 ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
2072 if (ret)
2073 goto out;
2074
2075 /*
2076 * Set pos so that sync_page_range_nolock() below understands
2077 * where to start from. We might've moved it around via the
2078 * calls above. The range we want to actually sync starts from
2079 * *ppos here.
2080 *
2081 */
2082 pos = *ppos;
2083
ccd979bd 2084 /* communicate with ocfs2_dio_end_io */
7cdfc3a1 2085 ocfs2_iocb_set_rw_locked(iocb, rw_level);
ccd979bd 2086
9517bac6
MF
2087 if (direct_io) {
2088 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2089 ppos, count, ocount);
2090 if (written < 0) {
2091 ret = written;
2092 goto out_dio;
2093 }
2094 } else {
2095 written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
2096 count, written);
2097 if (written < 0) {
2098 ret = written;
2099 if (ret != -EFAULT || ret != -ENOSPC)
2100 mlog_errno(ret);
2101 goto out;
2102 }
2103 }
ccd979bd 2104
9517bac6 2105out_dio:
ccd979bd 2106 /* buffered aio wouldn't have proper lock coverage today */
9517bac6 2107 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
ccd979bd
MF
2108
2109 /*
2110 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2111 * function pointer which is called when o_direct io completes so that
2112 * it can unlock our rw lock. (it's the clustered equivalent of
2113 * i_alloc_sem; protects truncate from racing with pending ios).
2114 * Unfortunately there are error cases which call end_io and others
2115 * that don't. so we don't have to unlock the rw_lock if either an
2116 * async dio is going to do it in the future or an end_io after an
2117 * error has already done it.
2118 */
2119 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2120 rw_level = -1;
2121 have_alloc_sem = 0;
2122 }
2123
2124out:
9517bac6
MF
2125 if (rw_level != -1)
2126 ocfs2_rw_unlock(inode, rw_level);
2127
2128out_sems:
ccd979bd
MF
2129 if (have_alloc_sem)
2130 up_read(&inode->i_alloc_sem);
9517bac6
MF
2131
2132 if (written > 0 && sync) {
2133 ssize_t err;
2134
2135 err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
2136 if (err < 0)
2137 written = err;
2138 }
2139
1b1dcc1b 2140 mutex_unlock(&inode->i_mutex);
ccd979bd
MF
2141
2142 mlog_exit(ret);
9517bac6 2143 return written ? written : ret;
ccd979bd
MF
2144}
2145
6af67d82
MF
2146static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
2147 struct pipe_buffer *buf,
2148 struct splice_desc *sd)
2149{
3a307ffc 2150 int ret, count;
6af67d82 2151 ssize_t copied = 0;
3a307ffc
MF
2152 struct file *file = sd->u.file;
2153 unsigned int offset;
2154 struct page *page = NULL;
2155 void *fsdata;
2156 char *src, *dst;
6af67d82 2157
cac36bb0 2158 ret = buf->ops->confirm(pipe, buf);
6af67d82
MF
2159 if (ret)
2160 goto out;
2161
3a307ffc 2162 offset = sd->pos & ~PAGE_CACHE_MASK;
6af67d82 2163 count = sd->len;
3a307ffc
MF
2164 if (count + offset > PAGE_CACHE_SIZE)
2165 count = PAGE_CACHE_SIZE - offset;
6af67d82 2166
3a307ffc
MF
2167 ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
2168 &page, &fsdata);
2169 if (ret) {
2170 mlog_errno(ret);
2171 goto out;
2172 }
6af67d82 2173
3a307ffc
MF
2174 src = buf->ops->map(pipe, buf, 1);
2175 dst = kmap_atomic(page, KM_USER1);
2176 memcpy(dst + offset, src + buf->offset, count);
3836df6b 2177 kunmap_atomic(dst, KM_USER1);
3a307ffc 2178 buf->ops->unmap(pipe, buf, src);
6af67d82 2179
3a307ffc
MF
2180 copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
2181 page, fsdata);
2182 if (copied < 0) {
2183 mlog_errno(copied);
2184 ret = copied;
2185 goto out;
2186 }
6af67d82
MF
2187out:
2188
3a307ffc 2189 return copied ? copied : ret;
6af67d82
MF
2190}
2191
2192static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2193 struct file *out,
2194 loff_t *ppos,
2195 size_t len,
2196 unsigned int flags)
2197{
2198 int ret, err;
2199 struct address_space *mapping = out->f_mapping;
2200 struct inode *inode = mapping->host;
c66ab6fa
JA
2201 struct splice_desc sd = {
2202 .total_len = len,
2203 .flags = flags,
2204 .pos = *ppos,
6a14b90b 2205 .u.file = out,
c66ab6fa
JA
2206 };
2207
2208 ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor);
6af67d82
MF
2209 if (ret > 0) {
2210 *ppos += ret;
2211
2212 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
2213 err = generic_osync_inode(inode, mapping,
2214 OSYNC_METADATA|OSYNC_DATA);
2215 if (err)
2216 ret = err;
2217 }
2218 }
2219
2220 return ret;
2221}
2222
8659ac25
TY
2223static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2224 struct file *out,
2225 loff_t *ppos,
2226 size_t len,
2227 unsigned int flags)
2228{
2229 int ret;
d28c9174 2230 struct inode *inode = out->f_path.dentry->d_inode;
8659ac25
TY
2231
2232 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2233 (unsigned int)len,
d28c9174
JS
2234 out->f_path.dentry->d_name.len,
2235 out->f_path.dentry->d_name.name);
8659ac25
TY
2236
2237 inode_double_lock(inode, pipe->inode);
2238
2239 ret = ocfs2_rw_lock(inode, 1);
2240 if (ret < 0) {
2241 mlog_errno(ret);
2242 goto out;
2243 }
2244
9517bac6
MF
2245 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2246 NULL);
8659ac25
TY
2247 if (ret < 0) {
2248 mlog_errno(ret);
2249 goto out_unlock;
2250 }
2251
2252 /* ok, we're done with i_size and alloc work */
6af67d82 2253 ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
8659ac25
TY
2254
2255out_unlock:
2256 ocfs2_rw_unlock(inode, 1);
2257out:
2258 inode_double_unlock(inode, pipe->inode);
2259
2260 mlog_exit(ret);
2261 return ret;
2262}
2263
2264static ssize_t ocfs2_file_splice_read(struct file *in,
2265 loff_t *ppos,
2266 struct pipe_inode_info *pipe,
2267 size_t len,
2268 unsigned int flags)
2269{
2270 int ret = 0;
d28c9174 2271 struct inode *inode = in->f_path.dentry->d_inode;
8659ac25
TY
2272
2273 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2274 (unsigned int)len,
d28c9174
JS
2275 in->f_path.dentry->d_name.len,
2276 in->f_path.dentry->d_name.name);
8659ac25
TY
2277
2278 /*
2279 * See the comment in ocfs2_file_aio_read()
2280 */
2281 ret = ocfs2_meta_lock(inode, NULL, 0);
2282 if (ret < 0) {
2283 mlog_errno(ret);
2284 goto bail;
2285 }
2286 ocfs2_meta_unlock(inode, 0);
2287
2288 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2289
2290bail:
2291 mlog_exit(ret);
2292 return ret;
2293}
2294
ccd979bd 2295static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
027445c3
BP
2296 const struct iovec *iov,
2297 unsigned long nr_segs,
ccd979bd
MF
2298 loff_t pos)
2299{
25899dee 2300 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
ccd979bd 2301 struct file *filp = iocb->ki_filp;
d28c9174 2302 struct inode *inode = filp->f_path.dentry->d_inode;
ccd979bd 2303
027445c3
BP
2304 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2305 (unsigned int)nr_segs,
d28c9174
JS
2306 filp->f_path.dentry->d_name.len,
2307 filp->f_path.dentry->d_name.name);
ccd979bd
MF
2308
2309 if (!inode) {
2310 ret = -EINVAL;
2311 mlog_errno(ret);
2312 goto bail;
2313 }
2314
ccd979bd
MF
2315 /*
2316 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2317 * need locks to protect pending reads from racing with truncate.
2318 */
2319 if (filp->f_flags & O_DIRECT) {
2320 down_read(&inode->i_alloc_sem);
2321 have_alloc_sem = 1;
2322
2323 ret = ocfs2_rw_lock(inode, 0);
2324 if (ret < 0) {
2325 mlog_errno(ret);
2326 goto bail;
2327 }
2328 rw_level = 0;
2329 /* communicate with ocfs2_dio_end_io */
7cdfc3a1 2330 ocfs2_iocb_set_rw_locked(iocb, rw_level);
ccd979bd
MF
2331 }
2332
c4374f8a
MF
2333 /*
2334 * We're fine letting folks race truncates and extending
2335 * writes with read across the cluster, just like they can
2336 * locally. Hence no rw_lock during read.
2337 *
2338 * Take and drop the meta data lock to update inode fields
2339 * like i_size. This allows the checks down below
2340 * generic_file_aio_read() a chance of actually working.
2341 */
25899dee 2342 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
c4374f8a
MF
2343 if (ret < 0) {
2344 mlog_errno(ret);
2345 goto bail;
2346 }
25899dee 2347 ocfs2_meta_unlock(inode, lock_level);
c4374f8a 2348
027445c3 2349 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
ccd979bd
MF
2350 if (ret == -EINVAL)
2351 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2352
2353 /* buffered aio wouldn't have proper lock coverage today */
2354 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2355
2356 /* see ocfs2_file_aio_write */
2357 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2358 rw_level = -1;
2359 have_alloc_sem = 0;
2360 }
2361
2362bail:
2363 if (have_alloc_sem)
2364 up_read(&inode->i_alloc_sem);
2365 if (rw_level != -1)
2366 ocfs2_rw_unlock(inode, rw_level);
2367 mlog_exit(ret);
2368
2369 return ret;
2370}
2371
92e1d5be 2372const struct inode_operations ocfs2_file_iops = {
ccd979bd
MF
2373 .setattr = ocfs2_setattr,
2374 .getattr = ocfs2_getattr,
d38eb8db 2375 .permission = ocfs2_permission,
385820a3 2376 .fallocate = ocfs2_fallocate,
ccd979bd
MF
2377};
2378
92e1d5be 2379const struct inode_operations ocfs2_special_file_iops = {
ccd979bd
MF
2380 .setattr = ocfs2_setattr,
2381 .getattr = ocfs2_getattr,
d38eb8db 2382 .permission = ocfs2_permission,
ccd979bd
MF
2383};
2384
4b6f5d20 2385const struct file_operations ocfs2_fops = {
ccd979bd
MF
2386 .read = do_sync_read,
2387 .write = do_sync_write,
ccd979bd
MF
2388 .mmap = ocfs2_mmap,
2389 .fsync = ocfs2_sync_file,
2390 .release = ocfs2_file_release,
2391 .open = ocfs2_file_open,
2392 .aio_read = ocfs2_file_aio_read,
2393 .aio_write = ocfs2_file_aio_write,
ca4d147e 2394 .ioctl = ocfs2_ioctl,
586d232b
MF
2395#ifdef CONFIG_COMPAT
2396 .compat_ioctl = ocfs2_compat_ioctl,
2397#endif
8659ac25
TY
2398 .splice_read = ocfs2_file_splice_read,
2399 .splice_write = ocfs2_file_splice_write,
ccd979bd
MF
2400};
2401
4b6f5d20 2402const struct file_operations ocfs2_dops = {
ccd979bd
MF
2403 .read = generic_read_dir,
2404 .readdir = ocfs2_readdir,
2405 .fsync = ocfs2_sync_file,
ca4d147e 2406 .ioctl = ocfs2_ioctl,
586d232b
MF
2407#ifdef CONFIG_COMPAT
2408 .compat_ioctl = ocfs2_compat_ioctl,
2409#endif
ccd979bd 2410};