]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/ocfs2/file.c
ocfs2: Modify ocfs2_num_free_extents for future xattr usage.
[mirror_ubuntu-bionic-kernel.git] / fs / ocfs2 / file.c
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
26 #include <linux/capability.h>
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>
33 #include <linux/sched.h>
34 #include <linux/splice.h>
35 #include <linux/mount.h>
36 #include <linux/writeback.h>
37 #include <linux/falloc.h>
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"
52 #include "ioctl.h"
53 #include "journal.h"
54 #include "locks.h"
55 #include "mmap.h"
56 #include "suballoc.h"
57 #include "super.h"
58
59 #include "buffer_head_io.h"
60
61 static int ocfs2_sync_inode(struct inode *inode)
62 {
63 filemap_fdatawrite(inode->i_mapping);
64 return sync_mapping_buffers(inode->i_mapping);
65 }
66
67 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
68 {
69 struct ocfs2_file_private *fp;
70
71 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
72 if (!fp)
73 return -ENOMEM;
74
75 fp->fp_file = file;
76 mutex_init(&fp->fp_mutex);
77 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
78 file->private_data = fp;
79
80 return 0;
81 }
82
83 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
84 {
85 struct ocfs2_file_private *fp = file->private_data;
86 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
87
88 if (fp) {
89 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
90 ocfs2_lock_res_free(&fp->fp_flock);
91 kfree(fp);
92 file->private_data = NULL;
93 }
94 }
95
96 static int ocfs2_file_open(struct inode *inode, struct file *file)
97 {
98 int status;
99 int mode = file->f_flags;
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
101
102 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
103 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
104
105 spin_lock(&oi->ip_lock);
106
107 /* Check that the inode hasn't been wiped from disk by another
108 * node. If it hasn't then we're safe as long as we hold the
109 * spin lock until our increment of open count. */
110 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
111 spin_unlock(&oi->ip_lock);
112
113 status = -ENOENT;
114 goto leave;
115 }
116
117 if (mode & O_DIRECT)
118 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
119
120 oi->ip_open_count++;
121 spin_unlock(&oi->ip_lock);
122
123 status = ocfs2_init_file_private(inode, file);
124 if (status) {
125 /*
126 * We want to set open count back if we're failing the
127 * open.
128 */
129 spin_lock(&oi->ip_lock);
130 oi->ip_open_count--;
131 spin_unlock(&oi->ip_lock);
132 }
133
134 leave:
135 mlog_exit(status);
136 return status;
137 }
138
139 static int ocfs2_file_release(struct inode *inode, struct file *file)
140 {
141 struct ocfs2_inode_info *oi = OCFS2_I(inode);
142
143 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
144 file->f_path.dentry->d_name.len,
145 file->f_path.dentry->d_name.name);
146
147 spin_lock(&oi->ip_lock);
148 if (!--oi->ip_open_count)
149 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
150 spin_unlock(&oi->ip_lock);
151
152 ocfs2_free_file_private(inode, file);
153
154 mlog_exit(0);
155
156 return 0;
157 }
158
159 static int ocfs2_dir_open(struct inode *inode, struct file *file)
160 {
161 return ocfs2_init_file_private(inode, file);
162 }
163
164 static int ocfs2_dir_release(struct inode *inode, struct file *file)
165 {
166 ocfs2_free_file_private(inode, file);
167 return 0;
168 }
169
170 static int ocfs2_sync_file(struct file *file,
171 struct dentry *dentry,
172 int datasync)
173 {
174 int err = 0;
175 journal_t *journal;
176 struct inode *inode = dentry->d_inode;
177 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
178
179 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
180 dentry->d_name.len, dentry->d_name.name);
181
182 err = ocfs2_sync_inode(dentry->d_inode);
183 if (err)
184 goto bail;
185
186 journal = osb->journal->j_journal;
187 err = journal_force_commit(journal);
188
189 bail:
190 mlog_exit(err);
191
192 return (err < 0) ? -EIO : 0;
193 }
194
195 int ocfs2_should_update_atime(struct inode *inode,
196 struct vfsmount *vfsmnt)
197 {
198 struct timespec now;
199 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
200
201 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
202 return 0;
203
204 if ((inode->i_flags & S_NOATIME) ||
205 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
206 return 0;
207
208 /*
209 * We can be called with no vfsmnt structure - NFSD will
210 * sometimes do this.
211 *
212 * Note that our action here is different than touch_atime() -
213 * if we can't tell whether this is a noatime mount, then we
214 * don't know whether to trust the value of s_atime_quantum.
215 */
216 if (vfsmnt == NULL)
217 return 0;
218
219 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
220 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
221 return 0;
222
223 if (vfsmnt->mnt_flags & MNT_RELATIME) {
224 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
225 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
226 return 1;
227
228 return 0;
229 }
230
231 now = CURRENT_TIME;
232 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
233 return 0;
234 else
235 return 1;
236 }
237
238 int ocfs2_update_inode_atime(struct inode *inode,
239 struct buffer_head *bh)
240 {
241 int ret;
242 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
243 handle_t *handle;
244 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
245
246 mlog_entry_void();
247
248 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
249 if (handle == NULL) {
250 ret = -ENOMEM;
251 mlog_errno(ret);
252 goto out;
253 }
254
255 ret = ocfs2_journal_access(handle, inode, bh,
256 OCFS2_JOURNAL_ACCESS_WRITE);
257 if (ret) {
258 mlog_errno(ret);
259 goto out_commit;
260 }
261
262 /*
263 * Don't use ocfs2_mark_inode_dirty() here as we don't always
264 * have i_mutex to guard against concurrent changes to other
265 * inode fields.
266 */
267 inode->i_atime = CURRENT_TIME;
268 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
269 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
270
271 ret = ocfs2_journal_dirty(handle, bh);
272 if (ret < 0)
273 mlog_errno(ret);
274
275 out_commit:
276 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
277 out:
278 mlog_exit(ret);
279 return ret;
280 }
281
282 static int ocfs2_set_inode_size(handle_t *handle,
283 struct inode *inode,
284 struct buffer_head *fe_bh,
285 u64 new_i_size)
286 {
287 int status;
288
289 mlog_entry_void();
290 i_size_write(inode, new_i_size);
291 inode->i_blocks = ocfs2_inode_sector_count(inode);
292 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
293
294 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
295 if (status < 0) {
296 mlog_errno(status);
297 goto bail;
298 }
299
300 bail:
301 mlog_exit(status);
302 return status;
303 }
304
305 static int ocfs2_simple_size_update(struct inode *inode,
306 struct buffer_head *di_bh,
307 u64 new_i_size)
308 {
309 int ret;
310 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
311 handle_t *handle = NULL;
312
313 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
314 if (handle == NULL) {
315 ret = -ENOMEM;
316 mlog_errno(ret);
317 goto out;
318 }
319
320 ret = ocfs2_set_inode_size(handle, inode, di_bh,
321 new_i_size);
322 if (ret < 0)
323 mlog_errno(ret);
324
325 ocfs2_commit_trans(osb, handle);
326 out:
327 return ret;
328 }
329
330 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
331 struct inode *inode,
332 struct buffer_head *fe_bh,
333 u64 new_i_size)
334 {
335 int status;
336 handle_t *handle;
337 struct ocfs2_dinode *di;
338 u64 cluster_bytes;
339
340 mlog_entry_void();
341
342 /* TODO: This needs to actually orphan the inode in this
343 * transaction. */
344
345 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
346 if (IS_ERR(handle)) {
347 status = PTR_ERR(handle);
348 mlog_errno(status);
349 goto out;
350 }
351
352 status = ocfs2_journal_access(handle, inode, fe_bh,
353 OCFS2_JOURNAL_ACCESS_WRITE);
354 if (status < 0) {
355 mlog_errno(status);
356 goto out_commit;
357 }
358
359 /*
360 * Do this before setting i_size.
361 */
362 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
363 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
364 cluster_bytes);
365 if (status) {
366 mlog_errno(status);
367 goto out_commit;
368 }
369
370 i_size_write(inode, new_i_size);
371 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
372
373 di = (struct ocfs2_dinode *) fe_bh->b_data;
374 di->i_size = cpu_to_le64(new_i_size);
375 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
376 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
377
378 status = ocfs2_journal_dirty(handle, fe_bh);
379 if (status < 0)
380 mlog_errno(status);
381
382 out_commit:
383 ocfs2_commit_trans(osb, handle);
384 out:
385
386 mlog_exit(status);
387 return status;
388 }
389
390 static int ocfs2_truncate_file(struct inode *inode,
391 struct buffer_head *di_bh,
392 u64 new_i_size)
393 {
394 int status = 0;
395 struct ocfs2_dinode *fe = NULL;
396 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
397 struct ocfs2_truncate_context *tc = NULL;
398
399 mlog_entry("(inode = %llu, new_i_size = %llu\n",
400 (unsigned long long)OCFS2_I(inode)->ip_blkno,
401 (unsigned long long)new_i_size);
402
403 fe = (struct ocfs2_dinode *) di_bh->b_data;
404 if (!OCFS2_IS_VALID_DINODE(fe)) {
405 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
406 status = -EIO;
407 goto bail;
408 }
409
410 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
411 "Inode %llu, inode i_size = %lld != di "
412 "i_size = %llu, i_flags = 0x%x\n",
413 (unsigned long long)OCFS2_I(inode)->ip_blkno,
414 i_size_read(inode),
415 (unsigned long long)le64_to_cpu(fe->i_size),
416 le32_to_cpu(fe->i_flags));
417
418 if (new_i_size > le64_to_cpu(fe->i_size)) {
419 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
420 (unsigned long long)le64_to_cpu(fe->i_size),
421 (unsigned long long)new_i_size);
422 status = -EINVAL;
423 mlog_errno(status);
424 goto bail;
425 }
426
427 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
428 (unsigned long long)le64_to_cpu(fe->i_blkno),
429 (unsigned long long)le64_to_cpu(fe->i_size),
430 (unsigned long long)new_i_size);
431
432 /* lets handle the simple truncate cases before doing any more
433 * cluster locking. */
434 if (new_i_size == le64_to_cpu(fe->i_size))
435 goto bail;
436
437 down_write(&OCFS2_I(inode)->ip_alloc_sem);
438
439 /*
440 * The inode lock forced other nodes to sync and drop their
441 * pages, which (correctly) happens even if we have a truncate
442 * without allocation change - ocfs2 cluster sizes can be much
443 * greater than page size, so we have to truncate them
444 * anyway.
445 */
446 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
447 truncate_inode_pages(inode->i_mapping, new_i_size);
448
449 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
450 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
451 i_size_read(inode), 1);
452 if (status)
453 mlog_errno(status);
454
455 goto bail_unlock_sem;
456 }
457
458 /* alright, we're going to need to do a full blown alloc size
459 * change. Orphan the inode so that recovery can complete the
460 * truncate if necessary. This does the task of marking
461 * i_size. */
462 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
463 if (status < 0) {
464 mlog_errno(status);
465 goto bail_unlock_sem;
466 }
467
468 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
469 if (status < 0) {
470 mlog_errno(status);
471 goto bail_unlock_sem;
472 }
473
474 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
475 if (status < 0) {
476 mlog_errno(status);
477 goto bail_unlock_sem;
478 }
479
480 /* TODO: orphan dir cleanup here. */
481 bail_unlock_sem:
482 up_write(&OCFS2_I(inode)->ip_alloc_sem);
483
484 bail:
485
486 mlog_exit(status);
487 return status;
488 }
489
490 /*
491 * extend allocation only here.
492 * we'll update all the disk stuff, and oip->alloc_size
493 *
494 * expect stuff to be locked, a transaction started and enough data /
495 * metadata reservations in the contexts.
496 *
497 * Will return -EAGAIN, and a reason if a restart is needed.
498 * If passed in, *reason will always be set, even in error.
499 */
500 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
501 struct inode *inode,
502 u32 *logical_offset,
503 u32 clusters_to_add,
504 int mark_unwritten,
505 struct buffer_head *fe_bh,
506 handle_t *handle,
507 struct ocfs2_alloc_context *data_ac,
508 struct ocfs2_alloc_context *meta_ac,
509 enum ocfs2_alloc_restarted *reason_ret)
510 {
511 int status = 0;
512 int free_extents;
513 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
514 enum ocfs2_alloc_restarted reason = RESTART_NONE;
515 u32 bit_off, num_bits;
516 u64 block;
517 u8 flags = 0;
518
519 BUG_ON(!clusters_to_add);
520
521 if (mark_unwritten)
522 flags = OCFS2_EXT_UNWRITTEN;
523
524 free_extents = ocfs2_num_free_extents(osb, inode, fe_bh);
525 if (free_extents < 0) {
526 status = free_extents;
527 mlog_errno(status);
528 goto leave;
529 }
530
531 /* there are two cases which could cause us to EAGAIN in the
532 * we-need-more-metadata case:
533 * 1) we haven't reserved *any*
534 * 2) we are so fragmented, we've needed to add metadata too
535 * many times. */
536 if (!free_extents && !meta_ac) {
537 mlog(0, "we haven't reserved any metadata!\n");
538 status = -EAGAIN;
539 reason = RESTART_META;
540 goto leave;
541 } else if ((!free_extents)
542 && (ocfs2_alloc_context_bits_left(meta_ac)
543 < ocfs2_extend_meta_needed(fe))) {
544 mlog(0, "filesystem is really fragmented...\n");
545 status = -EAGAIN;
546 reason = RESTART_META;
547 goto leave;
548 }
549
550 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
551 clusters_to_add, &bit_off, &num_bits);
552 if (status < 0) {
553 if (status != -ENOSPC)
554 mlog_errno(status);
555 goto leave;
556 }
557
558 BUG_ON(num_bits > clusters_to_add);
559
560 /* reserve our write early -- insert_extent may update the inode */
561 status = ocfs2_journal_access(handle, inode, fe_bh,
562 OCFS2_JOURNAL_ACCESS_WRITE);
563 if (status < 0) {
564 mlog_errno(status);
565 goto leave;
566 }
567
568 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
569 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
570 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
571 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
572 *logical_offset, block, num_bits,
573 flags, meta_ac);
574 if (status < 0) {
575 mlog_errno(status);
576 goto leave;
577 }
578
579 status = ocfs2_journal_dirty(handle, fe_bh);
580 if (status < 0) {
581 mlog_errno(status);
582 goto leave;
583 }
584
585 clusters_to_add -= num_bits;
586 *logical_offset += num_bits;
587
588 if (clusters_to_add) {
589 mlog(0, "need to alloc once more, clusters = %u, wanted = "
590 "%u\n", fe->i_clusters, clusters_to_add);
591 status = -EAGAIN;
592 reason = RESTART_TRANS;
593 }
594
595 leave:
596 mlog_exit(status);
597 if (reason_ret)
598 *reason_ret = reason;
599 return status;
600 }
601
602 /*
603 * For a given allocation, determine which allocators will need to be
604 * accessed, and lock them, reserving the appropriate number of bits.
605 *
606 * Sparse file systems call this from ocfs2_write_begin_nolock()
607 * and ocfs2_allocate_unwritten_extents().
608 *
609 * File systems which don't support holes call this from
610 * ocfs2_extend_allocation().
611 */
612 int ocfs2_lock_allocators(struct inode *inode, struct buffer_head *di_bh,
613 u32 clusters_to_add, u32 extents_to_split,
614 struct ocfs2_alloc_context **data_ac,
615 struct ocfs2_alloc_context **meta_ac)
616 {
617 int ret = 0, num_free_extents;
618 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
619 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
620 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
621
622 *meta_ac = NULL;
623 if (data_ac)
624 *data_ac = NULL;
625
626 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
627
628 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
629 "clusters_to_add = %u, extents_to_split = %u\n",
630 (unsigned long long)OCFS2_I(inode)->ip_blkno, (long long)i_size_read(inode),
631 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
632
633 num_free_extents = ocfs2_num_free_extents(osb, inode, di_bh);
634 if (num_free_extents < 0) {
635 ret = num_free_extents;
636 mlog_errno(ret);
637 goto out;
638 }
639
640 /*
641 * Sparse allocation file systems need to be more conservative
642 * with reserving room for expansion - the actual allocation
643 * happens while we've got a journal handle open so re-taking
644 * a cluster lock (because we ran out of room for another
645 * extent) will violate ordering rules.
646 *
647 * Most of the time we'll only be seeing this 1 cluster at a time
648 * anyway.
649 *
650 * Always lock for any unwritten extents - we might want to
651 * add blocks during a split.
652 */
653 if (!num_free_extents ||
654 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
655 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
656 if (ret < 0) {
657 if (ret != -ENOSPC)
658 mlog_errno(ret);
659 goto out;
660 }
661 }
662
663 if (clusters_to_add == 0)
664 goto out;
665
666 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
667 if (ret < 0) {
668 if (ret != -ENOSPC)
669 mlog_errno(ret);
670 goto out;
671 }
672
673 out:
674 if (ret) {
675 if (*meta_ac) {
676 ocfs2_free_alloc_context(*meta_ac);
677 *meta_ac = NULL;
678 }
679
680 /*
681 * We cannot have an error and a non null *data_ac.
682 */
683 }
684
685 return ret;
686 }
687
688 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
689 u32 clusters_to_add, int mark_unwritten)
690 {
691 int status = 0;
692 int restart_func = 0;
693 int credits;
694 u32 prev_clusters;
695 struct buffer_head *bh = NULL;
696 struct ocfs2_dinode *fe = NULL;
697 handle_t *handle = NULL;
698 struct ocfs2_alloc_context *data_ac = NULL;
699 struct ocfs2_alloc_context *meta_ac = NULL;
700 enum ocfs2_alloc_restarted why;
701 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
702
703 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
704
705 /*
706 * This function only exists for file systems which don't
707 * support holes.
708 */
709 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
710
711 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
712 OCFS2_BH_CACHED, inode);
713 if (status < 0) {
714 mlog_errno(status);
715 goto leave;
716 }
717
718 fe = (struct ocfs2_dinode *) bh->b_data;
719 if (!OCFS2_IS_VALID_DINODE(fe)) {
720 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
721 status = -EIO;
722 goto leave;
723 }
724
725 restart_all:
726 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
727
728 status = ocfs2_lock_allocators(inode, bh, clusters_to_add, 0, &data_ac,
729 &meta_ac);
730 if (status) {
731 mlog_errno(status);
732 goto leave;
733 }
734
735 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
736 handle = ocfs2_start_trans(osb, credits);
737 if (IS_ERR(handle)) {
738 status = PTR_ERR(handle);
739 handle = NULL;
740 mlog_errno(status);
741 goto leave;
742 }
743
744 restarted_transaction:
745 /* reserve a write to the file entry early on - that we if we
746 * run out of credits in the allocation path, we can still
747 * update i_size. */
748 status = ocfs2_journal_access(handle, inode, bh,
749 OCFS2_JOURNAL_ACCESS_WRITE);
750 if (status < 0) {
751 mlog_errno(status);
752 goto leave;
753 }
754
755 prev_clusters = OCFS2_I(inode)->ip_clusters;
756
757 status = ocfs2_do_extend_allocation(osb,
758 inode,
759 &logical_start,
760 clusters_to_add,
761 mark_unwritten,
762 bh,
763 handle,
764 data_ac,
765 meta_ac,
766 &why);
767 if ((status < 0) && (status != -EAGAIN)) {
768 if (status != -ENOSPC)
769 mlog_errno(status);
770 goto leave;
771 }
772
773 status = ocfs2_journal_dirty(handle, bh);
774 if (status < 0) {
775 mlog_errno(status);
776 goto leave;
777 }
778
779 spin_lock(&OCFS2_I(inode)->ip_lock);
780 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
781 spin_unlock(&OCFS2_I(inode)->ip_lock);
782
783 if (why != RESTART_NONE && clusters_to_add) {
784 if (why == RESTART_META) {
785 mlog(0, "restarting function.\n");
786 restart_func = 1;
787 } else {
788 BUG_ON(why != RESTART_TRANS);
789
790 mlog(0, "restarting transaction.\n");
791 /* TODO: This can be more intelligent. */
792 credits = ocfs2_calc_extend_credits(osb->sb,
793 fe,
794 clusters_to_add);
795 status = ocfs2_extend_trans(handle, credits);
796 if (status < 0) {
797 /* handle still has to be committed at
798 * this point. */
799 status = -ENOMEM;
800 mlog_errno(status);
801 goto leave;
802 }
803 goto restarted_transaction;
804 }
805 }
806
807 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
808 le32_to_cpu(fe->i_clusters),
809 (unsigned long long)le64_to_cpu(fe->i_size));
810 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
811 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
812
813 leave:
814 if (handle) {
815 ocfs2_commit_trans(osb, handle);
816 handle = NULL;
817 }
818 if (data_ac) {
819 ocfs2_free_alloc_context(data_ac);
820 data_ac = NULL;
821 }
822 if (meta_ac) {
823 ocfs2_free_alloc_context(meta_ac);
824 meta_ac = NULL;
825 }
826 if ((!status) && restart_func) {
827 restart_func = 0;
828 goto restart_all;
829 }
830 if (bh) {
831 brelse(bh);
832 bh = NULL;
833 }
834
835 mlog_exit(status);
836 return status;
837 }
838
839 /* Some parts of this taken from generic_cont_expand, which turned out
840 * to be too fragile to do exactly what we need without us having to
841 * worry about recursive locking in ->prepare_write() and
842 * ->commit_write(). */
843 static int ocfs2_write_zero_page(struct inode *inode,
844 u64 size)
845 {
846 struct address_space *mapping = inode->i_mapping;
847 struct page *page;
848 unsigned long index;
849 unsigned int offset;
850 handle_t *handle = NULL;
851 int ret;
852
853 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
854 /* ugh. in prepare/commit_write, if from==to==start of block, we
855 ** skip the prepare. make sure we never send an offset for the start
856 ** of a block
857 */
858 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
859 offset++;
860 }
861 index = size >> PAGE_CACHE_SHIFT;
862
863 page = grab_cache_page(mapping, index);
864 if (!page) {
865 ret = -ENOMEM;
866 mlog_errno(ret);
867 goto out;
868 }
869
870 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
871 if (ret < 0) {
872 mlog_errno(ret);
873 goto out_unlock;
874 }
875
876 if (ocfs2_should_order_data(inode)) {
877 handle = ocfs2_start_walk_page_trans(inode, page, offset,
878 offset);
879 if (IS_ERR(handle)) {
880 ret = PTR_ERR(handle);
881 handle = NULL;
882 goto out_unlock;
883 }
884 }
885
886 /* must not update i_size! */
887 ret = block_commit_write(page, offset, offset);
888 if (ret < 0)
889 mlog_errno(ret);
890 else
891 ret = 0;
892
893 if (handle)
894 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
895 out_unlock:
896 unlock_page(page);
897 page_cache_release(page);
898 out:
899 return ret;
900 }
901
902 static int ocfs2_zero_extend(struct inode *inode,
903 u64 zero_to_size)
904 {
905 int ret = 0;
906 u64 start_off;
907 struct super_block *sb = inode->i_sb;
908
909 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
910 while (start_off < zero_to_size) {
911 ret = ocfs2_write_zero_page(inode, start_off);
912 if (ret < 0) {
913 mlog_errno(ret);
914 goto out;
915 }
916
917 start_off += sb->s_blocksize;
918
919 /*
920 * Very large extends have the potential to lock up
921 * the cpu for extended periods of time.
922 */
923 cond_resched();
924 }
925
926 out:
927 return ret;
928 }
929
930 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
931 {
932 int ret;
933 u32 clusters_to_add;
934 struct ocfs2_inode_info *oi = OCFS2_I(inode);
935
936 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
937 if (clusters_to_add < oi->ip_clusters)
938 clusters_to_add = 0;
939 else
940 clusters_to_add -= oi->ip_clusters;
941
942 if (clusters_to_add) {
943 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
944 clusters_to_add, 0);
945 if (ret) {
946 mlog_errno(ret);
947 goto out;
948 }
949 }
950
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, zero_to);
957 if (ret < 0)
958 mlog_errno(ret);
959
960 out:
961 return ret;
962 }
963
964 static int ocfs2_extend_file(struct inode *inode,
965 struct buffer_head *di_bh,
966 u64 new_i_size)
967 {
968 int ret = 0;
969 struct ocfs2_inode_info *oi = OCFS2_I(inode);
970
971 BUG_ON(!di_bh);
972
973 /* setattr sometimes calls us like this. */
974 if (new_i_size == 0)
975 goto out;
976
977 if (i_size_read(inode) == new_i_size)
978 goto out;
979 BUG_ON(new_i_size < i_size_read(inode));
980
981 /*
982 * Fall through for converting inline data, even if the fs
983 * supports sparse files.
984 *
985 * The check for inline data here is legal - nobody can add
986 * the feature since we have i_mutex. We must check it again
987 * after acquiring ip_alloc_sem though, as paths like mmap
988 * might have raced us to converting the inode to extents.
989 */
990 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
991 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
992 goto out_update_size;
993
994 /*
995 * The alloc sem blocks people in read/write from reading our
996 * allocation until we're done changing it. We depend on
997 * i_mutex to block other extend/truncate calls while we're
998 * here.
999 */
1000 down_write(&oi->ip_alloc_sem);
1001
1002 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1003 /*
1004 * We can optimize small extends by keeping the inodes
1005 * inline data.
1006 */
1007 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1008 up_write(&oi->ip_alloc_sem);
1009 goto out_update_size;
1010 }
1011
1012 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1013 if (ret) {
1014 up_write(&oi->ip_alloc_sem);
1015
1016 mlog_errno(ret);
1017 goto out;
1018 }
1019 }
1020
1021 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1022 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
1023
1024 up_write(&oi->ip_alloc_sem);
1025
1026 if (ret < 0) {
1027 mlog_errno(ret);
1028 goto out;
1029 }
1030
1031 out_update_size:
1032 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1033 if (ret < 0)
1034 mlog_errno(ret);
1035
1036 out:
1037 return ret;
1038 }
1039
1040 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1041 {
1042 int status = 0, size_change;
1043 struct inode *inode = dentry->d_inode;
1044 struct super_block *sb = inode->i_sb;
1045 struct ocfs2_super *osb = OCFS2_SB(sb);
1046 struct buffer_head *bh = NULL;
1047 handle_t *handle = NULL;
1048
1049 mlog_entry("(0x%p, '%.*s')\n", dentry,
1050 dentry->d_name.len, dentry->d_name.name);
1051
1052 /* ensuring we don't even attempt to truncate a symlink */
1053 if (S_ISLNK(inode->i_mode))
1054 attr->ia_valid &= ~ATTR_SIZE;
1055
1056 if (attr->ia_valid & ATTR_MODE)
1057 mlog(0, "mode change: %d\n", attr->ia_mode);
1058 if (attr->ia_valid & ATTR_UID)
1059 mlog(0, "uid change: %d\n", attr->ia_uid);
1060 if (attr->ia_valid & ATTR_GID)
1061 mlog(0, "gid change: %d\n", attr->ia_gid);
1062 if (attr->ia_valid & ATTR_SIZE)
1063 mlog(0, "size change...\n");
1064 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1065 mlog(0, "time change...\n");
1066
1067 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1068 | ATTR_GID | ATTR_UID | ATTR_MODE)
1069 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1070 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1071 return 0;
1072 }
1073
1074 status = inode_change_ok(inode, attr);
1075 if (status)
1076 return status;
1077
1078 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1079 if (size_change) {
1080 status = ocfs2_rw_lock(inode, 1);
1081 if (status < 0) {
1082 mlog_errno(status);
1083 goto bail;
1084 }
1085 }
1086
1087 status = ocfs2_inode_lock(inode, &bh, 1);
1088 if (status < 0) {
1089 if (status != -ENOENT)
1090 mlog_errno(status);
1091 goto bail_unlock_rw;
1092 }
1093
1094 if (size_change && attr->ia_size != i_size_read(inode)) {
1095 if (attr->ia_size > sb->s_maxbytes) {
1096 status = -EFBIG;
1097 goto bail_unlock;
1098 }
1099
1100 if (i_size_read(inode) > attr->ia_size)
1101 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1102 else
1103 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1104 if (status < 0) {
1105 if (status != -ENOSPC)
1106 mlog_errno(status);
1107 status = -ENOSPC;
1108 goto bail_unlock;
1109 }
1110 }
1111
1112 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1113 if (IS_ERR(handle)) {
1114 status = PTR_ERR(handle);
1115 mlog_errno(status);
1116 goto bail_unlock;
1117 }
1118
1119 /*
1120 * This will intentionally not wind up calling vmtruncate(),
1121 * since all the work for a size change has been done above.
1122 * Otherwise, we could get into problems with truncate as
1123 * ip_alloc_sem is used there to protect against i_size
1124 * changes.
1125 */
1126 status = inode_setattr(inode, attr);
1127 if (status < 0) {
1128 mlog_errno(status);
1129 goto bail_commit;
1130 }
1131
1132 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1133 if (status < 0)
1134 mlog_errno(status);
1135
1136 bail_commit:
1137 ocfs2_commit_trans(osb, handle);
1138 bail_unlock:
1139 ocfs2_inode_unlock(inode, 1);
1140 bail_unlock_rw:
1141 if (size_change)
1142 ocfs2_rw_unlock(inode, 1);
1143 bail:
1144 if (bh)
1145 brelse(bh);
1146
1147 mlog_exit(status);
1148 return status;
1149 }
1150
1151 int ocfs2_getattr(struct vfsmount *mnt,
1152 struct dentry *dentry,
1153 struct kstat *stat)
1154 {
1155 struct inode *inode = dentry->d_inode;
1156 struct super_block *sb = dentry->d_inode->i_sb;
1157 struct ocfs2_super *osb = sb->s_fs_info;
1158 int err;
1159
1160 mlog_entry_void();
1161
1162 err = ocfs2_inode_revalidate(dentry);
1163 if (err) {
1164 if (err != -ENOENT)
1165 mlog_errno(err);
1166 goto bail;
1167 }
1168
1169 generic_fillattr(inode, stat);
1170
1171 /* We set the blksize from the cluster size for performance */
1172 stat->blksize = osb->s_clustersize;
1173
1174 bail:
1175 mlog_exit(err);
1176
1177 return err;
1178 }
1179
1180 int ocfs2_permission(struct inode *inode, int mask)
1181 {
1182 int ret;
1183
1184 mlog_entry_void();
1185
1186 ret = ocfs2_inode_lock(inode, NULL, 0);
1187 if (ret) {
1188 if (ret != -ENOENT)
1189 mlog_errno(ret);
1190 goto out;
1191 }
1192
1193 ret = generic_permission(inode, mask, NULL);
1194
1195 ocfs2_inode_unlock(inode, 0);
1196 out:
1197 mlog_exit(ret);
1198 return ret;
1199 }
1200
1201 static int __ocfs2_write_remove_suid(struct inode *inode,
1202 struct buffer_head *bh)
1203 {
1204 int ret;
1205 handle_t *handle;
1206 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1207 struct ocfs2_dinode *di;
1208
1209 mlog_entry("(Inode %llu, mode 0%o)\n",
1210 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1211
1212 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1213 if (handle == NULL) {
1214 ret = -ENOMEM;
1215 mlog_errno(ret);
1216 goto out;
1217 }
1218
1219 ret = ocfs2_journal_access(handle, inode, bh,
1220 OCFS2_JOURNAL_ACCESS_WRITE);
1221 if (ret < 0) {
1222 mlog_errno(ret);
1223 goto out_trans;
1224 }
1225
1226 inode->i_mode &= ~S_ISUID;
1227 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1228 inode->i_mode &= ~S_ISGID;
1229
1230 di = (struct ocfs2_dinode *) bh->b_data;
1231 di->i_mode = cpu_to_le16(inode->i_mode);
1232
1233 ret = ocfs2_journal_dirty(handle, bh);
1234 if (ret < 0)
1235 mlog_errno(ret);
1236
1237 out_trans:
1238 ocfs2_commit_trans(osb, handle);
1239 out:
1240 mlog_exit(ret);
1241 return ret;
1242 }
1243
1244 /*
1245 * Will look for holes and unwritten extents in the range starting at
1246 * pos for count bytes (inclusive).
1247 */
1248 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1249 size_t count)
1250 {
1251 int ret = 0;
1252 unsigned int extent_flags;
1253 u32 cpos, clusters, extent_len, phys_cpos;
1254 struct super_block *sb = inode->i_sb;
1255
1256 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1257 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1258
1259 while (clusters) {
1260 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1261 &extent_flags);
1262 if (ret < 0) {
1263 mlog_errno(ret);
1264 goto out;
1265 }
1266
1267 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1268 ret = 1;
1269 break;
1270 }
1271
1272 if (extent_len > clusters)
1273 extent_len = clusters;
1274
1275 clusters -= extent_len;
1276 cpos += extent_len;
1277 }
1278 out:
1279 return ret;
1280 }
1281
1282 static int ocfs2_write_remove_suid(struct inode *inode)
1283 {
1284 int ret;
1285 struct buffer_head *bh = NULL;
1286 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1287
1288 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1289 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1290 if (ret < 0) {
1291 mlog_errno(ret);
1292 goto out;
1293 }
1294
1295 ret = __ocfs2_write_remove_suid(inode, bh);
1296 out:
1297 brelse(bh);
1298 return ret;
1299 }
1300
1301 /*
1302 * Allocate enough extents to cover the region starting at byte offset
1303 * start for len bytes. Existing extents are skipped, any extents
1304 * added are marked as "unwritten".
1305 */
1306 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1307 u64 start, u64 len)
1308 {
1309 int ret;
1310 u32 cpos, phys_cpos, clusters, alloc_size;
1311 u64 end = start + len;
1312 struct buffer_head *di_bh = NULL;
1313
1314 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1315 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1316 OCFS2_I(inode)->ip_blkno, &di_bh,
1317 OCFS2_BH_CACHED, inode);
1318 if (ret) {
1319 mlog_errno(ret);
1320 goto out;
1321 }
1322
1323 /*
1324 * Nothing to do if the requested reservation range
1325 * fits within the inode.
1326 */
1327 if (ocfs2_size_fits_inline_data(di_bh, end))
1328 goto out;
1329
1330 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1331 if (ret) {
1332 mlog_errno(ret);
1333 goto out;
1334 }
1335 }
1336
1337 /*
1338 * We consider both start and len to be inclusive.
1339 */
1340 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1341 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1342 clusters -= cpos;
1343
1344 while (clusters) {
1345 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1346 &alloc_size, NULL);
1347 if (ret) {
1348 mlog_errno(ret);
1349 goto out;
1350 }
1351
1352 /*
1353 * Hole or existing extent len can be arbitrary, so
1354 * cap it to our own allocation request.
1355 */
1356 if (alloc_size > clusters)
1357 alloc_size = clusters;
1358
1359 if (phys_cpos) {
1360 /*
1361 * We already have an allocation at this
1362 * region so we can safely skip it.
1363 */
1364 goto next;
1365 }
1366
1367 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1368 if (ret) {
1369 if (ret != -ENOSPC)
1370 mlog_errno(ret);
1371 goto out;
1372 }
1373
1374 next:
1375 cpos += alloc_size;
1376 clusters -= alloc_size;
1377 }
1378
1379 ret = 0;
1380 out:
1381
1382 brelse(di_bh);
1383 return ret;
1384 }
1385
1386 static int __ocfs2_remove_inode_range(struct inode *inode,
1387 struct buffer_head *di_bh,
1388 u32 cpos, u32 phys_cpos, u32 len,
1389 struct ocfs2_cached_dealloc_ctxt *dealloc)
1390 {
1391 int ret;
1392 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1393 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1394 struct inode *tl_inode = osb->osb_tl_inode;
1395 handle_t *handle;
1396 struct ocfs2_alloc_context *meta_ac = NULL;
1397 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1398
1399 ret = ocfs2_lock_allocators(inode, di_bh, 0, 1, NULL, &meta_ac);
1400 if (ret) {
1401 mlog_errno(ret);
1402 return ret;
1403 }
1404
1405 mutex_lock(&tl_inode->i_mutex);
1406
1407 if (ocfs2_truncate_log_needs_flush(osb)) {
1408 ret = __ocfs2_flush_truncate_log(osb);
1409 if (ret < 0) {
1410 mlog_errno(ret);
1411 goto out;
1412 }
1413 }
1414
1415 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1416 if (handle == NULL) {
1417 ret = -ENOMEM;
1418 mlog_errno(ret);
1419 goto out;
1420 }
1421
1422 ret = ocfs2_journal_access(handle, inode, di_bh,
1423 OCFS2_JOURNAL_ACCESS_WRITE);
1424 if (ret) {
1425 mlog_errno(ret);
1426 goto out;
1427 }
1428
1429 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1430 dealloc);
1431 if (ret) {
1432 mlog_errno(ret);
1433 goto out_commit;
1434 }
1435
1436 OCFS2_I(inode)->ip_clusters -= len;
1437 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1438
1439 ret = ocfs2_journal_dirty(handle, di_bh);
1440 if (ret) {
1441 mlog_errno(ret);
1442 goto out_commit;
1443 }
1444
1445 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1446 if (ret)
1447 mlog_errno(ret);
1448
1449 out_commit:
1450 ocfs2_commit_trans(osb, handle);
1451 out:
1452 mutex_unlock(&tl_inode->i_mutex);
1453
1454 if (meta_ac)
1455 ocfs2_free_alloc_context(meta_ac);
1456
1457 return ret;
1458 }
1459
1460 /*
1461 * Truncate a byte range, avoiding pages within partial clusters. This
1462 * preserves those pages for the zeroing code to write to.
1463 */
1464 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1465 u64 byte_len)
1466 {
1467 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1468 loff_t start, end;
1469 struct address_space *mapping = inode->i_mapping;
1470
1471 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1472 end = byte_start + byte_len;
1473 end = end & ~(osb->s_clustersize - 1);
1474
1475 if (start < end) {
1476 unmap_mapping_range(mapping, start, end - start, 0);
1477 truncate_inode_pages_range(mapping, start, end - 1);
1478 }
1479 }
1480
1481 static int ocfs2_zero_partial_clusters(struct inode *inode,
1482 u64 start, u64 len)
1483 {
1484 int ret = 0;
1485 u64 tmpend, end = start + len;
1486 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1487 unsigned int csize = osb->s_clustersize;
1488 handle_t *handle;
1489
1490 /*
1491 * The "start" and "end" values are NOT necessarily part of
1492 * the range whose allocation is being deleted. Rather, this
1493 * is what the user passed in with the request. We must zero
1494 * partial clusters here. There's no need to worry about
1495 * physical allocation - the zeroing code knows to skip holes.
1496 */
1497 mlog(0, "byte start: %llu, end: %llu\n",
1498 (unsigned long long)start, (unsigned long long)end);
1499
1500 /*
1501 * If both edges are on a cluster boundary then there's no
1502 * zeroing required as the region is part of the allocation to
1503 * be truncated.
1504 */
1505 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1506 goto out;
1507
1508 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1509 if (handle == NULL) {
1510 ret = -ENOMEM;
1511 mlog_errno(ret);
1512 goto out;
1513 }
1514
1515 /*
1516 * We want to get the byte offset of the end of the 1st cluster.
1517 */
1518 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1519 if (tmpend > end)
1520 tmpend = end;
1521
1522 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1523 (unsigned long long)start, (unsigned long long)tmpend);
1524
1525 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1526 if (ret)
1527 mlog_errno(ret);
1528
1529 if (tmpend < end) {
1530 /*
1531 * This may make start and end equal, but the zeroing
1532 * code will skip any work in that case so there's no
1533 * need to catch it up here.
1534 */
1535 start = end & ~(osb->s_clustersize - 1);
1536
1537 mlog(0, "2nd range: start: %llu, end: %llu\n",
1538 (unsigned long long)start, (unsigned long long)end);
1539
1540 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1541 if (ret)
1542 mlog_errno(ret);
1543 }
1544
1545 ocfs2_commit_trans(osb, handle);
1546 out:
1547 return ret;
1548 }
1549
1550 static int ocfs2_remove_inode_range(struct inode *inode,
1551 struct buffer_head *di_bh, u64 byte_start,
1552 u64 byte_len)
1553 {
1554 int ret = 0;
1555 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1556 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1557 struct ocfs2_cached_dealloc_ctxt dealloc;
1558 struct address_space *mapping = inode->i_mapping;
1559
1560 ocfs2_init_dealloc_ctxt(&dealloc);
1561
1562 if (byte_len == 0)
1563 return 0;
1564
1565 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1566 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1567 byte_start + byte_len, 0);
1568 if (ret) {
1569 mlog_errno(ret);
1570 goto out;
1571 }
1572 /*
1573 * There's no need to get fancy with the page cache
1574 * truncate of an inline-data inode. We're talking
1575 * about less than a page here, which will be cached
1576 * in the dinode buffer anyway.
1577 */
1578 unmap_mapping_range(mapping, 0, 0, 0);
1579 truncate_inode_pages(mapping, 0);
1580 goto out;
1581 }
1582
1583 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1584 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1585 if (trunc_len >= trunc_start)
1586 trunc_len -= trunc_start;
1587 else
1588 trunc_len = 0;
1589
1590 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1591 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1592 (unsigned long long)byte_start,
1593 (unsigned long long)byte_len, trunc_start, trunc_len);
1594
1595 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1596 if (ret) {
1597 mlog_errno(ret);
1598 goto out;
1599 }
1600
1601 cpos = trunc_start;
1602 while (trunc_len) {
1603 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1604 &alloc_size, NULL);
1605 if (ret) {
1606 mlog_errno(ret);
1607 goto out;
1608 }
1609
1610 if (alloc_size > trunc_len)
1611 alloc_size = trunc_len;
1612
1613 /* Only do work for non-holes */
1614 if (phys_cpos != 0) {
1615 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1616 phys_cpos, alloc_size,
1617 &dealloc);
1618 if (ret) {
1619 mlog_errno(ret);
1620 goto out;
1621 }
1622 }
1623
1624 cpos += alloc_size;
1625 trunc_len -= alloc_size;
1626 }
1627
1628 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1629
1630 out:
1631 ocfs2_schedule_truncate_log_flush(osb, 1);
1632 ocfs2_run_deallocs(osb, &dealloc);
1633
1634 return ret;
1635 }
1636
1637 /*
1638 * Parts of this function taken from xfs_change_file_space()
1639 */
1640 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1641 loff_t f_pos, unsigned int cmd,
1642 struct ocfs2_space_resv *sr,
1643 int change_size)
1644 {
1645 int ret;
1646 s64 llen;
1647 loff_t size;
1648 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1649 struct buffer_head *di_bh = NULL;
1650 handle_t *handle;
1651 unsigned long long max_off = inode->i_sb->s_maxbytes;
1652
1653 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1654 return -EROFS;
1655
1656 mutex_lock(&inode->i_mutex);
1657
1658 /*
1659 * This prevents concurrent writes on other nodes
1660 */
1661 ret = ocfs2_rw_lock(inode, 1);
1662 if (ret) {
1663 mlog_errno(ret);
1664 goto out;
1665 }
1666
1667 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1668 if (ret) {
1669 mlog_errno(ret);
1670 goto out_rw_unlock;
1671 }
1672
1673 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1674 ret = -EPERM;
1675 goto out_inode_unlock;
1676 }
1677
1678 switch (sr->l_whence) {
1679 case 0: /*SEEK_SET*/
1680 break;
1681 case 1: /*SEEK_CUR*/
1682 sr->l_start += f_pos;
1683 break;
1684 case 2: /*SEEK_END*/
1685 sr->l_start += i_size_read(inode);
1686 break;
1687 default:
1688 ret = -EINVAL;
1689 goto out_inode_unlock;
1690 }
1691 sr->l_whence = 0;
1692
1693 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1694
1695 if (sr->l_start < 0
1696 || sr->l_start > max_off
1697 || (sr->l_start + llen) < 0
1698 || (sr->l_start + llen) > max_off) {
1699 ret = -EINVAL;
1700 goto out_inode_unlock;
1701 }
1702 size = sr->l_start + sr->l_len;
1703
1704 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1705 if (sr->l_len <= 0) {
1706 ret = -EINVAL;
1707 goto out_inode_unlock;
1708 }
1709 }
1710
1711 if (file && should_remove_suid(file->f_path.dentry)) {
1712 ret = __ocfs2_write_remove_suid(inode, di_bh);
1713 if (ret) {
1714 mlog_errno(ret);
1715 goto out_inode_unlock;
1716 }
1717 }
1718
1719 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1720 switch (cmd) {
1721 case OCFS2_IOC_RESVSP:
1722 case OCFS2_IOC_RESVSP64:
1723 /*
1724 * This takes unsigned offsets, but the signed ones we
1725 * pass have been checked against overflow above.
1726 */
1727 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1728 sr->l_len);
1729 break;
1730 case OCFS2_IOC_UNRESVSP:
1731 case OCFS2_IOC_UNRESVSP64:
1732 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1733 sr->l_len);
1734 break;
1735 default:
1736 ret = -EINVAL;
1737 }
1738 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1739 if (ret) {
1740 mlog_errno(ret);
1741 goto out_inode_unlock;
1742 }
1743
1744 /*
1745 * We update c/mtime for these changes
1746 */
1747 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1748 if (IS_ERR(handle)) {
1749 ret = PTR_ERR(handle);
1750 mlog_errno(ret);
1751 goto out_inode_unlock;
1752 }
1753
1754 if (change_size && i_size_read(inode) < size)
1755 i_size_write(inode, size);
1756
1757 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1758 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1759 if (ret < 0)
1760 mlog_errno(ret);
1761
1762 ocfs2_commit_trans(osb, handle);
1763
1764 out_inode_unlock:
1765 brelse(di_bh);
1766 ocfs2_inode_unlock(inode, 1);
1767 out_rw_unlock:
1768 ocfs2_rw_unlock(inode, 1);
1769
1770 out:
1771 mutex_unlock(&inode->i_mutex);
1772 return ret;
1773 }
1774
1775 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1776 struct ocfs2_space_resv *sr)
1777 {
1778 struct inode *inode = file->f_path.dentry->d_inode;
1779 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1780
1781 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1782 !ocfs2_writes_unwritten_extents(osb))
1783 return -ENOTTY;
1784 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1785 !ocfs2_sparse_alloc(osb))
1786 return -ENOTTY;
1787
1788 if (!S_ISREG(inode->i_mode))
1789 return -EINVAL;
1790
1791 if (!(file->f_mode & FMODE_WRITE))
1792 return -EBADF;
1793
1794 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1795 }
1796
1797 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1798 loff_t len)
1799 {
1800 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1801 struct ocfs2_space_resv sr;
1802 int change_size = 1;
1803
1804 if (!ocfs2_writes_unwritten_extents(osb))
1805 return -EOPNOTSUPP;
1806
1807 if (S_ISDIR(inode->i_mode))
1808 return -ENODEV;
1809
1810 if (mode & FALLOC_FL_KEEP_SIZE)
1811 change_size = 0;
1812
1813 sr.l_whence = 0;
1814 sr.l_start = (s64)offset;
1815 sr.l_len = (s64)len;
1816
1817 return __ocfs2_change_file_space(NULL, inode, offset,
1818 OCFS2_IOC_RESVSP64, &sr, change_size);
1819 }
1820
1821 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1822 loff_t *ppos,
1823 size_t count,
1824 int appending,
1825 int *direct_io)
1826 {
1827 int ret = 0, meta_level = 0;
1828 struct inode *inode = dentry->d_inode;
1829 loff_t saved_pos, end;
1830
1831 /*
1832 * We start with a read level meta lock and only jump to an ex
1833 * if we need to make modifications here.
1834 */
1835 for(;;) {
1836 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1837 if (ret < 0) {
1838 meta_level = -1;
1839 mlog_errno(ret);
1840 goto out;
1841 }
1842
1843 /* Clear suid / sgid if necessary. We do this here
1844 * instead of later in the write path because
1845 * remove_suid() calls ->setattr without any hint that
1846 * we may have already done our cluster locking. Since
1847 * ocfs2_setattr() *must* take cluster locks to
1848 * proceeed, this will lead us to recursively lock the
1849 * inode. There's also the dinode i_size state which
1850 * can be lost via setattr during extending writes (we
1851 * set inode->i_size at the end of a write. */
1852 if (should_remove_suid(dentry)) {
1853 if (meta_level == 0) {
1854 ocfs2_inode_unlock(inode, meta_level);
1855 meta_level = 1;
1856 continue;
1857 }
1858
1859 ret = ocfs2_write_remove_suid(inode);
1860 if (ret < 0) {
1861 mlog_errno(ret);
1862 goto out_unlock;
1863 }
1864 }
1865
1866 /* work on a copy of ppos until we're sure that we won't have
1867 * to recalculate it due to relocking. */
1868 if (appending) {
1869 saved_pos = i_size_read(inode);
1870 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1871 } else {
1872 saved_pos = *ppos;
1873 }
1874
1875 end = saved_pos + count;
1876
1877 /*
1878 * Skip the O_DIRECT checks if we don't need
1879 * them.
1880 */
1881 if (!direct_io || !(*direct_io))
1882 break;
1883
1884 /*
1885 * There's no sane way to do direct writes to an inode
1886 * with inline data.
1887 */
1888 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1889 *direct_io = 0;
1890 break;
1891 }
1892
1893 /*
1894 * Allowing concurrent direct writes means
1895 * i_size changes wouldn't be synchronized, so
1896 * one node could wind up truncating another
1897 * nodes writes.
1898 */
1899 if (end > i_size_read(inode)) {
1900 *direct_io = 0;
1901 break;
1902 }
1903
1904 /*
1905 * We don't fill holes during direct io, so
1906 * check for them here. If any are found, the
1907 * caller will have to retake some cluster
1908 * locks and initiate the io as buffered.
1909 */
1910 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1911 if (ret == 1) {
1912 *direct_io = 0;
1913 ret = 0;
1914 } else if (ret < 0)
1915 mlog_errno(ret);
1916 break;
1917 }
1918
1919 if (appending)
1920 *ppos = saved_pos;
1921
1922 out_unlock:
1923 ocfs2_inode_unlock(inode, meta_level);
1924
1925 out:
1926 return ret;
1927 }
1928
1929 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1930 const struct iovec *iov,
1931 unsigned long nr_segs,
1932 loff_t pos)
1933 {
1934 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1935 int can_do_direct;
1936 ssize_t written = 0;
1937 size_t ocount; /* original count */
1938 size_t count; /* after file limit checks */
1939 loff_t old_size, *ppos = &iocb->ki_pos;
1940 u32 old_clusters;
1941 struct file *file = iocb->ki_filp;
1942 struct inode *inode = file->f_path.dentry->d_inode;
1943 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1944
1945 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1946 (unsigned int)nr_segs,
1947 file->f_path.dentry->d_name.len,
1948 file->f_path.dentry->d_name.name);
1949
1950 if (iocb->ki_left == 0)
1951 return 0;
1952
1953 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1954
1955 appending = file->f_flags & O_APPEND ? 1 : 0;
1956 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1957
1958 mutex_lock(&inode->i_mutex);
1959
1960 relock:
1961 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1962 if (direct_io) {
1963 down_read(&inode->i_alloc_sem);
1964 have_alloc_sem = 1;
1965 }
1966
1967 /* concurrent O_DIRECT writes are allowed */
1968 rw_level = !direct_io;
1969 ret = ocfs2_rw_lock(inode, rw_level);
1970 if (ret < 0) {
1971 mlog_errno(ret);
1972 goto out_sems;
1973 }
1974
1975 can_do_direct = direct_io;
1976 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1977 iocb->ki_left, appending,
1978 &can_do_direct);
1979 if (ret < 0) {
1980 mlog_errno(ret);
1981 goto out;
1982 }
1983
1984 /*
1985 * We can't complete the direct I/O as requested, fall back to
1986 * buffered I/O.
1987 */
1988 if (direct_io && !can_do_direct) {
1989 ocfs2_rw_unlock(inode, rw_level);
1990 up_read(&inode->i_alloc_sem);
1991
1992 have_alloc_sem = 0;
1993 rw_level = -1;
1994
1995 direct_io = 0;
1996 goto relock;
1997 }
1998
1999 /*
2000 * To later detect whether a journal commit for sync writes is
2001 * necessary, we sample i_size, and cluster count here.
2002 */
2003 old_size = i_size_read(inode);
2004 old_clusters = OCFS2_I(inode)->ip_clusters;
2005
2006 /* communicate with ocfs2_dio_end_io */
2007 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2008
2009 if (direct_io) {
2010 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2011 VERIFY_READ);
2012 if (ret)
2013 goto out_dio;
2014
2015 ret = generic_write_checks(file, ppos, &count,
2016 S_ISBLK(inode->i_mode));
2017 if (ret)
2018 goto out_dio;
2019
2020 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2021 ppos, count, ocount);
2022 if (written < 0) {
2023 ret = written;
2024 goto out_dio;
2025 }
2026 } else {
2027 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
2028 *ppos);
2029 }
2030
2031 out_dio:
2032 /* buffered aio wouldn't have proper lock coverage today */
2033 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2034
2035 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
2036 /*
2037 * The generic write paths have handled getting data
2038 * to disk, but since we don't make use of the dirty
2039 * inode list, a manual journal commit is necessary
2040 * here.
2041 */
2042 if (old_size != i_size_read(inode) ||
2043 old_clusters != OCFS2_I(inode)->ip_clusters) {
2044 ret = journal_force_commit(osb->journal->j_journal);
2045 if (ret < 0)
2046 written = ret;
2047 }
2048 }
2049
2050 /*
2051 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2052 * function pointer which is called when o_direct io completes so that
2053 * it can unlock our rw lock. (it's the clustered equivalent of
2054 * i_alloc_sem; protects truncate from racing with pending ios).
2055 * Unfortunately there are error cases which call end_io and others
2056 * that don't. so we don't have to unlock the rw_lock if either an
2057 * async dio is going to do it in the future or an end_io after an
2058 * error has already done it.
2059 */
2060 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2061 rw_level = -1;
2062 have_alloc_sem = 0;
2063 }
2064
2065 out:
2066 if (rw_level != -1)
2067 ocfs2_rw_unlock(inode, rw_level);
2068
2069 out_sems:
2070 if (have_alloc_sem)
2071 up_read(&inode->i_alloc_sem);
2072
2073 mutex_unlock(&inode->i_mutex);
2074
2075 mlog_exit(ret);
2076 return written ? written : ret;
2077 }
2078
2079 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2080 struct file *out,
2081 loff_t *ppos,
2082 size_t len,
2083 unsigned int flags)
2084 {
2085 int ret;
2086 struct inode *inode = out->f_path.dentry->d_inode;
2087
2088 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2089 (unsigned int)len,
2090 out->f_path.dentry->d_name.len,
2091 out->f_path.dentry->d_name.name);
2092
2093 inode_double_lock(inode, pipe->inode);
2094
2095 ret = ocfs2_rw_lock(inode, 1);
2096 if (ret < 0) {
2097 mlog_errno(ret);
2098 goto out;
2099 }
2100
2101 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2102 NULL);
2103 if (ret < 0) {
2104 mlog_errno(ret);
2105 goto out_unlock;
2106 }
2107
2108 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
2109
2110 out_unlock:
2111 ocfs2_rw_unlock(inode, 1);
2112 out:
2113 inode_double_unlock(inode, pipe->inode);
2114
2115 mlog_exit(ret);
2116 return ret;
2117 }
2118
2119 static ssize_t ocfs2_file_splice_read(struct file *in,
2120 loff_t *ppos,
2121 struct pipe_inode_info *pipe,
2122 size_t len,
2123 unsigned int flags)
2124 {
2125 int ret = 0;
2126 struct inode *inode = in->f_path.dentry->d_inode;
2127
2128 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2129 (unsigned int)len,
2130 in->f_path.dentry->d_name.len,
2131 in->f_path.dentry->d_name.name);
2132
2133 /*
2134 * See the comment in ocfs2_file_aio_read()
2135 */
2136 ret = ocfs2_inode_lock(inode, NULL, 0);
2137 if (ret < 0) {
2138 mlog_errno(ret);
2139 goto bail;
2140 }
2141 ocfs2_inode_unlock(inode, 0);
2142
2143 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2144
2145 bail:
2146 mlog_exit(ret);
2147 return ret;
2148 }
2149
2150 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2151 const struct iovec *iov,
2152 unsigned long nr_segs,
2153 loff_t pos)
2154 {
2155 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2156 struct file *filp = iocb->ki_filp;
2157 struct inode *inode = filp->f_path.dentry->d_inode;
2158
2159 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2160 (unsigned int)nr_segs,
2161 filp->f_path.dentry->d_name.len,
2162 filp->f_path.dentry->d_name.name);
2163
2164 if (!inode) {
2165 ret = -EINVAL;
2166 mlog_errno(ret);
2167 goto bail;
2168 }
2169
2170 /*
2171 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2172 * need locks to protect pending reads from racing with truncate.
2173 */
2174 if (filp->f_flags & O_DIRECT) {
2175 down_read(&inode->i_alloc_sem);
2176 have_alloc_sem = 1;
2177
2178 ret = ocfs2_rw_lock(inode, 0);
2179 if (ret < 0) {
2180 mlog_errno(ret);
2181 goto bail;
2182 }
2183 rw_level = 0;
2184 /* communicate with ocfs2_dio_end_io */
2185 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2186 }
2187
2188 /*
2189 * We're fine letting folks race truncates and extending
2190 * writes with read across the cluster, just like they can
2191 * locally. Hence no rw_lock during read.
2192 *
2193 * Take and drop the meta data lock to update inode fields
2194 * like i_size. This allows the checks down below
2195 * generic_file_aio_read() a chance of actually working.
2196 */
2197 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2198 if (ret < 0) {
2199 mlog_errno(ret);
2200 goto bail;
2201 }
2202 ocfs2_inode_unlock(inode, lock_level);
2203
2204 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2205 if (ret == -EINVAL)
2206 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2207
2208 /* buffered aio wouldn't have proper lock coverage today */
2209 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2210
2211 /* see ocfs2_file_aio_write */
2212 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2213 rw_level = -1;
2214 have_alloc_sem = 0;
2215 }
2216
2217 bail:
2218 if (have_alloc_sem)
2219 up_read(&inode->i_alloc_sem);
2220 if (rw_level != -1)
2221 ocfs2_rw_unlock(inode, rw_level);
2222 mlog_exit(ret);
2223
2224 return ret;
2225 }
2226
2227 const struct inode_operations ocfs2_file_iops = {
2228 .setattr = ocfs2_setattr,
2229 .getattr = ocfs2_getattr,
2230 .permission = ocfs2_permission,
2231 .fallocate = ocfs2_fallocate,
2232 .fiemap = ocfs2_fiemap,
2233 };
2234
2235 const struct inode_operations ocfs2_special_file_iops = {
2236 .setattr = ocfs2_setattr,
2237 .getattr = ocfs2_getattr,
2238 .permission = ocfs2_permission,
2239 };
2240
2241 /*
2242 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2243 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2244 */
2245 const struct file_operations ocfs2_fops = {
2246 .llseek = generic_file_llseek,
2247 .read = do_sync_read,
2248 .write = do_sync_write,
2249 .mmap = ocfs2_mmap,
2250 .fsync = ocfs2_sync_file,
2251 .release = ocfs2_file_release,
2252 .open = ocfs2_file_open,
2253 .aio_read = ocfs2_file_aio_read,
2254 .aio_write = ocfs2_file_aio_write,
2255 .unlocked_ioctl = ocfs2_ioctl,
2256 #ifdef CONFIG_COMPAT
2257 .compat_ioctl = ocfs2_compat_ioctl,
2258 #endif
2259 .lock = ocfs2_lock,
2260 .flock = ocfs2_flock,
2261 .splice_read = ocfs2_file_splice_read,
2262 .splice_write = ocfs2_file_splice_write,
2263 };
2264
2265 const struct file_operations ocfs2_dops = {
2266 .llseek = generic_file_llseek,
2267 .read = generic_read_dir,
2268 .readdir = ocfs2_readdir,
2269 .fsync = ocfs2_sync_file,
2270 .release = ocfs2_dir_release,
2271 .open = ocfs2_dir_open,
2272 .unlocked_ioctl = ocfs2_ioctl,
2273 #ifdef CONFIG_COMPAT
2274 .compat_ioctl = ocfs2_compat_ioctl,
2275 #endif
2276 .lock = ocfs2_lock,
2277 .flock = ocfs2_flock,
2278 };
2279
2280 /*
2281 * POSIX-lockless variants of our file_operations.
2282 *
2283 * These will be used if the underlying cluster stack does not support
2284 * posix file locking, if the user passes the "localflocks" mount
2285 * option, or if we have a local-only fs.
2286 *
2287 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2288 * so we still want it in the case of no stack support for
2289 * plocks. Internally, it will do the right thing when asked to ignore
2290 * the cluster.
2291 */
2292 const struct file_operations ocfs2_fops_no_plocks = {
2293 .llseek = generic_file_llseek,
2294 .read = do_sync_read,
2295 .write = do_sync_write,
2296 .mmap = ocfs2_mmap,
2297 .fsync = ocfs2_sync_file,
2298 .release = ocfs2_file_release,
2299 .open = ocfs2_file_open,
2300 .aio_read = ocfs2_file_aio_read,
2301 .aio_write = ocfs2_file_aio_write,
2302 .unlocked_ioctl = ocfs2_ioctl,
2303 #ifdef CONFIG_COMPAT
2304 .compat_ioctl = ocfs2_compat_ioctl,
2305 #endif
2306 .flock = ocfs2_flock,
2307 .splice_read = ocfs2_file_splice_read,
2308 .splice_write = ocfs2_file_splice_write,
2309 };
2310
2311 const struct file_operations ocfs2_dops_no_plocks = {
2312 .llseek = generic_file_llseek,
2313 .read = generic_read_dir,
2314 .readdir = ocfs2_readdir,
2315 .fsync = ocfs2_sync_file,
2316 .release = ocfs2_dir_release,
2317 .open = ocfs2_dir_open,
2318 .unlocked_ioctl = ocfs2_ioctl,
2319 #ifdef CONFIG_COMPAT
2320 .compat_ioctl = ocfs2_compat_ioctl,
2321 #endif
2322 .flock = ocfs2_flock,
2323 };