projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ocfs2: Wrap inode block reads in a dedicated function.
[deliverable/linux.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 #include "xattr.h"
59 #include "acl.h"
60
61 #include "buffer_head_io.h"
62
63 static int ocfs2_sync_inode(struct inode *inode)
64 {
65 filemap_fdatawrite(inode->i_mapping);
66 return sync_mapping_buffers(inode->i_mapping);
67 }
68
69 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
70 {
71 struct ocfs2_file_private *fp;
72
73 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
74 if (!fp)
75 return -ENOMEM;
76
77 fp->fp_file = file;
78 mutex_init(&fp->fp_mutex);
79 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
80 file->private_data = fp;
81
82 return 0;
83 }
84
85 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
86 {
87 struct ocfs2_file_private *fp = file->private_data;
88 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
89
90 if (fp) {
91 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
92 ocfs2_lock_res_free(&fp->fp_flock);
93 kfree(fp);
94 file->private_data = NULL;
95 }
96 }
97
98 static int ocfs2_file_open(struct inode *inode, struct file *file)
99 {
100 int status;
101 int mode = file->f_flags;
102 struct ocfs2_inode_info *oi = OCFS2_I(inode);
103
104 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
105 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
106
107 spin_lock(&oi->ip_lock);
108
109 /* Check that the inode hasn't been wiped from disk by another
110 * node. If it hasn't then we're safe as long as we hold the
111 * spin lock until our increment of open count. */
112 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
113 spin_unlock(&oi->ip_lock);
114
115 status = -ENOENT;
116 goto leave;
117 }
118
119 if (mode & O_DIRECT)
120 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
121
122 oi->ip_open_count++;
123 spin_unlock(&oi->ip_lock);
124
125 status = ocfs2_init_file_private(inode, file);
126 if (status) {
127 /*
128 * We want to set open count back if we're failing the
129 * open.
130 */
131 spin_lock(&oi->ip_lock);
132 oi->ip_open_count--;
133 spin_unlock(&oi->ip_lock);
134 }
135
136 leave:
137 mlog_exit(status);
138 return status;
139 }
140
141 static int ocfs2_file_release(struct inode *inode, struct file *file)
142 {
143 struct ocfs2_inode_info *oi = OCFS2_I(inode);
144
145 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
146 file->f_path.dentry->d_name.len,
147 file->f_path.dentry->d_name.name);
148
149 spin_lock(&oi->ip_lock);
150 if (!--oi->ip_open_count)
151 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
152 spin_unlock(&oi->ip_lock);
153
154 ocfs2_free_file_private(inode, file);
155
156 mlog_exit(0);
157
158 return 0;
159 }
160
161 static int ocfs2_dir_open(struct inode *inode, struct file *file)
162 {
163 return ocfs2_init_file_private(inode, file);
164 }
165
166 static int ocfs2_dir_release(struct inode *inode, struct file *file)
167 {
168 ocfs2_free_file_private(inode, file);
169 return 0;
170 }
171
172 static int ocfs2_sync_file(struct file *file,
173 struct dentry *dentry,
174 int datasync)
175 {
176 int err = 0;
177 journal_t *journal;
178 struct inode *inode = dentry->d_inode;
179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180
181 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
182 dentry->d_name.len, dentry->d_name.name);
183
184 err = ocfs2_sync_inode(dentry->d_inode);
185 if (err)
186 goto bail;
187
188 journal = osb->journal->j_journal;
189 err = jbd2_journal_force_commit(journal);
190
191 bail:
192 mlog_exit(err);
193
194 return (err < 0) ? -EIO : 0;
195 }
196
197 int ocfs2_should_update_atime(struct inode *inode,
198 struct vfsmount *vfsmnt)
199 {
200 struct timespec now;
201 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
202
203 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
204 return 0;
205
206 if ((inode->i_flags & S_NOATIME) ||
207 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
208 return 0;
209
210 /*
211 * We can be called with no vfsmnt structure - NFSD will
212 * sometimes do this.
213 *
214 * Note that our action here is different than touch_atime() -
215 * if we can't tell whether this is a noatime mount, then we
216 * don't know whether to trust the value of s_atime_quantum.
217 */
218 if (vfsmnt == NULL)
219 return 0;
220
221 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
222 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
223 return 0;
224
225 if (vfsmnt->mnt_flags & MNT_RELATIME) {
226 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
227 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
228 return 1;
229
230 return 0;
231 }
232
233 now = CURRENT_TIME;
234 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
235 return 0;
236 else
237 return 1;
238 }
239
240 int ocfs2_update_inode_atime(struct inode *inode,
241 struct buffer_head *bh)
242 {
243 int ret;
244 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
245 handle_t *handle;
246 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
247
248 mlog_entry_void();
249
250 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
251 if (IS_ERR(handle)) {
252 ret = PTR_ERR(handle);
253 mlog_errno(ret);
254 goto out;
255 }
256
257 ret = ocfs2_journal_access(handle, inode, bh,
258 OCFS2_JOURNAL_ACCESS_WRITE);
259 if (ret) {
260 mlog_errno(ret);
261 goto out_commit;
262 }
263
264 /*
265 * Don't use ocfs2_mark_inode_dirty() here as we don't always
266 * have i_mutex to guard against concurrent changes to other
267 * inode fields.
268 */
269 inode->i_atime = CURRENT_TIME;
270 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
271 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
272
273 ret = ocfs2_journal_dirty(handle, bh);
274 if (ret < 0)
275 mlog_errno(ret);
276
277 out_commit:
278 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
279 out:
280 mlog_exit(ret);
281 return ret;
282 }
283
284 static int ocfs2_set_inode_size(handle_t *handle,
285 struct inode *inode,
286 struct buffer_head *fe_bh,
287 u64 new_i_size)
288 {
289 int status;
290
291 mlog_entry_void();
292 i_size_write(inode, new_i_size);
293 inode->i_blocks = ocfs2_inode_sector_count(inode);
294 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
295
296 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
297 if (status < 0) {
298 mlog_errno(status);
299 goto bail;
300 }
301
302 bail:
303 mlog_exit(status);
304 return status;
305 }
306
307 static int ocfs2_simple_size_update(struct inode *inode,
308 struct buffer_head *di_bh,
309 u64 new_i_size)
310 {
311 int ret;
312 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
313 handle_t *handle = NULL;
314
315 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
316 if (IS_ERR(handle)) {
317 ret = PTR_ERR(handle);
318 mlog_errno(ret);
319 goto out;
320 }
321
322 ret = ocfs2_set_inode_size(handle, inode, di_bh,
323 new_i_size);
324 if (ret < 0)
325 mlog_errno(ret);
326
327 ocfs2_commit_trans(osb, handle);
328 out:
329 return ret;
330 }
331
332 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
333 struct inode *inode,
334 struct buffer_head *fe_bh,
335 u64 new_i_size)
336 {
337 int status;
338 handle_t *handle;
339 struct ocfs2_dinode *di;
340 u64 cluster_bytes;
341
342 mlog_entry_void();
343
344 /* TODO: This needs to actually orphan the inode in this
345 * transaction. */
346
347 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
348 if (IS_ERR(handle)) {
349 status = PTR_ERR(handle);
350 mlog_errno(status);
351 goto out;
352 }
353
354 status = ocfs2_journal_access(handle, inode, fe_bh,
355 OCFS2_JOURNAL_ACCESS_WRITE);
356 if (status < 0) {
357 mlog_errno(status);
358 goto out_commit;
359 }
360
361 /*
362 * Do this before setting i_size.
363 */
364 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
365 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
366 cluster_bytes);
367 if (status) {
368 mlog_errno(status);
369 goto out_commit;
370 }
371
372 i_size_write(inode, new_i_size);
373 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
374
375 di = (struct ocfs2_dinode *) fe_bh->b_data;
376 di->i_size = cpu_to_le64(new_i_size);
377 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
378 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
379
380 status = ocfs2_journal_dirty(handle, fe_bh);
381 if (status < 0)
382 mlog_errno(status);
383
384 out_commit:
385 ocfs2_commit_trans(osb, handle);
386 out:
387
388 mlog_exit(status);
389 return status;
390 }
391
392 static int ocfs2_truncate_file(struct inode *inode,
393 struct buffer_head *di_bh,
394 u64 new_i_size)
395 {
396 int status = 0;
397 struct ocfs2_dinode *fe = NULL;
398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
399 struct ocfs2_truncate_context *tc = NULL;
400
401 mlog_entry("(inode = %llu, new_i_size = %llu\n",
402 (unsigned long long)OCFS2_I(inode)->ip_blkno,
403 (unsigned long long)new_i_size);
404
405 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
406 * already validated it */
407 fe = (struct ocfs2_dinode *) di_bh->b_data;
408
409 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
410 "Inode %llu, inode i_size = %lld != di "
411 "i_size = %llu, i_flags = 0x%x\n",
412 (unsigned long long)OCFS2_I(inode)->ip_blkno,
413 i_size_read(inode),
414 (unsigned long long)le64_to_cpu(fe->i_size),
415 le32_to_cpu(fe->i_flags));
416
417 if (new_i_size > le64_to_cpu(fe->i_size)) {
418 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
419 (unsigned long long)le64_to_cpu(fe->i_size),
420 (unsigned long long)new_i_size);
421 status = -EINVAL;
422 mlog_errno(status);
423 goto bail;
424 }
425
426 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
427 (unsigned long long)le64_to_cpu(fe->i_blkno),
428 (unsigned long long)le64_to_cpu(fe->i_size),
429 (unsigned long long)new_i_size);
430
431 /* lets handle the simple truncate cases before doing any more
432 * cluster locking. */
433 if (new_i_size == le64_to_cpu(fe->i_size))
434 goto bail;
435
436 down_write(&OCFS2_I(inode)->ip_alloc_sem);
437
438 /*
439 * The inode lock forced other nodes to sync and drop their
440 * pages, which (correctly) happens even if we have a truncate
441 * without allocation change - ocfs2 cluster sizes can be much
442 * greater than page size, so we have to truncate them
443 * anyway.
444 */
445 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
446 truncate_inode_pages(inode->i_mapping, new_i_size);
447
448 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
449 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
450 i_size_read(inode), 1);
451 if (status)
452 mlog_errno(status);
453
454 goto bail_unlock_sem;
455 }
456
457 /* alright, we're going to need to do a full blown alloc size
458 * change. Orphan the inode so that recovery can complete the
459 * truncate if necessary. This does the task of marking
460 * i_size. */
461 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
462 if (status < 0) {
463 mlog_errno(status);
464 goto bail_unlock_sem;
465 }
466
467 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
468 if (status < 0) {
469 mlog_errno(status);
470 goto bail_unlock_sem;
471 }
472
473 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
474 if (status < 0) {
475 mlog_errno(status);
476 goto bail_unlock_sem;
477 }
478
479 /* TODO: orphan dir cleanup here. */
480 bail_unlock_sem:
481 up_write(&OCFS2_I(inode)->ip_alloc_sem);
482
483 bail:
484
485 mlog_exit(status);
486 return status;
487 }
488
489 /*
490 * extend file allocation only here.
491 * we'll update all the disk stuff, and oip->alloc_size
492 *
493 * expect stuff to be locked, a transaction started and enough data /
494 * metadata reservations in the contexts.
495 *
496 * Will return -EAGAIN, and a reason if a restart is needed.
497 * If passed in, *reason will always be set, even in error.
498 */
499 int ocfs2_add_inode_data(struct ocfs2_super *osb,
500 struct inode *inode,
501 u32 *logical_offset,
502 u32 clusters_to_add,
503 int mark_unwritten,
504 struct buffer_head *fe_bh,
505 handle_t *handle,
506 struct ocfs2_alloc_context *data_ac,
507 struct ocfs2_alloc_context *meta_ac,
508 enum ocfs2_alloc_restarted *reason_ret)
509 {
510 int ret;
511 struct ocfs2_extent_tree et;
512
513 ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
514 ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
515 clusters_to_add, mark_unwritten,
516 &et, handle,
517 data_ac, meta_ac, reason_ret);
518
519 return ret;
520 }
521
522 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
523 u32 clusters_to_add, int mark_unwritten)
524 {
525 int status = 0;
526 int restart_func = 0;
527 int credits;
528 u32 prev_clusters;
529 struct buffer_head *bh = NULL;
530 struct ocfs2_dinode *fe = NULL;
531 handle_t *handle = NULL;
532 struct ocfs2_alloc_context *data_ac = NULL;
533 struct ocfs2_alloc_context *meta_ac = NULL;
534 enum ocfs2_alloc_restarted why;
535 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
536 struct ocfs2_extent_tree et;
537
538 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
539
540 /*
541 * This function only exists for file systems which don't
542 * support holes.
543 */
544 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
545
546 status = ocfs2_read_inode_block(inode, &bh);
547 if (status < 0) {
548 mlog_errno(status);
549 goto leave;
550 }
551 fe = (struct ocfs2_dinode *) bh->b_data;
552
553 restart_all:
554 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
555
556 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
557 "clusters_to_add = %u\n",
558 (unsigned long long)OCFS2_I(inode)->ip_blkno,
559 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
560 clusters_to_add);
561 ocfs2_init_dinode_extent_tree(&et, inode, bh);
562 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
563 &data_ac, &meta_ac);
564 if (status) {
565 mlog_errno(status);
566 goto leave;
567 }
568
569 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
570 clusters_to_add);
571 handle = ocfs2_start_trans(osb, credits);
572 if (IS_ERR(handle)) {
573 status = PTR_ERR(handle);
574 handle = NULL;
575 mlog_errno(status);
576 goto leave;
577 }
578
579 restarted_transaction:
580 /* reserve a write to the file entry early on - that we if we
581 * run out of credits in the allocation path, we can still
582 * update i_size. */
583 status = ocfs2_journal_access(handle, inode, bh,
584 OCFS2_JOURNAL_ACCESS_WRITE);
585 if (status < 0) {
586 mlog_errno(status);
587 goto leave;
588 }
589
590 prev_clusters = OCFS2_I(inode)->ip_clusters;
591
592 status = ocfs2_add_inode_data(osb,
593 inode,
594 &logical_start,
595 clusters_to_add,
596 mark_unwritten,
597 bh,
598 handle,
599 data_ac,
600 meta_ac,
601 &why);
602 if ((status < 0) && (status != -EAGAIN)) {
603 if (status != -ENOSPC)
604 mlog_errno(status);
605 goto leave;
606 }
607
608 status = ocfs2_journal_dirty(handle, bh);
609 if (status < 0) {
610 mlog_errno(status);
611 goto leave;
612 }
613
614 spin_lock(&OCFS2_I(inode)->ip_lock);
615 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
616 spin_unlock(&OCFS2_I(inode)->ip_lock);
617
618 if (why != RESTART_NONE && clusters_to_add) {
619 if (why == RESTART_META) {
620 mlog(0, "restarting function.\n");
621 restart_func = 1;
622 } else {
623 BUG_ON(why != RESTART_TRANS);
624
625 mlog(0, "restarting transaction.\n");
626 /* TODO: This can be more intelligent. */
627 credits = ocfs2_calc_extend_credits(osb->sb,
628 &fe->id2.i_list,
629 clusters_to_add);
630 status = ocfs2_extend_trans(handle, credits);
631 if (status < 0) {
632 /* handle still has to be committed at
633 * this point. */
634 status = -ENOMEM;
635 mlog_errno(status);
636 goto leave;
637 }
638 goto restarted_transaction;
639 }
640 }
641
642 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
643 le32_to_cpu(fe->i_clusters),
644 (unsigned long long)le64_to_cpu(fe->i_size));
645 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
646 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
647
648 leave:
649 if (handle) {
650 ocfs2_commit_trans(osb, handle);
651 handle = NULL;
652 }
653 if (data_ac) {
654 ocfs2_free_alloc_context(data_ac);
655 data_ac = NULL;
656 }
657 if (meta_ac) {
658 ocfs2_free_alloc_context(meta_ac);
659 meta_ac = NULL;
660 }
661 if ((!status) && restart_func) {
662 restart_func = 0;
663 goto restart_all;
664 }
665 brelse(bh);
666 bh = NULL;
667
668 mlog_exit(status);
669 return status;
670 }
671
672 /* Some parts of this taken from generic_cont_expand, which turned out
673 * to be too fragile to do exactly what we need without us having to
674 * worry about recursive locking in ->write_begin() and ->write_end(). */
675 static int ocfs2_write_zero_page(struct inode *inode,
676 u64 size)
677 {
678 struct address_space *mapping = inode->i_mapping;
679 struct page *page;
680 unsigned long index;
681 unsigned int offset;
682 handle_t *handle = NULL;
683 int ret;
684
685 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
686 /* ugh. in prepare/commit_write, if from==to==start of block, we
687 ** skip the prepare. make sure we never send an offset for the start
688 ** of a block
689 */
690 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
691 offset++;
692 }
693 index = size >> PAGE_CACHE_SHIFT;
694
695 page = grab_cache_page(mapping, index);
696 if (!page) {
697 ret = -ENOMEM;
698 mlog_errno(ret);
699 goto out;
700 }
701
702 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
703 if (ret < 0) {
704 mlog_errno(ret);
705 goto out_unlock;
706 }
707
708 if (ocfs2_should_order_data(inode)) {
709 handle = ocfs2_start_walk_page_trans(inode, page, offset,
710 offset);
711 if (IS_ERR(handle)) {
712 ret = PTR_ERR(handle);
713 handle = NULL;
714 goto out_unlock;
715 }
716 }
717
718 /* must not update i_size! */
719 ret = block_commit_write(page, offset, offset);
720 if (ret < 0)
721 mlog_errno(ret);
722 else
723 ret = 0;
724
725 if (handle)
726 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
727 out_unlock:
728 unlock_page(page);
729 page_cache_release(page);
730 out:
731 return ret;
732 }
733
734 static int ocfs2_zero_extend(struct inode *inode,
735 u64 zero_to_size)
736 {
737 int ret = 0;
738 u64 start_off;
739 struct super_block *sb = inode->i_sb;
740
741 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
742 while (start_off < zero_to_size) {
743 ret = ocfs2_write_zero_page(inode, start_off);
744 if (ret < 0) {
745 mlog_errno(ret);
746 goto out;
747 }
748
749 start_off += sb->s_blocksize;
750
751 /*
752 * Very large extends have the potential to lock up
753 * the cpu for extended periods of time.
754 */
755 cond_resched();
756 }
757
758 out:
759 return ret;
760 }
761
762 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
763 {
764 int ret;
765 u32 clusters_to_add;
766 struct ocfs2_inode_info *oi = OCFS2_I(inode);
767
768 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
769 if (clusters_to_add < oi->ip_clusters)
770 clusters_to_add = 0;
771 else
772 clusters_to_add -= oi->ip_clusters;
773
774 if (clusters_to_add) {
775 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
776 clusters_to_add, 0);
777 if (ret) {
778 mlog_errno(ret);
779 goto out;
780 }
781 }
782
783 /*
784 * Call this even if we don't add any clusters to the tree. We
785 * still need to zero the area between the old i_size and the
786 * new i_size.
787 */
788 ret = ocfs2_zero_extend(inode, zero_to);
789 if (ret < 0)
790 mlog_errno(ret);
791
792 out:
793 return ret;
794 }
795
796 static int ocfs2_extend_file(struct inode *inode,
797 struct buffer_head *di_bh,
798 u64 new_i_size)
799 {
800 int ret = 0;
801 struct ocfs2_inode_info *oi = OCFS2_I(inode);
802
803 BUG_ON(!di_bh);
804
805 /* setattr sometimes calls us like this. */
806 if (new_i_size == 0)
807 goto out;
808
809 if (i_size_read(inode) == new_i_size)
810 goto out;
811 BUG_ON(new_i_size < i_size_read(inode));
812
813 /*
814 * Fall through for converting inline data, even if the fs
815 * supports sparse files.
816 *
817 * The check for inline data here is legal - nobody can add
818 * the feature since we have i_mutex. We must check it again
819 * after acquiring ip_alloc_sem though, as paths like mmap
820 * might have raced us to converting the inode to extents.
821 */
822 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
823 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
824 goto out_update_size;
825
826 /*
827 * The alloc sem blocks people in read/write from reading our
828 * allocation until we're done changing it. We depend on
829 * i_mutex to block other extend/truncate calls while we're
830 * here.
831 */
832 down_write(&oi->ip_alloc_sem);
833
834 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
835 /*
836 * We can optimize small extends by keeping the inodes
837 * inline data.
838 */
839 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
840 up_write(&oi->ip_alloc_sem);
841 goto out_update_size;
842 }
843
844 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
845 if (ret) {
846 up_write(&oi->ip_alloc_sem);
847
848 mlog_errno(ret);
849 goto out;
850 }
851 }
852
853 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
854 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
855
856 up_write(&oi->ip_alloc_sem);
857
858 if (ret < 0) {
859 mlog_errno(ret);
860 goto out;
861 }
862
863 out_update_size:
864 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
865 if (ret < 0)
866 mlog_errno(ret);
867
868 out:
869 return ret;
870 }
871
872 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
873 {
874 int status = 0, size_change;
875 struct inode *inode = dentry->d_inode;
876 struct super_block *sb = inode->i_sb;
877 struct ocfs2_super *osb = OCFS2_SB(sb);
878 struct buffer_head *bh = NULL;
879 handle_t *handle = NULL;
880
881 mlog_entry("(0x%p, '%.*s')\n", dentry,
882 dentry->d_name.len, dentry->d_name.name);
883
884 /* ensuring we don't even attempt to truncate a symlink */
885 if (S_ISLNK(inode->i_mode))
886 attr->ia_valid &= ~ATTR_SIZE;
887
888 if (attr->ia_valid & ATTR_MODE)
889 mlog(0, "mode change: %d\n", attr->ia_mode);
890 if (attr->ia_valid & ATTR_UID)
891 mlog(0, "uid change: %d\n", attr->ia_uid);
892 if (attr->ia_valid & ATTR_GID)
893 mlog(0, "gid change: %d\n", attr->ia_gid);
894 if (attr->ia_valid & ATTR_SIZE)
895 mlog(0, "size change...\n");
896 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
897 mlog(0, "time change...\n");
898
899 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
900 | ATTR_GID | ATTR_UID | ATTR_MODE)
901 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
902 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
903 return 0;
904 }
905
906 status = inode_change_ok(inode, attr);
907 if (status)
908 return status;
909
910 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
911 if (size_change) {
912 status = ocfs2_rw_lock(inode, 1);
913 if (status < 0) {
914 mlog_errno(status);
915 goto bail;
916 }
917 }
918
919 status = ocfs2_inode_lock(inode, &bh, 1);
920 if (status < 0) {
921 if (status != -ENOENT)
922 mlog_errno(status);
923 goto bail_unlock_rw;
924 }
925
926 if (size_change && attr->ia_size != i_size_read(inode)) {
927 if (attr->ia_size > sb->s_maxbytes) {
928 status = -EFBIG;
929 goto bail_unlock;
930 }
931
932 if (i_size_read(inode) > attr->ia_size) {
933 if (ocfs2_should_order_data(inode)) {
934 status = ocfs2_begin_ordered_truncate(inode,
935 attr->ia_size);
936 if (status)
937 goto bail_unlock;
938 }
939 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
940 } else
941 status = ocfs2_extend_file(inode, bh, attr->ia_size);
942 if (status < 0) {
943 if (status != -ENOSPC)
944 mlog_errno(status);
945 status = -ENOSPC;
946 goto bail_unlock;
947 }
948 }
949
950 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
951 if (IS_ERR(handle)) {
952 status = PTR_ERR(handle);
953 mlog_errno(status);
954 goto bail_unlock;
955 }
956
957 /*
958 * This will intentionally not wind up calling vmtruncate(),
959 * since all the work for a size change has been done above.
960 * Otherwise, we could get into problems with truncate as
961 * ip_alloc_sem is used there to protect against i_size
962 * changes.
963 */
964 status = inode_setattr(inode, attr);
965 if (status < 0) {
966 mlog_errno(status);
967 goto bail_commit;
968 }
969
970 status = ocfs2_mark_inode_dirty(handle, inode, bh);
971 if (status < 0)
972 mlog_errno(status);
973
974 bail_commit:
975 ocfs2_commit_trans(osb, handle);
976 bail_unlock:
977 ocfs2_inode_unlock(inode, 1);
978 bail_unlock_rw:
979 if (size_change)
980 ocfs2_rw_unlock(inode, 1);
981 bail:
982 brelse(bh);
983
984 if (!status && attr->ia_valid & ATTR_MODE) {
985 status = ocfs2_acl_chmod(inode);
986 if (status < 0)
987 mlog_errno(status);
988 }
989
990 mlog_exit(status);
991 return status;
992 }
993
994 int ocfs2_getattr(struct vfsmount *mnt,
995 struct dentry *dentry,
996 struct kstat *stat)
997 {
998 struct inode *inode = dentry->d_inode;
999 struct super_block *sb = dentry->d_inode->i_sb;
1000 struct ocfs2_super *osb = sb->s_fs_info;
1001 int err;
1002
1003 mlog_entry_void();
1004
1005 err = ocfs2_inode_revalidate(dentry);
1006 if (err) {
1007 if (err != -ENOENT)
1008 mlog_errno(err);
1009 goto bail;
1010 }
1011
1012 generic_fillattr(inode, stat);
1013
1014 /* We set the blksize from the cluster size for performance */
1015 stat->blksize = osb->s_clustersize;
1016
1017 bail:
1018 mlog_exit(err);
1019
1020 return err;
1021 }
1022
1023 int ocfs2_permission(struct inode *inode, int mask)
1024 {
1025 int ret;
1026
1027 mlog_entry_void();
1028
1029 ret = ocfs2_inode_lock(inode, NULL, 0);
1030 if (ret) {
1031 if (ret != -ENOENT)
1032 mlog_errno(ret);
1033 goto out;
1034 }
1035
1036 ret = generic_permission(inode, mask, ocfs2_check_acl);
1037
1038 ocfs2_inode_unlock(inode, 0);
1039 out:
1040 mlog_exit(ret);
1041 return ret;
1042 }
1043
1044 static int __ocfs2_write_remove_suid(struct inode *inode,
1045 struct buffer_head *bh)
1046 {
1047 int ret;
1048 handle_t *handle;
1049 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1050 struct ocfs2_dinode *di;
1051
1052 mlog_entry("(Inode %llu, mode 0%o)\n",
1053 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1054
1055 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1056 if (IS_ERR(handle)) {
1057 ret = PTR_ERR(handle);
1058 mlog_errno(ret);
1059 goto out;
1060 }
1061
1062 ret = ocfs2_journal_access(handle, inode, bh,
1063 OCFS2_JOURNAL_ACCESS_WRITE);
1064 if (ret < 0) {
1065 mlog_errno(ret);
1066 goto out_trans;
1067 }
1068
1069 inode->i_mode &= ~S_ISUID;
1070 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1071 inode->i_mode &= ~S_ISGID;
1072
1073 di = (struct ocfs2_dinode *) bh->b_data;
1074 di->i_mode = cpu_to_le16(inode->i_mode);
1075
1076 ret = ocfs2_journal_dirty(handle, bh);
1077 if (ret < 0)
1078 mlog_errno(ret);
1079
1080 out_trans:
1081 ocfs2_commit_trans(osb, handle);
1082 out:
1083 mlog_exit(ret);
1084 return ret;
1085 }
1086
1087 /*
1088 * Will look for holes and unwritten extents in the range starting at
1089 * pos for count bytes (inclusive).
1090 */
1091 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1092 size_t count)
1093 {
1094 int ret = 0;
1095 unsigned int extent_flags;
1096 u32 cpos, clusters, extent_len, phys_cpos;
1097 struct super_block *sb = inode->i_sb;
1098
1099 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1100 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1101
1102 while (clusters) {
1103 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1104 &extent_flags);
1105 if (ret < 0) {
1106 mlog_errno(ret);
1107 goto out;
1108 }
1109
1110 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1111 ret = 1;
1112 break;
1113 }
1114
1115 if (extent_len > clusters)
1116 extent_len = clusters;
1117
1118 clusters -= extent_len;
1119 cpos += extent_len;
1120 }
1121 out:
1122 return ret;
1123 }
1124
1125 static int ocfs2_write_remove_suid(struct inode *inode)
1126 {
1127 int ret;
1128 struct buffer_head *bh = NULL;
1129
1130 ret = ocfs2_read_inode_block(inode, &bh);
1131 if (ret < 0) {
1132 mlog_errno(ret);
1133 goto out;
1134 }
1135
1136 ret = __ocfs2_write_remove_suid(inode, bh);
1137 out:
1138 brelse(bh);
1139 return ret;
1140 }
1141
1142 /*
1143 * Allocate enough extents to cover the region starting at byte offset
1144 * start for len bytes. Existing extents are skipped, any extents
1145 * added are marked as "unwritten".
1146 */
1147 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1148 u64 start, u64 len)
1149 {
1150 int ret;
1151 u32 cpos, phys_cpos, clusters, alloc_size;
1152 u64 end = start + len;
1153 struct buffer_head *di_bh = NULL;
1154
1155 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1156 ret = ocfs2_read_inode_block(inode, &di_bh);
1157 if (ret) {
1158 mlog_errno(ret);
1159 goto out;
1160 }
1161
1162 /*
1163 * Nothing to do if the requested reservation range
1164 * fits within the inode.
1165 */
1166 if (ocfs2_size_fits_inline_data(di_bh, end))
1167 goto out;
1168
1169 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1170 if (ret) {
1171 mlog_errno(ret);
1172 goto out;
1173 }
1174 }
1175
1176 /*
1177 * We consider both start and len to be inclusive.
1178 */
1179 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1180 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1181 clusters -= cpos;
1182
1183 while (clusters) {
1184 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1185 &alloc_size, NULL);
1186 if (ret) {
1187 mlog_errno(ret);
1188 goto out;
1189 }
1190
1191 /*
1192 * Hole or existing extent len can be arbitrary, so
1193 * cap it to our own allocation request.
1194 */
1195 if (alloc_size > clusters)
1196 alloc_size = clusters;
1197
1198 if (phys_cpos) {
1199 /*
1200 * We already have an allocation at this
1201 * region so we can safely skip it.
1202 */
1203 goto next;
1204 }
1205
1206 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1207 if (ret) {
1208 if (ret != -ENOSPC)
1209 mlog_errno(ret);
1210 goto out;
1211 }
1212
1213 next:
1214 cpos += alloc_size;
1215 clusters -= alloc_size;
1216 }
1217
1218 ret = 0;
1219 out:
1220
1221 brelse(di_bh);
1222 return ret;
1223 }
1224
1225 /*
1226 * Truncate a byte range, avoiding pages within partial clusters. This
1227 * preserves those pages for the zeroing code to write to.
1228 */
1229 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1230 u64 byte_len)
1231 {
1232 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1233 loff_t start, end;
1234 struct address_space *mapping = inode->i_mapping;
1235
1236 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1237 end = byte_start + byte_len;
1238 end = end & ~(osb->s_clustersize - 1);
1239
1240 if (start < end) {
1241 unmap_mapping_range(mapping, start, end - start, 0);
1242 truncate_inode_pages_range(mapping, start, end - 1);
1243 }
1244 }
1245
1246 static int ocfs2_zero_partial_clusters(struct inode *inode,
1247 u64 start, u64 len)
1248 {
1249 int ret = 0;
1250 u64 tmpend, end = start + len;
1251 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1252 unsigned int csize = osb->s_clustersize;
1253 handle_t *handle;
1254
1255 /*
1256 * The "start" and "end" values are NOT necessarily part of
1257 * the range whose allocation is being deleted. Rather, this
1258 * is what the user passed in with the request. We must zero
1259 * partial clusters here. There's no need to worry about
1260 * physical allocation - the zeroing code knows to skip holes.
1261 */
1262 mlog(0, "byte start: %llu, end: %llu\n",
1263 (unsigned long long)start, (unsigned long long)end);
1264
1265 /*
1266 * If both edges are on a cluster boundary then there's no
1267 * zeroing required as the region is part of the allocation to
1268 * be truncated.
1269 */
1270 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1271 goto out;
1272
1273 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1274 if (IS_ERR(handle)) {
1275 ret = PTR_ERR(handle);
1276 mlog_errno(ret);
1277 goto out;
1278 }
1279
1280 /*
1281 * We want to get the byte offset of the end of the 1st cluster.
1282 */
1283 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1284 if (tmpend > end)
1285 tmpend = end;
1286
1287 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1288 (unsigned long long)start, (unsigned long long)tmpend);
1289
1290 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1291 if (ret)
1292 mlog_errno(ret);
1293
1294 if (tmpend < end) {
1295 /*
1296 * This may make start and end equal, but the zeroing
1297 * code will skip any work in that case so there's no
1298 * need to catch it up here.
1299 */
1300 start = end & ~(osb->s_clustersize - 1);
1301
1302 mlog(0, "2nd range: start: %llu, end: %llu\n",
1303 (unsigned long long)start, (unsigned long long)end);
1304
1305 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1306 if (ret)
1307 mlog_errno(ret);
1308 }
1309
1310 ocfs2_commit_trans(osb, handle);
1311 out:
1312 return ret;
1313 }
1314
1315 static int ocfs2_remove_inode_range(struct inode *inode,
1316 struct buffer_head *di_bh, u64 byte_start,
1317 u64 byte_len)
1318 {
1319 int ret = 0;
1320 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1321 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1322 struct ocfs2_cached_dealloc_ctxt dealloc;
1323 struct address_space *mapping = inode->i_mapping;
1324 struct ocfs2_extent_tree et;
1325
1326 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
1327 ocfs2_init_dealloc_ctxt(&dealloc);
1328
1329 if (byte_len == 0)
1330 return 0;
1331
1332 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1333 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1334 byte_start + byte_len, 0);
1335 if (ret) {
1336 mlog_errno(ret);
1337 goto out;
1338 }
1339 /*
1340 * There's no need to get fancy with the page cache
1341 * truncate of an inline-data inode. We're talking
1342 * about less than a page here, which will be cached
1343 * in the dinode buffer anyway.
1344 */
1345 unmap_mapping_range(mapping, 0, 0, 0);
1346 truncate_inode_pages(mapping, 0);
1347 goto out;
1348 }
1349
1350 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1351 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1352 if (trunc_len >= trunc_start)
1353 trunc_len -= trunc_start;
1354 else
1355 trunc_len = 0;
1356
1357 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1358 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1359 (unsigned long long)byte_start,
1360 (unsigned long long)byte_len, trunc_start, trunc_len);
1361
1362 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1363 if (ret) {
1364 mlog_errno(ret);
1365 goto out;
1366 }
1367
1368 cpos = trunc_start;
1369 while (trunc_len) {
1370 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1371 &alloc_size, NULL);
1372 if (ret) {
1373 mlog_errno(ret);
1374 goto out;
1375 }
1376
1377 if (alloc_size > trunc_len)
1378 alloc_size = trunc_len;
1379
1380 /* Only do work for non-holes */
1381 if (phys_cpos != 0) {
1382 ret = ocfs2_remove_btree_range(inode, &et, cpos,
1383 phys_cpos, alloc_size,
1384 &dealloc);
1385 if (ret) {
1386 mlog_errno(ret);
1387 goto out;
1388 }
1389 }
1390
1391 cpos += alloc_size;
1392 trunc_len -= alloc_size;
1393 }
1394
1395 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1396
1397 out:
1398 ocfs2_schedule_truncate_log_flush(osb, 1);
1399 ocfs2_run_deallocs(osb, &dealloc);
1400
1401 return ret;
1402 }
1403
1404 /*
1405 * Parts of this function taken from xfs_change_file_space()
1406 */
1407 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1408 loff_t f_pos, unsigned int cmd,
1409 struct ocfs2_space_resv *sr,
1410 int change_size)
1411 {
1412 int ret;
1413 s64 llen;
1414 loff_t size;
1415 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1416 struct buffer_head *di_bh = NULL;
1417 handle_t *handle;
1418 unsigned long long max_off = inode->i_sb->s_maxbytes;
1419
1420 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1421 return -EROFS;
1422
1423 mutex_lock(&inode->i_mutex);
1424
1425 /*
1426 * This prevents concurrent writes on other nodes
1427 */
1428 ret = ocfs2_rw_lock(inode, 1);
1429 if (ret) {
1430 mlog_errno(ret);
1431 goto out;
1432 }
1433
1434 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1435 if (ret) {
1436 mlog_errno(ret);
1437 goto out_rw_unlock;
1438 }
1439
1440 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1441 ret = -EPERM;
1442 goto out_inode_unlock;
1443 }
1444
1445 switch (sr->l_whence) {
1446 case 0: /*SEEK_SET*/
1447 break;
1448 case 1: /*SEEK_CUR*/
1449 sr->l_start += f_pos;
1450 break;
1451 case 2: /*SEEK_END*/
1452 sr->l_start += i_size_read(inode);
1453 break;
1454 default:
1455 ret = -EINVAL;
1456 goto out_inode_unlock;
1457 }
1458 sr->l_whence = 0;
1459
1460 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1461
1462 if (sr->l_start < 0
1463 || sr->l_start > max_off
1464 || (sr->l_start + llen) < 0
1465 || (sr->l_start + llen) > max_off) {
1466 ret = -EINVAL;
1467 goto out_inode_unlock;
1468 }
1469 size = sr->l_start + sr->l_len;
1470
1471 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1472 if (sr->l_len <= 0) {
1473 ret = -EINVAL;
1474 goto out_inode_unlock;
1475 }
1476 }
1477
1478 if (file && should_remove_suid(file->f_path.dentry)) {
1479 ret = __ocfs2_write_remove_suid(inode, di_bh);
1480 if (ret) {
1481 mlog_errno(ret);
1482 goto out_inode_unlock;
1483 }
1484 }
1485
1486 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1487 switch (cmd) {
1488 case OCFS2_IOC_RESVSP:
1489 case OCFS2_IOC_RESVSP64:
1490 /*
1491 * This takes unsigned offsets, but the signed ones we
1492 * pass have been checked against overflow above.
1493 */
1494 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1495 sr->l_len);
1496 break;
1497 case OCFS2_IOC_UNRESVSP:
1498 case OCFS2_IOC_UNRESVSP64:
1499 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1500 sr->l_len);
1501 break;
1502 default:
1503 ret = -EINVAL;
1504 }
1505 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1506 if (ret) {
1507 mlog_errno(ret);
1508 goto out_inode_unlock;
1509 }
1510
1511 /*
1512 * We update c/mtime for these changes
1513 */
1514 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1515 if (IS_ERR(handle)) {
1516 ret = PTR_ERR(handle);
1517 mlog_errno(ret);
1518 goto out_inode_unlock;
1519 }
1520
1521 if (change_size && i_size_read(inode) < size)
1522 i_size_write(inode, size);
1523
1524 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1525 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1526 if (ret < 0)
1527 mlog_errno(ret);
1528
1529 ocfs2_commit_trans(osb, handle);
1530
1531 out_inode_unlock:
1532 brelse(di_bh);
1533 ocfs2_inode_unlock(inode, 1);
1534 out_rw_unlock:
1535 ocfs2_rw_unlock(inode, 1);
1536
1537 out:
1538 mutex_unlock(&inode->i_mutex);
1539 return ret;
1540 }
1541
1542 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1543 struct ocfs2_space_resv *sr)
1544 {
1545 struct inode *inode = file->f_path.dentry->d_inode;
1546 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1547
1548 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1549 !ocfs2_writes_unwritten_extents(osb))
1550 return -ENOTTY;
1551 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1552 !ocfs2_sparse_alloc(osb))
1553 return -ENOTTY;
1554
1555 if (!S_ISREG(inode->i_mode))
1556 return -EINVAL;
1557
1558 if (!(file->f_mode & FMODE_WRITE))
1559 return -EBADF;
1560
1561 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1562 }
1563
1564 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1565 loff_t len)
1566 {
1567 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1568 struct ocfs2_space_resv sr;
1569 int change_size = 1;
1570
1571 if (!ocfs2_writes_unwritten_extents(osb))
1572 return -EOPNOTSUPP;
1573
1574 if (S_ISDIR(inode->i_mode))
1575 return -ENODEV;
1576
1577 if (mode & FALLOC_FL_KEEP_SIZE)
1578 change_size = 0;
1579
1580 sr.l_whence = 0;
1581 sr.l_start = (s64)offset;
1582 sr.l_len = (s64)len;
1583
1584 return __ocfs2_change_file_space(NULL, inode, offset,
1585 OCFS2_IOC_RESVSP64, &sr, change_size);
1586 }
1587
1588 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1589 loff_t *ppos,
1590 size_t count,
1591 int appending,
1592 int *direct_io)
1593 {
1594 int ret = 0, meta_level = 0;
1595 struct inode *inode = dentry->d_inode;
1596 loff_t saved_pos, end;
1597
1598 /*
1599 * We start with a read level meta lock and only jump to an ex
1600 * if we need to make modifications here.
1601 */
1602 for(;;) {
1603 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1604 if (ret < 0) {
1605 meta_level = -1;
1606 mlog_errno(ret);
1607 goto out;
1608 }
1609
1610 /* Clear suid / sgid if necessary. We do this here
1611 * instead of later in the write path because
1612 * remove_suid() calls ->setattr without any hint that
1613 * we may have already done our cluster locking. Since
1614 * ocfs2_setattr() *must* take cluster locks to
1615 * proceeed, this will lead us to recursively lock the
1616 * inode. There's also the dinode i_size state which
1617 * can be lost via setattr during extending writes (we
1618 * set inode->i_size at the end of a write. */
1619 if (should_remove_suid(dentry)) {
1620 if (meta_level == 0) {
1621 ocfs2_inode_unlock(inode, meta_level);
1622 meta_level = 1;
1623 continue;
1624 }
1625
1626 ret = ocfs2_write_remove_suid(inode);
1627 if (ret < 0) {
1628 mlog_errno(ret);
1629 goto out_unlock;
1630 }
1631 }
1632
1633 /* work on a copy of ppos until we're sure that we won't have
1634 * to recalculate it due to relocking. */
1635 if (appending) {
1636 saved_pos = i_size_read(inode);
1637 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1638 } else {
1639 saved_pos = *ppos;
1640 }
1641
1642 end = saved_pos + count;
1643
1644 /*
1645 * Skip the O_DIRECT checks if we don't need
1646 * them.
1647 */
1648 if (!direct_io || !(*direct_io))
1649 break;
1650
1651 /*
1652 * There's no sane way to do direct writes to an inode
1653 * with inline data.
1654 */
1655 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1656 *direct_io = 0;
1657 break;
1658 }
1659
1660 /*
1661 * Allowing concurrent direct writes means
1662 * i_size changes wouldn't be synchronized, so
1663 * one node could wind up truncating another
1664 * nodes writes.
1665 */
1666 if (end > i_size_read(inode)) {
1667 *direct_io = 0;
1668 break;
1669 }
1670
1671 /*
1672 * We don't fill holes during direct io, so
1673 * check for them here. If any are found, the
1674 * caller will have to retake some cluster
1675 * locks and initiate the io as buffered.
1676 */
1677 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1678 if (ret == 1) {
1679 *direct_io = 0;
1680 ret = 0;
1681 } else if (ret < 0)
1682 mlog_errno(ret);
1683 break;
1684 }
1685
1686 if (appending)
1687 *ppos = saved_pos;
1688
1689 out_unlock:
1690 ocfs2_inode_unlock(inode, meta_level);
1691
1692 out:
1693 return ret;
1694 }
1695
1696 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1697 const struct iovec *iov,
1698 unsigned long nr_segs,
1699 loff_t pos)
1700 {
1701 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1702 int can_do_direct;
1703 ssize_t written = 0;
1704 size_t ocount; /* original count */
1705 size_t count; /* after file limit checks */
1706 loff_t old_size, *ppos = &iocb->ki_pos;
1707 u32 old_clusters;
1708 struct file *file = iocb->ki_filp;
1709 struct inode *inode = file->f_path.dentry->d_inode;
1710 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1711
1712 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1713 (unsigned int)nr_segs,
1714 file->f_path.dentry->d_name.len,
1715 file->f_path.dentry->d_name.name);
1716
1717 if (iocb->ki_left == 0)
1718 return 0;
1719
1720 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1721
1722 appending = file->f_flags & O_APPEND ? 1 : 0;
1723 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1724
1725 mutex_lock(&inode->i_mutex);
1726
1727 relock:
1728 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1729 if (direct_io) {
1730 down_read(&inode->i_alloc_sem);
1731 have_alloc_sem = 1;
1732 }
1733
1734 /* concurrent O_DIRECT writes are allowed */
1735 rw_level = !direct_io;
1736 ret = ocfs2_rw_lock(inode, rw_level);
1737 if (ret < 0) {
1738 mlog_errno(ret);
1739 goto out_sems;
1740 }
1741
1742 can_do_direct = direct_io;
1743 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1744 iocb->ki_left, appending,
1745 &can_do_direct);
1746 if (ret < 0) {
1747 mlog_errno(ret);
1748 goto out;
1749 }
1750
1751 /*
1752 * We can't complete the direct I/O as requested, fall back to
1753 * buffered I/O.
1754 */
1755 if (direct_io && !can_do_direct) {
1756 ocfs2_rw_unlock(inode, rw_level);
1757 up_read(&inode->i_alloc_sem);
1758
1759 have_alloc_sem = 0;
1760 rw_level = -1;
1761
1762 direct_io = 0;
1763 goto relock;
1764 }
1765
1766 /*
1767 * To later detect whether a journal commit for sync writes is
1768 * necessary, we sample i_size, and cluster count here.
1769 */
1770 old_size = i_size_read(inode);
1771 old_clusters = OCFS2_I(inode)->ip_clusters;
1772
1773 /* communicate with ocfs2_dio_end_io */
1774 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1775
1776 if (direct_io) {
1777 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1778 VERIFY_READ);
1779 if (ret)
1780 goto out_dio;
1781
1782 ret = generic_write_checks(file, ppos, &count,
1783 S_ISBLK(inode->i_mode));
1784 if (ret)
1785 goto out_dio;
1786
1787 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1788 ppos, count, ocount);
1789 if (written < 0) {
1790 /*
1791 * direct write may have instantiated a few
1792 * blocks outside i_size. Trim these off again.
1793 * Don't need i_size_read because we hold i_mutex.
1794 */
1795 if (*ppos + count > inode->i_size)
1796 vmtruncate(inode, inode->i_size);
1797 ret = written;
1798 goto out_dio;
1799 }
1800 } else {
1801 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1802 *ppos);
1803 }
1804
1805 out_dio:
1806 /* buffered aio wouldn't have proper lock coverage today */
1807 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1808
1809 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1810 /*
1811 * The generic write paths have handled getting data
1812 * to disk, but since we don't make use of the dirty
1813 * inode list, a manual journal commit is necessary
1814 * here.
1815 */
1816 if (old_size != i_size_read(inode) ||
1817 old_clusters != OCFS2_I(inode)->ip_clusters) {
1818 ret = jbd2_journal_force_commit(osb->journal->j_journal);
1819 if (ret < 0)
1820 written = ret;
1821 }
1822 }
1823
1824 /*
1825 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1826 * function pointer which is called when o_direct io completes so that
1827 * it can unlock our rw lock. (it's the clustered equivalent of
1828 * i_alloc_sem; protects truncate from racing with pending ios).
1829 * Unfortunately there are error cases which call end_io and others
1830 * that don't. so we don't have to unlock the rw_lock if either an
1831 * async dio is going to do it in the future or an end_io after an
1832 * error has already done it.
1833 */
1834 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1835 rw_level = -1;
1836 have_alloc_sem = 0;
1837 }
1838
1839 out:
1840 if (rw_level != -1)
1841 ocfs2_rw_unlock(inode, rw_level);
1842
1843 out_sems:
1844 if (have_alloc_sem)
1845 up_read(&inode->i_alloc_sem);
1846
1847 mutex_unlock(&inode->i_mutex);
1848
1849 mlog_exit(ret);
1850 return written ? written : ret;
1851 }
1852
1853 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1854 struct file *out,
1855 loff_t *ppos,
1856 size_t len,
1857 unsigned int flags)
1858 {
1859 int ret;
1860 struct inode *inode = out->f_path.dentry->d_inode;
1861
1862 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1863 (unsigned int)len,
1864 out->f_path.dentry->d_name.len,
1865 out->f_path.dentry->d_name.name);
1866
1867 inode_double_lock(inode, pipe->inode);
1868
1869 ret = ocfs2_rw_lock(inode, 1);
1870 if (ret < 0) {
1871 mlog_errno(ret);
1872 goto out;
1873 }
1874
1875 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
1876 NULL);
1877 if (ret < 0) {
1878 mlog_errno(ret);
1879 goto out_unlock;
1880 }
1881
1882 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
1883
1884 out_unlock:
1885 ocfs2_rw_unlock(inode, 1);
1886 out:
1887 inode_double_unlock(inode, pipe->inode);
1888
1889 mlog_exit(ret);
1890 return ret;
1891 }
1892
1893 static ssize_t ocfs2_file_splice_read(struct file *in,
1894 loff_t *ppos,
1895 struct pipe_inode_info *pipe,
1896 size_t len,
1897 unsigned int flags)
1898 {
1899 int ret = 0;
1900 struct inode *inode = in->f_path.dentry->d_inode;
1901
1902 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1903 (unsigned int)len,
1904 in->f_path.dentry->d_name.len,
1905 in->f_path.dentry->d_name.name);
1906
1907 /*
1908 * See the comment in ocfs2_file_aio_read()
1909 */
1910 ret = ocfs2_inode_lock(inode, NULL, 0);
1911 if (ret < 0) {
1912 mlog_errno(ret);
1913 goto bail;
1914 }
1915 ocfs2_inode_unlock(inode, 0);
1916
1917 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1918
1919 bail:
1920 mlog_exit(ret);
1921 return ret;
1922 }
1923
1924 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
1925 const struct iovec *iov,
1926 unsigned long nr_segs,
1927 loff_t pos)
1928 {
1929 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
1930 struct file *filp = iocb->ki_filp;
1931 struct inode *inode = filp->f_path.dentry->d_inode;
1932
1933 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1934 (unsigned int)nr_segs,
1935 filp->f_path.dentry->d_name.len,
1936 filp->f_path.dentry->d_name.name);
1937
1938 if (!inode) {
1939 ret = -EINVAL;
1940 mlog_errno(ret);
1941 goto bail;
1942 }
1943
1944 /*
1945 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1946 * need locks to protect pending reads from racing with truncate.
1947 */
1948 if (filp->f_flags & O_DIRECT) {
1949 down_read(&inode->i_alloc_sem);
1950 have_alloc_sem = 1;
1951
1952 ret = ocfs2_rw_lock(inode, 0);
1953 if (ret < 0) {
1954 mlog_errno(ret);
1955 goto bail;
1956 }
1957 rw_level = 0;
1958 /* communicate with ocfs2_dio_end_io */
1959 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1960 }
1961
1962 /*
1963 * We're fine letting folks race truncates and extending
1964 * writes with read across the cluster, just like they can
1965 * locally. Hence no rw_lock during read.
1966 *
1967 * Take and drop the meta data lock to update inode fields
1968 * like i_size. This allows the checks down below
1969 * generic_file_aio_read() a chance of actually working.
1970 */
1971 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
1972 if (ret < 0) {
1973 mlog_errno(ret);
1974 goto bail;
1975 }
1976 ocfs2_inode_unlock(inode, lock_level);
1977
1978 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
1979 if (ret == -EINVAL)
1980 mlog(0, "generic_file_aio_read returned -EINVAL\n");
1981
1982 /* buffered aio wouldn't have proper lock coverage today */
1983 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1984
1985 /* see ocfs2_file_aio_write */
1986 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1987 rw_level = -1;
1988 have_alloc_sem = 0;
1989 }
1990
1991 bail:
1992 if (have_alloc_sem)
1993 up_read(&inode->i_alloc_sem);
1994 if (rw_level != -1)
1995 ocfs2_rw_unlock(inode, rw_level);
1996 mlog_exit(ret);
1997
1998 return ret;
1999 }
2000
2001 const struct inode_operations ocfs2_file_iops = {
2002 .setattr = ocfs2_setattr,
2003 .getattr = ocfs2_getattr,
2004 .permission = ocfs2_permission,
2005 .setxattr = generic_setxattr,
2006 .getxattr = generic_getxattr,
2007 .listxattr = ocfs2_listxattr,
2008 .removexattr = generic_removexattr,
2009 .fallocate = ocfs2_fallocate,
2010 .fiemap = ocfs2_fiemap,
2011 };
2012
2013 const struct inode_operations ocfs2_special_file_iops = {
2014 .setattr = ocfs2_setattr,
2015 .getattr = ocfs2_getattr,
2016 .permission = ocfs2_permission,
2017 };
2018
2019 /*
2020 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2021 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2022 */
2023 const struct file_operations ocfs2_fops = {
2024 .llseek = generic_file_llseek,
2025 .read = do_sync_read,
2026 .write = do_sync_write,
2027 .mmap = ocfs2_mmap,
2028 .fsync = ocfs2_sync_file,
2029 .release = ocfs2_file_release,
2030 .open = ocfs2_file_open,
2031 .aio_read = ocfs2_file_aio_read,
2032 .aio_write = ocfs2_file_aio_write,
2033 .unlocked_ioctl = ocfs2_ioctl,
2034 #ifdef CONFIG_COMPAT
2035 .compat_ioctl = ocfs2_compat_ioctl,
2036 #endif
2037 .lock = ocfs2_lock,
2038 .flock = ocfs2_flock,
2039 .splice_read = ocfs2_file_splice_read,
2040 .splice_write = ocfs2_file_splice_write,
2041 };
2042
2043 const struct file_operations ocfs2_dops = {
2044 .llseek = generic_file_llseek,
2045 .read = generic_read_dir,
2046 .readdir = ocfs2_readdir,
2047 .fsync = ocfs2_sync_file,
2048 .release = ocfs2_dir_release,
2049 .open = ocfs2_dir_open,
2050 .unlocked_ioctl = ocfs2_ioctl,
2051 #ifdef CONFIG_COMPAT
2052 .compat_ioctl = ocfs2_compat_ioctl,
2053 #endif
2054 .lock = ocfs2_lock,
2055 .flock = ocfs2_flock,
2056 };
2057
2058 /*
2059 * POSIX-lockless variants of our file_operations.
2060 *
2061 * These will be used if the underlying cluster stack does not support
2062 * posix file locking, if the user passes the "localflocks" mount
2063 * option, or if we have a local-only fs.
2064 *
2065 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2066 * so we still want it in the case of no stack support for
2067 * plocks. Internally, it will do the right thing when asked to ignore
2068 * the cluster.
2069 */
2070 const struct file_operations ocfs2_fops_no_plocks = {
2071 .llseek = generic_file_llseek,
2072 .read = do_sync_read,
2073 .write = do_sync_write,
2074 .mmap = ocfs2_mmap,
2075 .fsync = ocfs2_sync_file,
2076 .release = ocfs2_file_release,
2077 .open = ocfs2_file_open,
2078 .aio_read = ocfs2_file_aio_read,
2079 .aio_write = ocfs2_file_aio_write,
2080 .unlocked_ioctl = ocfs2_ioctl,
2081 #ifdef CONFIG_COMPAT
2082 .compat_ioctl = ocfs2_compat_ioctl,
2083 #endif
2084 .flock = ocfs2_flock,
2085 .splice_read = ocfs2_file_splice_read,
2086 .splice_write = ocfs2_file_splice_write,
2087 };
2088
2089 const struct file_operations ocfs2_dops_no_plocks = {
2090 .llseek = generic_file_llseek,
2091 .read = generic_read_dir,
2092 .readdir = ocfs2_readdir,
2093 .fsync = ocfs2_sync_file,
2094 .release = ocfs2_dir_release,
2095 .open = ocfs2_dir_open,
2096 .unlocked_ioctl = ocfs2_ioctl,
2097 #ifdef CONFIG_COMPAT
2098 .compat_ioctl = ocfs2_compat_ioctl,
2099 #endif
2100 .flock = ocfs2_flock,
2101 };
Linux kernel - Deliverables
This page took 0.090475 seconds and 5 git commands to generate.