projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xfs: fix issues that cause userspace warnings
[deliverable/linux.git] / fs / xfs / xfs_vnodeops.c
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_format.h"
23 #include "xfs_bit.h"
24 #include "xfs_log.h"
25 #include "xfs_trans.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2_format.h"
31 #include "xfs_dir2.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_itable.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_acl.h"
43 #include "xfs_attr.h"
44 #include "xfs_error.h"
45 #include "xfs_quota.h"
46 #include "xfs_utils.h"
47 #include "xfs_rtalloc.h"
48 #include "xfs_trans_space.h"
49 #include "xfs_log_priv.h"
50 #include "xfs_filestream.h"
51 #include "xfs_vnodeops.h"
52 #include "xfs_trace.h"
53 #include "xfs_icache.h"
54 #include "xfs_symlink.h"
55
56
57 /*
58 * This is called by xfs_inactive to free any blocks beyond eof
59 * when the link count isn't zero and by xfs_dm_punch_hole() when
60 * punching a hole to EOF.
61 */
62 int
63 xfs_free_eofblocks(
64 xfs_mount_t *mp,
65 xfs_inode_t *ip,
66 bool need_iolock)
67 {
68 xfs_trans_t *tp;
69 int error;
70 xfs_fileoff_t end_fsb;
71 xfs_fileoff_t last_fsb;
72 xfs_filblks_t map_len;
73 int nimaps;
74 xfs_bmbt_irec_t imap;
75
76 /*
77 * Figure out if there are any blocks beyond the end
78 * of the file. If not, then there is nothing to do.
79 */
80 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
81 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
82 if (last_fsb <= end_fsb)
83 return 0;
84 map_len = last_fsb - end_fsb;
85
86 nimaps = 1;
87 xfs_ilock(ip, XFS_ILOCK_SHARED);
88 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
89 xfs_iunlock(ip, XFS_ILOCK_SHARED);
90
91 if (!error && (nimaps != 0) &&
92 (imap.br_startblock != HOLESTARTBLOCK ||
93 ip->i_delayed_blks)) {
94 /*
95 * Attach the dquots to the inode up front.
96 */
97 error = xfs_qm_dqattach(ip, 0);
98 if (error)
99 return error;
100
101 /*
102 * There are blocks after the end of file.
103 * Free them up now by truncating the file to
104 * its current size.
105 */
106 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
107
108 if (need_iolock) {
109 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
110 xfs_trans_cancel(tp, 0);
111 return EAGAIN;
112 }
113 }
114
115 error = xfs_trans_reserve(tp, 0,
116 XFS_ITRUNCATE_LOG_RES(mp),
117 0, XFS_TRANS_PERM_LOG_RES,
118 XFS_ITRUNCATE_LOG_COUNT);
119 if (error) {
120 ASSERT(XFS_FORCED_SHUTDOWN(mp));
121 xfs_trans_cancel(tp, 0);
122 if (need_iolock)
123 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
124 return error;
125 }
126
127 xfs_ilock(ip, XFS_ILOCK_EXCL);
128 xfs_trans_ijoin(tp, ip, 0);
129
130 /*
131 * Do not update the on-disk file size. If we update the
132 * on-disk file size and then the system crashes before the
133 * contents of the file are flushed to disk then the files
134 * may be full of holes (ie NULL files bug).
135 */
136 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
137 XFS_ISIZE(ip));
138 if (error) {
139 /*
140 * If we get an error at this point we simply don't
141 * bother truncating the file.
142 */
143 xfs_trans_cancel(tp,
144 (XFS_TRANS_RELEASE_LOG_RES |
145 XFS_TRANS_ABORT));
146 } else {
147 error = xfs_trans_commit(tp,
148 XFS_TRANS_RELEASE_LOG_RES);
149 if (!error)
150 xfs_inode_clear_eofblocks_tag(ip);
151 }
152
153 xfs_iunlock(ip, XFS_ILOCK_EXCL);
154 if (need_iolock)
155 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
156 }
157 return error;
158 }
159
160 int
161 xfs_release(
162 xfs_inode_t *ip)
163 {
164 xfs_mount_t *mp = ip->i_mount;
165 int error;
166
167 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
168 return 0;
169
170 /* If this is a read-only mount, don't do this (would generate I/O) */
171 if (mp->m_flags & XFS_MOUNT_RDONLY)
172 return 0;
173
174 if (!XFS_FORCED_SHUTDOWN(mp)) {
175 int truncated;
176
177 /*
178 * If we are using filestreams, and we have an unlinked
179 * file that we are processing the last close on, then nothing
180 * will be able to reopen and write to this file. Purge this
181 * inode from the filestreams cache so that it doesn't delay
182 * teardown of the inode.
183 */
184 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
185 xfs_filestream_deassociate(ip);
186
187 /*
188 * If we previously truncated this file and removed old data
189 * in the process, we want to initiate "early" writeout on
190 * the last close. This is an attempt to combat the notorious
191 * NULL files problem which is particularly noticeable from a
192 * truncate down, buffered (re-)write (delalloc), followed by
193 * a crash. What we are effectively doing here is
194 * significantly reducing the time window where we'd otherwise
195 * be exposed to that problem.
196 */
197 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
198 if (truncated) {
199 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
200 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
201 error = -filemap_flush(VFS_I(ip)->i_mapping);
202 if (error)
203 return error;
204 }
205 }
206 }
207
208 if (ip->i_d.di_nlink == 0)
209 return 0;
210
211 if (xfs_can_free_eofblocks(ip, false)) {
212
213 /*
214 * If we can't get the iolock just skip truncating the blocks
215 * past EOF because we could deadlock with the mmap_sem
216 * otherwise. We'll get another chance to drop them once the
217 * last reference to the inode is dropped, so we'll never leak
218 * blocks permanently.
219 *
220 * Further, check if the inode is being opened, written and
221 * closed frequently and we have delayed allocation blocks
222 * outstanding (e.g. streaming writes from the NFS server),
223 * truncating the blocks past EOF will cause fragmentation to
224 * occur.
225 *
226 * In this case don't do the truncation, either, but we have to
227 * be careful how we detect this case. Blocks beyond EOF show
228 * up as i_delayed_blks even when the inode is clean, so we
229 * need to truncate them away first before checking for a dirty
230 * release. Hence on the first dirty close we will still remove
231 * the speculative allocation, but after that we will leave it
232 * in place.
233 */
234 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
235 return 0;
236
237 error = xfs_free_eofblocks(mp, ip, true);
238 if (error && error != EAGAIN)
239 return error;
240
241 /* delalloc blocks after truncation means it really is dirty */
242 if (ip->i_delayed_blks)
243 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
244 }
245 return 0;
246 }
247
248 /*
249 * xfs_inactive
250 *
251 * This is called when the vnode reference count for the vnode
252 * goes to zero. If the file has been unlinked, then it must
253 * now be truncated. Also, we clear all of the read-ahead state
254 * kept for the inode here since the file is now closed.
255 */
256 int
257 xfs_inactive(
258 xfs_inode_t *ip)
259 {
260 xfs_bmap_free_t free_list;
261 xfs_fsblock_t first_block;
262 int committed;
263 xfs_trans_t *tp;
264 xfs_mount_t *mp;
265 int error;
266 int truncate = 0;
267
268 /*
269 * If the inode is already free, then there can be nothing
270 * to clean up here.
271 */
272 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
273 ASSERT(ip->i_df.if_real_bytes == 0);
274 ASSERT(ip->i_df.if_broot_bytes == 0);
275 return VN_INACTIVE_CACHE;
276 }
277
278 mp = ip->i_mount;
279
280 error = 0;
281
282 /* If this is a read-only mount, don't do this (would generate I/O) */
283 if (mp->m_flags & XFS_MOUNT_RDONLY)
284 goto out;
285
286 if (ip->i_d.di_nlink != 0) {
287 /*
288 * force is true because we are evicting an inode from the
289 * cache. Post-eof blocks must be freed, lest we end up with
290 * broken free space accounting.
291 */
292 if (xfs_can_free_eofblocks(ip, true)) {
293 error = xfs_free_eofblocks(mp, ip, false);
294 if (error)
295 return VN_INACTIVE_CACHE;
296 }
297 goto out;
298 }
299
300 if (S_ISREG(ip->i_d.di_mode) &&
301 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
302 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
303 truncate = 1;
304
305 error = xfs_qm_dqattach(ip, 0);
306 if (error)
307 return VN_INACTIVE_CACHE;
308
309 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
310 error = xfs_trans_reserve(tp, 0,
311 (truncate || S_ISLNK(ip->i_d.di_mode)) ?
312 XFS_ITRUNCATE_LOG_RES(mp) :
313 XFS_IFREE_LOG_RES(mp),
314 0,
315 XFS_TRANS_PERM_LOG_RES,
316 XFS_ITRUNCATE_LOG_COUNT);
317 if (error) {
318 ASSERT(XFS_FORCED_SHUTDOWN(mp));
319 xfs_trans_cancel(tp, 0);
320 return VN_INACTIVE_CACHE;
321 }
322
323 xfs_ilock(ip, XFS_ILOCK_EXCL);
324 xfs_trans_ijoin(tp, ip, 0);
325
326 if (S_ISLNK(ip->i_d.di_mode)) {
327 error = xfs_inactive_symlink(ip, &tp);
328 if (error)
329 goto out_cancel;
330 } else if (truncate) {
331 ip->i_d.di_size = 0;
332 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
333
334 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
335 if (error)
336 goto out_cancel;
337
338 ASSERT(ip->i_d.di_nextents == 0);
339 }
340
341 /*
342 * If there are attributes associated with the file then blow them away
343 * now. The code calls a routine that recursively deconstructs the
344 * attribute fork. We need to just commit the current transaction
345 * because we can't use it for xfs_attr_inactive().
346 */
347 if (ip->i_d.di_anextents > 0) {
348 ASSERT(ip->i_d.di_forkoff != 0);
349
350 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
351 if (error)
352 goto out_unlock;
353
354 xfs_iunlock(ip, XFS_ILOCK_EXCL);
355
356 error = xfs_attr_inactive(ip);
357 if (error)
358 goto out;
359
360 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
361 error = xfs_trans_reserve(tp, 0,
362 XFS_IFREE_LOG_RES(mp),
363 0, XFS_TRANS_PERM_LOG_RES,
364 XFS_INACTIVE_LOG_COUNT);
365 if (error) {
366 xfs_trans_cancel(tp, 0);
367 goto out;
368 }
369
370 xfs_ilock(ip, XFS_ILOCK_EXCL);
371 xfs_trans_ijoin(tp, ip, 0);
372 }
373
374 if (ip->i_afp)
375 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
376
377 ASSERT(ip->i_d.di_anextents == 0);
378
379 /*
380 * Free the inode.
381 */
382 xfs_bmap_init(&free_list, &first_block);
383 error = xfs_ifree(tp, ip, &free_list);
384 if (error) {
385 /*
386 * If we fail to free the inode, shut down. The cancel
387 * might do that, we need to make sure. Otherwise the
388 * inode might be lost for a long time or forever.
389 */
390 if (!XFS_FORCED_SHUTDOWN(mp)) {
391 xfs_notice(mp, "%s: xfs_ifree returned error %d",
392 __func__, error);
393 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
394 }
395 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
396 } else {
397 /*
398 * Credit the quota account(s). The inode is gone.
399 */
400 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
401
402 /*
403 * Just ignore errors at this point. There is nothing we can
404 * do except to try to keep going. Make sure it's not a silent
405 * error.
406 */
407 error = xfs_bmap_finish(&tp, &free_list, &committed);
408 if (error)
409 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
410 __func__, error);
411 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
412 if (error)
413 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
414 __func__, error);
415 }
416
417 /*
418 * Release the dquots held by inode, if any.
419 */
420 xfs_qm_dqdetach(ip);
421 out_unlock:
422 xfs_iunlock(ip, XFS_ILOCK_EXCL);
423 out:
424 return VN_INACTIVE_CACHE;
425 out_cancel:
426 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
427 goto out_unlock;
428 }
429
430 /*
431 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
432 * is allowed, otherwise it has to be an exact match. If a CI match is found,
433 * ci_name->name will point to a the actual name (caller must free) or
434 * will be set to NULL if an exact match is found.
435 */
436 int
437 xfs_lookup(
438 xfs_inode_t *dp,
439 struct xfs_name *name,
440 xfs_inode_t **ipp,
441 struct xfs_name *ci_name)
442 {
443 xfs_ino_t inum;
444 int error;
445 uint lock_mode;
446
447 trace_xfs_lookup(dp, name);
448
449 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
450 return XFS_ERROR(EIO);
451
452 lock_mode = xfs_ilock_map_shared(dp);
453 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
454 xfs_iunlock_map_shared(dp, lock_mode);
455
456 if (error)
457 goto out;
458
459 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
460 if (error)
461 goto out_free_name;
462
463 return 0;
464
465 out_free_name:
466 if (ci_name)
467 kmem_free(ci_name->name);
468 out:
469 *ipp = NULL;
470 return error;
471 }
472
473 int
474 xfs_create(
475 xfs_inode_t *dp,
476 struct xfs_name *name,
477 umode_t mode,
478 xfs_dev_t rdev,
479 xfs_inode_t **ipp)
480 {
481 int is_dir = S_ISDIR(mode);
482 struct xfs_mount *mp = dp->i_mount;
483 struct xfs_inode *ip = NULL;
484 struct xfs_trans *tp = NULL;
485 int error;
486 xfs_bmap_free_t free_list;
487 xfs_fsblock_t first_block;
488 bool unlock_dp_on_error = false;
489 uint cancel_flags;
490 int committed;
491 prid_t prid;
492 struct xfs_dquot *udqp = NULL;
493 struct xfs_dquot *gdqp = NULL;
494 struct xfs_dquot *pdqp = NULL;
495 uint resblks;
496 uint log_res;
497 uint log_count;
498
499 trace_xfs_create(dp, name);
500
501 if (XFS_FORCED_SHUTDOWN(mp))
502 return XFS_ERROR(EIO);
503
504 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
505 prid = xfs_get_projid(dp);
506 else
507 prid = XFS_PROJID_DEFAULT;
508
509 /*
510 * Make sure that we have allocated dquot(s) on disk.
511 */
512 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
513 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
514 &udqp, &gdqp, &pdqp);
515 if (error)
516 return error;
517
518 if (is_dir) {
519 rdev = 0;
520 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
521 log_res = XFS_MKDIR_LOG_RES(mp);
522 log_count = XFS_MKDIR_LOG_COUNT;
523 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
524 } else {
525 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
526 log_res = XFS_CREATE_LOG_RES(mp);
527 log_count = XFS_CREATE_LOG_COUNT;
528 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
529 }
530
531 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
532
533 /*
534 * Initially assume that the file does not exist and
535 * reserve the resources for that case. If that is not
536 * the case we'll drop the one we have and get a more
537 * appropriate transaction later.
538 */
539 error = xfs_trans_reserve(tp, resblks, log_res, 0,
540 XFS_TRANS_PERM_LOG_RES, log_count);
541 if (error == ENOSPC) {
542 /* flush outstanding delalloc blocks and retry */
543 xfs_flush_inodes(mp);
544 error = xfs_trans_reserve(tp, resblks, log_res, 0,
545 XFS_TRANS_PERM_LOG_RES, log_count);
546 }
547 if (error == ENOSPC) {
548 /* No space at all so try a "no-allocation" reservation */
549 resblks = 0;
550 error = xfs_trans_reserve(tp, 0, log_res, 0,
551 XFS_TRANS_PERM_LOG_RES, log_count);
552 }
553 if (error) {
554 cancel_flags = 0;
555 goto out_trans_cancel;
556 }
557
558 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
559 unlock_dp_on_error = true;
560
561 xfs_bmap_init(&free_list, &first_block);
562
563 /*
564 * Reserve disk quota and the inode.
565 */
566 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
567 pdqp, resblks, 1, 0);
568 if (error)
569 goto out_trans_cancel;
570
571 error = xfs_dir_canenter(tp, dp, name, resblks);
572 if (error)
573 goto out_trans_cancel;
574
575 /*
576 * A newly created regular or special file just has one directory
577 * entry pointing to them, but a directory also the "." entry
578 * pointing to itself.
579 */
580 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
581 prid, resblks > 0, &ip, &committed);
582 if (error) {
583 if (error == ENOSPC)
584 goto out_trans_cancel;
585 goto out_trans_abort;
586 }
587
588 /*
589 * Now we join the directory inode to the transaction. We do not do it
590 * earlier because xfs_dir_ialloc might commit the previous transaction
591 * (and release all the locks). An error from here on will result in
592 * the transaction cancel unlocking dp so don't do it explicitly in the
593 * error path.
594 */
595 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
596 unlock_dp_on_error = false;
597
598 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
599 &first_block, &free_list, resblks ?
600 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
601 if (error) {
602 ASSERT(error != ENOSPC);
603 goto out_trans_abort;
604 }
605 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
606 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
607
608 if (is_dir) {
609 error = xfs_dir_init(tp, ip, dp);
610 if (error)
611 goto out_bmap_cancel;
612
613 error = xfs_bumplink(tp, dp);
614 if (error)
615 goto out_bmap_cancel;
616 }
617
618 /*
619 * If this is a synchronous mount, make sure that the
620 * create transaction goes to disk before returning to
621 * the user.
622 */
623 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
624 xfs_trans_set_sync(tp);
625
626 /*
627 * Attach the dquot(s) to the inodes and modify them incore.
628 * These ids of the inode couldn't have changed since the new
629 * inode has been locked ever since it was created.
630 */
631 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
632
633 error = xfs_bmap_finish(&tp, &free_list, &committed);
634 if (error)
635 goto out_bmap_cancel;
636
637 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
638 if (error)
639 goto out_release_inode;
640
641 xfs_qm_dqrele(udqp);
642 xfs_qm_dqrele(gdqp);
643 xfs_qm_dqrele(pdqp);
644
645 *ipp = ip;
646 return 0;
647
648 out_bmap_cancel:
649 xfs_bmap_cancel(&free_list);
650 out_trans_abort:
651 cancel_flags |= XFS_TRANS_ABORT;
652 out_trans_cancel:
653 xfs_trans_cancel(tp, cancel_flags);
654 out_release_inode:
655 /*
656 * Wait until after the current transaction is aborted to
657 * release the inode. This prevents recursive transactions
658 * and deadlocks from xfs_inactive.
659 */
660 if (ip)
661 IRELE(ip);
662
663 xfs_qm_dqrele(udqp);
664 xfs_qm_dqrele(gdqp);
665 xfs_qm_dqrele(pdqp);
666
667 if (unlock_dp_on_error)
668 xfs_iunlock(dp, XFS_ILOCK_EXCL);
669 return error;
670 }
671
672 #ifdef DEBUG
673 int xfs_locked_n;
674 int xfs_small_retries;
675 int xfs_middle_retries;
676 int xfs_lots_retries;
677 int xfs_lock_delays;
678 #endif
679
680 /*
681 * Bump the subclass so xfs_lock_inodes() acquires each lock with
682 * a different value
683 */
684 static inline int
685 xfs_lock_inumorder(int lock_mode, int subclass)
686 {
687 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
688 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
689 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
690 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
691
692 return lock_mode;
693 }
694
695 /*
696 * The following routine will lock n inodes in exclusive mode.
697 * We assume the caller calls us with the inodes in i_ino order.
698 *
699 * We need to detect deadlock where an inode that we lock
700 * is in the AIL and we start waiting for another inode that is locked
701 * by a thread in a long running transaction (such as truncate). This can
702 * result in deadlock since the long running trans might need to wait
703 * for the inode we just locked in order to push the tail and free space
704 * in the log.
705 */
706 void
707 xfs_lock_inodes(
708 xfs_inode_t **ips,
709 int inodes,
710 uint lock_mode)
711 {
712 int attempts = 0, i, j, try_lock;
713 xfs_log_item_t *lp;
714
715 ASSERT(ips && (inodes >= 2)); /* we need at least two */
716
717 try_lock = 0;
718 i = 0;
719
720 again:
721 for (; i < inodes; i++) {
722 ASSERT(ips[i]);
723
724 if (i && (ips[i] == ips[i-1])) /* Already locked */
725 continue;
726
727 /*
728 * If try_lock is not set yet, make sure all locked inodes
729 * are not in the AIL.
730 * If any are, set try_lock to be used later.
731 */
732
733 if (!try_lock) {
734 for (j = (i - 1); j >= 0 && !try_lock; j--) {
735 lp = (xfs_log_item_t *)ips[j]->i_itemp;
736 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
737 try_lock++;
738 }
739 }
740 }
741
742 /*
743 * If any of the previous locks we have locked is in the AIL,
744 * we must TRY to get the second and subsequent locks. If
745 * we can't get any, we must release all we have
746 * and try again.
747 */
748
749 if (try_lock) {
750 /* try_lock must be 0 if i is 0. */
751 /*
752 * try_lock means we have an inode locked
753 * that is in the AIL.
754 */
755 ASSERT(i != 0);
756 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
757 attempts++;
758
759 /*
760 * Unlock all previous guys and try again.
761 * xfs_iunlock will try to push the tail
762 * if the inode is in the AIL.
763 */
764
765 for(j = i - 1; j >= 0; j--) {
766
767 /*
768 * Check to see if we've already
769 * unlocked this one.
770 * Not the first one going back,
771 * and the inode ptr is the same.
772 */
773 if ((j != (i - 1)) && ips[j] ==
774 ips[j+1])
775 continue;
776
777 xfs_iunlock(ips[j], lock_mode);
778 }
779
780 if ((attempts % 5) == 0) {
781 delay(1); /* Don't just spin the CPU */
782 #ifdef DEBUG
783 xfs_lock_delays++;
784 #endif
785 }
786 i = 0;
787 try_lock = 0;
788 goto again;
789 }
790 } else {
791 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
792 }
793 }
794
795 #ifdef DEBUG
796 if (attempts) {
797 if (attempts < 5) xfs_small_retries++;
798 else if (attempts < 100) xfs_middle_retries++;
799 else xfs_lots_retries++;
800 } else {
801 xfs_locked_n++;
802 }
803 #endif
804 }
805
806 /*
807 * xfs_lock_two_inodes() can only be used to lock one type of lock
808 * at a time - the iolock or the ilock, but not both at once. If
809 * we lock both at once, lockdep will report false positives saying
810 * we have violated locking orders.
811 */
812 void
813 xfs_lock_two_inodes(
814 xfs_inode_t *ip0,
815 xfs_inode_t *ip1,
816 uint lock_mode)
817 {
818 xfs_inode_t *temp;
819 int attempts = 0;
820 xfs_log_item_t *lp;
821
822 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
823 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
824 ASSERT(ip0->i_ino != ip1->i_ino);
825
826 if (ip0->i_ino > ip1->i_ino) {
827 temp = ip0;
828 ip0 = ip1;
829 ip1 = temp;
830 }
831
832 again:
833 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
834
835 /*
836 * If the first lock we have locked is in the AIL, we must TRY to get
837 * the second lock. If we can't get it, we must release the first one
838 * and try again.
839 */
840 lp = (xfs_log_item_t *)ip0->i_itemp;
841 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
842 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
843 xfs_iunlock(ip0, lock_mode);
844 if ((++attempts % 5) == 0)
845 delay(1); /* Don't just spin the CPU */
846 goto again;
847 }
848 } else {
849 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
850 }
851 }
852
853 int
854 xfs_remove(
855 xfs_inode_t *dp,
856 struct xfs_name *name,
857 xfs_inode_t *ip)
858 {
859 xfs_mount_t *mp = dp->i_mount;
860 xfs_trans_t *tp = NULL;
861 int is_dir = S_ISDIR(ip->i_d.di_mode);
862 int error = 0;
863 xfs_bmap_free_t free_list;
864 xfs_fsblock_t first_block;
865 int cancel_flags;
866 int committed;
867 int link_zero;
868 uint resblks;
869 uint log_count;
870
871 trace_xfs_remove(dp, name);
872
873 if (XFS_FORCED_SHUTDOWN(mp))
874 return XFS_ERROR(EIO);
875
876 error = xfs_qm_dqattach(dp, 0);
877 if (error)
878 goto std_return;
879
880 error = xfs_qm_dqattach(ip, 0);
881 if (error)
882 goto std_return;
883
884 if (is_dir) {
885 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
886 log_count = XFS_DEFAULT_LOG_COUNT;
887 } else {
888 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
889 log_count = XFS_REMOVE_LOG_COUNT;
890 }
891 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
892
893 /*
894 * We try to get the real space reservation first,
895 * allowing for directory btree deletion(s) implying
896 * possible bmap insert(s). If we can't get the space
897 * reservation then we use 0 instead, and avoid the bmap
898 * btree insert(s) in the directory code by, if the bmap
899 * insert tries to happen, instead trimming the LAST
900 * block from the directory.
901 */
902 resblks = XFS_REMOVE_SPACE_RES(mp);
903 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
904 XFS_TRANS_PERM_LOG_RES, log_count);
905 if (error == ENOSPC) {
906 resblks = 0;
907 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
908 XFS_TRANS_PERM_LOG_RES, log_count);
909 }
910 if (error) {
911 ASSERT(error != ENOSPC);
912 cancel_flags = 0;
913 goto out_trans_cancel;
914 }
915
916 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
917
918 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
919 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
920
921 /*
922 * If we're removing a directory perform some additional validation.
923 */
924 if (is_dir) {
925 ASSERT(ip->i_d.di_nlink >= 2);
926 if (ip->i_d.di_nlink != 2) {
927 error = XFS_ERROR(ENOTEMPTY);
928 goto out_trans_cancel;
929 }
930 if (!xfs_dir_isempty(ip)) {
931 error = XFS_ERROR(ENOTEMPTY);
932 goto out_trans_cancel;
933 }
934 }
935
936 xfs_bmap_init(&free_list, &first_block);
937 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
938 &first_block, &free_list, resblks);
939 if (error) {
940 ASSERT(error != ENOENT);
941 goto out_bmap_cancel;
942 }
943 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
944
945 if (is_dir) {
946 /*
947 * Drop the link from ip's "..".
948 */
949 error = xfs_droplink(tp, dp);
950 if (error)
951 goto out_bmap_cancel;
952
953 /*
954 * Drop the "." link from ip to self.
955 */
956 error = xfs_droplink(tp, ip);
957 if (error)
958 goto out_bmap_cancel;
959 } else {
960 /*
961 * When removing a non-directory we need to log the parent
962 * inode here. For a directory this is done implicitly
963 * by the xfs_droplink call for the ".." entry.
964 */
965 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
966 }
967
968 /*
969 * Drop the link from dp to ip.
970 */
971 error = xfs_droplink(tp, ip);
972 if (error)
973 goto out_bmap_cancel;
974
975 /*
976 * Determine if this is the last link while
977 * we are in the transaction.
978 */
979 link_zero = (ip->i_d.di_nlink == 0);
980
981 /*
982 * If this is a synchronous mount, make sure that the
983 * remove transaction goes to disk before returning to
984 * the user.
985 */
986 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
987 xfs_trans_set_sync(tp);
988
989 error = xfs_bmap_finish(&tp, &free_list, &committed);
990 if (error)
991 goto out_bmap_cancel;
992
993 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
994 if (error)
995 goto std_return;
996
997 /*
998 * If we are using filestreams, kill the stream association.
999 * If the file is still open it may get a new one but that
1000 * will get killed on last close in xfs_close() so we don't
1001 * have to worry about that.
1002 */
1003 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1004 xfs_filestream_deassociate(ip);
1005
1006 return 0;
1007
1008 out_bmap_cancel:
1009 xfs_bmap_cancel(&free_list);
1010 cancel_flags |= XFS_TRANS_ABORT;
1011 out_trans_cancel:
1012 xfs_trans_cancel(tp, cancel_flags);
1013 std_return:
1014 return error;
1015 }
1016
1017 int
1018 xfs_link(
1019 xfs_inode_t *tdp,
1020 xfs_inode_t *sip,
1021 struct xfs_name *target_name)
1022 {
1023 xfs_mount_t *mp = tdp->i_mount;
1024 xfs_trans_t *tp;
1025 int error;
1026 xfs_bmap_free_t free_list;
1027 xfs_fsblock_t first_block;
1028 int cancel_flags;
1029 int committed;
1030 int resblks;
1031
1032 trace_xfs_link(tdp, target_name);
1033
1034 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1035
1036 if (XFS_FORCED_SHUTDOWN(mp))
1037 return XFS_ERROR(EIO);
1038
1039 error = xfs_qm_dqattach(sip, 0);
1040 if (error)
1041 goto std_return;
1042
1043 error = xfs_qm_dqattach(tdp, 0);
1044 if (error)
1045 goto std_return;
1046
1047 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1048 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1049 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1050 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1051 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1052 if (error == ENOSPC) {
1053 resblks = 0;
1054 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1055 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1056 }
1057 if (error) {
1058 cancel_flags = 0;
1059 goto error_return;
1060 }
1061
1062 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1063
1064 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1065 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1066
1067 /*
1068 * If we are using project inheritance, we only allow hard link
1069 * creation in our tree when the project IDs are the same; else
1070 * the tree quota mechanism could be circumvented.
1071 */
1072 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1073 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1074 error = XFS_ERROR(EXDEV);
1075 goto error_return;
1076 }
1077
1078 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1079 if (error)
1080 goto error_return;
1081
1082 xfs_bmap_init(&free_list, &first_block);
1083
1084 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1085 &first_block, &free_list, resblks);
1086 if (error)
1087 goto abort_return;
1088 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1089 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1090
1091 error = xfs_bumplink(tp, sip);
1092 if (error)
1093 goto abort_return;
1094
1095 /*
1096 * If this is a synchronous mount, make sure that the
1097 * link transaction goes to disk before returning to
1098 * the user.
1099 */
1100 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1101 xfs_trans_set_sync(tp);
1102 }
1103
1104 error = xfs_bmap_finish (&tp, &free_list, &committed);
1105 if (error) {
1106 xfs_bmap_cancel(&free_list);
1107 goto abort_return;
1108 }
1109
1110 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1111
1112 abort_return:
1113 cancel_flags |= XFS_TRANS_ABORT;
1114 error_return:
1115 xfs_trans_cancel(tp, cancel_flags);
1116 std_return:
1117 return error;
1118 }
1119
1120 int
1121 xfs_set_dmattrs(
1122 xfs_inode_t *ip,
1123 u_int evmask,
1124 u_int16_t state)
1125 {
1126 xfs_mount_t *mp = ip->i_mount;
1127 xfs_trans_t *tp;
1128 int error;
1129
1130 if (!capable(CAP_SYS_ADMIN))
1131 return XFS_ERROR(EPERM);
1132
1133 if (XFS_FORCED_SHUTDOWN(mp))
1134 return XFS_ERROR(EIO);
1135
1136 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1137 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1138 if (error) {
1139 xfs_trans_cancel(tp, 0);
1140 return error;
1141 }
1142 xfs_ilock(ip, XFS_ILOCK_EXCL);
1143 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1144
1145 ip->i_d.di_dmevmask = evmask;
1146 ip->i_d.di_dmstate = state;
1147
1148 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1149 error = xfs_trans_commit(tp, 0);
1150
1151 return error;
1152 }
1153
1154 /*
1155 * xfs_alloc_file_space()
1156 * This routine allocates disk space for the given file.
1157 *
1158 * If alloc_type == 0, this request is for an ALLOCSP type
1159 * request which will change the file size. In this case, no
1160 * DMAPI event will be generated by the call. A TRUNCATE event
1161 * will be generated later by xfs_setattr.
1162 *
1163 * If alloc_type != 0, this request is for a RESVSP type
1164 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1165 * lower block boundary byte address is less than the file's
1166 * length.
1167 *
1168 * RETURNS:
1169 * 0 on success
1170 * errno on error
1171 *
1172 */
1173 STATIC int
1174 xfs_alloc_file_space(
1175 xfs_inode_t *ip,
1176 xfs_off_t offset,
1177 xfs_off_t len,
1178 int alloc_type,
1179 int attr_flags)
1180 {
1181 xfs_mount_t *mp = ip->i_mount;
1182 xfs_off_t count;
1183 xfs_filblks_t allocated_fsb;
1184 xfs_filblks_t allocatesize_fsb;
1185 xfs_extlen_t extsz, temp;
1186 xfs_fileoff_t startoffset_fsb;
1187 xfs_fsblock_t firstfsb;
1188 int nimaps;
1189 int quota_flag;
1190 int rt;
1191 xfs_trans_t *tp;
1192 xfs_bmbt_irec_t imaps[1], *imapp;
1193 xfs_bmap_free_t free_list;
1194 uint qblocks, resblks, resrtextents;
1195 int committed;
1196 int error;
1197
1198 trace_xfs_alloc_file_space(ip);
1199
1200 if (XFS_FORCED_SHUTDOWN(mp))
1201 return XFS_ERROR(EIO);
1202
1203 error = xfs_qm_dqattach(ip, 0);
1204 if (error)
1205 return error;
1206
1207 if (len <= 0)
1208 return XFS_ERROR(EINVAL);
1209
1210 rt = XFS_IS_REALTIME_INODE(ip);
1211 extsz = xfs_get_extsz_hint(ip);
1212
1213 count = len;
1214 imapp = &imaps[0];
1215 nimaps = 1;
1216 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1217 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1218
1219 /*
1220 * Allocate file space until done or until there is an error
1221 */
1222 while (allocatesize_fsb && !error) {
1223 xfs_fileoff_t s, e;
1224
1225 /*
1226 * Determine space reservations for data/realtime.
1227 */
1228 if (unlikely(extsz)) {
1229 s = startoffset_fsb;
1230 do_div(s, extsz);
1231 s *= extsz;
1232 e = startoffset_fsb + allocatesize_fsb;
1233 if ((temp = do_mod(startoffset_fsb, extsz)))
1234 e += temp;
1235 if ((temp = do_mod(e, extsz)))
1236 e += extsz - temp;
1237 } else {
1238 s = 0;
1239 e = allocatesize_fsb;
1240 }
1241
1242 /*
1243 * The transaction reservation is limited to a 32-bit block
1244 * count, hence we need to limit the number of blocks we are
1245 * trying to reserve to avoid an overflow. We can't allocate
1246 * more than @nimaps extents, and an extent is limited on disk
1247 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1248 */
1249 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1250 if (unlikely(rt)) {
1251 resrtextents = qblocks = resblks;
1252 resrtextents /= mp->m_sb.sb_rextsize;
1253 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1254 quota_flag = XFS_QMOPT_RES_RTBLKS;
1255 } else {
1256 resrtextents = 0;
1257 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1258 quota_flag = XFS_QMOPT_RES_REGBLKS;
1259 }
1260
1261 /*
1262 * Allocate and setup the transaction.
1263 */
1264 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1265 error = xfs_trans_reserve(tp, resblks,
1266 XFS_WRITE_LOG_RES(mp), resrtextents,
1267 XFS_TRANS_PERM_LOG_RES,
1268 XFS_WRITE_LOG_COUNT);
1269 /*
1270 * Check for running out of space
1271 */
1272 if (error) {
1273 /*
1274 * Free the transaction structure.
1275 */
1276 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1277 xfs_trans_cancel(tp, 0);
1278 break;
1279 }
1280 xfs_ilock(ip, XFS_ILOCK_EXCL);
1281 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1282 0, quota_flag);
1283 if (error)
1284 goto error1;
1285
1286 xfs_trans_ijoin(tp, ip, 0);
1287
1288 xfs_bmap_init(&free_list, &firstfsb);
1289 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1290 allocatesize_fsb, alloc_type, &firstfsb,
1291 0, imapp, &nimaps, &free_list);
1292 if (error) {
1293 goto error0;
1294 }
1295
1296 /*
1297 * Complete the transaction
1298 */
1299 error = xfs_bmap_finish(&tp, &free_list, &committed);
1300 if (error) {
1301 goto error0;
1302 }
1303
1304 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1305 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1306 if (error) {
1307 break;
1308 }
1309
1310 allocated_fsb = imapp->br_blockcount;
1311
1312 if (nimaps == 0) {
1313 error = XFS_ERROR(ENOSPC);
1314 break;
1315 }
1316
1317 startoffset_fsb += allocated_fsb;
1318 allocatesize_fsb -= allocated_fsb;
1319 }
1320
1321 return error;
1322
1323 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1324 xfs_bmap_cancel(&free_list);
1325 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1326
1327 error1: /* Just cancel transaction */
1328 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1329 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1330 return error;
1331 }
1332
1333 /*
1334 * Zero file bytes between startoff and endoff inclusive.
1335 * The iolock is held exclusive and no blocks are buffered.
1336 *
1337 * This function is used by xfs_free_file_space() to zero
1338 * partial blocks when the range to free is not block aligned.
1339 * When unreserving space with boundaries that are not block
1340 * aligned we round up the start and round down the end
1341 * boundaries and then use this function to zero the parts of
1342 * the blocks that got dropped during the rounding.
1343 */
1344 STATIC int
1345 xfs_zero_remaining_bytes(
1346 xfs_inode_t *ip,
1347 xfs_off_t startoff,
1348 xfs_off_t endoff)
1349 {
1350 xfs_bmbt_irec_t imap;
1351 xfs_fileoff_t offset_fsb;
1352 xfs_off_t lastoffset;
1353 xfs_off_t offset;
1354 xfs_buf_t *bp;
1355 xfs_mount_t *mp = ip->i_mount;
1356 int nimap;
1357 int error = 0;
1358
1359 /*
1360 * Avoid doing I/O beyond eof - it's not necessary
1361 * since nothing can read beyond eof. The space will
1362 * be zeroed when the file is extended anyway.
1363 */
1364 if (startoff >= XFS_ISIZE(ip))
1365 return 0;
1366
1367 if (endoff > XFS_ISIZE(ip))
1368 endoff = XFS_ISIZE(ip);
1369
1370 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1371 mp->m_rtdev_targp : mp->m_ddev_targp,
1372 BTOBB(mp->m_sb.sb_blocksize), 0);
1373 if (!bp)
1374 return XFS_ERROR(ENOMEM);
1375
1376 xfs_buf_unlock(bp);
1377
1378 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1379 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1380 nimap = 1;
1381 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1382 if (error || nimap < 1)
1383 break;
1384 ASSERT(imap.br_blockcount >= 1);
1385 ASSERT(imap.br_startoff == offset_fsb);
1386 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1387 if (lastoffset > endoff)
1388 lastoffset = endoff;
1389 if (imap.br_startblock == HOLESTARTBLOCK)
1390 continue;
1391 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1392 if (imap.br_state == XFS_EXT_UNWRITTEN)
1393 continue;
1394 XFS_BUF_UNDONE(bp);
1395 XFS_BUF_UNWRITE(bp);
1396 XFS_BUF_READ(bp);
1397 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1398 xfsbdstrat(mp, bp);
1399 error = xfs_buf_iowait(bp);
1400 if (error) {
1401 xfs_buf_ioerror_alert(bp,
1402 "xfs_zero_remaining_bytes(read)");
1403 break;
1404 }
1405 memset(bp->b_addr +
1406 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1407 0, lastoffset - offset + 1);
1408 XFS_BUF_UNDONE(bp);
1409 XFS_BUF_UNREAD(bp);
1410 XFS_BUF_WRITE(bp);
1411 xfsbdstrat(mp, bp);
1412 error = xfs_buf_iowait(bp);
1413 if (error) {
1414 xfs_buf_ioerror_alert(bp,
1415 "xfs_zero_remaining_bytes(write)");
1416 break;
1417 }
1418 }
1419 xfs_buf_free(bp);
1420 return error;
1421 }
1422
1423 /*
1424 * xfs_free_file_space()
1425 * This routine frees disk space for the given file.
1426 *
1427 * This routine is only called by xfs_change_file_space
1428 * for an UNRESVSP type call.
1429 *
1430 * RETURNS:
1431 * 0 on success
1432 * errno on error
1433 *
1434 */
1435 STATIC int
1436 xfs_free_file_space(
1437 xfs_inode_t *ip,
1438 xfs_off_t offset,
1439 xfs_off_t len,
1440 int attr_flags)
1441 {
1442 int committed;
1443 int done;
1444 xfs_fileoff_t endoffset_fsb;
1445 int error;
1446 xfs_fsblock_t firstfsb;
1447 xfs_bmap_free_t free_list;
1448 xfs_bmbt_irec_t imap;
1449 xfs_off_t ioffset;
1450 xfs_extlen_t mod=0;
1451 xfs_mount_t *mp;
1452 int nimap;
1453 uint resblks;
1454 xfs_off_t rounding;
1455 int rt;
1456 xfs_fileoff_t startoffset_fsb;
1457 xfs_trans_t *tp;
1458 int need_iolock = 1;
1459
1460 mp = ip->i_mount;
1461
1462 trace_xfs_free_file_space(ip);
1463
1464 error = xfs_qm_dqattach(ip, 0);
1465 if (error)
1466 return error;
1467
1468 error = 0;
1469 if (len <= 0) /* if nothing being freed */
1470 return error;
1471 rt = XFS_IS_REALTIME_INODE(ip);
1472 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1473 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1474
1475 if (attr_flags & XFS_ATTR_NOLOCK)
1476 need_iolock = 0;
1477 if (need_iolock) {
1478 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1479 /* wait for the completion of any pending DIOs */
1480 inode_dio_wait(VFS_I(ip));
1481 }
1482
1483 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1484 ioffset = offset & ~(rounding - 1);
1485 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1486 ioffset, -1);
1487 if (error)
1488 goto out_unlock_iolock;
1489 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1490
1491 /*
1492 * Need to zero the stuff we're not freeing, on disk.
1493 * If it's a realtime file & can't use unwritten extents then we
1494 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1495 * will take care of it for us.
1496 */
1497 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1498 nimap = 1;
1499 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1500 &imap, &nimap, 0);
1501 if (error)
1502 goto out_unlock_iolock;
1503 ASSERT(nimap == 0 || nimap == 1);
1504 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1505 xfs_daddr_t block;
1506
1507 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1508 block = imap.br_startblock;
1509 mod = do_div(block, mp->m_sb.sb_rextsize);
1510 if (mod)
1511 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1512 }
1513 nimap = 1;
1514 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1515 &imap, &nimap, 0);
1516 if (error)
1517 goto out_unlock_iolock;
1518 ASSERT(nimap == 0 || nimap == 1);
1519 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1520 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1521 mod++;
1522 if (mod && (mod != mp->m_sb.sb_rextsize))
1523 endoffset_fsb -= mod;
1524 }
1525 }
1526 if ((done = (endoffset_fsb <= startoffset_fsb)))
1527 /*
1528 * One contiguous piece to clear
1529 */
1530 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1531 else {
1532 /*
1533 * Some full blocks, possibly two pieces to clear
1534 */
1535 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1536 error = xfs_zero_remaining_bytes(ip, offset,
1537 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1538 if (!error &&
1539 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1540 error = xfs_zero_remaining_bytes(ip,
1541 XFS_FSB_TO_B(mp, endoffset_fsb),
1542 offset + len - 1);
1543 }
1544
1545 /*
1546 * free file space until done or until there is an error
1547 */
1548 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1549 while (!error && !done) {
1550
1551 /*
1552 * allocate and setup the transaction. Allow this
1553 * transaction to dip into the reserve blocks to ensure
1554 * the freeing of the space succeeds at ENOSPC.
1555 */
1556 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1557 tp->t_flags |= XFS_TRANS_RESERVE;
1558 error = xfs_trans_reserve(tp,
1559 resblks,
1560 XFS_WRITE_LOG_RES(mp),
1561 0,
1562 XFS_TRANS_PERM_LOG_RES,
1563 XFS_WRITE_LOG_COUNT);
1564
1565 /*
1566 * check for running out of space
1567 */
1568 if (error) {
1569 /*
1570 * Free the transaction structure.
1571 */
1572 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1573 xfs_trans_cancel(tp, 0);
1574 break;
1575 }
1576 xfs_ilock(ip, XFS_ILOCK_EXCL);
1577 error = xfs_trans_reserve_quota(tp, mp,
1578 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1579 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1580 if (error)
1581 goto error1;
1582
1583 xfs_trans_ijoin(tp, ip, 0);
1584
1585 /*
1586 * issue the bunmapi() call to free the blocks
1587 */
1588 xfs_bmap_init(&free_list, &firstfsb);
1589 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1590 endoffset_fsb - startoffset_fsb,
1591 0, 2, &firstfsb, &free_list, &done);
1592 if (error) {
1593 goto error0;
1594 }
1595
1596 /*
1597 * complete the transaction
1598 */
1599 error = xfs_bmap_finish(&tp, &free_list, &committed);
1600 if (error) {
1601 goto error0;
1602 }
1603
1604 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1605 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1606 }
1607
1608 out_unlock_iolock:
1609 if (need_iolock)
1610 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1611 return error;
1612
1613 error0:
1614 xfs_bmap_cancel(&free_list);
1615 error1:
1616 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1617 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
1618 XFS_ILOCK_EXCL);
1619 return error;
1620 }
1621
1622
1623 STATIC int
1624 xfs_zero_file_space(
1625 struct xfs_inode *ip,
1626 xfs_off_t offset,
1627 xfs_off_t len,
1628 int attr_flags)
1629 {
1630 struct xfs_mount *mp = ip->i_mount;
1631 uint granularity;
1632 xfs_off_t start_boundary;
1633 xfs_off_t end_boundary;
1634 int error;
1635
1636 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1637
1638 /*
1639 * Round the range of extents we are going to convert inwards. If the
1640 * offset is aligned, then it doesn't get changed so we zero from the
1641 * start of the block offset points to.
1642 */
1643 start_boundary = round_up(offset, granularity);
1644 end_boundary = round_down(offset + len, granularity);
1645
1646 ASSERT(start_boundary >= offset);
1647 ASSERT(end_boundary <= offset + len);
1648
1649 if (!(attr_flags & XFS_ATTR_NOLOCK))
1650 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1651
1652 if (start_boundary < end_boundary - 1) {
1653 /* punch out the page cache over the conversion range */
1654 truncate_pagecache_range(VFS_I(ip), start_boundary,
1655 end_boundary - 1);
1656 /* convert the blocks */
1657 error = xfs_alloc_file_space(ip, start_boundary,
1658 end_boundary - start_boundary - 1,
1659 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT,
1660 attr_flags);
1661 if (error)
1662 goto out_unlock;
1663
1664 /* We've handled the interior of the range, now for the edges */
1665 if (start_boundary != offset)
1666 error = xfs_iozero(ip, offset, start_boundary - offset);
1667 if (error)
1668 goto out_unlock;
1669
1670 if (end_boundary != offset + len)
1671 error = xfs_iozero(ip, end_boundary,
1672 offset + len - end_boundary);
1673
1674 } else {
1675 /*
1676 * It's either a sub-granularity range or the range spanned lies
1677 * partially across two adjacent blocks.
1678 */
1679 error = xfs_iozero(ip, offset, len);
1680 }
1681
1682 out_unlock:
1683 if (!(attr_flags & XFS_ATTR_NOLOCK))
1684 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1685 return error;
1686
1687 }
1688
1689 /*
1690 * xfs_change_file_space()
1691 * This routine allocates or frees disk space for the given file.
1692 * The user specified parameters are checked for alignment and size
1693 * limitations.
1694 *
1695 * RETURNS:
1696 * 0 on success
1697 * errno on error
1698 *
1699 */
1700 int
1701 xfs_change_file_space(
1702 xfs_inode_t *ip,
1703 int cmd,
1704 xfs_flock64_t *bf,
1705 xfs_off_t offset,
1706 int attr_flags)
1707 {
1708 xfs_mount_t *mp = ip->i_mount;
1709 int clrprealloc;
1710 int error;
1711 xfs_fsize_t fsize;
1712 int setprealloc;
1713 xfs_off_t startoffset;
1714 xfs_trans_t *tp;
1715 struct iattr iattr;
1716
1717 if (!S_ISREG(ip->i_d.di_mode))
1718 return XFS_ERROR(EINVAL);
1719
1720 switch (bf->l_whence) {
1721 case 0: /*SEEK_SET*/
1722 break;
1723 case 1: /*SEEK_CUR*/
1724 bf->l_start += offset;
1725 break;
1726 case 2: /*SEEK_END*/
1727 bf->l_start += XFS_ISIZE(ip);
1728 break;
1729 default:
1730 return XFS_ERROR(EINVAL);
1731 }
1732
1733 /*
1734 * length of <= 0 for resv/unresv/zero is invalid. length for
1735 * alloc/free is ignored completely and we have no idea what userspace
1736 * might have set it to, so set it to zero to allow range
1737 * checks to pass.
1738 */
1739 switch (cmd) {
1740 case XFS_IOC_ZERO_RANGE:
1741 case XFS_IOC_RESVSP:
1742 case XFS_IOC_RESVSP64:
1743 case XFS_IOC_UNRESVSP:
1744 case XFS_IOC_UNRESVSP64:
1745 if (bf->l_len <= 0)
1746 return XFS_ERROR(EINVAL);
1747 break;
1748 default:
1749 bf->l_len = 0;
1750 break;
1751 }
1752
1753 if (bf->l_start < 0 ||
1754 bf->l_start > mp->m_super->s_maxbytes ||
1755 bf->l_start + bf->l_len < 0 ||
1756 bf->l_start + bf->l_len >= mp->m_super->s_maxbytes)
1757 return XFS_ERROR(EINVAL);
1758
1759 bf->l_whence = 0;
1760
1761 startoffset = bf->l_start;
1762 fsize = XFS_ISIZE(ip);
1763
1764 setprealloc = clrprealloc = 0;
1765 switch (cmd) {
1766 case XFS_IOC_ZERO_RANGE:
1767 error = xfs_zero_file_space(ip, startoffset, bf->l_len,
1768 attr_flags);
1769 if (error)
1770 return error;
1771 setprealloc = 1;
1772 break;
1773
1774 case XFS_IOC_RESVSP:
1775 case XFS_IOC_RESVSP64:
1776 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
1777 XFS_BMAPI_PREALLOC, attr_flags);
1778 if (error)
1779 return error;
1780 setprealloc = 1;
1781 break;
1782
1783 case XFS_IOC_UNRESVSP:
1784 case XFS_IOC_UNRESVSP64:
1785 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
1786 attr_flags)))
1787 return error;
1788 break;
1789
1790 case XFS_IOC_ALLOCSP:
1791 case XFS_IOC_ALLOCSP64:
1792 case XFS_IOC_FREESP:
1793 case XFS_IOC_FREESP64:
1794 /*
1795 * These operations actually do IO when extending the file, but
1796 * the allocation is done seperately to the zeroing that is
1797 * done. This set of operations need to be serialised against
1798 * other IO operations, such as truncate and buffered IO. We
1799 * need to take the IOLOCK here to serialise the allocation and
1800 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
1801 * truncate, direct IO) from racing against the transient
1802 * allocated but not written state we can have here.
1803 */
1804 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1805 if (startoffset > fsize) {
1806 error = xfs_alloc_file_space(ip, fsize,
1807 startoffset - fsize, 0,
1808 attr_flags | XFS_ATTR_NOLOCK);
1809 if (error) {
1810 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1811 break;
1812 }
1813 }
1814
1815 iattr.ia_valid = ATTR_SIZE;
1816 iattr.ia_size = startoffset;
1817
1818 error = xfs_setattr_size(ip, &iattr,
1819 attr_flags | XFS_ATTR_NOLOCK);
1820 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1821
1822 if (error)
1823 return error;
1824
1825 clrprealloc = 1;
1826 break;
1827
1828 default:
1829 ASSERT(0);
1830 return XFS_ERROR(EINVAL);
1831 }
1832
1833 /*
1834 * update the inode timestamp, mode, and prealloc flag bits
1835 */
1836 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
1837
1838 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
1839 0, 0, 0))) {
1840 /* ASSERT(0); */
1841 xfs_trans_cancel(tp, 0);
1842 return error;
1843 }
1844
1845 xfs_ilock(ip, XFS_ILOCK_EXCL);
1846 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1847
1848 if ((attr_flags & XFS_ATTR_DMI) == 0) {
1849 ip->i_d.di_mode &= ~S_ISUID;
1850
1851 /*
1852 * Note that we don't have to worry about mandatory
1853 * file locking being disabled here because we only
1854 * clear the S_ISGID bit if the Group execute bit is
1855 * on, but if it was on then mandatory locking wouldn't
1856 * have been enabled.
1857 */
1858 if (ip->i_d.di_mode & S_IXGRP)
1859 ip->i_d.di_mode &= ~S_ISGID;
1860
1861 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1862 }
1863 if (setprealloc)
1864 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
1865 else if (clrprealloc)
1866 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
1867
1868 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1869 if (attr_flags & XFS_ATTR_SYNC)
1870 xfs_trans_set_sync(tp);
1871 return xfs_trans_commit(tp, 0);
1872 }
Linux kernel - Deliverables
This page took 0.07316 seconds and 5 git commands to generate.