projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xfs: clean up unused codes at xfs_bulkstat()
[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_types.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_dir2.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_itable.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_alloc.h"
39 #include "xfs_bmap.h"
40 #include "xfs_acl.h"
41 #include "xfs_attr.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_rtalloc.h"
46 #include "xfs_trans_space.h"
47 #include "xfs_log_priv.h"
48 #include "xfs_filestream.h"
49 #include "xfs_vnodeops.h"
50 #include "xfs_trace.h"
51 #include "xfs_icache.h"
52 #include "xfs_symlink.h"
53
54
55 /*
56 * This is called by xfs_inactive to free any blocks beyond eof
57 * when the link count isn't zero and by xfs_dm_punch_hole() when
58 * punching a hole to EOF.
59 */
60 int
61 xfs_free_eofblocks(
62 xfs_mount_t *mp,
63 xfs_inode_t *ip,
64 bool need_iolock)
65 {
66 xfs_trans_t *tp;
67 int error;
68 xfs_fileoff_t end_fsb;
69 xfs_fileoff_t last_fsb;
70 xfs_filblks_t map_len;
71 int nimaps;
72 xfs_bmbt_irec_t imap;
73
74 /*
75 * Figure out if there are any blocks beyond the end
76 * of the file. If not, then there is nothing to do.
77 */
78 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
79 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
80 if (last_fsb <= end_fsb)
81 return 0;
82 map_len = last_fsb - end_fsb;
83
84 nimaps = 1;
85 xfs_ilock(ip, XFS_ILOCK_SHARED);
86 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
87 xfs_iunlock(ip, XFS_ILOCK_SHARED);
88
89 if (!error && (nimaps != 0) &&
90 (imap.br_startblock != HOLESTARTBLOCK ||
91 ip->i_delayed_blks)) {
92 /*
93 * Attach the dquots to the inode up front.
94 */
95 error = xfs_qm_dqattach(ip, 0);
96 if (error)
97 return error;
98
99 /*
100 * There are blocks after the end of file.
101 * Free them up now by truncating the file to
102 * its current size.
103 */
104 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
105
106 if (need_iolock) {
107 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
108 xfs_trans_cancel(tp, 0);
109 return EAGAIN;
110 }
111 }
112
113 error = xfs_trans_reserve(tp, 0,
114 XFS_ITRUNCATE_LOG_RES(mp),
115 0, XFS_TRANS_PERM_LOG_RES,
116 XFS_ITRUNCATE_LOG_COUNT);
117 if (error) {
118 ASSERT(XFS_FORCED_SHUTDOWN(mp));
119 xfs_trans_cancel(tp, 0);
120 if (need_iolock)
121 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
122 return error;
123 }
124
125 xfs_ilock(ip, XFS_ILOCK_EXCL);
126 xfs_trans_ijoin(tp, ip, 0);
127
128 /*
129 * Do not update the on-disk file size. If we update the
130 * on-disk file size and then the system crashes before the
131 * contents of the file are flushed to disk then the files
132 * may be full of holes (ie NULL files bug).
133 */
134 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
135 XFS_ISIZE(ip));
136 if (error) {
137 /*
138 * If we get an error at this point we simply don't
139 * bother truncating the file.
140 */
141 xfs_trans_cancel(tp,
142 (XFS_TRANS_RELEASE_LOG_RES |
143 XFS_TRANS_ABORT));
144 } else {
145 error = xfs_trans_commit(tp,
146 XFS_TRANS_RELEASE_LOG_RES);
147 if (!error)
148 xfs_inode_clear_eofblocks_tag(ip);
149 }
150
151 xfs_iunlock(ip, XFS_ILOCK_EXCL);
152 if (need_iolock)
153 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
154 }
155 return error;
156 }
157
158 int
159 xfs_release(
160 xfs_inode_t *ip)
161 {
162 xfs_mount_t *mp = ip->i_mount;
163 int error;
164
165 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
166 return 0;
167
168 /* If this is a read-only mount, don't do this (would generate I/O) */
169 if (mp->m_flags & XFS_MOUNT_RDONLY)
170 return 0;
171
172 if (!XFS_FORCED_SHUTDOWN(mp)) {
173 int truncated;
174
175 /*
176 * If we are using filestreams, and we have an unlinked
177 * file that we are processing the last close on, then nothing
178 * will be able to reopen and write to this file. Purge this
179 * inode from the filestreams cache so that it doesn't delay
180 * teardown of the inode.
181 */
182 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
183 xfs_filestream_deassociate(ip);
184
185 /*
186 * If we previously truncated this file and removed old data
187 * in the process, we want to initiate "early" writeout on
188 * the last close. This is an attempt to combat the notorious
189 * NULL files problem which is particularly noticeable from a
190 * truncate down, buffered (re-)write (delalloc), followed by
191 * a crash. What we are effectively doing here is
192 * significantly reducing the time window where we'd otherwise
193 * be exposed to that problem.
194 */
195 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
196 if (truncated) {
197 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
198 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
199 error = -filemap_flush(VFS_I(ip)->i_mapping);
200 if (error)
201 return error;
202 }
203 }
204 }
205
206 if (ip->i_d.di_nlink == 0)
207 return 0;
208
209 if (xfs_can_free_eofblocks(ip, false)) {
210
211 /*
212 * If we can't get the iolock just skip truncating the blocks
213 * past EOF because we could deadlock with the mmap_sem
214 * otherwise. We'll get another chance to drop them once the
215 * last reference to the inode is dropped, so we'll never leak
216 * blocks permanently.
217 *
218 * Further, check if the inode is being opened, written and
219 * closed frequently and we have delayed allocation blocks
220 * outstanding (e.g. streaming writes from the NFS server),
221 * truncating the blocks past EOF will cause fragmentation to
222 * occur.
223 *
224 * In this case don't do the truncation, either, but we have to
225 * be careful how we detect this case. Blocks beyond EOF show
226 * up as i_delayed_blks even when the inode is clean, so we
227 * need to truncate them away first before checking for a dirty
228 * release. Hence on the first dirty close we will still remove
229 * the speculative allocation, but after that we will leave it
230 * in place.
231 */
232 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
233 return 0;
234
235 error = xfs_free_eofblocks(mp, ip, true);
236 if (error && error != EAGAIN)
237 return error;
238
239 /* delalloc blocks after truncation means it really is dirty */
240 if (ip->i_delayed_blks)
241 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
242 }
243 return 0;
244 }
245
246 /*
247 * xfs_inactive
248 *
249 * This is called when the vnode reference count for the vnode
250 * goes to zero. If the file has been unlinked, then it must
251 * now be truncated. Also, we clear all of the read-ahead state
252 * kept for the inode here since the file is now closed.
253 */
254 int
255 xfs_inactive(
256 xfs_inode_t *ip)
257 {
258 xfs_bmap_free_t free_list;
259 xfs_fsblock_t first_block;
260 int committed;
261 xfs_trans_t *tp;
262 xfs_mount_t *mp;
263 int error;
264 int truncate = 0;
265
266 /*
267 * If the inode is already free, then there can be nothing
268 * to clean up here.
269 */
270 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
271 ASSERT(ip->i_df.if_real_bytes == 0);
272 ASSERT(ip->i_df.if_broot_bytes == 0);
273 return VN_INACTIVE_CACHE;
274 }
275
276 mp = ip->i_mount;
277
278 error = 0;
279
280 /* If this is a read-only mount, don't do this (would generate I/O) */
281 if (mp->m_flags & XFS_MOUNT_RDONLY)
282 goto out;
283
284 if (ip->i_d.di_nlink != 0) {
285 /*
286 * force is true because we are evicting an inode from the
287 * cache. Post-eof blocks must be freed, lest we end up with
288 * broken free space accounting.
289 */
290 if (xfs_can_free_eofblocks(ip, true)) {
291 error = xfs_free_eofblocks(mp, ip, false);
292 if (error)
293 return VN_INACTIVE_CACHE;
294 }
295 goto out;
296 }
297
298 if (S_ISREG(ip->i_d.di_mode) &&
299 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
300 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
301 truncate = 1;
302
303 error = xfs_qm_dqattach(ip, 0);
304 if (error)
305 return VN_INACTIVE_CACHE;
306
307 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
308 error = xfs_trans_reserve(tp, 0,
309 (truncate || S_ISLNK(ip->i_d.di_mode)) ?
310 XFS_ITRUNCATE_LOG_RES(mp) :
311 XFS_IFREE_LOG_RES(mp),
312 0,
313 XFS_TRANS_PERM_LOG_RES,
314 XFS_ITRUNCATE_LOG_COUNT);
315 if (error) {
316 ASSERT(XFS_FORCED_SHUTDOWN(mp));
317 xfs_trans_cancel(tp, 0);
318 return VN_INACTIVE_CACHE;
319 }
320
321 xfs_ilock(ip, XFS_ILOCK_EXCL);
322 xfs_trans_ijoin(tp, ip, 0);
323
324 if (S_ISLNK(ip->i_d.di_mode)) {
325 error = xfs_inactive_symlink(ip, &tp);
326 if (error)
327 goto out_cancel;
328 } else if (truncate) {
329 ip->i_d.di_size = 0;
330 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
331
332 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
333 if (error)
334 goto out_cancel;
335
336 ASSERT(ip->i_d.di_nextents == 0);
337 }
338
339 /*
340 * If there are attributes associated with the file then blow them away
341 * now. The code calls a routine that recursively deconstructs the
342 * attribute fork. We need to just commit the current transaction
343 * because we can't use it for xfs_attr_inactive().
344 */
345 if (ip->i_d.di_anextents > 0) {
346 ASSERT(ip->i_d.di_forkoff != 0);
347
348 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
349 if (error)
350 goto out_unlock;
351
352 xfs_iunlock(ip, XFS_ILOCK_EXCL);
353
354 error = xfs_attr_inactive(ip);
355 if (error)
356 goto out;
357
358 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
359 error = xfs_trans_reserve(tp, 0,
360 XFS_IFREE_LOG_RES(mp),
361 0, XFS_TRANS_PERM_LOG_RES,
362 XFS_INACTIVE_LOG_COUNT);
363 if (error) {
364 xfs_trans_cancel(tp, 0);
365 goto out;
366 }
367
368 xfs_ilock(ip, XFS_ILOCK_EXCL);
369 xfs_trans_ijoin(tp, ip, 0);
370 }
371
372 if (ip->i_afp)
373 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
374
375 ASSERT(ip->i_d.di_anextents == 0);
376
377 /*
378 * Free the inode.
379 */
380 xfs_bmap_init(&free_list, &first_block);
381 error = xfs_ifree(tp, ip, &free_list);
382 if (error) {
383 /*
384 * If we fail to free the inode, shut down. The cancel
385 * might do that, we need to make sure. Otherwise the
386 * inode might be lost for a long time or forever.
387 */
388 if (!XFS_FORCED_SHUTDOWN(mp)) {
389 xfs_notice(mp, "%s: xfs_ifree returned error %d",
390 __func__, error);
391 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
392 }
393 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
394 } else {
395 /*
396 * Credit the quota account(s). The inode is gone.
397 */
398 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
399
400 /*
401 * Just ignore errors at this point. There is nothing we can
402 * do except to try to keep going. Make sure it's not a silent
403 * error.
404 */
405 error = xfs_bmap_finish(&tp, &free_list, &committed);
406 if (error)
407 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
408 __func__, error);
409 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
410 if (error)
411 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
412 __func__, error);
413 }
414
415 /*
416 * Release the dquots held by inode, if any.
417 */
418 xfs_qm_dqdetach(ip);
419 out_unlock:
420 xfs_iunlock(ip, XFS_ILOCK_EXCL);
421 out:
422 return VN_INACTIVE_CACHE;
423 out_cancel:
424 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
425 goto out_unlock;
426 }
427
428 /*
429 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
430 * is allowed, otherwise it has to be an exact match. If a CI match is found,
431 * ci_name->name will point to a the actual name (caller must free) or
432 * will be set to NULL if an exact match is found.
433 */
434 int
435 xfs_lookup(
436 xfs_inode_t *dp,
437 struct xfs_name *name,
438 xfs_inode_t **ipp,
439 struct xfs_name *ci_name)
440 {
441 xfs_ino_t inum;
442 int error;
443 uint lock_mode;
444
445 trace_xfs_lookup(dp, name);
446
447 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
448 return XFS_ERROR(EIO);
449
450 lock_mode = xfs_ilock_map_shared(dp);
451 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
452 xfs_iunlock_map_shared(dp, lock_mode);
453
454 if (error)
455 goto out;
456
457 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
458 if (error)
459 goto out_free_name;
460
461 return 0;
462
463 out_free_name:
464 if (ci_name)
465 kmem_free(ci_name->name);
466 out:
467 *ipp = NULL;
468 return error;
469 }
470
471 int
472 xfs_create(
473 xfs_inode_t *dp,
474 struct xfs_name *name,
475 umode_t mode,
476 xfs_dev_t rdev,
477 xfs_inode_t **ipp)
478 {
479 int is_dir = S_ISDIR(mode);
480 struct xfs_mount *mp = dp->i_mount;
481 struct xfs_inode *ip = NULL;
482 struct xfs_trans *tp = NULL;
483 int error;
484 xfs_bmap_free_t free_list;
485 xfs_fsblock_t first_block;
486 bool unlock_dp_on_error = false;
487 uint cancel_flags;
488 int committed;
489 prid_t prid;
490 struct xfs_dquot *udqp = NULL;
491 struct xfs_dquot *gdqp = NULL;
492 uint resblks;
493 uint log_res;
494 uint log_count;
495
496 trace_xfs_create(dp, name);
497
498 if (XFS_FORCED_SHUTDOWN(mp))
499 return XFS_ERROR(EIO);
500
501 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
502 prid = xfs_get_projid(dp);
503 else
504 prid = XFS_PROJID_DEFAULT;
505
506 /*
507 * Make sure that we have allocated dquot(s) on disk.
508 */
509 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
510 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
511 if (error)
512 return error;
513
514 if (is_dir) {
515 rdev = 0;
516 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
517 log_res = XFS_MKDIR_LOG_RES(mp);
518 log_count = XFS_MKDIR_LOG_COUNT;
519 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
520 } else {
521 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
522 log_res = XFS_CREATE_LOG_RES(mp);
523 log_count = XFS_CREATE_LOG_COUNT;
524 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
525 }
526
527 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
528
529 /*
530 * Initially assume that the file does not exist and
531 * reserve the resources for that case. If that is not
532 * the case we'll drop the one we have and get a more
533 * appropriate transaction later.
534 */
535 error = xfs_trans_reserve(tp, resblks, log_res, 0,
536 XFS_TRANS_PERM_LOG_RES, log_count);
537 if (error == ENOSPC) {
538 /* flush outstanding delalloc blocks and retry */
539 xfs_flush_inodes(mp);
540 error = xfs_trans_reserve(tp, resblks, log_res, 0,
541 XFS_TRANS_PERM_LOG_RES, log_count);
542 }
543 if (error == ENOSPC) {
544 /* No space at all so try a "no-allocation" reservation */
545 resblks = 0;
546 error = xfs_trans_reserve(tp, 0, log_res, 0,
547 XFS_TRANS_PERM_LOG_RES, log_count);
548 }
549 if (error) {
550 cancel_flags = 0;
551 goto out_trans_cancel;
552 }
553
554 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
555 unlock_dp_on_error = true;
556
557 xfs_bmap_init(&free_list, &first_block);
558
559 /*
560 * Reserve disk quota and the inode.
561 */
562 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
563 if (error)
564 goto out_trans_cancel;
565
566 error = xfs_dir_canenter(tp, dp, name, resblks);
567 if (error)
568 goto out_trans_cancel;
569
570 /*
571 * A newly created regular or special file just has one directory
572 * entry pointing to them, but a directory also the "." entry
573 * pointing to itself.
574 */
575 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
576 prid, resblks > 0, &ip, &committed);
577 if (error) {
578 if (error == ENOSPC)
579 goto out_trans_cancel;
580 goto out_trans_abort;
581 }
582
583 /*
584 * Now we join the directory inode to the transaction. We do not do it
585 * earlier because xfs_dir_ialloc might commit the previous transaction
586 * (and release all the locks). An error from here on will result in
587 * the transaction cancel unlocking dp so don't do it explicitly in the
588 * error path.
589 */
590 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
591 unlock_dp_on_error = false;
592
593 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
594 &first_block, &free_list, resblks ?
595 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
596 if (error) {
597 ASSERT(error != ENOSPC);
598 goto out_trans_abort;
599 }
600 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
601 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
602
603 if (is_dir) {
604 error = xfs_dir_init(tp, ip, dp);
605 if (error)
606 goto out_bmap_cancel;
607
608 error = xfs_bumplink(tp, dp);
609 if (error)
610 goto out_bmap_cancel;
611 }
612
613 /*
614 * If this is a synchronous mount, make sure that the
615 * create transaction goes to disk before returning to
616 * the user.
617 */
618 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
619 xfs_trans_set_sync(tp);
620
621 /*
622 * Attach the dquot(s) to the inodes and modify them incore.
623 * These ids of the inode couldn't have changed since the new
624 * inode has been locked ever since it was created.
625 */
626 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
627
628 error = xfs_bmap_finish(&tp, &free_list, &committed);
629 if (error)
630 goto out_bmap_cancel;
631
632 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
633 if (error)
634 goto out_release_inode;
635
636 xfs_qm_dqrele(udqp);
637 xfs_qm_dqrele(gdqp);
638
639 *ipp = ip;
640 return 0;
641
642 out_bmap_cancel:
643 xfs_bmap_cancel(&free_list);
644 out_trans_abort:
645 cancel_flags |= XFS_TRANS_ABORT;
646 out_trans_cancel:
647 xfs_trans_cancel(tp, cancel_flags);
648 out_release_inode:
649 /*
650 * Wait until after the current transaction is aborted to
651 * release the inode. This prevents recursive transactions
652 * and deadlocks from xfs_inactive.
653 */
654 if (ip)
655 IRELE(ip);
656
657 xfs_qm_dqrele(udqp);
658 xfs_qm_dqrele(gdqp);
659
660 if (unlock_dp_on_error)
661 xfs_iunlock(dp, XFS_ILOCK_EXCL);
662 return error;
663 }
664
665 #ifdef DEBUG
666 int xfs_locked_n;
667 int xfs_small_retries;
668 int xfs_middle_retries;
669 int xfs_lots_retries;
670 int xfs_lock_delays;
671 #endif
672
673 /*
674 * Bump the subclass so xfs_lock_inodes() acquires each lock with
675 * a different value
676 */
677 static inline int
678 xfs_lock_inumorder(int lock_mode, int subclass)
679 {
680 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
681 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
682 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
683 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
684
685 return lock_mode;
686 }
687
688 /*
689 * The following routine will lock n inodes in exclusive mode.
690 * We assume the caller calls us with the inodes in i_ino order.
691 *
692 * We need to detect deadlock where an inode that we lock
693 * is in the AIL and we start waiting for another inode that is locked
694 * by a thread in a long running transaction (such as truncate). This can
695 * result in deadlock since the long running trans might need to wait
696 * for the inode we just locked in order to push the tail and free space
697 * in the log.
698 */
699 void
700 xfs_lock_inodes(
701 xfs_inode_t **ips,
702 int inodes,
703 uint lock_mode)
704 {
705 int attempts = 0, i, j, try_lock;
706 xfs_log_item_t *lp;
707
708 ASSERT(ips && (inodes >= 2)); /* we need at least two */
709
710 try_lock = 0;
711 i = 0;
712
713 again:
714 for (; i < inodes; i++) {
715 ASSERT(ips[i]);
716
717 if (i && (ips[i] == ips[i-1])) /* Already locked */
718 continue;
719
720 /*
721 * If try_lock is not set yet, make sure all locked inodes
722 * are not in the AIL.
723 * If any are, set try_lock to be used later.
724 */
725
726 if (!try_lock) {
727 for (j = (i - 1); j >= 0 && !try_lock; j--) {
728 lp = (xfs_log_item_t *)ips[j]->i_itemp;
729 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
730 try_lock++;
731 }
732 }
733 }
734
735 /*
736 * If any of the previous locks we have locked is in the AIL,
737 * we must TRY to get the second and subsequent locks. If
738 * we can't get any, we must release all we have
739 * and try again.
740 */
741
742 if (try_lock) {
743 /* try_lock must be 0 if i is 0. */
744 /*
745 * try_lock means we have an inode locked
746 * that is in the AIL.
747 */
748 ASSERT(i != 0);
749 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
750 attempts++;
751
752 /*
753 * Unlock all previous guys and try again.
754 * xfs_iunlock will try to push the tail
755 * if the inode is in the AIL.
756 */
757
758 for(j = i - 1; j >= 0; j--) {
759
760 /*
761 * Check to see if we've already
762 * unlocked this one.
763 * Not the first one going back,
764 * and the inode ptr is the same.
765 */
766 if ((j != (i - 1)) && ips[j] ==
767 ips[j+1])
768 continue;
769
770 xfs_iunlock(ips[j], lock_mode);
771 }
772
773 if ((attempts % 5) == 0) {
774 delay(1); /* Don't just spin the CPU */
775 #ifdef DEBUG
776 xfs_lock_delays++;
777 #endif
778 }
779 i = 0;
780 try_lock = 0;
781 goto again;
782 }
783 } else {
784 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
785 }
786 }
787
788 #ifdef DEBUG
789 if (attempts) {
790 if (attempts < 5) xfs_small_retries++;
791 else if (attempts < 100) xfs_middle_retries++;
792 else xfs_lots_retries++;
793 } else {
794 xfs_locked_n++;
795 }
796 #endif
797 }
798
799 /*
800 * xfs_lock_two_inodes() can only be used to lock one type of lock
801 * at a time - the iolock or the ilock, but not both at once. If
802 * we lock both at once, lockdep will report false positives saying
803 * we have violated locking orders.
804 */
805 void
806 xfs_lock_two_inodes(
807 xfs_inode_t *ip0,
808 xfs_inode_t *ip1,
809 uint lock_mode)
810 {
811 xfs_inode_t *temp;
812 int attempts = 0;
813 xfs_log_item_t *lp;
814
815 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
816 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
817 ASSERT(ip0->i_ino != ip1->i_ino);
818
819 if (ip0->i_ino > ip1->i_ino) {
820 temp = ip0;
821 ip0 = ip1;
822 ip1 = temp;
823 }
824
825 again:
826 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
827
828 /*
829 * If the first lock we have locked is in the AIL, we must TRY to get
830 * the second lock. If we can't get it, we must release the first one
831 * and try again.
832 */
833 lp = (xfs_log_item_t *)ip0->i_itemp;
834 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
835 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
836 xfs_iunlock(ip0, lock_mode);
837 if ((++attempts % 5) == 0)
838 delay(1); /* Don't just spin the CPU */
839 goto again;
840 }
841 } else {
842 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
843 }
844 }
845
846 int
847 xfs_remove(
848 xfs_inode_t *dp,
849 struct xfs_name *name,
850 xfs_inode_t *ip)
851 {
852 xfs_mount_t *mp = dp->i_mount;
853 xfs_trans_t *tp = NULL;
854 int is_dir = S_ISDIR(ip->i_d.di_mode);
855 int error = 0;
856 xfs_bmap_free_t free_list;
857 xfs_fsblock_t first_block;
858 int cancel_flags;
859 int committed;
860 int link_zero;
861 uint resblks;
862 uint log_count;
863
864 trace_xfs_remove(dp, name);
865
866 if (XFS_FORCED_SHUTDOWN(mp))
867 return XFS_ERROR(EIO);
868
869 error = xfs_qm_dqattach(dp, 0);
870 if (error)
871 goto std_return;
872
873 error = xfs_qm_dqattach(ip, 0);
874 if (error)
875 goto std_return;
876
877 if (is_dir) {
878 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
879 log_count = XFS_DEFAULT_LOG_COUNT;
880 } else {
881 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
882 log_count = XFS_REMOVE_LOG_COUNT;
883 }
884 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
885
886 /*
887 * We try to get the real space reservation first,
888 * allowing for directory btree deletion(s) implying
889 * possible bmap insert(s). If we can't get the space
890 * reservation then we use 0 instead, and avoid the bmap
891 * btree insert(s) in the directory code by, if the bmap
892 * insert tries to happen, instead trimming the LAST
893 * block from the directory.
894 */
895 resblks = XFS_REMOVE_SPACE_RES(mp);
896 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
897 XFS_TRANS_PERM_LOG_RES, log_count);
898 if (error == ENOSPC) {
899 resblks = 0;
900 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
901 XFS_TRANS_PERM_LOG_RES, log_count);
902 }
903 if (error) {
904 ASSERT(error != ENOSPC);
905 cancel_flags = 0;
906 goto out_trans_cancel;
907 }
908
909 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
910
911 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
912 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
913
914 /*
915 * If we're removing a directory perform some additional validation.
916 */
917 if (is_dir) {
918 ASSERT(ip->i_d.di_nlink >= 2);
919 if (ip->i_d.di_nlink != 2) {
920 error = XFS_ERROR(ENOTEMPTY);
921 goto out_trans_cancel;
922 }
923 if (!xfs_dir_isempty(ip)) {
924 error = XFS_ERROR(ENOTEMPTY);
925 goto out_trans_cancel;
926 }
927 }
928
929 xfs_bmap_init(&free_list, &first_block);
930 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
931 &first_block, &free_list, resblks);
932 if (error) {
933 ASSERT(error != ENOENT);
934 goto out_bmap_cancel;
935 }
936 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
937
938 if (is_dir) {
939 /*
940 * Drop the link from ip's "..".
941 */
942 error = xfs_droplink(tp, dp);
943 if (error)
944 goto out_bmap_cancel;
945
946 /*
947 * Drop the "." link from ip to self.
948 */
949 error = xfs_droplink(tp, ip);
950 if (error)
951 goto out_bmap_cancel;
952 } else {
953 /*
954 * When removing a non-directory we need to log the parent
955 * inode here. For a directory this is done implicitly
956 * by the xfs_droplink call for the ".." entry.
957 */
958 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
959 }
960
961 /*
962 * Drop the link from dp to ip.
963 */
964 error = xfs_droplink(tp, ip);
965 if (error)
966 goto out_bmap_cancel;
967
968 /*
969 * Determine if this is the last link while
970 * we are in the transaction.
971 */
972 link_zero = (ip->i_d.di_nlink == 0);
973
974 /*
975 * If this is a synchronous mount, make sure that the
976 * remove transaction goes to disk before returning to
977 * the user.
978 */
979 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
980 xfs_trans_set_sync(tp);
981
982 error = xfs_bmap_finish(&tp, &free_list, &committed);
983 if (error)
984 goto out_bmap_cancel;
985
986 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
987 if (error)
988 goto std_return;
989
990 /*
991 * If we are using filestreams, kill the stream association.
992 * If the file is still open it may get a new one but that
993 * will get killed on last close in xfs_close() so we don't
994 * have to worry about that.
995 */
996 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
997 xfs_filestream_deassociate(ip);
998
999 return 0;
1000
1001 out_bmap_cancel:
1002 xfs_bmap_cancel(&free_list);
1003 cancel_flags |= XFS_TRANS_ABORT;
1004 out_trans_cancel:
1005 xfs_trans_cancel(tp, cancel_flags);
1006 std_return:
1007 return error;
1008 }
1009
1010 int
1011 xfs_link(
1012 xfs_inode_t *tdp,
1013 xfs_inode_t *sip,
1014 struct xfs_name *target_name)
1015 {
1016 xfs_mount_t *mp = tdp->i_mount;
1017 xfs_trans_t *tp;
1018 int error;
1019 xfs_bmap_free_t free_list;
1020 xfs_fsblock_t first_block;
1021 int cancel_flags;
1022 int committed;
1023 int resblks;
1024
1025 trace_xfs_link(tdp, target_name);
1026
1027 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1028
1029 if (XFS_FORCED_SHUTDOWN(mp))
1030 return XFS_ERROR(EIO);
1031
1032 error = xfs_qm_dqattach(sip, 0);
1033 if (error)
1034 goto std_return;
1035
1036 error = xfs_qm_dqattach(tdp, 0);
1037 if (error)
1038 goto std_return;
1039
1040 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1041 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1042 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1043 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1044 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1045 if (error == ENOSPC) {
1046 resblks = 0;
1047 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1048 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1049 }
1050 if (error) {
1051 cancel_flags = 0;
1052 goto error_return;
1053 }
1054
1055 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1056
1057 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1058 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1059
1060 /*
1061 * If we are using project inheritance, we only allow hard link
1062 * creation in our tree when the project IDs are the same; else
1063 * the tree quota mechanism could be circumvented.
1064 */
1065 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1066 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1067 error = XFS_ERROR(EXDEV);
1068 goto error_return;
1069 }
1070
1071 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1072 if (error)
1073 goto error_return;
1074
1075 xfs_bmap_init(&free_list, &first_block);
1076
1077 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1078 &first_block, &free_list, resblks);
1079 if (error)
1080 goto abort_return;
1081 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1082 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1083
1084 error = xfs_bumplink(tp, sip);
1085 if (error)
1086 goto abort_return;
1087
1088 /*
1089 * If this is a synchronous mount, make sure that the
1090 * link transaction goes to disk before returning to
1091 * the user.
1092 */
1093 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1094 xfs_trans_set_sync(tp);
1095 }
1096
1097 error = xfs_bmap_finish (&tp, &free_list, &committed);
1098 if (error) {
1099 xfs_bmap_cancel(&free_list);
1100 goto abort_return;
1101 }
1102
1103 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1104
1105 abort_return:
1106 cancel_flags |= XFS_TRANS_ABORT;
1107 error_return:
1108 xfs_trans_cancel(tp, cancel_flags);
1109 std_return:
1110 return error;
1111 }
1112
1113 int
1114 xfs_set_dmattrs(
1115 xfs_inode_t *ip,
1116 u_int evmask,
1117 u_int16_t state)
1118 {
1119 xfs_mount_t *mp = ip->i_mount;
1120 xfs_trans_t *tp;
1121 int error;
1122
1123 if (!capable(CAP_SYS_ADMIN))
1124 return XFS_ERROR(EPERM);
1125
1126 if (XFS_FORCED_SHUTDOWN(mp))
1127 return XFS_ERROR(EIO);
1128
1129 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1130 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1131 if (error) {
1132 xfs_trans_cancel(tp, 0);
1133 return error;
1134 }
1135 xfs_ilock(ip, XFS_ILOCK_EXCL);
1136 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1137
1138 ip->i_d.di_dmevmask = evmask;
1139 ip->i_d.di_dmstate = state;
1140
1141 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1142 error = xfs_trans_commit(tp, 0);
1143
1144 return error;
1145 }
1146
1147 /*
1148 * xfs_alloc_file_space()
1149 * This routine allocates disk space for the given file.
1150 *
1151 * If alloc_type == 0, this request is for an ALLOCSP type
1152 * request which will change the file size. In this case, no
1153 * DMAPI event will be generated by the call. A TRUNCATE event
1154 * will be generated later by xfs_setattr.
1155 *
1156 * If alloc_type != 0, this request is for a RESVSP type
1157 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1158 * lower block boundary byte address is less than the file's
1159 * length.
1160 *
1161 * RETURNS:
1162 * 0 on success
1163 * errno on error
1164 *
1165 */
1166 STATIC int
1167 xfs_alloc_file_space(
1168 xfs_inode_t *ip,
1169 xfs_off_t offset,
1170 xfs_off_t len,
1171 int alloc_type,
1172 int attr_flags)
1173 {
1174 xfs_mount_t *mp = ip->i_mount;
1175 xfs_off_t count;
1176 xfs_filblks_t allocated_fsb;
1177 xfs_filblks_t allocatesize_fsb;
1178 xfs_extlen_t extsz, temp;
1179 xfs_fileoff_t startoffset_fsb;
1180 xfs_fsblock_t firstfsb;
1181 int nimaps;
1182 int quota_flag;
1183 int rt;
1184 xfs_trans_t *tp;
1185 xfs_bmbt_irec_t imaps[1], *imapp;
1186 xfs_bmap_free_t free_list;
1187 uint qblocks, resblks, resrtextents;
1188 int committed;
1189 int error;
1190
1191 trace_xfs_alloc_file_space(ip);
1192
1193 if (XFS_FORCED_SHUTDOWN(mp))
1194 return XFS_ERROR(EIO);
1195
1196 error = xfs_qm_dqattach(ip, 0);
1197 if (error)
1198 return error;
1199
1200 if (len <= 0)
1201 return XFS_ERROR(EINVAL);
1202
1203 rt = XFS_IS_REALTIME_INODE(ip);
1204 extsz = xfs_get_extsz_hint(ip);
1205
1206 count = len;
1207 imapp = &imaps[0];
1208 nimaps = 1;
1209 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1210 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1211
1212 /*
1213 * Allocate file space until done or until there is an error
1214 */
1215 while (allocatesize_fsb && !error) {
1216 xfs_fileoff_t s, e;
1217
1218 /*
1219 * Determine space reservations for data/realtime.
1220 */
1221 if (unlikely(extsz)) {
1222 s = startoffset_fsb;
1223 do_div(s, extsz);
1224 s *= extsz;
1225 e = startoffset_fsb + allocatesize_fsb;
1226 if ((temp = do_mod(startoffset_fsb, extsz)))
1227 e += temp;
1228 if ((temp = do_mod(e, extsz)))
1229 e += extsz - temp;
1230 } else {
1231 s = 0;
1232 e = allocatesize_fsb;
1233 }
1234
1235 /*
1236 * The transaction reservation is limited to a 32-bit block
1237 * count, hence we need to limit the number of blocks we are
1238 * trying to reserve to avoid an overflow. We can't allocate
1239 * more than @nimaps extents, and an extent is limited on disk
1240 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1241 */
1242 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1243 if (unlikely(rt)) {
1244 resrtextents = qblocks = resblks;
1245 resrtextents /= mp->m_sb.sb_rextsize;
1246 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1247 quota_flag = XFS_QMOPT_RES_RTBLKS;
1248 } else {
1249 resrtextents = 0;
1250 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1251 quota_flag = XFS_QMOPT_RES_REGBLKS;
1252 }
1253
1254 /*
1255 * Allocate and setup the transaction.
1256 */
1257 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1258 error = xfs_trans_reserve(tp, resblks,
1259 XFS_WRITE_LOG_RES(mp), resrtextents,
1260 XFS_TRANS_PERM_LOG_RES,
1261 XFS_WRITE_LOG_COUNT);
1262 /*
1263 * Check for running out of space
1264 */
1265 if (error) {
1266 /*
1267 * Free the transaction structure.
1268 */
1269 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1270 xfs_trans_cancel(tp, 0);
1271 break;
1272 }
1273 xfs_ilock(ip, XFS_ILOCK_EXCL);
1274 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1275 0, quota_flag);
1276 if (error)
1277 goto error1;
1278
1279 xfs_trans_ijoin(tp, ip, 0);
1280
1281 xfs_bmap_init(&free_list, &firstfsb);
1282 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1283 allocatesize_fsb, alloc_type, &firstfsb,
1284 0, imapp, &nimaps, &free_list);
1285 if (error) {
1286 goto error0;
1287 }
1288
1289 /*
1290 * Complete the transaction
1291 */
1292 error = xfs_bmap_finish(&tp, &free_list, &committed);
1293 if (error) {
1294 goto error0;
1295 }
1296
1297 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1298 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1299 if (error) {
1300 break;
1301 }
1302
1303 allocated_fsb = imapp->br_blockcount;
1304
1305 if (nimaps == 0) {
1306 error = XFS_ERROR(ENOSPC);
1307 break;
1308 }
1309
1310 startoffset_fsb += allocated_fsb;
1311 allocatesize_fsb -= allocated_fsb;
1312 }
1313
1314 return error;
1315
1316 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1317 xfs_bmap_cancel(&free_list);
1318 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1319
1320 error1: /* Just cancel transaction */
1321 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1322 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1323 return error;
1324 }
1325
1326 /*
1327 * Zero file bytes between startoff and endoff inclusive.
1328 * The iolock is held exclusive and no blocks are buffered.
1329 *
1330 * This function is used by xfs_free_file_space() to zero
1331 * partial blocks when the range to free is not block aligned.
1332 * When unreserving space with boundaries that are not block
1333 * aligned we round up the start and round down the end
1334 * boundaries and then use this function to zero the parts of
1335 * the blocks that got dropped during the rounding.
1336 */
1337 STATIC int
1338 xfs_zero_remaining_bytes(
1339 xfs_inode_t *ip,
1340 xfs_off_t startoff,
1341 xfs_off_t endoff)
1342 {
1343 xfs_bmbt_irec_t imap;
1344 xfs_fileoff_t offset_fsb;
1345 xfs_off_t lastoffset;
1346 xfs_off_t offset;
1347 xfs_buf_t *bp;
1348 xfs_mount_t *mp = ip->i_mount;
1349 int nimap;
1350 int error = 0;
1351
1352 /*
1353 * Avoid doing I/O beyond eof - it's not necessary
1354 * since nothing can read beyond eof. The space will
1355 * be zeroed when the file is extended anyway.
1356 */
1357 if (startoff >= XFS_ISIZE(ip))
1358 return 0;
1359
1360 if (endoff > XFS_ISIZE(ip))
1361 endoff = XFS_ISIZE(ip);
1362
1363 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1364 mp->m_rtdev_targp : mp->m_ddev_targp,
1365 BTOBB(mp->m_sb.sb_blocksize), 0);
1366 if (!bp)
1367 return XFS_ERROR(ENOMEM);
1368
1369 xfs_buf_unlock(bp);
1370
1371 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1372 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1373 nimap = 1;
1374 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1375 if (error || nimap < 1)
1376 break;
1377 ASSERT(imap.br_blockcount >= 1);
1378 ASSERT(imap.br_startoff == offset_fsb);
1379 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1380 if (lastoffset > endoff)
1381 lastoffset = endoff;
1382 if (imap.br_startblock == HOLESTARTBLOCK)
1383 continue;
1384 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1385 if (imap.br_state == XFS_EXT_UNWRITTEN)
1386 continue;
1387 XFS_BUF_UNDONE(bp);
1388 XFS_BUF_UNWRITE(bp);
1389 XFS_BUF_READ(bp);
1390 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1391 xfsbdstrat(mp, bp);
1392 error = xfs_buf_iowait(bp);
1393 if (error) {
1394 xfs_buf_ioerror_alert(bp,
1395 "xfs_zero_remaining_bytes(read)");
1396 break;
1397 }
1398 memset(bp->b_addr +
1399 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1400 0, lastoffset - offset + 1);
1401 XFS_BUF_UNDONE(bp);
1402 XFS_BUF_UNREAD(bp);
1403 XFS_BUF_WRITE(bp);
1404 xfsbdstrat(mp, bp);
1405 error = xfs_buf_iowait(bp);
1406 if (error) {
1407 xfs_buf_ioerror_alert(bp,
1408 "xfs_zero_remaining_bytes(write)");
1409 break;
1410 }
1411 }
1412 xfs_buf_free(bp);
1413 return error;
1414 }
1415
1416 /*
1417 * xfs_free_file_space()
1418 * This routine frees disk space for the given file.
1419 *
1420 * This routine is only called by xfs_change_file_space
1421 * for an UNRESVSP type call.
1422 *
1423 * RETURNS:
1424 * 0 on success
1425 * errno on error
1426 *
1427 */
1428 STATIC int
1429 xfs_free_file_space(
1430 xfs_inode_t *ip,
1431 xfs_off_t offset,
1432 xfs_off_t len,
1433 int attr_flags)
1434 {
1435 int committed;
1436 int done;
1437 xfs_fileoff_t endoffset_fsb;
1438 int error;
1439 xfs_fsblock_t firstfsb;
1440 xfs_bmap_free_t free_list;
1441 xfs_bmbt_irec_t imap;
1442 xfs_off_t ioffset;
1443 xfs_extlen_t mod=0;
1444 xfs_mount_t *mp;
1445 int nimap;
1446 uint resblks;
1447 xfs_off_t rounding;
1448 int rt;
1449 xfs_fileoff_t startoffset_fsb;
1450 xfs_trans_t *tp;
1451 int need_iolock = 1;
1452
1453 mp = ip->i_mount;
1454
1455 trace_xfs_free_file_space(ip);
1456
1457 error = xfs_qm_dqattach(ip, 0);
1458 if (error)
1459 return error;
1460
1461 error = 0;
1462 if (len <= 0) /* if nothing being freed */
1463 return error;
1464 rt = XFS_IS_REALTIME_INODE(ip);
1465 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1466 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1467
1468 if (attr_flags & XFS_ATTR_NOLOCK)
1469 need_iolock = 0;
1470 if (need_iolock) {
1471 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1472 /* wait for the completion of any pending DIOs */
1473 inode_dio_wait(VFS_I(ip));
1474 }
1475
1476 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1477 ioffset = offset & ~(rounding - 1);
1478 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1479 ioffset, -1);
1480 if (error)
1481 goto out_unlock_iolock;
1482 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1483
1484 /*
1485 * Need to zero the stuff we're not freeing, on disk.
1486 * If it's a realtime file & can't use unwritten extents then we
1487 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1488 * will take care of it for us.
1489 */
1490 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1491 nimap = 1;
1492 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1493 &imap, &nimap, 0);
1494 if (error)
1495 goto out_unlock_iolock;
1496 ASSERT(nimap == 0 || nimap == 1);
1497 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1498 xfs_daddr_t block;
1499
1500 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1501 block = imap.br_startblock;
1502 mod = do_div(block, mp->m_sb.sb_rextsize);
1503 if (mod)
1504 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1505 }
1506 nimap = 1;
1507 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1508 &imap, &nimap, 0);
1509 if (error)
1510 goto out_unlock_iolock;
1511 ASSERT(nimap == 0 || nimap == 1);
1512 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1513 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1514 mod++;
1515 if (mod && (mod != mp->m_sb.sb_rextsize))
1516 endoffset_fsb -= mod;
1517 }
1518 }
1519 if ((done = (endoffset_fsb <= startoffset_fsb)))
1520 /*
1521 * One contiguous piece to clear
1522 */
1523 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1524 else {
1525 /*
1526 * Some full blocks, possibly two pieces to clear
1527 */
1528 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1529 error = xfs_zero_remaining_bytes(ip, offset,
1530 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1531 if (!error &&
1532 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1533 error = xfs_zero_remaining_bytes(ip,
1534 XFS_FSB_TO_B(mp, endoffset_fsb),
1535 offset + len - 1);
1536 }
1537
1538 /*
1539 * free file space until done or until there is an error
1540 */
1541 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1542 while (!error && !done) {
1543
1544 /*
1545 * allocate and setup the transaction. Allow this
1546 * transaction to dip into the reserve blocks to ensure
1547 * the freeing of the space succeeds at ENOSPC.
1548 */
1549 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1550 tp->t_flags |= XFS_TRANS_RESERVE;
1551 error = xfs_trans_reserve(tp,
1552 resblks,
1553 XFS_WRITE_LOG_RES(mp),
1554 0,
1555 XFS_TRANS_PERM_LOG_RES,
1556 XFS_WRITE_LOG_COUNT);
1557
1558 /*
1559 * check for running out of space
1560 */
1561 if (error) {
1562 /*
1563 * Free the transaction structure.
1564 */
1565 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1566 xfs_trans_cancel(tp, 0);
1567 break;
1568 }
1569 xfs_ilock(ip, XFS_ILOCK_EXCL);
1570 error = xfs_trans_reserve_quota(tp, mp,
1571 ip->i_udquot, ip->i_gdquot,
1572 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1573 if (error)
1574 goto error1;
1575
1576 xfs_trans_ijoin(tp, ip, 0);
1577
1578 /*
1579 * issue the bunmapi() call to free the blocks
1580 */
1581 xfs_bmap_init(&free_list, &firstfsb);
1582 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1583 endoffset_fsb - startoffset_fsb,
1584 0, 2, &firstfsb, &free_list, &done);
1585 if (error) {
1586 goto error0;
1587 }
1588
1589 /*
1590 * complete the transaction
1591 */
1592 error = xfs_bmap_finish(&tp, &free_list, &committed);
1593 if (error) {
1594 goto error0;
1595 }
1596
1597 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1598 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1599 }
1600
1601 out_unlock_iolock:
1602 if (need_iolock)
1603 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1604 return error;
1605
1606 error0:
1607 xfs_bmap_cancel(&free_list);
1608 error1:
1609 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1610 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
1611 XFS_ILOCK_EXCL);
1612 return error;
1613 }
1614
1615
1616 STATIC int
1617 xfs_zero_file_space(
1618 struct xfs_inode *ip,
1619 xfs_off_t offset,
1620 xfs_off_t len,
1621 int attr_flags)
1622 {
1623 struct xfs_mount *mp = ip->i_mount;
1624 uint granularity;
1625 xfs_off_t start_boundary;
1626 xfs_off_t end_boundary;
1627 int error;
1628
1629 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1630
1631 /*
1632 * Round the range of extents we are going to convert inwards. If the
1633 * offset is aligned, then it doesn't get changed so we zero from the
1634 * start of the block offset points to.
1635 */
1636 start_boundary = round_up(offset, granularity);
1637 end_boundary = round_down(offset + len, granularity);
1638
1639 ASSERT(start_boundary >= offset);
1640 ASSERT(end_boundary <= offset + len);
1641
1642 if (!(attr_flags & XFS_ATTR_NOLOCK))
1643 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1644
1645 if (start_boundary < end_boundary - 1) {
1646 /* punch out the page cache over the conversion range */
1647 truncate_pagecache_range(VFS_I(ip), start_boundary,
1648 end_boundary - 1);
1649 /* convert the blocks */
1650 error = xfs_alloc_file_space(ip, start_boundary,
1651 end_boundary - start_boundary - 1,
1652 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT,
1653 attr_flags);
1654 if (error)
1655 goto out_unlock;
1656
1657 /* We've handled the interior of the range, now for the edges */
1658 if (start_boundary != offset)
1659 error = xfs_iozero(ip, offset, start_boundary - offset);
1660 if (error)
1661 goto out_unlock;
1662
1663 if (end_boundary != offset + len)
1664 error = xfs_iozero(ip, end_boundary,
1665 offset + len - end_boundary);
1666
1667 } else {
1668 /*
1669 * It's either a sub-granularity range or the range spanned lies
1670 * partially across two adjacent blocks.
1671 */
1672 error = xfs_iozero(ip, offset, len);
1673 }
1674
1675 out_unlock:
1676 if (!(attr_flags & XFS_ATTR_NOLOCK))
1677 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1678 return error;
1679
1680 }
1681
1682 /*
1683 * xfs_change_file_space()
1684 * This routine allocates or frees disk space for the given file.
1685 * The user specified parameters are checked for alignment and size
1686 * limitations.
1687 *
1688 * RETURNS:
1689 * 0 on success
1690 * errno on error
1691 *
1692 */
1693 int
1694 xfs_change_file_space(
1695 xfs_inode_t *ip,
1696 int cmd,
1697 xfs_flock64_t *bf,
1698 xfs_off_t offset,
1699 int attr_flags)
1700 {
1701 xfs_mount_t *mp = ip->i_mount;
1702 int clrprealloc;
1703 int error;
1704 xfs_fsize_t fsize;
1705 int setprealloc;
1706 xfs_off_t startoffset;
1707 xfs_trans_t *tp;
1708 struct iattr iattr;
1709
1710 if (!S_ISREG(ip->i_d.di_mode))
1711 return XFS_ERROR(EINVAL);
1712
1713 switch (bf->l_whence) {
1714 case 0: /*SEEK_SET*/
1715 break;
1716 case 1: /*SEEK_CUR*/
1717 bf->l_start += offset;
1718 break;
1719 case 2: /*SEEK_END*/
1720 bf->l_start += XFS_ISIZE(ip);
1721 break;
1722 default:
1723 return XFS_ERROR(EINVAL);
1724 }
1725
1726 /*
1727 * length of <= 0 for resv/unresv/zero is invalid. length for
1728 * alloc/free is ignored completely and we have no idea what userspace
1729 * might have set it to, so set it to zero to allow range
1730 * checks to pass.
1731 */
1732 switch (cmd) {
1733 case XFS_IOC_ZERO_RANGE:
1734 case XFS_IOC_RESVSP:
1735 case XFS_IOC_RESVSP64:
1736 case XFS_IOC_UNRESVSP:
1737 case XFS_IOC_UNRESVSP64:
1738 if (bf->l_len <= 0)
1739 return XFS_ERROR(EINVAL);
1740 break;
1741 default:
1742 bf->l_len = 0;
1743 break;
1744 }
1745
1746 if (bf->l_start < 0 ||
1747 bf->l_start > mp->m_super->s_maxbytes ||
1748 bf->l_start + bf->l_len < 0 ||
1749 bf->l_start + bf->l_len >= mp->m_super->s_maxbytes)
1750 return XFS_ERROR(EINVAL);
1751
1752 bf->l_whence = 0;
1753
1754 startoffset = bf->l_start;
1755 fsize = XFS_ISIZE(ip);
1756
1757 setprealloc = clrprealloc = 0;
1758 switch (cmd) {
1759 case XFS_IOC_ZERO_RANGE:
1760 error = xfs_zero_file_space(ip, startoffset, bf->l_len,
1761 attr_flags);
1762 if (error)
1763 return error;
1764 setprealloc = 1;
1765 break;
1766
1767 case XFS_IOC_RESVSP:
1768 case XFS_IOC_RESVSP64:
1769 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
1770 XFS_BMAPI_PREALLOC, attr_flags);
1771 if (error)
1772 return error;
1773 setprealloc = 1;
1774 break;
1775
1776 case XFS_IOC_UNRESVSP:
1777 case XFS_IOC_UNRESVSP64:
1778 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
1779 attr_flags)))
1780 return error;
1781 break;
1782
1783 case XFS_IOC_ALLOCSP:
1784 case XFS_IOC_ALLOCSP64:
1785 case XFS_IOC_FREESP:
1786 case XFS_IOC_FREESP64:
1787 /*
1788 * These operations actually do IO when extending the file, but
1789 * the allocation is done seperately to the zeroing that is
1790 * done. This set of operations need to be serialised against
1791 * other IO operations, such as truncate and buffered IO. We
1792 * need to take the IOLOCK here to serialise the allocation and
1793 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
1794 * truncate, direct IO) from racing against the transient
1795 * allocated but not written state we can have here.
1796 */
1797 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1798 if (startoffset > fsize) {
1799 error = xfs_alloc_file_space(ip, fsize,
1800 startoffset - fsize, 0,
1801 attr_flags | XFS_ATTR_NOLOCK);
1802 if (error) {
1803 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1804 break;
1805 }
1806 }
1807
1808 iattr.ia_valid = ATTR_SIZE;
1809 iattr.ia_size = startoffset;
1810
1811 error = xfs_setattr_size(ip, &iattr,
1812 attr_flags | XFS_ATTR_NOLOCK);
1813 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1814
1815 if (error)
1816 return error;
1817
1818 clrprealloc = 1;
1819 break;
1820
1821 default:
1822 ASSERT(0);
1823 return XFS_ERROR(EINVAL);
1824 }
1825
1826 /*
1827 * update the inode timestamp, mode, and prealloc flag bits
1828 */
1829 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
1830
1831 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
1832 0, 0, 0))) {
1833 /* ASSERT(0); */
1834 xfs_trans_cancel(tp, 0);
1835 return error;
1836 }
1837
1838 xfs_ilock(ip, XFS_ILOCK_EXCL);
1839 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1840
1841 if ((attr_flags & XFS_ATTR_DMI) == 0) {
1842 ip->i_d.di_mode &= ~S_ISUID;
1843
1844 /*
1845 * Note that we don't have to worry about mandatory
1846 * file locking being disabled here because we only
1847 * clear the S_ISGID bit if the Group execute bit is
1848 * on, but if it was on then mandatory locking wouldn't
1849 * have been enabled.
1850 */
1851 if (ip->i_d.di_mode & S_IXGRP)
1852 ip->i_d.di_mode &= ~S_ISGID;
1853
1854 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1855 }
1856 if (setprealloc)
1857 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
1858 else if (clrprealloc)
1859 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
1860
1861 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1862 if (attr_flags & XFS_ATTR_SYNC)
1863 xfs_trans_set_sync(tp);
1864 return xfs_trans_commit(tp, 0);
1865 }
Linux kernel - Deliverables
This page took 0.070954 seconds and 5 git commands to generate.