1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
16 #include "xfs_dir2.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_rtalloc.h"
20 #include "xfs_bmap.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_priv.h"
23 #include "xfs_log.h"
24 #include "xfs_log_priv.h"
25 #include "xfs_error.h"
26 #include "xfs_quota.h"
27 #include "xfs_fsops.h"
28 #include "xfs_icache.h"
29 #include "xfs_sysfs.h"
30 #include "xfs_rmap_btree.h"
31 #include "xfs_refcount_btree.h"
32 #include "xfs_reflink.h"
33 #include "xfs_extent_busy.h"
34 #include "xfs_health.h"
35 #include "xfs_trace.h"
36 #include "xfs_ag.h"
37 #include "xfs_rtbitmap.h"
38 #include "scrub/stats.h"
39
40 static DEFINE_MUTEX(xfs_uuid_table_mutex);
41 static int xfs_uuid_table_size;
42 static uuid_t *xfs_uuid_table;
43
44 void
45 xfs_uuid_table_free(void)
46 {
47 if (xfs_uuid_table_size == 0)
48 return;
49 kfree(xfs_uuid_table);
50 xfs_uuid_table = NULL;
51 xfs_uuid_table_size = 0;
52 }
53
54 /*
55 * See if the UUID is unique among mounted XFS filesystems.
56 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
57 */
58 STATIC int
59 xfs_uuid_mount(
60 struct xfs_mount *mp)
61 {
62 uuid_t *uuid = &mp->m_sb.sb_uuid;
63 int hole, i;
64
65 /* Publish UUID in struct super_block */
66 super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
67
68 if (xfs_has_nouuid(mp))
69 return 0;
70
71 if (uuid_is_null(uuid)) {
72 xfs_warn(mp, "Filesystem has null UUID - can't mount");
73 return -EINVAL;
74 }
75
76 mutex_lock(&xfs_uuid_table_mutex);
77 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
78 if (uuid_is_null(&xfs_uuid_table[i])) {
79 hole = i;
80 continue;
81 }
82 if (uuid_equal(uuid, &xfs_uuid_table[i]))
83 goto out_duplicate;
84 }
85
86 if (hole < 0) {
87 xfs_uuid_table = krealloc(xfs_uuid_table,
88 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
89 GFP_KERNEL | __GFP_NOFAIL);
90 hole = xfs_uuid_table_size++;
91 }
92 xfs_uuid_table[hole] = *uuid;
93 mutex_unlock(&xfs_uuid_table_mutex);
94
95 return 0;
96
97 out_duplicate:
98 mutex_unlock(&xfs_uuid_table_mutex);
99 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
100 return -EINVAL;
101 }
102
103 STATIC void
104 xfs_uuid_unmount(
105 struct xfs_mount *mp)
106 {
107 uuid_t *uuid = &mp->m_sb.sb_uuid;
108 int i;
109
110 if (xfs_has_nouuid(mp))
111 return;
112
113 mutex_lock(&xfs_uuid_table_mutex);
114 for (i = 0; i < xfs_uuid_table_size; i++) {
115 if (uuid_is_null(&xfs_uuid_table[i]))
116 continue;
117 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
118 continue;
119 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
120 break;
121 }
122 ASSERT(i < xfs_uuid_table_size);
123 mutex_unlock(&xfs_uuid_table_mutex);
124 }
125
126 /*
127 * Check size of device based on the (data/realtime) block count.
128 * Note: this check is used by the growfs code as well as mount.
129 */
130 int
131 xfs_sb_validate_fsb_count(
132 xfs_sb_t *sbp,
133 uint64_t nblocks)
134 {
135 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
136 ASSERT(sbp->sb_blocklog >= BBSHIFT);
137
138 /* Limited by ULONG_MAX of page cache index */
139 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
140 return -EFBIG;
141 return 0;
142 }
143
144 /*
145 * xfs_readsb
146 *
147 * Does the initial read of the superblock.
148 */
149 int
150 xfs_readsb(
151 struct xfs_mount *mp,
152 int flags)
153 {
154 unsigned int sector_size;
155 struct xfs_buf *bp;
156 struct xfs_sb *sbp = &mp->m_sb;
157 int error;
158 int loud = !(flags & XFS_MFSI_QUIET);
159 const struct xfs_buf_ops *buf_ops;
160
161 ASSERT(mp->m_sb_bp == NULL);
162 ASSERT(mp->m_ddev_targp != NULL);
163
164 /*
165 * For the initial read, we must guess at the sector
166 * size based on the block device. It's enough to
167 * get the sb_sectsize out of the superblock and
168 * then reread with the proper length.
169 * We don't verify it yet, because it may not be complete.
170 */
171 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
172 buf_ops = NULL;
173
174 /*
175 * Allocate a (locked) buffer to hold the superblock. This will be kept
176 * around at all times to optimize access to the superblock. Therefore,
177 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
178 * elevated.
179 */
180 reread:
181 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
182 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
183 buf_ops);
184 if (error) {
185 if (loud)
186 xfs_warn(mp, "SB validate failed with error %d.", error);
187 /* bad CRC means corrupted metadata */
188 if (error == -EFSBADCRC)
189 error = -EFSCORRUPTED;
190 return error;
191 }
192
193 /*
194 * Initialize the mount structure from the superblock.
195 */
196 xfs_sb_from_disk(sbp, bp->b_addr);
197
198 /*
199 * If we haven't validated the superblock, do so now before we try
200 * to check the sector size and reread the superblock appropriately.
201 */
202 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
203 if (loud)
204 xfs_warn(mp, "Invalid superblock magic number");
205 error = -EINVAL;
206 goto release_buf;
207 }
208
209 /*
210 * We must be able to do sector-sized and sector-aligned IO.
211 */
212 if (sector_size > sbp->sb_sectsize) {
213 if (loud)
214 xfs_warn(mp, "device supports %u byte sectors (not %u)",
215 sector_size, sbp->sb_sectsize);
216 error = -ENOSYS;
217 goto release_buf;
218 }
219
220 if (buf_ops == NULL) {
221 /*
222 * Re-read the superblock so the buffer is correctly sized,
223 * and properly verified.
224 */
225 xfs_buf_relse(bp);
226 sector_size = sbp->sb_sectsize;
227 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
228 goto reread;
229 }
230
231 mp->m_features |= xfs_sb_version_to_features(sbp);
232 xfs_reinit_percpu_counters(mp);
233
234 /*
235 * If logged xattrs are enabled after log recovery finishes, then set
236 * the opstate so that log recovery will work properly.
237 */
238 if (xfs_sb_version_haslogxattrs(&mp->m_sb))
239 xfs_set_using_logged_xattrs(mp);
240
241 /* no need to be quiet anymore, so reset the buf ops */
242 bp->b_ops = &xfs_sb_buf_ops;
243
244 mp->m_sb_bp = bp;
245 xfs_buf_unlock(bp);
246 return 0;
247
248 release_buf:
249 xfs_buf_relse(bp);
250 return error;
251 }
252
253 /*
254 * If the sunit/swidth change would move the precomputed root inode value, we
255 * must reject the ondisk change because repair will stumble over that.
256 * However, we allow the mount to proceed because we never rejected this
257 * combination before. Returns true to update the sb, false otherwise.
258 */
259 static inline int
260 xfs_check_new_dalign(
261 struct xfs_mount *mp,
262 int new_dalign,
263 bool *update_sb)
264 {
265 struct xfs_sb *sbp = &mp->m_sb;
266 xfs_ino_t calc_ino;
267
268 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
269 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
270
271 if (sbp->sb_rootino == calc_ino) {
272 *update_sb = true;
273 return 0;
274 }
275
276 xfs_warn(mp,
277 "Cannot change stripe alignment; would require moving root inode.");
278
279 /*
280 * XXX: Next time we add a new incompat feature, this should start
281 * returning -EINVAL to fail the mount. Until then, spit out a warning
282 * that we're ignoring the administrator's instructions.
283 */
284 xfs_warn(mp, "Skipping superblock stripe alignment update.");
285 *update_sb = false;
286 return 0;
287 }
288
289 /*
290 * If we were provided with new sunit/swidth values as mount options, make sure
291 * that they pass basic alignment and superblock feature checks, and convert
292 * them into the same units (FSB) that everything else expects. This step
293 * /must/ be done before computing the inode geometry.
294 */
295 STATIC int
296 xfs_validate_new_dalign(
297 struct xfs_mount *mp)
298 {
299 if (mp->m_dalign == 0)
300 return 0;
301
302 /*
303 * If stripe unit and stripe width are not multiples
304 * of the fs blocksize turn off alignment.
305 */
306 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
307 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
308 xfs_warn(mp,
309 "alignment check failed: sunit/swidth vs. blocksize(%d)",
310 mp->m_sb.sb_blocksize);
311 return -EINVAL;
312 }
313
314 /*
315 * Convert the stripe unit and width to FSBs.
316 */
317 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
318 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
319 xfs_warn(mp,
320 "alignment check failed: sunit/swidth vs. agsize(%d)",
321 mp->m_sb.sb_agblocks);
322 return -EINVAL;
323 }
324
325 if (!mp->m_dalign) {
326 xfs_warn(mp,
327 "alignment check failed: sunit(%d) less than bsize(%d)",
328 mp->m_dalign, mp->m_sb.sb_blocksize);
329 return -EINVAL;
330 }
331
332 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
333
334 if (!xfs_has_dalign(mp)) {
335 xfs_warn(mp,
336 "cannot change alignment: superblock does not support data alignment");
337 return -EINVAL;
338 }
339
340 return 0;
341 }
342
343 /* Update alignment values based on mount options and sb values. */
344 STATIC int
345 xfs_update_alignment(
346 struct xfs_mount *mp)
347 {
348 struct xfs_sb *sbp = &mp->m_sb;
349
350 if (mp->m_dalign) {
351 bool update_sb;
352 int error;
353
354 if (sbp->sb_unit == mp->m_dalign &&
355 sbp->sb_width == mp->m_swidth)
356 return 0;
357
358 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
359 if (error || !update_sb)
360 return error;
361
362 sbp->sb_unit = mp->m_dalign;
363 sbp->sb_width = mp->m_swidth;
364 mp->m_update_sb = true;
365 } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
366 mp->m_dalign = sbp->sb_unit;
367 mp->m_swidth = sbp->sb_width;
368 }
369
370 return 0;
371 }
372
373 /*
374 * precalculate the low space thresholds for dynamic speculative preallocation.
375 */
376 void
377 xfs_set_low_space_thresholds(
378 struct xfs_mount *mp)
379 {
380 uint64_t dblocks = mp->m_sb.sb_dblocks;
381 uint64_t rtexts = mp->m_sb.sb_rextents;
382 int i;
383
384 do_div(dblocks, 100);
385 do_div(rtexts, 100);
386
387 for (i = 0; i < XFS_LOWSP_MAX; i++) {
388 mp->m_low_space[i] = dblocks * (i + 1);
389 mp->m_low_rtexts[i] = rtexts * (i + 1);
390 }
391 }
392
393 /*
394 * Check that the data (and log if separate) is an ok size.
395 */
396 STATIC int
397 xfs_check_sizes(
398 struct xfs_mount *mp)
399 {
400 struct xfs_buf *bp;
401 xfs_daddr_t d;
402 int error;
403
404 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
405 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
406 xfs_warn(mp, "filesystem size mismatch detected");
407 return -EFBIG;
408 }
409 error = xfs_buf_read_uncached(mp->m_ddev_targp,
410 d - XFS_FSS_TO_BB(mp, 1),
411 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
412 if (error) {
413 xfs_warn(mp, "last sector read failed");
414 return error;
415 }
416 xfs_buf_relse(bp);
417
418 if (mp->m_logdev_targp == mp->m_ddev_targp)
419 return 0;
420
421 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
422 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
423 xfs_warn(mp, "log size mismatch detected");
424 return -EFBIG;
425 }
426 error = xfs_buf_read_uncached(mp->m_logdev_targp,
427 d - XFS_FSB_TO_BB(mp, 1),
428 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
429 if (error) {
430 xfs_warn(mp, "log device read failed");
431 return error;
432 }
433 xfs_buf_relse(bp);
434 return 0;
435 }
436
437 /*
438 * Clear the quotaflags in memory and in the superblock.
439 */
440 int
441 xfs_mount_reset_sbqflags(
442 struct xfs_mount *mp)
443 {
444 mp->m_qflags = 0;
445
446 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
447 if (mp->m_sb.sb_qflags == 0)
448 return 0;
449 spin_lock(&mp->m_sb_lock);
450 mp->m_sb.sb_qflags = 0;
451 spin_unlock(&mp->m_sb_lock);
452
453 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
454 return 0;
455
456 return xfs_sync_sb(mp, false);
457 }
458
459 uint64_t
460 xfs_default_resblks(xfs_mount_t *mp)
461 {
462 uint64_t resblks;
463
464 /*
465 * We default to 5% or 8192 fsbs of space reserved, whichever is
466 * smaller. This is intended to cover concurrent allocation
467 * transactions when we initially hit enospc. These each require a 4
468 * block reservation. Hence by default we cover roughly 2000 concurrent
469 * allocation reservations.
470 */
471 resblks = mp->m_sb.sb_dblocks;
472 do_div(resblks, 20);
473 resblks = min_t(uint64_t, resblks, 8192);
474 return resblks;
475 }
476
477 /* Ensure the summary counts are correct. */
478 STATIC int
479 xfs_check_summary_counts(
480 struct xfs_mount *mp)
481 {
482 int error = 0;
483
484 /*
485 * The AG0 superblock verifier rejects in-progress filesystems,
486 * so we should never see the flag set this far into mounting.
487 */
488 if (mp->m_sb.sb_inprogress) {
489 xfs_err(mp, "sb_inprogress set after log recovery??");
490 WARN_ON(1);
491 return -EFSCORRUPTED;
492 }
493
494 /*
495 * Now the log is mounted, we know if it was an unclean shutdown or
496 * not. If it was, with the first phase of recovery has completed, we
497 * have consistent AG blocks on disk. We have not recovered EFIs yet,
498 * but they are recovered transactionally in the second recovery phase
499 * later.
500 *
501 * If the log was clean when we mounted, we can check the summary
502 * counters. If any of them are obviously incorrect, we can recompute
503 * them from the AGF headers in the next step.
504 */
505 if (xfs_is_clean(mp) &&
506 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
507 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
508 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
509 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
510
511 /*
512 * We can safely re-initialise incore superblock counters from the
513 * per-ag data. These may not be correct if the filesystem was not
514 * cleanly unmounted, so we waited for recovery to finish before doing
515 * this.
516 *
517 * If the filesystem was cleanly unmounted or the previous check did
518 * not flag anything weird, then we can trust the values in the
519 * superblock to be correct and we don't need to do anything here.
520 * Otherwise, recalculate the summary counters.
521 */
522 if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
523 xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
524 error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
525 if (error)
526 return error;
527 }
528
529 /*
530 * Older kernels misused sb_frextents to reflect both incore
531 * reservations made by running transactions and the actual count of
532 * free rt extents in the ondisk metadata. Transactions committed
533 * during runtime can therefore contain a superblock update that
534 * undercounts the number of free rt extents tracked in the rt bitmap.
535 * A clean unmount record will have the correct frextents value since
536 * there can be no other transactions running at that point.
537 *
538 * If we're mounting the rt volume after recovering the log, recompute
539 * frextents from the rtbitmap file to fix the inconsistency.
540 */
541 if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
542 error = xfs_rtalloc_reinit_frextents(mp);
543 if (error)
544 return error;
545 }
546
547 return 0;
548 }
549
550 static void
551 xfs_unmount_check(
552 struct xfs_mount *mp)
553 {
554 if (xfs_is_shutdown(mp))
555 return;
556
557 if (percpu_counter_sum(&mp->m_ifree) >
558 percpu_counter_sum(&mp->m_icount)) {
559 xfs_alert(mp, "ifree/icount mismatch at unmount");
560 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
561 }
562 }
563
564 /*
565 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
566 * internal inode structures can be sitting in the CIL and AIL at this point,
567 * so we need to unpin them, write them back and/or reclaim them before unmount
568 * can proceed. In other words, callers are required to have inactivated all
569 * inodes.
570 *
571 * An inode cluster that has been freed can have its buffer still pinned in
572 * memory because the transaction is still sitting in a iclog. The stale inodes
573 * on that buffer will be pinned to the buffer until the transaction hits the
574 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
575 * may never see the pinned buffer, so nothing will push out the iclog and
576 * unpin the buffer.
577 *
578 * Hence we need to force the log to unpin everything first. However, log
579 * forces don't wait for the discards they issue to complete, so we have to
580 * explicitly wait for them to complete here as well.
581 *
582 * Then we can tell the world we are unmounting so that error handling knows
583 * that the filesystem is going away and we should error out anything that we
584 * have been retrying in the background. This will prevent never-ending
585 * retries in AIL pushing from hanging the unmount.
586 *
587 * Finally, we can push the AIL to clean all the remaining dirty objects, then
588 * reclaim the remaining inodes that are still in memory at this point in time.
589 */
590 static void
591 xfs_unmount_flush_inodes(
592 struct xfs_mount *mp)
593 {
594 xfs_log_force(mp, XFS_LOG_SYNC);
595 xfs_extent_busy_wait_all(mp);
596 flush_workqueue(xfs_discard_wq);
597
598 set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate);
599
600 xfs_ail_push_all_sync(mp->m_ail);
601 xfs_inodegc_stop(mp);
602 cancel_delayed_work_sync(&mp->m_reclaim_work);
603 xfs_reclaim_inodes(mp);
604 xfs_health_unmount(mp);
605 }
606
607 static void
608 xfs_mount_setup_inode_geom(
609 struct xfs_mount *mp)
610 {
611 struct xfs_ino_geometry *igeo = M_IGEO(mp);
612
613 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
614 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
615
616 xfs_ialloc_setup_geometry(mp);
617 }
618
619 /* Compute maximum possible height for per-AG btree types for this fs. */
620 static inline void
621 xfs_agbtree_compute_maxlevels(
622 struct xfs_mount *mp)
623 {
624 unsigned int levels;
625
626 levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
627 levels = max(levels, mp->m_rmap_maxlevels);
628 mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
629 }
630
631 /*
632 * This function does the following on an initial mount of a file system:
633 * - reads the superblock from disk and init the mount struct
634 * - if we're a 32-bit kernel, do a size check on the superblock
635 * so we don't mount terabyte filesystems
636 * - init mount struct realtime fields
637 * - allocate inode hash table for fs
638 * - init directory manager
639 * - perform recovery and init the log manager
640 */
641 int
642 xfs_mountfs(
643 struct xfs_mount *mp)
644 {
645 struct xfs_sb *sbp = &(mp->m_sb);
646 struct xfs_inode *rip;
647 struct xfs_ino_geometry *igeo = M_IGEO(mp);
648 uint quotamount = 0;
649 uint quotaflags = 0;
650 int error = 0;
651
652 xfs_sb_mount_common(mp, sbp);
653
654 /*
655 * Check for a mismatched features2 values. Older kernels read & wrote
656 * into the wrong sb offset for sb_features2 on some platforms due to
657 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
658 * which made older superblock reading/writing routines swap it as a
659 * 64-bit value.
660 *
661 * For backwards compatibility, we make both slots equal.
662 *
663 * If we detect a mismatched field, we OR the set bits into the existing
664 * features2 field in case it has already been modified; we don't want
665 * to lose any features. We then update the bad location with the ORed
666 * value so that older kernels will see any features2 flags. The
667 * superblock writeback code ensures the new sb_features2 is copied to
668 * sb_bad_features2 before it is logged or written to disk.
669 */
670 if (xfs_sb_has_mismatched_features2(sbp)) {
671 xfs_warn(mp, "correcting sb_features alignment problem");
672 sbp->sb_features2 |= sbp->sb_bad_features2;
673 mp->m_update_sb = true;
674 }
675
676
677 /* always use v2 inodes by default now */
678 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
679 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
680 mp->m_features |= XFS_FEAT_NLINK;
681 mp->m_update_sb = true;
682 }
683
684 /*
685 * If we were given new sunit/swidth options, do some basic validation
686 * checks and convert the incore dalign and swidth values to the
687 * same units (FSB) that everything else uses. This /must/ happen
688 * before computing the inode geometry.
689 */
690 error = xfs_validate_new_dalign(mp);
691 if (error)
692 goto out;
693
694 xfs_alloc_compute_maxlevels(mp);
695 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
696 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
697 xfs_mount_setup_inode_geom(mp);
698 xfs_rmapbt_compute_maxlevels(mp);
699 xfs_refcountbt_compute_maxlevels(mp);
700
701 xfs_agbtree_compute_maxlevels(mp);
702
703 /*
704 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
705 * is NOT aligned turn off m_dalign since allocator alignment is within
706 * an ag, therefore ag has to be aligned at stripe boundary. Note that
707 * we must compute the free space and rmap btree geometry before doing
708 * this.
709 */
710 error = xfs_update_alignment(mp);
711 if (error)
712 goto out;
713
714 /* enable fail_at_unmount as default */
715 mp->m_fail_unmount = true;
716
717 super_set_sysfs_name_id(mp->m_super);
718
719 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
720 NULL, mp->m_super->s_id);
721 if (error)
722 goto out;
723
724 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
725 &mp->m_kobj, "stats");
726 if (error)
727 goto out_remove_sysfs;
728
729 xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
730
731 error = xfs_error_sysfs_init(mp);
732 if (error)
733 goto out_remove_scrub_stats;
734
735 error = xfs_errortag_init(mp);
736 if (error)
737 goto out_remove_error_sysfs;
738
739 error = xfs_uuid_mount(mp);
740 if (error)
741 goto out_remove_errortag;
742
743 /*
744 * Update the preferred write size based on the information from the
745 * on-disk superblock.
746 */
747 mp->m_allocsize_log =
748 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
749 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
750
751 /* set the low space thresholds for dynamic preallocation */
752 xfs_set_low_space_thresholds(mp);
753
754 /*
755 * If enabled, sparse inode chunk alignment is expected to match the
756 * cluster size. Full inode chunk alignment must match the chunk size,
757 * but that is checked on sb read verification...
758 */
759 if (xfs_has_sparseinodes(mp) &&
760 mp->m_sb.sb_spino_align !=
761 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
762 xfs_warn(mp,
763 "Sparse inode block alignment (%u) must match cluster size (%llu).",
764 mp->m_sb.sb_spino_align,
765 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
766 error = -EINVAL;
767 goto out_remove_uuid;
768 }
769
770 /*
771 * Check that the data (and log if separate) is an ok size.
772 */
773 error = xfs_check_sizes(mp);
774 if (error)
775 goto out_remove_uuid;
776
777 /*
778 * Initialize realtime fields in the mount structure
779 */
780 error = xfs_rtmount_init(mp);
781 if (error) {
782 xfs_warn(mp, "RT mount failed");
783 goto out_remove_uuid;
784 }
785
786 /*
787 * Copies the low order bits of the timestamp and the randomly
788 * set "sequence" number out of a UUID.
789 */
790 mp->m_fixedfsid[0] =
791 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
792 get_unaligned_be16(&sbp->sb_uuid.b[4]);
793 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
794
795 error = xfs_da_mount(mp);
796 if (error) {
797 xfs_warn(mp, "Failed dir/attr init: %d", error);
798 goto out_remove_uuid;
799 }
800
801 /*
802 * Initialize the precomputed transaction reservations values.
803 */
804 xfs_trans_init(mp);
805
806 /*
807 * Allocate and initialize the per-ag data.
808 */
809 error = xfs_initialize_perag(mp, sbp->sb_agcount, mp->m_sb.sb_dblocks,
810 &mp->m_maxagi);
811 if (error) {
812 xfs_warn(mp, "Failed per-ag init: %d", error);
813 goto out_free_dir;
814 }
815
816 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
817 xfs_warn(mp, "no log defined");
818 error = -EFSCORRUPTED;
819 goto out_free_perag;
820 }
821
822 error = xfs_inodegc_register_shrinker(mp);
823 if (error)
824 goto out_fail_wait;
825
826 /*
827 * Log's mount-time initialization. The first part of recovery can place
828 * some items on the AIL, to be handled when recovery is finished or
829 * cancelled.
830 */
831 error = xfs_log_mount(mp, mp->m_logdev_targp,
832 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
833 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
834 if (error) {
835 xfs_warn(mp, "log mount failed");
836 goto out_inodegc_shrinker;
837 }
838
839 /*
840 * If logged xattrs are still enabled after log recovery finishes, then
841 * they'll be available until unmount. Otherwise, turn them off.
842 */
843 if (xfs_sb_version_haslogxattrs(&mp->m_sb))
844 xfs_set_using_logged_xattrs(mp);
845 else
846 xfs_clear_using_logged_xattrs(mp);
847
848 /* Enable background inode inactivation workers. */
849 xfs_inodegc_start(mp);
850 xfs_blockgc_start(mp);
851
852 /*
853 * Now that we've recovered any pending superblock feature bit
854 * additions, we can finish setting up the attr2 behaviour for the
855 * mount. The noattr2 option overrides the superblock flag, so only
856 * check the superblock feature flag if the mount option is not set.
857 */
858 if (xfs_has_noattr2(mp)) {
859 mp->m_features &= ~XFS_FEAT_ATTR2;
860 } else if (!xfs_has_attr2(mp) &&
861 (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) {
862 mp->m_features |= XFS_FEAT_ATTR2;
863 }
864
865 /*
866 * Get and sanity-check the root inode.
867 * Save the pointer to it in the mount structure.
868 */
869 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
870 XFS_ILOCK_EXCL, &rip);
871 if (error) {
872 xfs_warn(mp,
873 "Failed to read root inode 0x%llx, error %d",
874 sbp->sb_rootino, -error);
875 goto out_log_dealloc;
876 }
877
878 ASSERT(rip != NULL);
879
880 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
881 xfs_warn(mp, "corrupted root inode %llu: not a directory",
882 (unsigned long long)rip->i_ino);
883 xfs_iunlock(rip, XFS_ILOCK_EXCL);
884 error = -EFSCORRUPTED;
885 goto out_rele_rip;
886 }
887 mp->m_rootip = rip; /* save it */
888
889 xfs_iunlock(rip, XFS_ILOCK_EXCL);
890
891 /*
892 * Initialize realtime inode pointers in the mount structure
893 */
894 error = xfs_rtmount_inodes(mp);
895 if (error) {
896 /*
897 * Free up the root inode.
898 */
899 xfs_warn(mp, "failed to read RT inodes");
900 goto out_rele_rip;
901 }
902
903 /* Make sure the summary counts are ok. */
904 error = xfs_check_summary_counts(mp);
905 if (error)
906 goto out_rtunmount;
907
908 /*
909 * If this is a read-only mount defer the superblock updates until
910 * the next remount into writeable mode. Otherwise we would never
911 * perform the update e.g. for the root filesystem.
912 */
913 if (mp->m_update_sb && !xfs_is_readonly(mp)) {
914 error = xfs_sync_sb(mp, false);
915 if (error) {
916 xfs_warn(mp, "failed to write sb changes");
917 goto out_rtunmount;
918 }
919 }
920
921 /*
922 * Initialise the XFS quota management subsystem for this mount
923 */
924 if (XFS_IS_QUOTA_ON(mp)) {
925 error = xfs_qm_newmount(mp, "amount, "aflags);
926 if (error)
927 goto out_rtunmount;
928 } else {
929 /*
930 * If a file system had quotas running earlier, but decided to
931 * mount without -o uquota/pquota/gquota options, revoke the
932 * quotachecked license.
933 */
934 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
935 xfs_notice(mp, "resetting quota flags");
936 error = xfs_mount_reset_sbqflags(mp);
937 if (error)
938 goto out_rtunmount;
939 }
940 }
941
942 /*
943 * Finish recovering the file system. This part needed to be delayed
944 * until after the root and real-time bitmap inodes were consistently
945 * read in. Temporarily create per-AG space reservations for metadata
946 * btree shape changes because space freeing transactions (for inode
947 * inactivation) require the per-AG reservation in lieu of reserving
948 * blocks.
949 */
950 error = xfs_fs_reserve_ag_blocks(mp);
951 if (error && error == -ENOSPC)
952 xfs_warn(mp,
953 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
954 error = xfs_log_mount_finish(mp);
955 xfs_fs_unreserve_ag_blocks(mp);
956 if (error) {
957 xfs_warn(mp, "log mount finish failed");
958 goto out_rtunmount;
959 }
960
961 /*
962 * Now the log is fully replayed, we can transition to full read-only
963 * mode for read-only mounts. This will sync all the metadata and clean
964 * the log so that the recovery we just performed does not have to be
965 * replayed again on the next mount.
966 *
967 * We use the same quiesce mechanism as the rw->ro remount, as they are
968 * semantically identical operations.
969 */
970 if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
971 xfs_log_clean(mp);
972
973 /*
974 * Complete the quota initialisation, post-log-replay component.
975 */
976 if (quotamount) {
977 ASSERT(mp->m_qflags == 0);
978 mp->m_qflags = quotaflags;
979
980 xfs_qm_mount_quotas(mp);
981 }
982
983 /*
984 * Now we are mounted, reserve a small amount of unused space for
985 * privileged transactions. This is needed so that transaction
986 * space required for critical operations can dip into this pool
987 * when at ENOSPC. This is needed for operations like create with
988 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
989 * are not allowed to use this reserved space.
990 *
991 * This may drive us straight to ENOSPC on mount, but that implies
992 * we were already there on the last unmount. Warn if this occurs.
993 */
994 if (!xfs_is_readonly(mp)) {
995 error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
996 if (error)
997 xfs_warn(mp,
998 "Unable to allocate reserve blocks. Continuing without reserve pool.");
999
1000 /* Reserve AG blocks for future btree expansion. */
1001 error = xfs_fs_reserve_ag_blocks(mp);
1002 if (error && error != -ENOSPC)
1003 goto out_agresv;
1004 }
1005
1006 return 0;
1007
1008 out_agresv:
1009 xfs_fs_unreserve_ag_blocks(mp);
1010 xfs_qm_unmount_quotas(mp);
1011 out_rtunmount:
1012 xfs_rtunmount_inodes(mp);
1013 out_rele_rip:
1014 xfs_irele(rip);
1015 /* Clean out dquots that might be in memory after quotacheck. */
1016 xfs_qm_unmount(mp);
1017
1018 /*
1019 * Inactivate all inodes that might still be in memory after a log
1020 * intent recovery failure so that reclaim can free them. Metadata
1021 * inodes and the root directory shouldn't need inactivation, but the
1022 * mount failed for some reason, so pull down all the state and flee.
1023 */
1024 xfs_inodegc_flush(mp);
1025
1026 /*
1027 * Flush all inode reclamation work and flush the log.
1028 * We have to do this /after/ rtunmount and qm_unmount because those
1029 * two will have scheduled delayed reclaim for the rt/quota inodes.
1030 *
1031 * This is slightly different from the unmountfs call sequence
1032 * because we could be tearing down a partially set up mount. In
1033 * particular, if log_mount_finish fails we bail out without calling
1034 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1035 * quota inodes.
1036 */
1037 xfs_unmount_flush_inodes(mp);
1038 out_log_dealloc:
1039 xfs_log_mount_cancel(mp);
1040 out_inodegc_shrinker:
1041 shrinker_free(mp->m_inodegc_shrinker);
1042 out_fail_wait:
1043 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1044 xfs_buftarg_drain(mp->m_logdev_targp);
1045 xfs_buftarg_drain(mp->m_ddev_targp);
1046 out_free_perag:
1047 xfs_free_perag(mp);
1048 out_free_dir:
1049 xfs_da_unmount(mp);
1050 out_remove_uuid:
1051 xfs_uuid_unmount(mp);
1052 out_remove_errortag:
1053 xfs_errortag_del(mp);
1054 out_remove_error_sysfs:
1055 xfs_error_sysfs_del(mp);
1056 out_remove_scrub_stats:
1057 xchk_stats_unregister(mp->m_scrub_stats);
1058 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1059 out_remove_sysfs:
1060 xfs_sysfs_del(&mp->m_kobj);
1061 out:
1062 return error;
1063 }
1064
1065 /*
1066 * This flushes out the inodes,dquots and the superblock, unmounts the
1067 * log and makes sure that incore structures are freed.
1068 */
1069 void
1070 xfs_unmountfs(
1071 struct xfs_mount *mp)
1072 {
1073 int error;
1074
1075 /*
1076 * Perform all on-disk metadata updates required to inactivate inodes
1077 * that the VFS evicted earlier in the unmount process. Freeing inodes
1078 * and discarding CoW fork preallocations can cause shape changes to
1079 * the free inode and refcount btrees, respectively, so we must finish
1080 * this before we discard the metadata space reservations. Metadata
1081 * inodes and the root directory do not require inactivation.
1082 */
1083 xfs_inodegc_flush(mp);
1084
1085 xfs_blockgc_stop(mp);
1086 xfs_fs_unreserve_ag_blocks(mp);
1087 xfs_qm_unmount_quotas(mp);
1088 xfs_rtunmount_inodes(mp);
1089 xfs_irele(mp->m_rootip);
1090
1091 xfs_unmount_flush_inodes(mp);
1092
1093 xfs_qm_unmount(mp);
1094
1095 /*
1096 * Unreserve any blocks we have so that when we unmount we don't account
1097 * the reserved free space as used. This is really only necessary for
1098 * lazy superblock counting because it trusts the incore superblock
1099 * counters to be absolutely correct on clean unmount.
1100 *
1101 * We don't bother correcting this elsewhere for lazy superblock
1102 * counting because on mount of an unclean filesystem we reconstruct the
1103 * correct counter value and this is irrelevant.
1104 *
1105 * For non-lazy counter filesystems, this doesn't matter at all because
1106 * we only every apply deltas to the superblock and hence the incore
1107 * value does not matter....
1108 */
1109 error = xfs_reserve_blocks(mp, 0);
1110 if (error)
1111 xfs_warn(mp, "Unable to free reserved block pool. "
1112 "Freespace may not be correct on next mount.");
1113 xfs_unmount_check(mp);
1114
1115 /*
1116 * Indicate that it's ok to clear log incompat bits before cleaning
1117 * the log and writing the unmount record.
1118 */
1119 xfs_set_done_with_log_incompat(mp);
1120 xfs_log_unmount(mp);
1121 xfs_da_unmount(mp);
1122 xfs_uuid_unmount(mp);
1123
1124 #if defined(DEBUG)
1125 xfs_errortag_clearall(mp);
1126 #endif
1127 shrinker_free(mp->m_inodegc_shrinker);
1128 xfs_free_perag(mp);
1129
1130 xfs_errortag_del(mp);
1131 xfs_error_sysfs_del(mp);
1132 xchk_stats_unregister(mp->m_scrub_stats);
1133 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1134 xfs_sysfs_del(&mp->m_kobj);
1135 }
1136
1137 /*
1138 * Determine whether modifications can proceed. The caller specifies the minimum
1139 * freeze level for which modifications should not be allowed. This allows
1140 * certain operations to proceed while the freeze sequence is in progress, if
1141 * necessary.
1142 */
1143 bool
1144 xfs_fs_writable(
1145 struct xfs_mount *mp,
1146 int level)
1147 {
1148 ASSERT(level > SB_UNFROZEN);
1149 if ((mp->m_super->s_writers.frozen >= level) ||
1150 xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1151 return false;
1152
1153 return true;
1154 }
1155
1156 void
1157 xfs_add_freecounter(
1158 struct xfs_mount *mp,
1159 struct percpu_counter *counter,
1160 uint64_t delta)
1161 {
1162 bool has_resv_pool = (counter == &mp->m_fdblocks);
1163 uint64_t res_used;
1164
1165 /*
1166 * If the reserve pool is depleted, put blocks back into it first.
1167 * Most of the time the pool is full.
1168 */
1169 if (!has_resv_pool || mp->m_resblks == mp->m_resblks_avail) {
1170 percpu_counter_add(counter, delta);
1171 return;
1172 }
1173
1174 spin_lock(&mp->m_sb_lock);
1175 res_used = mp->m_resblks - mp->m_resblks_avail;
1176 if (res_used > delta) {
1177 mp->m_resblks_avail += delta;
1178 } else {
1179 delta -= res_used;
1180 mp->m_resblks_avail = mp->m_resblks;
1181 percpu_counter_add(counter, delta);
1182 }
1183 spin_unlock(&mp->m_sb_lock);
1184 }
1185
1186 int
1187 xfs_dec_freecounter(
1188 struct xfs_mount *mp,
1189 struct percpu_counter *counter,
1190 uint64_t delta,
1191 bool rsvd)
1192 {
1193 int64_t lcounter;
1194 uint64_t set_aside = 0;
1195 s32 batch;
1196 bool has_resv_pool;
1197
1198 ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1199 has_resv_pool = (counter == &mp->m_fdblocks);
1200 if (rsvd)
1201 ASSERT(has_resv_pool);
1202
1203 /*
1204 * Taking blocks away, need to be more accurate the closer we
1205 * are to zero.
1206 *
1207 * If the counter has a value of less than 2 * max batch size,
1208 * then make everything serialise as we are real close to
1209 * ENOSPC.
1210 */
1211 if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1212 XFS_FDBLOCKS_BATCH) < 0)
1213 batch = 1;
1214 else
1215 batch = XFS_FDBLOCKS_BATCH;
1216
1217 /*
1218 * Set aside allocbt blocks because these blocks are tracked as free
1219 * space but not available for allocation. Technically this means that a
1220 * single reservation cannot consume all remaining free space, but the
1221 * ratio of allocbt blocks to usable free blocks should be rather small.
1222 * The tradeoff without this is that filesystems that maintain high
1223 * perag block reservations can over reserve physical block availability
1224 * and fail physical allocation, which leads to much more serious
1225 * problems (i.e. transaction abort, pagecache discards, etc.) than
1226 * slightly premature -ENOSPC.
1227 */
1228 if (has_resv_pool)
1229 set_aside = xfs_fdblocks_unavailable(mp);
1230 percpu_counter_add_batch(counter, -((int64_t)delta), batch);
1231 if (__percpu_counter_compare(counter, set_aside,
1232 XFS_FDBLOCKS_BATCH) >= 0) {
1233 /* we had space! */
1234 return 0;
1235 }
1236
1237 /*
1238 * lock up the sb for dipping into reserves before releasing the space
1239 * that took us to ENOSPC.
1240 */
1241 spin_lock(&mp->m_sb_lock);
1242 percpu_counter_add(counter, delta);
1243 if (!has_resv_pool || !rsvd)
1244 goto fdblocks_enospc;
1245
1246 lcounter = (long long)mp->m_resblks_avail - delta;
1247 if (lcounter >= 0) {
1248 mp->m_resblks_avail = lcounter;
1249 spin_unlock(&mp->m_sb_lock);
1250 return 0;
1251 }
1252 xfs_warn_once(mp,
1253 "Reserve blocks depleted! Consider increasing reserve pool size.");
1254
1255 fdblocks_enospc:
1256 spin_unlock(&mp->m_sb_lock);
1257 return -ENOSPC;
1258 }
1259
1260 /*
1261 * Used to free the superblock along various error paths.
1262 */
1263 void
1264 xfs_freesb(
1265 struct xfs_mount *mp)
1266 {
1267 struct xfs_buf *bp = mp->m_sb_bp;
1268
1269 xfs_buf_lock(bp);
1270 mp->m_sb_bp = NULL;
1271 xfs_buf_relse(bp);
1272 }
1273
1274 /*
1275 * If the underlying (data/log/rt) device is readonly, there are some
1276 * operations that cannot proceed.
1277 */
1278 int
1279 xfs_dev_is_read_only(
1280 struct xfs_mount *mp,
1281 char *message)
1282 {
1283 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1284 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1285 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1286 xfs_notice(mp, "%s required on read-only device.", message);
1287 xfs_notice(mp, "write access unavailable, cannot proceed.");
1288 return -EROFS;
1289 }
1290 return 0;
1291 }
1292
1293 /* Force the summary counters to be recalculated at next mount. */
1294 void
1295 xfs_force_summary_recalc(
1296 struct xfs_mount *mp)
1297 {
1298 if (!xfs_has_lazysbcount(mp))
1299 return;
1300
1301 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1302 }
1303
1304 /*
1305 * Enable a log incompat feature flag in the primary superblock. The caller
1306 * cannot have any other transactions in progress.
1307 */
1308 int
1309 xfs_add_incompat_log_feature(
1310 struct xfs_mount *mp,
1311 uint32_t feature)
1312 {
1313 struct xfs_dsb *dsb;
1314 int error;
1315
1316 ASSERT(hweight32(feature) == 1);
1317 ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1318
1319 /*
1320 * Force the log to disk and kick the background AIL thread to reduce
1321 * the chances that the bwrite will stall waiting for the AIL to unpin
1322 * the primary superblock buffer. This isn't a data integrity
1323 * operation, so we don't need a synchronous push.
1324 */
1325 error = xfs_log_force(mp, XFS_LOG_SYNC);
1326 if (error)
1327 return error;
1328 xfs_ail_push_all(mp->m_ail);
1329
1330 /*
1331 * Lock the primary superblock buffer to serialize all callers that
1332 * are trying to set feature bits.
1333 */
1334 xfs_buf_lock(mp->m_sb_bp);
1335 xfs_buf_hold(mp->m_sb_bp);
1336
1337 if (xfs_is_shutdown(mp)) {
1338 error = -EIO;
1339 goto rele;
1340 }
1341
1342 if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1343 goto rele;
1344
1345 /*
1346 * Write the primary superblock to disk immediately, because we need
1347 * the log_incompat bit to be set in the primary super now to protect
1348 * the log items that we're going to commit later.
1349 */
1350 dsb = mp->m_sb_bp->b_addr;
1351 xfs_sb_to_disk(dsb, &mp->m_sb);
1352 dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1353 error = xfs_bwrite(mp->m_sb_bp);
1354 if (error)
1355 goto shutdown;
1356
1357 /*
1358 * Add the feature bits to the incore superblock before we unlock the
1359 * buffer.
1360 */
1361 xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1362 xfs_buf_relse(mp->m_sb_bp);
1363
1364 /* Log the superblock to disk. */
1365 return xfs_sync_sb(mp, false);
1366 shutdown:
1367 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1368 rele:
1369 xfs_buf_relse(mp->m_sb_bp);
1370 return error;
1371 }
1372
1373 /*
1374 * Clear all the log incompat flags from the superblock.
1375 *
1376 * The caller cannot be in a transaction, must ensure that the log does not
1377 * contain any log items protected by any log incompat bit, and must ensure
1378 * that there are no other threads that depend on the state of the log incompat
1379 * feature flags in the primary super.
1380 *
1381 * Returns true if the superblock is dirty.
1382 */
1383 bool
1384 xfs_clear_incompat_log_features(
1385 struct xfs_mount *mp)
1386 {
1387 bool ret = false;
1388
1389 if (!xfs_has_crc(mp) ||
1390 !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1391 XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1392 xfs_is_shutdown(mp) ||
1393 !xfs_is_done_with_log_incompat(mp))
1394 return false;
1395
1396 /*
1397 * Update the incore superblock. We synchronize on the primary super
1398 * buffer lock to be consistent with the add function, though at least
1399 * in theory this shouldn't be necessary.
1400 */
1401 xfs_buf_lock(mp->m_sb_bp);
1402 xfs_buf_hold(mp->m_sb_bp);
1403
1404 if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1405 XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1406 xfs_sb_remove_incompat_log_features(&mp->m_sb);
1407 ret = true;
1408 }
1409
1410 xfs_buf_relse(mp->m_sb_bp);
1411 return ret;
1412 }
1413
1414 /*
1415 * Update the in-core delayed block counter.
1416 *
1417 * We prefer to update the counter without having to take a spinlock for every
1418 * counter update (i.e. batching). Each change to delayed allocation
1419 * reservations can change can easily exceed the default percpu counter
1420 * batching, so we use a larger batch factor here.
1421 *
1422 * Note that we don't currently have any callers requiring fast summation
1423 * (e.g. percpu_counter_read) so we can use a big batch value here.
1424 */
1425 #define XFS_DELALLOC_BATCH (4096)
1426 void
1427 xfs_mod_delalloc(
1428 struct xfs_inode *ip,
1429 int64_t data_delta,
1430 int64_t ind_delta)
1431 {
1432 struct xfs_mount *mp = ip->i_mount;
1433
1434 if (XFS_IS_REALTIME_INODE(ip)) {
1435 percpu_counter_add_batch(&mp->m_delalloc_rtextents,
1436 xfs_rtb_to_rtx(mp, data_delta),
1437 XFS_DELALLOC_BATCH);
1438 if (!ind_delta)
1439 return;
1440 data_delta = 0;
1441 }
1442 percpu_counter_add_batch(&mp->m_delalloc_blks, data_delta + ind_delta,
1443 XFS_DELALLOC_BATCH);
1444 }
1445
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.