Loading...
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 "scrub/stats.h"
38
39static DEFINE_MUTEX(xfs_uuid_table_mutex);
40static int xfs_uuid_table_size;
41static uuid_t *xfs_uuid_table;
42
43void
44xfs_uuid_table_free(void)
45{
46 if (xfs_uuid_table_size == 0)
47 return;
48 kfree(xfs_uuid_table);
49 xfs_uuid_table = NULL;
50 xfs_uuid_table_size = 0;
51}
52
53/*
54 * See if the UUID is unique among mounted XFS filesystems.
55 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
56 */
57STATIC int
58xfs_uuid_mount(
59 struct xfs_mount *mp)
60{
61 uuid_t *uuid = &mp->m_sb.sb_uuid;
62 int hole, i;
63
64 /* Publish UUID in struct super_block */
65 super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
66
67 if (xfs_has_nouuid(mp))
68 return 0;
69
70 if (uuid_is_null(uuid)) {
71 xfs_warn(mp, "Filesystem has null UUID - can't mount");
72 return -EINVAL;
73 }
74
75 mutex_lock(&xfs_uuid_table_mutex);
76 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
77 if (uuid_is_null(&xfs_uuid_table[i])) {
78 hole = i;
79 continue;
80 }
81 if (uuid_equal(uuid, &xfs_uuid_table[i]))
82 goto out_duplicate;
83 }
84
85 if (hole < 0) {
86 xfs_uuid_table = krealloc(xfs_uuid_table,
87 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
88 GFP_KERNEL | __GFP_NOFAIL);
89 hole = xfs_uuid_table_size++;
90 }
91 xfs_uuid_table[hole] = *uuid;
92 mutex_unlock(&xfs_uuid_table_mutex);
93
94 return 0;
95
96 out_duplicate:
97 mutex_unlock(&xfs_uuid_table_mutex);
98 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
99 return -EINVAL;
100}
101
102STATIC void
103xfs_uuid_unmount(
104 struct xfs_mount *mp)
105{
106 uuid_t *uuid = &mp->m_sb.sb_uuid;
107 int i;
108
109 if (xfs_has_nouuid(mp))
110 return;
111
112 mutex_lock(&xfs_uuid_table_mutex);
113 for (i = 0; i < xfs_uuid_table_size; i++) {
114 if (uuid_is_null(&xfs_uuid_table[i]))
115 continue;
116 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
117 continue;
118 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
119 break;
120 }
121 ASSERT(i < xfs_uuid_table_size);
122 mutex_unlock(&xfs_uuid_table_mutex);
123}
124
125/*
126 * Check size of device based on the (data/realtime) block count.
127 * Note: this check is used by the growfs code as well as mount.
128 */
129int
130xfs_sb_validate_fsb_count(
131 xfs_sb_t *sbp,
132 uint64_t nblocks)
133{
134 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
135 ASSERT(sbp->sb_blocklog >= BBSHIFT);
136
137 /* Limited by ULONG_MAX of page cache index */
138 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
139 return -EFBIG;
140 return 0;
141}
142
143/*
144 * xfs_readsb
145 *
146 * Does the initial read of the superblock.
147 */
148int
149xfs_readsb(
150 struct xfs_mount *mp,
151 int flags)
152{
153 unsigned int sector_size;
154 struct xfs_buf *bp;
155 struct xfs_sb *sbp = &mp->m_sb;
156 int error;
157 int loud = !(flags & XFS_MFSI_QUIET);
158 const struct xfs_buf_ops *buf_ops;
159
160 ASSERT(mp->m_sb_bp == NULL);
161 ASSERT(mp->m_ddev_targp != NULL);
162
163 /*
164 * For the initial read, we must guess at the sector
165 * size based on the block device. It's enough to
166 * get the sb_sectsize out of the superblock and
167 * then reread with the proper length.
168 * We don't verify it yet, because it may not be complete.
169 */
170 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
171 buf_ops = NULL;
172
173 /*
174 * Allocate a (locked) buffer to hold the superblock. This will be kept
175 * around at all times to optimize access to the superblock. Therefore,
176 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
177 * elevated.
178 */
179reread:
180 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
181 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
182 buf_ops);
183 if (error) {
184 if (loud)
185 xfs_warn(mp, "SB validate failed with error %d.", error);
186 /* bad CRC means corrupted metadata */
187 if (error == -EFSBADCRC)
188 error = -EFSCORRUPTED;
189 return error;
190 }
191
192 /*
193 * Initialize the mount structure from the superblock.
194 */
195 xfs_sb_from_disk(sbp, bp->b_addr);
196
197 /*
198 * If we haven't validated the superblock, do so now before we try
199 * to check the sector size and reread the superblock appropriately.
200 */
201 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
202 if (loud)
203 xfs_warn(mp, "Invalid superblock magic number");
204 error = -EINVAL;
205 goto release_buf;
206 }
207
208 /*
209 * We must be able to do sector-sized and sector-aligned IO.
210 */
211 if (sector_size > sbp->sb_sectsize) {
212 if (loud)
213 xfs_warn(mp, "device supports %u byte sectors (not %u)",
214 sector_size, sbp->sb_sectsize);
215 error = -ENOSYS;
216 goto release_buf;
217 }
218
219 if (buf_ops == NULL) {
220 /*
221 * Re-read the superblock so the buffer is correctly sized,
222 * and properly verified.
223 */
224 xfs_buf_relse(bp);
225 sector_size = sbp->sb_sectsize;
226 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
227 goto reread;
228 }
229
230 mp->m_features |= xfs_sb_version_to_features(sbp);
231 xfs_reinit_percpu_counters(mp);
232
233 /* no need to be quiet anymore, so reset the buf ops */
234 bp->b_ops = &xfs_sb_buf_ops;
235
236 mp->m_sb_bp = bp;
237 xfs_buf_unlock(bp);
238 return 0;
239
240release_buf:
241 xfs_buf_relse(bp);
242 return error;
243}
244
245/*
246 * If the sunit/swidth change would move the precomputed root inode value, we
247 * must reject the ondisk change because repair will stumble over that.
248 * However, we allow the mount to proceed because we never rejected this
249 * combination before. Returns true to update the sb, false otherwise.
250 */
251static inline int
252xfs_check_new_dalign(
253 struct xfs_mount *mp,
254 int new_dalign,
255 bool *update_sb)
256{
257 struct xfs_sb *sbp = &mp->m_sb;
258 xfs_ino_t calc_ino;
259
260 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
261 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
262
263 if (sbp->sb_rootino == calc_ino) {
264 *update_sb = true;
265 return 0;
266 }
267
268 xfs_warn(mp,
269"Cannot change stripe alignment; would require moving root inode.");
270
271 /*
272 * XXX: Next time we add a new incompat feature, this should start
273 * returning -EINVAL to fail the mount. Until then, spit out a warning
274 * that we're ignoring the administrator's instructions.
275 */
276 xfs_warn(mp, "Skipping superblock stripe alignment update.");
277 *update_sb = false;
278 return 0;
279}
280
281/*
282 * If we were provided with new sunit/swidth values as mount options, make sure
283 * that they pass basic alignment and superblock feature checks, and convert
284 * them into the same units (FSB) that everything else expects. This step
285 * /must/ be done before computing the inode geometry.
286 */
287STATIC int
288xfs_validate_new_dalign(
289 struct xfs_mount *mp)
290{
291 if (mp->m_dalign == 0)
292 return 0;
293
294 /*
295 * If stripe unit and stripe width are not multiples
296 * of the fs blocksize turn off alignment.
297 */
298 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
299 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
300 xfs_warn(mp,
301 "alignment check failed: sunit/swidth vs. blocksize(%d)",
302 mp->m_sb.sb_blocksize);
303 return -EINVAL;
304 }
305
306 /*
307 * Convert the stripe unit and width to FSBs.
308 */
309 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
310 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
311 xfs_warn(mp,
312 "alignment check failed: sunit/swidth vs. agsize(%d)",
313 mp->m_sb.sb_agblocks);
314 return -EINVAL;
315 }
316
317 if (!mp->m_dalign) {
318 xfs_warn(mp,
319 "alignment check failed: sunit(%d) less than bsize(%d)",
320 mp->m_dalign, mp->m_sb.sb_blocksize);
321 return -EINVAL;
322 }
323
324 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
325
326 if (!xfs_has_dalign(mp)) {
327 xfs_warn(mp,
328"cannot change alignment: superblock does not support data alignment");
329 return -EINVAL;
330 }
331
332 return 0;
333}
334
335/* Update alignment values based on mount options and sb values. */
336STATIC int
337xfs_update_alignment(
338 struct xfs_mount *mp)
339{
340 struct xfs_sb *sbp = &mp->m_sb;
341
342 if (mp->m_dalign) {
343 bool update_sb;
344 int error;
345
346 if (sbp->sb_unit == mp->m_dalign &&
347 sbp->sb_width == mp->m_swidth)
348 return 0;
349
350 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
351 if (error || !update_sb)
352 return error;
353
354 sbp->sb_unit = mp->m_dalign;
355 sbp->sb_width = mp->m_swidth;
356 mp->m_update_sb = true;
357 } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
358 mp->m_dalign = sbp->sb_unit;
359 mp->m_swidth = sbp->sb_width;
360 }
361
362 return 0;
363}
364
365/*
366 * precalculate the low space thresholds for dynamic speculative preallocation.
367 */
368void
369xfs_set_low_space_thresholds(
370 struct xfs_mount *mp)
371{
372 uint64_t dblocks = mp->m_sb.sb_dblocks;
373 uint64_t rtexts = mp->m_sb.sb_rextents;
374 int i;
375
376 do_div(dblocks, 100);
377 do_div(rtexts, 100);
378
379 for (i = 0; i < XFS_LOWSP_MAX; i++) {
380 mp->m_low_space[i] = dblocks * (i + 1);
381 mp->m_low_rtexts[i] = rtexts * (i + 1);
382 }
383}
384
385/*
386 * Check that the data (and log if separate) is an ok size.
387 */
388STATIC int
389xfs_check_sizes(
390 struct xfs_mount *mp)
391{
392 struct xfs_buf *bp;
393 xfs_daddr_t d;
394 int error;
395
396 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
397 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
398 xfs_warn(mp, "filesystem size mismatch detected");
399 return -EFBIG;
400 }
401 error = xfs_buf_read_uncached(mp->m_ddev_targp,
402 d - XFS_FSS_TO_BB(mp, 1),
403 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
404 if (error) {
405 xfs_warn(mp, "last sector read failed");
406 return error;
407 }
408 xfs_buf_relse(bp);
409
410 if (mp->m_logdev_targp == mp->m_ddev_targp)
411 return 0;
412
413 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
414 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
415 xfs_warn(mp, "log size mismatch detected");
416 return -EFBIG;
417 }
418 error = xfs_buf_read_uncached(mp->m_logdev_targp,
419 d - XFS_FSB_TO_BB(mp, 1),
420 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
421 if (error) {
422 xfs_warn(mp, "log device read failed");
423 return error;
424 }
425 xfs_buf_relse(bp);
426 return 0;
427}
428
429/*
430 * Clear the quotaflags in memory and in the superblock.
431 */
432int
433xfs_mount_reset_sbqflags(
434 struct xfs_mount *mp)
435{
436 mp->m_qflags = 0;
437
438 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
439 if (mp->m_sb.sb_qflags == 0)
440 return 0;
441 spin_lock(&mp->m_sb_lock);
442 mp->m_sb.sb_qflags = 0;
443 spin_unlock(&mp->m_sb_lock);
444
445 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
446 return 0;
447
448 return xfs_sync_sb(mp, false);
449}
450
451uint64_t
452xfs_default_resblks(xfs_mount_t *mp)
453{
454 uint64_t resblks;
455
456 /*
457 * We default to 5% or 8192 fsbs of space reserved, whichever is
458 * smaller. This is intended to cover concurrent allocation
459 * transactions when we initially hit enospc. These each require a 4
460 * block reservation. Hence by default we cover roughly 2000 concurrent
461 * allocation reservations.
462 */
463 resblks = mp->m_sb.sb_dblocks;
464 do_div(resblks, 20);
465 resblks = min_t(uint64_t, resblks, 8192);
466 return resblks;
467}
468
469/* Ensure the summary counts are correct. */
470STATIC int
471xfs_check_summary_counts(
472 struct xfs_mount *mp)
473{
474 int error = 0;
475
476 /*
477 * The AG0 superblock verifier rejects in-progress filesystems,
478 * so we should never see the flag set this far into mounting.
479 */
480 if (mp->m_sb.sb_inprogress) {
481 xfs_err(mp, "sb_inprogress set after log recovery??");
482 WARN_ON(1);
483 return -EFSCORRUPTED;
484 }
485
486 /*
487 * Now the log is mounted, we know if it was an unclean shutdown or
488 * not. If it was, with the first phase of recovery has completed, we
489 * have consistent AG blocks on disk. We have not recovered EFIs yet,
490 * but they are recovered transactionally in the second recovery phase
491 * later.
492 *
493 * If the log was clean when we mounted, we can check the summary
494 * counters. If any of them are obviously incorrect, we can recompute
495 * them from the AGF headers in the next step.
496 */
497 if (xfs_is_clean(mp) &&
498 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
499 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
500 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
501 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
502
503 /*
504 * We can safely re-initialise incore superblock counters from the
505 * per-ag data. These may not be correct if the filesystem was not
506 * cleanly unmounted, so we waited for recovery to finish before doing
507 * this.
508 *
509 * If the filesystem was cleanly unmounted or the previous check did
510 * not flag anything weird, then we can trust the values in the
511 * superblock to be correct and we don't need to do anything here.
512 * Otherwise, recalculate the summary counters.
513 */
514 if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
515 xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
516 error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
517 if (error)
518 return error;
519 }
520
521 /*
522 * Older kernels misused sb_frextents to reflect both incore
523 * reservations made by running transactions and the actual count of
524 * free rt extents in the ondisk metadata. Transactions committed
525 * during runtime can therefore contain a superblock update that
526 * undercounts the number of free rt extents tracked in the rt bitmap.
527 * A clean unmount record will have the correct frextents value since
528 * there can be no other transactions running at that point.
529 *
530 * If we're mounting the rt volume after recovering the log, recompute
531 * frextents from the rtbitmap file to fix the inconsistency.
532 */
533 if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
534 error = xfs_rtalloc_reinit_frextents(mp);
535 if (error)
536 return error;
537 }
538
539 return 0;
540}
541
542static void
543xfs_unmount_check(
544 struct xfs_mount *mp)
545{
546 if (xfs_is_shutdown(mp))
547 return;
548
549 if (percpu_counter_sum(&mp->m_ifree) >
550 percpu_counter_sum(&mp->m_icount)) {
551 xfs_alert(mp, "ifree/icount mismatch at unmount");
552 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
553 }
554}
555
556/*
557 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
558 * internal inode structures can be sitting in the CIL and AIL at this point,
559 * so we need to unpin them, write them back and/or reclaim them before unmount
560 * can proceed. In other words, callers are required to have inactivated all
561 * inodes.
562 *
563 * An inode cluster that has been freed can have its buffer still pinned in
564 * memory because the transaction is still sitting in a iclog. The stale inodes
565 * on that buffer will be pinned to the buffer until the transaction hits the
566 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
567 * may never see the pinned buffer, so nothing will push out the iclog and
568 * unpin the buffer.
569 *
570 * Hence we need to force the log to unpin everything first. However, log
571 * forces don't wait for the discards they issue to complete, so we have to
572 * explicitly wait for them to complete here as well.
573 *
574 * Then we can tell the world we are unmounting so that error handling knows
575 * that the filesystem is going away and we should error out anything that we
576 * have been retrying in the background. This will prevent never-ending
577 * retries in AIL pushing from hanging the unmount.
578 *
579 * Finally, we can push the AIL to clean all the remaining dirty objects, then
580 * reclaim the remaining inodes that are still in memory at this point in time.
581 */
582static void
583xfs_unmount_flush_inodes(
584 struct xfs_mount *mp)
585{
586 xfs_log_force(mp, XFS_LOG_SYNC);
587 xfs_extent_busy_wait_all(mp);
588 flush_workqueue(xfs_discard_wq);
589
590 set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate);
591
592 xfs_ail_push_all_sync(mp->m_ail);
593 xfs_inodegc_stop(mp);
594 cancel_delayed_work_sync(&mp->m_reclaim_work);
595 xfs_reclaim_inodes(mp);
596 xfs_health_unmount(mp);
597}
598
599static void
600xfs_mount_setup_inode_geom(
601 struct xfs_mount *mp)
602{
603 struct xfs_ino_geometry *igeo = M_IGEO(mp);
604
605 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
606 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
607
608 xfs_ialloc_setup_geometry(mp);
609}
610
611/* Compute maximum possible height for per-AG btree types for this fs. */
612static inline void
613xfs_agbtree_compute_maxlevels(
614 struct xfs_mount *mp)
615{
616 unsigned int levels;
617
618 levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
619 levels = max(levels, mp->m_rmap_maxlevels);
620 mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
621}
622
623/*
624 * This function does the following on an initial mount of a file system:
625 * - reads the superblock from disk and init the mount struct
626 * - if we're a 32-bit kernel, do a size check on the superblock
627 * so we don't mount terabyte filesystems
628 * - init mount struct realtime fields
629 * - allocate inode hash table for fs
630 * - init directory manager
631 * - perform recovery and init the log manager
632 */
633int
634xfs_mountfs(
635 struct xfs_mount *mp)
636{
637 struct xfs_sb *sbp = &(mp->m_sb);
638 struct xfs_inode *rip;
639 struct xfs_ino_geometry *igeo = M_IGEO(mp);
640 uint quotamount = 0;
641 uint quotaflags = 0;
642 int error = 0;
643
644 xfs_sb_mount_common(mp, sbp);
645
646 /*
647 * Check for a mismatched features2 values. Older kernels read & wrote
648 * into the wrong sb offset for sb_features2 on some platforms due to
649 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
650 * which made older superblock reading/writing routines swap it as a
651 * 64-bit value.
652 *
653 * For backwards compatibility, we make both slots equal.
654 *
655 * If we detect a mismatched field, we OR the set bits into the existing
656 * features2 field in case it has already been modified; we don't want
657 * to lose any features. We then update the bad location with the ORed
658 * value so that older kernels will see any features2 flags. The
659 * superblock writeback code ensures the new sb_features2 is copied to
660 * sb_bad_features2 before it is logged or written to disk.
661 */
662 if (xfs_sb_has_mismatched_features2(sbp)) {
663 xfs_warn(mp, "correcting sb_features alignment problem");
664 sbp->sb_features2 |= sbp->sb_bad_features2;
665 mp->m_update_sb = true;
666 }
667
668
669 /* always use v2 inodes by default now */
670 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
671 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
672 mp->m_features |= XFS_FEAT_NLINK;
673 mp->m_update_sb = true;
674 }
675
676 /*
677 * If we were given new sunit/swidth options, do some basic validation
678 * checks and convert the incore dalign and swidth values to the
679 * same units (FSB) that everything else uses. This /must/ happen
680 * before computing the inode geometry.
681 */
682 error = xfs_validate_new_dalign(mp);
683 if (error)
684 goto out;
685
686 xfs_alloc_compute_maxlevels(mp);
687 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
688 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
689 xfs_mount_setup_inode_geom(mp);
690 xfs_rmapbt_compute_maxlevels(mp);
691 xfs_refcountbt_compute_maxlevels(mp);
692
693 xfs_agbtree_compute_maxlevels(mp);
694
695 /*
696 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
697 * is NOT aligned turn off m_dalign since allocator alignment is within
698 * an ag, therefore ag has to be aligned at stripe boundary. Note that
699 * we must compute the free space and rmap btree geometry before doing
700 * this.
701 */
702 error = xfs_update_alignment(mp);
703 if (error)
704 goto out;
705
706 /* enable fail_at_unmount as default */
707 mp->m_fail_unmount = true;
708
709 super_set_sysfs_name_id(mp->m_super);
710
711 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
712 NULL, mp->m_super->s_id);
713 if (error)
714 goto out;
715
716 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
717 &mp->m_kobj, "stats");
718 if (error)
719 goto out_remove_sysfs;
720
721 xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
722
723 error = xfs_error_sysfs_init(mp);
724 if (error)
725 goto out_remove_scrub_stats;
726
727 error = xfs_errortag_init(mp);
728 if (error)
729 goto out_remove_error_sysfs;
730
731 error = xfs_uuid_mount(mp);
732 if (error)
733 goto out_remove_errortag;
734
735 /*
736 * Update the preferred write size based on the information from the
737 * on-disk superblock.
738 */
739 mp->m_allocsize_log =
740 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
741 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
742
743 /* set the low space thresholds for dynamic preallocation */
744 xfs_set_low_space_thresholds(mp);
745
746 /*
747 * If enabled, sparse inode chunk alignment is expected to match the
748 * cluster size. Full inode chunk alignment must match the chunk size,
749 * but that is checked on sb read verification...
750 */
751 if (xfs_has_sparseinodes(mp) &&
752 mp->m_sb.sb_spino_align !=
753 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
754 xfs_warn(mp,
755 "Sparse inode block alignment (%u) must match cluster size (%llu).",
756 mp->m_sb.sb_spino_align,
757 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
758 error = -EINVAL;
759 goto out_remove_uuid;
760 }
761
762 /*
763 * Check that the data (and log if separate) is an ok size.
764 */
765 error = xfs_check_sizes(mp);
766 if (error)
767 goto out_remove_uuid;
768
769 /*
770 * Initialize realtime fields in the mount structure
771 */
772 error = xfs_rtmount_init(mp);
773 if (error) {
774 xfs_warn(mp, "RT mount failed");
775 goto out_remove_uuid;
776 }
777
778 /*
779 * Copies the low order bits of the timestamp and the randomly
780 * set "sequence" number out of a UUID.
781 */
782 mp->m_fixedfsid[0] =
783 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
784 get_unaligned_be16(&sbp->sb_uuid.b[4]);
785 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
786
787 error = xfs_da_mount(mp);
788 if (error) {
789 xfs_warn(mp, "Failed dir/attr init: %d", error);
790 goto out_remove_uuid;
791 }
792
793 /*
794 * Initialize the precomputed transaction reservations values.
795 */
796 xfs_trans_init(mp);
797
798 /*
799 * Allocate and initialize the per-ag data.
800 */
801 error = xfs_initialize_perag(mp, sbp->sb_agcount, mp->m_sb.sb_dblocks,
802 &mp->m_maxagi);
803 if (error) {
804 xfs_warn(mp, "Failed per-ag init: %d", error);
805 goto out_free_dir;
806 }
807
808 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
809 xfs_warn(mp, "no log defined");
810 error = -EFSCORRUPTED;
811 goto out_free_perag;
812 }
813
814 error = xfs_inodegc_register_shrinker(mp);
815 if (error)
816 goto out_fail_wait;
817
818 /*
819 * Log's mount-time initialization. The first part of recovery can place
820 * some items on the AIL, to be handled when recovery is finished or
821 * cancelled.
822 */
823 error = xfs_log_mount(mp, mp->m_logdev_targp,
824 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
825 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
826 if (error) {
827 xfs_warn(mp, "log mount failed");
828 goto out_inodegc_shrinker;
829 }
830
831 /* Enable background inode inactivation workers. */
832 xfs_inodegc_start(mp);
833 xfs_blockgc_start(mp);
834
835 /*
836 * Now that we've recovered any pending superblock feature bit
837 * additions, we can finish setting up the attr2 behaviour for the
838 * mount. The noattr2 option overrides the superblock flag, so only
839 * check the superblock feature flag if the mount option is not set.
840 */
841 if (xfs_has_noattr2(mp)) {
842 mp->m_features &= ~XFS_FEAT_ATTR2;
843 } else if (!xfs_has_attr2(mp) &&
844 (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) {
845 mp->m_features |= XFS_FEAT_ATTR2;
846 }
847
848 /*
849 * Get and sanity-check the root inode.
850 * Save the pointer to it in the mount structure.
851 */
852 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
853 XFS_ILOCK_EXCL, &rip);
854 if (error) {
855 xfs_warn(mp,
856 "Failed to read root inode 0x%llx, error %d",
857 sbp->sb_rootino, -error);
858 goto out_log_dealloc;
859 }
860
861 ASSERT(rip != NULL);
862
863 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
864 xfs_warn(mp, "corrupted root inode %llu: not a directory",
865 (unsigned long long)rip->i_ino);
866 xfs_iunlock(rip, XFS_ILOCK_EXCL);
867 error = -EFSCORRUPTED;
868 goto out_rele_rip;
869 }
870 mp->m_rootip = rip; /* save it */
871
872 xfs_iunlock(rip, XFS_ILOCK_EXCL);
873
874 /*
875 * Initialize realtime inode pointers in the mount structure
876 */
877 error = xfs_rtmount_inodes(mp);
878 if (error) {
879 /*
880 * Free up the root inode.
881 */
882 xfs_warn(mp, "failed to read RT inodes");
883 goto out_rele_rip;
884 }
885
886 /* Make sure the summary counts are ok. */
887 error = xfs_check_summary_counts(mp);
888 if (error)
889 goto out_rtunmount;
890
891 /*
892 * If this is a read-only mount defer the superblock updates until
893 * the next remount into writeable mode. Otherwise we would never
894 * perform the update e.g. for the root filesystem.
895 */
896 if (mp->m_update_sb && !xfs_is_readonly(mp)) {
897 error = xfs_sync_sb(mp, false);
898 if (error) {
899 xfs_warn(mp, "failed to write sb changes");
900 goto out_rtunmount;
901 }
902 }
903
904 /*
905 * Initialise the XFS quota management subsystem for this mount
906 */
907 if (XFS_IS_QUOTA_ON(mp)) {
908 error = xfs_qm_newmount(mp, "amount, "aflags);
909 if (error)
910 goto out_rtunmount;
911 } else {
912 /*
913 * If a file system had quotas running earlier, but decided to
914 * mount without -o uquota/pquota/gquota options, revoke the
915 * quotachecked license.
916 */
917 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
918 xfs_notice(mp, "resetting quota flags");
919 error = xfs_mount_reset_sbqflags(mp);
920 if (error)
921 goto out_rtunmount;
922 }
923 }
924
925 /*
926 * Finish recovering the file system. This part needed to be delayed
927 * until after the root and real-time bitmap inodes were consistently
928 * read in. Temporarily create per-AG space reservations for metadata
929 * btree shape changes because space freeing transactions (for inode
930 * inactivation) require the per-AG reservation in lieu of reserving
931 * blocks.
932 */
933 error = xfs_fs_reserve_ag_blocks(mp);
934 if (error && error == -ENOSPC)
935 xfs_warn(mp,
936 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
937 error = xfs_log_mount_finish(mp);
938 xfs_fs_unreserve_ag_blocks(mp);
939 if (error) {
940 xfs_warn(mp, "log mount finish failed");
941 goto out_rtunmount;
942 }
943
944 /*
945 * Now the log is fully replayed, we can transition to full read-only
946 * mode for read-only mounts. This will sync all the metadata and clean
947 * the log so that the recovery we just performed does not have to be
948 * replayed again on the next mount.
949 *
950 * We use the same quiesce mechanism as the rw->ro remount, as they are
951 * semantically identical operations.
952 */
953 if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
954 xfs_log_clean(mp);
955
956 /*
957 * Complete the quota initialisation, post-log-replay component.
958 */
959 if (quotamount) {
960 ASSERT(mp->m_qflags == 0);
961 mp->m_qflags = quotaflags;
962
963 xfs_qm_mount_quotas(mp);
964 }
965
966 /*
967 * Now we are mounted, reserve a small amount of unused space for
968 * privileged transactions. This is needed so that transaction
969 * space required for critical operations can dip into this pool
970 * when at ENOSPC. This is needed for operations like create with
971 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
972 * are not allowed to use this reserved space.
973 *
974 * This may drive us straight to ENOSPC on mount, but that implies
975 * we were already there on the last unmount. Warn if this occurs.
976 */
977 if (!xfs_is_readonly(mp)) {
978 error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
979 if (error)
980 xfs_warn(mp,
981 "Unable to allocate reserve blocks. Continuing without reserve pool.");
982
983 /* Reserve AG blocks for future btree expansion. */
984 error = xfs_fs_reserve_ag_blocks(mp);
985 if (error && error != -ENOSPC)
986 goto out_agresv;
987 }
988
989 return 0;
990
991 out_agresv:
992 xfs_fs_unreserve_ag_blocks(mp);
993 xfs_qm_unmount_quotas(mp);
994 out_rtunmount:
995 xfs_rtunmount_inodes(mp);
996 out_rele_rip:
997 xfs_irele(rip);
998 /* Clean out dquots that might be in memory after quotacheck. */
999 xfs_qm_unmount(mp);
1000
1001 /*
1002 * Inactivate all inodes that might still be in memory after a log
1003 * intent recovery failure so that reclaim can free them. Metadata
1004 * inodes and the root directory shouldn't need inactivation, but the
1005 * mount failed for some reason, so pull down all the state and flee.
1006 */
1007 xfs_inodegc_flush(mp);
1008
1009 /*
1010 * Flush all inode reclamation work and flush the log.
1011 * We have to do this /after/ rtunmount and qm_unmount because those
1012 * two will have scheduled delayed reclaim for the rt/quota inodes.
1013 *
1014 * This is slightly different from the unmountfs call sequence
1015 * because we could be tearing down a partially set up mount. In
1016 * particular, if log_mount_finish fails we bail out without calling
1017 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1018 * quota inodes.
1019 */
1020 xfs_unmount_flush_inodes(mp);
1021 out_log_dealloc:
1022 xfs_log_mount_cancel(mp);
1023 out_inodegc_shrinker:
1024 shrinker_free(mp->m_inodegc_shrinker);
1025 out_fail_wait:
1026 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1027 xfs_buftarg_drain(mp->m_logdev_targp);
1028 xfs_buftarg_drain(mp->m_ddev_targp);
1029 out_free_perag:
1030 xfs_free_perag(mp);
1031 out_free_dir:
1032 xfs_da_unmount(mp);
1033 out_remove_uuid:
1034 xfs_uuid_unmount(mp);
1035 out_remove_errortag:
1036 xfs_errortag_del(mp);
1037 out_remove_error_sysfs:
1038 xfs_error_sysfs_del(mp);
1039 out_remove_scrub_stats:
1040 xchk_stats_unregister(mp->m_scrub_stats);
1041 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1042 out_remove_sysfs:
1043 xfs_sysfs_del(&mp->m_kobj);
1044 out:
1045 return error;
1046}
1047
1048/*
1049 * This flushes out the inodes,dquots and the superblock, unmounts the
1050 * log and makes sure that incore structures are freed.
1051 */
1052void
1053xfs_unmountfs(
1054 struct xfs_mount *mp)
1055{
1056 int error;
1057
1058 /*
1059 * Perform all on-disk metadata updates required to inactivate inodes
1060 * that the VFS evicted earlier in the unmount process. Freeing inodes
1061 * and discarding CoW fork preallocations can cause shape changes to
1062 * the free inode and refcount btrees, respectively, so we must finish
1063 * this before we discard the metadata space reservations. Metadata
1064 * inodes and the root directory do not require inactivation.
1065 */
1066 xfs_inodegc_flush(mp);
1067
1068 xfs_blockgc_stop(mp);
1069 xfs_fs_unreserve_ag_blocks(mp);
1070 xfs_qm_unmount_quotas(mp);
1071 xfs_rtunmount_inodes(mp);
1072 xfs_irele(mp->m_rootip);
1073
1074 xfs_unmount_flush_inodes(mp);
1075
1076 xfs_qm_unmount(mp);
1077
1078 /*
1079 * Unreserve any blocks we have so that when we unmount we don't account
1080 * the reserved free space as used. This is really only necessary for
1081 * lazy superblock counting because it trusts the incore superblock
1082 * counters to be absolutely correct on clean unmount.
1083 *
1084 * We don't bother correcting this elsewhere for lazy superblock
1085 * counting because on mount of an unclean filesystem we reconstruct the
1086 * correct counter value and this is irrelevant.
1087 *
1088 * For non-lazy counter filesystems, this doesn't matter at all because
1089 * we only every apply deltas to the superblock and hence the incore
1090 * value does not matter....
1091 */
1092 error = xfs_reserve_blocks(mp, 0);
1093 if (error)
1094 xfs_warn(mp, "Unable to free reserved block pool. "
1095 "Freespace may not be correct on next mount.");
1096 xfs_unmount_check(mp);
1097
1098 xfs_log_unmount(mp);
1099 xfs_da_unmount(mp);
1100 xfs_uuid_unmount(mp);
1101
1102#if defined(DEBUG)
1103 xfs_errortag_clearall(mp);
1104#endif
1105 shrinker_free(mp->m_inodegc_shrinker);
1106 xfs_free_perag(mp);
1107
1108 xfs_errortag_del(mp);
1109 xfs_error_sysfs_del(mp);
1110 xchk_stats_unregister(mp->m_scrub_stats);
1111 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1112 xfs_sysfs_del(&mp->m_kobj);
1113}
1114
1115/*
1116 * Determine whether modifications can proceed. The caller specifies the minimum
1117 * freeze level for which modifications should not be allowed. This allows
1118 * certain operations to proceed while the freeze sequence is in progress, if
1119 * necessary.
1120 */
1121bool
1122xfs_fs_writable(
1123 struct xfs_mount *mp,
1124 int level)
1125{
1126 ASSERT(level > SB_UNFROZEN);
1127 if ((mp->m_super->s_writers.frozen >= level) ||
1128 xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1129 return false;
1130
1131 return true;
1132}
1133
1134/* Adjust m_fdblocks or m_frextents. */
1135int
1136xfs_mod_freecounter(
1137 struct xfs_mount *mp,
1138 struct percpu_counter *counter,
1139 int64_t delta,
1140 bool rsvd)
1141{
1142 int64_t lcounter;
1143 long long res_used;
1144 uint64_t set_aside = 0;
1145 s32 batch;
1146 bool has_resv_pool;
1147
1148 ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1149 has_resv_pool = (counter == &mp->m_fdblocks);
1150 if (rsvd)
1151 ASSERT(has_resv_pool);
1152
1153 if (delta > 0) {
1154 /*
1155 * If the reserve pool is depleted, put blocks back into it
1156 * first. Most of the time the pool is full.
1157 */
1158 if (likely(!has_resv_pool ||
1159 mp->m_resblks == mp->m_resblks_avail)) {
1160 percpu_counter_add(counter, delta);
1161 return 0;
1162 }
1163
1164 spin_lock(&mp->m_sb_lock);
1165 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1166
1167 if (res_used > delta) {
1168 mp->m_resblks_avail += delta;
1169 } else {
1170 delta -= res_used;
1171 mp->m_resblks_avail = mp->m_resblks;
1172 percpu_counter_add(counter, delta);
1173 }
1174 spin_unlock(&mp->m_sb_lock);
1175 return 0;
1176 }
1177
1178 /*
1179 * Taking blocks away, need to be more accurate the closer we
1180 * are to zero.
1181 *
1182 * If the counter has a value of less than 2 * max batch size,
1183 * then make everything serialise as we are real close to
1184 * ENOSPC.
1185 */
1186 if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1187 XFS_FDBLOCKS_BATCH) < 0)
1188 batch = 1;
1189 else
1190 batch = XFS_FDBLOCKS_BATCH;
1191
1192 /*
1193 * Set aside allocbt blocks because these blocks are tracked as free
1194 * space but not available for allocation. Technically this means that a
1195 * single reservation cannot consume all remaining free space, but the
1196 * ratio of allocbt blocks to usable free blocks should be rather small.
1197 * The tradeoff without this is that filesystems that maintain high
1198 * perag block reservations can over reserve physical block availability
1199 * and fail physical allocation, which leads to much more serious
1200 * problems (i.e. transaction abort, pagecache discards, etc.) than
1201 * slightly premature -ENOSPC.
1202 */
1203 if (has_resv_pool)
1204 set_aside = xfs_fdblocks_unavailable(mp);
1205 percpu_counter_add_batch(counter, delta, batch);
1206 if (__percpu_counter_compare(counter, set_aside,
1207 XFS_FDBLOCKS_BATCH) >= 0) {
1208 /* we had space! */
1209 return 0;
1210 }
1211
1212 /*
1213 * lock up the sb for dipping into reserves before releasing the space
1214 * that took us to ENOSPC.
1215 */
1216 spin_lock(&mp->m_sb_lock);
1217 percpu_counter_add(counter, -delta);
1218 if (!has_resv_pool || !rsvd)
1219 goto fdblocks_enospc;
1220
1221 lcounter = (long long)mp->m_resblks_avail + delta;
1222 if (lcounter >= 0) {
1223 mp->m_resblks_avail = lcounter;
1224 spin_unlock(&mp->m_sb_lock);
1225 return 0;
1226 }
1227 xfs_warn_once(mp,
1228"Reserve blocks depleted! Consider increasing reserve pool size.");
1229
1230fdblocks_enospc:
1231 spin_unlock(&mp->m_sb_lock);
1232 return -ENOSPC;
1233}
1234
1235/*
1236 * Used to free the superblock along various error paths.
1237 */
1238void
1239xfs_freesb(
1240 struct xfs_mount *mp)
1241{
1242 struct xfs_buf *bp = mp->m_sb_bp;
1243
1244 xfs_buf_lock(bp);
1245 mp->m_sb_bp = NULL;
1246 xfs_buf_relse(bp);
1247}
1248
1249/*
1250 * If the underlying (data/log/rt) device is readonly, there are some
1251 * operations that cannot proceed.
1252 */
1253int
1254xfs_dev_is_read_only(
1255 struct xfs_mount *mp,
1256 char *message)
1257{
1258 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1259 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1260 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1261 xfs_notice(mp, "%s required on read-only device.", message);
1262 xfs_notice(mp, "write access unavailable, cannot proceed.");
1263 return -EROFS;
1264 }
1265 return 0;
1266}
1267
1268/* Force the summary counters to be recalculated at next mount. */
1269void
1270xfs_force_summary_recalc(
1271 struct xfs_mount *mp)
1272{
1273 if (!xfs_has_lazysbcount(mp))
1274 return;
1275
1276 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1277}
1278
1279/*
1280 * Enable a log incompat feature flag in the primary superblock. The caller
1281 * cannot have any other transactions in progress.
1282 */
1283int
1284xfs_add_incompat_log_feature(
1285 struct xfs_mount *mp,
1286 uint32_t feature)
1287{
1288 struct xfs_dsb *dsb;
1289 int error;
1290
1291 ASSERT(hweight32(feature) == 1);
1292 ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1293
1294 /*
1295 * Force the log to disk and kick the background AIL thread to reduce
1296 * the chances that the bwrite will stall waiting for the AIL to unpin
1297 * the primary superblock buffer. This isn't a data integrity
1298 * operation, so we don't need a synchronous push.
1299 */
1300 error = xfs_log_force(mp, XFS_LOG_SYNC);
1301 if (error)
1302 return error;
1303 xfs_ail_push_all(mp->m_ail);
1304
1305 /*
1306 * Lock the primary superblock buffer to serialize all callers that
1307 * are trying to set feature bits.
1308 */
1309 xfs_buf_lock(mp->m_sb_bp);
1310 xfs_buf_hold(mp->m_sb_bp);
1311
1312 if (xfs_is_shutdown(mp)) {
1313 error = -EIO;
1314 goto rele;
1315 }
1316
1317 if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1318 goto rele;
1319
1320 /*
1321 * Write the primary superblock to disk immediately, because we need
1322 * the log_incompat bit to be set in the primary super now to protect
1323 * the log items that we're going to commit later.
1324 */
1325 dsb = mp->m_sb_bp->b_addr;
1326 xfs_sb_to_disk(dsb, &mp->m_sb);
1327 dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1328 error = xfs_bwrite(mp->m_sb_bp);
1329 if (error)
1330 goto shutdown;
1331
1332 /*
1333 * Add the feature bits to the incore superblock before we unlock the
1334 * buffer.
1335 */
1336 xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1337 xfs_buf_relse(mp->m_sb_bp);
1338
1339 /* Log the superblock to disk. */
1340 return xfs_sync_sb(mp, false);
1341shutdown:
1342 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1343rele:
1344 xfs_buf_relse(mp->m_sb_bp);
1345 return error;
1346}
1347
1348/*
1349 * Clear all the log incompat flags from the superblock.
1350 *
1351 * The caller cannot be in a transaction, must ensure that the log does not
1352 * contain any log items protected by any log incompat bit, and must ensure
1353 * that there are no other threads that depend on the state of the log incompat
1354 * feature flags in the primary super.
1355 *
1356 * Returns true if the superblock is dirty.
1357 */
1358bool
1359xfs_clear_incompat_log_features(
1360 struct xfs_mount *mp)
1361{
1362 bool ret = false;
1363
1364 if (!xfs_has_crc(mp) ||
1365 !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1366 XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1367 xfs_is_shutdown(mp))
1368 return false;
1369
1370 /*
1371 * Update the incore superblock. We synchronize on the primary super
1372 * buffer lock to be consistent with the add function, though at least
1373 * in theory this shouldn't be necessary.
1374 */
1375 xfs_buf_lock(mp->m_sb_bp);
1376 xfs_buf_hold(mp->m_sb_bp);
1377
1378 if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1379 XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1380 xfs_sb_remove_incompat_log_features(&mp->m_sb);
1381 ret = true;
1382 }
1383
1384 xfs_buf_relse(mp->m_sb_bp);
1385 return ret;
1386}
1387
1388/*
1389 * Update the in-core delayed block counter.
1390 *
1391 * We prefer to update the counter without having to take a spinlock for every
1392 * counter update (i.e. batching). Each change to delayed allocation
1393 * reservations can change can easily exceed the default percpu counter
1394 * batching, so we use a larger batch factor here.
1395 *
1396 * Note that we don't currently have any callers requiring fast summation
1397 * (e.g. percpu_counter_read) so we can use a big batch value here.
1398 */
1399#define XFS_DELALLOC_BATCH (4096)
1400void
1401xfs_mod_delalloc(
1402 struct xfs_mount *mp,
1403 int64_t delta)
1404{
1405 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1406 XFS_DELALLOC_BATCH);
1407}
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 "xfs_metafile.h"
39#include "xfs_rtgroup.h"
40#include "scrub/stats.h"
41
42static DEFINE_MUTEX(xfs_uuid_table_mutex);
43static int xfs_uuid_table_size;
44static uuid_t *xfs_uuid_table;
45
46void
47xfs_uuid_table_free(void)
48{
49 if (xfs_uuid_table_size == 0)
50 return;
51 kfree(xfs_uuid_table);
52 xfs_uuid_table = NULL;
53 xfs_uuid_table_size = 0;
54}
55
56/*
57 * See if the UUID is unique among mounted XFS filesystems.
58 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
59 */
60STATIC int
61xfs_uuid_mount(
62 struct xfs_mount *mp)
63{
64 uuid_t *uuid = &mp->m_sb.sb_uuid;
65 int hole, i;
66
67 /* Publish UUID in struct super_block */
68 super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
69
70 if (xfs_has_nouuid(mp))
71 return 0;
72
73 if (uuid_is_null(uuid)) {
74 xfs_warn(mp, "Filesystem has null UUID - can't mount");
75 return -EINVAL;
76 }
77
78 mutex_lock(&xfs_uuid_table_mutex);
79 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
80 if (uuid_is_null(&xfs_uuid_table[i])) {
81 hole = i;
82 continue;
83 }
84 if (uuid_equal(uuid, &xfs_uuid_table[i]))
85 goto out_duplicate;
86 }
87
88 if (hole < 0) {
89 xfs_uuid_table = krealloc(xfs_uuid_table,
90 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
91 GFP_KERNEL | __GFP_NOFAIL);
92 hole = xfs_uuid_table_size++;
93 }
94 xfs_uuid_table[hole] = *uuid;
95 mutex_unlock(&xfs_uuid_table_mutex);
96
97 return 0;
98
99 out_duplicate:
100 mutex_unlock(&xfs_uuid_table_mutex);
101 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
102 return -EINVAL;
103}
104
105STATIC void
106xfs_uuid_unmount(
107 struct xfs_mount *mp)
108{
109 uuid_t *uuid = &mp->m_sb.sb_uuid;
110 int i;
111
112 if (xfs_has_nouuid(mp))
113 return;
114
115 mutex_lock(&xfs_uuid_table_mutex);
116 for (i = 0; i < xfs_uuid_table_size; i++) {
117 if (uuid_is_null(&xfs_uuid_table[i]))
118 continue;
119 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
120 continue;
121 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
122 break;
123 }
124 ASSERT(i < xfs_uuid_table_size);
125 mutex_unlock(&xfs_uuid_table_mutex);
126}
127
128/*
129 * Check size of device based on the (data/realtime) block count.
130 * Note: this check is used by the growfs code as well as mount.
131 */
132int
133xfs_sb_validate_fsb_count(
134 xfs_sb_t *sbp,
135 uint64_t nblocks)
136{
137 uint64_t max_bytes;
138
139 ASSERT(sbp->sb_blocklog >= BBSHIFT);
140
141 if (check_shl_overflow(nblocks, sbp->sb_blocklog, &max_bytes))
142 return -EFBIG;
143
144 /* Limited by ULONG_MAX of page cache index */
145 if (max_bytes >> PAGE_SHIFT > ULONG_MAX)
146 return -EFBIG;
147 return 0;
148}
149
150/*
151 * xfs_readsb
152 *
153 * Does the initial read of the superblock.
154 */
155int
156xfs_readsb(
157 struct xfs_mount *mp,
158 int flags)
159{
160 unsigned int sector_size;
161 struct xfs_buf *bp;
162 struct xfs_sb *sbp = &mp->m_sb;
163 int error;
164 int loud = !(flags & XFS_MFSI_QUIET);
165 const struct xfs_buf_ops *buf_ops;
166
167 ASSERT(mp->m_sb_bp == NULL);
168 ASSERT(mp->m_ddev_targp != NULL);
169
170 /*
171 * For the initial read, we must guess at the sector
172 * size based on the block device. It's enough to
173 * get the sb_sectsize out of the superblock and
174 * then reread with the proper length.
175 * We don't verify it yet, because it may not be complete.
176 */
177 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
178 buf_ops = NULL;
179
180 /*
181 * Allocate a (locked) buffer to hold the superblock. This will be kept
182 * around at all times to optimize access to the superblock. Therefore,
183 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
184 * elevated.
185 */
186reread:
187 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
188 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
189 buf_ops);
190 if (error) {
191 if (loud)
192 xfs_warn(mp, "SB validate failed with error %d.", error);
193 /* bad CRC means corrupted metadata */
194 if (error == -EFSBADCRC)
195 error = -EFSCORRUPTED;
196 return error;
197 }
198
199 /*
200 * Initialize the mount structure from the superblock.
201 */
202 xfs_sb_from_disk(sbp, bp->b_addr);
203
204 /*
205 * If we haven't validated the superblock, do so now before we try
206 * to check the sector size and reread the superblock appropriately.
207 */
208 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
209 if (loud)
210 xfs_warn(mp, "Invalid superblock magic number");
211 error = -EINVAL;
212 goto release_buf;
213 }
214
215 /*
216 * We must be able to do sector-sized and sector-aligned IO.
217 */
218 if (sector_size > sbp->sb_sectsize) {
219 if (loud)
220 xfs_warn(mp, "device supports %u byte sectors (not %u)",
221 sector_size, sbp->sb_sectsize);
222 error = -ENOSYS;
223 goto release_buf;
224 }
225
226 if (buf_ops == NULL) {
227 /*
228 * Re-read the superblock so the buffer is correctly sized,
229 * and properly verified.
230 */
231 xfs_buf_relse(bp);
232 sector_size = sbp->sb_sectsize;
233 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
234 goto reread;
235 }
236
237 mp->m_features |= xfs_sb_version_to_features(sbp);
238 xfs_reinit_percpu_counters(mp);
239
240 /*
241 * If logged xattrs are enabled after log recovery finishes, then set
242 * the opstate so that log recovery will work properly.
243 */
244 if (xfs_sb_version_haslogxattrs(&mp->m_sb))
245 xfs_set_using_logged_xattrs(mp);
246
247 /* no need to be quiet anymore, so reset the buf ops */
248 bp->b_ops = &xfs_sb_buf_ops;
249
250 mp->m_sb_bp = bp;
251 xfs_buf_unlock(bp);
252 return 0;
253
254release_buf:
255 xfs_buf_relse(bp);
256 return error;
257}
258
259/*
260 * If the sunit/swidth change would move the precomputed root inode value, we
261 * must reject the ondisk change because repair will stumble over that.
262 * However, we allow the mount to proceed because we never rejected this
263 * combination before. Returns true to update the sb, false otherwise.
264 */
265static inline int
266xfs_check_new_dalign(
267 struct xfs_mount *mp,
268 int new_dalign,
269 bool *update_sb)
270{
271 struct xfs_sb *sbp = &mp->m_sb;
272 xfs_ino_t calc_ino;
273
274 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
275 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
276
277 if (sbp->sb_rootino == calc_ino) {
278 *update_sb = true;
279 return 0;
280 }
281
282 xfs_warn(mp,
283"Cannot change stripe alignment; would require moving root inode.");
284
285 /*
286 * XXX: Next time we add a new incompat feature, this should start
287 * returning -EINVAL to fail the mount. Until then, spit out a warning
288 * that we're ignoring the administrator's instructions.
289 */
290 xfs_warn(mp, "Skipping superblock stripe alignment update.");
291 *update_sb = false;
292 return 0;
293}
294
295/*
296 * If we were provided with new sunit/swidth values as mount options, make sure
297 * that they pass basic alignment and superblock feature checks, and convert
298 * them into the same units (FSB) that everything else expects. This step
299 * /must/ be done before computing the inode geometry.
300 */
301STATIC int
302xfs_validate_new_dalign(
303 struct xfs_mount *mp)
304{
305 if (mp->m_dalign == 0)
306 return 0;
307
308 /*
309 * If stripe unit and stripe width are not multiples
310 * of the fs blocksize turn off alignment.
311 */
312 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
313 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
314 xfs_warn(mp,
315 "alignment check failed: sunit/swidth vs. blocksize(%d)",
316 mp->m_sb.sb_blocksize);
317 return -EINVAL;
318 }
319
320 /*
321 * Convert the stripe unit and width to FSBs.
322 */
323 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
324 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
325 xfs_warn(mp,
326 "alignment check failed: sunit/swidth vs. agsize(%d)",
327 mp->m_sb.sb_agblocks);
328 return -EINVAL;
329 }
330
331 if (!mp->m_dalign) {
332 xfs_warn(mp,
333 "alignment check failed: sunit(%d) less than bsize(%d)",
334 mp->m_dalign, mp->m_sb.sb_blocksize);
335 return -EINVAL;
336 }
337
338 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
339
340 if (!xfs_has_dalign(mp)) {
341 xfs_warn(mp,
342"cannot change alignment: superblock does not support data alignment");
343 return -EINVAL;
344 }
345
346 return 0;
347}
348
349/* Update alignment values based on mount options and sb values. */
350STATIC int
351xfs_update_alignment(
352 struct xfs_mount *mp)
353{
354 struct xfs_sb *sbp = &mp->m_sb;
355
356 if (mp->m_dalign) {
357 bool update_sb;
358 int error;
359
360 if (sbp->sb_unit == mp->m_dalign &&
361 sbp->sb_width == mp->m_swidth)
362 return 0;
363
364 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
365 if (error || !update_sb)
366 return error;
367
368 sbp->sb_unit = mp->m_dalign;
369 sbp->sb_width = mp->m_swidth;
370 mp->m_update_sb = true;
371 } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
372 mp->m_dalign = sbp->sb_unit;
373 mp->m_swidth = sbp->sb_width;
374 }
375
376 return 0;
377}
378
379/*
380 * precalculate the low space thresholds for dynamic speculative preallocation.
381 */
382void
383xfs_set_low_space_thresholds(
384 struct xfs_mount *mp)
385{
386 uint64_t dblocks = mp->m_sb.sb_dblocks;
387 uint64_t rtexts = mp->m_sb.sb_rextents;
388 int i;
389
390 do_div(dblocks, 100);
391 do_div(rtexts, 100);
392
393 for (i = 0; i < XFS_LOWSP_MAX; i++) {
394 mp->m_low_space[i] = dblocks * (i + 1);
395 mp->m_low_rtexts[i] = rtexts * (i + 1);
396 }
397}
398
399/*
400 * Check that the data (and log if separate) is an ok size.
401 */
402STATIC int
403xfs_check_sizes(
404 struct xfs_mount *mp)
405{
406 struct xfs_buf *bp;
407 xfs_daddr_t d;
408 int error;
409
410 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
411 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
412 xfs_warn(mp, "filesystem size mismatch detected");
413 return -EFBIG;
414 }
415 error = xfs_buf_read_uncached(mp->m_ddev_targp,
416 d - XFS_FSS_TO_BB(mp, 1),
417 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
418 if (error) {
419 xfs_warn(mp, "last sector read failed");
420 return error;
421 }
422 xfs_buf_relse(bp);
423
424 if (mp->m_logdev_targp == mp->m_ddev_targp)
425 return 0;
426
427 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
428 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
429 xfs_warn(mp, "log size mismatch detected");
430 return -EFBIG;
431 }
432 error = xfs_buf_read_uncached(mp->m_logdev_targp,
433 d - XFS_FSB_TO_BB(mp, 1),
434 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
435 if (error) {
436 xfs_warn(mp, "log device read failed");
437 return error;
438 }
439 xfs_buf_relse(bp);
440 return 0;
441}
442
443/*
444 * Clear the quotaflags in memory and in the superblock.
445 */
446int
447xfs_mount_reset_sbqflags(
448 struct xfs_mount *mp)
449{
450 mp->m_qflags = 0;
451
452 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
453 if (mp->m_sb.sb_qflags == 0)
454 return 0;
455 spin_lock(&mp->m_sb_lock);
456 mp->m_sb.sb_qflags = 0;
457 spin_unlock(&mp->m_sb_lock);
458
459 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
460 return 0;
461
462 return xfs_sync_sb(mp, false);
463}
464
465uint64_t
466xfs_default_resblks(xfs_mount_t *mp)
467{
468 uint64_t resblks;
469
470 /*
471 * We default to 5% or 8192 fsbs of space reserved, whichever is
472 * smaller. This is intended to cover concurrent allocation
473 * transactions when we initially hit enospc. These each require a 4
474 * block reservation. Hence by default we cover roughly 2000 concurrent
475 * allocation reservations.
476 */
477 resblks = mp->m_sb.sb_dblocks;
478 do_div(resblks, 20);
479 resblks = min_t(uint64_t, resblks, 8192);
480 return resblks;
481}
482
483/* Ensure the summary counts are correct. */
484STATIC int
485xfs_check_summary_counts(
486 struct xfs_mount *mp)
487{
488 int error = 0;
489
490 /*
491 * The AG0 superblock verifier rejects in-progress filesystems,
492 * so we should never see the flag set this far into mounting.
493 */
494 if (mp->m_sb.sb_inprogress) {
495 xfs_err(mp, "sb_inprogress set after log recovery??");
496 WARN_ON(1);
497 return -EFSCORRUPTED;
498 }
499
500 /*
501 * Now the log is mounted, we know if it was an unclean shutdown or
502 * not. If it was, with the first phase of recovery has completed, we
503 * have consistent AG blocks on disk. We have not recovered EFIs yet,
504 * but they are recovered transactionally in the second recovery phase
505 * later.
506 *
507 * If the log was clean when we mounted, we can check the summary
508 * counters. If any of them are obviously incorrect, we can recompute
509 * them from the AGF headers in the next step.
510 */
511 if (xfs_is_clean(mp) &&
512 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
513 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
514 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
515 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
516
517 /*
518 * We can safely re-initialise incore superblock counters from the
519 * per-ag data. These may not be correct if the filesystem was not
520 * cleanly unmounted, so we waited for recovery to finish before doing
521 * this.
522 *
523 * If the filesystem was cleanly unmounted or the previous check did
524 * not flag anything weird, then we can trust the values in the
525 * superblock to be correct and we don't need to do anything here.
526 * Otherwise, recalculate the summary counters.
527 */
528 if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
529 xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
530 error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
531 if (error)
532 return error;
533 }
534
535 /*
536 * Older kernels misused sb_frextents to reflect both incore
537 * reservations made by running transactions and the actual count of
538 * free rt extents in the ondisk metadata. Transactions committed
539 * during runtime can therefore contain a superblock update that
540 * undercounts the number of free rt extents tracked in the rt bitmap.
541 * A clean unmount record will have the correct frextents value since
542 * there can be no other transactions running at that point.
543 *
544 * If we're mounting the rt volume after recovering the log, recompute
545 * frextents from the rtbitmap file to fix the inconsistency.
546 */
547 if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
548 error = xfs_rtalloc_reinit_frextents(mp);
549 if (error)
550 return error;
551 }
552
553 return 0;
554}
555
556static void
557xfs_unmount_check(
558 struct xfs_mount *mp)
559{
560 if (xfs_is_shutdown(mp))
561 return;
562
563 if (percpu_counter_sum(&mp->m_ifree) >
564 percpu_counter_sum(&mp->m_icount)) {
565 xfs_alert(mp, "ifree/icount mismatch at unmount");
566 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
567 }
568}
569
570/*
571 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
572 * internal inode structures can be sitting in the CIL and AIL at this point,
573 * so we need to unpin them, write them back and/or reclaim them before unmount
574 * can proceed. In other words, callers are required to have inactivated all
575 * inodes.
576 *
577 * An inode cluster that has been freed can have its buffer still pinned in
578 * memory because the transaction is still sitting in a iclog. The stale inodes
579 * on that buffer will be pinned to the buffer until the transaction hits the
580 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
581 * may never see the pinned buffer, so nothing will push out the iclog and
582 * unpin the buffer.
583 *
584 * Hence we need to force the log to unpin everything first. However, log
585 * forces don't wait for the discards they issue to complete, so we have to
586 * explicitly wait for them to complete here as well.
587 *
588 * Then we can tell the world we are unmounting so that error handling knows
589 * that the filesystem is going away and we should error out anything that we
590 * have been retrying in the background. This will prevent never-ending
591 * retries in AIL pushing from hanging the unmount.
592 *
593 * Finally, we can push the AIL to clean all the remaining dirty objects, then
594 * reclaim the remaining inodes that are still in memory at this point in time.
595 */
596static void
597xfs_unmount_flush_inodes(
598 struct xfs_mount *mp)
599{
600 xfs_log_force(mp, XFS_LOG_SYNC);
601 xfs_extent_busy_wait_all(mp);
602 flush_workqueue(xfs_discard_wq);
603
604 xfs_set_unmounting(mp);
605
606 xfs_ail_push_all_sync(mp->m_ail);
607 xfs_inodegc_stop(mp);
608 cancel_delayed_work_sync(&mp->m_reclaim_work);
609 xfs_reclaim_inodes(mp);
610 xfs_health_unmount(mp);
611}
612
613static void
614xfs_mount_setup_inode_geom(
615 struct xfs_mount *mp)
616{
617 struct xfs_ino_geometry *igeo = M_IGEO(mp);
618
619 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
620 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
621
622 xfs_ialloc_setup_geometry(mp);
623}
624
625/* Mount the metadata directory tree root. */
626STATIC int
627xfs_mount_setup_metadir(
628 struct xfs_mount *mp)
629{
630 int error;
631
632 /* Load the metadata directory root inode into memory. */
633 error = xfs_metafile_iget(mp, mp->m_sb.sb_metadirino, XFS_METAFILE_DIR,
634 &mp->m_metadirip);
635 if (error)
636 xfs_warn(mp, "Failed to load metadir root directory, error %d",
637 error);
638 return error;
639}
640
641/* Compute maximum possible height for per-AG btree types for this fs. */
642static inline void
643xfs_agbtree_compute_maxlevels(
644 struct xfs_mount *mp)
645{
646 unsigned int levels;
647
648 levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
649 levels = max(levels, mp->m_rmap_maxlevels);
650 mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
651}
652
653/*
654 * This function does the following on an initial mount of a file system:
655 * - reads the superblock from disk and init the mount struct
656 * - if we're a 32-bit kernel, do a size check on the superblock
657 * so we don't mount terabyte filesystems
658 * - init mount struct realtime fields
659 * - allocate inode hash table for fs
660 * - init directory manager
661 * - perform recovery and init the log manager
662 */
663int
664xfs_mountfs(
665 struct xfs_mount *mp)
666{
667 struct xfs_sb *sbp = &(mp->m_sb);
668 struct xfs_inode *rip;
669 struct xfs_ino_geometry *igeo = M_IGEO(mp);
670 uint quotamount = 0;
671 uint quotaflags = 0;
672 int error = 0;
673
674 xfs_sb_mount_common(mp, sbp);
675
676 /*
677 * Check for a mismatched features2 values. Older kernels read & wrote
678 * into the wrong sb offset for sb_features2 on some platforms due to
679 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
680 * which made older superblock reading/writing routines swap it as a
681 * 64-bit value.
682 *
683 * For backwards compatibility, we make both slots equal.
684 *
685 * If we detect a mismatched field, we OR the set bits into the existing
686 * features2 field in case it has already been modified; we don't want
687 * to lose any features. We then update the bad location with the ORed
688 * value so that older kernels will see any features2 flags. The
689 * superblock writeback code ensures the new sb_features2 is copied to
690 * sb_bad_features2 before it is logged or written to disk.
691 */
692 if (xfs_sb_has_mismatched_features2(sbp)) {
693 xfs_warn(mp, "correcting sb_features alignment problem");
694 sbp->sb_features2 |= sbp->sb_bad_features2;
695 mp->m_update_sb = true;
696 }
697
698
699 /* always use v2 inodes by default now */
700 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
701 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
702 mp->m_features |= XFS_FEAT_NLINK;
703 mp->m_update_sb = true;
704 }
705
706 /*
707 * If we were given new sunit/swidth options, do some basic validation
708 * checks and convert the incore dalign and swidth values to the
709 * same units (FSB) that everything else uses. This /must/ happen
710 * before computing the inode geometry.
711 */
712 error = xfs_validate_new_dalign(mp);
713 if (error)
714 goto out;
715
716 xfs_alloc_compute_maxlevels(mp);
717 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
718 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
719 xfs_mount_setup_inode_geom(mp);
720 xfs_rmapbt_compute_maxlevels(mp);
721 xfs_refcountbt_compute_maxlevels(mp);
722
723 xfs_agbtree_compute_maxlevels(mp);
724
725 /*
726 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
727 * is NOT aligned turn off m_dalign since allocator alignment is within
728 * an ag, therefore ag has to be aligned at stripe boundary. Note that
729 * we must compute the free space and rmap btree geometry before doing
730 * this.
731 */
732 error = xfs_update_alignment(mp);
733 if (error)
734 goto out;
735
736 /* enable fail_at_unmount as default */
737 mp->m_fail_unmount = true;
738
739 super_set_sysfs_name_id(mp->m_super);
740
741 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
742 NULL, mp->m_super->s_id);
743 if (error)
744 goto out;
745
746 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
747 &mp->m_kobj, "stats");
748 if (error)
749 goto out_remove_sysfs;
750
751 xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
752
753 error = xfs_error_sysfs_init(mp);
754 if (error)
755 goto out_remove_scrub_stats;
756
757 error = xfs_errortag_init(mp);
758 if (error)
759 goto out_remove_error_sysfs;
760
761 error = xfs_uuid_mount(mp);
762 if (error)
763 goto out_remove_errortag;
764
765 /*
766 * Update the preferred write size based on the information from the
767 * on-disk superblock.
768 */
769 mp->m_allocsize_log =
770 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
771 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
772
773 /* set the low space thresholds for dynamic preallocation */
774 xfs_set_low_space_thresholds(mp);
775
776 /*
777 * If enabled, sparse inode chunk alignment is expected to match the
778 * cluster size. Full inode chunk alignment must match the chunk size,
779 * but that is checked on sb read verification...
780 */
781 if (xfs_has_sparseinodes(mp) &&
782 mp->m_sb.sb_spino_align !=
783 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
784 xfs_warn(mp,
785 "Sparse inode block alignment (%u) must match cluster size (%llu).",
786 mp->m_sb.sb_spino_align,
787 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
788 error = -EINVAL;
789 goto out_remove_uuid;
790 }
791
792 /*
793 * Check that the data (and log if separate) is an ok size.
794 */
795 error = xfs_check_sizes(mp);
796 if (error)
797 goto out_remove_uuid;
798
799 /*
800 * Initialize realtime fields in the mount structure
801 */
802 error = xfs_rtmount_init(mp);
803 if (error) {
804 xfs_warn(mp, "RT mount failed");
805 goto out_remove_uuid;
806 }
807
808 /*
809 * Copies the low order bits of the timestamp and the randomly
810 * set "sequence" number out of a UUID.
811 */
812 mp->m_fixedfsid[0] =
813 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
814 get_unaligned_be16(&sbp->sb_uuid.b[4]);
815 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
816
817 error = xfs_da_mount(mp);
818 if (error) {
819 xfs_warn(mp, "Failed dir/attr init: %d", error);
820 goto out_remove_uuid;
821 }
822
823 /*
824 * Initialize the precomputed transaction reservations values.
825 */
826 xfs_trans_init(mp);
827
828 /*
829 * Allocate and initialize the per-ag data.
830 */
831 error = xfs_initialize_perag(mp, 0, sbp->sb_agcount,
832 mp->m_sb.sb_dblocks, &mp->m_maxagi);
833 if (error) {
834 xfs_warn(mp, "Failed per-ag init: %d", error);
835 goto out_free_dir;
836 }
837
838 error = xfs_initialize_rtgroups(mp, 0, sbp->sb_rgcount,
839 mp->m_sb.sb_rextents);
840 if (error) {
841 xfs_warn(mp, "Failed rtgroup init: %d", error);
842 goto out_free_perag;
843 }
844
845 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
846 xfs_warn(mp, "no log defined");
847 error = -EFSCORRUPTED;
848 goto out_free_rtgroup;
849 }
850
851 error = xfs_inodegc_register_shrinker(mp);
852 if (error)
853 goto out_fail_wait;
854
855 /*
856 * If we're resuming quota status, pick up the preliminary qflags from
857 * the ondisk superblock so that we know if we should recover dquots.
858 */
859 if (xfs_is_resuming_quotaon(mp))
860 xfs_qm_resume_quotaon(mp);
861
862 /*
863 * Log's mount-time initialization. The first part of recovery can place
864 * some items on the AIL, to be handled when recovery is finished or
865 * cancelled.
866 */
867 error = xfs_log_mount(mp, mp->m_logdev_targp,
868 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
869 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
870 if (error) {
871 xfs_warn(mp, "log mount failed");
872 goto out_inodegc_shrinker;
873 }
874
875 /*
876 * If we're resuming quota status and recovered the log, re-sample the
877 * qflags from the ondisk superblock now that we've recovered it, just
878 * in case someone shut down enforcement just before a crash.
879 */
880 if (xfs_clear_resuming_quotaon(mp) && xlog_recovery_needed(mp->m_log))
881 xfs_qm_resume_quotaon(mp);
882
883 /*
884 * If logged xattrs are still enabled after log recovery finishes, then
885 * they'll be available until unmount. Otherwise, turn them off.
886 */
887 if (xfs_sb_version_haslogxattrs(&mp->m_sb))
888 xfs_set_using_logged_xattrs(mp);
889 else
890 xfs_clear_using_logged_xattrs(mp);
891
892 /* Enable background inode inactivation workers. */
893 xfs_inodegc_start(mp);
894 xfs_blockgc_start(mp);
895
896 /*
897 * Now that we've recovered any pending superblock feature bit
898 * additions, we can finish setting up the attr2 behaviour for the
899 * mount. The noattr2 option overrides the superblock flag, so only
900 * check the superblock feature flag if the mount option is not set.
901 */
902 if (xfs_has_noattr2(mp)) {
903 mp->m_features &= ~XFS_FEAT_ATTR2;
904 } else if (!xfs_has_attr2(mp) &&
905 (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) {
906 mp->m_features |= XFS_FEAT_ATTR2;
907 }
908
909 if (xfs_has_metadir(mp)) {
910 error = xfs_mount_setup_metadir(mp);
911 if (error)
912 goto out_free_metadir;
913 }
914
915 /*
916 * Get and sanity-check the root inode.
917 * Save the pointer to it in the mount structure.
918 */
919 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
920 XFS_ILOCK_EXCL, &rip);
921 if (error) {
922 xfs_warn(mp,
923 "Failed to read root inode 0x%llx, error %d",
924 sbp->sb_rootino, -error);
925 goto out_free_metadir;
926 }
927
928 ASSERT(rip != NULL);
929
930 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
931 xfs_warn(mp, "corrupted root inode %llu: not a directory",
932 (unsigned long long)rip->i_ino);
933 xfs_iunlock(rip, XFS_ILOCK_EXCL);
934 error = -EFSCORRUPTED;
935 goto out_rele_rip;
936 }
937 mp->m_rootip = rip; /* save it */
938
939 xfs_iunlock(rip, XFS_ILOCK_EXCL);
940
941 /*
942 * Initialize realtime inode pointers in the mount structure
943 */
944 error = xfs_rtmount_inodes(mp);
945 if (error) {
946 /*
947 * Free up the root inode.
948 */
949 xfs_warn(mp, "failed to read RT inodes");
950 goto out_rele_rip;
951 }
952
953 /* Make sure the summary counts are ok. */
954 error = xfs_check_summary_counts(mp);
955 if (error)
956 goto out_rtunmount;
957
958 /*
959 * If this is a read-only mount defer the superblock updates until
960 * the next remount into writeable mode. Otherwise we would never
961 * perform the update e.g. for the root filesystem.
962 */
963 if (mp->m_update_sb && !xfs_is_readonly(mp)) {
964 error = xfs_sync_sb(mp, false);
965 if (error) {
966 xfs_warn(mp, "failed to write sb changes");
967 goto out_rtunmount;
968 }
969 }
970
971 /*
972 * Initialise the XFS quota management subsystem for this mount
973 */
974 if (XFS_IS_QUOTA_ON(mp)) {
975 error = xfs_qm_newmount(mp, "amount, "aflags);
976 if (error)
977 goto out_rtunmount;
978 } else {
979 /*
980 * If a file system had quotas running earlier, but decided to
981 * mount without -o uquota/pquota/gquota options, revoke the
982 * quotachecked license.
983 */
984 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
985 xfs_notice(mp, "resetting quota flags");
986 error = xfs_mount_reset_sbqflags(mp);
987 if (error)
988 goto out_rtunmount;
989 }
990 }
991
992 /*
993 * Finish recovering the file system. This part needed to be delayed
994 * until after the root and real-time bitmap inodes were consistently
995 * read in. Temporarily create per-AG space reservations for metadata
996 * btree shape changes because space freeing transactions (for inode
997 * inactivation) require the per-AG reservation in lieu of reserving
998 * blocks.
999 */
1000 error = xfs_fs_reserve_ag_blocks(mp);
1001 if (error && error == -ENOSPC)
1002 xfs_warn(mp,
1003 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
1004 error = xfs_log_mount_finish(mp);
1005 xfs_fs_unreserve_ag_blocks(mp);
1006 if (error) {
1007 xfs_warn(mp, "log mount finish failed");
1008 goto out_rtunmount;
1009 }
1010
1011 /*
1012 * Now the log is fully replayed, we can transition to full read-only
1013 * mode for read-only mounts. This will sync all the metadata and clean
1014 * the log so that the recovery we just performed does not have to be
1015 * replayed again on the next mount.
1016 *
1017 * We use the same quiesce mechanism as the rw->ro remount, as they are
1018 * semantically identical operations.
1019 */
1020 if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
1021 xfs_log_clean(mp);
1022
1023 /*
1024 * Complete the quota initialisation, post-log-replay component.
1025 */
1026 if (quotamount) {
1027 ASSERT(mp->m_qflags == 0);
1028 mp->m_qflags = quotaflags;
1029
1030 xfs_qm_mount_quotas(mp);
1031 }
1032
1033 /*
1034 * Now we are mounted, reserve a small amount of unused space for
1035 * privileged transactions. This is needed so that transaction
1036 * space required for critical operations can dip into this pool
1037 * when at ENOSPC. This is needed for operations like create with
1038 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1039 * are not allowed to use this reserved space.
1040 *
1041 * This may drive us straight to ENOSPC on mount, but that implies
1042 * we were already there on the last unmount. Warn if this occurs.
1043 */
1044 if (!xfs_is_readonly(mp)) {
1045 error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
1046 if (error)
1047 xfs_warn(mp,
1048 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1049
1050 /* Reserve AG blocks for future btree expansion. */
1051 error = xfs_fs_reserve_ag_blocks(mp);
1052 if (error && error != -ENOSPC)
1053 goto out_agresv;
1054 }
1055
1056 return 0;
1057
1058 out_agresv:
1059 xfs_fs_unreserve_ag_blocks(mp);
1060 xfs_qm_unmount_quotas(mp);
1061 out_rtunmount:
1062 xfs_rtunmount_inodes(mp);
1063 out_rele_rip:
1064 xfs_irele(rip);
1065 /* Clean out dquots that might be in memory after quotacheck. */
1066 xfs_qm_unmount(mp);
1067 out_free_metadir:
1068 if (mp->m_metadirip)
1069 xfs_irele(mp->m_metadirip);
1070
1071 /*
1072 * Inactivate all inodes that might still be in memory after a log
1073 * intent recovery failure so that reclaim can free them. Metadata
1074 * inodes and the root directory shouldn't need inactivation, but the
1075 * mount failed for some reason, so pull down all the state and flee.
1076 */
1077 xfs_inodegc_flush(mp);
1078
1079 /*
1080 * Flush all inode reclamation work and flush the log.
1081 * We have to do this /after/ rtunmount and qm_unmount because those
1082 * two will have scheduled delayed reclaim for the rt/quota inodes.
1083 *
1084 * This is slightly different from the unmountfs call sequence
1085 * because we could be tearing down a partially set up mount. In
1086 * particular, if log_mount_finish fails we bail out without calling
1087 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1088 * quota inodes.
1089 */
1090 xfs_unmount_flush_inodes(mp);
1091 xfs_log_mount_cancel(mp);
1092 out_inodegc_shrinker:
1093 shrinker_free(mp->m_inodegc_shrinker);
1094 out_fail_wait:
1095 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1096 xfs_buftarg_drain(mp->m_logdev_targp);
1097 xfs_buftarg_drain(mp->m_ddev_targp);
1098 out_free_rtgroup:
1099 xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
1100 out_free_perag:
1101 xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
1102 out_free_dir:
1103 xfs_da_unmount(mp);
1104 out_remove_uuid:
1105 xfs_uuid_unmount(mp);
1106 out_remove_errortag:
1107 xfs_errortag_del(mp);
1108 out_remove_error_sysfs:
1109 xfs_error_sysfs_del(mp);
1110 out_remove_scrub_stats:
1111 xchk_stats_unregister(mp->m_scrub_stats);
1112 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1113 out_remove_sysfs:
1114 xfs_sysfs_del(&mp->m_kobj);
1115 out:
1116 return error;
1117}
1118
1119/*
1120 * This flushes out the inodes,dquots and the superblock, unmounts the
1121 * log and makes sure that incore structures are freed.
1122 */
1123void
1124xfs_unmountfs(
1125 struct xfs_mount *mp)
1126{
1127 int error;
1128
1129 /*
1130 * Perform all on-disk metadata updates required to inactivate inodes
1131 * that the VFS evicted earlier in the unmount process. Freeing inodes
1132 * and discarding CoW fork preallocations can cause shape changes to
1133 * the free inode and refcount btrees, respectively, so we must finish
1134 * this before we discard the metadata space reservations. Metadata
1135 * inodes and the root directory do not require inactivation.
1136 */
1137 xfs_inodegc_flush(mp);
1138
1139 xfs_blockgc_stop(mp);
1140 xfs_fs_unreserve_ag_blocks(mp);
1141 xfs_qm_unmount_quotas(mp);
1142 xfs_rtunmount_inodes(mp);
1143 xfs_irele(mp->m_rootip);
1144 if (mp->m_metadirip)
1145 xfs_irele(mp->m_metadirip);
1146
1147 xfs_unmount_flush_inodes(mp);
1148
1149 xfs_qm_unmount(mp);
1150
1151 /*
1152 * Unreserve any blocks we have so that when we unmount we don't account
1153 * the reserved free space as used. This is really only necessary for
1154 * lazy superblock counting because it trusts the incore superblock
1155 * counters to be absolutely correct on clean unmount.
1156 *
1157 * We don't bother correcting this elsewhere for lazy superblock
1158 * counting because on mount of an unclean filesystem we reconstruct the
1159 * correct counter value and this is irrelevant.
1160 *
1161 * For non-lazy counter filesystems, this doesn't matter at all because
1162 * we only every apply deltas to the superblock and hence the incore
1163 * value does not matter....
1164 */
1165 error = xfs_reserve_blocks(mp, 0);
1166 if (error)
1167 xfs_warn(mp, "Unable to free reserved block pool. "
1168 "Freespace may not be correct on next mount.");
1169 xfs_unmount_check(mp);
1170
1171 /*
1172 * Indicate that it's ok to clear log incompat bits before cleaning
1173 * the log and writing the unmount record.
1174 */
1175 xfs_set_done_with_log_incompat(mp);
1176 xfs_log_unmount(mp);
1177 xfs_da_unmount(mp);
1178 xfs_uuid_unmount(mp);
1179
1180#if defined(DEBUG)
1181 xfs_errortag_clearall(mp);
1182#endif
1183 shrinker_free(mp->m_inodegc_shrinker);
1184 xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
1185 xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
1186 xfs_errortag_del(mp);
1187 xfs_error_sysfs_del(mp);
1188 xchk_stats_unregister(mp->m_scrub_stats);
1189 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1190 xfs_sysfs_del(&mp->m_kobj);
1191}
1192
1193/*
1194 * Determine whether modifications can proceed. The caller specifies the minimum
1195 * freeze level for which modifications should not be allowed. This allows
1196 * certain operations to proceed while the freeze sequence is in progress, if
1197 * necessary.
1198 */
1199bool
1200xfs_fs_writable(
1201 struct xfs_mount *mp,
1202 int level)
1203{
1204 ASSERT(level > SB_UNFROZEN);
1205 if ((mp->m_super->s_writers.frozen >= level) ||
1206 xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1207 return false;
1208
1209 return true;
1210}
1211
1212void
1213xfs_add_freecounter(
1214 struct xfs_mount *mp,
1215 struct percpu_counter *counter,
1216 uint64_t delta)
1217{
1218 bool has_resv_pool = (counter == &mp->m_fdblocks);
1219 uint64_t res_used;
1220
1221 /*
1222 * If the reserve pool is depleted, put blocks back into it first.
1223 * Most of the time the pool is full.
1224 */
1225 if (!has_resv_pool || mp->m_resblks == mp->m_resblks_avail) {
1226 percpu_counter_add(counter, delta);
1227 return;
1228 }
1229
1230 spin_lock(&mp->m_sb_lock);
1231 res_used = mp->m_resblks - mp->m_resblks_avail;
1232 if (res_used > delta) {
1233 mp->m_resblks_avail += delta;
1234 } else {
1235 delta -= res_used;
1236 mp->m_resblks_avail = mp->m_resblks;
1237 percpu_counter_add(counter, delta);
1238 }
1239 spin_unlock(&mp->m_sb_lock);
1240}
1241
1242int
1243xfs_dec_freecounter(
1244 struct xfs_mount *mp,
1245 struct percpu_counter *counter,
1246 uint64_t delta,
1247 bool rsvd)
1248{
1249 int64_t lcounter;
1250 uint64_t set_aside = 0;
1251 s32 batch;
1252 bool has_resv_pool;
1253
1254 ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1255 has_resv_pool = (counter == &mp->m_fdblocks);
1256 if (rsvd)
1257 ASSERT(has_resv_pool);
1258
1259 /*
1260 * Taking blocks away, need to be more accurate the closer we
1261 * are to zero.
1262 *
1263 * If the counter has a value of less than 2 * max batch size,
1264 * then make everything serialise as we are real close to
1265 * ENOSPC.
1266 */
1267 if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1268 XFS_FDBLOCKS_BATCH) < 0)
1269 batch = 1;
1270 else
1271 batch = XFS_FDBLOCKS_BATCH;
1272
1273 /*
1274 * Set aside allocbt blocks because these blocks are tracked as free
1275 * space but not available for allocation. Technically this means that a
1276 * single reservation cannot consume all remaining free space, but the
1277 * ratio of allocbt blocks to usable free blocks should be rather small.
1278 * The tradeoff without this is that filesystems that maintain high
1279 * perag block reservations can over reserve physical block availability
1280 * and fail physical allocation, which leads to much more serious
1281 * problems (i.e. transaction abort, pagecache discards, etc.) than
1282 * slightly premature -ENOSPC.
1283 */
1284 if (has_resv_pool)
1285 set_aside = xfs_fdblocks_unavailable(mp);
1286 percpu_counter_add_batch(counter, -((int64_t)delta), batch);
1287 if (__percpu_counter_compare(counter, set_aside,
1288 XFS_FDBLOCKS_BATCH) >= 0) {
1289 /* we had space! */
1290 return 0;
1291 }
1292
1293 /*
1294 * lock up the sb for dipping into reserves before releasing the space
1295 * that took us to ENOSPC.
1296 */
1297 spin_lock(&mp->m_sb_lock);
1298 percpu_counter_add(counter, delta);
1299 if (!has_resv_pool || !rsvd)
1300 goto fdblocks_enospc;
1301
1302 lcounter = (long long)mp->m_resblks_avail - delta;
1303 if (lcounter >= 0) {
1304 mp->m_resblks_avail = lcounter;
1305 spin_unlock(&mp->m_sb_lock);
1306 return 0;
1307 }
1308 xfs_warn_once(mp,
1309"Reserve blocks depleted! Consider increasing reserve pool size.");
1310
1311fdblocks_enospc:
1312 spin_unlock(&mp->m_sb_lock);
1313 return -ENOSPC;
1314}
1315
1316/*
1317 * Used to free the superblock along various error paths.
1318 */
1319void
1320xfs_freesb(
1321 struct xfs_mount *mp)
1322{
1323 struct xfs_buf *bp = mp->m_sb_bp;
1324
1325 xfs_buf_lock(bp);
1326 mp->m_sb_bp = NULL;
1327 xfs_buf_relse(bp);
1328}
1329
1330/*
1331 * If the underlying (data/log/rt) device is readonly, there are some
1332 * operations that cannot proceed.
1333 */
1334int
1335xfs_dev_is_read_only(
1336 struct xfs_mount *mp,
1337 char *message)
1338{
1339 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1340 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1341 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1342 xfs_notice(mp, "%s required on read-only device.", message);
1343 xfs_notice(mp, "write access unavailable, cannot proceed.");
1344 return -EROFS;
1345 }
1346 return 0;
1347}
1348
1349/* Force the summary counters to be recalculated at next mount. */
1350void
1351xfs_force_summary_recalc(
1352 struct xfs_mount *mp)
1353{
1354 if (!xfs_has_lazysbcount(mp))
1355 return;
1356
1357 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1358}
1359
1360/*
1361 * Enable a log incompat feature flag in the primary superblock. The caller
1362 * cannot have any other transactions in progress.
1363 */
1364int
1365xfs_add_incompat_log_feature(
1366 struct xfs_mount *mp,
1367 uint32_t feature)
1368{
1369 struct xfs_dsb *dsb;
1370 int error;
1371
1372 ASSERT(hweight32(feature) == 1);
1373 ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1374
1375 /*
1376 * Force the log to disk and kick the background AIL thread to reduce
1377 * the chances that the bwrite will stall waiting for the AIL to unpin
1378 * the primary superblock buffer. This isn't a data integrity
1379 * operation, so we don't need a synchronous push.
1380 */
1381 error = xfs_log_force(mp, XFS_LOG_SYNC);
1382 if (error)
1383 return error;
1384 xfs_ail_push_all(mp->m_ail);
1385
1386 /*
1387 * Lock the primary superblock buffer to serialize all callers that
1388 * are trying to set feature bits.
1389 */
1390 xfs_buf_lock(mp->m_sb_bp);
1391 xfs_buf_hold(mp->m_sb_bp);
1392
1393 if (xfs_is_shutdown(mp)) {
1394 error = -EIO;
1395 goto rele;
1396 }
1397
1398 if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1399 goto rele;
1400
1401 /*
1402 * Write the primary superblock to disk immediately, because we need
1403 * the log_incompat bit to be set in the primary super now to protect
1404 * the log items that we're going to commit later.
1405 */
1406 dsb = mp->m_sb_bp->b_addr;
1407 xfs_sb_to_disk(dsb, &mp->m_sb);
1408 dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1409 error = xfs_bwrite(mp->m_sb_bp);
1410 if (error)
1411 goto shutdown;
1412
1413 /*
1414 * Add the feature bits to the incore superblock before we unlock the
1415 * buffer.
1416 */
1417 xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1418 xfs_buf_relse(mp->m_sb_bp);
1419
1420 /* Log the superblock to disk. */
1421 return xfs_sync_sb(mp, false);
1422shutdown:
1423 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1424rele:
1425 xfs_buf_relse(mp->m_sb_bp);
1426 return error;
1427}
1428
1429/*
1430 * Clear all the log incompat flags from the superblock.
1431 *
1432 * The caller cannot be in a transaction, must ensure that the log does not
1433 * contain any log items protected by any log incompat bit, and must ensure
1434 * that there are no other threads that depend on the state of the log incompat
1435 * feature flags in the primary super.
1436 *
1437 * Returns true if the superblock is dirty.
1438 */
1439bool
1440xfs_clear_incompat_log_features(
1441 struct xfs_mount *mp)
1442{
1443 bool ret = false;
1444
1445 if (!xfs_has_crc(mp) ||
1446 !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1447 XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1448 xfs_is_shutdown(mp) ||
1449 !xfs_is_done_with_log_incompat(mp))
1450 return false;
1451
1452 /*
1453 * Update the incore superblock. We synchronize on the primary super
1454 * buffer lock to be consistent with the add function, though at least
1455 * in theory this shouldn't be necessary.
1456 */
1457 xfs_buf_lock(mp->m_sb_bp);
1458 xfs_buf_hold(mp->m_sb_bp);
1459
1460 if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1461 XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1462 xfs_sb_remove_incompat_log_features(&mp->m_sb);
1463 ret = true;
1464 }
1465
1466 xfs_buf_relse(mp->m_sb_bp);
1467 return ret;
1468}
1469
1470/*
1471 * Update the in-core delayed block counter.
1472 *
1473 * We prefer to update the counter without having to take a spinlock for every
1474 * counter update (i.e. batching). Each change to delayed allocation
1475 * reservations can change can easily exceed the default percpu counter
1476 * batching, so we use a larger batch factor here.
1477 *
1478 * Note that we don't currently have any callers requiring fast summation
1479 * (e.g. percpu_counter_read) so we can use a big batch value here.
1480 */
1481#define XFS_DELALLOC_BATCH (4096)
1482void
1483xfs_mod_delalloc(
1484 struct xfs_inode *ip,
1485 int64_t data_delta,
1486 int64_t ind_delta)
1487{
1488 struct xfs_mount *mp = ip->i_mount;
1489
1490 if (XFS_IS_REALTIME_INODE(ip)) {
1491 percpu_counter_add_batch(&mp->m_delalloc_rtextents,
1492 xfs_blen_to_rtbxlen(mp, data_delta),
1493 XFS_DELALLOC_BATCH);
1494 if (!ind_delta)
1495 return;
1496 data_delta = 0;
1497 }
1498 percpu_counter_add_batch(&mp->m_delalloc_blks, data_delta + ind_delta,
1499 XFS_DELALLOC_BATCH);
1500}