1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_rtalloc.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_priv.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"
37 #include "scrub/stats.h"
39 static DEFINE_MUTEX(xfs_uuid_table_mutex);
40 static int xfs_uuid_table_size;
41 static uuid_t *xfs_uuid_table;
44 xfs_uuid_table_free(void)
46 if (xfs_uuid_table_size == 0)
48 kfree(xfs_uuid_table);
49 xfs_uuid_table = NULL;
50 xfs_uuid_table_size = 0;
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.
61 uuid_t *uuid = &mp->m_sb.sb_uuid;
64 /* Publish UUID in struct super_block */
65 super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
67 if (xfs_has_nouuid(mp))
70 if (uuid_is_null(uuid)) {
71 xfs_warn(mp, "Filesystem has null UUID - can't mount");
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])) {
81 if (uuid_equal(uuid, &xfs_uuid_table[i]))
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++;
91 xfs_uuid_table[hole] = *uuid;
92 mutex_unlock(&xfs_uuid_table_mutex);
97 mutex_unlock(&xfs_uuid_table_mutex);
98 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
104 struct xfs_mount *mp)
106 uuid_t *uuid = &mp->m_sb.sb_uuid;
109 if (xfs_has_nouuid(mp))
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]))
116 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
118 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
121 ASSERT(i < xfs_uuid_table_size);
122 mutex_unlock(&xfs_uuid_table_mutex);
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.
130 xfs_sb_validate_fsb_count(
134 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
135 ASSERT(sbp->sb_blocklog >= BBSHIFT);
137 /* Limited by ULONG_MAX of page cache index */
138 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
146 * Does the initial read of the superblock.
150 struct xfs_mount *mp,
153 unsigned int sector_size;
155 struct xfs_sb *sbp = &mp->m_sb;
157 int loud = !(flags & XFS_MFSI_QUIET);
158 const struct xfs_buf_ops *buf_ops;
160 ASSERT(mp->m_sb_bp == NULL);
161 ASSERT(mp->m_ddev_targp != NULL);
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.
170 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
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
180 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
181 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
185 xfs_warn(mp, "SB validate failed with error %d.", error);
186 /* bad CRC means corrupted metadata */
187 if (error == -EFSBADCRC)
188 error = -EFSCORRUPTED;
193 * Initialize the mount structure from the superblock.
195 xfs_sb_from_disk(sbp, bp->b_addr);
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.
201 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
203 xfs_warn(mp, "Invalid superblock magic number");
209 * We must be able to do sector-sized and sector-aligned IO.
211 if (sector_size > sbp->sb_sectsize) {
213 xfs_warn(mp, "device supports %u byte sectors (not %u)",
214 sector_size, sbp->sb_sectsize);
219 if (buf_ops == NULL) {
221 * Re-read the superblock so the buffer is correctly sized,
222 * and properly verified.
225 sector_size = sbp->sb_sectsize;
226 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
230 mp->m_features |= xfs_sb_version_to_features(sbp);
231 xfs_reinit_percpu_counters(mp);
233 /* no need to be quiet anymore, so reset the buf ops */
234 bp->b_ops = &xfs_sb_buf_ops;
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.
252 xfs_check_new_dalign(
253 struct xfs_mount *mp,
257 struct xfs_sb *sbp = &mp->m_sb;
260 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
261 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
263 if (sbp->sb_rootino == calc_ino) {
269 "Cannot change stripe alignment; would require moving root inode.");
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.
276 xfs_warn(mp, "Skipping superblock stripe alignment update.");
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.
288 xfs_validate_new_dalign(
289 struct xfs_mount *mp)
291 if (mp->m_dalign == 0)
295 * If stripe unit and stripe width are not multiples
296 * of the fs blocksize turn off alignment.
298 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
299 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
301 "alignment check failed: sunit/swidth vs. blocksize(%d)",
302 mp->m_sb.sb_blocksize);
307 * Convert the stripe unit and width to FSBs.
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)) {
312 "alignment check failed: sunit/swidth vs. agsize(%d)",
313 mp->m_sb.sb_agblocks);
319 "alignment check failed: sunit(%d) less than bsize(%d)",
320 mp->m_dalign, mp->m_sb.sb_blocksize);
324 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
326 if (!xfs_has_dalign(mp)) {
328 "cannot change alignment: superblock does not support data alignment");
335 /* Update alignment values based on mount options and sb values. */
337 xfs_update_alignment(
338 struct xfs_mount *mp)
340 struct xfs_sb *sbp = &mp->m_sb;
346 if (sbp->sb_unit == mp->m_dalign &&
347 sbp->sb_width == mp->m_swidth)
350 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
351 if (error || !update_sb)
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;
366 * precalculate the low space thresholds for dynamic speculative preallocation.
369 xfs_set_low_space_thresholds(
370 struct xfs_mount *mp)
372 uint64_t dblocks = mp->m_sb.sb_dblocks;
373 uint64_t rtexts = mp->m_sb.sb_rextents;
376 do_div(dblocks, 100);
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);
386 * Check that the data (and log if separate) is an ok size.
390 struct xfs_mount *mp)
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");
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);
405 xfs_warn(mp, "last sector read failed");
410 if (mp->m_logdev_targp == mp->m_ddev_targp)
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");
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);
422 xfs_warn(mp, "log device read failed");
430 * Clear the quotaflags in memory and in the superblock.
433 xfs_mount_reset_sbqflags(
434 struct xfs_mount *mp)
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)
441 spin_lock(&mp->m_sb_lock);
442 mp->m_sb.sb_qflags = 0;
443 spin_unlock(&mp->m_sb_lock);
445 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
448 return xfs_sync_sb(mp, false);
452 xfs_default_resblks(xfs_mount_t *mp)
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.
463 resblks = mp->m_sb.sb_dblocks;
465 resblks = min_t(uint64_t, resblks, 8192);
469 /* Ensure the summary counts are correct. */
471 xfs_check_summary_counts(
472 struct xfs_mount *mp)
477 * The AG0 superblock verifier rejects in-progress filesystems,
478 * so we should never see the flag set this far into mounting.
480 if (mp->m_sb.sb_inprogress) {
481 xfs_err(mp, "sb_inprogress set after log recovery??");
483 return -EFSCORRUPTED;
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
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.
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);
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
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.
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);
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.
530 * If we're mounting the rt volume after recovering the log, recompute
531 * frextents from the rtbitmap file to fix the inconsistency.
533 if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
534 error = xfs_rtalloc_reinit_frextents(mp);
544 struct xfs_mount *mp)
546 if (xfs_is_shutdown(mp))
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);
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
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
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.
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.
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.
583 xfs_unmount_flush_inodes(
584 struct xfs_mount *mp)
586 xfs_log_force(mp, XFS_LOG_SYNC);
587 xfs_extent_busy_wait_all(mp);
588 flush_workqueue(xfs_discard_wq);
590 set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate);
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);
600 xfs_mount_setup_inode_geom(
601 struct xfs_mount *mp)
603 struct xfs_ino_geometry *igeo = M_IGEO(mp);
605 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
606 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
608 xfs_ialloc_setup_geometry(mp);
611 /* Compute maximum possible height for per-AG btree types for this fs. */
613 xfs_agbtree_compute_maxlevels(
614 struct xfs_mount *mp)
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);
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
635 struct xfs_mount *mp)
637 struct xfs_sb *sbp = &(mp->m_sb);
638 struct xfs_inode *rip;
639 struct xfs_ino_geometry *igeo = M_IGEO(mp);
644 xfs_sb_mount_common(mp, sbp);
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
653 * For backwards compatibility, we make both slots equal.
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.
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;
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;
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.
682 error = xfs_validate_new_dalign(mp);
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);
693 xfs_agbtree_compute_maxlevels(mp);
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
702 error = xfs_update_alignment(mp);
706 /* enable fail_at_unmount as default */
707 mp->m_fail_unmount = true;
709 super_set_sysfs_name_id(mp->m_super);
711 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
712 NULL, mp->m_super->s_id);
716 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
717 &mp->m_kobj, "stats");
719 goto out_remove_sysfs;
721 xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
723 error = xfs_error_sysfs_init(mp);
725 goto out_remove_scrub_stats;
727 error = xfs_errortag_init(mp);
729 goto out_remove_error_sysfs;
731 error = xfs_uuid_mount(mp);
733 goto out_remove_errortag;
736 * Update the preferred write size based on the information from the
737 * on-disk superblock.
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);
743 /* set the low space thresholds for dynamic preallocation */
744 xfs_set_low_space_thresholds(mp);
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...
751 if (xfs_has_sparseinodes(mp) &&
752 mp->m_sb.sb_spino_align !=
753 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
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));
759 goto out_remove_uuid;
763 * Check that the data (and log if separate) is an ok size.
765 error = xfs_check_sizes(mp);
767 goto out_remove_uuid;
770 * Initialize realtime fields in the mount structure
772 error = xfs_rtmount_init(mp);
774 xfs_warn(mp, "RT mount failed");
775 goto out_remove_uuid;
779 * Copies the low order bits of the timestamp and the randomly
780 * set "sequence" number out of a UUID.
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]);
787 error = xfs_da_mount(mp);
789 xfs_warn(mp, "Failed dir/attr init: %d", error);
790 goto out_remove_uuid;
794 * Initialize the precomputed transaction reservations values.
799 * Allocate and initialize the per-ag data.
801 error = xfs_initialize_perag(mp, sbp->sb_agcount, mp->m_sb.sb_dblocks,
804 xfs_warn(mp, "Failed per-ag init: %d", error);
808 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
809 xfs_warn(mp, "no log defined");
810 error = -EFSCORRUPTED;
814 error = xfs_inodegc_register_shrinker(mp);
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
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));
827 xfs_warn(mp, "log mount failed");
828 goto out_inodegc_shrinker;
831 /* Enable background inode inactivation workers. */
832 xfs_inodegc_start(mp);
833 xfs_blockgc_start(mp);
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.
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;
849 * Get and sanity-check the root inode.
850 * Save the pointer to it in the mount structure.
852 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
853 XFS_ILOCK_EXCL, &rip);
856 "Failed to read root inode 0x%llx, error %d",
857 sbp->sb_rootino, -error);
858 goto out_log_dealloc;
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;
870 mp->m_rootip = rip; /* save it */
872 xfs_iunlock(rip, XFS_ILOCK_EXCL);
875 * Initialize realtime inode pointers in the mount structure
877 error = xfs_rtmount_inodes(mp);
880 * Free up the root inode.
882 xfs_warn(mp, "failed to read RT inodes");
886 /* Make sure the summary counts are ok. */
887 error = xfs_check_summary_counts(mp);
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.
896 if (mp->m_update_sb && !xfs_is_readonly(mp)) {
897 error = xfs_sync_sb(mp, false);
899 xfs_warn(mp, "failed to write sb changes");
905 * Initialise the XFS quota management subsystem for this mount
907 if (XFS_IS_QUOTA_ON(mp)) {
908 error = xfs_qm_newmount(mp, "amount, "aflags);
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.
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);
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
933 error = xfs_fs_reserve_ag_blocks(mp);
934 if (error && error == -ENOSPC)
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);
940 xfs_warn(mp, "log mount finish failed");
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.
950 * We use the same quiesce mechanism as the rw->ro remount, as they are
951 * semantically identical operations.
953 if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
957 * Complete the quota initialisation, post-log-replay component.
960 ASSERT(mp->m_qflags == 0);
961 mp->m_qflags = quotaflags;
963 xfs_qm_mount_quotas(mp);
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.
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.
977 if (!xfs_is_readonly(mp)) {
978 error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
981 "Unable to allocate reserve blocks. Continuing without reserve pool.");
983 /* Reserve AG blocks for future btree expansion. */
984 error = xfs_fs_reserve_ag_blocks(mp);
985 if (error && error != -ENOSPC)
992 xfs_fs_unreserve_ag_blocks(mp);
993 xfs_qm_unmount_quotas(mp);
995 xfs_rtunmount_inodes(mp);
998 /* Clean out dquots that might be in memory after quotacheck. */
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.
1007 xfs_inodegc_flush(mp);
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.
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
1020 xfs_unmount_flush_inodes(mp);
1022 xfs_log_mount_cancel(mp);
1023 out_inodegc_shrinker:
1024 shrinker_free(mp->m_inodegc_shrinker);
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);
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);
1043 xfs_sysfs_del(&mp->m_kobj);
1049 * This flushes out the inodes,dquots and the superblock, unmounts the
1050 * log and makes sure that incore structures are freed.
1054 struct xfs_mount *mp)
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.
1066 xfs_inodegc_flush(mp);
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);
1074 xfs_unmount_flush_inodes(mp);
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.
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.
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....
1092 error = xfs_reserve_blocks(mp, 0);
1094 xfs_warn(mp, "Unable to free reserved block pool. "
1095 "Freespace may not be correct on next mount.");
1096 xfs_unmount_check(mp);
1098 xfs_log_unmount(mp);
1100 xfs_uuid_unmount(mp);
1103 xfs_errortag_clearall(mp);
1105 shrinker_free(mp->m_inodegc_shrinker);
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);
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
1123 struct xfs_mount *mp,
1126 ASSERT(level > SB_UNFROZEN);
1127 if ((mp->m_super->s_writers.frozen >= level) ||
1128 xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1134 /* Adjust m_fdblocks or m_frextents. */
1136 xfs_mod_freecounter(
1137 struct xfs_mount *mp,
1138 struct percpu_counter *counter,
1144 uint64_t set_aside = 0;
1148 ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1149 has_resv_pool = (counter == &mp->m_fdblocks);
1151 ASSERT(has_resv_pool);
1155 * If the reserve pool is depleted, put blocks back into it
1156 * first. Most of the time the pool is full.
1158 if (likely(!has_resv_pool ||
1159 mp->m_resblks == mp->m_resblks_avail)) {
1160 percpu_counter_add(counter, delta);
1164 spin_lock(&mp->m_sb_lock);
1165 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1167 if (res_used > delta) {
1168 mp->m_resblks_avail += delta;
1171 mp->m_resblks_avail = mp->m_resblks;
1172 percpu_counter_add(counter, delta);
1174 spin_unlock(&mp->m_sb_lock);
1179 * Taking blocks away, need to be more accurate the closer we
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
1186 if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1187 XFS_FDBLOCKS_BATCH) < 0)
1190 batch = XFS_FDBLOCKS_BATCH;
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.
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) {
1213 * lock up the sb for dipping into reserves before releasing the space
1214 * that took us to ENOSPC.
1216 spin_lock(&mp->m_sb_lock);
1217 percpu_counter_add(counter, -delta);
1218 if (!has_resv_pool || !rsvd)
1219 goto fdblocks_enospc;
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);
1228 "Reserve blocks depleted! Consider increasing reserve pool size.");
1231 spin_unlock(&mp->m_sb_lock);
1236 * Used to free the superblock along various error paths.
1240 struct xfs_mount *mp)
1242 struct xfs_buf *bp = mp->m_sb_bp;
1250 * If the underlying (data/log/rt) device is readonly, there are some
1251 * operations that cannot proceed.
1254 xfs_dev_is_read_only(
1255 struct xfs_mount *mp,
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.");
1268 /* Force the summary counters to be recalculated at next mount. */
1270 xfs_force_summary_recalc(
1271 struct xfs_mount *mp)
1273 if (!xfs_has_lazysbcount(mp))
1276 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1280 * Enable a log incompat feature flag in the primary superblock. The caller
1281 * cannot have any other transactions in progress.
1284 xfs_add_incompat_log_feature(
1285 struct xfs_mount *mp,
1288 struct xfs_dsb *dsb;
1291 ASSERT(hweight32(feature) == 1);
1292 ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
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.
1300 error = xfs_log_force(mp, XFS_LOG_SYNC);
1303 xfs_ail_push_all(mp->m_ail);
1306 * Lock the primary superblock buffer to serialize all callers that
1307 * are trying to set feature bits.
1309 xfs_buf_lock(mp->m_sb_bp);
1310 xfs_buf_hold(mp->m_sb_bp);
1312 if (xfs_is_shutdown(mp)) {
1317 if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
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.
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);
1333 * Add the feature bits to the incore superblock before we unlock the
1336 xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1337 xfs_buf_relse(mp->m_sb_bp);
1339 /* Log the superblock to disk. */
1340 return xfs_sync_sb(mp, false);
1342 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1344 xfs_buf_relse(mp->m_sb_bp);
1349 * Clear all the log incompat flags from the superblock.
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.
1356 * Returns true if the superblock is dirty.
1359 xfs_clear_incompat_log_features(
1360 struct xfs_mount *mp)
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))
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.
1375 xfs_buf_lock(mp->m_sb_bp);
1376 xfs_buf_hold(mp->m_sb_bp);
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);
1384 xfs_buf_relse(mp->m_sb_bp);
1389 * Update the in-core delayed block counter.
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.
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.
1399 #define XFS_DELALLOC_BATCH (4096)
1402 struct xfs_mount *mp,
1405 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1406 XFS_DELALLOC_BATCH);