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_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_fsops.h"
27 #include "xfs_icache.h"
28 #include "xfs_sysfs.h"
29 #include "xfs_rmap_btree.h"
30 #include "xfs_refcount_btree.h"
31 #include "xfs_reflink.h"
32 #include "xfs_extent_busy.h"
33 #include "xfs_health.h"
34 #include "xfs_trace.h"
37 static DEFINE_MUTEX(xfs_uuid_table_mutex);
38 static int xfs_uuid_table_size;
39 static uuid_t *xfs_uuid_table;
42 xfs_uuid_table_free(void)
44 if (xfs_uuid_table_size == 0)
46 kmem_free(xfs_uuid_table);
47 xfs_uuid_table = NULL;
48 xfs_uuid_table_size = 0;
52 * See if the UUID is unique among mounted XFS filesystems.
53 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
59 uuid_t *uuid = &mp->m_sb.sb_uuid;
62 /* Publish UUID in struct super_block */
63 uuid_copy(&mp->m_super->s_uuid, uuid);
65 if (mp->m_flags & XFS_MOUNT_NOUUID)
68 if (uuid_is_null(uuid)) {
69 xfs_warn(mp, "Filesystem has null UUID - can't mount");
73 mutex_lock(&xfs_uuid_table_mutex);
74 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
75 if (uuid_is_null(&xfs_uuid_table[i])) {
79 if (uuid_equal(uuid, &xfs_uuid_table[i]))
84 xfs_uuid_table = krealloc(xfs_uuid_table,
85 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
86 GFP_KERNEL | __GFP_NOFAIL);
87 hole = xfs_uuid_table_size++;
89 xfs_uuid_table[hole] = *uuid;
90 mutex_unlock(&xfs_uuid_table_mutex);
95 mutex_unlock(&xfs_uuid_table_mutex);
96 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
102 struct xfs_mount *mp)
104 uuid_t *uuid = &mp->m_sb.sb_uuid;
107 if (mp->m_flags & XFS_MOUNT_NOUUID)
110 mutex_lock(&xfs_uuid_table_mutex);
111 for (i = 0; i < xfs_uuid_table_size; i++) {
112 if (uuid_is_null(&xfs_uuid_table[i]))
114 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
116 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
119 ASSERT(i < xfs_uuid_table_size);
120 mutex_unlock(&xfs_uuid_table_mutex);
124 * Check size of device based on the (data/realtime) block count.
125 * Note: this check is used by the growfs code as well as mount.
128 xfs_sb_validate_fsb_count(
132 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
133 ASSERT(sbp->sb_blocklog >= BBSHIFT);
135 /* Limited by ULONG_MAX of page cache index */
136 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
144 * Does the initial read of the superblock.
148 struct xfs_mount *mp,
151 unsigned int sector_size;
153 struct xfs_sb *sbp = &mp->m_sb;
155 int loud = !(flags & XFS_MFSI_QUIET);
156 const struct xfs_buf_ops *buf_ops;
158 ASSERT(mp->m_sb_bp == NULL);
159 ASSERT(mp->m_ddev_targp != NULL);
162 * For the initial read, we must guess at the sector
163 * size based on the block device. It's enough to
164 * get the sb_sectsize out of the superblock and
165 * then reread with the proper length.
166 * We don't verify it yet, because it may not be complete.
168 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
172 * Allocate a (locked) buffer to hold the superblock. This will be kept
173 * around at all times to optimize access to the superblock. Therefore,
174 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
178 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
179 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
183 xfs_warn(mp, "SB validate failed with error %d.", error);
184 /* bad CRC means corrupted metadata */
185 if (error == -EFSBADCRC)
186 error = -EFSCORRUPTED;
191 * Initialize the mount structure from the superblock.
193 xfs_sb_from_disk(sbp, bp->b_addr);
196 * If we haven't validated the superblock, do so now before we try
197 * to check the sector size and reread the superblock appropriately.
199 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
201 xfs_warn(mp, "Invalid superblock magic number");
207 * We must be able to do sector-sized and sector-aligned IO.
209 if (sector_size > sbp->sb_sectsize) {
211 xfs_warn(mp, "device supports %u byte sectors (not %u)",
212 sector_size, sbp->sb_sectsize);
217 if (buf_ops == NULL) {
219 * Re-read the superblock so the buffer is correctly sized,
220 * and properly verified.
223 sector_size = sbp->sb_sectsize;
224 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
228 xfs_reinit_percpu_counters(mp);
230 /* no need to be quiet anymore, so reset the buf ops */
231 bp->b_ops = &xfs_sb_buf_ops;
243 * If the sunit/swidth change would move the precomputed root inode value, we
244 * must reject the ondisk change because repair will stumble over that.
245 * However, we allow the mount to proceed because we never rejected this
246 * combination before. Returns true to update the sb, false otherwise.
249 xfs_check_new_dalign(
250 struct xfs_mount *mp,
254 struct xfs_sb *sbp = &mp->m_sb;
257 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
258 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
260 if (sbp->sb_rootino == calc_ino) {
266 "Cannot change stripe alignment; would require moving root inode.");
269 * XXX: Next time we add a new incompat feature, this should start
270 * returning -EINVAL to fail the mount. Until then, spit out a warning
271 * that we're ignoring the administrator's instructions.
273 xfs_warn(mp, "Skipping superblock stripe alignment update.");
279 * If we were provided with new sunit/swidth values as mount options, make sure
280 * that they pass basic alignment and superblock feature checks, and convert
281 * them into the same units (FSB) that everything else expects. This step
282 * /must/ be done before computing the inode geometry.
285 xfs_validate_new_dalign(
286 struct xfs_mount *mp)
288 if (mp->m_dalign == 0)
292 * If stripe unit and stripe width are not multiples
293 * of the fs blocksize turn off alignment.
295 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
296 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
298 "alignment check failed: sunit/swidth vs. blocksize(%d)",
299 mp->m_sb.sb_blocksize);
303 * Convert the stripe unit and width to FSBs.
305 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
306 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
308 "alignment check failed: sunit/swidth vs. agsize(%d)",
309 mp->m_sb.sb_agblocks);
311 } else if (mp->m_dalign) {
312 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
315 "alignment check failed: sunit(%d) less than bsize(%d)",
316 mp->m_dalign, mp->m_sb.sb_blocksize);
321 if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
323 "cannot change alignment: superblock does not support data alignment");
330 /* Update alignment values based on mount options and sb values. */
332 xfs_update_alignment(
333 struct xfs_mount *mp)
335 struct xfs_sb *sbp = &mp->m_sb;
341 if (sbp->sb_unit == mp->m_dalign &&
342 sbp->sb_width == mp->m_swidth)
345 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
346 if (error || !update_sb)
349 sbp->sb_unit = mp->m_dalign;
350 sbp->sb_width = mp->m_swidth;
351 mp->m_update_sb = true;
352 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
353 xfs_sb_version_hasdalign(&mp->m_sb)) {
354 mp->m_dalign = sbp->sb_unit;
355 mp->m_swidth = sbp->sb_width;
362 * precalculate the low space thresholds for dynamic speculative preallocation.
365 xfs_set_low_space_thresholds(
366 struct xfs_mount *mp)
370 for (i = 0; i < XFS_LOWSP_MAX; i++) {
371 uint64_t space = mp->m_sb.sb_dblocks;
374 mp->m_low_space[i] = space * (i + 1);
379 * Check that the data (and log if separate) is an ok size.
383 struct xfs_mount *mp)
389 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
390 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
391 xfs_warn(mp, "filesystem size mismatch detected");
394 error = xfs_buf_read_uncached(mp->m_ddev_targp,
395 d - XFS_FSS_TO_BB(mp, 1),
396 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
398 xfs_warn(mp, "last sector read failed");
403 if (mp->m_logdev_targp == mp->m_ddev_targp)
406 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
407 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
408 xfs_warn(mp, "log size mismatch detected");
411 error = xfs_buf_read_uncached(mp->m_logdev_targp,
412 d - XFS_FSB_TO_BB(mp, 1),
413 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
415 xfs_warn(mp, "log device read failed");
423 * Clear the quotaflags in memory and in the superblock.
426 xfs_mount_reset_sbqflags(
427 struct xfs_mount *mp)
431 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
432 if (mp->m_sb.sb_qflags == 0)
434 spin_lock(&mp->m_sb_lock);
435 mp->m_sb.sb_qflags = 0;
436 spin_unlock(&mp->m_sb_lock);
438 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
441 return xfs_sync_sb(mp, false);
445 xfs_default_resblks(xfs_mount_t *mp)
450 * We default to 5% or 8192 fsbs of space reserved, whichever is
451 * smaller. This is intended to cover concurrent allocation
452 * transactions when we initially hit enospc. These each require a 4
453 * block reservation. Hence by default we cover roughly 2000 concurrent
454 * allocation reservations.
456 resblks = mp->m_sb.sb_dblocks;
458 resblks = min_t(uint64_t, resblks, 8192);
462 /* Ensure the summary counts are correct. */
464 xfs_check_summary_counts(
465 struct xfs_mount *mp)
468 * The AG0 superblock verifier rejects in-progress filesystems,
469 * so we should never see the flag set this far into mounting.
471 if (mp->m_sb.sb_inprogress) {
472 xfs_err(mp, "sb_inprogress set after log recovery??");
474 return -EFSCORRUPTED;
478 * Now the log is mounted, we know if it was an unclean shutdown or
479 * not. If it was, with the first phase of recovery has completed, we
480 * have consistent AG blocks on disk. We have not recovered EFIs yet,
481 * but they are recovered transactionally in the second recovery phase
484 * If the log was clean when we mounted, we can check the summary
485 * counters. If any of them are obviously incorrect, we can recompute
486 * them from the AGF headers in the next step.
488 if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
489 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
490 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
491 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
492 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
495 * We can safely re-initialise incore superblock counters from the
496 * per-ag data. These may not be correct if the filesystem was not
497 * cleanly unmounted, so we waited for recovery to finish before doing
500 * If the filesystem was cleanly unmounted or the previous check did
501 * not flag anything weird, then we can trust the values in the
502 * superblock to be correct and we don't need to do anything here.
503 * Otherwise, recalculate the summary counters.
505 if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
506 XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
507 !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
510 return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
514 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
515 * internal inode structures can be sitting in the CIL and AIL at this point,
516 * so we need to unpin them, write them back and/or reclaim them before unmount
519 * An inode cluster that has been freed can have its buffer still pinned in
520 * memory because the transaction is still sitting in a iclog. The stale inodes
521 * on that buffer will be pinned to the buffer until the transaction hits the
522 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
523 * may never see the pinned buffer, so nothing will push out the iclog and
526 * Hence we need to force the log to unpin everything first. However, log
527 * forces don't wait for the discards they issue to complete, so we have to
528 * explicitly wait for them to complete here as well.
530 * Then we can tell the world we are unmounting so that error handling knows
531 * that the filesystem is going away and we should error out anything that we
532 * have been retrying in the background. This will prevent never-ending
533 * retries in AIL pushing from hanging the unmount.
535 * Finally, we can push the AIL to clean all the remaining dirty objects, then
536 * reclaim the remaining inodes that are still in memory at this point in time.
539 xfs_unmount_flush_inodes(
540 struct xfs_mount *mp)
542 xfs_log_force(mp, XFS_LOG_SYNC);
543 xfs_extent_busy_wait_all(mp);
544 flush_workqueue(xfs_discard_wq);
546 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
548 xfs_ail_push_all_sync(mp->m_ail);
549 cancel_delayed_work_sync(&mp->m_reclaim_work);
550 xfs_reclaim_inodes(mp);
551 xfs_health_unmount(mp);
555 xfs_mount_setup_inode_geom(
556 struct xfs_mount *mp)
558 struct xfs_ino_geometry *igeo = M_IGEO(mp);
560 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
561 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
563 xfs_ialloc_setup_geometry(mp);
567 * This function does the following on an initial mount of a file system:
568 * - reads the superblock from disk and init the mount struct
569 * - if we're a 32-bit kernel, do a size check on the superblock
570 * so we don't mount terabyte filesystems
571 * - init mount struct realtime fields
572 * - allocate inode hash table for fs
573 * - init directory manager
574 * - perform recovery and init the log manager
578 struct xfs_mount *mp)
580 struct xfs_sb *sbp = &(mp->m_sb);
581 struct xfs_inode *rip;
582 struct xfs_ino_geometry *igeo = M_IGEO(mp);
588 xfs_sb_mount_common(mp, sbp);
591 * Check for a mismatched features2 values. Older kernels read & wrote
592 * into the wrong sb offset for sb_features2 on some platforms due to
593 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
594 * which made older superblock reading/writing routines swap it as a
597 * For backwards compatibility, we make both slots equal.
599 * If we detect a mismatched field, we OR the set bits into the existing
600 * features2 field in case it has already been modified; we don't want
601 * to lose any features. We then update the bad location with the ORed
602 * value so that older kernels will see any features2 flags. The
603 * superblock writeback code ensures the new sb_features2 is copied to
604 * sb_bad_features2 before it is logged or written to disk.
606 if (xfs_sb_has_mismatched_features2(sbp)) {
607 xfs_warn(mp, "correcting sb_features alignment problem");
608 sbp->sb_features2 |= sbp->sb_bad_features2;
609 mp->m_update_sb = true;
612 * Re-check for ATTR2 in case it was found in bad_features2
615 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
616 !(mp->m_flags & XFS_MOUNT_NOATTR2))
617 mp->m_flags |= XFS_MOUNT_ATTR2;
620 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
621 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
622 xfs_sb_version_removeattr2(&mp->m_sb);
623 mp->m_update_sb = true;
625 /* update sb_versionnum for the clearing of the morebits */
626 if (!sbp->sb_features2)
627 mp->m_update_sb = true;
630 /* always use v2 inodes by default now */
631 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
632 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
633 mp->m_update_sb = true;
637 * If we were given new sunit/swidth options, do some basic validation
638 * checks and convert the incore dalign and swidth values to the
639 * same units (FSB) that everything else uses. This /must/ happen
640 * before computing the inode geometry.
642 error = xfs_validate_new_dalign(mp);
646 xfs_alloc_compute_maxlevels(mp);
647 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
648 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
649 xfs_mount_setup_inode_geom(mp);
650 xfs_rmapbt_compute_maxlevels(mp);
651 xfs_refcountbt_compute_maxlevels(mp);
654 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
655 * is NOT aligned turn off m_dalign since allocator alignment is within
656 * an ag, therefore ag has to be aligned at stripe boundary. Note that
657 * we must compute the free space and rmap btree geometry before doing
660 error = xfs_update_alignment(mp);
664 /* enable fail_at_unmount as default */
665 mp->m_fail_unmount = true;
667 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
668 NULL, mp->m_super->s_id);
672 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
673 &mp->m_kobj, "stats");
675 goto out_remove_sysfs;
677 error = xfs_error_sysfs_init(mp);
681 error = xfs_errortag_init(mp);
683 goto out_remove_error_sysfs;
685 error = xfs_uuid_mount(mp);
687 goto out_remove_errortag;
690 * Update the preferred write size based on the information from the
691 * on-disk superblock.
693 mp->m_allocsize_log =
694 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
695 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
697 /* set the low space thresholds for dynamic preallocation */
698 xfs_set_low_space_thresholds(mp);
701 * If enabled, sparse inode chunk alignment is expected to match the
702 * cluster size. Full inode chunk alignment must match the chunk size,
703 * but that is checked on sb read verification...
705 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
706 mp->m_sb.sb_spino_align !=
707 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
709 "Sparse inode block alignment (%u) must match cluster size (%llu).",
710 mp->m_sb.sb_spino_align,
711 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
713 goto out_remove_uuid;
717 * Check that the data (and log if separate) is an ok size.
719 error = xfs_check_sizes(mp);
721 goto out_remove_uuid;
724 * Initialize realtime fields in the mount structure
726 error = xfs_rtmount_init(mp);
728 xfs_warn(mp, "RT mount failed");
729 goto out_remove_uuid;
733 * Copies the low order bits of the timestamp and the randomly
734 * set "sequence" number out of a UUID.
737 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
738 get_unaligned_be16(&sbp->sb_uuid.b[4]);
739 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
741 error = xfs_da_mount(mp);
743 xfs_warn(mp, "Failed dir/attr init: %d", error);
744 goto out_remove_uuid;
748 * Initialize the precomputed transaction reservations values.
753 * Allocate and initialize the per-ag data.
755 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
757 xfs_warn(mp, "Failed per-ag init: %d", error);
761 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
762 xfs_warn(mp, "no log defined");
763 error = -EFSCORRUPTED;
768 * Log's mount-time initialization. The first part of recovery can place
769 * some items on the AIL, to be handled when recovery is finished or
772 error = xfs_log_mount(mp, mp->m_logdev_targp,
773 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
774 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
776 xfs_warn(mp, "log mount failed");
780 /* Make sure the summary counts are ok. */
781 error = xfs_check_summary_counts(mp);
783 goto out_log_dealloc;
786 * Get and sanity-check the root inode.
787 * Save the pointer to it in the mount structure.
789 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
790 XFS_ILOCK_EXCL, &rip);
793 "Failed to read root inode 0x%llx, error %d",
794 sbp->sb_rootino, -error);
795 goto out_log_dealloc;
800 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
801 xfs_warn(mp, "corrupted root inode %llu: not a directory",
802 (unsigned long long)rip->i_ino);
803 xfs_iunlock(rip, XFS_ILOCK_EXCL);
804 error = -EFSCORRUPTED;
807 mp->m_rootip = rip; /* save it */
809 xfs_iunlock(rip, XFS_ILOCK_EXCL);
812 * Initialize realtime inode pointers in the mount structure
814 error = xfs_rtmount_inodes(mp);
817 * Free up the root inode.
819 xfs_warn(mp, "failed to read RT inodes");
824 * If this is a read-only mount defer the superblock updates until
825 * the next remount into writeable mode. Otherwise we would never
826 * perform the update e.g. for the root filesystem.
828 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
829 error = xfs_sync_sb(mp, false);
831 xfs_warn(mp, "failed to write sb changes");
837 * Initialise the XFS quota management subsystem for this mount
839 if (XFS_IS_QUOTA_RUNNING(mp)) {
840 error = xfs_qm_newmount(mp, "amount, "aflags);
844 ASSERT(!XFS_IS_QUOTA_ON(mp));
847 * If a file system had quotas running earlier, but decided to
848 * mount without -o uquota/pquota/gquota options, revoke the
849 * quotachecked license.
851 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
852 xfs_notice(mp, "resetting quota flags");
853 error = xfs_mount_reset_sbqflags(mp);
860 * Finish recovering the file system. This part needed to be delayed
861 * until after the root and real-time bitmap inodes were consistently
862 * read in. Temporarily create per-AG space reservations for metadata
863 * btree shape changes because space freeing transactions (for inode
864 * inactivation) require the per-AG reservation in lieu of reserving
867 error = xfs_fs_reserve_ag_blocks(mp);
868 if (error && error == -ENOSPC)
870 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
871 error = xfs_log_mount_finish(mp);
872 xfs_fs_unreserve_ag_blocks(mp);
874 xfs_warn(mp, "log mount finish failed");
879 * Now the log is fully replayed, we can transition to full read-only
880 * mode for read-only mounts. This will sync all the metadata and clean
881 * the log so that the recovery we just performed does not have to be
882 * replayed again on the next mount.
884 * We use the same quiesce mechanism as the rw->ro remount, as they are
885 * semantically identical operations.
887 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
893 * Complete the quota initialisation, post-log-replay component.
896 ASSERT(mp->m_qflags == 0);
897 mp->m_qflags = quotaflags;
899 xfs_qm_mount_quotas(mp);
903 * Now we are mounted, reserve a small amount of unused space for
904 * privileged transactions. This is needed so that transaction
905 * space required for critical operations can dip into this pool
906 * when at ENOSPC. This is needed for operations like create with
907 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
908 * are not allowed to use this reserved space.
910 * This may drive us straight to ENOSPC on mount, but that implies
911 * we were already there on the last unmount. Warn if this occurs.
913 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
914 resblks = xfs_default_resblks(mp);
915 error = xfs_reserve_blocks(mp, &resblks, NULL);
918 "Unable to allocate reserve blocks. Continuing without reserve pool.");
920 /* Recover any CoW blocks that never got remapped. */
921 error = xfs_reflink_recover_cow(mp);
924 "Error %d recovering leftover CoW allocations.", error);
925 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
929 /* Reserve AG blocks for future btree expansion. */
930 error = xfs_fs_reserve_ag_blocks(mp);
931 if (error && error != -ENOSPC)
938 xfs_fs_unreserve_ag_blocks(mp);
940 xfs_qm_unmount_quotas(mp);
942 xfs_rtunmount_inodes(mp);
945 /* Clean out dquots that might be in memory after quotacheck. */
948 * Flush all inode reclamation work and flush the log.
949 * We have to do this /after/ rtunmount and qm_unmount because those
950 * two will have scheduled delayed reclaim for the rt/quota inodes.
952 * This is slightly different from the unmountfs call sequence
953 * because we could be tearing down a partially set up mount. In
954 * particular, if log_mount_finish fails we bail out without calling
955 * qm_unmount_quotas and therefore rely on qm_unmount to release the
958 xfs_unmount_flush_inodes(mp);
960 xfs_log_mount_cancel(mp);
962 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
963 xfs_buftarg_drain(mp->m_logdev_targp);
964 xfs_buftarg_drain(mp->m_ddev_targp);
970 xfs_uuid_unmount(mp);
972 xfs_errortag_del(mp);
973 out_remove_error_sysfs:
974 xfs_error_sysfs_del(mp);
976 xfs_sysfs_del(&mp->m_stats.xs_kobj);
978 xfs_sysfs_del(&mp->m_kobj);
984 * This flushes out the inodes,dquots and the superblock, unmounts the
985 * log and makes sure that incore structures are freed.
989 struct xfs_mount *mp)
994 xfs_blockgc_stop(mp);
995 xfs_fs_unreserve_ag_blocks(mp);
996 xfs_qm_unmount_quotas(mp);
997 xfs_rtunmount_inodes(mp);
998 xfs_irele(mp->m_rootip);
1000 xfs_unmount_flush_inodes(mp);
1005 * Unreserve any blocks we have so that when we unmount we don't account
1006 * the reserved free space as used. This is really only necessary for
1007 * lazy superblock counting because it trusts the incore superblock
1008 * counters to be absolutely correct on clean unmount.
1010 * We don't bother correcting this elsewhere for lazy superblock
1011 * counting because on mount of an unclean filesystem we reconstruct the
1012 * correct counter value and this is irrelevant.
1014 * For non-lazy counter filesystems, this doesn't matter at all because
1015 * we only every apply deltas to the superblock and hence the incore
1016 * value does not matter....
1019 error = xfs_reserve_blocks(mp, &resblks, NULL);
1021 xfs_warn(mp, "Unable to free reserved block pool. "
1022 "Freespace may not be correct on next mount.");
1024 xfs_log_unmount(mp);
1026 xfs_uuid_unmount(mp);
1029 xfs_errortag_clearall(mp);
1033 xfs_errortag_del(mp);
1034 xfs_error_sysfs_del(mp);
1035 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1036 xfs_sysfs_del(&mp->m_kobj);
1040 * Determine whether modifications can proceed. The caller specifies the minimum
1041 * freeze level for which modifications should not be allowed. This allows
1042 * certain operations to proceed while the freeze sequence is in progress, if
1047 struct xfs_mount *mp,
1050 ASSERT(level > SB_UNFROZEN);
1051 if ((mp->m_super->s_writers.frozen >= level) ||
1052 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1059 * Deltas for the block count can vary from 1 to very large, but lock contention
1060 * only occurs on frequent small block count updates such as in the delayed
1061 * allocation path for buffered writes (page a time updates). Hence we set
1062 * a large batch count (1024) to minimise global counter updates except when
1063 * we get near to ENOSPC and we have to be very accurate with our updates.
1065 #define XFS_FDBLOCKS_BATCH 1024
1068 struct xfs_mount *mp,
1079 * If the reserve pool is depleted, put blocks back into it
1080 * first. Most of the time the pool is full.
1082 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1083 percpu_counter_add(&mp->m_fdblocks, delta);
1087 spin_lock(&mp->m_sb_lock);
1088 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1090 if (res_used > delta) {
1091 mp->m_resblks_avail += delta;
1094 mp->m_resblks_avail = mp->m_resblks;
1095 percpu_counter_add(&mp->m_fdblocks, delta);
1097 spin_unlock(&mp->m_sb_lock);
1102 * Taking blocks away, need to be more accurate the closer we
1105 * If the counter has a value of less than 2 * max batch size,
1106 * then make everything serialise as we are real close to
1109 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1110 XFS_FDBLOCKS_BATCH) < 0)
1113 batch = XFS_FDBLOCKS_BATCH;
1116 * Set aside allocbt blocks because these blocks are tracked as free
1117 * space but not available for allocation. Technically this means that a
1118 * single reservation cannot consume all remaining free space, but the
1119 * ratio of allocbt blocks to usable free blocks should be rather small.
1120 * The tradeoff without this is that filesystems that maintain high
1121 * perag block reservations can over reserve physical block availability
1122 * and fail physical allocation, which leads to much more serious
1123 * problems (i.e. transaction abort, pagecache discards, etc.) than
1124 * slightly premature -ENOSPC.
1126 set_aside = mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks);
1127 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1128 if (__percpu_counter_compare(&mp->m_fdblocks, set_aside,
1129 XFS_FDBLOCKS_BATCH) >= 0) {
1135 * lock up the sb for dipping into reserves before releasing the space
1136 * that took us to ENOSPC.
1138 spin_lock(&mp->m_sb_lock);
1139 percpu_counter_add(&mp->m_fdblocks, -delta);
1141 goto fdblocks_enospc;
1143 lcounter = (long long)mp->m_resblks_avail + delta;
1144 if (lcounter >= 0) {
1145 mp->m_resblks_avail = lcounter;
1146 spin_unlock(&mp->m_sb_lock);
1150 "Reserve blocks depleted! Consider increasing reserve pool size.");
1153 spin_unlock(&mp->m_sb_lock);
1159 struct xfs_mount *mp,
1165 spin_lock(&mp->m_sb_lock);
1166 lcounter = mp->m_sb.sb_frextents + delta;
1170 mp->m_sb.sb_frextents = lcounter;
1171 spin_unlock(&mp->m_sb_lock);
1176 * Used to free the superblock along various error paths.
1180 struct xfs_mount *mp)
1182 struct xfs_buf *bp = mp->m_sb_bp;
1190 * If the underlying (data/log/rt) device is readonly, there are some
1191 * operations that cannot proceed.
1194 xfs_dev_is_read_only(
1195 struct xfs_mount *mp,
1198 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1199 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1200 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1201 xfs_notice(mp, "%s required on read-only device.", message);
1202 xfs_notice(mp, "write access unavailable, cannot proceed.");
1208 /* Force the summary counters to be recalculated at next mount. */
1210 xfs_force_summary_recalc(
1211 struct xfs_mount *mp)
1213 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1216 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1220 * Update the in-core delayed block counter.
1222 * We prefer to update the counter without having to take a spinlock for every
1223 * counter update (i.e. batching). Each change to delayed allocation
1224 * reservations can change can easily exceed the default percpu counter
1225 * batching, so we use a larger batch factor here.
1227 * Note that we don't currently have any callers requiring fast summation
1228 * (e.g. percpu_counter_read) so we can use a big batch value here.
1230 #define XFS_DELALLOC_BATCH (4096)
1233 struct xfs_mount *mp,
1236 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1237 XFS_DELALLOC_BATCH);