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"
36 static DEFINE_MUTEX(xfs_uuid_table_mutex);
37 static int xfs_uuid_table_size;
38 static uuid_t *xfs_uuid_table;
41 xfs_uuid_table_free(void)
43 if (xfs_uuid_table_size == 0)
45 kmem_free(xfs_uuid_table);
46 xfs_uuid_table = NULL;
47 xfs_uuid_table_size = 0;
51 * See if the UUID is unique among mounted XFS filesystems.
52 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
58 uuid_t *uuid = &mp->m_sb.sb_uuid;
61 /* Publish UUID in struct super_block */
62 uuid_copy(&mp->m_super->s_uuid, uuid);
64 if (mp->m_flags & XFS_MOUNT_NOUUID)
67 if (uuid_is_null(uuid)) {
68 xfs_warn(mp, "Filesystem has null UUID - can't mount");
72 mutex_lock(&xfs_uuid_table_mutex);
73 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
74 if (uuid_is_null(&xfs_uuid_table[i])) {
78 if (uuid_equal(uuid, &xfs_uuid_table[i]))
83 xfs_uuid_table = krealloc(xfs_uuid_table,
84 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
85 GFP_KERNEL | __GFP_NOFAIL);
86 hole = xfs_uuid_table_size++;
88 xfs_uuid_table[hole] = *uuid;
89 mutex_unlock(&xfs_uuid_table_mutex);
94 mutex_unlock(&xfs_uuid_table_mutex);
95 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
101 struct xfs_mount *mp)
103 uuid_t *uuid = &mp->m_sb.sb_uuid;
106 if (mp->m_flags & XFS_MOUNT_NOUUID)
109 mutex_lock(&xfs_uuid_table_mutex);
110 for (i = 0; i < xfs_uuid_table_size; i++) {
111 if (uuid_is_null(&xfs_uuid_table[i]))
113 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
115 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
118 ASSERT(i < xfs_uuid_table_size);
119 mutex_unlock(&xfs_uuid_table_mutex);
125 struct rcu_head *head)
127 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
129 ASSERT(atomic_read(&pag->pag_ref) == 0);
134 * Free up the per-ag resources associated with the mount structure.
141 struct xfs_perag *pag;
143 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
144 spin_lock(&mp->m_perag_lock);
145 pag = radix_tree_delete(&mp->m_perag_tree, agno);
146 spin_unlock(&mp->m_perag_lock);
148 ASSERT(atomic_read(&pag->pag_ref) == 0);
149 xfs_iunlink_destroy(pag);
150 xfs_buf_hash_destroy(pag);
151 call_rcu(&pag->rcu_head, __xfs_free_perag);
156 * Check size of device based on the (data/realtime) block count.
157 * Note: this check is used by the growfs code as well as mount.
160 xfs_sb_validate_fsb_count(
164 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
165 ASSERT(sbp->sb_blocklog >= BBSHIFT);
167 /* Limited by ULONG_MAX of page cache index */
168 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
174 xfs_initialize_perag(
176 xfs_agnumber_t agcount,
177 xfs_agnumber_t *maxagi)
179 xfs_agnumber_t index;
180 xfs_agnumber_t first_initialised = NULLAGNUMBER;
185 * Walk the current per-ag tree so we don't try to initialise AGs
186 * that already exist (growfs case). Allocate and insert all the
187 * AGs we don't find ready for initialisation.
189 for (index = 0; index < agcount; index++) {
190 pag = xfs_perag_get(mp, index);
196 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
198 goto out_unwind_new_pags;
199 pag->pag_agno = index;
201 spin_lock_init(&pag->pag_ici_lock);
202 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
203 if (xfs_buf_hash_init(pag))
205 init_waitqueue_head(&pag->pagb_wait);
206 spin_lock_init(&pag->pagb_lock);
208 pag->pagb_tree = RB_ROOT;
210 if (radix_tree_preload(GFP_NOFS))
211 goto out_hash_destroy;
213 spin_lock(&mp->m_perag_lock);
214 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
216 spin_unlock(&mp->m_perag_lock);
217 radix_tree_preload_end();
219 goto out_hash_destroy;
221 spin_unlock(&mp->m_perag_lock);
222 radix_tree_preload_end();
223 /* first new pag is fully initialized */
224 if (first_initialised == NULLAGNUMBER)
225 first_initialised = index;
226 error = xfs_iunlink_init(pag);
228 goto out_hash_destroy;
229 spin_lock_init(&pag->pag_state_lock);
232 index = xfs_set_inode_alloc(mp, agcount);
237 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
241 xfs_buf_hash_destroy(pag);
245 /* unwind any prior newly initialized pags */
246 for (index = first_initialised; index < agcount; index++) {
247 pag = radix_tree_delete(&mp->m_perag_tree, index);
250 xfs_buf_hash_destroy(pag);
251 xfs_iunlink_destroy(pag);
260 * Does the initial read of the superblock.
264 struct xfs_mount *mp,
267 unsigned int sector_size;
269 struct xfs_sb *sbp = &mp->m_sb;
271 int loud = !(flags & XFS_MFSI_QUIET);
272 const struct xfs_buf_ops *buf_ops;
274 ASSERT(mp->m_sb_bp == NULL);
275 ASSERT(mp->m_ddev_targp != NULL);
278 * For the initial read, we must guess at the sector
279 * size based on the block device. It's enough to
280 * get the sb_sectsize out of the superblock and
281 * then reread with the proper length.
282 * We don't verify it yet, because it may not be complete.
284 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
288 * Allocate a (locked) buffer to hold the superblock. This will be kept
289 * around at all times to optimize access to the superblock. Therefore,
290 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
294 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
295 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
299 xfs_warn(mp, "SB validate failed with error %d.", error);
300 /* bad CRC means corrupted metadata */
301 if (error == -EFSBADCRC)
302 error = -EFSCORRUPTED;
307 * Initialize the mount structure from the superblock.
309 xfs_sb_from_disk(sbp, bp->b_addr);
312 * If we haven't validated the superblock, do so now before we try
313 * to check the sector size and reread the superblock appropriately.
315 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
317 xfs_warn(mp, "Invalid superblock magic number");
323 * We must be able to do sector-sized and sector-aligned IO.
325 if (sector_size > sbp->sb_sectsize) {
327 xfs_warn(mp, "device supports %u byte sectors (not %u)",
328 sector_size, sbp->sb_sectsize);
333 if (buf_ops == NULL) {
335 * Re-read the superblock so the buffer is correctly sized,
336 * and properly verified.
339 sector_size = sbp->sb_sectsize;
340 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
344 xfs_reinit_percpu_counters(mp);
346 /* no need to be quiet anymore, so reset the buf ops */
347 bp->b_ops = &xfs_sb_buf_ops;
359 * If the sunit/swidth change would move the precomputed root inode value, we
360 * must reject the ondisk change because repair will stumble over that.
361 * However, we allow the mount to proceed because we never rejected this
362 * combination before. Returns true to update the sb, false otherwise.
365 xfs_check_new_dalign(
366 struct xfs_mount *mp,
370 struct xfs_sb *sbp = &mp->m_sb;
373 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
374 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
376 if (sbp->sb_rootino == calc_ino) {
382 "Cannot change stripe alignment; would require moving root inode.");
385 * XXX: Next time we add a new incompat feature, this should start
386 * returning -EINVAL to fail the mount. Until then, spit out a warning
387 * that we're ignoring the administrator's instructions.
389 xfs_warn(mp, "Skipping superblock stripe alignment update.");
395 * If we were provided with new sunit/swidth values as mount options, make sure
396 * that they pass basic alignment and superblock feature checks, and convert
397 * them into the same units (FSB) that everything else expects. This step
398 * /must/ be done before computing the inode geometry.
401 xfs_validate_new_dalign(
402 struct xfs_mount *mp)
404 if (mp->m_dalign == 0)
408 * If stripe unit and stripe width are not multiples
409 * of the fs blocksize turn off alignment.
411 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
412 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
414 "alignment check failed: sunit/swidth vs. blocksize(%d)",
415 mp->m_sb.sb_blocksize);
419 * Convert the stripe unit and width to FSBs.
421 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
422 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
424 "alignment check failed: sunit/swidth vs. agsize(%d)",
425 mp->m_sb.sb_agblocks);
427 } else if (mp->m_dalign) {
428 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
431 "alignment check failed: sunit(%d) less than bsize(%d)",
432 mp->m_dalign, mp->m_sb.sb_blocksize);
437 if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
439 "cannot change alignment: superblock does not support data alignment");
446 /* Update alignment values based on mount options and sb values. */
448 xfs_update_alignment(
449 struct xfs_mount *mp)
451 struct xfs_sb *sbp = &mp->m_sb;
457 if (sbp->sb_unit == mp->m_dalign &&
458 sbp->sb_width == mp->m_swidth)
461 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
462 if (error || !update_sb)
465 sbp->sb_unit = mp->m_dalign;
466 sbp->sb_width = mp->m_swidth;
467 mp->m_update_sb = true;
468 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
469 xfs_sb_version_hasdalign(&mp->m_sb)) {
470 mp->m_dalign = sbp->sb_unit;
471 mp->m_swidth = sbp->sb_width;
478 * precalculate the low space thresholds for dynamic speculative preallocation.
481 xfs_set_low_space_thresholds(
482 struct xfs_mount *mp)
486 for (i = 0; i < XFS_LOWSP_MAX; i++) {
487 uint64_t space = mp->m_sb.sb_dblocks;
490 mp->m_low_space[i] = space * (i + 1);
495 * Check that the data (and log if separate) is an ok size.
499 struct xfs_mount *mp)
505 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
506 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
507 xfs_warn(mp, "filesystem size mismatch detected");
510 error = xfs_buf_read_uncached(mp->m_ddev_targp,
511 d - XFS_FSS_TO_BB(mp, 1),
512 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
514 xfs_warn(mp, "last sector read failed");
519 if (mp->m_logdev_targp == mp->m_ddev_targp)
522 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
523 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
524 xfs_warn(mp, "log size mismatch detected");
527 error = xfs_buf_read_uncached(mp->m_logdev_targp,
528 d - XFS_FSB_TO_BB(mp, 1),
529 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
531 xfs_warn(mp, "log device read failed");
539 * Clear the quotaflags in memory and in the superblock.
542 xfs_mount_reset_sbqflags(
543 struct xfs_mount *mp)
547 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
548 if (mp->m_sb.sb_qflags == 0)
550 spin_lock(&mp->m_sb_lock);
551 mp->m_sb.sb_qflags = 0;
552 spin_unlock(&mp->m_sb_lock);
554 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
557 return xfs_sync_sb(mp, false);
561 xfs_default_resblks(xfs_mount_t *mp)
566 * We default to 5% or 8192 fsbs of space reserved, whichever is
567 * smaller. This is intended to cover concurrent allocation
568 * transactions when we initially hit enospc. These each require a 4
569 * block reservation. Hence by default we cover roughly 2000 concurrent
570 * allocation reservations.
572 resblks = mp->m_sb.sb_dblocks;
574 resblks = min_t(uint64_t, resblks, 8192);
578 /* Ensure the summary counts are correct. */
580 xfs_check_summary_counts(
581 struct xfs_mount *mp)
584 * The AG0 superblock verifier rejects in-progress filesystems,
585 * so we should never see the flag set this far into mounting.
587 if (mp->m_sb.sb_inprogress) {
588 xfs_err(mp, "sb_inprogress set after log recovery??");
590 return -EFSCORRUPTED;
594 * Now the log is mounted, we know if it was an unclean shutdown or
595 * not. If it was, with the first phase of recovery has completed, we
596 * have consistent AG blocks on disk. We have not recovered EFIs yet,
597 * but they are recovered transactionally in the second recovery phase
600 * If the log was clean when we mounted, we can check the summary
601 * counters. If any of them are obviously incorrect, we can recompute
602 * them from the AGF headers in the next step.
604 if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
605 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
606 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
607 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
608 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
611 * We can safely re-initialise incore superblock counters from the
612 * per-ag data. These may not be correct if the filesystem was not
613 * cleanly unmounted, so we waited for recovery to finish before doing
616 * If the filesystem was cleanly unmounted or the previous check did
617 * not flag anything weird, then we can trust the values in the
618 * superblock to be correct and we don't need to do anything here.
619 * Otherwise, recalculate the summary counters.
621 if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
622 XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
623 !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
626 return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
630 * This function does the following on an initial mount of a file system:
631 * - reads the superblock from disk and init the mount struct
632 * - if we're a 32-bit kernel, do a size check on the superblock
633 * so we don't mount terabyte filesystems
634 * - init mount struct realtime fields
635 * - allocate inode hash table for fs
636 * - init directory manager
637 * - perform recovery and init the log manager
641 struct xfs_mount *mp)
643 struct xfs_sb *sbp = &(mp->m_sb);
644 struct xfs_inode *rip;
645 struct xfs_ino_geometry *igeo = M_IGEO(mp);
651 xfs_sb_mount_common(mp, sbp);
654 * Check for a mismatched features2 values. Older kernels read & wrote
655 * into the wrong sb offset for sb_features2 on some platforms due to
656 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
657 * which made older superblock reading/writing routines swap it as a
660 * For backwards compatibility, we make both slots equal.
662 * If we detect a mismatched field, we OR the set bits into the existing
663 * features2 field in case it has already been modified; we don't want
664 * to lose any features. We then update the bad location with the ORed
665 * value so that older kernels will see any features2 flags. The
666 * superblock writeback code ensures the new sb_features2 is copied to
667 * sb_bad_features2 before it is logged or written to disk.
669 if (xfs_sb_has_mismatched_features2(sbp)) {
670 xfs_warn(mp, "correcting sb_features alignment problem");
671 sbp->sb_features2 |= sbp->sb_bad_features2;
672 mp->m_update_sb = true;
675 * Re-check for ATTR2 in case it was found in bad_features2
678 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
679 !(mp->m_flags & XFS_MOUNT_NOATTR2))
680 mp->m_flags |= XFS_MOUNT_ATTR2;
683 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
684 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
685 xfs_sb_version_removeattr2(&mp->m_sb);
686 mp->m_update_sb = true;
688 /* update sb_versionnum for the clearing of the morebits */
689 if (!sbp->sb_features2)
690 mp->m_update_sb = true;
693 /* always use v2 inodes by default now */
694 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
695 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
696 mp->m_update_sb = true;
700 * If we were given new sunit/swidth options, do some basic validation
701 * checks and convert the incore dalign and swidth values to the
702 * same units (FSB) that everything else uses. This /must/ happen
703 * before computing the inode geometry.
705 error = xfs_validate_new_dalign(mp);
709 xfs_alloc_compute_maxlevels(mp);
710 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
711 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
712 xfs_ialloc_setup_geometry(mp);
713 xfs_rmapbt_compute_maxlevels(mp);
714 xfs_refcountbt_compute_maxlevels(mp);
717 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
718 * is NOT aligned turn off m_dalign since allocator alignment is within
719 * an ag, therefore ag has to be aligned at stripe boundary. Note that
720 * we must compute the free space and rmap btree geometry before doing
723 error = xfs_update_alignment(mp);
727 /* enable fail_at_unmount as default */
728 mp->m_fail_unmount = true;
730 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
731 NULL, mp->m_super->s_id);
735 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
736 &mp->m_kobj, "stats");
738 goto out_remove_sysfs;
740 error = xfs_error_sysfs_init(mp);
744 error = xfs_errortag_init(mp);
746 goto out_remove_error_sysfs;
748 error = xfs_uuid_mount(mp);
750 goto out_remove_errortag;
753 * Update the preferred write size based on the information from the
754 * on-disk superblock.
756 mp->m_allocsize_log =
757 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
758 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
760 /* set the low space thresholds for dynamic preallocation */
761 xfs_set_low_space_thresholds(mp);
764 * If enabled, sparse inode chunk alignment is expected to match the
765 * cluster size. Full inode chunk alignment must match the chunk size,
766 * but that is checked on sb read verification...
768 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
769 mp->m_sb.sb_spino_align !=
770 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
772 "Sparse inode block alignment (%u) must match cluster size (%llu).",
773 mp->m_sb.sb_spino_align,
774 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
776 goto out_remove_uuid;
780 * Check that the data (and log if separate) is an ok size.
782 error = xfs_check_sizes(mp);
784 goto out_remove_uuid;
787 * Initialize realtime fields in the mount structure
789 error = xfs_rtmount_init(mp);
791 xfs_warn(mp, "RT mount failed");
792 goto out_remove_uuid;
796 * Copies the low order bits of the timestamp and the randomly
797 * set "sequence" number out of a UUID.
800 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
801 get_unaligned_be16(&sbp->sb_uuid.b[4]);
802 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
804 error = xfs_da_mount(mp);
806 xfs_warn(mp, "Failed dir/attr init: %d", error);
807 goto out_remove_uuid;
811 * Initialize the precomputed transaction reservations values.
816 * Allocate and initialize the per-ag data.
818 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
820 xfs_warn(mp, "Failed per-ag init: %d", error);
824 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
825 xfs_warn(mp, "no log defined");
826 error = -EFSCORRUPTED;
831 * Log's mount-time initialization. The first part of recovery can place
832 * some items on the AIL, to be handled when recovery is finished or
835 error = xfs_log_mount(mp, mp->m_logdev_targp,
836 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
837 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
839 xfs_warn(mp, "log mount failed");
843 /* Make sure the summary counts are ok. */
844 error = xfs_check_summary_counts(mp);
846 goto out_log_dealloc;
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");
887 * If this is a read-only mount defer the superblock updates until
888 * the next remount into writeable mode. Otherwise we would never
889 * perform the update e.g. for the root filesystem.
891 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
892 error = xfs_sync_sb(mp, false);
894 xfs_warn(mp, "failed to write sb changes");
900 * Initialise the XFS quota management subsystem for this mount
902 if (XFS_IS_QUOTA_RUNNING(mp)) {
903 error = xfs_qm_newmount(mp, "amount, "aflags);
907 ASSERT(!XFS_IS_QUOTA_ON(mp));
910 * If a file system had quotas running earlier, but decided to
911 * mount without -o uquota/pquota/gquota options, revoke the
912 * quotachecked license.
914 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
915 xfs_notice(mp, "resetting quota flags");
916 error = xfs_mount_reset_sbqflags(mp);
923 * Finish recovering the file system. This part needed to be delayed
924 * until after the root and real-time bitmap inodes were consistently
927 error = xfs_log_mount_finish(mp);
929 xfs_warn(mp, "log mount finish failed");
934 * Now the log is fully replayed, we can transition to full read-only
935 * mode for read-only mounts. This will sync all the metadata and clean
936 * the log so that the recovery we just performed does not have to be
937 * replayed again on the next mount.
939 * We use the same quiesce mechanism as the rw->ro remount, as they are
940 * semantically identical operations.
942 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
944 xfs_quiesce_attr(mp);
948 * Complete the quota initialisation, post-log-replay component.
951 ASSERT(mp->m_qflags == 0);
952 mp->m_qflags = quotaflags;
954 xfs_qm_mount_quotas(mp);
958 * Now we are mounted, reserve a small amount of unused space for
959 * privileged transactions. This is needed so that transaction
960 * space required for critical operations can dip into this pool
961 * when at ENOSPC. This is needed for operations like create with
962 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
963 * are not allowed to use this reserved space.
965 * This may drive us straight to ENOSPC on mount, but that implies
966 * we were already there on the last unmount. Warn if this occurs.
968 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
969 resblks = xfs_default_resblks(mp);
970 error = xfs_reserve_blocks(mp, &resblks, NULL);
973 "Unable to allocate reserve blocks. Continuing without reserve pool.");
975 /* Recover any CoW blocks that never got remapped. */
976 error = xfs_reflink_recover_cow(mp);
979 "Error %d recovering leftover CoW allocations.", error);
980 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
984 /* Reserve AG blocks for future btree expansion. */
985 error = xfs_fs_reserve_ag_blocks(mp);
986 if (error && error != -ENOSPC)
993 xfs_fs_unreserve_ag_blocks(mp);
995 xfs_qm_unmount_quotas(mp);
997 xfs_rtunmount_inodes(mp);
1000 /* Clean out dquots that might be in memory after quotacheck. */
1003 * Cancel all delayed reclaim work and reclaim the inodes directly.
1004 * We have to do this /after/ rtunmount and qm_unmount because those
1005 * two will have scheduled delayed reclaim for the rt/quota inodes.
1007 * This is slightly different from the unmountfs call sequence
1008 * because we could be tearing down a partially set up mount. In
1009 * particular, if log_mount_finish fails we bail out without calling
1010 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1013 cancel_delayed_work_sync(&mp->m_reclaim_work);
1014 xfs_reclaim_inodes(mp);
1015 xfs_health_unmount(mp);
1017 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1018 xfs_log_mount_cancel(mp);
1020 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1021 xfs_wait_buftarg(mp->m_logdev_targp);
1022 xfs_wait_buftarg(mp->m_ddev_targp);
1028 xfs_uuid_unmount(mp);
1029 out_remove_errortag:
1030 xfs_errortag_del(mp);
1031 out_remove_error_sysfs:
1032 xfs_error_sysfs_del(mp);
1034 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1036 xfs_sysfs_del(&mp->m_kobj);
1042 * This flushes out the inodes,dquots and the superblock, unmounts the
1043 * log and makes sure that incore structures are freed.
1047 struct xfs_mount *mp)
1052 xfs_stop_block_reaping(mp);
1053 xfs_fs_unreserve_ag_blocks(mp);
1054 xfs_qm_unmount_quotas(mp);
1055 xfs_rtunmount_inodes(mp);
1056 xfs_irele(mp->m_rootip);
1059 * We can potentially deadlock here if we have an inode cluster
1060 * that has been freed has its buffer still pinned in memory because
1061 * the transaction is still sitting in a iclog. The stale inodes
1062 * on that buffer will be pinned to the buffer until the
1063 * transaction hits the disk and the callbacks run. Pushing the AIL will
1064 * skip the stale inodes and may never see the pinned buffer, so
1065 * nothing will push out the iclog and unpin the buffer. Hence we
1066 * need to force the log here to ensure all items are flushed into the
1067 * AIL before we go any further.
1069 xfs_log_force(mp, XFS_LOG_SYNC);
1072 * Wait for all busy extents to be freed, including completion of
1073 * any discard operation.
1075 xfs_extent_busy_wait_all(mp);
1076 flush_workqueue(xfs_discard_wq);
1079 * We now need to tell the world we are unmounting. This will allow
1080 * us to detect that the filesystem is going away and we should error
1081 * out anything that we have been retrying in the background. This will
1082 * prevent neverending retries in AIL pushing from hanging the unmount.
1084 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1087 * Flush all pending changes from the AIL.
1089 xfs_ail_push_all_sync(mp->m_ail);
1092 * Reclaim all inodes. At this point there should be no dirty inodes and
1093 * none should be pinned or locked. Stop background inode reclaim here
1094 * if it is still running.
1096 cancel_delayed_work_sync(&mp->m_reclaim_work);
1097 xfs_reclaim_inodes(mp);
1098 xfs_health_unmount(mp);
1103 * Unreserve any blocks we have so that when we unmount we don't account
1104 * the reserved free space as used. This is really only necessary for
1105 * lazy superblock counting because it trusts the incore superblock
1106 * counters to be absolutely correct on clean unmount.
1108 * We don't bother correcting this elsewhere for lazy superblock
1109 * counting because on mount of an unclean filesystem we reconstruct the
1110 * correct counter value and this is irrelevant.
1112 * For non-lazy counter filesystems, this doesn't matter at all because
1113 * we only every apply deltas to the superblock and hence the incore
1114 * value does not matter....
1117 error = xfs_reserve_blocks(mp, &resblks, NULL);
1119 xfs_warn(mp, "Unable to free reserved block pool. "
1120 "Freespace may not be correct on next mount.");
1122 error = xfs_log_sbcount(mp);
1124 xfs_warn(mp, "Unable to update superblock counters. "
1125 "Freespace may not be correct on next mount.");
1128 xfs_log_unmount(mp);
1130 xfs_uuid_unmount(mp);
1133 xfs_errortag_clearall(mp);
1137 xfs_errortag_del(mp);
1138 xfs_error_sysfs_del(mp);
1139 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1140 xfs_sysfs_del(&mp->m_kobj);
1144 * Determine whether modifications can proceed. The caller specifies the minimum
1145 * freeze level for which modifications should not be allowed. This allows
1146 * certain operations to proceed while the freeze sequence is in progress, if
1151 struct xfs_mount *mp,
1154 ASSERT(level > SB_UNFROZEN);
1155 if ((mp->m_super->s_writers.frozen >= level) ||
1156 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1165 * Sync the superblock counters to disk.
1167 * Note this code can be called during the process of freezing, so we use the
1168 * transaction allocator that does not block when the transaction subsystem is
1169 * in its frozen state.
1172 xfs_log_sbcount(xfs_mount_t *mp)
1174 /* allow this to proceed during the freeze sequence... */
1175 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1179 * we don't need to do this if we are updating the superblock
1180 * counters on every modification.
1182 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1185 return xfs_sync_sb(mp, true);
1189 * Deltas for the block count can vary from 1 to very large, but lock contention
1190 * only occurs on frequent small block count updates such as in the delayed
1191 * allocation path for buffered writes (page a time updates). Hence we set
1192 * a large batch count (1024) to minimise global counter updates except when
1193 * we get near to ENOSPC and we have to be very accurate with our updates.
1195 #define XFS_FDBLOCKS_BATCH 1024
1198 struct xfs_mount *mp,
1208 * If the reserve pool is depleted, put blocks back into it
1209 * first. Most of the time the pool is full.
1211 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1212 percpu_counter_add(&mp->m_fdblocks, delta);
1216 spin_lock(&mp->m_sb_lock);
1217 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1219 if (res_used > delta) {
1220 mp->m_resblks_avail += delta;
1223 mp->m_resblks_avail = mp->m_resblks;
1224 percpu_counter_add(&mp->m_fdblocks, delta);
1226 spin_unlock(&mp->m_sb_lock);
1231 * Taking blocks away, need to be more accurate the closer we
1234 * If the counter has a value of less than 2 * max batch size,
1235 * then make everything serialise as we are real close to
1238 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1239 XFS_FDBLOCKS_BATCH) < 0)
1242 batch = XFS_FDBLOCKS_BATCH;
1244 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1245 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1246 XFS_FDBLOCKS_BATCH) >= 0) {
1252 * lock up the sb for dipping into reserves before releasing the space
1253 * that took us to ENOSPC.
1255 spin_lock(&mp->m_sb_lock);
1256 percpu_counter_add(&mp->m_fdblocks, -delta);
1258 goto fdblocks_enospc;
1260 lcounter = (long long)mp->m_resblks_avail + delta;
1261 if (lcounter >= 0) {
1262 mp->m_resblks_avail = lcounter;
1263 spin_unlock(&mp->m_sb_lock);
1267 "Reserve blocks depleted! Consider increasing reserve pool size.");
1270 spin_unlock(&mp->m_sb_lock);
1276 struct xfs_mount *mp,
1282 spin_lock(&mp->m_sb_lock);
1283 lcounter = mp->m_sb.sb_frextents + delta;
1287 mp->m_sb.sb_frextents = lcounter;
1288 spin_unlock(&mp->m_sb_lock);
1293 * Used to free the superblock along various error paths.
1297 struct xfs_mount *mp)
1299 struct xfs_buf *bp = mp->m_sb_bp;
1307 * If the underlying (data/log/rt) device is readonly, there are some
1308 * operations that cannot proceed.
1311 xfs_dev_is_read_only(
1312 struct xfs_mount *mp,
1315 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1316 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1317 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1318 xfs_notice(mp, "%s required on read-only device.", message);
1319 xfs_notice(mp, "write access unavailable, cannot proceed.");
1325 /* Force the summary counters to be recalculated at next mount. */
1327 xfs_force_summary_recalc(
1328 struct xfs_mount *mp)
1330 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1333 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1337 * Update the in-core delayed block counter.
1339 * We prefer to update the counter without having to take a spinlock for every
1340 * counter update (i.e. batching). Each change to delayed allocation
1341 * reservations can change can easily exceed the default percpu counter
1342 * batching, so we use a larger batch factor here.
1344 * Note that we don't currently have any callers requiring fast summation
1345 * (e.g. percpu_counter_read) so we can use a big batch value here.
1347 #define XFS_DELALLOC_BATCH (4096)
1350 struct xfs_mount *mp,
1353 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1354 XFS_DELALLOC_BATCH);