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"
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 = kmem_realloc(xfs_uuid_table,
84 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
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 mutex_destroy(&pag->pag_ici_reclaim_lock);
152 call_rcu(&pag->rcu_head, __xfs_free_perag);
157 * Check size of device based on the (data/realtime) block count.
158 * Note: this check is used by the growfs code as well as mount.
161 xfs_sb_validate_fsb_count(
165 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
166 ASSERT(sbp->sb_blocklog >= BBSHIFT);
168 /* Limited by ULONG_MAX of page cache index */
169 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
175 xfs_initialize_perag(
177 xfs_agnumber_t agcount,
178 xfs_agnumber_t *maxagi)
180 xfs_agnumber_t index;
181 xfs_agnumber_t first_initialised = NULLAGNUMBER;
186 * Walk the current per-ag tree so we don't try to initialise AGs
187 * that already exist (growfs case). Allocate and insert all the
188 * AGs we don't find ready for initialisation.
190 for (index = 0; index < agcount; index++) {
191 pag = xfs_perag_get(mp, index);
197 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
199 goto out_unwind_new_pags;
200 pag->pag_agno = index;
202 spin_lock_init(&pag->pag_ici_lock);
203 mutex_init(&pag->pag_ici_reclaim_lock);
204 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
205 if (xfs_buf_hash_init(pag))
207 init_waitqueue_head(&pag->pagb_wait);
208 spin_lock_init(&pag->pagb_lock);
210 pag->pagb_tree = RB_ROOT;
212 if (radix_tree_preload(GFP_NOFS))
213 goto out_hash_destroy;
215 spin_lock(&mp->m_perag_lock);
216 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
218 spin_unlock(&mp->m_perag_lock);
219 radix_tree_preload_end();
221 goto out_hash_destroy;
223 spin_unlock(&mp->m_perag_lock);
224 radix_tree_preload_end();
225 /* first new pag is fully initialized */
226 if (first_initialised == NULLAGNUMBER)
227 first_initialised = index;
228 error = xfs_iunlink_init(pag);
230 goto out_hash_destroy;
231 spin_lock_init(&pag->pag_state_lock);
234 index = xfs_set_inode_alloc(mp, agcount);
239 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
243 xfs_buf_hash_destroy(pag);
245 mutex_destroy(&pag->pag_ici_reclaim_lock);
248 /* unwind any prior newly initialized pags */
249 for (index = first_initialised; index < agcount; index++) {
250 pag = radix_tree_delete(&mp->m_perag_tree, index);
253 xfs_buf_hash_destroy(pag);
254 xfs_iunlink_destroy(pag);
255 mutex_destroy(&pag->pag_ici_reclaim_lock);
264 * Does the initial read of the superblock.
268 struct xfs_mount *mp,
271 unsigned int sector_size;
273 struct xfs_sb *sbp = &mp->m_sb;
275 int loud = !(flags & XFS_MFSI_QUIET);
276 const struct xfs_buf_ops *buf_ops;
278 ASSERT(mp->m_sb_bp == NULL);
279 ASSERT(mp->m_ddev_targp != NULL);
282 * For the initial read, we must guess at the sector
283 * size based on the block device. It's enough to
284 * get the sb_sectsize out of the superblock and
285 * then reread with the proper length.
286 * We don't verify it yet, because it may not be complete.
288 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
292 * Allocate a (locked) buffer to hold the superblock. This will be kept
293 * around at all times to optimize access to the superblock. Therefore,
294 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
298 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
299 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
303 xfs_warn(mp, "SB validate failed with error %d.", error);
304 /* bad CRC means corrupted metadata */
305 if (error == -EFSBADCRC)
306 error = -EFSCORRUPTED;
311 * Initialize the mount structure from the superblock.
313 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
316 * If we haven't validated the superblock, do so now before we try
317 * to check the sector size and reread the superblock appropriately.
319 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
321 xfs_warn(mp, "Invalid superblock magic number");
327 * We must be able to do sector-sized and sector-aligned IO.
329 if (sector_size > sbp->sb_sectsize) {
331 xfs_warn(mp, "device supports %u byte sectors (not %u)",
332 sector_size, sbp->sb_sectsize);
337 if (buf_ops == NULL) {
339 * Re-read the superblock so the buffer is correctly sized,
340 * and properly verified.
343 sector_size = sbp->sb_sectsize;
344 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
348 xfs_reinit_percpu_counters(mp);
350 /* no need to be quiet anymore, so reset the buf ops */
351 bp->b_ops = &xfs_sb_buf_ops;
363 * Update alignment values based on mount options and sb values
366 xfs_update_alignment(xfs_mount_t *mp)
368 xfs_sb_t *sbp = &(mp->m_sb);
372 * If stripe unit and stripe width are not multiples
373 * of the fs blocksize turn off alignment.
375 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
376 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
378 "alignment check failed: sunit/swidth vs. blocksize(%d)",
383 * Convert the stripe unit and width to FSBs.
385 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
386 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
388 "alignment check failed: sunit/swidth vs. agsize(%d)",
391 } else if (mp->m_dalign) {
392 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
395 "alignment check failed: sunit(%d) less than bsize(%d)",
396 mp->m_dalign, sbp->sb_blocksize);
402 * Update superblock with new values
405 if (xfs_sb_version_hasdalign(sbp)) {
406 if (sbp->sb_unit != mp->m_dalign) {
407 sbp->sb_unit = mp->m_dalign;
408 mp->m_update_sb = true;
410 if (sbp->sb_width != mp->m_swidth) {
411 sbp->sb_width = mp->m_swidth;
412 mp->m_update_sb = true;
416 "cannot change alignment: superblock does not support data alignment");
419 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
420 xfs_sb_version_hasdalign(&mp->m_sb)) {
421 mp->m_dalign = sbp->sb_unit;
422 mp->m_swidth = sbp->sb_width;
429 * precalculate the low space thresholds for dynamic speculative preallocation.
432 xfs_set_low_space_thresholds(
433 struct xfs_mount *mp)
437 for (i = 0; i < XFS_LOWSP_MAX; i++) {
438 uint64_t space = mp->m_sb.sb_dblocks;
441 mp->m_low_space[i] = space * (i + 1);
446 * Check that the data (and log if separate) is an ok size.
450 struct xfs_mount *mp)
456 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
457 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
458 xfs_warn(mp, "filesystem size mismatch detected");
461 error = xfs_buf_read_uncached(mp->m_ddev_targp,
462 d - XFS_FSS_TO_BB(mp, 1),
463 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
465 xfs_warn(mp, "last sector read failed");
470 if (mp->m_logdev_targp == mp->m_ddev_targp)
473 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
474 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
475 xfs_warn(mp, "log size mismatch detected");
478 error = xfs_buf_read_uncached(mp->m_logdev_targp,
479 d - XFS_FSB_TO_BB(mp, 1),
480 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
482 xfs_warn(mp, "log device read failed");
490 * Clear the quotaflags in memory and in the superblock.
493 xfs_mount_reset_sbqflags(
494 struct xfs_mount *mp)
498 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
499 if (mp->m_sb.sb_qflags == 0)
501 spin_lock(&mp->m_sb_lock);
502 mp->m_sb.sb_qflags = 0;
503 spin_unlock(&mp->m_sb_lock);
505 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
508 return xfs_sync_sb(mp, false);
512 xfs_default_resblks(xfs_mount_t *mp)
517 * We default to 5% or 8192 fsbs of space reserved, whichever is
518 * smaller. This is intended to cover concurrent allocation
519 * transactions when we initially hit enospc. These each require a 4
520 * block reservation. Hence by default we cover roughly 2000 concurrent
521 * allocation reservations.
523 resblks = mp->m_sb.sb_dblocks;
525 resblks = min_t(uint64_t, resblks, 8192);
529 /* Ensure the summary counts are correct. */
531 xfs_check_summary_counts(
532 struct xfs_mount *mp)
535 * The AG0 superblock verifier rejects in-progress filesystems,
536 * so we should never see the flag set this far into mounting.
538 if (mp->m_sb.sb_inprogress) {
539 xfs_err(mp, "sb_inprogress set after log recovery??");
541 return -EFSCORRUPTED;
545 * Now the log is mounted, we know if it was an unclean shutdown or
546 * not. If it was, with the first phase of recovery has completed, we
547 * have consistent AG blocks on disk. We have not recovered EFIs yet,
548 * but they are recovered transactionally in the second recovery phase
551 * If the log was clean when we mounted, we can check the summary
552 * counters. If any of them are obviously incorrect, we can recompute
553 * them from the AGF headers in the next step.
555 if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
556 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
557 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
558 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
559 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
562 * We can safely re-initialise incore superblock counters from the
563 * per-ag data. These may not be correct if the filesystem was not
564 * cleanly unmounted, so we waited for recovery to finish before doing
567 * If the filesystem was cleanly unmounted or the previous check did
568 * not flag anything weird, then we can trust the values in the
569 * superblock to be correct and we don't need to do anything here.
570 * Otherwise, recalculate the summary counters.
572 if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
573 XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
574 !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
577 return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
581 * This function does the following on an initial mount of a file system:
582 * - reads the superblock from disk and init the mount struct
583 * - if we're a 32-bit kernel, do a size check on the superblock
584 * so we don't mount terabyte filesystems
585 * - init mount struct realtime fields
586 * - allocate inode hash table for fs
587 * - init directory manager
588 * - perform recovery and init the log manager
592 struct xfs_mount *mp)
594 struct xfs_sb *sbp = &(mp->m_sb);
595 struct xfs_inode *rip;
596 struct xfs_ino_geometry *igeo = M_IGEO(mp);
602 xfs_sb_mount_common(mp, sbp);
605 * Check for a mismatched features2 values. Older kernels read & wrote
606 * into the wrong sb offset for sb_features2 on some platforms due to
607 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
608 * which made older superblock reading/writing routines swap it as a
611 * For backwards compatibility, we make both slots equal.
613 * If we detect a mismatched field, we OR the set bits into the existing
614 * features2 field in case it has already been modified; we don't want
615 * to lose any features. We then update the bad location with the ORed
616 * value so that older kernels will see any features2 flags. The
617 * superblock writeback code ensures the new sb_features2 is copied to
618 * sb_bad_features2 before it is logged or written to disk.
620 if (xfs_sb_has_mismatched_features2(sbp)) {
621 xfs_warn(mp, "correcting sb_features alignment problem");
622 sbp->sb_features2 |= sbp->sb_bad_features2;
623 mp->m_update_sb = true;
626 * Re-check for ATTR2 in case it was found in bad_features2
629 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
630 !(mp->m_flags & XFS_MOUNT_NOATTR2))
631 mp->m_flags |= XFS_MOUNT_ATTR2;
634 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
635 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
636 xfs_sb_version_removeattr2(&mp->m_sb);
637 mp->m_update_sb = true;
639 /* update sb_versionnum for the clearing of the morebits */
640 if (!sbp->sb_features2)
641 mp->m_update_sb = true;
644 /* always use v2 inodes by default now */
645 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
646 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
647 mp->m_update_sb = true;
651 * Check if sb_agblocks is aligned at stripe boundary
652 * If sb_agblocks is NOT aligned turn off m_dalign since
653 * allocator alignment is within an ag, therefore ag has
654 * to be aligned at stripe boundary.
656 error = xfs_update_alignment(mp);
660 xfs_alloc_compute_maxlevels(mp);
661 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
662 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
663 xfs_ialloc_setup_geometry(mp);
664 xfs_rmapbt_compute_maxlevels(mp);
665 xfs_refcountbt_compute_maxlevels(mp);
667 /* enable fail_at_unmount as default */
668 mp->m_fail_unmount = true;
670 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
671 NULL, mp->m_super->s_id);
675 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
676 &mp->m_kobj, "stats");
678 goto out_remove_sysfs;
680 error = xfs_error_sysfs_init(mp);
684 error = xfs_errortag_init(mp);
686 goto out_remove_error_sysfs;
688 error = xfs_uuid_mount(mp);
690 goto out_remove_errortag;
693 * Update the preferred write size based on the information from the
694 * on-disk superblock.
696 mp->m_allocsize_log =
697 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
698 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
700 /* set the low space thresholds for dynamic preallocation */
701 xfs_set_low_space_thresholds(mp);
704 * If enabled, sparse inode chunk alignment is expected to match the
705 * cluster size. Full inode chunk alignment must match the chunk size,
706 * but that is checked on sb read verification...
708 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
709 mp->m_sb.sb_spino_align !=
710 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
712 "Sparse inode block alignment (%u) must match cluster size (%llu).",
713 mp->m_sb.sb_spino_align,
714 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
716 goto out_remove_uuid;
720 * Check that the data (and log if separate) is an ok size.
722 error = xfs_check_sizes(mp);
724 goto out_remove_uuid;
727 * Initialize realtime fields in the mount structure
729 error = xfs_rtmount_init(mp);
731 xfs_warn(mp, "RT mount failed");
732 goto out_remove_uuid;
736 * Copies the low order bits of the timestamp and the randomly
737 * set "sequence" number out of a UUID.
740 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
741 get_unaligned_be16(&sbp->sb_uuid.b[4]);
742 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
744 error = xfs_da_mount(mp);
746 xfs_warn(mp, "Failed dir/attr init: %d", error);
747 goto out_remove_uuid;
751 * Initialize the precomputed transaction reservations values.
756 * Allocate and initialize the per-ag data.
758 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
760 xfs_warn(mp, "Failed per-ag init: %d", error);
764 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
765 xfs_warn(mp, "no log defined");
766 error = -EFSCORRUPTED;
771 * Log's mount-time initialization. The first part of recovery can place
772 * some items on the AIL, to be handled when recovery is finished or
775 error = xfs_log_mount(mp, mp->m_logdev_targp,
776 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
777 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
779 xfs_warn(mp, "log mount failed");
783 /* Make sure the summary counts are ok. */
784 error = xfs_check_summary_counts(mp);
786 goto out_log_dealloc;
789 * Get and sanity-check the root inode.
790 * Save the pointer to it in the mount structure.
792 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
793 XFS_ILOCK_EXCL, &rip);
796 "Failed to read root inode 0x%llx, error %d",
797 sbp->sb_rootino, -error);
798 goto out_log_dealloc;
803 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
804 xfs_warn(mp, "corrupted root inode %llu: not a directory",
805 (unsigned long long)rip->i_ino);
806 xfs_iunlock(rip, XFS_ILOCK_EXCL);
807 error = -EFSCORRUPTED;
810 mp->m_rootip = rip; /* save it */
812 xfs_iunlock(rip, XFS_ILOCK_EXCL);
815 * Initialize realtime inode pointers in the mount structure
817 error = xfs_rtmount_inodes(mp);
820 * Free up the root inode.
822 xfs_warn(mp, "failed to read RT inodes");
827 * If this is a read-only mount defer the superblock updates until
828 * the next remount into writeable mode. Otherwise we would never
829 * perform the update e.g. for the root filesystem.
831 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
832 error = xfs_sync_sb(mp, false);
834 xfs_warn(mp, "failed to write sb changes");
840 * Initialise the XFS quota management subsystem for this mount
842 if (XFS_IS_QUOTA_RUNNING(mp)) {
843 error = xfs_qm_newmount(mp, "amount, "aflags);
847 ASSERT(!XFS_IS_QUOTA_ON(mp));
850 * If a file system had quotas running earlier, but decided to
851 * mount without -o uquota/pquota/gquota options, revoke the
852 * quotachecked license.
854 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
855 xfs_notice(mp, "resetting quota flags");
856 error = xfs_mount_reset_sbqflags(mp);
863 * Finish recovering the file system. This part needed to be delayed
864 * until after the root and real-time bitmap inodes were consistently
867 error = xfs_log_mount_finish(mp);
869 xfs_warn(mp, "log mount finish failed");
874 * Now the log is fully replayed, we can transition to full read-only
875 * mode for read-only mounts. This will sync all the metadata and clean
876 * the log so that the recovery we just performed does not have to be
877 * replayed again on the next mount.
879 * We use the same quiesce mechanism as the rw->ro remount, as they are
880 * semantically identical operations.
882 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
884 xfs_quiesce_attr(mp);
888 * Complete the quota initialisation, post-log-replay component.
891 ASSERT(mp->m_qflags == 0);
892 mp->m_qflags = quotaflags;
894 xfs_qm_mount_quotas(mp);
898 * Now we are mounted, reserve a small amount of unused space for
899 * privileged transactions. This is needed so that transaction
900 * space required for critical operations can dip into this pool
901 * when at ENOSPC. This is needed for operations like create with
902 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
903 * are not allowed to use this reserved space.
905 * This may drive us straight to ENOSPC on mount, but that implies
906 * we were already there on the last unmount. Warn if this occurs.
908 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
909 resblks = xfs_default_resblks(mp);
910 error = xfs_reserve_blocks(mp, &resblks, NULL);
913 "Unable to allocate reserve blocks. Continuing without reserve pool.");
915 /* Recover any CoW blocks that never got remapped. */
916 error = xfs_reflink_recover_cow(mp);
919 "Error %d recovering leftover CoW allocations.", error);
920 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
924 /* Reserve AG blocks for future btree expansion. */
925 error = xfs_fs_reserve_ag_blocks(mp);
926 if (error && error != -ENOSPC)
933 xfs_fs_unreserve_ag_blocks(mp);
935 xfs_qm_unmount_quotas(mp);
937 xfs_rtunmount_inodes(mp);
940 /* Clean out dquots that might be in memory after quotacheck. */
943 * Cancel all delayed reclaim work and reclaim the inodes directly.
944 * We have to do this /after/ rtunmount and qm_unmount because those
945 * two will have scheduled delayed reclaim for the rt/quota inodes.
947 * This is slightly different from the unmountfs call sequence
948 * because we could be tearing down a partially set up mount. In
949 * particular, if log_mount_finish fails we bail out without calling
950 * qm_unmount_quotas and therefore rely on qm_unmount to release the
953 cancel_delayed_work_sync(&mp->m_reclaim_work);
954 xfs_reclaim_inodes(mp, SYNC_WAIT);
955 xfs_health_unmount(mp);
957 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
958 xfs_log_mount_cancel(mp);
960 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
961 xfs_wait_buftarg(mp->m_logdev_targp);
962 xfs_wait_buftarg(mp->m_ddev_targp);
968 xfs_uuid_unmount(mp);
970 xfs_errortag_del(mp);
971 out_remove_error_sysfs:
972 xfs_error_sysfs_del(mp);
974 xfs_sysfs_del(&mp->m_stats.xs_kobj);
976 xfs_sysfs_del(&mp->m_kobj);
982 * This flushes out the inodes,dquots and the superblock, unmounts the
983 * log and makes sure that incore structures are freed.
987 struct xfs_mount *mp)
992 xfs_stop_block_reaping(mp);
993 xfs_fs_unreserve_ag_blocks(mp);
994 xfs_qm_unmount_quotas(mp);
995 xfs_rtunmount_inodes(mp);
996 xfs_irele(mp->m_rootip);
999 * We can potentially deadlock here if we have an inode cluster
1000 * that has been freed has its buffer still pinned in memory because
1001 * the transaction is still sitting in a iclog. The stale inodes
1002 * on that buffer will have their flush locks held until the
1003 * transaction hits the disk and the callbacks run. the inode
1004 * flush takes the flush lock unconditionally and with nothing to
1005 * push out the iclog we will never get that unlocked. hence we
1006 * need to force the log first.
1008 xfs_log_force(mp, XFS_LOG_SYNC);
1011 * Wait for all busy extents to be freed, including completion of
1012 * any discard operation.
1014 xfs_extent_busy_wait_all(mp);
1015 flush_workqueue(xfs_discard_wq);
1018 * We now need to tell the world we are unmounting. This will allow
1019 * us to detect that the filesystem is going away and we should error
1020 * out anything that we have been retrying in the background. This will
1021 * prevent neverending retries in AIL pushing from hanging the unmount.
1023 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1026 * Flush all pending changes from the AIL.
1028 xfs_ail_push_all_sync(mp->m_ail);
1031 * And reclaim all inodes. At this point there should be no dirty
1032 * inodes and none should be pinned or locked, but use synchronous
1033 * reclaim just to be sure. We can stop background inode reclaim
1034 * here as well if it is still running.
1036 cancel_delayed_work_sync(&mp->m_reclaim_work);
1037 xfs_reclaim_inodes(mp, SYNC_WAIT);
1038 xfs_health_unmount(mp);
1043 * Unreserve any blocks we have so that when we unmount we don't account
1044 * the reserved free space as used. This is really only necessary for
1045 * lazy superblock counting because it trusts the incore superblock
1046 * counters to be absolutely correct on clean unmount.
1048 * We don't bother correcting this elsewhere for lazy superblock
1049 * counting because on mount of an unclean filesystem we reconstruct the
1050 * correct counter value and this is irrelevant.
1052 * For non-lazy counter filesystems, this doesn't matter at all because
1053 * we only every apply deltas to the superblock and hence the incore
1054 * value does not matter....
1057 error = xfs_reserve_blocks(mp, &resblks, NULL);
1059 xfs_warn(mp, "Unable to free reserved block pool. "
1060 "Freespace may not be correct on next mount.");
1062 error = xfs_log_sbcount(mp);
1064 xfs_warn(mp, "Unable to update superblock counters. "
1065 "Freespace may not be correct on next mount.");
1068 xfs_log_unmount(mp);
1070 xfs_uuid_unmount(mp);
1073 xfs_errortag_clearall(mp);
1077 xfs_errortag_del(mp);
1078 xfs_error_sysfs_del(mp);
1079 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1080 xfs_sysfs_del(&mp->m_kobj);
1084 * Determine whether modifications can proceed. The caller specifies the minimum
1085 * freeze level for which modifications should not be allowed. This allows
1086 * certain operations to proceed while the freeze sequence is in progress, if
1091 struct xfs_mount *mp,
1094 ASSERT(level > SB_UNFROZEN);
1095 if ((mp->m_super->s_writers.frozen >= level) ||
1096 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1105 * Sync the superblock counters to disk.
1107 * Note this code can be called during the process of freezing, so we use the
1108 * transaction allocator that does not block when the transaction subsystem is
1109 * in its frozen state.
1112 xfs_log_sbcount(xfs_mount_t *mp)
1114 /* allow this to proceed during the freeze sequence... */
1115 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1119 * we don't need to do this if we are updating the superblock
1120 * counters on every modification.
1122 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1125 return xfs_sync_sb(mp, true);
1129 * Deltas for the inode count are +/-64, hence we use a large batch size
1130 * of 128 so we don't need to take the counter lock on every update.
1132 #define XFS_ICOUNT_BATCH 128
1135 struct xfs_mount *mp,
1138 percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1139 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1141 percpu_counter_add(&mp->m_icount, -delta);
1149 struct xfs_mount *mp,
1152 percpu_counter_add(&mp->m_ifree, delta);
1153 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1155 percpu_counter_add(&mp->m_ifree, -delta);
1162 * Deltas for the block count can vary from 1 to very large, but lock contention
1163 * only occurs on frequent small block count updates such as in the delayed
1164 * allocation path for buffered writes (page a time updates). Hence we set
1165 * a large batch count (1024) to minimise global counter updates except when
1166 * we get near to ENOSPC and we have to be very accurate with our updates.
1168 #define XFS_FDBLOCKS_BATCH 1024
1171 struct xfs_mount *mp,
1181 * If the reserve pool is depleted, put blocks back into it
1182 * first. Most of the time the pool is full.
1184 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1185 percpu_counter_add(&mp->m_fdblocks, delta);
1189 spin_lock(&mp->m_sb_lock);
1190 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1192 if (res_used > delta) {
1193 mp->m_resblks_avail += delta;
1196 mp->m_resblks_avail = mp->m_resblks;
1197 percpu_counter_add(&mp->m_fdblocks, delta);
1199 spin_unlock(&mp->m_sb_lock);
1204 * Taking blocks away, need to be more accurate the closer we
1207 * If the counter has a value of less than 2 * max batch size,
1208 * then make everything serialise as we are real close to
1211 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1212 XFS_FDBLOCKS_BATCH) < 0)
1215 batch = XFS_FDBLOCKS_BATCH;
1217 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1218 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1219 XFS_FDBLOCKS_BATCH) >= 0) {
1225 * lock up the sb for dipping into reserves before releasing the space
1226 * that took us to ENOSPC.
1228 spin_lock(&mp->m_sb_lock);
1229 percpu_counter_add(&mp->m_fdblocks, -delta);
1231 goto fdblocks_enospc;
1233 lcounter = (long long)mp->m_resblks_avail + delta;
1234 if (lcounter >= 0) {
1235 mp->m_resblks_avail = lcounter;
1236 spin_unlock(&mp->m_sb_lock);
1239 printk_once(KERN_WARNING
1240 "Filesystem \"%s\": reserve blocks depleted! "
1241 "Consider increasing reserve pool size.",
1244 spin_unlock(&mp->m_sb_lock);
1250 struct xfs_mount *mp,
1256 spin_lock(&mp->m_sb_lock);
1257 lcounter = mp->m_sb.sb_frextents + delta;
1261 mp->m_sb.sb_frextents = lcounter;
1262 spin_unlock(&mp->m_sb_lock);
1267 * xfs_getsb() is called to obtain the buffer for the superblock.
1268 * The buffer is returned locked and read in from disk.
1269 * The buffer should be released with a call to xfs_brelse().
1273 struct xfs_mount *mp)
1275 struct xfs_buf *bp = mp->m_sb_bp;
1279 ASSERT(bp->b_flags & XBF_DONE);
1284 * Used to free the superblock along various error paths.
1288 struct xfs_mount *mp)
1290 struct xfs_buf *bp = mp->m_sb_bp;
1298 * If the underlying (data/log/rt) device is readonly, there are some
1299 * operations that cannot proceed.
1302 xfs_dev_is_read_only(
1303 struct xfs_mount *mp,
1306 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1307 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1308 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1309 xfs_notice(mp, "%s required on read-only device.", message);
1310 xfs_notice(mp, "write access unavailable, cannot proceed.");
1316 /* Force the summary counters to be recalculated at next mount. */
1318 xfs_force_summary_recalc(
1319 struct xfs_mount *mp)
1321 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1324 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1328 * Update the in-core delayed block counter.
1330 * We prefer to update the counter without having to take a spinlock for every
1331 * counter update (i.e. batching). Each change to delayed allocation
1332 * reservations can change can easily exceed the default percpu counter
1333 * batching, so we use a larger batch factor here.
1335 * Note that we don't currently have any callers requiring fast summation
1336 * (e.g. percpu_counter_read) so we can use a big batch value here.
1338 #define XFS_DELALLOC_BATCH (4096)
1341 struct xfs_mount *mp,
1344 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1345 XFS_DELALLOC_BATCH);