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_defer.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
20 #include "xfs_ialloc.h"
21 #include "xfs_alloc.h"
22 #include "xfs_rtalloc.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
27 #include "xfs_error.h"
28 #include "xfs_quota.h"
29 #include "xfs_fsops.h"
30 #include "xfs_trace.h"
31 #include "xfs_icache.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_rmap_btree.h"
34 #include "xfs_refcount_btree.h"
35 #include "xfs_reflink.h"
36 #include "xfs_extent_busy.h"
37 #include "xfs_health.h"
40 static DEFINE_MUTEX(xfs_uuid_table_mutex);
41 static int xfs_uuid_table_size;
42 static uuid_t *xfs_uuid_table;
45 xfs_uuid_table_free(void)
47 if (xfs_uuid_table_size == 0)
49 kmem_free(xfs_uuid_table);
50 xfs_uuid_table = NULL;
51 xfs_uuid_table_size = 0;
55 * See if the UUID is unique among mounted XFS filesystems.
56 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
62 uuid_t *uuid = &mp->m_sb.sb_uuid;
65 /* Publish UUID in struct super_block */
66 uuid_copy(&mp->m_super->s_uuid, uuid);
68 if (mp->m_flags & XFS_MOUNT_NOUUID)
71 if (uuid_is_null(uuid)) {
72 xfs_warn(mp, "Filesystem has null UUID - can't mount");
76 mutex_lock(&xfs_uuid_table_mutex);
77 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
78 if (uuid_is_null(&xfs_uuid_table[i])) {
82 if (uuid_equal(uuid, &xfs_uuid_table[i]))
87 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
88 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
90 hole = xfs_uuid_table_size++;
92 xfs_uuid_table[hole] = *uuid;
93 mutex_unlock(&xfs_uuid_table_mutex);
98 mutex_unlock(&xfs_uuid_table_mutex);
99 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
105 struct xfs_mount *mp)
107 uuid_t *uuid = &mp->m_sb.sb_uuid;
110 if (mp->m_flags & XFS_MOUNT_NOUUID)
113 mutex_lock(&xfs_uuid_table_mutex);
114 for (i = 0; i < xfs_uuid_table_size; i++) {
115 if (uuid_is_null(&xfs_uuid_table[i]))
117 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
119 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
122 ASSERT(i < xfs_uuid_table_size);
123 mutex_unlock(&xfs_uuid_table_mutex);
129 struct rcu_head *head)
131 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
133 ASSERT(atomic_read(&pag->pag_ref) == 0);
138 * Free up the per-ag resources associated with the mount structure.
145 struct xfs_perag *pag;
147 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
148 spin_lock(&mp->m_perag_lock);
149 pag = radix_tree_delete(&mp->m_perag_tree, agno);
150 spin_unlock(&mp->m_perag_lock);
152 ASSERT(atomic_read(&pag->pag_ref) == 0);
153 xfs_iunlink_destroy(pag);
154 xfs_buf_hash_destroy(pag);
155 mutex_destroy(&pag->pag_ici_reclaim_lock);
156 call_rcu(&pag->rcu_head, __xfs_free_perag);
161 * Check size of device based on the (data/realtime) block count.
162 * Note: this check is used by the growfs code as well as mount.
165 xfs_sb_validate_fsb_count(
169 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
170 ASSERT(sbp->sb_blocklog >= BBSHIFT);
172 /* Limited by ULONG_MAX of page cache index */
173 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
179 xfs_initialize_perag(
181 xfs_agnumber_t agcount,
182 xfs_agnumber_t *maxagi)
184 xfs_agnumber_t index;
185 xfs_agnumber_t first_initialised = NULLAGNUMBER;
190 * Walk the current per-ag tree so we don't try to initialise AGs
191 * that already exist (growfs case). Allocate and insert all the
192 * AGs we don't find ready for initialisation.
194 for (index = 0; index < agcount; index++) {
195 pag = xfs_perag_get(mp, index);
201 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
203 goto out_unwind_new_pags;
204 pag->pag_agno = index;
206 spin_lock_init(&pag->pag_ici_lock);
207 mutex_init(&pag->pag_ici_reclaim_lock);
208 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
209 if (xfs_buf_hash_init(pag))
211 init_waitqueue_head(&pag->pagb_wait);
212 spin_lock_init(&pag->pagb_lock);
214 pag->pagb_tree = RB_ROOT;
216 if (radix_tree_preload(GFP_NOFS))
217 goto out_hash_destroy;
219 spin_lock(&mp->m_perag_lock);
220 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
222 spin_unlock(&mp->m_perag_lock);
223 radix_tree_preload_end();
225 goto out_hash_destroy;
227 spin_unlock(&mp->m_perag_lock);
228 radix_tree_preload_end();
229 /* first new pag is fully initialized */
230 if (first_initialised == NULLAGNUMBER)
231 first_initialised = index;
232 error = xfs_iunlink_init(pag);
234 goto out_hash_destroy;
235 spin_lock_init(&pag->pag_state_lock);
238 index = xfs_set_inode_alloc(mp, agcount);
243 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
247 xfs_buf_hash_destroy(pag);
249 mutex_destroy(&pag->pag_ici_reclaim_lock);
252 /* unwind any prior newly initialized pags */
253 for (index = first_initialised; index < agcount; index++) {
254 pag = radix_tree_delete(&mp->m_perag_tree, index);
257 xfs_buf_hash_destroy(pag);
258 xfs_iunlink_destroy(pag);
259 mutex_destroy(&pag->pag_ici_reclaim_lock);
268 * Does the initial read of the superblock.
272 struct xfs_mount *mp,
275 unsigned int sector_size;
277 struct xfs_sb *sbp = &mp->m_sb;
279 int loud = !(flags & XFS_MFSI_QUIET);
280 const struct xfs_buf_ops *buf_ops;
282 ASSERT(mp->m_sb_bp == NULL);
283 ASSERT(mp->m_ddev_targp != NULL);
286 * For the initial read, we must guess at the sector
287 * size based on the block device. It's enough to
288 * get the sb_sectsize out of the superblock and
289 * then reread with the proper length.
290 * We don't verify it yet, because it may not be complete.
292 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
296 * Allocate a (locked) buffer to hold the superblock. This will be kept
297 * around at all times to optimize access to the superblock. Therefore,
298 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
302 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
303 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
307 xfs_warn(mp, "SB validate failed with error %d.", error);
308 /* bad CRC means corrupted metadata */
309 if (error == -EFSBADCRC)
310 error = -EFSCORRUPTED;
315 * Initialize the mount structure from the superblock.
317 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
320 * If we haven't validated the superblock, do so now before we try
321 * to check the sector size and reread the superblock appropriately.
323 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
325 xfs_warn(mp, "Invalid superblock magic number");
331 * We must be able to do sector-sized and sector-aligned IO.
333 if (sector_size > sbp->sb_sectsize) {
335 xfs_warn(mp, "device supports %u byte sectors (not %u)",
336 sector_size, sbp->sb_sectsize);
341 if (buf_ops == NULL) {
343 * Re-read the superblock so the buffer is correctly sized,
344 * and properly verified.
347 sector_size = sbp->sb_sectsize;
348 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
352 xfs_reinit_percpu_counters(mp);
354 /* no need to be quiet anymore, so reset the buf ops */
355 bp->b_ops = &xfs_sb_buf_ops;
367 * Update alignment values based on mount options and sb values
370 xfs_update_alignment(xfs_mount_t *mp)
372 xfs_sb_t *sbp = &(mp->m_sb);
376 * If stripe unit and stripe width are not multiples
377 * of the fs blocksize turn off alignment.
379 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
380 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
382 "alignment check failed: sunit/swidth vs. blocksize(%d)",
387 * Convert the stripe unit and width to FSBs.
389 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
390 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
392 "alignment check failed: sunit/swidth vs. agsize(%d)",
395 } else if (mp->m_dalign) {
396 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
399 "alignment check failed: sunit(%d) less than bsize(%d)",
400 mp->m_dalign, sbp->sb_blocksize);
406 * Update superblock with new values
409 if (xfs_sb_version_hasdalign(sbp)) {
410 if (sbp->sb_unit != mp->m_dalign) {
411 sbp->sb_unit = mp->m_dalign;
412 mp->m_update_sb = true;
414 if (sbp->sb_width != mp->m_swidth) {
415 sbp->sb_width = mp->m_swidth;
416 mp->m_update_sb = true;
420 "cannot change alignment: superblock does not support data alignment");
423 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
424 xfs_sb_version_hasdalign(&mp->m_sb)) {
425 mp->m_dalign = sbp->sb_unit;
426 mp->m_swidth = sbp->sb_width;
433 * Set the maximum inode count for this filesystem
436 xfs_set_maxicount(xfs_mount_t *mp)
438 xfs_sb_t *sbp = &(mp->m_sb);
441 if (sbp->sb_imax_pct) {
443 * Make sure the maximum inode count is a multiple
444 * of the units we allocate inodes in.
446 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
448 do_div(icount, mp->m_ialloc_blks);
449 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
457 * Set the default minimum read and write sizes unless
458 * already specified in a mount option.
459 * We use smaller I/O sizes when the file system
460 * is being used for NFS service (wsync mount option).
463 xfs_set_rw_sizes(xfs_mount_t *mp)
465 xfs_sb_t *sbp = &(mp->m_sb);
466 int readio_log, writeio_log;
468 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
469 if (mp->m_flags & XFS_MOUNT_WSYNC) {
470 readio_log = XFS_WSYNC_READIO_LOG;
471 writeio_log = XFS_WSYNC_WRITEIO_LOG;
473 readio_log = XFS_READIO_LOG_LARGE;
474 writeio_log = XFS_WRITEIO_LOG_LARGE;
477 readio_log = mp->m_readio_log;
478 writeio_log = mp->m_writeio_log;
481 if (sbp->sb_blocklog > readio_log) {
482 mp->m_readio_log = sbp->sb_blocklog;
484 mp->m_readio_log = readio_log;
486 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
487 if (sbp->sb_blocklog > writeio_log) {
488 mp->m_writeio_log = sbp->sb_blocklog;
490 mp->m_writeio_log = writeio_log;
492 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
496 * precalculate the low space thresholds for dynamic speculative preallocation.
499 xfs_set_low_space_thresholds(
500 struct xfs_mount *mp)
504 for (i = 0; i < XFS_LOWSP_MAX; i++) {
505 uint64_t space = mp->m_sb.sb_dblocks;
508 mp->m_low_space[i] = space * (i + 1);
514 * Set whether we're using inode alignment.
517 xfs_set_inoalignment(xfs_mount_t *mp)
519 if (xfs_sb_version_hasalign(&mp->m_sb) &&
520 mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
521 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
523 mp->m_inoalign_mask = 0;
525 * If we are using stripe alignment, check whether
526 * the stripe unit is a multiple of the inode alignment
528 if (mp->m_dalign && mp->m_inoalign_mask &&
529 !(mp->m_dalign & mp->m_inoalign_mask))
530 mp->m_sinoalign = mp->m_dalign;
536 * Check that the data (and log if separate) is an ok size.
540 struct xfs_mount *mp)
546 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
547 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
548 xfs_warn(mp, "filesystem size mismatch detected");
551 error = xfs_buf_read_uncached(mp->m_ddev_targp,
552 d - XFS_FSS_TO_BB(mp, 1),
553 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
555 xfs_warn(mp, "last sector read failed");
560 if (mp->m_logdev_targp == mp->m_ddev_targp)
563 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
564 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
565 xfs_warn(mp, "log size mismatch detected");
568 error = xfs_buf_read_uncached(mp->m_logdev_targp,
569 d - XFS_FSB_TO_BB(mp, 1),
570 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
572 xfs_warn(mp, "log device read failed");
580 * Clear the quotaflags in memory and in the superblock.
583 xfs_mount_reset_sbqflags(
584 struct xfs_mount *mp)
588 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
589 if (mp->m_sb.sb_qflags == 0)
591 spin_lock(&mp->m_sb_lock);
592 mp->m_sb.sb_qflags = 0;
593 spin_unlock(&mp->m_sb_lock);
595 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
598 return xfs_sync_sb(mp, false);
602 xfs_default_resblks(xfs_mount_t *mp)
607 * We default to 5% or 8192 fsbs of space reserved, whichever is
608 * smaller. This is intended to cover concurrent allocation
609 * transactions when we initially hit enospc. These each require a 4
610 * block reservation. Hence by default we cover roughly 2000 concurrent
611 * allocation reservations.
613 resblks = mp->m_sb.sb_dblocks;
615 resblks = min_t(uint64_t, resblks, 8192);
619 /* Ensure the summary counts are correct. */
621 xfs_check_summary_counts(
622 struct xfs_mount *mp)
625 * The AG0 superblock verifier rejects in-progress filesystems,
626 * so we should never see the flag set this far into mounting.
628 if (mp->m_sb.sb_inprogress) {
629 xfs_err(mp, "sb_inprogress set after log recovery??");
631 return -EFSCORRUPTED;
635 * Now the log is mounted, we know if it was an unclean shutdown or
636 * not. If it was, with the first phase of recovery has completed, we
637 * have consistent AG blocks on disk. We have not recovered EFIs yet,
638 * but they are recovered transactionally in the second recovery phase
641 * If the log was clean when we mounted, we can check the summary
642 * counters. If any of them are obviously incorrect, we can recompute
643 * them from the AGF headers in the next step.
645 if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
646 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
647 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
648 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
649 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
652 * We can safely re-initialise incore superblock counters from the
653 * per-ag data. These may not be correct if the filesystem was not
654 * cleanly unmounted, so we waited for recovery to finish before doing
657 * If the filesystem was cleanly unmounted or the previous check did
658 * not flag anything weird, then we can trust the values in the
659 * superblock to be correct and we don't need to do anything here.
660 * Otherwise, recalculate the summary counters.
662 if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
663 XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
664 !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
667 return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
671 * This function does the following on an initial mount of a file system:
672 * - reads the superblock from disk and init the mount struct
673 * - if we're a 32-bit kernel, do a size check on the superblock
674 * so we don't mount terabyte filesystems
675 * - init mount struct realtime fields
676 * - allocate inode hash table for fs
677 * - init directory manager
678 * - perform recovery and init the log manager
682 struct xfs_mount *mp)
684 struct xfs_sb *sbp = &(mp->m_sb);
685 struct xfs_inode *rip;
691 xfs_sb_mount_common(mp, sbp);
694 * Check for a mismatched features2 values. Older kernels read & wrote
695 * into the wrong sb offset for sb_features2 on some platforms due to
696 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
697 * which made older superblock reading/writing routines swap it as a
700 * For backwards compatibility, we make both slots equal.
702 * If we detect a mismatched field, we OR the set bits into the existing
703 * features2 field in case it has already been modified; we don't want
704 * to lose any features. We then update the bad location with the ORed
705 * value so that older kernels will see any features2 flags. The
706 * superblock writeback code ensures the new sb_features2 is copied to
707 * sb_bad_features2 before it is logged or written to disk.
709 if (xfs_sb_has_mismatched_features2(sbp)) {
710 xfs_warn(mp, "correcting sb_features alignment problem");
711 sbp->sb_features2 |= sbp->sb_bad_features2;
712 mp->m_update_sb = true;
715 * Re-check for ATTR2 in case it was found in bad_features2
718 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
719 !(mp->m_flags & XFS_MOUNT_NOATTR2))
720 mp->m_flags |= XFS_MOUNT_ATTR2;
723 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
724 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
725 xfs_sb_version_removeattr2(&mp->m_sb);
726 mp->m_update_sb = true;
728 /* update sb_versionnum for the clearing of the morebits */
729 if (!sbp->sb_features2)
730 mp->m_update_sb = true;
733 /* always use v2 inodes by default now */
734 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
735 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
736 mp->m_update_sb = true;
740 * Check if sb_agblocks is aligned at stripe boundary
741 * If sb_agblocks is NOT aligned turn off m_dalign since
742 * allocator alignment is within an ag, therefore ag has
743 * to be aligned at stripe boundary.
745 error = xfs_update_alignment(mp);
749 xfs_alloc_compute_maxlevels(mp);
750 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
751 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
752 xfs_ialloc_compute_maxlevels(mp);
753 xfs_rmapbt_compute_maxlevels(mp);
754 xfs_refcountbt_compute_maxlevels(mp);
756 xfs_set_maxicount(mp);
758 /* enable fail_at_unmount as default */
759 mp->m_fail_unmount = true;
761 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
765 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
766 &mp->m_kobj, "stats");
768 goto out_remove_sysfs;
770 error = xfs_error_sysfs_init(mp);
774 error = xfs_errortag_init(mp);
776 goto out_remove_error_sysfs;
778 error = xfs_uuid_mount(mp);
780 goto out_remove_errortag;
783 * Set the minimum read and write sizes
785 xfs_set_rw_sizes(mp);
787 /* set the low space thresholds for dynamic preallocation */
788 xfs_set_low_space_thresholds(mp);
791 * Set the inode cluster size.
792 * This may still be overridden by the file system
793 * block size if it is larger than the chosen cluster size.
795 * For v5 filesystems, scale the cluster size with the inode size to
796 * keep a constant ratio of inode per cluster buffer, but only if mkfs
797 * has set the inode alignment value appropriately for larger cluster
800 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
801 if (xfs_sb_version_hascrc(&mp->m_sb)) {
802 int new_size = mp->m_inode_cluster_size;
804 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
805 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
806 mp->m_inode_cluster_size = new_size;
808 mp->m_blocks_per_cluster = xfs_icluster_size_fsb(mp);
809 mp->m_inodes_per_cluster = XFS_FSB_TO_INO(mp, mp->m_blocks_per_cluster);
810 mp->m_cluster_align = xfs_ialloc_cluster_alignment(mp);
811 mp->m_cluster_align_inodes = XFS_FSB_TO_INO(mp, mp->m_cluster_align);
814 * If enabled, sparse inode chunk alignment is expected to match the
815 * cluster size. Full inode chunk alignment must match the chunk size,
816 * but that is checked on sb read verification...
818 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
819 mp->m_sb.sb_spino_align !=
820 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
822 "Sparse inode block alignment (%u) must match cluster size (%llu).",
823 mp->m_sb.sb_spino_align,
824 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
826 goto out_remove_uuid;
830 * Set inode alignment fields
832 xfs_set_inoalignment(mp);
835 * Check that the data (and log if separate) is an ok size.
837 error = xfs_check_sizes(mp);
839 goto out_remove_uuid;
842 * Initialize realtime fields in the mount structure
844 error = xfs_rtmount_init(mp);
846 xfs_warn(mp, "RT mount failed");
847 goto out_remove_uuid;
851 * Copies the low order bits of the timestamp and the randomly
852 * set "sequence" number out of a UUID.
855 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
856 get_unaligned_be16(&sbp->sb_uuid.b[4]);
857 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
859 error = xfs_da_mount(mp);
861 xfs_warn(mp, "Failed dir/attr init: %d", error);
862 goto out_remove_uuid;
866 * Initialize the precomputed transaction reservations values.
871 * Allocate and initialize the per-ag data.
873 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
875 xfs_warn(mp, "Failed per-ag init: %d", error);
879 if (!sbp->sb_logblocks) {
880 xfs_warn(mp, "no log defined");
881 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
882 error = -EFSCORRUPTED;
887 * Log's mount-time initialization. The first part of recovery can place
888 * some items on the AIL, to be handled when recovery is finished or
891 error = xfs_log_mount(mp, mp->m_logdev_targp,
892 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
893 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
895 xfs_warn(mp, "log mount failed");
899 /* Make sure the summary counts are ok. */
900 error = xfs_check_summary_counts(mp);
902 goto out_log_dealloc;
905 * Get and sanity-check the root inode.
906 * Save the pointer to it in the mount structure.
908 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
909 XFS_ILOCK_EXCL, &rip);
912 "Failed to read root inode 0x%llx, error %d",
913 sbp->sb_rootino, -error);
914 goto out_log_dealloc;
919 if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
920 xfs_warn(mp, "corrupted root inode %llu: not a directory",
921 (unsigned long long)rip->i_ino);
922 xfs_iunlock(rip, XFS_ILOCK_EXCL);
923 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
925 error = -EFSCORRUPTED;
928 mp->m_rootip = rip; /* save it */
930 xfs_iunlock(rip, XFS_ILOCK_EXCL);
933 * Initialize realtime inode pointers in the mount structure
935 error = xfs_rtmount_inodes(mp);
938 * Free up the root inode.
940 xfs_warn(mp, "failed to read RT inodes");
945 * If this is a read-only mount defer the superblock updates until
946 * the next remount into writeable mode. Otherwise we would never
947 * perform the update e.g. for the root filesystem.
949 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
950 error = xfs_sync_sb(mp, false);
952 xfs_warn(mp, "failed to write sb changes");
958 * Initialise the XFS quota management subsystem for this mount
960 if (XFS_IS_QUOTA_RUNNING(mp)) {
961 error = xfs_qm_newmount(mp, "amount, "aflags);
965 ASSERT(!XFS_IS_QUOTA_ON(mp));
968 * If a file system had quotas running earlier, but decided to
969 * mount without -o uquota/pquota/gquota options, revoke the
970 * quotachecked license.
972 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
973 xfs_notice(mp, "resetting quota flags");
974 error = xfs_mount_reset_sbqflags(mp);
981 * Finish recovering the file system. This part needed to be delayed
982 * until after the root and real-time bitmap inodes were consistently
985 error = xfs_log_mount_finish(mp);
987 xfs_warn(mp, "log mount finish failed");
992 * Now the log is fully replayed, we can transition to full read-only
993 * mode for read-only mounts. This will sync all the metadata and clean
994 * the log so that the recovery we just performed does not have to be
995 * replayed again on the next mount.
997 * We use the same quiesce mechanism as the rw->ro remount, as they are
998 * semantically identical operations.
1000 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
1002 xfs_quiesce_attr(mp);
1006 * Complete the quota initialisation, post-log-replay component.
1009 ASSERT(mp->m_qflags == 0);
1010 mp->m_qflags = quotaflags;
1012 xfs_qm_mount_quotas(mp);
1016 * Now we are mounted, reserve a small amount of unused space for
1017 * privileged transactions. This is needed so that transaction
1018 * space required for critical operations can dip into this pool
1019 * when at ENOSPC. This is needed for operations like create with
1020 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1021 * are not allowed to use this reserved space.
1023 * This may drive us straight to ENOSPC on mount, but that implies
1024 * we were already there on the last unmount. Warn if this occurs.
1026 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1027 resblks = xfs_default_resblks(mp);
1028 error = xfs_reserve_blocks(mp, &resblks, NULL);
1031 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1033 /* Recover any CoW blocks that never got remapped. */
1034 error = xfs_reflink_recover_cow(mp);
1037 "Error %d recovering leftover CoW allocations.", error);
1038 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1042 /* Reserve AG blocks for future btree expansion. */
1043 error = xfs_fs_reserve_ag_blocks(mp);
1044 if (error && error != -ENOSPC)
1051 xfs_fs_unreserve_ag_blocks(mp);
1053 xfs_qm_unmount_quotas(mp);
1055 xfs_rtunmount_inodes(mp);
1058 /* Clean out dquots that might be in memory after quotacheck. */
1061 * Cancel all delayed reclaim work and reclaim the inodes directly.
1062 * We have to do this /after/ rtunmount and qm_unmount because those
1063 * two will have scheduled delayed reclaim for the rt/quota inodes.
1065 * This is slightly different from the unmountfs call sequence
1066 * because we could be tearing down a partially set up mount. In
1067 * particular, if log_mount_finish fails we bail out without calling
1068 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1071 cancel_delayed_work_sync(&mp->m_reclaim_work);
1072 xfs_reclaim_inodes(mp, SYNC_WAIT);
1073 xfs_health_unmount(mp);
1075 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1076 xfs_log_mount_cancel(mp);
1078 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1079 xfs_wait_buftarg(mp->m_logdev_targp);
1080 xfs_wait_buftarg(mp->m_ddev_targp);
1086 xfs_uuid_unmount(mp);
1087 out_remove_errortag:
1088 xfs_errortag_del(mp);
1089 out_remove_error_sysfs:
1090 xfs_error_sysfs_del(mp);
1092 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1094 xfs_sysfs_del(&mp->m_kobj);
1100 * This flushes out the inodes,dquots and the superblock, unmounts the
1101 * log and makes sure that incore structures are freed.
1105 struct xfs_mount *mp)
1110 xfs_stop_block_reaping(mp);
1111 xfs_fs_unreserve_ag_blocks(mp);
1112 xfs_qm_unmount_quotas(mp);
1113 xfs_rtunmount_inodes(mp);
1114 xfs_irele(mp->m_rootip);
1117 * We can potentially deadlock here if we have an inode cluster
1118 * that has been freed has its buffer still pinned in memory because
1119 * the transaction is still sitting in a iclog. The stale inodes
1120 * on that buffer will have their flush locks held until the
1121 * transaction hits the disk and the callbacks run. the inode
1122 * flush takes the flush lock unconditionally and with nothing to
1123 * push out the iclog we will never get that unlocked. hence we
1124 * need to force the log first.
1126 xfs_log_force(mp, XFS_LOG_SYNC);
1129 * Wait for all busy extents to be freed, including completion of
1130 * any discard operation.
1132 xfs_extent_busy_wait_all(mp);
1133 flush_workqueue(xfs_discard_wq);
1136 * We now need to tell the world we are unmounting. This will allow
1137 * us to detect that the filesystem is going away and we should error
1138 * out anything that we have been retrying in the background. This will
1139 * prevent neverending retries in AIL pushing from hanging the unmount.
1141 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1144 * Flush all pending changes from the AIL.
1146 xfs_ail_push_all_sync(mp->m_ail);
1149 * And reclaim all inodes. At this point there should be no dirty
1150 * inodes and none should be pinned or locked, but use synchronous
1151 * reclaim just to be sure. We can stop background inode reclaim
1152 * here as well if it is still running.
1154 cancel_delayed_work_sync(&mp->m_reclaim_work);
1155 xfs_reclaim_inodes(mp, SYNC_WAIT);
1156 xfs_health_unmount(mp);
1161 * Unreserve any blocks we have so that when we unmount we don't account
1162 * the reserved free space as used. This is really only necessary for
1163 * lazy superblock counting because it trusts the incore superblock
1164 * counters to be absolutely correct on clean unmount.
1166 * We don't bother correcting this elsewhere for lazy superblock
1167 * counting because on mount of an unclean filesystem we reconstruct the
1168 * correct counter value and this is irrelevant.
1170 * For non-lazy counter filesystems, this doesn't matter at all because
1171 * we only every apply deltas to the superblock and hence the incore
1172 * value does not matter....
1175 error = xfs_reserve_blocks(mp, &resblks, NULL);
1177 xfs_warn(mp, "Unable to free reserved block pool. "
1178 "Freespace may not be correct on next mount.");
1180 error = xfs_log_sbcount(mp);
1182 xfs_warn(mp, "Unable to update superblock counters. "
1183 "Freespace may not be correct on next mount.");
1186 xfs_log_unmount(mp);
1188 xfs_uuid_unmount(mp);
1191 xfs_errortag_clearall(mp);
1195 xfs_errortag_del(mp);
1196 xfs_error_sysfs_del(mp);
1197 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1198 xfs_sysfs_del(&mp->m_kobj);
1202 * Determine whether modifications can proceed. The caller specifies the minimum
1203 * freeze level for which modifications should not be allowed. This allows
1204 * certain operations to proceed while the freeze sequence is in progress, if
1209 struct xfs_mount *mp,
1212 ASSERT(level > SB_UNFROZEN);
1213 if ((mp->m_super->s_writers.frozen >= level) ||
1214 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1223 * Sync the superblock counters to disk.
1225 * Note this code can be called during the process of freezing, so we use the
1226 * transaction allocator that does not block when the transaction subsystem is
1227 * in its frozen state.
1230 xfs_log_sbcount(xfs_mount_t *mp)
1232 /* allow this to proceed during the freeze sequence... */
1233 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1237 * we don't need to do this if we are updating the superblock
1238 * counters on every modification.
1240 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1243 return xfs_sync_sb(mp, true);
1247 * Deltas for the inode count are +/-64, hence we use a large batch size
1248 * of 128 so we don't need to take the counter lock on every update.
1250 #define XFS_ICOUNT_BATCH 128
1253 struct xfs_mount *mp,
1256 percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1257 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1259 percpu_counter_add(&mp->m_icount, -delta);
1267 struct xfs_mount *mp,
1270 percpu_counter_add(&mp->m_ifree, delta);
1271 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1273 percpu_counter_add(&mp->m_ifree, -delta);
1280 * Deltas for the block count can vary from 1 to very large, but lock contention
1281 * only occurs on frequent small block count updates such as in the delayed
1282 * allocation path for buffered writes (page a time updates). Hence we set
1283 * a large batch count (1024) to minimise global counter updates except when
1284 * we get near to ENOSPC and we have to be very accurate with our updates.
1286 #define XFS_FDBLOCKS_BATCH 1024
1289 struct xfs_mount *mp,
1299 * If the reserve pool is depleted, put blocks back into it
1300 * first. Most of the time the pool is full.
1302 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1303 percpu_counter_add(&mp->m_fdblocks, delta);
1307 spin_lock(&mp->m_sb_lock);
1308 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1310 if (res_used > delta) {
1311 mp->m_resblks_avail += delta;
1314 mp->m_resblks_avail = mp->m_resblks;
1315 percpu_counter_add(&mp->m_fdblocks, delta);
1317 spin_unlock(&mp->m_sb_lock);
1322 * Taking blocks away, need to be more accurate the closer we
1325 * If the counter has a value of less than 2 * max batch size,
1326 * then make everything serialise as we are real close to
1329 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1330 XFS_FDBLOCKS_BATCH) < 0)
1333 batch = XFS_FDBLOCKS_BATCH;
1335 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1336 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1337 XFS_FDBLOCKS_BATCH) >= 0) {
1343 * lock up the sb for dipping into reserves before releasing the space
1344 * that took us to ENOSPC.
1346 spin_lock(&mp->m_sb_lock);
1347 percpu_counter_add(&mp->m_fdblocks, -delta);
1349 goto fdblocks_enospc;
1351 lcounter = (long long)mp->m_resblks_avail + delta;
1352 if (lcounter >= 0) {
1353 mp->m_resblks_avail = lcounter;
1354 spin_unlock(&mp->m_sb_lock);
1357 printk_once(KERN_WARNING
1358 "Filesystem \"%s\": reserve blocks depleted! "
1359 "Consider increasing reserve pool size.",
1362 spin_unlock(&mp->m_sb_lock);
1368 struct xfs_mount *mp,
1374 spin_lock(&mp->m_sb_lock);
1375 lcounter = mp->m_sb.sb_frextents + delta;
1379 mp->m_sb.sb_frextents = lcounter;
1380 spin_unlock(&mp->m_sb_lock);
1385 * xfs_getsb() is called to obtain the buffer for the superblock.
1386 * The buffer is returned locked and read in from disk.
1387 * The buffer should be released with a call to xfs_brelse().
1389 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1390 * the superblock buffer if it can be locked without sleeping.
1391 * If it can't then we'll return NULL.
1395 struct xfs_mount *mp,
1398 struct xfs_buf *bp = mp->m_sb_bp;
1400 if (!xfs_buf_trylock(bp)) {
1401 if (flags & XBF_TRYLOCK)
1407 ASSERT(bp->b_flags & XBF_DONE);
1412 * Used to free the superblock along various error paths.
1416 struct xfs_mount *mp)
1418 struct xfs_buf *bp = mp->m_sb_bp;
1426 * If the underlying (data/log/rt) device is readonly, there are some
1427 * operations that cannot proceed.
1430 xfs_dev_is_read_only(
1431 struct xfs_mount *mp,
1434 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1435 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1436 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1437 xfs_notice(mp, "%s required on read-only device.", message);
1438 xfs_notice(mp, "write access unavailable, cannot proceed.");
1444 /* Force the summary counters to be recalculated at next mount. */
1446 xfs_force_summary_recalc(
1447 struct xfs_mount *mp)
1449 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1452 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1456 * Update the in-core delayed block counter.
1458 * We prefer to update the counter without having to take a spinlock for every
1459 * counter update (i.e. batching). Each change to delayed allocation
1460 * reservations can change can easily exceed the default percpu counter
1461 * batching, so we use a larger batch factor here.
1463 * Note that we don't currently have any callers requiring fast summation
1464 * (e.g. percpu_counter_read) so we can use a big batch value here.
1466 #define XFS_DELALLOC_BATCH (4096)
1469 struct xfs_mount *mp,
1472 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1473 XFS_DELALLOC_BATCH);