2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
36 #include "xfs_trans.h"
37 #include "xfs_trans_priv.h"
39 #include "xfs_error.h"
40 #include "xfs_quota.h"
41 #include "xfs_fsops.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_sysfs.h"
45 #include "xfs_rmap_btree.h"
46 #include "xfs_refcount_btree.h"
47 #include "xfs_reflink.h"
48 #include "xfs_extent_busy.h"
51 static DEFINE_MUTEX(xfs_uuid_table_mutex);
52 static int xfs_uuid_table_size;
53 static uuid_t *xfs_uuid_table;
56 xfs_uuid_table_free(void)
58 if (xfs_uuid_table_size == 0)
60 kmem_free(xfs_uuid_table);
61 xfs_uuid_table = NULL;
62 xfs_uuid_table_size = 0;
66 * See if the UUID is unique among mounted XFS filesystems.
67 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
73 uuid_t *uuid = &mp->m_sb.sb_uuid;
76 /* Publish UUID in struct super_block */
77 uuid_copy(&mp->m_super->s_uuid, uuid);
79 if (mp->m_flags & XFS_MOUNT_NOUUID)
82 if (uuid_is_null(uuid)) {
83 xfs_warn(mp, "Filesystem has null UUID - can't mount");
87 mutex_lock(&xfs_uuid_table_mutex);
88 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
89 if (uuid_is_null(&xfs_uuid_table[i])) {
93 if (uuid_equal(uuid, &xfs_uuid_table[i]))
98 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
99 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
101 hole = xfs_uuid_table_size++;
103 xfs_uuid_table[hole] = *uuid;
104 mutex_unlock(&xfs_uuid_table_mutex);
109 mutex_unlock(&xfs_uuid_table_mutex);
110 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
116 struct xfs_mount *mp)
118 uuid_t *uuid = &mp->m_sb.sb_uuid;
121 if (mp->m_flags & XFS_MOUNT_NOUUID)
124 mutex_lock(&xfs_uuid_table_mutex);
125 for (i = 0; i < xfs_uuid_table_size; i++) {
126 if (uuid_is_null(&xfs_uuid_table[i]))
128 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
130 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
133 ASSERT(i < xfs_uuid_table_size);
134 mutex_unlock(&xfs_uuid_table_mutex);
140 struct rcu_head *head)
142 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
144 ASSERT(atomic_read(&pag->pag_ref) == 0);
149 * Free up the per-ag resources associated with the mount structure.
156 struct xfs_perag *pag;
158 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
159 spin_lock(&mp->m_perag_lock);
160 pag = radix_tree_delete(&mp->m_perag_tree, agno);
161 spin_unlock(&mp->m_perag_lock);
163 ASSERT(atomic_read(&pag->pag_ref) == 0);
164 xfs_buf_hash_destroy(pag);
165 mutex_destroy(&pag->pag_ici_reclaim_lock);
166 call_rcu(&pag->rcu_head, __xfs_free_perag);
171 * Check size of device based on the (data/realtime) block count.
172 * Note: this check is used by the growfs code as well as mount.
175 xfs_sb_validate_fsb_count(
179 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
180 ASSERT(sbp->sb_blocklog >= BBSHIFT);
182 /* Limited by ULONG_MAX of page cache index */
183 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
189 xfs_initialize_perag(
191 xfs_agnumber_t agcount,
192 xfs_agnumber_t *maxagi)
194 xfs_agnumber_t index;
195 xfs_agnumber_t first_initialised = NULLAGNUMBER;
200 * Walk the current per-ag tree so we don't try to initialise AGs
201 * that already exist (growfs case). Allocate and insert all the
202 * AGs we don't find ready for initialisation.
204 for (index = 0; index < agcount; index++) {
205 pag = xfs_perag_get(mp, index);
211 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
213 goto out_unwind_new_pags;
214 pag->pag_agno = index;
216 spin_lock_init(&pag->pag_ici_lock);
217 mutex_init(&pag->pag_ici_reclaim_lock);
218 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
219 if (xfs_buf_hash_init(pag))
221 init_waitqueue_head(&pag->pagb_wait);
223 if (radix_tree_preload(GFP_NOFS))
224 goto out_hash_destroy;
226 spin_lock(&mp->m_perag_lock);
227 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
229 spin_unlock(&mp->m_perag_lock);
230 radix_tree_preload_end();
232 goto out_hash_destroy;
234 spin_unlock(&mp->m_perag_lock);
235 radix_tree_preload_end();
236 /* first new pag is fully initialized */
237 if (first_initialised == NULLAGNUMBER)
238 first_initialised = index;
241 index = xfs_set_inode_alloc(mp, agcount);
246 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
250 xfs_buf_hash_destroy(pag);
252 mutex_destroy(&pag->pag_ici_reclaim_lock);
255 /* unwind any prior newly initialized pags */
256 for (index = first_initialised; index < agcount; index++) {
257 pag = radix_tree_delete(&mp->m_perag_tree, index);
260 xfs_buf_hash_destroy(pag);
261 mutex_destroy(&pag->pag_ici_reclaim_lock);
270 * Does the initial read of the superblock.
274 struct xfs_mount *mp,
277 unsigned int sector_size;
279 struct xfs_sb *sbp = &mp->m_sb;
281 int loud = !(flags & XFS_MFSI_QUIET);
282 const struct xfs_buf_ops *buf_ops;
284 ASSERT(mp->m_sb_bp == NULL);
285 ASSERT(mp->m_ddev_targp != NULL);
288 * For the initial read, we must guess at the sector
289 * size based on the block device. It's enough to
290 * get the sb_sectsize out of the superblock and
291 * then reread with the proper length.
292 * We don't verify it yet, because it may not be complete.
294 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
298 * Allocate a (locked) buffer to hold the superblock. This will be kept
299 * around at all times to optimize access to the superblock. Therefore,
300 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
304 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
305 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
309 xfs_warn(mp, "SB validate failed with error %d.", error);
310 /* bad CRC means corrupted metadata */
311 if (error == -EFSBADCRC)
312 error = -EFSCORRUPTED;
317 * Initialize the mount structure from the superblock.
319 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
322 * If we haven't validated the superblock, do so now before we try
323 * to check the sector size and reread the superblock appropriately.
325 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
327 xfs_warn(mp, "Invalid superblock magic number");
333 * We must be able to do sector-sized and sector-aligned IO.
335 if (sector_size > sbp->sb_sectsize) {
337 xfs_warn(mp, "device supports %u byte sectors (not %u)",
338 sector_size, sbp->sb_sectsize);
343 if (buf_ops == NULL) {
345 * Re-read the superblock so the buffer is correctly sized,
346 * and properly verified.
349 sector_size = sbp->sb_sectsize;
350 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
354 xfs_reinit_percpu_counters(mp);
356 /* no need to be quiet anymore, so reset the buf ops */
357 bp->b_ops = &xfs_sb_buf_ops;
369 * Update alignment values based on mount options and sb values
372 xfs_update_alignment(xfs_mount_t *mp)
374 xfs_sb_t *sbp = &(mp->m_sb);
378 * If stripe unit and stripe width are not multiples
379 * of the fs blocksize turn off alignment.
381 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
382 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
384 "alignment check failed: sunit/swidth vs. blocksize(%d)",
389 * Convert the stripe unit and width to FSBs.
391 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
392 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
394 "alignment check failed: sunit/swidth vs. agsize(%d)",
397 } else if (mp->m_dalign) {
398 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
401 "alignment check failed: sunit(%d) less than bsize(%d)",
402 mp->m_dalign, sbp->sb_blocksize);
408 * Update superblock with new values
411 if (xfs_sb_version_hasdalign(sbp)) {
412 if (sbp->sb_unit != mp->m_dalign) {
413 sbp->sb_unit = mp->m_dalign;
414 mp->m_update_sb = true;
416 if (sbp->sb_width != mp->m_swidth) {
417 sbp->sb_width = mp->m_swidth;
418 mp->m_update_sb = true;
422 "cannot change alignment: superblock does not support data alignment");
425 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
426 xfs_sb_version_hasdalign(&mp->m_sb)) {
427 mp->m_dalign = sbp->sb_unit;
428 mp->m_swidth = sbp->sb_width;
435 * Set the maximum inode count for this filesystem
438 xfs_set_maxicount(xfs_mount_t *mp)
440 xfs_sb_t *sbp = &(mp->m_sb);
443 if (sbp->sb_imax_pct) {
445 * Make sure the maximum inode count is a multiple
446 * of the units we allocate inodes in.
448 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
450 do_div(icount, mp->m_ialloc_blks);
451 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
459 * Set the default minimum read and write sizes unless
460 * already specified in a mount option.
461 * We use smaller I/O sizes when the file system
462 * is being used for NFS service (wsync mount option).
465 xfs_set_rw_sizes(xfs_mount_t *mp)
467 xfs_sb_t *sbp = &(mp->m_sb);
468 int readio_log, writeio_log;
470 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
471 if (mp->m_flags & XFS_MOUNT_WSYNC) {
472 readio_log = XFS_WSYNC_READIO_LOG;
473 writeio_log = XFS_WSYNC_WRITEIO_LOG;
475 readio_log = XFS_READIO_LOG_LARGE;
476 writeio_log = XFS_WRITEIO_LOG_LARGE;
479 readio_log = mp->m_readio_log;
480 writeio_log = mp->m_writeio_log;
483 if (sbp->sb_blocklog > readio_log) {
484 mp->m_readio_log = sbp->sb_blocklog;
486 mp->m_readio_log = readio_log;
488 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
489 if (sbp->sb_blocklog > writeio_log) {
490 mp->m_writeio_log = sbp->sb_blocklog;
492 mp->m_writeio_log = writeio_log;
494 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
498 * precalculate the low space thresholds for dynamic speculative preallocation.
501 xfs_set_low_space_thresholds(
502 struct xfs_mount *mp)
506 for (i = 0; i < XFS_LOWSP_MAX; i++) {
507 uint64_t space = mp->m_sb.sb_dblocks;
510 mp->m_low_space[i] = space * (i + 1);
516 * Set whether we're using inode alignment.
519 xfs_set_inoalignment(xfs_mount_t *mp)
521 if (xfs_sb_version_hasalign(&mp->m_sb) &&
522 mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
523 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
525 mp->m_inoalign_mask = 0;
527 * If we are using stripe alignment, check whether
528 * the stripe unit is a multiple of the inode alignment
530 if (mp->m_dalign && mp->m_inoalign_mask &&
531 !(mp->m_dalign & mp->m_inoalign_mask))
532 mp->m_sinoalign = mp->m_dalign;
538 * Check that the data (and log if separate) is an ok size.
542 struct xfs_mount *mp)
548 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
549 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
550 xfs_warn(mp, "filesystem size mismatch detected");
553 error = xfs_buf_read_uncached(mp->m_ddev_targp,
554 d - XFS_FSS_TO_BB(mp, 1),
555 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
557 xfs_warn(mp, "last sector read failed");
562 if (mp->m_logdev_targp == mp->m_ddev_targp)
565 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
566 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
567 xfs_warn(mp, "log size mismatch detected");
570 error = xfs_buf_read_uncached(mp->m_logdev_targp,
571 d - XFS_FSB_TO_BB(mp, 1),
572 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
574 xfs_warn(mp, "log device read failed");
582 * Clear the quotaflags in memory and in the superblock.
585 xfs_mount_reset_sbqflags(
586 struct xfs_mount *mp)
590 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
591 if (mp->m_sb.sb_qflags == 0)
593 spin_lock(&mp->m_sb_lock);
594 mp->m_sb.sb_qflags = 0;
595 spin_unlock(&mp->m_sb_lock);
597 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
600 return xfs_sync_sb(mp, false);
604 xfs_default_resblks(xfs_mount_t *mp)
609 * We default to 5% or 8192 fsbs of space reserved, whichever is
610 * smaller. This is intended to cover concurrent allocation
611 * transactions when we initially hit enospc. These each require a 4
612 * block reservation. Hence by default we cover roughly 2000 concurrent
613 * allocation reservations.
615 resblks = mp->m_sb.sb_dblocks;
617 resblks = min_t(uint64_t, resblks, 8192);
622 * This function does the following on an initial mount of a file system:
623 * - reads the superblock from disk and init the mount struct
624 * - if we're a 32-bit kernel, do a size check on the superblock
625 * so we don't mount terabyte filesystems
626 * - init mount struct realtime fields
627 * - allocate inode hash table for fs
628 * - init directory manager
629 * - perform recovery and init the log manager
633 struct xfs_mount *mp)
635 struct xfs_sb *sbp = &(mp->m_sb);
636 struct xfs_inode *rip;
642 xfs_sb_mount_common(mp, sbp);
645 * Check for a mismatched features2 values. Older kernels read & wrote
646 * into the wrong sb offset for sb_features2 on some platforms due to
647 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
648 * which made older superblock reading/writing routines swap it as a
651 * For backwards compatibility, we make both slots equal.
653 * If we detect a mismatched field, we OR the set bits into the existing
654 * features2 field in case it has already been modified; we don't want
655 * to lose any features. We then update the bad location with the ORed
656 * value so that older kernels will see any features2 flags. The
657 * superblock writeback code ensures the new sb_features2 is copied to
658 * sb_bad_features2 before it is logged or written to disk.
660 if (xfs_sb_has_mismatched_features2(sbp)) {
661 xfs_warn(mp, "correcting sb_features alignment problem");
662 sbp->sb_features2 |= sbp->sb_bad_features2;
663 mp->m_update_sb = true;
666 * Re-check for ATTR2 in case it was found in bad_features2
669 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
670 !(mp->m_flags & XFS_MOUNT_NOATTR2))
671 mp->m_flags |= XFS_MOUNT_ATTR2;
674 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
675 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
676 xfs_sb_version_removeattr2(&mp->m_sb);
677 mp->m_update_sb = true;
679 /* update sb_versionnum for the clearing of the morebits */
680 if (!sbp->sb_features2)
681 mp->m_update_sb = true;
684 /* always use v2 inodes by default now */
685 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
686 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
687 mp->m_update_sb = true;
691 * Check if sb_agblocks is aligned at stripe boundary
692 * If sb_agblocks is NOT aligned turn off m_dalign since
693 * allocator alignment is within an ag, therefore ag has
694 * to be aligned at stripe boundary.
696 error = xfs_update_alignment(mp);
700 xfs_alloc_compute_maxlevels(mp);
701 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
702 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
703 xfs_ialloc_compute_maxlevels(mp);
704 xfs_rmapbt_compute_maxlevels(mp);
705 xfs_refcountbt_compute_maxlevels(mp);
707 xfs_set_maxicount(mp);
709 /* enable fail_at_unmount as default */
710 mp->m_fail_unmount = true;
712 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
716 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
717 &mp->m_kobj, "stats");
719 goto out_remove_sysfs;
721 error = xfs_error_sysfs_init(mp);
725 error = xfs_errortag_init(mp);
727 goto out_remove_error_sysfs;
729 error = xfs_uuid_mount(mp);
731 goto out_remove_errortag;
734 * Set the minimum read and write sizes
736 xfs_set_rw_sizes(mp);
738 /* set the low space thresholds for dynamic preallocation */
739 xfs_set_low_space_thresholds(mp);
742 * Set the inode cluster size.
743 * This may still be overridden by the file system
744 * block size if it is larger than the chosen cluster size.
746 * For v5 filesystems, scale the cluster size with the inode size to
747 * keep a constant ratio of inode per cluster buffer, but only if mkfs
748 * has set the inode alignment value appropriately for larger cluster
751 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
752 if (xfs_sb_version_hascrc(&mp->m_sb)) {
753 int new_size = mp->m_inode_cluster_size;
755 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
756 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
757 mp->m_inode_cluster_size = new_size;
761 * If enabled, sparse inode chunk alignment is expected to match the
762 * cluster size. Full inode chunk alignment must match the chunk size,
763 * but that is checked on sb read verification...
765 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
766 mp->m_sb.sb_spino_align !=
767 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
769 "Sparse inode block alignment (%u) must match cluster size (%llu).",
770 mp->m_sb.sb_spino_align,
771 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
773 goto out_remove_uuid;
777 * Set inode alignment fields
779 xfs_set_inoalignment(mp);
782 * Check that the data (and log if separate) is an ok size.
784 error = xfs_check_sizes(mp);
786 goto out_remove_uuid;
789 * Initialize realtime fields in the mount structure
791 error = xfs_rtmount_init(mp);
793 xfs_warn(mp, "RT mount failed");
794 goto out_remove_uuid;
798 * Copies the low order bits of the timestamp and the randomly
799 * set "sequence" number out of a UUID.
802 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
803 get_unaligned_be16(&sbp->sb_uuid.b[4]);
804 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
806 error = xfs_da_mount(mp);
808 xfs_warn(mp, "Failed dir/attr init: %d", error);
809 goto out_remove_uuid;
813 * Initialize the precomputed transaction reservations values.
818 * Allocate and initialize the per-ag data.
820 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
822 xfs_warn(mp, "Failed per-ag init: %d", error);
826 if (!sbp->sb_logblocks) {
827 xfs_warn(mp, "no log defined");
828 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
829 error = -EFSCORRUPTED;
834 * Log's mount-time initialization. The first part of recovery can place
835 * some items on the AIL, to be handled when recovery is finished or
838 error = xfs_log_mount(mp, mp->m_logdev_targp,
839 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
840 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
842 xfs_warn(mp, "log mount failed");
847 * Now the log is mounted, we know if it was an unclean shutdown or
848 * not. If it was, with the first phase of recovery has completed, we
849 * have consistent AG blocks on disk. We have not recovered EFIs yet,
850 * but they are recovered transactionally in the second recovery phase
853 * Hence we can safely re-initialise incore superblock counters from
854 * the per-ag data. These may not be correct if the filesystem was not
855 * cleanly unmounted, so we need to wait for recovery to finish before
858 * If the filesystem was cleanly unmounted, then we can trust the
859 * values in the superblock to be correct and we don't need to do
862 * If we are currently making the filesystem, the initialisation will
863 * fail as the perag data is in an undefined state.
865 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
866 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
867 !mp->m_sb.sb_inprogress) {
868 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
870 goto out_log_dealloc;
874 * Get and sanity-check the root inode.
875 * Save the pointer to it in the mount structure.
877 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
879 xfs_warn(mp, "failed to read root inode");
880 goto out_log_dealloc;
885 if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
886 xfs_warn(mp, "corrupted root inode %llu: not a directory",
887 (unsigned long long)rip->i_ino);
888 xfs_iunlock(rip, XFS_ILOCK_EXCL);
889 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
891 error = -EFSCORRUPTED;
894 mp->m_rootip = rip; /* save it */
896 xfs_iunlock(rip, XFS_ILOCK_EXCL);
899 * Initialize realtime inode pointers in the mount structure
901 error = xfs_rtmount_inodes(mp);
904 * Free up the root inode.
906 xfs_warn(mp, "failed to read RT inodes");
911 * If this is a read-only mount defer the superblock updates until
912 * the next remount into writeable mode. Otherwise we would never
913 * perform the update e.g. for the root filesystem.
915 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
916 error = xfs_sync_sb(mp, false);
918 xfs_warn(mp, "failed to write sb changes");
924 * Initialise the XFS quota management subsystem for this mount
926 if (XFS_IS_QUOTA_RUNNING(mp)) {
927 error = xfs_qm_newmount(mp, "amount, "aflags);
931 ASSERT(!XFS_IS_QUOTA_ON(mp));
934 * If a file system had quotas running earlier, but decided to
935 * mount without -o uquota/pquota/gquota options, revoke the
936 * quotachecked license.
938 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
939 xfs_notice(mp, "resetting quota flags");
940 error = xfs_mount_reset_sbqflags(mp);
947 * Finish recovering the file system. This part needed to be delayed
948 * until after the root and real-time bitmap inodes were consistently
951 error = xfs_log_mount_finish(mp);
953 xfs_warn(mp, "log mount finish failed");
958 * Now the log is fully replayed, we can transition to full read-only
959 * mode for read-only mounts. This will sync all the metadata and clean
960 * the log so that the recovery we just performed does not have to be
961 * replayed again on the next mount.
963 * We use the same quiesce mechanism as the rw->ro remount, as they are
964 * semantically identical operations.
966 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
968 xfs_quiesce_attr(mp);
972 * Complete the quota initialisation, post-log-replay component.
975 ASSERT(mp->m_qflags == 0);
976 mp->m_qflags = quotaflags;
978 xfs_qm_mount_quotas(mp);
982 * Now we are mounted, reserve a small amount of unused space for
983 * privileged transactions. This is needed so that transaction
984 * space required for critical operations can dip into this pool
985 * when at ENOSPC. This is needed for operations like create with
986 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
987 * are not allowed to use this reserved space.
989 * This may drive us straight to ENOSPC on mount, but that implies
990 * we were already there on the last unmount. Warn if this occurs.
992 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
993 resblks = xfs_default_resblks(mp);
994 error = xfs_reserve_blocks(mp, &resblks, NULL);
997 "Unable to allocate reserve blocks. Continuing without reserve pool.");
999 /* Recover any CoW blocks that never got remapped. */
1000 error = xfs_reflink_recover_cow(mp);
1003 "Error %d recovering leftover CoW allocations.", error);
1004 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1008 /* Reserve AG blocks for future btree expansion. */
1009 error = xfs_fs_reserve_ag_blocks(mp);
1010 if (error && error != -ENOSPC)
1017 xfs_fs_unreserve_ag_blocks(mp);
1019 xfs_qm_unmount_quotas(mp);
1021 xfs_rtunmount_inodes(mp);
1024 /* Clean out dquots that might be in memory after quotacheck. */
1027 * Cancel all delayed reclaim work and reclaim the inodes directly.
1028 * We have to do this /after/ rtunmount and qm_unmount because those
1029 * two will have scheduled delayed reclaim for the rt/quota inodes.
1031 * This is slightly different from the unmountfs call sequence
1032 * because we could be tearing down a partially set up mount. In
1033 * particular, if log_mount_finish fails we bail out without calling
1034 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1037 cancel_delayed_work_sync(&mp->m_reclaim_work);
1038 xfs_reclaim_inodes(mp, SYNC_WAIT);
1040 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1041 xfs_log_mount_cancel(mp);
1043 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1044 xfs_wait_buftarg(mp->m_logdev_targp);
1045 xfs_wait_buftarg(mp->m_ddev_targp);
1051 xfs_uuid_unmount(mp);
1052 out_remove_errortag:
1053 xfs_errortag_del(mp);
1054 out_remove_error_sysfs:
1055 xfs_error_sysfs_del(mp);
1057 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1059 xfs_sysfs_del(&mp->m_kobj);
1065 * This flushes out the inodes,dquots and the superblock, unmounts the
1066 * log and makes sure that incore structures are freed.
1070 struct xfs_mount *mp)
1075 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1076 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1078 xfs_fs_unreserve_ag_blocks(mp);
1079 xfs_qm_unmount_quotas(mp);
1080 xfs_rtunmount_inodes(mp);
1081 IRELE(mp->m_rootip);
1084 * We can potentially deadlock here if we have an inode cluster
1085 * that has been freed has its buffer still pinned in memory because
1086 * the transaction is still sitting in a iclog. The stale inodes
1087 * on that buffer will have their flush locks held until the
1088 * transaction hits the disk and the callbacks run. the inode
1089 * flush takes the flush lock unconditionally and with nothing to
1090 * push out the iclog we will never get that unlocked. hence we
1091 * need to force the log first.
1093 xfs_log_force(mp, XFS_LOG_SYNC);
1096 * Wait for all busy extents to be freed, including completion of
1097 * any discard operation.
1099 xfs_extent_busy_wait_all(mp);
1100 flush_workqueue(xfs_discard_wq);
1103 * We now need to tell the world we are unmounting. This will allow
1104 * us to detect that the filesystem is going away and we should error
1105 * out anything that we have been retrying in the background. This will
1106 * prevent neverending retries in AIL pushing from hanging the unmount.
1108 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1111 * Flush all pending changes from the AIL.
1113 xfs_ail_push_all_sync(mp->m_ail);
1116 * And reclaim all inodes. At this point there should be no dirty
1117 * inodes and none should be pinned or locked, but use synchronous
1118 * reclaim just to be sure. We can stop background inode reclaim
1119 * here as well if it is still running.
1121 cancel_delayed_work_sync(&mp->m_reclaim_work);
1122 xfs_reclaim_inodes(mp, SYNC_WAIT);
1127 * Unreserve any blocks we have so that when we unmount we don't account
1128 * the reserved free space as used. This is really only necessary for
1129 * lazy superblock counting because it trusts the incore superblock
1130 * counters to be absolutely correct on clean unmount.
1132 * We don't bother correcting this elsewhere for lazy superblock
1133 * counting because on mount of an unclean filesystem we reconstruct the
1134 * correct counter value and this is irrelevant.
1136 * For non-lazy counter filesystems, this doesn't matter at all because
1137 * we only every apply deltas to the superblock and hence the incore
1138 * value does not matter....
1141 error = xfs_reserve_blocks(mp, &resblks, NULL);
1143 xfs_warn(mp, "Unable to free reserved block pool. "
1144 "Freespace may not be correct on next mount.");
1146 error = xfs_log_sbcount(mp);
1148 xfs_warn(mp, "Unable to update superblock counters. "
1149 "Freespace may not be correct on next mount.");
1152 xfs_log_unmount(mp);
1154 xfs_uuid_unmount(mp);
1157 xfs_errortag_clearall(mp);
1161 xfs_errortag_del(mp);
1162 xfs_error_sysfs_del(mp);
1163 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1164 xfs_sysfs_del(&mp->m_kobj);
1168 * Determine whether modifications can proceed. The caller specifies the minimum
1169 * freeze level for which modifications should not be allowed. This allows
1170 * certain operations to proceed while the freeze sequence is in progress, if
1175 struct xfs_mount *mp,
1178 ASSERT(level > SB_UNFROZEN);
1179 if ((mp->m_super->s_writers.frozen >= level) ||
1180 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1189 * Sync the superblock counters to disk.
1191 * Note this code can be called during the process of freezing, so we use the
1192 * transaction allocator that does not block when the transaction subsystem is
1193 * in its frozen state.
1196 xfs_log_sbcount(xfs_mount_t *mp)
1198 /* allow this to proceed during the freeze sequence... */
1199 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1203 * we don't need to do this if we are updating the superblock
1204 * counters on every modification.
1206 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1209 return xfs_sync_sb(mp, true);
1213 * Deltas for the inode count are +/-64, hence we use a large batch size
1214 * of 128 so we don't need to take the counter lock on every update.
1216 #define XFS_ICOUNT_BATCH 128
1219 struct xfs_mount *mp,
1222 percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1223 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1225 percpu_counter_add(&mp->m_icount, -delta);
1233 struct xfs_mount *mp,
1236 percpu_counter_add(&mp->m_ifree, delta);
1237 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1239 percpu_counter_add(&mp->m_ifree, -delta);
1246 * Deltas for the block count can vary from 1 to very large, but lock contention
1247 * only occurs on frequent small block count updates such as in the delayed
1248 * allocation path for buffered writes (page a time updates). Hence we set
1249 * a large batch count (1024) to minimise global counter updates except when
1250 * we get near to ENOSPC and we have to be very accurate with our updates.
1252 #define XFS_FDBLOCKS_BATCH 1024
1255 struct xfs_mount *mp,
1265 * If the reserve pool is depleted, put blocks back into it
1266 * first. Most of the time the pool is full.
1268 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1269 percpu_counter_add(&mp->m_fdblocks, delta);
1273 spin_lock(&mp->m_sb_lock);
1274 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1276 if (res_used > delta) {
1277 mp->m_resblks_avail += delta;
1280 mp->m_resblks_avail = mp->m_resblks;
1281 percpu_counter_add(&mp->m_fdblocks, delta);
1283 spin_unlock(&mp->m_sb_lock);
1288 * Taking blocks away, need to be more accurate the closer we
1291 * If the counter has a value of less than 2 * max batch size,
1292 * then make everything serialise as we are real close to
1295 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1296 XFS_FDBLOCKS_BATCH) < 0)
1299 batch = XFS_FDBLOCKS_BATCH;
1301 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1302 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1303 XFS_FDBLOCKS_BATCH) >= 0) {
1309 * lock up the sb for dipping into reserves before releasing the space
1310 * that took us to ENOSPC.
1312 spin_lock(&mp->m_sb_lock);
1313 percpu_counter_add(&mp->m_fdblocks, -delta);
1315 goto fdblocks_enospc;
1317 lcounter = (long long)mp->m_resblks_avail + delta;
1318 if (lcounter >= 0) {
1319 mp->m_resblks_avail = lcounter;
1320 spin_unlock(&mp->m_sb_lock);
1323 printk_once(KERN_WARNING
1324 "Filesystem \"%s\": reserve blocks depleted! "
1325 "Consider increasing reserve pool size.",
1328 spin_unlock(&mp->m_sb_lock);
1334 struct xfs_mount *mp,
1340 spin_lock(&mp->m_sb_lock);
1341 lcounter = mp->m_sb.sb_frextents + delta;
1345 mp->m_sb.sb_frextents = lcounter;
1346 spin_unlock(&mp->m_sb_lock);
1351 * xfs_getsb() is called to obtain the buffer for the superblock.
1352 * The buffer is returned locked and read in from disk.
1353 * The buffer should be released with a call to xfs_brelse().
1355 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1356 * the superblock buffer if it can be locked without sleeping.
1357 * If it can't then we'll return NULL.
1361 struct xfs_mount *mp,
1364 struct xfs_buf *bp = mp->m_sb_bp;
1366 if (!xfs_buf_trylock(bp)) {
1367 if (flags & XBF_TRYLOCK)
1373 ASSERT(bp->b_flags & XBF_DONE);
1378 * Used to free the superblock along various error paths.
1382 struct xfs_mount *mp)
1384 struct xfs_buf *bp = mp->m_sb_bp;
1392 * If the underlying (data/log/rt) device is readonly, there are some
1393 * operations that cannot proceed.
1396 xfs_dev_is_read_only(
1397 struct xfs_mount *mp,
1400 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1401 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1402 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1403 xfs_notice(mp, "%s required on read-only device.", message);
1404 xfs_notice(mp, "write access unavailable, cannot proceed.");