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_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_alloc.h"
33 #include "xfs_rtalloc.h"
35 #include "xfs_trans.h"
36 #include "xfs_trans_priv.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_fsops.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_sysfs.h"
46 static DEFINE_MUTEX(xfs_uuid_table_mutex);
47 static int xfs_uuid_table_size;
48 static uuid_t *xfs_uuid_table;
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 if (mp->m_flags & XFS_MOUNT_NOUUID)
64 if (uuid_is_nil(uuid)) {
65 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
69 mutex_lock(&xfs_uuid_table_mutex);
70 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
71 if (uuid_is_nil(&xfs_uuid_table[i])) {
75 if (uuid_equal(uuid, &xfs_uuid_table[i]))
80 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
81 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
82 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
84 hole = xfs_uuid_table_size++;
86 xfs_uuid_table[hole] = *uuid;
87 mutex_unlock(&xfs_uuid_table_mutex);
92 mutex_unlock(&xfs_uuid_table_mutex);
93 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
101 uuid_t *uuid = &mp->m_sb.sb_uuid;
104 if (mp->m_flags & XFS_MOUNT_NOUUID)
107 mutex_lock(&xfs_uuid_table_mutex);
108 for (i = 0; i < xfs_uuid_table_size; i++) {
109 if (uuid_is_nil(&xfs_uuid_table[i]))
111 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
113 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
116 ASSERT(i < xfs_uuid_table_size);
117 mutex_unlock(&xfs_uuid_table_mutex);
123 struct rcu_head *head)
125 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
127 ASSERT(atomic_read(&pag->pag_ref) == 0);
132 * Free up the per-ag resources associated with the mount structure.
139 struct xfs_perag *pag;
141 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
142 spin_lock(&mp->m_perag_lock);
143 pag = radix_tree_delete(&mp->m_perag_tree, agno);
144 spin_unlock(&mp->m_perag_lock);
146 ASSERT(atomic_read(&pag->pag_ref) == 0);
147 call_rcu(&pag->rcu_head, __xfs_free_perag);
152 * Check size of device based on the (data/realtime) block count.
153 * Note: this check is used by the growfs code as well as mount.
156 xfs_sb_validate_fsb_count(
160 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
161 ASSERT(sbp->sb_blocklog >= BBSHIFT);
163 /* Limited by ULONG_MAX of page cache index */
164 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
170 xfs_initialize_perag(
172 xfs_agnumber_t agcount,
173 xfs_agnumber_t *maxagi)
175 xfs_agnumber_t index;
176 xfs_agnumber_t first_initialised = 0;
180 xfs_sb_t *sbp = &mp->m_sb;
184 * Walk the current per-ag tree so we don't try to initialise AGs
185 * that already exist (growfs case). Allocate and insert all the
186 * AGs we don't find ready for initialisation.
188 for (index = 0; index < agcount; index++) {
189 pag = xfs_perag_get(mp, index);
194 if (!first_initialised)
195 first_initialised = index;
197 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
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 spin_lock_init(&pag->pag_buf_lock);
206 pag->pag_buf_tree = RB_ROOT;
208 if (radix_tree_preload(GFP_NOFS))
211 spin_lock(&mp->m_perag_lock);
212 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
214 spin_unlock(&mp->m_perag_lock);
215 radix_tree_preload_end();
219 spin_unlock(&mp->m_perag_lock);
220 radix_tree_preload_end();
224 * If we mount with the inode64 option, or no inode overflows
225 * the legacy 32-bit address space clear the inode32 option.
227 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
228 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
230 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
231 mp->m_flags |= XFS_MOUNT_32BITINODES;
233 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
235 if (mp->m_flags & XFS_MOUNT_32BITINODES)
236 index = xfs_set_inode32(mp, agcount);
238 index = xfs_set_inode64(mp, agcount);
246 for (; index > first_initialised; index--) {
247 pag = radix_tree_delete(&mp->m_perag_tree, index);
256 * Does the initial read of the superblock.
260 struct xfs_mount *mp,
263 unsigned int sector_size;
265 struct xfs_sb *sbp = &mp->m_sb;
267 int loud = !(flags & XFS_MFSI_QUIET);
268 const struct xfs_buf_ops *buf_ops;
270 ASSERT(mp->m_sb_bp == NULL);
271 ASSERT(mp->m_ddev_targp != NULL);
274 * For the initial read, we must guess at the sector
275 * size based on the block device. It's enough to
276 * get the sb_sectsize out of the superblock and
277 * then reread with the proper length.
278 * We don't verify it yet, because it may not be complete.
280 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
284 * Allocate a (locked) buffer to hold the superblock.
285 * This will be kept around at all times to optimize
286 * access to the superblock.
289 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
290 BTOBB(sector_size), 0, &bp, buf_ops);
293 xfs_warn(mp, "SB validate failed with error %d.", error);
294 /* bad CRC means corrupted metadata */
295 if (error == -EFSBADCRC)
296 error = -EFSCORRUPTED;
301 * Initialize the mount structure from the superblock.
303 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
306 * If we haven't validated the superblock, do so now before we try
307 * to check the sector size and reread the superblock appropriately.
309 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
311 xfs_warn(mp, "Invalid superblock magic number");
317 * We must be able to do sector-sized and sector-aligned IO.
319 if (sector_size > sbp->sb_sectsize) {
321 xfs_warn(mp, "device supports %u byte sectors (not %u)",
322 sector_size, sbp->sb_sectsize);
327 if (buf_ops == NULL) {
329 * Re-read the superblock so the buffer is correctly sized,
330 * and properly verified.
333 sector_size = sbp->sb_sectsize;
334 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
338 xfs_reinit_percpu_counters(mp);
340 /* no need to be quiet anymore, so reset the buf ops */
341 bp->b_ops = &xfs_sb_buf_ops;
353 * Update alignment values based on mount options and sb values
356 xfs_update_alignment(xfs_mount_t *mp)
358 xfs_sb_t *sbp = &(mp->m_sb);
362 * If stripe unit and stripe width are not multiples
363 * of the fs blocksize turn off alignment.
365 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
366 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
368 "alignment check failed: sunit/swidth vs. blocksize(%d)",
373 * Convert the stripe unit and width to FSBs.
375 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
376 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
378 "alignment check failed: sunit/swidth vs. agsize(%d)",
381 } else if (mp->m_dalign) {
382 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
385 "alignment check failed: sunit(%d) less than bsize(%d)",
386 mp->m_dalign, sbp->sb_blocksize);
392 * Update superblock with new values
395 if (xfs_sb_version_hasdalign(sbp)) {
396 if (sbp->sb_unit != mp->m_dalign) {
397 sbp->sb_unit = mp->m_dalign;
398 mp->m_update_sb = true;
400 if (sbp->sb_width != mp->m_swidth) {
401 sbp->sb_width = mp->m_swidth;
402 mp->m_update_sb = true;
406 "cannot change alignment: superblock does not support data alignment");
409 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
410 xfs_sb_version_hasdalign(&mp->m_sb)) {
411 mp->m_dalign = sbp->sb_unit;
412 mp->m_swidth = sbp->sb_width;
419 * Set the maximum inode count for this filesystem
422 xfs_set_maxicount(xfs_mount_t *mp)
424 xfs_sb_t *sbp = &(mp->m_sb);
427 if (sbp->sb_imax_pct) {
429 * Make sure the maximum inode count is a multiple
430 * of the units we allocate inodes in.
432 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
434 do_div(icount, mp->m_ialloc_blks);
435 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
443 * Set the default minimum read and write sizes unless
444 * already specified in a mount option.
445 * We use smaller I/O sizes when the file system
446 * is being used for NFS service (wsync mount option).
449 xfs_set_rw_sizes(xfs_mount_t *mp)
451 xfs_sb_t *sbp = &(mp->m_sb);
452 int readio_log, writeio_log;
454 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
455 if (mp->m_flags & XFS_MOUNT_WSYNC) {
456 readio_log = XFS_WSYNC_READIO_LOG;
457 writeio_log = XFS_WSYNC_WRITEIO_LOG;
459 readio_log = XFS_READIO_LOG_LARGE;
460 writeio_log = XFS_WRITEIO_LOG_LARGE;
463 readio_log = mp->m_readio_log;
464 writeio_log = mp->m_writeio_log;
467 if (sbp->sb_blocklog > readio_log) {
468 mp->m_readio_log = sbp->sb_blocklog;
470 mp->m_readio_log = readio_log;
472 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
473 if (sbp->sb_blocklog > writeio_log) {
474 mp->m_writeio_log = sbp->sb_blocklog;
476 mp->m_writeio_log = writeio_log;
478 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
482 * precalculate the low space thresholds for dynamic speculative preallocation.
485 xfs_set_low_space_thresholds(
486 struct xfs_mount *mp)
490 for (i = 0; i < XFS_LOWSP_MAX; i++) {
491 __uint64_t space = mp->m_sb.sb_dblocks;
494 mp->m_low_space[i] = space * (i + 1);
500 * Set whether we're using inode alignment.
503 xfs_set_inoalignment(xfs_mount_t *mp)
505 if (xfs_sb_version_hasalign(&mp->m_sb) &&
506 mp->m_sb.sb_inoalignmt >=
507 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
508 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
510 mp->m_inoalign_mask = 0;
512 * If we are using stripe alignment, check whether
513 * the stripe unit is a multiple of the inode alignment
515 if (mp->m_dalign && mp->m_inoalign_mask &&
516 !(mp->m_dalign & mp->m_inoalign_mask))
517 mp->m_sinoalign = mp->m_dalign;
523 * Check that the data (and log if separate) is an ok size.
527 struct xfs_mount *mp)
533 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
534 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
535 xfs_warn(mp, "filesystem size mismatch detected");
538 error = xfs_buf_read_uncached(mp->m_ddev_targp,
539 d - XFS_FSS_TO_BB(mp, 1),
540 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
542 xfs_warn(mp, "last sector read failed");
547 if (mp->m_logdev_targp == mp->m_ddev_targp)
550 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
551 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
552 xfs_warn(mp, "log size mismatch detected");
555 error = xfs_buf_read_uncached(mp->m_logdev_targp,
556 d - XFS_FSB_TO_BB(mp, 1),
557 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
559 xfs_warn(mp, "log device read failed");
567 * Clear the quotaflags in memory and in the superblock.
570 xfs_mount_reset_sbqflags(
571 struct xfs_mount *mp)
575 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
576 if (mp->m_sb.sb_qflags == 0)
578 spin_lock(&mp->m_sb_lock);
579 mp->m_sb.sb_qflags = 0;
580 spin_unlock(&mp->m_sb_lock);
582 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
585 return xfs_sync_sb(mp, false);
589 xfs_default_resblks(xfs_mount_t *mp)
594 * We default to 5% or 8192 fsbs of space reserved, whichever is
595 * smaller. This is intended to cover concurrent allocation
596 * transactions when we initially hit enospc. These each require a 4
597 * block reservation. Hence by default we cover roughly 2000 concurrent
598 * allocation reservations.
600 resblks = mp->m_sb.sb_dblocks;
602 resblks = min_t(__uint64_t, resblks, 8192);
607 * This function does the following on an initial mount of a file system:
608 * - reads the superblock from disk and init the mount struct
609 * - if we're a 32-bit kernel, do a size check on the superblock
610 * so we don't mount terabyte filesystems
611 * - init mount struct realtime fields
612 * - allocate inode hash table for fs
613 * - init directory manager
614 * - perform recovery and init the log manager
620 xfs_sb_t *sbp = &(mp->m_sb);
627 xfs_sb_mount_common(mp, sbp);
630 * Check for a mismatched features2 values. Older kernels read & wrote
631 * into the wrong sb offset for sb_features2 on some platforms due to
632 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
633 * which made older superblock reading/writing routines swap it as a
636 * For backwards compatibility, we make both slots equal.
638 * If we detect a mismatched field, we OR the set bits into the existing
639 * features2 field in case it has already been modified; we don't want
640 * to lose any features. We then update the bad location with the ORed
641 * value so that older kernels will see any features2 flags. The
642 * superblock writeback code ensures the new sb_features2 is copied to
643 * sb_bad_features2 before it is logged or written to disk.
645 if (xfs_sb_has_mismatched_features2(sbp)) {
646 xfs_warn(mp, "correcting sb_features alignment problem");
647 sbp->sb_features2 |= sbp->sb_bad_features2;
648 mp->m_update_sb = true;
651 * Re-check for ATTR2 in case it was found in bad_features2
654 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
655 !(mp->m_flags & XFS_MOUNT_NOATTR2))
656 mp->m_flags |= XFS_MOUNT_ATTR2;
659 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
660 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
661 xfs_sb_version_removeattr2(&mp->m_sb);
662 mp->m_update_sb = true;
664 /* update sb_versionnum for the clearing of the morebits */
665 if (!sbp->sb_features2)
666 mp->m_update_sb = true;
669 /* always use v2 inodes by default now */
670 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
671 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
672 mp->m_update_sb = true;
676 * Check if sb_agblocks is aligned at stripe boundary
677 * If sb_agblocks is NOT aligned turn off m_dalign since
678 * allocator alignment is within an ag, therefore ag has
679 * to be aligned at stripe boundary.
681 error = xfs_update_alignment(mp);
685 xfs_alloc_compute_maxlevels(mp);
686 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
687 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
688 xfs_ialloc_compute_maxlevels(mp);
690 xfs_set_maxicount(mp);
692 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
696 error = xfs_uuid_mount(mp);
698 goto out_remove_sysfs;
701 * Set the minimum read and write sizes
703 xfs_set_rw_sizes(mp);
705 /* set the low space thresholds for dynamic preallocation */
706 xfs_set_low_space_thresholds(mp);
709 * Set the inode cluster size.
710 * This may still be overridden by the file system
711 * block size if it is larger than the chosen cluster size.
713 * For v5 filesystems, scale the cluster size with the inode size to
714 * keep a constant ratio of inode per cluster buffer, but only if mkfs
715 * has set the inode alignment value appropriately for larger cluster
718 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
719 if (xfs_sb_version_hascrc(&mp->m_sb)) {
720 int new_size = mp->m_inode_cluster_size;
722 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
723 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
724 mp->m_inode_cluster_size = new_size;
728 * Set inode alignment fields
730 xfs_set_inoalignment(mp);
733 * Check that the data (and log if separate) is an ok size.
735 error = xfs_check_sizes(mp);
737 goto out_remove_uuid;
740 * Initialize realtime fields in the mount structure
742 error = xfs_rtmount_init(mp);
744 xfs_warn(mp, "RT mount failed");
745 goto out_remove_uuid;
749 * Copies the low order bits of the timestamp and the randomly
750 * set "sequence" number out of a UUID.
752 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
754 mp->m_dmevmask = 0; /* not persistent; set after each mount */
756 error = xfs_da_mount(mp);
758 xfs_warn(mp, "Failed dir/attr init: %d", error);
759 goto out_remove_uuid;
763 * Initialize the precomputed transaction reservations values.
768 * Allocate and initialize the per-ag data.
770 spin_lock_init(&mp->m_perag_lock);
771 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
772 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
774 xfs_warn(mp, "Failed per-ag init: %d", error);
778 if (!sbp->sb_logblocks) {
779 xfs_warn(mp, "no log defined");
780 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
781 error = -EFSCORRUPTED;
786 * log's mount-time initialization. Perform 1st part recovery if needed
788 error = xfs_log_mount(mp, mp->m_logdev_targp,
789 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
790 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
792 xfs_warn(mp, "log mount failed");
797 * Now the log is mounted, we know if it was an unclean shutdown or
798 * not. If it was, with the first phase of recovery has completed, we
799 * have consistent AG blocks on disk. We have not recovered EFIs yet,
800 * but they are recovered transactionally in the second recovery phase
803 * Hence we can safely re-initialise incore superblock counters from
804 * the per-ag data. These may not be correct if the filesystem was not
805 * cleanly unmounted, so we need to wait for recovery to finish before
808 * If the filesystem was cleanly unmounted, then we can trust the
809 * values in the superblock to be correct and we don't need to do
812 * If we are currently making the filesystem, the initialisation will
813 * fail as the perag data is in an undefined state.
815 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
816 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
817 !mp->m_sb.sb_inprogress) {
818 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
820 goto out_log_dealloc;
824 * Get and sanity-check the root inode.
825 * Save the pointer to it in the mount structure.
827 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
829 xfs_warn(mp, "failed to read root inode");
830 goto out_log_dealloc;
835 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
836 xfs_warn(mp, "corrupted root inode %llu: not a directory",
837 (unsigned long long)rip->i_ino);
838 xfs_iunlock(rip, XFS_ILOCK_EXCL);
839 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
841 error = -EFSCORRUPTED;
844 mp->m_rootip = rip; /* save it */
846 xfs_iunlock(rip, XFS_ILOCK_EXCL);
849 * Initialize realtime inode pointers in the mount structure
851 error = xfs_rtmount_inodes(mp);
854 * Free up the root inode.
856 xfs_warn(mp, "failed to read RT inodes");
861 * If this is a read-only mount defer the superblock updates until
862 * the next remount into writeable mode. Otherwise we would never
863 * perform the update e.g. for the root filesystem.
865 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
866 error = xfs_sync_sb(mp, false);
868 xfs_warn(mp, "failed to write sb changes");
874 * Initialise the XFS quota management subsystem for this mount
876 if (XFS_IS_QUOTA_RUNNING(mp)) {
877 error = xfs_qm_newmount(mp, "amount, "aflags);
881 ASSERT(!XFS_IS_QUOTA_ON(mp));
884 * If a file system had quotas running earlier, but decided to
885 * mount without -o uquota/pquota/gquota options, revoke the
886 * quotachecked license.
888 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
889 xfs_notice(mp, "resetting quota flags");
890 error = xfs_mount_reset_sbqflags(mp);
897 * Finish recovering the file system. This part needed to be
898 * delayed until after the root and real-time bitmap inodes
899 * were consistently read in.
901 error = xfs_log_mount_finish(mp);
903 xfs_warn(mp, "log mount finish failed");
908 * Complete the quota initialisation, post-log-replay component.
911 ASSERT(mp->m_qflags == 0);
912 mp->m_qflags = quotaflags;
914 xfs_qm_mount_quotas(mp);
918 * Now we are mounted, reserve a small amount of unused space for
919 * privileged transactions. This is needed so that transaction
920 * space required for critical operations can dip into this pool
921 * when at ENOSPC. This is needed for operations like create with
922 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
923 * are not allowed to use this reserved space.
925 * This may drive us straight to ENOSPC on mount, but that implies
926 * we were already there on the last unmount. Warn if this occurs.
928 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
929 resblks = xfs_default_resblks(mp);
930 error = xfs_reserve_blocks(mp, &resblks, NULL);
933 "Unable to allocate reserve blocks. Continuing without reserve pool.");
939 xfs_rtunmount_inodes(mp);
945 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
946 xfs_wait_buftarg(mp->m_logdev_targp);
947 xfs_wait_buftarg(mp->m_ddev_targp);
953 xfs_uuid_unmount(mp);
955 xfs_sysfs_del(&mp->m_kobj);
961 * This flushes out the inodes,dquots and the superblock, unmounts the
962 * log and makes sure that incore structures are freed.
966 struct xfs_mount *mp)
971 cancel_delayed_work_sync(&mp->m_eofblocks_work);
973 xfs_qm_unmount_quotas(mp);
974 xfs_rtunmount_inodes(mp);
978 * We can potentially deadlock here if we have an inode cluster
979 * that has been freed has its buffer still pinned in memory because
980 * the transaction is still sitting in a iclog. The stale inodes
981 * on that buffer will have their flush locks held until the
982 * transaction hits the disk and the callbacks run. the inode
983 * flush takes the flush lock unconditionally and with nothing to
984 * push out the iclog we will never get that unlocked. hence we
985 * need to force the log first.
987 xfs_log_force(mp, XFS_LOG_SYNC);
990 * Flush all pending changes from the AIL.
992 xfs_ail_push_all_sync(mp->m_ail);
995 * And reclaim all inodes. At this point there should be no dirty
996 * inodes and none should be pinned or locked, but use synchronous
997 * reclaim just to be sure. We can stop background inode reclaim
998 * here as well if it is still running.
1000 cancel_delayed_work_sync(&mp->m_reclaim_work);
1001 xfs_reclaim_inodes(mp, SYNC_WAIT);
1006 * Unreserve any blocks we have so that when we unmount we don't account
1007 * the reserved free space as used. This is really only necessary for
1008 * lazy superblock counting because it trusts the incore superblock
1009 * counters to be absolutely correct on clean unmount.
1011 * We don't bother correcting this elsewhere for lazy superblock
1012 * counting because on mount of an unclean filesystem we reconstruct the
1013 * correct counter value and this is irrelevant.
1015 * For non-lazy counter filesystems, this doesn't matter at all because
1016 * we only every apply deltas to the superblock and hence the incore
1017 * value does not matter....
1020 error = xfs_reserve_blocks(mp, &resblks, NULL);
1022 xfs_warn(mp, "Unable to free reserved block pool. "
1023 "Freespace may not be correct on next mount.");
1025 error = xfs_log_sbcount(mp);
1027 xfs_warn(mp, "Unable to update superblock counters. "
1028 "Freespace may not be correct on next mount.");
1030 xfs_log_unmount(mp);
1032 xfs_uuid_unmount(mp);
1035 xfs_errortag_clearall(mp, 0);
1039 xfs_sysfs_del(&mp->m_kobj);
1043 * Determine whether modifications can proceed. The caller specifies the minimum
1044 * freeze level for which modifications should not be allowed. This allows
1045 * certain operations to proceed while the freeze sequence is in progress, if
1050 struct xfs_mount *mp,
1053 ASSERT(level > SB_UNFROZEN);
1054 if ((mp->m_super->s_writers.frozen >= level) ||
1055 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1064 * Sync the superblock counters to disk.
1066 * Note this code can be called during the process of freezing, so we use the
1067 * transaction allocator that does not block when the transaction subsystem is
1068 * in its frozen state.
1071 xfs_log_sbcount(xfs_mount_t *mp)
1073 /* allow this to proceed during the freeze sequence... */
1074 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1078 * we don't need to do this if we are updating the superblock
1079 * counters on every modification.
1081 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1084 return xfs_sync_sb(mp, true);
1089 struct xfs_mount *mp,
1092 /* deltas are +/-64, hence the large batch size of 128. */
1093 __percpu_counter_add(&mp->m_icount, delta, 128);
1094 if (percpu_counter_compare(&mp->m_icount, 0) < 0) {
1096 percpu_counter_add(&mp->m_icount, -delta);
1104 struct xfs_mount *mp,
1107 percpu_counter_add(&mp->m_ifree, delta);
1108 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1110 percpu_counter_add(&mp->m_ifree, -delta);
1118 struct xfs_mount *mp,
1128 * If the reserve pool is depleted, put blocks back into it
1129 * first. Most of the time the pool is full.
1131 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1132 percpu_counter_add(&mp->m_fdblocks, delta);
1136 spin_lock(&mp->m_sb_lock);
1137 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1139 if (res_used > delta) {
1140 mp->m_resblks_avail += delta;
1143 mp->m_resblks_avail = mp->m_resblks;
1144 percpu_counter_add(&mp->m_fdblocks, delta);
1146 spin_unlock(&mp->m_sb_lock);
1151 * Taking blocks away, need to be more accurate the closer we
1154 * batch size is set to a maximum of 1024 blocks - if we are
1155 * allocating of freeing extents larger than this then we aren't
1156 * going to be hammering the counter lock so a lock per update
1159 * If the counter has a value of less than 2 * max batch size,
1160 * then make everything serialise as we are real close to
1163 #define __BATCH 1024
1164 if (percpu_counter_compare(&mp->m_fdblocks, 2 * __BATCH) < 0)
1170 __percpu_counter_add(&mp->m_fdblocks, delta, batch);
1171 if (percpu_counter_compare(&mp->m_fdblocks,
1172 XFS_ALLOC_SET_ASIDE(mp)) >= 0) {
1178 * lock up the sb for dipping into reserves before releasing the space
1179 * that took us to ENOSPC.
1181 spin_lock(&mp->m_sb_lock);
1182 percpu_counter_add(&mp->m_fdblocks, -delta);
1184 goto fdblocks_enospc;
1186 lcounter = (long long)mp->m_resblks_avail + delta;
1187 if (lcounter >= 0) {
1188 mp->m_resblks_avail = lcounter;
1189 spin_unlock(&mp->m_sb_lock);
1192 printk_once(KERN_WARNING
1193 "Filesystem \"%s\": reserve blocks depleted! "
1194 "Consider increasing reserve pool size.",
1197 spin_unlock(&mp->m_sb_lock);
1203 struct xfs_mount *mp,
1209 spin_lock(&mp->m_sb_lock);
1210 lcounter = mp->m_sb.sb_frextents + delta;
1214 mp->m_sb.sb_frextents = lcounter;
1215 spin_unlock(&mp->m_sb_lock);
1220 * xfs_getsb() is called to obtain the buffer for the superblock.
1221 * The buffer is returned locked and read in from disk.
1222 * The buffer should be released with a call to xfs_brelse().
1224 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1225 * the superblock buffer if it can be locked without sleeping.
1226 * If it can't then we'll return NULL.
1230 struct xfs_mount *mp,
1233 struct xfs_buf *bp = mp->m_sb_bp;
1235 if (!xfs_buf_trylock(bp)) {
1236 if (flags & XBF_TRYLOCK)
1242 ASSERT(XFS_BUF_ISDONE(bp));
1247 * Used to free the superblock along various error paths.
1251 struct xfs_mount *mp)
1253 struct xfs_buf *bp = mp->m_sb_bp;
1261 * If the underlying (data/log/rt) device is readonly, there are some
1262 * operations that cannot proceed.
1265 xfs_dev_is_read_only(
1266 struct xfs_mount *mp,
1269 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1270 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1271 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1272 xfs_notice(mp, "%s required on read-only device.", message);
1273 xfs_notice(mp, "write access unavailable, cannot proceed.");