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"
50 static DEFINE_MUTEX(xfs_uuid_table_mutex);
51 static int xfs_uuid_table_size;
52 static uuid_t *xfs_uuid_table;
55 xfs_uuid_table_free(void)
57 if (xfs_uuid_table_size == 0)
59 kmem_free(xfs_uuid_table);
60 xfs_uuid_table = NULL;
61 xfs_uuid_table_size = 0;
65 * See if the UUID is unique among mounted XFS filesystems.
66 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
72 uuid_t *uuid = &mp->m_sb.sb_uuid;
75 if (mp->m_flags & XFS_MOUNT_NOUUID)
78 if (uuid_is_nil(uuid)) {
79 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
83 mutex_lock(&xfs_uuid_table_mutex);
84 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
85 if (uuid_is_nil(&xfs_uuid_table[i])) {
89 if (uuid_equal(uuid, &xfs_uuid_table[i]))
94 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
95 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
97 hole = xfs_uuid_table_size++;
99 xfs_uuid_table[hole] = *uuid;
100 mutex_unlock(&xfs_uuid_table_mutex);
105 mutex_unlock(&xfs_uuid_table_mutex);
106 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
112 struct xfs_mount *mp)
114 uuid_t *uuid = &mp->m_sb.sb_uuid;
117 if (mp->m_flags & XFS_MOUNT_NOUUID)
120 mutex_lock(&xfs_uuid_table_mutex);
121 for (i = 0; i < xfs_uuid_table_size; i++) {
122 if (uuid_is_nil(&xfs_uuid_table[i]))
124 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
126 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
129 ASSERT(i < xfs_uuid_table_size);
130 mutex_unlock(&xfs_uuid_table_mutex);
136 struct rcu_head *head)
138 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
140 ASSERT(atomic_read(&pag->pag_ref) == 0);
145 * Free up the per-ag resources associated with the mount structure.
152 struct xfs_perag *pag;
154 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
155 spin_lock(&mp->m_perag_lock);
156 pag = radix_tree_delete(&mp->m_perag_tree, agno);
157 spin_unlock(&mp->m_perag_lock);
159 ASSERT(atomic_read(&pag->pag_ref) == 0);
160 xfs_buf_hash_destroy(pag);
161 call_rcu(&pag->rcu_head, __xfs_free_perag);
166 * Check size of device based on the (data/realtime) block count.
167 * Note: this check is used by the growfs code as well as mount.
170 xfs_sb_validate_fsb_count(
174 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
175 ASSERT(sbp->sb_blocklog >= BBSHIFT);
177 /* Limited by ULONG_MAX of page cache index */
178 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
184 xfs_initialize_perag(
186 xfs_agnumber_t agcount,
187 xfs_agnumber_t *maxagi)
189 xfs_agnumber_t index;
190 xfs_agnumber_t first_initialised = 0;
195 * Walk the current per-ag tree so we don't try to initialise AGs
196 * that already exist (growfs case). Allocate and insert all the
197 * AGs we don't find ready for initialisation.
199 for (index = 0; index < agcount; index++) {
200 pag = xfs_perag_get(mp, index);
205 if (!first_initialised)
206 first_initialised = index;
208 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
211 pag->pag_agno = index;
213 spin_lock_init(&pag->pag_ici_lock);
214 mutex_init(&pag->pag_ici_reclaim_lock);
215 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
216 if (xfs_buf_hash_init(pag))
219 if (radix_tree_preload(GFP_NOFS))
222 spin_lock(&mp->m_perag_lock);
223 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
225 spin_unlock(&mp->m_perag_lock);
226 radix_tree_preload_end();
230 spin_unlock(&mp->m_perag_lock);
231 radix_tree_preload_end();
234 index = xfs_set_inode_alloc(mp, agcount);
239 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
243 xfs_buf_hash_destroy(pag);
245 for (; index > first_initialised; index--) {
246 pag = radix_tree_delete(&mp->m_perag_tree, index);
247 xfs_buf_hash_destroy(pag);
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. This will be kept
285 * around at all times to optimize access to the superblock. Therefore,
286 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
290 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
291 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
295 xfs_warn(mp, "SB validate failed with error %d.", error);
296 /* bad CRC means corrupted metadata */
297 if (error == -EFSBADCRC)
298 error = -EFSCORRUPTED;
303 * Initialize the mount structure from the superblock.
305 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
308 * If we haven't validated the superblock, do so now before we try
309 * to check the sector size and reread the superblock appropriately.
311 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
313 xfs_warn(mp, "Invalid superblock magic number");
319 * We must be able to do sector-sized and sector-aligned IO.
321 if (sector_size > sbp->sb_sectsize) {
323 xfs_warn(mp, "device supports %u byte sectors (not %u)",
324 sector_size, sbp->sb_sectsize);
329 if (buf_ops == NULL) {
331 * Re-read the superblock so the buffer is correctly sized,
332 * and properly verified.
335 sector_size = sbp->sb_sectsize;
336 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
340 xfs_reinit_percpu_counters(mp);
342 /* no need to be quiet anymore, so reset the buf ops */
343 bp->b_ops = &xfs_sb_buf_ops;
355 * Update alignment values based on mount options and sb values
358 xfs_update_alignment(xfs_mount_t *mp)
360 xfs_sb_t *sbp = &(mp->m_sb);
364 * If stripe unit and stripe width are not multiples
365 * of the fs blocksize turn off alignment.
367 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
368 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
370 "alignment check failed: sunit/swidth vs. blocksize(%d)",
375 * Convert the stripe unit and width to FSBs.
377 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
378 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
380 "alignment check failed: sunit/swidth vs. agsize(%d)",
383 } else if (mp->m_dalign) {
384 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
387 "alignment check failed: sunit(%d) less than bsize(%d)",
388 mp->m_dalign, sbp->sb_blocksize);
394 * Update superblock with new values
397 if (xfs_sb_version_hasdalign(sbp)) {
398 if (sbp->sb_unit != mp->m_dalign) {
399 sbp->sb_unit = mp->m_dalign;
400 mp->m_update_sb = true;
402 if (sbp->sb_width != mp->m_swidth) {
403 sbp->sb_width = mp->m_swidth;
404 mp->m_update_sb = true;
408 "cannot change alignment: superblock does not support data alignment");
411 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
412 xfs_sb_version_hasdalign(&mp->m_sb)) {
413 mp->m_dalign = sbp->sb_unit;
414 mp->m_swidth = sbp->sb_width;
421 * Set the maximum inode count for this filesystem
424 xfs_set_maxicount(xfs_mount_t *mp)
426 xfs_sb_t *sbp = &(mp->m_sb);
429 if (sbp->sb_imax_pct) {
431 * Make sure the maximum inode count is a multiple
432 * of the units we allocate inodes in.
434 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
436 do_div(icount, mp->m_ialloc_blks);
437 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
445 * Set the default minimum read and write sizes unless
446 * already specified in a mount option.
447 * We use smaller I/O sizes when the file system
448 * is being used for NFS service (wsync mount option).
451 xfs_set_rw_sizes(xfs_mount_t *mp)
453 xfs_sb_t *sbp = &(mp->m_sb);
454 int readio_log, writeio_log;
456 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
457 if (mp->m_flags & XFS_MOUNT_WSYNC) {
458 readio_log = XFS_WSYNC_READIO_LOG;
459 writeio_log = XFS_WSYNC_WRITEIO_LOG;
461 readio_log = XFS_READIO_LOG_LARGE;
462 writeio_log = XFS_WRITEIO_LOG_LARGE;
465 readio_log = mp->m_readio_log;
466 writeio_log = mp->m_writeio_log;
469 if (sbp->sb_blocklog > readio_log) {
470 mp->m_readio_log = sbp->sb_blocklog;
472 mp->m_readio_log = readio_log;
474 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
475 if (sbp->sb_blocklog > writeio_log) {
476 mp->m_writeio_log = sbp->sb_blocklog;
478 mp->m_writeio_log = writeio_log;
480 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
484 * precalculate the low space thresholds for dynamic speculative preallocation.
487 xfs_set_low_space_thresholds(
488 struct xfs_mount *mp)
492 for (i = 0; i < XFS_LOWSP_MAX; i++) {
493 __uint64_t space = mp->m_sb.sb_dblocks;
496 mp->m_low_space[i] = space * (i + 1);
502 * Set whether we're using inode alignment.
505 xfs_set_inoalignment(xfs_mount_t *mp)
507 if (xfs_sb_version_hasalign(&mp->m_sb) &&
508 mp->m_sb.sb_inoalignmt >=
509 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
510 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
512 mp->m_inoalign_mask = 0;
514 * If we are using stripe alignment, check whether
515 * the stripe unit is a multiple of the inode alignment
517 if (mp->m_dalign && mp->m_inoalign_mask &&
518 !(mp->m_dalign & mp->m_inoalign_mask))
519 mp->m_sinoalign = mp->m_dalign;
525 * Check that the data (and log if separate) is an ok size.
529 struct xfs_mount *mp)
535 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
536 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
537 xfs_warn(mp, "filesystem size mismatch detected");
540 error = xfs_buf_read_uncached(mp->m_ddev_targp,
541 d - XFS_FSS_TO_BB(mp, 1),
542 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
544 xfs_warn(mp, "last sector read failed");
549 if (mp->m_logdev_targp == mp->m_ddev_targp)
552 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
553 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
554 xfs_warn(mp, "log size mismatch detected");
557 error = xfs_buf_read_uncached(mp->m_logdev_targp,
558 d - XFS_FSB_TO_BB(mp, 1),
559 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
561 xfs_warn(mp, "log device read failed");
569 * Clear the quotaflags in memory and in the superblock.
572 xfs_mount_reset_sbqflags(
573 struct xfs_mount *mp)
577 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
578 if (mp->m_sb.sb_qflags == 0)
580 spin_lock(&mp->m_sb_lock);
581 mp->m_sb.sb_qflags = 0;
582 spin_unlock(&mp->m_sb_lock);
584 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
587 return xfs_sync_sb(mp, false);
591 xfs_default_resblks(xfs_mount_t *mp)
596 * We default to 5% or 8192 fsbs of space reserved, whichever is
597 * smaller. This is intended to cover concurrent allocation
598 * transactions when we initially hit enospc. These each require a 4
599 * block reservation. Hence by default we cover roughly 2000 concurrent
600 * allocation reservations.
602 resblks = mp->m_sb.sb_dblocks;
604 resblks = min_t(__uint64_t, resblks, 8192);
609 * This function does the following on an initial mount of a file system:
610 * - reads the superblock from disk and init the mount struct
611 * - if we're a 32-bit kernel, do a size check on the superblock
612 * so we don't mount terabyte filesystems
613 * - init mount struct realtime fields
614 * - allocate inode hash table for fs
615 * - init directory manager
616 * - perform recovery and init the log manager
620 struct xfs_mount *mp)
622 struct xfs_sb *sbp = &(mp->m_sb);
623 struct xfs_inode *rip;
629 xfs_sb_mount_common(mp, sbp);
632 * Check for a mismatched features2 values. Older kernels read & wrote
633 * into the wrong sb offset for sb_features2 on some platforms due to
634 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
635 * which made older superblock reading/writing routines swap it as a
638 * For backwards compatibility, we make both slots equal.
640 * If we detect a mismatched field, we OR the set bits into the existing
641 * features2 field in case it has already been modified; we don't want
642 * to lose any features. We then update the bad location with the ORed
643 * value so that older kernels will see any features2 flags. The
644 * superblock writeback code ensures the new sb_features2 is copied to
645 * sb_bad_features2 before it is logged or written to disk.
647 if (xfs_sb_has_mismatched_features2(sbp)) {
648 xfs_warn(mp, "correcting sb_features alignment problem");
649 sbp->sb_features2 |= sbp->sb_bad_features2;
650 mp->m_update_sb = true;
653 * Re-check for ATTR2 in case it was found in bad_features2
656 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
657 !(mp->m_flags & XFS_MOUNT_NOATTR2))
658 mp->m_flags |= XFS_MOUNT_ATTR2;
661 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
662 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
663 xfs_sb_version_removeattr2(&mp->m_sb);
664 mp->m_update_sb = true;
666 /* update sb_versionnum for the clearing of the morebits */
667 if (!sbp->sb_features2)
668 mp->m_update_sb = true;
671 /* always use v2 inodes by default now */
672 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
673 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
674 mp->m_update_sb = true;
678 * Check if sb_agblocks is aligned at stripe boundary
679 * If sb_agblocks is NOT aligned turn off m_dalign since
680 * allocator alignment is within an ag, therefore ag has
681 * to be aligned at stripe boundary.
683 error = xfs_update_alignment(mp);
687 xfs_alloc_compute_maxlevels(mp);
688 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
689 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
690 xfs_ialloc_compute_maxlevels(mp);
691 xfs_rmapbt_compute_maxlevels(mp);
692 xfs_refcountbt_compute_maxlevels(mp);
694 xfs_set_maxicount(mp);
696 /* enable fail_at_unmount as default */
697 mp->m_fail_unmount = 1;
699 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
703 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
704 &mp->m_kobj, "stats");
706 goto out_remove_sysfs;
708 error = xfs_error_sysfs_init(mp);
713 error = xfs_uuid_mount(mp);
715 goto out_remove_error_sysfs;
718 * Set the minimum read and write sizes
720 xfs_set_rw_sizes(mp);
722 /* set the low space thresholds for dynamic preallocation */
723 xfs_set_low_space_thresholds(mp);
726 * Set the inode cluster size.
727 * This may still be overridden by the file system
728 * block size if it is larger than the chosen cluster size.
730 * For v5 filesystems, scale the cluster size with the inode size to
731 * keep a constant ratio of inode per cluster buffer, but only if mkfs
732 * has set the inode alignment value appropriately for larger cluster
735 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
736 if (xfs_sb_version_hascrc(&mp->m_sb)) {
737 int new_size = mp->m_inode_cluster_size;
739 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
740 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
741 mp->m_inode_cluster_size = new_size;
745 * If enabled, sparse inode chunk alignment is expected to match the
746 * cluster size. Full inode chunk alignment must match the chunk size,
747 * but that is checked on sb read verification...
749 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
750 mp->m_sb.sb_spino_align !=
751 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
753 "Sparse inode block alignment (%u) must match cluster size (%llu).",
754 mp->m_sb.sb_spino_align,
755 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
757 goto out_remove_uuid;
761 * Set inode alignment fields
763 xfs_set_inoalignment(mp);
766 * Check that the data (and log if separate) is an ok size.
768 error = xfs_check_sizes(mp);
770 goto out_remove_uuid;
773 * Initialize realtime fields in the mount structure
775 error = xfs_rtmount_init(mp);
777 xfs_warn(mp, "RT mount failed");
778 goto out_remove_uuid;
782 * Copies the low order bits of the timestamp and the randomly
783 * set "sequence" number out of a UUID.
785 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
787 mp->m_dmevmask = 0; /* not persistent; set after each mount */
789 error = xfs_da_mount(mp);
791 xfs_warn(mp, "Failed dir/attr init: %d", error);
792 goto out_remove_uuid;
796 * Initialize the precomputed transaction reservations values.
801 * Allocate and initialize the per-ag data.
803 spin_lock_init(&mp->m_perag_lock);
804 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
805 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
807 xfs_warn(mp, "Failed per-ag init: %d", error);
811 if (!sbp->sb_logblocks) {
812 xfs_warn(mp, "no log defined");
813 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
814 error = -EFSCORRUPTED;
819 * Log's mount-time initialization. The first part of recovery can place
820 * some items on the AIL, to be handled when recovery is finished or
823 error = xfs_log_mount(mp, mp->m_logdev_targp,
824 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
825 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
827 xfs_warn(mp, "log mount failed");
832 * Now the log is mounted, we know if it was an unclean shutdown or
833 * not. If it was, with the first phase of recovery has completed, we
834 * have consistent AG blocks on disk. We have not recovered EFIs yet,
835 * but they are recovered transactionally in the second recovery phase
838 * Hence we can safely re-initialise incore superblock counters from
839 * the per-ag data. These may not be correct if the filesystem was not
840 * cleanly unmounted, so we need to wait for recovery to finish before
843 * If the filesystem was cleanly unmounted, then we can trust the
844 * values in the superblock to be correct and we don't need to do
847 * If we are currently making the filesystem, the initialisation will
848 * fail as the perag data is in an undefined state.
850 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
851 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
852 !mp->m_sb.sb_inprogress) {
853 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
855 goto out_log_dealloc;
859 * Get and sanity-check the root inode.
860 * Save the pointer to it in the mount structure.
862 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
864 xfs_warn(mp, "failed to read root inode");
865 goto out_log_dealloc;
870 if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
871 xfs_warn(mp, "corrupted root inode %llu: not a directory",
872 (unsigned long long)rip->i_ino);
873 xfs_iunlock(rip, XFS_ILOCK_EXCL);
874 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
876 error = -EFSCORRUPTED;
879 mp->m_rootip = rip; /* save it */
881 xfs_iunlock(rip, XFS_ILOCK_EXCL);
884 * Initialize realtime inode pointers in the mount structure
886 error = xfs_rtmount_inodes(mp);
889 * Free up the root inode.
891 xfs_warn(mp, "failed to read RT inodes");
896 * If this is a read-only mount defer the superblock updates until
897 * the next remount into writeable mode. Otherwise we would never
898 * perform the update e.g. for the root filesystem.
900 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
901 error = xfs_sync_sb(mp, false);
903 xfs_warn(mp, "failed to write sb changes");
909 * Initialise the XFS quota management subsystem for this mount
911 if (XFS_IS_QUOTA_RUNNING(mp)) {
912 error = xfs_qm_newmount(mp, "amount, "aflags);
916 ASSERT(!XFS_IS_QUOTA_ON(mp));
919 * If a file system had quotas running earlier, but decided to
920 * mount without -o uquota/pquota/gquota options, revoke the
921 * quotachecked license.
923 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
924 xfs_notice(mp, "resetting quota flags");
925 error = xfs_mount_reset_sbqflags(mp);
932 * During the second phase of log recovery, we need iget and
933 * iput to behave like they do for an active filesystem.
934 * xfs_fs_drop_inode needs to be able to prevent the deletion
935 * of inodes before we're done replaying log items on those
938 mp->m_super->s_flags |= MS_ACTIVE;
941 * Finish recovering the file system. This part needed to be delayed
942 * until after the root and real-time bitmap inodes were consistently
945 error = xfs_log_mount_finish(mp);
947 xfs_warn(mp, "log mount finish failed");
952 * Now the log is fully replayed, we can transition to full read-only
953 * mode for read-only mounts. This will sync all the metadata and clean
954 * the log so that the recovery we just performed does not have to be
955 * replayed again on the next mount.
957 * We use the same quiesce mechanism as the rw->ro remount, as they are
958 * semantically identical operations.
960 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
962 xfs_quiesce_attr(mp);
966 * Complete the quota initialisation, post-log-replay component.
969 ASSERT(mp->m_qflags == 0);
970 mp->m_qflags = quotaflags;
972 xfs_qm_mount_quotas(mp);
976 * Now we are mounted, reserve a small amount of unused space for
977 * privileged transactions. This is needed so that transaction
978 * space required for critical operations can dip into this pool
979 * when at ENOSPC. This is needed for operations like create with
980 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
981 * are not allowed to use this reserved space.
983 * This may drive us straight to ENOSPC on mount, but that implies
984 * we were already there on the last unmount. Warn if this occurs.
986 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
987 resblks = xfs_default_resblks(mp);
988 error = xfs_reserve_blocks(mp, &resblks, NULL);
991 "Unable to allocate reserve blocks. Continuing without reserve pool.");
993 /* Recover any CoW blocks that never got remapped. */
994 error = xfs_reflink_recover_cow(mp);
997 "Error %d recovering leftover CoW allocations.", error);
998 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1002 /* Reserve AG blocks for future btree expansion. */
1003 error = xfs_fs_reserve_ag_blocks(mp);
1004 if (error && error != -ENOSPC)
1011 xfs_fs_unreserve_ag_blocks(mp);
1013 xfs_qm_unmount_quotas(mp);
1015 mp->m_super->s_flags &= ~MS_ACTIVE;
1016 xfs_rtunmount_inodes(mp);
1019 cancel_delayed_work_sync(&mp->m_reclaim_work);
1020 xfs_reclaim_inodes(mp, SYNC_WAIT);
1022 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1023 xfs_log_mount_cancel(mp);
1025 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1026 xfs_wait_buftarg(mp->m_logdev_targp);
1027 xfs_wait_buftarg(mp->m_ddev_targp);
1033 xfs_uuid_unmount(mp);
1034 out_remove_error_sysfs:
1035 xfs_error_sysfs_del(mp);
1037 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1039 xfs_sysfs_del(&mp->m_kobj);
1045 * This flushes out the inodes,dquots and the superblock, unmounts the
1046 * log and makes sure that incore structures are freed.
1050 struct xfs_mount *mp)
1055 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1056 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1058 xfs_fs_unreserve_ag_blocks(mp);
1059 xfs_qm_unmount_quotas(mp);
1060 xfs_rtunmount_inodes(mp);
1061 IRELE(mp->m_rootip);
1064 * We can potentially deadlock here if we have an inode cluster
1065 * that has been freed has its buffer still pinned in memory because
1066 * the transaction is still sitting in a iclog. The stale inodes
1067 * on that buffer will have their flush locks held until the
1068 * transaction hits the disk and the callbacks run. the inode
1069 * flush takes the flush lock unconditionally and with nothing to
1070 * push out the iclog we will never get that unlocked. hence we
1071 * need to force the log first.
1073 xfs_log_force(mp, XFS_LOG_SYNC);
1076 * We now need to tell the world we are unmounting. This will allow
1077 * us to detect that the filesystem is going away and we should error
1078 * out anything that we have been retrying in the background. This will
1079 * prevent neverending retries in AIL pushing from hanging the unmount.
1081 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1084 * Flush all pending changes from the AIL.
1086 xfs_ail_push_all_sync(mp->m_ail);
1089 * And reclaim all inodes. At this point there should be no dirty
1090 * inodes and none should be pinned or locked, but use synchronous
1091 * reclaim just to be sure. We can stop background inode reclaim
1092 * here as well if it is still running.
1094 cancel_delayed_work_sync(&mp->m_reclaim_work);
1095 xfs_reclaim_inodes(mp, SYNC_WAIT);
1100 * Unreserve any blocks we have so that when we unmount we don't account
1101 * the reserved free space as used. This is really only necessary for
1102 * lazy superblock counting because it trusts the incore superblock
1103 * counters to be absolutely correct on clean unmount.
1105 * We don't bother correcting this elsewhere for lazy superblock
1106 * counting because on mount of an unclean filesystem we reconstruct the
1107 * correct counter value and this is irrelevant.
1109 * For non-lazy counter filesystems, this doesn't matter at all because
1110 * we only every apply deltas to the superblock and hence the incore
1111 * value does not matter....
1114 error = xfs_reserve_blocks(mp, &resblks, NULL);
1116 xfs_warn(mp, "Unable to free reserved block pool. "
1117 "Freespace may not be correct on next mount.");
1119 error = xfs_log_sbcount(mp);
1121 xfs_warn(mp, "Unable to update superblock counters. "
1122 "Freespace may not be correct on next mount.");
1125 xfs_log_unmount(mp);
1127 xfs_uuid_unmount(mp);
1130 xfs_errortag_clearall(mp, 0);
1134 xfs_error_sysfs_del(mp);
1135 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1136 xfs_sysfs_del(&mp->m_kobj);
1140 * Determine whether modifications can proceed. The caller specifies the minimum
1141 * freeze level for which modifications should not be allowed. This allows
1142 * certain operations to proceed while the freeze sequence is in progress, if
1147 struct xfs_mount *mp,
1150 ASSERT(level > SB_UNFROZEN);
1151 if ((mp->m_super->s_writers.frozen >= level) ||
1152 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1161 * Sync the superblock counters to disk.
1163 * Note this code can be called during the process of freezing, so we use the
1164 * transaction allocator that does not block when the transaction subsystem is
1165 * in its frozen state.
1168 xfs_log_sbcount(xfs_mount_t *mp)
1170 /* allow this to proceed during the freeze sequence... */
1171 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1175 * we don't need to do this if we are updating the superblock
1176 * counters on every modification.
1178 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1181 return xfs_sync_sb(mp, true);
1185 * Deltas for the inode count are +/-64, hence we use a large batch size
1186 * of 128 so we don't need to take the counter lock on every update.
1188 #define XFS_ICOUNT_BATCH 128
1191 struct xfs_mount *mp,
1194 __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1195 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1197 percpu_counter_add(&mp->m_icount, -delta);
1205 struct xfs_mount *mp,
1208 percpu_counter_add(&mp->m_ifree, delta);
1209 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1211 percpu_counter_add(&mp->m_ifree, -delta);
1218 * Deltas for the block count can vary from 1 to very large, but lock contention
1219 * only occurs on frequent small block count updates such as in the delayed
1220 * allocation path for buffered writes (page a time updates). Hence we set
1221 * a large batch count (1024) to minimise global counter updates except when
1222 * we get near to ENOSPC and we have to be very accurate with our updates.
1224 #define XFS_FDBLOCKS_BATCH 1024
1227 struct xfs_mount *mp,
1237 * If the reserve pool is depleted, put blocks back into it
1238 * first. Most of the time the pool is full.
1240 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1241 percpu_counter_add(&mp->m_fdblocks, delta);
1245 spin_lock(&mp->m_sb_lock);
1246 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1248 if (res_used > delta) {
1249 mp->m_resblks_avail += delta;
1252 mp->m_resblks_avail = mp->m_resblks;
1253 percpu_counter_add(&mp->m_fdblocks, delta);
1255 spin_unlock(&mp->m_sb_lock);
1260 * Taking blocks away, need to be more accurate the closer we
1263 * If the counter has a value of less than 2 * max batch size,
1264 * then make everything serialise as we are real close to
1267 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1268 XFS_FDBLOCKS_BATCH) < 0)
1271 batch = XFS_FDBLOCKS_BATCH;
1273 __percpu_counter_add(&mp->m_fdblocks, delta, batch);
1274 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1275 XFS_FDBLOCKS_BATCH) >= 0) {
1281 * lock up the sb for dipping into reserves before releasing the space
1282 * that took us to ENOSPC.
1284 spin_lock(&mp->m_sb_lock);
1285 percpu_counter_add(&mp->m_fdblocks, -delta);
1287 goto fdblocks_enospc;
1289 lcounter = (long long)mp->m_resblks_avail + delta;
1290 if (lcounter >= 0) {
1291 mp->m_resblks_avail = lcounter;
1292 spin_unlock(&mp->m_sb_lock);
1295 printk_once(KERN_WARNING
1296 "Filesystem \"%s\": reserve blocks depleted! "
1297 "Consider increasing reserve pool size.",
1300 spin_unlock(&mp->m_sb_lock);
1306 struct xfs_mount *mp,
1312 spin_lock(&mp->m_sb_lock);
1313 lcounter = mp->m_sb.sb_frextents + delta;
1317 mp->m_sb.sb_frextents = lcounter;
1318 spin_unlock(&mp->m_sb_lock);
1323 * xfs_getsb() is called to obtain the buffer for the superblock.
1324 * The buffer is returned locked and read in from disk.
1325 * The buffer should be released with a call to xfs_brelse().
1327 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1328 * the superblock buffer if it can be locked without sleeping.
1329 * If it can't then we'll return NULL.
1333 struct xfs_mount *mp,
1336 struct xfs_buf *bp = mp->m_sb_bp;
1338 if (!xfs_buf_trylock(bp)) {
1339 if (flags & XBF_TRYLOCK)
1345 ASSERT(bp->b_flags & XBF_DONE);
1350 * Used to free the superblock along various error paths.
1354 struct xfs_mount *mp)
1356 struct xfs_buf *bp = mp->m_sb_bp;
1364 * If the underlying (data/log/rt) device is readonly, there are some
1365 * operations that cannot proceed.
1368 xfs_dev_is_read_only(
1369 struct xfs_mount *mp,
1372 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1373 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1374 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1375 xfs_notice(mp, "%s required on read-only device.", message);
1376 xfs_notice(mp, "write access unavailable, cannot proceed.");