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 xfs_uuid_table_free(void)
53 if (xfs_uuid_table_size
== 0)
55 kmem_free(xfs_uuid_table
);
56 xfs_uuid_table
= NULL
;
57 xfs_uuid_table_size
= 0;
61 * See if the UUID is unique among mounted XFS filesystems.
62 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
68 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
71 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
74 if (uuid_is_nil(uuid
)) {
75 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
79 mutex_lock(&xfs_uuid_table_mutex
);
80 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
81 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
85 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
90 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
91 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
93 hole
= xfs_uuid_table_size
++;
95 xfs_uuid_table
[hole
] = *uuid
;
96 mutex_unlock(&xfs_uuid_table_mutex
);
101 mutex_unlock(&xfs_uuid_table_mutex
);
102 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
108 struct xfs_mount
*mp
)
110 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
113 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
116 mutex_lock(&xfs_uuid_table_mutex
);
117 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
118 if (uuid_is_nil(&xfs_uuid_table
[i
]))
120 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
122 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
125 ASSERT(i
< xfs_uuid_table_size
);
126 mutex_unlock(&xfs_uuid_table_mutex
);
132 struct rcu_head
*head
)
134 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
136 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
141 * Free up the per-ag resources associated with the mount structure.
148 struct xfs_perag
*pag
;
150 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
151 spin_lock(&mp
->m_perag_lock
);
152 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
153 spin_unlock(&mp
->m_perag_lock
);
155 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
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
= 0;
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
);
200 if (!first_initialised
)
201 first_initialised
= index
;
203 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
206 pag
->pag_agno
= index
;
208 spin_lock_init(&pag
->pag_ici_lock
);
209 mutex_init(&pag
->pag_ici_reclaim_lock
);
210 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
211 spin_lock_init(&pag
->pag_buf_lock
);
212 pag
->pag_buf_tree
= RB_ROOT
;
214 if (radix_tree_preload(GFP_NOFS
))
217 spin_lock(&mp
->m_perag_lock
);
218 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
220 spin_unlock(&mp
->m_perag_lock
);
221 radix_tree_preload_end();
225 spin_unlock(&mp
->m_perag_lock
);
226 radix_tree_preload_end();
229 index
= xfs_set_inode_alloc(mp
, agcount
);
237 for (; index
> first_initialised
; index
--) {
238 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
247 * Does the initial read of the superblock.
251 struct xfs_mount
*mp
,
254 unsigned int sector_size
;
256 struct xfs_sb
*sbp
= &mp
->m_sb
;
258 int loud
= !(flags
& XFS_MFSI_QUIET
);
259 const struct xfs_buf_ops
*buf_ops
;
261 ASSERT(mp
->m_sb_bp
== NULL
);
262 ASSERT(mp
->m_ddev_targp
!= NULL
);
265 * For the initial read, we must guess at the sector
266 * size based on the block device. It's enough to
267 * get the sb_sectsize out of the superblock and
268 * then reread with the proper length.
269 * We don't verify it yet, because it may not be complete.
271 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
275 * Allocate a (locked) buffer to hold the superblock. This will be kept
276 * around at all times to optimize access to the superblock. Therefore,
277 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
281 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
282 BTOBB(sector_size
), XBF_NO_IOACCT
, &bp
,
286 xfs_warn(mp
, "SB validate failed with error %d.", error
);
287 /* bad CRC means corrupted metadata */
288 if (error
== -EFSBADCRC
)
289 error
= -EFSCORRUPTED
;
294 * Initialize the mount structure from the superblock.
296 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
299 * If we haven't validated the superblock, do so now before we try
300 * to check the sector size and reread the superblock appropriately.
302 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
304 xfs_warn(mp
, "Invalid superblock magic number");
310 * We must be able to do sector-sized and sector-aligned IO.
312 if (sector_size
> sbp
->sb_sectsize
) {
314 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
315 sector_size
, sbp
->sb_sectsize
);
320 if (buf_ops
== NULL
) {
322 * Re-read the superblock so the buffer is correctly sized,
323 * and properly verified.
326 sector_size
= sbp
->sb_sectsize
;
327 buf_ops
= loud
? &xfs_sb_buf_ops
: &xfs_sb_quiet_buf_ops
;
331 xfs_reinit_percpu_counters(mp
);
333 /* no need to be quiet anymore, so reset the buf ops */
334 bp
->b_ops
= &xfs_sb_buf_ops
;
346 * Update alignment values based on mount options and sb values
349 xfs_update_alignment(xfs_mount_t
*mp
)
351 xfs_sb_t
*sbp
= &(mp
->m_sb
);
355 * If stripe unit and stripe width are not multiples
356 * of the fs blocksize turn off alignment.
358 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
359 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
361 "alignment check failed: sunit/swidth vs. blocksize(%d)",
366 * Convert the stripe unit and width to FSBs.
368 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
369 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
371 "alignment check failed: sunit/swidth vs. agsize(%d)",
374 } else if (mp
->m_dalign
) {
375 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
378 "alignment check failed: sunit(%d) less than bsize(%d)",
379 mp
->m_dalign
, sbp
->sb_blocksize
);
385 * Update superblock with new values
388 if (xfs_sb_version_hasdalign(sbp
)) {
389 if (sbp
->sb_unit
!= mp
->m_dalign
) {
390 sbp
->sb_unit
= mp
->m_dalign
;
391 mp
->m_update_sb
= true;
393 if (sbp
->sb_width
!= mp
->m_swidth
) {
394 sbp
->sb_width
= mp
->m_swidth
;
395 mp
->m_update_sb
= true;
399 "cannot change alignment: superblock does not support data alignment");
402 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
403 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
404 mp
->m_dalign
= sbp
->sb_unit
;
405 mp
->m_swidth
= sbp
->sb_width
;
412 * Set the maximum inode count for this filesystem
415 xfs_set_maxicount(xfs_mount_t
*mp
)
417 xfs_sb_t
*sbp
= &(mp
->m_sb
);
420 if (sbp
->sb_imax_pct
) {
422 * Make sure the maximum inode count is a multiple
423 * of the units we allocate inodes in.
425 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
427 do_div(icount
, mp
->m_ialloc_blks
);
428 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
436 * Set the default minimum read and write sizes unless
437 * already specified in a mount option.
438 * We use smaller I/O sizes when the file system
439 * is being used for NFS service (wsync mount option).
442 xfs_set_rw_sizes(xfs_mount_t
*mp
)
444 xfs_sb_t
*sbp
= &(mp
->m_sb
);
445 int readio_log
, writeio_log
;
447 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
448 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
449 readio_log
= XFS_WSYNC_READIO_LOG
;
450 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
452 readio_log
= XFS_READIO_LOG_LARGE
;
453 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
456 readio_log
= mp
->m_readio_log
;
457 writeio_log
= mp
->m_writeio_log
;
460 if (sbp
->sb_blocklog
> readio_log
) {
461 mp
->m_readio_log
= sbp
->sb_blocklog
;
463 mp
->m_readio_log
= readio_log
;
465 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
466 if (sbp
->sb_blocklog
> writeio_log
) {
467 mp
->m_writeio_log
= sbp
->sb_blocklog
;
469 mp
->m_writeio_log
= writeio_log
;
471 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
475 * precalculate the low space thresholds for dynamic speculative preallocation.
478 xfs_set_low_space_thresholds(
479 struct xfs_mount
*mp
)
483 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
484 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
487 mp
->m_low_space
[i
] = space
* (i
+ 1);
493 * Set whether we're using inode alignment.
496 xfs_set_inoalignment(xfs_mount_t
*mp
)
498 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
499 mp
->m_sb
.sb_inoalignmt
>=
500 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
501 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
503 mp
->m_inoalign_mask
= 0;
505 * If we are using stripe alignment, check whether
506 * the stripe unit is a multiple of the inode alignment
508 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
509 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
510 mp
->m_sinoalign
= mp
->m_dalign
;
516 * Check that the data (and log if separate) is an ok size.
520 struct xfs_mount
*mp
)
526 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
527 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
528 xfs_warn(mp
, "filesystem size mismatch detected");
531 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
532 d
- XFS_FSS_TO_BB(mp
, 1),
533 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, NULL
);
535 xfs_warn(mp
, "last sector read failed");
540 if (mp
->m_logdev_targp
== mp
->m_ddev_targp
)
543 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
544 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
545 xfs_warn(mp
, "log size mismatch detected");
548 error
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
549 d
- XFS_FSB_TO_BB(mp
, 1),
550 XFS_FSB_TO_BB(mp
, 1), 0, &bp
, NULL
);
552 xfs_warn(mp
, "log device read failed");
560 * Clear the quotaflags in memory and in the superblock.
563 xfs_mount_reset_sbqflags(
564 struct xfs_mount
*mp
)
568 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
569 if (mp
->m_sb
.sb_qflags
== 0)
571 spin_lock(&mp
->m_sb_lock
);
572 mp
->m_sb
.sb_qflags
= 0;
573 spin_unlock(&mp
->m_sb_lock
);
575 if (!xfs_fs_writable(mp
, SB_FREEZE_WRITE
))
578 return xfs_sync_sb(mp
, false);
582 xfs_default_resblks(xfs_mount_t
*mp
)
587 * We default to 5% or 8192 fsbs of space reserved, whichever is
588 * smaller. This is intended to cover concurrent allocation
589 * transactions when we initially hit enospc. These each require a 4
590 * block reservation. Hence by default we cover roughly 2000 concurrent
591 * allocation reservations.
593 resblks
= mp
->m_sb
.sb_dblocks
;
595 resblks
= min_t(__uint64_t
, resblks
, 8192);
600 * This function does the following on an initial mount of a file system:
601 * - reads the superblock from disk and init the mount struct
602 * - if we're a 32-bit kernel, do a size check on the superblock
603 * so we don't mount terabyte filesystems
604 * - init mount struct realtime fields
605 * - allocate inode hash table for fs
606 * - init directory manager
607 * - perform recovery and init the log manager
611 struct xfs_mount
*mp
)
613 struct xfs_sb
*sbp
= &(mp
->m_sb
);
614 struct xfs_inode
*rip
;
620 xfs_sb_mount_common(mp
, sbp
);
623 * Check for a mismatched features2 values. Older kernels read & wrote
624 * into the wrong sb offset for sb_features2 on some platforms due to
625 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
626 * which made older superblock reading/writing routines swap it as a
629 * For backwards compatibility, we make both slots equal.
631 * If we detect a mismatched field, we OR the set bits into the existing
632 * features2 field in case it has already been modified; we don't want
633 * to lose any features. We then update the bad location with the ORed
634 * value so that older kernels will see any features2 flags. The
635 * superblock writeback code ensures the new sb_features2 is copied to
636 * sb_bad_features2 before it is logged or written to disk.
638 if (xfs_sb_has_mismatched_features2(sbp
)) {
639 xfs_warn(mp
, "correcting sb_features alignment problem");
640 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
641 mp
->m_update_sb
= true;
644 * Re-check for ATTR2 in case it was found in bad_features2
647 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
648 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
649 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
652 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
653 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
654 xfs_sb_version_removeattr2(&mp
->m_sb
);
655 mp
->m_update_sb
= true;
657 /* update sb_versionnum for the clearing of the morebits */
658 if (!sbp
->sb_features2
)
659 mp
->m_update_sb
= true;
662 /* always use v2 inodes by default now */
663 if (!(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_NLINKBIT
)) {
664 mp
->m_sb
.sb_versionnum
|= XFS_SB_VERSION_NLINKBIT
;
665 mp
->m_update_sb
= true;
669 * Check if sb_agblocks is aligned at stripe boundary
670 * If sb_agblocks is NOT aligned turn off m_dalign since
671 * allocator alignment is within an ag, therefore ag has
672 * to be aligned at stripe boundary.
674 error
= xfs_update_alignment(mp
);
678 xfs_alloc_compute_maxlevels(mp
);
679 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
680 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
681 xfs_ialloc_compute_maxlevels(mp
);
683 xfs_set_maxicount(mp
);
685 /* enable fail_at_unmount as default */
686 mp
->m_fail_unmount
= 1;
688 error
= xfs_sysfs_init(&mp
->m_kobj
, &xfs_mp_ktype
, NULL
, mp
->m_fsname
);
692 error
= xfs_sysfs_init(&mp
->m_stats
.xs_kobj
, &xfs_stats_ktype
,
693 &mp
->m_kobj
, "stats");
695 goto out_remove_sysfs
;
697 error
= xfs_error_sysfs_init(mp
);
702 error
= xfs_uuid_mount(mp
);
704 goto out_remove_error_sysfs
;
707 * Set the minimum read and write sizes
709 xfs_set_rw_sizes(mp
);
711 /* set the low space thresholds for dynamic preallocation */
712 xfs_set_low_space_thresholds(mp
);
715 * Set the inode cluster size.
716 * This may still be overridden by the file system
717 * block size if it is larger than the chosen cluster size.
719 * For v5 filesystems, scale the cluster size with the inode size to
720 * keep a constant ratio of inode per cluster buffer, but only if mkfs
721 * has set the inode alignment value appropriately for larger cluster
724 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
725 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
726 int new_size
= mp
->m_inode_cluster_size
;
728 new_size
*= mp
->m_sb
.sb_inodesize
/ XFS_DINODE_MIN_SIZE
;
729 if (mp
->m_sb
.sb_inoalignmt
>= XFS_B_TO_FSBT(mp
, new_size
))
730 mp
->m_inode_cluster_size
= new_size
;
734 * If enabled, sparse inode chunk alignment is expected to match the
735 * cluster size. Full inode chunk alignment must match the chunk size,
736 * but that is checked on sb read verification...
738 if (xfs_sb_version_hassparseinodes(&mp
->m_sb
) &&
739 mp
->m_sb
.sb_spino_align
!=
740 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
)) {
742 "Sparse inode block alignment (%u) must match cluster size (%llu).",
743 mp
->m_sb
.sb_spino_align
,
744 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
));
746 goto out_remove_uuid
;
750 * Set inode alignment fields
752 xfs_set_inoalignment(mp
);
755 * Check that the data (and log if separate) is an ok size.
757 error
= xfs_check_sizes(mp
);
759 goto out_remove_uuid
;
762 * Initialize realtime fields in the mount structure
764 error
= xfs_rtmount_init(mp
);
766 xfs_warn(mp
, "RT mount failed");
767 goto out_remove_uuid
;
771 * Copies the low order bits of the timestamp and the randomly
772 * set "sequence" number out of a UUID.
774 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
776 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
778 error
= xfs_da_mount(mp
);
780 xfs_warn(mp
, "Failed dir/attr init: %d", error
);
781 goto out_remove_uuid
;
785 * Initialize the precomputed transaction reservations values.
790 * Allocate and initialize the per-ag data.
792 spin_lock_init(&mp
->m_perag_lock
);
793 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
794 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
796 xfs_warn(mp
, "Failed per-ag init: %d", error
);
800 if (!sbp
->sb_logblocks
) {
801 xfs_warn(mp
, "no log defined");
802 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
803 error
= -EFSCORRUPTED
;
808 * Log's mount-time initialization. The first part of recovery can place
809 * some items on the AIL, to be handled when recovery is finished or
812 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
813 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
814 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
816 xfs_warn(mp
, "log mount failed");
821 * Now the log is mounted, we know if it was an unclean shutdown or
822 * not. If it was, with the first phase of recovery has completed, we
823 * have consistent AG blocks on disk. We have not recovered EFIs yet,
824 * but they are recovered transactionally in the second recovery phase
827 * Hence we can safely re-initialise incore superblock counters from
828 * the per-ag data. These may not be correct if the filesystem was not
829 * cleanly unmounted, so we need to wait for recovery to finish before
832 * If the filesystem was cleanly unmounted, then we can trust the
833 * values in the superblock to be correct and we don't need to do
836 * If we are currently making the filesystem, the initialisation will
837 * fail as the perag data is in an undefined state.
839 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
840 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
841 !mp
->m_sb
.sb_inprogress
) {
842 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
844 goto out_log_dealloc
;
848 * Get and sanity-check the root inode.
849 * Save the pointer to it in the mount structure.
851 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
853 xfs_warn(mp
, "failed to read root inode");
854 goto out_log_dealloc
;
859 if (unlikely(!S_ISDIR(VFS_I(rip
)->i_mode
))) {
860 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
861 (unsigned long long)rip
->i_ino
);
862 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
863 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
865 error
= -EFSCORRUPTED
;
868 mp
->m_rootip
= rip
; /* save it */
870 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
873 * Initialize realtime inode pointers in the mount structure
875 error
= xfs_rtmount_inodes(mp
);
878 * Free up the root inode.
880 xfs_warn(mp
, "failed to read RT inodes");
885 * If this is a read-only mount defer the superblock updates until
886 * the next remount into writeable mode. Otherwise we would never
887 * perform the update e.g. for the root filesystem.
889 if (mp
->m_update_sb
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
890 error
= xfs_sync_sb(mp
, false);
892 xfs_warn(mp
, "failed to write sb changes");
898 * Initialise the XFS quota management subsystem for this mount
900 if (XFS_IS_QUOTA_RUNNING(mp
)) {
901 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
905 ASSERT(!XFS_IS_QUOTA_ON(mp
));
908 * If a file system had quotas running earlier, but decided to
909 * mount without -o uquota/pquota/gquota options, revoke the
910 * quotachecked license.
912 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
913 xfs_notice(mp
, "resetting quota flags");
914 error
= xfs_mount_reset_sbqflags(mp
);
921 * Finish recovering the file system. This part needed to be delayed
922 * until after the root and real-time bitmap inodes were consistently
925 error
= xfs_log_mount_finish(mp
);
927 xfs_warn(mp
, "log mount finish failed");
932 * Complete the quota initialisation, post-log-replay component.
935 ASSERT(mp
->m_qflags
== 0);
936 mp
->m_qflags
= quotaflags
;
938 xfs_qm_mount_quotas(mp
);
942 * Now we are mounted, reserve a small amount of unused space for
943 * privileged transactions. This is needed so that transaction
944 * space required for critical operations can dip into this pool
945 * when at ENOSPC. This is needed for operations like create with
946 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
947 * are not allowed to use this reserved space.
949 * This may drive us straight to ENOSPC on mount, but that implies
950 * we were already there on the last unmount. Warn if this occurs.
952 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
953 resblks
= xfs_default_resblks(mp
);
954 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
957 "Unable to allocate reserve blocks. Continuing without reserve pool.");
963 xfs_rtunmount_inodes(mp
);
966 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
967 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
969 mp
->m_flags
|= XFS_MOUNT_UNMOUNTING
;
970 xfs_log_mount_cancel(mp
);
972 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
973 xfs_wait_buftarg(mp
->m_logdev_targp
);
974 xfs_wait_buftarg(mp
->m_ddev_targp
);
980 xfs_uuid_unmount(mp
);
981 out_remove_error_sysfs
:
982 xfs_error_sysfs_del(mp
);
984 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
986 xfs_sysfs_del(&mp
->m_kobj
);
992 * This flushes out the inodes,dquots and the superblock, unmounts the
993 * log and makes sure that incore structures are freed.
997 struct xfs_mount
*mp
)
1002 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
1004 xfs_qm_unmount_quotas(mp
);
1005 xfs_rtunmount_inodes(mp
);
1006 IRELE(mp
->m_rootip
);
1009 * We can potentially deadlock here if we have an inode cluster
1010 * that has been freed has its buffer still pinned in memory because
1011 * the transaction is still sitting in a iclog. The stale inodes
1012 * on that buffer will have their flush locks held until the
1013 * transaction hits the disk and the callbacks run. the inode
1014 * flush takes the flush lock unconditionally and with nothing to
1015 * push out the iclog we will never get that unlocked. hence we
1016 * need to force the log first.
1018 xfs_log_force(mp
, XFS_LOG_SYNC
);
1021 * We now need to tell the world we are unmounting. This will allow
1022 * us to detect that the filesystem is going away and we should error
1023 * out anything that we have been retrying in the background. This will
1024 * prevent neverending retries in AIL pushing from hanging the unmount.
1026 mp
->m_flags
|= XFS_MOUNT_UNMOUNTING
;
1029 * Flush all pending changes from the AIL.
1031 xfs_ail_push_all_sync(mp
->m_ail
);
1034 * And reclaim all inodes. At this point there should be no dirty
1035 * inodes and none should be pinned or locked, but use synchronous
1036 * reclaim just to be sure. We can stop background inode reclaim
1037 * here as well if it is still running.
1039 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1040 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1045 * Unreserve any blocks we have so that when we unmount we don't account
1046 * the reserved free space as used. This is really only necessary for
1047 * lazy superblock counting because it trusts the incore superblock
1048 * counters to be absolutely correct on clean unmount.
1050 * We don't bother correcting this elsewhere for lazy superblock
1051 * counting because on mount of an unclean filesystem we reconstruct the
1052 * correct counter value and this is irrelevant.
1054 * For non-lazy counter filesystems, this doesn't matter at all because
1055 * we only every apply deltas to the superblock and hence the incore
1056 * value does not matter....
1059 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1061 xfs_warn(mp
, "Unable to free reserved block pool. "
1062 "Freespace may not be correct on next mount.");
1064 error
= xfs_log_sbcount(mp
);
1066 xfs_warn(mp
, "Unable to update superblock counters. "
1067 "Freespace may not be correct on next mount.");
1070 xfs_log_unmount(mp
);
1072 xfs_uuid_unmount(mp
);
1075 xfs_errortag_clearall(mp
, 0);
1079 xfs_error_sysfs_del(mp
);
1080 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1081 xfs_sysfs_del(&mp
->m_kobj
);
1085 * Determine whether modifications can proceed. The caller specifies the minimum
1086 * freeze level for which modifications should not be allowed. This allows
1087 * certain operations to proceed while the freeze sequence is in progress, if
1092 struct xfs_mount
*mp
,
1095 ASSERT(level
> SB_UNFROZEN
);
1096 if ((mp
->m_super
->s_writers
.frozen
>= level
) ||
1097 XFS_FORCED_SHUTDOWN(mp
) || (mp
->m_flags
& XFS_MOUNT_RDONLY
))
1106 * Sync the superblock counters to disk.
1108 * Note this code can be called during the process of freezing, so we use the
1109 * transaction allocator that does not block when the transaction subsystem is
1110 * in its frozen state.
1113 xfs_log_sbcount(xfs_mount_t
*mp
)
1115 /* allow this to proceed during the freeze sequence... */
1116 if (!xfs_fs_writable(mp
, SB_FREEZE_COMPLETE
))
1120 * we don't need to do this if we are updating the superblock
1121 * counters on every modification.
1123 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1126 return xfs_sync_sb(mp
, true);
1130 * Deltas for the inode count are +/-64, hence we use a large batch size
1131 * of 128 so we don't need to take the counter lock on every update.
1133 #define XFS_ICOUNT_BATCH 128
1136 struct xfs_mount
*mp
,
1139 __percpu_counter_add(&mp
->m_icount
, delta
, XFS_ICOUNT_BATCH
);
1140 if (__percpu_counter_compare(&mp
->m_icount
, 0, XFS_ICOUNT_BATCH
) < 0) {
1142 percpu_counter_add(&mp
->m_icount
, -delta
);
1150 struct xfs_mount
*mp
,
1153 percpu_counter_add(&mp
->m_ifree
, delta
);
1154 if (percpu_counter_compare(&mp
->m_ifree
, 0) < 0) {
1156 percpu_counter_add(&mp
->m_ifree
, -delta
);
1163 * Deltas for the block count can vary from 1 to very large, but lock contention
1164 * only occurs on frequent small block count updates such as in the delayed
1165 * allocation path for buffered writes (page a time updates). Hence we set
1166 * a large batch count (1024) to minimise global counter updates except when
1167 * we get near to ENOSPC and we have to be very accurate with our updates.
1169 #define XFS_FDBLOCKS_BATCH 1024
1172 struct xfs_mount
*mp
,
1182 * If the reserve pool is depleted, put blocks back into it
1183 * first. Most of the time the pool is full.
1185 if (likely(mp
->m_resblks
== mp
->m_resblks_avail
)) {
1186 percpu_counter_add(&mp
->m_fdblocks
, delta
);
1190 spin_lock(&mp
->m_sb_lock
);
1191 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1193 if (res_used
> delta
) {
1194 mp
->m_resblks_avail
+= delta
;
1197 mp
->m_resblks_avail
= mp
->m_resblks
;
1198 percpu_counter_add(&mp
->m_fdblocks
, delta
);
1200 spin_unlock(&mp
->m_sb_lock
);
1205 * Taking blocks away, need to be more accurate the closer we
1208 * If the counter has a value of less than 2 * max batch size,
1209 * then make everything serialise as we are real close to
1212 if (__percpu_counter_compare(&mp
->m_fdblocks
, 2 * XFS_FDBLOCKS_BATCH
,
1213 XFS_FDBLOCKS_BATCH
) < 0)
1216 batch
= XFS_FDBLOCKS_BATCH
;
1218 __percpu_counter_add(&mp
->m_fdblocks
, delta
, batch
);
1219 if (__percpu_counter_compare(&mp
->m_fdblocks
, XFS_ALLOC_SET_ASIDE(mp
),
1220 XFS_FDBLOCKS_BATCH
) >= 0) {
1226 * lock up the sb for dipping into reserves before releasing the space
1227 * that took us to ENOSPC.
1229 spin_lock(&mp
->m_sb_lock
);
1230 percpu_counter_add(&mp
->m_fdblocks
, -delta
);
1232 goto fdblocks_enospc
;
1234 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1235 if (lcounter
>= 0) {
1236 mp
->m_resblks_avail
= lcounter
;
1237 spin_unlock(&mp
->m_sb_lock
);
1240 printk_once(KERN_WARNING
1241 "Filesystem \"%s\": reserve blocks depleted! "
1242 "Consider increasing reserve pool size.",
1245 spin_unlock(&mp
->m_sb_lock
);
1251 struct xfs_mount
*mp
,
1257 spin_lock(&mp
->m_sb_lock
);
1258 lcounter
= mp
->m_sb
.sb_frextents
+ delta
;
1262 mp
->m_sb
.sb_frextents
= lcounter
;
1263 spin_unlock(&mp
->m_sb_lock
);
1268 * xfs_getsb() is called to obtain the buffer for the superblock.
1269 * The buffer is returned locked and read in from disk.
1270 * The buffer should be released with a call to xfs_brelse().
1272 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1273 * the superblock buffer if it can be locked without sleeping.
1274 * If it can't then we'll return NULL.
1278 struct xfs_mount
*mp
,
1281 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1283 if (!xfs_buf_trylock(bp
)) {
1284 if (flags
& XBF_TRYLOCK
)
1290 ASSERT(bp
->b_flags
& XBF_DONE
);
1295 * Used to free the superblock along various error paths.
1299 struct xfs_mount
*mp
)
1301 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1309 * If the underlying (data/log/rt) device is readonly, there are some
1310 * operations that cannot proceed.
1313 xfs_dev_is_read_only(
1314 struct xfs_mount
*mp
,
1317 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1318 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1319 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1320 xfs_notice(mp
, "%s required on read-only device.", message
);
1321 xfs_notice(mp
, "write access unavailable, cannot proceed.");