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 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
79 if (uuid_is_nil(uuid
)) {
80 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
84 mutex_lock(&xfs_uuid_table_mutex
);
85 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
86 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
90 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
95 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
96 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
98 hole
= xfs_uuid_table_size
++;
100 xfs_uuid_table
[hole
] = *uuid
;
101 mutex_unlock(&xfs_uuid_table_mutex
);
106 mutex_unlock(&xfs_uuid_table_mutex
);
107 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
113 struct xfs_mount
*mp
)
115 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
118 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
121 mutex_lock(&xfs_uuid_table_mutex
);
122 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
123 if (uuid_is_nil(&xfs_uuid_table
[i
]))
125 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
127 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
130 ASSERT(i
< xfs_uuid_table_size
);
131 mutex_unlock(&xfs_uuid_table_mutex
);
137 struct rcu_head
*head
)
139 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
141 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
146 * Free up the per-ag resources associated with the mount structure.
153 struct xfs_perag
*pag
;
155 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
156 spin_lock(&mp
->m_perag_lock
);
157 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
158 spin_unlock(&mp
->m_perag_lock
);
160 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
161 xfs_buf_hash_destroy(pag
);
162 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
167 * Check size of device based on the (data/realtime) block count.
168 * Note: this check is used by the growfs code as well as mount.
171 xfs_sb_validate_fsb_count(
175 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
176 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
178 /* Limited by ULONG_MAX of page cache index */
179 if (nblocks
>> (PAGE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
185 xfs_initialize_perag(
187 xfs_agnumber_t agcount
,
188 xfs_agnumber_t
*maxagi
)
190 xfs_agnumber_t index
;
191 xfs_agnumber_t first_initialised
= NULLAGNUMBER
;
196 * Walk the current per-ag tree so we don't try to initialise AGs
197 * that already exist (growfs case). Allocate and insert all the
198 * AGs we don't find ready for initialisation.
200 for (index
= 0; index
< agcount
; index
++) {
201 pag
= xfs_perag_get(mp
, index
);
207 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
209 goto out_unwind_new_pags
;
210 pag
->pag_agno
= index
;
212 spin_lock_init(&pag
->pag_ici_lock
);
213 mutex_init(&pag
->pag_ici_reclaim_lock
);
214 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
215 if (xfs_buf_hash_init(pag
))
217 init_waitqueue_head(&pag
->pagb_wait
);
219 if (radix_tree_preload(GFP_NOFS
))
220 goto out_hash_destroy
;
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();
228 goto out_hash_destroy
;
230 spin_unlock(&mp
->m_perag_lock
);
231 radix_tree_preload_end();
232 /* first new pag is fully initialized */
233 if (first_initialised
== NULLAGNUMBER
)
234 first_initialised
= index
;
237 index
= xfs_set_inode_alloc(mp
, agcount
);
242 mp
->m_ag_prealloc_blocks
= xfs_prealloc_blocks(mp
);
246 xfs_buf_hash_destroy(pag
);
250 /* unwind any prior newly initialized pags */
251 for (index
= first_initialised
; index
< agcount
; index
++) {
252 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
255 xfs_buf_hash_destroy(pag
);
264 * Does the initial read of the superblock.
268 struct xfs_mount
*mp
,
271 unsigned int sector_size
;
273 struct xfs_sb
*sbp
= &mp
->m_sb
;
275 int loud
= !(flags
& XFS_MFSI_QUIET
);
276 const struct xfs_buf_ops
*buf_ops
;
278 ASSERT(mp
->m_sb_bp
== NULL
);
279 ASSERT(mp
->m_ddev_targp
!= NULL
);
282 * For the initial read, we must guess at the sector
283 * size based on the block device. It's enough to
284 * get the sb_sectsize out of the superblock and
285 * then reread with the proper length.
286 * We don't verify it yet, because it may not be complete.
288 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
292 * Allocate a (locked) buffer to hold the superblock. This will be kept
293 * around at all times to optimize access to the superblock. Therefore,
294 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
298 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
299 BTOBB(sector_size
), XBF_NO_IOACCT
, &bp
,
303 xfs_warn(mp
, "SB validate failed with error %d.", error
);
304 /* bad CRC means corrupted metadata */
305 if (error
== -EFSBADCRC
)
306 error
= -EFSCORRUPTED
;
311 * Initialize the mount structure from the superblock.
313 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
316 * If we haven't validated the superblock, do so now before we try
317 * to check the sector size and reread the superblock appropriately.
319 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
321 xfs_warn(mp
, "Invalid superblock magic number");
327 * We must be able to do sector-sized and sector-aligned IO.
329 if (sector_size
> sbp
->sb_sectsize
) {
331 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
332 sector_size
, sbp
->sb_sectsize
);
337 if (buf_ops
== NULL
) {
339 * Re-read the superblock so the buffer is correctly sized,
340 * and properly verified.
343 sector_size
= sbp
->sb_sectsize
;
344 buf_ops
= loud
? &xfs_sb_buf_ops
: &xfs_sb_quiet_buf_ops
;
348 xfs_reinit_percpu_counters(mp
);
350 /* no need to be quiet anymore, so reset the buf ops */
351 bp
->b_ops
= &xfs_sb_buf_ops
;
363 * Update alignment values based on mount options and sb values
366 xfs_update_alignment(xfs_mount_t
*mp
)
368 xfs_sb_t
*sbp
= &(mp
->m_sb
);
372 * If stripe unit and stripe width are not multiples
373 * of the fs blocksize turn off alignment.
375 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
376 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
378 "alignment check failed: sunit/swidth vs. blocksize(%d)",
383 * Convert the stripe unit and width to FSBs.
385 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
386 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
388 "alignment check failed: sunit/swidth vs. agsize(%d)",
391 } else if (mp
->m_dalign
) {
392 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
395 "alignment check failed: sunit(%d) less than bsize(%d)",
396 mp
->m_dalign
, sbp
->sb_blocksize
);
402 * Update superblock with new values
405 if (xfs_sb_version_hasdalign(sbp
)) {
406 if (sbp
->sb_unit
!= mp
->m_dalign
) {
407 sbp
->sb_unit
= mp
->m_dalign
;
408 mp
->m_update_sb
= true;
410 if (sbp
->sb_width
!= mp
->m_swidth
) {
411 sbp
->sb_width
= mp
->m_swidth
;
412 mp
->m_update_sb
= true;
416 "cannot change alignment: superblock does not support data alignment");
419 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
420 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
421 mp
->m_dalign
= sbp
->sb_unit
;
422 mp
->m_swidth
= sbp
->sb_width
;
429 * Set the maximum inode count for this filesystem
432 xfs_set_maxicount(xfs_mount_t
*mp
)
434 xfs_sb_t
*sbp
= &(mp
->m_sb
);
437 if (sbp
->sb_imax_pct
) {
439 * Make sure the maximum inode count is a multiple
440 * of the units we allocate inodes in.
442 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
444 do_div(icount
, mp
->m_ialloc_blks
);
445 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
453 * Set the default minimum read and write sizes unless
454 * already specified in a mount option.
455 * We use smaller I/O sizes when the file system
456 * is being used for NFS service (wsync mount option).
459 xfs_set_rw_sizes(xfs_mount_t
*mp
)
461 xfs_sb_t
*sbp
= &(mp
->m_sb
);
462 int readio_log
, writeio_log
;
464 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
465 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
466 readio_log
= XFS_WSYNC_READIO_LOG
;
467 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
469 readio_log
= XFS_READIO_LOG_LARGE
;
470 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
473 readio_log
= mp
->m_readio_log
;
474 writeio_log
= mp
->m_writeio_log
;
477 if (sbp
->sb_blocklog
> readio_log
) {
478 mp
->m_readio_log
= sbp
->sb_blocklog
;
480 mp
->m_readio_log
= readio_log
;
482 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
483 if (sbp
->sb_blocklog
> writeio_log
) {
484 mp
->m_writeio_log
= sbp
->sb_blocklog
;
486 mp
->m_writeio_log
= writeio_log
;
488 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
492 * precalculate the low space thresholds for dynamic speculative preallocation.
495 xfs_set_low_space_thresholds(
496 struct xfs_mount
*mp
)
500 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
501 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
504 mp
->m_low_space
[i
] = space
* (i
+ 1);
510 * Set whether we're using inode alignment.
513 xfs_set_inoalignment(xfs_mount_t
*mp
)
515 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
516 mp
->m_sb
.sb_inoalignmt
>=
517 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
518 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
520 mp
->m_inoalign_mask
= 0;
522 * If we are using stripe alignment, check whether
523 * the stripe unit is a multiple of the inode alignment
525 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
526 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
527 mp
->m_sinoalign
= mp
->m_dalign
;
533 * Check that the data (and log if separate) is an ok size.
537 struct xfs_mount
*mp
)
543 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
544 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
545 xfs_warn(mp
, "filesystem size mismatch detected");
548 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
549 d
- XFS_FSS_TO_BB(mp
, 1),
550 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, NULL
);
552 xfs_warn(mp
, "last sector read failed");
557 if (mp
->m_logdev_targp
== mp
->m_ddev_targp
)
560 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
561 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
562 xfs_warn(mp
, "log size mismatch detected");
565 error
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
566 d
- XFS_FSB_TO_BB(mp
, 1),
567 XFS_FSB_TO_BB(mp
, 1), 0, &bp
, NULL
);
569 xfs_warn(mp
, "log device read failed");
577 * Clear the quotaflags in memory and in the superblock.
580 xfs_mount_reset_sbqflags(
581 struct xfs_mount
*mp
)
585 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
586 if (mp
->m_sb
.sb_qflags
== 0)
588 spin_lock(&mp
->m_sb_lock
);
589 mp
->m_sb
.sb_qflags
= 0;
590 spin_unlock(&mp
->m_sb_lock
);
592 if (!xfs_fs_writable(mp
, SB_FREEZE_WRITE
))
595 return xfs_sync_sb(mp
, false);
599 xfs_default_resblks(xfs_mount_t
*mp
)
604 * We default to 5% or 8192 fsbs of space reserved, whichever is
605 * smaller. This is intended to cover concurrent allocation
606 * transactions when we initially hit enospc. These each require a 4
607 * block reservation. Hence by default we cover roughly 2000 concurrent
608 * allocation reservations.
610 resblks
= mp
->m_sb
.sb_dblocks
;
612 resblks
= min_t(__uint64_t
, resblks
, 8192);
617 * This function does the following on an initial mount of a file system:
618 * - reads the superblock from disk and init the mount struct
619 * - if we're a 32-bit kernel, do a size check on the superblock
620 * so we don't mount terabyte filesystems
621 * - init mount struct realtime fields
622 * - allocate inode hash table for fs
623 * - init directory manager
624 * - perform recovery and init the log manager
628 struct xfs_mount
*mp
)
630 struct xfs_sb
*sbp
= &(mp
->m_sb
);
631 struct xfs_inode
*rip
;
637 xfs_sb_mount_common(mp
, sbp
);
640 * Check for a mismatched features2 values. Older kernels read & wrote
641 * into the wrong sb offset for sb_features2 on some platforms due to
642 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
643 * which made older superblock reading/writing routines swap it as a
646 * For backwards compatibility, we make both slots equal.
648 * If we detect a mismatched field, we OR the set bits into the existing
649 * features2 field in case it has already been modified; we don't want
650 * to lose any features. We then update the bad location with the ORed
651 * value so that older kernels will see any features2 flags. The
652 * superblock writeback code ensures the new sb_features2 is copied to
653 * sb_bad_features2 before it is logged or written to disk.
655 if (xfs_sb_has_mismatched_features2(sbp
)) {
656 xfs_warn(mp
, "correcting sb_features alignment problem");
657 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
658 mp
->m_update_sb
= true;
661 * Re-check for ATTR2 in case it was found in bad_features2
664 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
665 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
666 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
669 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
670 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
671 xfs_sb_version_removeattr2(&mp
->m_sb
);
672 mp
->m_update_sb
= true;
674 /* update sb_versionnum for the clearing of the morebits */
675 if (!sbp
->sb_features2
)
676 mp
->m_update_sb
= true;
679 /* always use v2 inodes by default now */
680 if (!(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_NLINKBIT
)) {
681 mp
->m_sb
.sb_versionnum
|= XFS_SB_VERSION_NLINKBIT
;
682 mp
->m_update_sb
= true;
686 * Check if sb_agblocks is aligned at stripe boundary
687 * If sb_agblocks is NOT aligned turn off m_dalign since
688 * allocator alignment is within an ag, therefore ag has
689 * to be aligned at stripe boundary.
691 error
= xfs_update_alignment(mp
);
695 xfs_alloc_compute_maxlevels(mp
);
696 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
697 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
698 xfs_ialloc_compute_maxlevels(mp
);
699 xfs_rmapbt_compute_maxlevels(mp
);
700 xfs_refcountbt_compute_maxlevels(mp
);
702 xfs_set_maxicount(mp
);
704 /* enable fail_at_unmount as default */
705 mp
->m_fail_unmount
= 1;
707 error
= xfs_sysfs_init(&mp
->m_kobj
, &xfs_mp_ktype
, NULL
, mp
->m_fsname
);
711 error
= xfs_sysfs_init(&mp
->m_stats
.xs_kobj
, &xfs_stats_ktype
,
712 &mp
->m_kobj
, "stats");
714 goto out_remove_sysfs
;
716 error
= xfs_error_sysfs_init(mp
);
721 error
= xfs_uuid_mount(mp
);
723 goto out_remove_error_sysfs
;
726 * Set the minimum read and write sizes
728 xfs_set_rw_sizes(mp
);
730 /* set the low space thresholds for dynamic preallocation */
731 xfs_set_low_space_thresholds(mp
);
734 * Set the inode cluster size.
735 * This may still be overridden by the file system
736 * block size if it is larger than the chosen cluster size.
738 * For v5 filesystems, scale the cluster size with the inode size to
739 * keep a constant ratio of inode per cluster buffer, but only if mkfs
740 * has set the inode alignment value appropriately for larger cluster
743 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
744 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
745 int new_size
= mp
->m_inode_cluster_size
;
747 new_size
*= mp
->m_sb
.sb_inodesize
/ XFS_DINODE_MIN_SIZE
;
748 if (mp
->m_sb
.sb_inoalignmt
>= XFS_B_TO_FSBT(mp
, new_size
))
749 mp
->m_inode_cluster_size
= new_size
;
753 * If enabled, sparse inode chunk alignment is expected to match the
754 * cluster size. Full inode chunk alignment must match the chunk size,
755 * but that is checked on sb read verification...
757 if (xfs_sb_version_hassparseinodes(&mp
->m_sb
) &&
758 mp
->m_sb
.sb_spino_align
!=
759 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
)) {
761 "Sparse inode block alignment (%u) must match cluster size (%llu).",
762 mp
->m_sb
.sb_spino_align
,
763 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
));
765 goto out_remove_uuid
;
769 * Set inode alignment fields
771 xfs_set_inoalignment(mp
);
774 * Check that the data (and log if separate) is an ok size.
776 error
= xfs_check_sizes(mp
);
778 goto out_remove_uuid
;
781 * Initialize realtime fields in the mount structure
783 error
= xfs_rtmount_init(mp
);
785 xfs_warn(mp
, "RT mount failed");
786 goto out_remove_uuid
;
790 * Copies the low order bits of the timestamp and the randomly
791 * set "sequence" number out of a UUID.
793 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
795 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
797 error
= xfs_da_mount(mp
);
799 xfs_warn(mp
, "Failed dir/attr init: %d", error
);
800 goto out_remove_uuid
;
804 * Initialize the precomputed transaction reservations values.
809 * Allocate and initialize the per-ag data.
811 spin_lock_init(&mp
->m_perag_lock
);
812 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
813 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
815 xfs_warn(mp
, "Failed per-ag init: %d", error
);
819 if (!sbp
->sb_logblocks
) {
820 xfs_warn(mp
, "no log defined");
821 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
822 error
= -EFSCORRUPTED
;
827 * Log's mount-time initialization. The first part of recovery can place
828 * some items on the AIL, to be handled when recovery is finished or
831 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
832 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
833 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
835 xfs_warn(mp
, "log mount failed");
840 * Now the log is mounted, we know if it was an unclean shutdown or
841 * not. If it was, with the first phase of recovery has completed, we
842 * have consistent AG blocks on disk. We have not recovered EFIs yet,
843 * but they are recovered transactionally in the second recovery phase
846 * Hence we can safely re-initialise incore superblock counters from
847 * the per-ag data. These may not be correct if the filesystem was not
848 * cleanly unmounted, so we need to wait for recovery to finish before
851 * If the filesystem was cleanly unmounted, then we can trust the
852 * values in the superblock to be correct and we don't need to do
855 * If we are currently making the filesystem, the initialisation will
856 * fail as the perag data is in an undefined state.
858 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
859 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
860 !mp
->m_sb
.sb_inprogress
) {
861 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
863 goto out_log_dealloc
;
867 * Get and sanity-check the root inode.
868 * Save the pointer to it in the mount structure.
870 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
872 xfs_warn(mp
, "failed to read root inode");
873 goto out_log_dealloc
;
878 if (unlikely(!S_ISDIR(VFS_I(rip
)->i_mode
))) {
879 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
880 (unsigned long long)rip
->i_ino
);
881 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
882 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
884 error
= -EFSCORRUPTED
;
887 mp
->m_rootip
= rip
; /* save it */
889 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
892 * Initialize realtime inode pointers in the mount structure
894 error
= xfs_rtmount_inodes(mp
);
897 * Free up the root inode.
899 xfs_warn(mp
, "failed to read RT inodes");
904 * If this is a read-only mount defer the superblock updates until
905 * the next remount into writeable mode. Otherwise we would never
906 * perform the update e.g. for the root filesystem.
908 if (mp
->m_update_sb
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
909 error
= xfs_sync_sb(mp
, false);
911 xfs_warn(mp
, "failed to write sb changes");
917 * Initialise the XFS quota management subsystem for this mount
919 if (XFS_IS_QUOTA_RUNNING(mp
)) {
920 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
924 ASSERT(!XFS_IS_QUOTA_ON(mp
));
927 * If a file system had quotas running earlier, but decided to
928 * mount without -o uquota/pquota/gquota options, revoke the
929 * quotachecked license.
931 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
932 xfs_notice(mp
, "resetting quota flags");
933 error
= xfs_mount_reset_sbqflags(mp
);
940 * During the second phase of log recovery, we need iget and
941 * iput to behave like they do for an active filesystem.
942 * xfs_fs_drop_inode needs to be able to prevent the deletion
943 * of inodes before we're done replaying log items on those
946 mp
->m_super
->s_flags
|= MS_ACTIVE
;
949 * Finish recovering the file system. This part needed to be delayed
950 * until after the root and real-time bitmap inodes were consistently
953 error
= xfs_log_mount_finish(mp
);
955 xfs_warn(mp
, "log mount finish failed");
960 * Now the log is fully replayed, we can transition to full read-only
961 * mode for read-only mounts. This will sync all the metadata and clean
962 * the log so that the recovery we just performed does not have to be
963 * replayed again on the next mount.
965 * We use the same quiesce mechanism as the rw->ro remount, as they are
966 * semantically identical operations.
968 if ((mp
->m_flags
& (XFS_MOUNT_RDONLY
|XFS_MOUNT_NORECOVERY
)) ==
970 xfs_quiesce_attr(mp
);
974 * Complete the quota initialisation, post-log-replay component.
977 ASSERT(mp
->m_qflags
== 0);
978 mp
->m_qflags
= quotaflags
;
980 xfs_qm_mount_quotas(mp
);
984 * Now we are mounted, reserve a small amount of unused space for
985 * privileged transactions. This is needed so that transaction
986 * space required for critical operations can dip into this pool
987 * when at ENOSPC. This is needed for operations like create with
988 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
989 * are not allowed to use this reserved space.
991 * This may drive us straight to ENOSPC on mount, but that implies
992 * we were already there on the last unmount. Warn if this occurs.
994 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
995 resblks
= xfs_default_resblks(mp
);
996 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
999 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1001 /* Recover any CoW blocks that never got remapped. */
1002 error
= xfs_reflink_recover_cow(mp
);
1005 "Error %d recovering leftover CoW allocations.", error
);
1006 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1010 /* Reserve AG blocks for future btree expansion. */
1011 error
= xfs_fs_reserve_ag_blocks(mp
);
1012 if (error
&& error
!= -ENOSPC
)
1019 xfs_fs_unreserve_ag_blocks(mp
);
1021 xfs_qm_unmount_quotas(mp
);
1023 mp
->m_super
->s_flags
&= ~MS_ACTIVE
;
1024 xfs_rtunmount_inodes(mp
);
1027 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1028 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1030 mp
->m_flags
|= XFS_MOUNT_UNMOUNTING
;
1031 xfs_log_mount_cancel(mp
);
1033 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1034 xfs_wait_buftarg(mp
->m_logdev_targp
);
1035 xfs_wait_buftarg(mp
->m_ddev_targp
);
1041 xfs_uuid_unmount(mp
);
1042 out_remove_error_sysfs
:
1043 xfs_error_sysfs_del(mp
);
1045 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1047 xfs_sysfs_del(&mp
->m_kobj
);
1053 * This flushes out the inodes,dquots and the superblock, unmounts the
1054 * log and makes sure that incore structures are freed.
1058 struct xfs_mount
*mp
)
1063 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
1064 cancel_delayed_work_sync(&mp
->m_cowblocks_work
);
1066 xfs_fs_unreserve_ag_blocks(mp
);
1067 xfs_qm_unmount_quotas(mp
);
1068 xfs_rtunmount_inodes(mp
);
1069 IRELE(mp
->m_rootip
);
1072 * We can potentially deadlock here if we have an inode cluster
1073 * that has been freed has its buffer still pinned in memory because
1074 * the transaction is still sitting in a iclog. The stale inodes
1075 * on that buffer will have their flush locks held until the
1076 * transaction hits the disk and the callbacks run. the inode
1077 * flush takes the flush lock unconditionally and with nothing to
1078 * push out the iclog we will never get that unlocked. hence we
1079 * need to force the log first.
1081 xfs_log_force(mp
, XFS_LOG_SYNC
);
1084 * Wait for all busy extents to be freed, including completion of
1085 * any discard operation.
1087 xfs_extent_busy_wait_all(mp
);
1088 flush_workqueue(xfs_discard_wq
);
1091 * We now need to tell the world we are unmounting. This will allow
1092 * us to detect that the filesystem is going away and we should error
1093 * out anything that we have been retrying in the background. This will
1094 * prevent neverending retries in AIL pushing from hanging the unmount.
1096 mp
->m_flags
|= XFS_MOUNT_UNMOUNTING
;
1099 * Flush all pending changes from the AIL.
1101 xfs_ail_push_all_sync(mp
->m_ail
);
1104 * And reclaim all inodes. At this point there should be no dirty
1105 * inodes and none should be pinned or locked, but use synchronous
1106 * reclaim just to be sure. We can stop background inode reclaim
1107 * here as well if it is still running.
1109 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1110 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1115 * Unreserve any blocks we have so that when we unmount we don't account
1116 * the reserved free space as used. This is really only necessary for
1117 * lazy superblock counting because it trusts the incore superblock
1118 * counters to be absolutely correct on clean unmount.
1120 * We don't bother correcting this elsewhere for lazy superblock
1121 * counting because on mount of an unclean filesystem we reconstruct the
1122 * correct counter value and this is irrelevant.
1124 * For non-lazy counter filesystems, this doesn't matter at all because
1125 * we only every apply deltas to the superblock and hence the incore
1126 * value does not matter....
1129 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1131 xfs_warn(mp
, "Unable to free reserved block pool. "
1132 "Freespace may not be correct on next mount.");
1134 error
= xfs_log_sbcount(mp
);
1136 xfs_warn(mp
, "Unable to update superblock counters. "
1137 "Freespace may not be correct on next mount.");
1140 xfs_log_unmount(mp
);
1142 xfs_uuid_unmount(mp
);
1145 xfs_errortag_clearall(mp
, 0);
1149 xfs_error_sysfs_del(mp
);
1150 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1151 xfs_sysfs_del(&mp
->m_kobj
);
1155 * Determine whether modifications can proceed. The caller specifies the minimum
1156 * freeze level for which modifications should not be allowed. This allows
1157 * certain operations to proceed while the freeze sequence is in progress, if
1162 struct xfs_mount
*mp
,
1165 ASSERT(level
> SB_UNFROZEN
);
1166 if ((mp
->m_super
->s_writers
.frozen
>= level
) ||
1167 XFS_FORCED_SHUTDOWN(mp
) || (mp
->m_flags
& XFS_MOUNT_RDONLY
))
1176 * Sync the superblock counters to disk.
1178 * Note this code can be called during the process of freezing, so we use the
1179 * transaction allocator that does not block when the transaction subsystem is
1180 * in its frozen state.
1183 xfs_log_sbcount(xfs_mount_t
*mp
)
1185 /* allow this to proceed during the freeze sequence... */
1186 if (!xfs_fs_writable(mp
, SB_FREEZE_COMPLETE
))
1190 * we don't need to do this if we are updating the superblock
1191 * counters on every modification.
1193 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1196 return xfs_sync_sb(mp
, true);
1200 * Deltas for the inode count are +/-64, hence we use a large batch size
1201 * of 128 so we don't need to take the counter lock on every update.
1203 #define XFS_ICOUNT_BATCH 128
1206 struct xfs_mount
*mp
,
1209 __percpu_counter_add(&mp
->m_icount
, delta
, XFS_ICOUNT_BATCH
);
1210 if (__percpu_counter_compare(&mp
->m_icount
, 0, XFS_ICOUNT_BATCH
) < 0) {
1212 percpu_counter_add(&mp
->m_icount
, -delta
);
1220 struct xfs_mount
*mp
,
1223 percpu_counter_add(&mp
->m_ifree
, delta
);
1224 if (percpu_counter_compare(&mp
->m_ifree
, 0) < 0) {
1226 percpu_counter_add(&mp
->m_ifree
, -delta
);
1233 * Deltas for the block count can vary from 1 to very large, but lock contention
1234 * only occurs on frequent small block count updates such as in the delayed
1235 * allocation path for buffered writes (page a time updates). Hence we set
1236 * a large batch count (1024) to minimise global counter updates except when
1237 * we get near to ENOSPC and we have to be very accurate with our updates.
1239 #define XFS_FDBLOCKS_BATCH 1024
1242 struct xfs_mount
*mp
,
1252 * If the reserve pool is depleted, put blocks back into it
1253 * first. Most of the time the pool is full.
1255 if (likely(mp
->m_resblks
== mp
->m_resblks_avail
)) {
1256 percpu_counter_add(&mp
->m_fdblocks
, delta
);
1260 spin_lock(&mp
->m_sb_lock
);
1261 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1263 if (res_used
> delta
) {
1264 mp
->m_resblks_avail
+= delta
;
1267 mp
->m_resblks_avail
= mp
->m_resblks
;
1268 percpu_counter_add(&mp
->m_fdblocks
, delta
);
1270 spin_unlock(&mp
->m_sb_lock
);
1275 * Taking blocks away, need to be more accurate the closer we
1278 * If the counter has a value of less than 2 * max batch size,
1279 * then make everything serialise as we are real close to
1282 if (__percpu_counter_compare(&mp
->m_fdblocks
, 2 * XFS_FDBLOCKS_BATCH
,
1283 XFS_FDBLOCKS_BATCH
) < 0)
1286 batch
= XFS_FDBLOCKS_BATCH
;
1288 __percpu_counter_add(&mp
->m_fdblocks
, delta
, batch
);
1289 if (__percpu_counter_compare(&mp
->m_fdblocks
, mp
->m_alloc_set_aside
,
1290 XFS_FDBLOCKS_BATCH
) >= 0) {
1296 * lock up the sb for dipping into reserves before releasing the space
1297 * that took us to ENOSPC.
1299 spin_lock(&mp
->m_sb_lock
);
1300 percpu_counter_add(&mp
->m_fdblocks
, -delta
);
1302 goto fdblocks_enospc
;
1304 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1305 if (lcounter
>= 0) {
1306 mp
->m_resblks_avail
= lcounter
;
1307 spin_unlock(&mp
->m_sb_lock
);
1310 printk_once(KERN_WARNING
1311 "Filesystem \"%s\": reserve blocks depleted! "
1312 "Consider increasing reserve pool size.",
1315 spin_unlock(&mp
->m_sb_lock
);
1321 struct xfs_mount
*mp
,
1327 spin_lock(&mp
->m_sb_lock
);
1328 lcounter
= mp
->m_sb
.sb_frextents
+ delta
;
1332 mp
->m_sb
.sb_frextents
= lcounter
;
1333 spin_unlock(&mp
->m_sb_lock
);
1338 * xfs_getsb() is called to obtain the buffer for the superblock.
1339 * The buffer is returned locked and read in from disk.
1340 * The buffer should be released with a call to xfs_brelse().
1342 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1343 * the superblock buffer if it can be locked without sleeping.
1344 * If it can't then we'll return NULL.
1348 struct xfs_mount
*mp
,
1351 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1353 if (!xfs_buf_trylock(bp
)) {
1354 if (flags
& XBF_TRYLOCK
)
1360 ASSERT(bp
->b_flags
& XBF_DONE
);
1365 * Used to free the superblock along various error paths.
1369 struct xfs_mount
*mp
)
1371 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1379 * If the underlying (data/log/rt) device is readonly, there are some
1380 * operations that cannot proceed.
1383 xfs_dev_is_read_only(
1384 struct xfs_mount
*mp
,
1387 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1388 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1389 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1390 xfs_notice(mp
, "%s required on read-only device.", message
);
1391 xfs_notice(mp
, "write access unavailable, cannot proceed.");