2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
36 #include "xfs_trans.h"
37 #include "xfs_trans_priv.h"
39 #include "xfs_error.h"
40 #include "xfs_quota.h"
41 #include "xfs_fsops.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_sysfs.h"
45 #include "xfs_rmap_btree.h"
46 #include "xfs_refcount_btree.h"
47 #include "xfs_reflink.h"
48 #include "xfs_extent_busy.h"
51 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
52 static int xfs_uuid_table_size
;
53 static uuid_t
*xfs_uuid_table
;
56 xfs_uuid_table_free(void)
58 if (xfs_uuid_table_size
== 0)
60 kmem_free(xfs_uuid_table
);
61 xfs_uuid_table
= NULL
;
62 xfs_uuid_table_size
= 0;
66 * See if the UUID is unique among mounted XFS filesystems.
67 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
73 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
76 /* Publish UUID in struct super_block */
77 BUILD_BUG_ON(sizeof(mp
->m_super
->s_uuid
) != sizeof(uuid_t
));
78 memcpy(&mp
->m_super
->s_uuid
, uuid
, sizeof(uuid_t
));
80 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
83 if (uuid_is_nil(uuid
)) {
84 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
88 mutex_lock(&xfs_uuid_table_mutex
);
89 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
90 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
94 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
99 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
100 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
102 hole
= xfs_uuid_table_size
++;
104 xfs_uuid_table
[hole
] = *uuid
;
105 mutex_unlock(&xfs_uuid_table_mutex
);
110 mutex_unlock(&xfs_uuid_table_mutex
);
111 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
117 struct xfs_mount
*mp
)
119 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
122 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
125 mutex_lock(&xfs_uuid_table_mutex
);
126 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
127 if (uuid_is_nil(&xfs_uuid_table
[i
]))
129 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
131 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
134 ASSERT(i
< xfs_uuid_table_size
);
135 mutex_unlock(&xfs_uuid_table_mutex
);
141 struct rcu_head
*head
)
143 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
145 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
150 * Free up the per-ag resources associated with the mount structure.
157 struct xfs_perag
*pag
;
159 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
160 spin_lock(&mp
->m_perag_lock
);
161 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
162 spin_unlock(&mp
->m_perag_lock
);
164 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
165 xfs_buf_hash_destroy(pag
);
166 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
171 * Check size of device based on the (data/realtime) block count.
172 * Note: this check is used by the growfs code as well as mount.
175 xfs_sb_validate_fsb_count(
179 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
180 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
182 /* Limited by ULONG_MAX of page cache index */
183 if (nblocks
>> (PAGE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
189 xfs_initialize_perag(
191 xfs_agnumber_t agcount
,
192 xfs_agnumber_t
*maxagi
)
194 xfs_agnumber_t index
;
195 xfs_agnumber_t first_initialised
= NULLAGNUMBER
;
200 * Walk the current per-ag tree so we don't try to initialise AGs
201 * that already exist (growfs case). Allocate and insert all the
202 * AGs we don't find ready for initialisation.
204 for (index
= 0; index
< agcount
; index
++) {
205 pag
= xfs_perag_get(mp
, index
);
211 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
213 goto out_unwind_new_pags
;
214 pag
->pag_agno
= index
;
216 spin_lock_init(&pag
->pag_ici_lock
);
217 mutex_init(&pag
->pag_ici_reclaim_lock
);
218 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
219 if (xfs_buf_hash_init(pag
))
221 init_waitqueue_head(&pag
->pagb_wait
);
223 if (radix_tree_preload(GFP_NOFS
))
224 goto out_hash_destroy
;
226 spin_lock(&mp
->m_perag_lock
);
227 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
229 spin_unlock(&mp
->m_perag_lock
);
230 radix_tree_preload_end();
232 goto out_hash_destroy
;
234 spin_unlock(&mp
->m_perag_lock
);
235 radix_tree_preload_end();
236 /* first new pag is fully initialized */
237 if (first_initialised
== NULLAGNUMBER
)
238 first_initialised
= index
;
241 index
= xfs_set_inode_alloc(mp
, agcount
);
246 mp
->m_ag_prealloc_blocks
= xfs_prealloc_blocks(mp
);
250 xfs_buf_hash_destroy(pag
);
254 /* unwind any prior newly initialized pags */
255 for (index
= first_initialised
; index
< agcount
; index
++) {
256 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
259 xfs_buf_hash_destroy(pag
);
268 * Does the initial read of the superblock.
272 struct xfs_mount
*mp
,
275 unsigned int sector_size
;
277 struct xfs_sb
*sbp
= &mp
->m_sb
;
279 int loud
= !(flags
& XFS_MFSI_QUIET
);
280 const struct xfs_buf_ops
*buf_ops
;
282 ASSERT(mp
->m_sb_bp
== NULL
);
283 ASSERT(mp
->m_ddev_targp
!= NULL
);
286 * For the initial read, we must guess at the sector
287 * size based on the block device. It's enough to
288 * get the sb_sectsize out of the superblock and
289 * then reread with the proper length.
290 * We don't verify it yet, because it may not be complete.
292 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
296 * Allocate a (locked) buffer to hold the superblock. This will be kept
297 * around at all times to optimize access to the superblock. Therefore,
298 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
302 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
303 BTOBB(sector_size
), XBF_NO_IOACCT
, &bp
,
307 xfs_warn(mp
, "SB validate failed with error %d.", error
);
308 /* bad CRC means corrupted metadata */
309 if (error
== -EFSBADCRC
)
310 error
= -EFSCORRUPTED
;
315 * Initialize the mount structure from the superblock.
317 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
320 * If we haven't validated the superblock, do so now before we try
321 * to check the sector size and reread the superblock appropriately.
323 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
325 xfs_warn(mp
, "Invalid superblock magic number");
331 * We must be able to do sector-sized and sector-aligned IO.
333 if (sector_size
> sbp
->sb_sectsize
) {
335 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
336 sector_size
, sbp
->sb_sectsize
);
341 if (buf_ops
== NULL
) {
343 * Re-read the superblock so the buffer is correctly sized,
344 * and properly verified.
347 sector_size
= sbp
->sb_sectsize
;
348 buf_ops
= loud
? &xfs_sb_buf_ops
: &xfs_sb_quiet_buf_ops
;
352 xfs_reinit_percpu_counters(mp
);
354 /* no need to be quiet anymore, so reset the buf ops */
355 bp
->b_ops
= &xfs_sb_buf_ops
;
367 * Update alignment values based on mount options and sb values
370 xfs_update_alignment(xfs_mount_t
*mp
)
372 xfs_sb_t
*sbp
= &(mp
->m_sb
);
376 * If stripe unit and stripe width are not multiples
377 * of the fs blocksize turn off alignment.
379 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
380 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
382 "alignment check failed: sunit/swidth vs. blocksize(%d)",
387 * Convert the stripe unit and width to FSBs.
389 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
390 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
392 "alignment check failed: sunit/swidth vs. agsize(%d)",
395 } else if (mp
->m_dalign
) {
396 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
399 "alignment check failed: sunit(%d) less than bsize(%d)",
400 mp
->m_dalign
, sbp
->sb_blocksize
);
406 * Update superblock with new values
409 if (xfs_sb_version_hasdalign(sbp
)) {
410 if (sbp
->sb_unit
!= mp
->m_dalign
) {
411 sbp
->sb_unit
= mp
->m_dalign
;
412 mp
->m_update_sb
= true;
414 if (sbp
->sb_width
!= mp
->m_swidth
) {
415 sbp
->sb_width
= mp
->m_swidth
;
416 mp
->m_update_sb
= true;
420 "cannot change alignment: superblock does not support data alignment");
423 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
424 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
425 mp
->m_dalign
= sbp
->sb_unit
;
426 mp
->m_swidth
= sbp
->sb_width
;
433 * Set the maximum inode count for this filesystem
436 xfs_set_maxicount(xfs_mount_t
*mp
)
438 xfs_sb_t
*sbp
= &(mp
->m_sb
);
441 if (sbp
->sb_imax_pct
) {
443 * Make sure the maximum inode count is a multiple
444 * of the units we allocate inodes in.
446 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
448 do_div(icount
, mp
->m_ialloc_blks
);
449 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
457 * Set the default minimum read and write sizes unless
458 * already specified in a mount option.
459 * We use smaller I/O sizes when the file system
460 * is being used for NFS service (wsync mount option).
463 xfs_set_rw_sizes(xfs_mount_t
*mp
)
465 xfs_sb_t
*sbp
= &(mp
->m_sb
);
466 int readio_log
, writeio_log
;
468 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
469 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
470 readio_log
= XFS_WSYNC_READIO_LOG
;
471 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
473 readio_log
= XFS_READIO_LOG_LARGE
;
474 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
477 readio_log
= mp
->m_readio_log
;
478 writeio_log
= mp
->m_writeio_log
;
481 if (sbp
->sb_blocklog
> readio_log
) {
482 mp
->m_readio_log
= sbp
->sb_blocklog
;
484 mp
->m_readio_log
= readio_log
;
486 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
487 if (sbp
->sb_blocklog
> writeio_log
) {
488 mp
->m_writeio_log
= sbp
->sb_blocklog
;
490 mp
->m_writeio_log
= writeio_log
;
492 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
496 * precalculate the low space thresholds for dynamic speculative preallocation.
499 xfs_set_low_space_thresholds(
500 struct xfs_mount
*mp
)
504 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
505 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
508 mp
->m_low_space
[i
] = space
* (i
+ 1);
514 * Set whether we're using inode alignment.
517 xfs_set_inoalignment(xfs_mount_t
*mp
)
519 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
520 mp
->m_sb
.sb_inoalignmt
>= xfs_icluster_size_fsb(mp
))
521 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
523 mp
->m_inoalign_mask
= 0;
525 * If we are using stripe alignment, check whether
526 * the stripe unit is a multiple of the inode alignment
528 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
529 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
530 mp
->m_sinoalign
= mp
->m_dalign
;
536 * Check that the data (and log if separate) is an ok size.
540 struct xfs_mount
*mp
)
546 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
547 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
548 xfs_warn(mp
, "filesystem size mismatch detected");
551 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
552 d
- XFS_FSS_TO_BB(mp
, 1),
553 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, NULL
);
555 xfs_warn(mp
, "last sector read failed");
560 if (mp
->m_logdev_targp
== mp
->m_ddev_targp
)
563 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
564 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
565 xfs_warn(mp
, "log size mismatch detected");
568 error
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
569 d
- XFS_FSB_TO_BB(mp
, 1),
570 XFS_FSB_TO_BB(mp
, 1), 0, &bp
, NULL
);
572 xfs_warn(mp
, "log device read failed");
580 * Clear the quotaflags in memory and in the superblock.
583 xfs_mount_reset_sbqflags(
584 struct xfs_mount
*mp
)
588 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
589 if (mp
->m_sb
.sb_qflags
== 0)
591 spin_lock(&mp
->m_sb_lock
);
592 mp
->m_sb
.sb_qflags
= 0;
593 spin_unlock(&mp
->m_sb_lock
);
595 if (!xfs_fs_writable(mp
, SB_FREEZE_WRITE
))
598 return xfs_sync_sb(mp
, false);
602 xfs_default_resblks(xfs_mount_t
*mp
)
607 * We default to 5% or 8192 fsbs of space reserved, whichever is
608 * smaller. This is intended to cover concurrent allocation
609 * transactions when we initially hit enospc. These each require a 4
610 * block reservation. Hence by default we cover roughly 2000 concurrent
611 * allocation reservations.
613 resblks
= mp
->m_sb
.sb_dblocks
;
615 resblks
= min_t(__uint64_t
, resblks
, 8192);
620 * This function does the following on an initial mount of a file system:
621 * - reads the superblock from disk and init the mount struct
622 * - if we're a 32-bit kernel, do a size check on the superblock
623 * so we don't mount terabyte filesystems
624 * - init mount struct realtime fields
625 * - allocate inode hash table for fs
626 * - init directory manager
627 * - perform recovery and init the log manager
631 struct xfs_mount
*mp
)
633 struct xfs_sb
*sbp
= &(mp
->m_sb
);
634 struct xfs_inode
*rip
;
640 xfs_sb_mount_common(mp
, sbp
);
643 * Check for a mismatched features2 values. Older kernels read & wrote
644 * into the wrong sb offset for sb_features2 on some platforms due to
645 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
646 * which made older superblock reading/writing routines swap it as a
649 * For backwards compatibility, we make both slots equal.
651 * If we detect a mismatched field, we OR the set bits into the existing
652 * features2 field in case it has already been modified; we don't want
653 * to lose any features. We then update the bad location with the ORed
654 * value so that older kernels will see any features2 flags. The
655 * superblock writeback code ensures the new sb_features2 is copied to
656 * sb_bad_features2 before it is logged or written to disk.
658 if (xfs_sb_has_mismatched_features2(sbp
)) {
659 xfs_warn(mp
, "correcting sb_features alignment problem");
660 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
661 mp
->m_update_sb
= true;
664 * Re-check for ATTR2 in case it was found in bad_features2
667 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
668 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
669 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
672 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
673 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
674 xfs_sb_version_removeattr2(&mp
->m_sb
);
675 mp
->m_update_sb
= true;
677 /* update sb_versionnum for the clearing of the morebits */
678 if (!sbp
->sb_features2
)
679 mp
->m_update_sb
= true;
682 /* always use v2 inodes by default now */
683 if (!(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_NLINKBIT
)) {
684 mp
->m_sb
.sb_versionnum
|= XFS_SB_VERSION_NLINKBIT
;
685 mp
->m_update_sb
= true;
689 * Check if sb_agblocks is aligned at stripe boundary
690 * If sb_agblocks is NOT aligned turn off m_dalign since
691 * allocator alignment is within an ag, therefore ag has
692 * to be aligned at stripe boundary.
694 error
= xfs_update_alignment(mp
);
698 xfs_alloc_compute_maxlevels(mp
);
699 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
700 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
701 xfs_ialloc_compute_maxlevels(mp
);
702 xfs_rmapbt_compute_maxlevels(mp
);
703 xfs_refcountbt_compute_maxlevels(mp
);
705 xfs_set_maxicount(mp
);
707 /* enable fail_at_unmount as default */
708 mp
->m_fail_unmount
= 1;
710 error
= xfs_sysfs_init(&mp
->m_kobj
, &xfs_mp_ktype
, NULL
, mp
->m_fsname
);
714 error
= xfs_sysfs_init(&mp
->m_stats
.xs_kobj
, &xfs_stats_ktype
,
715 &mp
->m_kobj
, "stats");
717 goto out_remove_sysfs
;
719 error
= xfs_error_sysfs_init(mp
);
724 error
= xfs_uuid_mount(mp
);
726 goto out_remove_error_sysfs
;
729 * Set the minimum read and write sizes
731 xfs_set_rw_sizes(mp
);
733 /* set the low space thresholds for dynamic preallocation */
734 xfs_set_low_space_thresholds(mp
);
737 * Set the inode cluster size.
738 * This may still be overridden by the file system
739 * block size if it is larger than the chosen cluster size.
741 * For v5 filesystems, scale the cluster size with the inode size to
742 * keep a constant ratio of inode per cluster buffer, but only if mkfs
743 * has set the inode alignment value appropriately for larger cluster
746 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
747 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
748 int new_size
= mp
->m_inode_cluster_size
;
750 new_size
*= mp
->m_sb
.sb_inodesize
/ XFS_DINODE_MIN_SIZE
;
751 if (mp
->m_sb
.sb_inoalignmt
>= XFS_B_TO_FSBT(mp
, new_size
))
752 mp
->m_inode_cluster_size
= new_size
;
756 * If enabled, sparse inode chunk alignment is expected to match the
757 * cluster size. Full inode chunk alignment must match the chunk size,
758 * but that is checked on sb read verification...
760 if (xfs_sb_version_hassparseinodes(&mp
->m_sb
) &&
761 mp
->m_sb
.sb_spino_align
!=
762 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
)) {
764 "Sparse inode block alignment (%u) must match cluster size (%llu).",
765 mp
->m_sb
.sb_spino_align
,
766 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
));
768 goto out_remove_uuid
;
772 * Set inode alignment fields
774 xfs_set_inoalignment(mp
);
777 * Check that the data (and log if separate) is an ok size.
779 error
= xfs_check_sizes(mp
);
781 goto out_remove_uuid
;
784 * Initialize realtime fields in the mount structure
786 error
= xfs_rtmount_init(mp
);
788 xfs_warn(mp
, "RT mount failed");
789 goto out_remove_uuid
;
793 * Copies the low order bits of the timestamp and the randomly
794 * set "sequence" number out of a UUID.
796 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
798 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
800 error
= xfs_da_mount(mp
);
802 xfs_warn(mp
, "Failed dir/attr init: %d", error
);
803 goto out_remove_uuid
;
807 * Initialize the precomputed transaction reservations values.
812 * Allocate and initialize the per-ag data.
814 spin_lock_init(&mp
->m_perag_lock
);
815 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
816 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
818 xfs_warn(mp
, "Failed per-ag init: %d", error
);
822 if (!sbp
->sb_logblocks
) {
823 xfs_warn(mp
, "no log defined");
824 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
825 error
= -EFSCORRUPTED
;
830 * Log's mount-time initialization. The first part of recovery can place
831 * some items on the AIL, to be handled when recovery is finished or
834 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
835 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
836 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
838 xfs_warn(mp
, "log mount failed");
843 * Now the log is mounted, we know if it was an unclean shutdown or
844 * not. If it was, with the first phase of recovery has completed, we
845 * have consistent AG blocks on disk. We have not recovered EFIs yet,
846 * but they are recovered transactionally in the second recovery phase
849 * Hence we can safely re-initialise incore superblock counters from
850 * the per-ag data. These may not be correct if the filesystem was not
851 * cleanly unmounted, so we need to wait for recovery to finish before
854 * If the filesystem was cleanly unmounted, then we can trust the
855 * values in the superblock to be correct and we don't need to do
858 * If we are currently making the filesystem, the initialisation will
859 * fail as the perag data is in an undefined state.
861 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
862 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
863 !mp
->m_sb
.sb_inprogress
) {
864 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
866 goto out_log_dealloc
;
870 * Get and sanity-check the root inode.
871 * Save the pointer to it in the mount structure.
873 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
875 xfs_warn(mp
, "failed to read root inode");
876 goto out_log_dealloc
;
881 if (unlikely(!S_ISDIR(VFS_I(rip
)->i_mode
))) {
882 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
883 (unsigned long long)rip
->i_ino
);
884 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
885 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
887 error
= -EFSCORRUPTED
;
890 mp
->m_rootip
= rip
; /* save it */
892 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
895 * Initialize realtime inode pointers in the mount structure
897 error
= xfs_rtmount_inodes(mp
);
900 * Free up the root inode.
902 xfs_warn(mp
, "failed to read RT inodes");
907 * If this is a read-only mount defer the superblock updates until
908 * the next remount into writeable mode. Otherwise we would never
909 * perform the update e.g. for the root filesystem.
911 if (mp
->m_update_sb
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
912 error
= xfs_sync_sb(mp
, false);
914 xfs_warn(mp
, "failed to write sb changes");
920 * Initialise the XFS quota management subsystem for this mount
922 if (XFS_IS_QUOTA_RUNNING(mp
)) {
923 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
927 ASSERT(!XFS_IS_QUOTA_ON(mp
));
930 * If a file system had quotas running earlier, but decided to
931 * mount without -o uquota/pquota/gquota options, revoke the
932 * quotachecked license.
934 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
935 xfs_notice(mp
, "resetting quota flags");
936 error
= xfs_mount_reset_sbqflags(mp
);
943 * During the second phase of log recovery, we need iget and
944 * iput to behave like they do for an active filesystem.
945 * xfs_fs_drop_inode needs to be able to prevent the deletion
946 * of inodes before we're done replaying log items on those
949 mp
->m_super
->s_flags
|= MS_ACTIVE
;
952 * Finish recovering the file system. This part needed to be delayed
953 * until after the root and real-time bitmap inodes were consistently
956 error
= xfs_log_mount_finish(mp
);
958 xfs_warn(mp
, "log mount finish failed");
963 * Now the log is fully replayed, we can transition to full read-only
964 * mode for read-only mounts. This will sync all the metadata and clean
965 * the log so that the recovery we just performed does not have to be
966 * replayed again on the next mount.
968 * We use the same quiesce mechanism as the rw->ro remount, as they are
969 * semantically identical operations.
971 if ((mp
->m_flags
& (XFS_MOUNT_RDONLY
|XFS_MOUNT_NORECOVERY
)) ==
973 xfs_quiesce_attr(mp
);
977 * Complete the quota initialisation, post-log-replay component.
980 ASSERT(mp
->m_qflags
== 0);
981 mp
->m_qflags
= quotaflags
;
983 xfs_qm_mount_quotas(mp
);
987 * Now we are mounted, reserve a small amount of unused space for
988 * privileged transactions. This is needed so that transaction
989 * space required for critical operations can dip into this pool
990 * when at ENOSPC. This is needed for operations like create with
991 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
992 * are not allowed to use this reserved space.
994 * This may drive us straight to ENOSPC on mount, but that implies
995 * we were already there on the last unmount. Warn if this occurs.
997 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
998 resblks
= xfs_default_resblks(mp
);
999 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1002 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1004 /* Recover any CoW blocks that never got remapped. */
1005 error
= xfs_reflink_recover_cow(mp
);
1008 "Error %d recovering leftover CoW allocations.", error
);
1009 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1013 /* Reserve AG blocks for future btree expansion. */
1014 error
= xfs_fs_reserve_ag_blocks(mp
);
1015 if (error
&& error
!= -ENOSPC
)
1022 xfs_fs_unreserve_ag_blocks(mp
);
1024 xfs_qm_unmount_quotas(mp
);
1026 mp
->m_super
->s_flags
&= ~MS_ACTIVE
;
1027 xfs_rtunmount_inodes(mp
);
1030 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1031 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1033 mp
->m_flags
|= XFS_MOUNT_UNMOUNTING
;
1034 xfs_log_mount_cancel(mp
);
1036 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1037 xfs_wait_buftarg(mp
->m_logdev_targp
);
1038 xfs_wait_buftarg(mp
->m_ddev_targp
);
1044 xfs_uuid_unmount(mp
);
1045 out_remove_error_sysfs
:
1046 xfs_error_sysfs_del(mp
);
1048 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1050 xfs_sysfs_del(&mp
->m_kobj
);
1056 * This flushes out the inodes,dquots and the superblock, unmounts the
1057 * log and makes sure that incore structures are freed.
1061 struct xfs_mount
*mp
)
1066 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
1067 cancel_delayed_work_sync(&mp
->m_cowblocks_work
);
1069 xfs_fs_unreserve_ag_blocks(mp
);
1070 xfs_qm_unmount_quotas(mp
);
1071 xfs_rtunmount_inodes(mp
);
1072 IRELE(mp
->m_rootip
);
1075 * We can potentially deadlock here if we have an inode cluster
1076 * that has been freed has its buffer still pinned in memory because
1077 * the transaction is still sitting in a iclog. The stale inodes
1078 * on that buffer will have their flush locks held until the
1079 * transaction hits the disk and the callbacks run. the inode
1080 * flush takes the flush lock unconditionally and with nothing to
1081 * push out the iclog we will never get that unlocked. hence we
1082 * need to force the log first.
1084 xfs_log_force(mp
, XFS_LOG_SYNC
);
1087 * Wait for all busy extents to be freed, including completion of
1088 * any discard operation.
1090 xfs_extent_busy_wait_all(mp
);
1091 flush_workqueue(xfs_discard_wq
);
1094 * We now need to tell the world we are unmounting. This will allow
1095 * us to detect that the filesystem is going away and we should error
1096 * out anything that we have been retrying in the background. This will
1097 * prevent neverending retries in AIL pushing from hanging the unmount.
1099 mp
->m_flags
|= XFS_MOUNT_UNMOUNTING
;
1102 * Flush all pending changes from the AIL.
1104 xfs_ail_push_all_sync(mp
->m_ail
);
1107 * And reclaim all inodes. At this point there should be no dirty
1108 * inodes and none should be pinned or locked, but use synchronous
1109 * reclaim just to be sure. We can stop background inode reclaim
1110 * here as well if it is still running.
1112 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1113 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1118 * Unreserve any blocks we have so that when we unmount we don't account
1119 * the reserved free space as used. This is really only necessary for
1120 * lazy superblock counting because it trusts the incore superblock
1121 * counters to be absolutely correct on clean unmount.
1123 * We don't bother correcting this elsewhere for lazy superblock
1124 * counting because on mount of an unclean filesystem we reconstruct the
1125 * correct counter value and this is irrelevant.
1127 * For non-lazy counter filesystems, this doesn't matter at all because
1128 * we only every apply deltas to the superblock and hence the incore
1129 * value does not matter....
1132 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1134 xfs_warn(mp
, "Unable to free reserved block pool. "
1135 "Freespace may not be correct on next mount.");
1137 error
= xfs_log_sbcount(mp
);
1139 xfs_warn(mp
, "Unable to update superblock counters. "
1140 "Freespace may not be correct on next mount.");
1143 xfs_log_unmount(mp
);
1145 xfs_uuid_unmount(mp
);
1148 xfs_errortag_clearall(mp
, 0);
1152 xfs_error_sysfs_del(mp
);
1153 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1154 xfs_sysfs_del(&mp
->m_kobj
);
1158 * Determine whether modifications can proceed. The caller specifies the minimum
1159 * freeze level for which modifications should not be allowed. This allows
1160 * certain operations to proceed while the freeze sequence is in progress, if
1165 struct xfs_mount
*mp
,
1168 ASSERT(level
> SB_UNFROZEN
);
1169 if ((mp
->m_super
->s_writers
.frozen
>= level
) ||
1170 XFS_FORCED_SHUTDOWN(mp
) || (mp
->m_flags
& XFS_MOUNT_RDONLY
))
1179 * Sync the superblock counters to disk.
1181 * Note this code can be called during the process of freezing, so we use the
1182 * transaction allocator that does not block when the transaction subsystem is
1183 * in its frozen state.
1186 xfs_log_sbcount(xfs_mount_t
*mp
)
1188 /* allow this to proceed during the freeze sequence... */
1189 if (!xfs_fs_writable(mp
, SB_FREEZE_COMPLETE
))
1193 * we don't need to do this if we are updating the superblock
1194 * counters on every modification.
1196 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1199 return xfs_sync_sb(mp
, true);
1203 * Deltas for the inode count are +/-64, hence we use a large batch size
1204 * of 128 so we don't need to take the counter lock on every update.
1206 #define XFS_ICOUNT_BATCH 128
1209 struct xfs_mount
*mp
,
1212 __percpu_counter_add(&mp
->m_icount
, delta
, XFS_ICOUNT_BATCH
);
1213 if (__percpu_counter_compare(&mp
->m_icount
, 0, XFS_ICOUNT_BATCH
) < 0) {
1215 percpu_counter_add(&mp
->m_icount
, -delta
);
1223 struct xfs_mount
*mp
,
1226 percpu_counter_add(&mp
->m_ifree
, delta
);
1227 if (percpu_counter_compare(&mp
->m_ifree
, 0) < 0) {
1229 percpu_counter_add(&mp
->m_ifree
, -delta
);
1236 * Deltas for the block count can vary from 1 to very large, but lock contention
1237 * only occurs on frequent small block count updates such as in the delayed
1238 * allocation path for buffered writes (page a time updates). Hence we set
1239 * a large batch count (1024) to minimise global counter updates except when
1240 * we get near to ENOSPC and we have to be very accurate with our updates.
1242 #define XFS_FDBLOCKS_BATCH 1024
1245 struct xfs_mount
*mp
,
1255 * If the reserve pool is depleted, put blocks back into it
1256 * first. Most of the time the pool is full.
1258 if (likely(mp
->m_resblks
== mp
->m_resblks_avail
)) {
1259 percpu_counter_add(&mp
->m_fdblocks
, delta
);
1263 spin_lock(&mp
->m_sb_lock
);
1264 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1266 if (res_used
> delta
) {
1267 mp
->m_resblks_avail
+= delta
;
1270 mp
->m_resblks_avail
= mp
->m_resblks
;
1271 percpu_counter_add(&mp
->m_fdblocks
, delta
);
1273 spin_unlock(&mp
->m_sb_lock
);
1278 * Taking blocks away, need to be more accurate the closer we
1281 * If the counter has a value of less than 2 * max batch size,
1282 * then make everything serialise as we are real close to
1285 if (__percpu_counter_compare(&mp
->m_fdblocks
, 2 * XFS_FDBLOCKS_BATCH
,
1286 XFS_FDBLOCKS_BATCH
) < 0)
1289 batch
= XFS_FDBLOCKS_BATCH
;
1291 __percpu_counter_add(&mp
->m_fdblocks
, delta
, batch
);
1292 if (__percpu_counter_compare(&mp
->m_fdblocks
, mp
->m_alloc_set_aside
,
1293 XFS_FDBLOCKS_BATCH
) >= 0) {
1299 * lock up the sb for dipping into reserves before releasing the space
1300 * that took us to ENOSPC.
1302 spin_lock(&mp
->m_sb_lock
);
1303 percpu_counter_add(&mp
->m_fdblocks
, -delta
);
1305 goto fdblocks_enospc
;
1307 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1308 if (lcounter
>= 0) {
1309 mp
->m_resblks_avail
= lcounter
;
1310 spin_unlock(&mp
->m_sb_lock
);
1313 printk_once(KERN_WARNING
1314 "Filesystem \"%s\": reserve blocks depleted! "
1315 "Consider increasing reserve pool size.",
1318 spin_unlock(&mp
->m_sb_lock
);
1324 struct xfs_mount
*mp
,
1330 spin_lock(&mp
->m_sb_lock
);
1331 lcounter
= mp
->m_sb
.sb_frextents
+ delta
;
1335 mp
->m_sb
.sb_frextents
= lcounter
;
1336 spin_unlock(&mp
->m_sb_lock
);
1341 * xfs_getsb() is called to obtain the buffer for the superblock.
1342 * The buffer is returned locked and read in from disk.
1343 * The buffer should be released with a call to xfs_brelse().
1345 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1346 * the superblock buffer if it can be locked without sleeping.
1347 * If it can't then we'll return NULL.
1351 struct xfs_mount
*mp
,
1354 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1356 if (!xfs_buf_trylock(bp
)) {
1357 if (flags
& XBF_TRYLOCK
)
1363 ASSERT(bp
->b_flags
& XBF_DONE
);
1368 * Used to free the superblock along various error paths.
1372 struct xfs_mount
*mp
)
1374 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1382 * If the underlying (data/log/rt) device is readonly, there are some
1383 * operations that cannot proceed.
1386 xfs_dev_is_read_only(
1387 struct xfs_mount
*mp
,
1390 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1391 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1392 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1393 xfs_notice(mp
, "%s required on read-only device.", message
);
1394 xfs_notice(mp
, "write access unavailable, cannot proceed.");