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"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.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_dinode.h"
45 #include "xfs_sysfs.h"
49 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
51 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
53 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
56 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
57 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
60 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
61 static int xfs_uuid_table_size
;
62 static uuid_t
*xfs_uuid_table
;
64 extern struct kset
*xfs_kset
;
67 * See if the UUID is unique among mounted XFS filesystems.
68 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
74 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
77 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
80 if (uuid_is_nil(uuid
)) {
81 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
85 mutex_lock(&xfs_uuid_table_mutex
);
86 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
87 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
91 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
96 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
97 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
98 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
100 hole
= xfs_uuid_table_size
++;
102 xfs_uuid_table
[hole
] = *uuid
;
103 mutex_unlock(&xfs_uuid_table_mutex
);
108 mutex_unlock(&xfs_uuid_table_mutex
);
109 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
115 struct xfs_mount
*mp
)
117 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
120 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
123 mutex_lock(&xfs_uuid_table_mutex
);
124 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
125 if (uuid_is_nil(&xfs_uuid_table
[i
]))
127 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
129 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
132 ASSERT(i
< xfs_uuid_table_size
);
133 mutex_unlock(&xfs_uuid_table_mutex
);
139 struct rcu_head
*head
)
141 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
143 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
148 * Free up the per-ag resources associated with the mount structure.
155 struct xfs_perag
*pag
;
157 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
158 spin_lock(&mp
->m_perag_lock
);
159 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
160 spin_unlock(&mp
->m_perag_lock
);
162 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
163 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
168 * Check size of device based on the (data/realtime) block count.
169 * Note: this check is used by the growfs code as well as mount.
172 xfs_sb_validate_fsb_count(
176 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
177 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
179 /* Limited by ULONG_MAX of page cache index */
180 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
186 xfs_initialize_perag(
188 xfs_agnumber_t agcount
,
189 xfs_agnumber_t
*maxagi
)
191 xfs_agnumber_t index
;
192 xfs_agnumber_t first_initialised
= 0;
196 xfs_sb_t
*sbp
= &mp
->m_sb
;
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
);
210 if (!first_initialised
)
211 first_initialised
= index
;
213 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
216 pag
->pag_agno
= index
;
218 spin_lock_init(&pag
->pag_ici_lock
);
219 mutex_init(&pag
->pag_ici_reclaim_lock
);
220 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
221 spin_lock_init(&pag
->pag_buf_lock
);
222 pag
->pag_buf_tree
= RB_ROOT
;
224 if (radix_tree_preload(GFP_NOFS
))
227 spin_lock(&mp
->m_perag_lock
);
228 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
230 spin_unlock(&mp
->m_perag_lock
);
231 radix_tree_preload_end();
235 spin_unlock(&mp
->m_perag_lock
);
236 radix_tree_preload_end();
240 * If we mount with the inode64 option, or no inode overflows
241 * the legacy 32-bit address space clear the inode32 option.
243 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
244 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
246 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
247 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
249 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
251 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
)
252 index
= xfs_set_inode32(mp
, agcount
);
254 index
= xfs_set_inode64(mp
, agcount
);
262 for (; index
> first_initialised
; index
--) {
263 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
272 * Does the initial read of the superblock.
276 struct xfs_mount
*mp
,
279 unsigned int sector_size
;
281 struct xfs_sb
*sbp
= &mp
->m_sb
;
283 int loud
= !(flags
& XFS_MFSI_QUIET
);
284 const struct xfs_buf_ops
*buf_ops
;
286 ASSERT(mp
->m_sb_bp
== NULL
);
287 ASSERT(mp
->m_ddev_targp
!= NULL
);
290 * For the initial read, we must guess at the sector
291 * size based on the block device. It's enough to
292 * get the sb_sectsize out of the superblock and
293 * then reread with the proper length.
294 * We don't verify it yet, because it may not be complete.
296 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
300 * Allocate a (locked) buffer to hold the superblock.
301 * This will be kept around at all times to optimize
302 * access to the superblock.
305 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
306 BTOBB(sector_size
), 0, &bp
, buf_ops
);
309 xfs_warn(mp
, "SB validate failed with error %d.", error
);
310 /* bad CRC means corrupted metadata */
311 if (error
== -EFSBADCRC
)
312 error
= -EFSCORRUPTED
;
317 * Initialize the mount structure from the superblock.
319 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
322 * If we haven't validated the superblock, do so now before we try
323 * to check the sector size and reread the superblock appropriately.
325 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
327 xfs_warn(mp
, "Invalid superblock magic number");
333 * We must be able to do sector-sized and sector-aligned IO.
335 if (sector_size
> sbp
->sb_sectsize
) {
337 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
338 sector_size
, sbp
->sb_sectsize
);
343 if (buf_ops
== NULL
) {
345 * Re-read the superblock so the buffer is correctly sized,
346 * and properly verified.
349 sector_size
= sbp
->sb_sectsize
;
350 buf_ops
= loud
? &xfs_sb_buf_ops
: &xfs_sb_quiet_buf_ops
;
354 /* Initialize per-cpu counters */
355 xfs_icsb_reinit_counters(mp
);
357 /* no need to be quiet anymore, so reset the buf ops */
358 bp
->b_ops
= &xfs_sb_buf_ops
;
370 * Update alignment values based on mount options and sb values
373 xfs_update_alignment(xfs_mount_t
*mp
)
375 xfs_sb_t
*sbp
= &(mp
->m_sb
);
379 * If stripe unit and stripe width are not multiples
380 * of the fs blocksize turn off alignment.
382 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
383 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
385 "alignment check failed: sunit/swidth vs. blocksize(%d)",
390 * Convert the stripe unit and width to FSBs.
392 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
393 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
395 "alignment check failed: sunit/swidth vs. agsize(%d)",
398 } else if (mp
->m_dalign
) {
399 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
402 "alignment check failed: sunit(%d) less than bsize(%d)",
403 mp
->m_dalign
, sbp
->sb_blocksize
);
409 * Update superblock with new values
412 if (xfs_sb_version_hasdalign(sbp
)) {
413 if (sbp
->sb_unit
!= mp
->m_dalign
) {
414 sbp
->sb_unit
= mp
->m_dalign
;
415 mp
->m_update_flags
|= XFS_SB_UNIT
;
417 if (sbp
->sb_width
!= mp
->m_swidth
) {
418 sbp
->sb_width
= mp
->m_swidth
;
419 mp
->m_update_flags
|= XFS_SB_WIDTH
;
423 "cannot change alignment: superblock does not support data alignment");
426 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
427 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
428 mp
->m_dalign
= sbp
->sb_unit
;
429 mp
->m_swidth
= sbp
->sb_width
;
436 * Set the maximum inode count for this filesystem
439 xfs_set_maxicount(xfs_mount_t
*mp
)
441 xfs_sb_t
*sbp
= &(mp
->m_sb
);
444 if (sbp
->sb_imax_pct
) {
446 * Make sure the maximum inode count is a multiple
447 * of the units we allocate inodes in.
449 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
451 do_div(icount
, mp
->m_ialloc_blks
);
452 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
460 * Set the default minimum read and write sizes unless
461 * already specified in a mount option.
462 * We use smaller I/O sizes when the file system
463 * is being used for NFS service (wsync mount option).
466 xfs_set_rw_sizes(xfs_mount_t
*mp
)
468 xfs_sb_t
*sbp
= &(mp
->m_sb
);
469 int readio_log
, writeio_log
;
471 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
472 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
473 readio_log
= XFS_WSYNC_READIO_LOG
;
474 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
476 readio_log
= XFS_READIO_LOG_LARGE
;
477 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
480 readio_log
= mp
->m_readio_log
;
481 writeio_log
= mp
->m_writeio_log
;
484 if (sbp
->sb_blocklog
> readio_log
) {
485 mp
->m_readio_log
= sbp
->sb_blocklog
;
487 mp
->m_readio_log
= readio_log
;
489 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
490 if (sbp
->sb_blocklog
> writeio_log
) {
491 mp
->m_writeio_log
= sbp
->sb_blocklog
;
493 mp
->m_writeio_log
= writeio_log
;
495 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
499 * precalculate the low space thresholds for dynamic speculative preallocation.
502 xfs_set_low_space_thresholds(
503 struct xfs_mount
*mp
)
507 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
508 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
511 mp
->m_low_space
[i
] = space
* (i
+ 1);
517 * Set whether we're using inode alignment.
520 xfs_set_inoalignment(xfs_mount_t
*mp
)
522 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
523 mp
->m_sb
.sb_inoalignmt
>=
524 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
525 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
527 mp
->m_inoalign_mask
= 0;
529 * If we are using stripe alignment, check whether
530 * the stripe unit is a multiple of the inode alignment
532 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
533 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
534 mp
->m_sinoalign
= mp
->m_dalign
;
540 * Check that the data (and log if separate) is an ok size.
544 struct xfs_mount
*mp
)
550 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
551 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
552 xfs_warn(mp
, "filesystem size mismatch detected");
555 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
556 d
- XFS_FSS_TO_BB(mp
, 1),
557 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, NULL
);
559 xfs_warn(mp
, "last sector read failed");
564 if (mp
->m_logdev_targp
== mp
->m_ddev_targp
)
567 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
568 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
569 xfs_warn(mp
, "log size mismatch detected");
572 error
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
573 d
- XFS_FSB_TO_BB(mp
, 1),
574 XFS_FSB_TO_BB(mp
, 1), 0, &bp
, NULL
);
576 xfs_warn(mp
, "log device read failed");
584 * Clear the quotaflags in memory and in the superblock.
587 xfs_mount_reset_sbqflags(
588 struct xfs_mount
*mp
)
591 struct xfs_trans
*tp
;
596 * It is OK to look at sb_qflags here in mount path,
599 if (mp
->m_sb
.sb_qflags
== 0)
601 spin_lock(&mp
->m_sb_lock
);
602 mp
->m_sb
.sb_qflags
= 0;
603 spin_unlock(&mp
->m_sb_lock
);
606 * If the fs is readonly, let the incore superblock run
607 * with quotas off but don't flush the update out to disk
609 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
612 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
613 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_qm_sbchange
, 0, 0);
615 xfs_trans_cancel(tp
, 0);
616 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
620 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
621 return xfs_trans_commit(tp
, 0);
625 xfs_default_resblks(xfs_mount_t
*mp
)
630 * We default to 5% or 8192 fsbs of space reserved, whichever is
631 * smaller. This is intended to cover concurrent allocation
632 * transactions when we initially hit enospc. These each require a 4
633 * block reservation. Hence by default we cover roughly 2000 concurrent
634 * allocation reservations.
636 resblks
= mp
->m_sb
.sb_dblocks
;
638 resblks
= min_t(__uint64_t
, resblks
, 8192);
643 * This function does the following on an initial mount of a file system:
644 * - reads the superblock from disk and init the mount struct
645 * - if we're a 32-bit kernel, do a size check on the superblock
646 * so we don't mount terabyte filesystems
647 * - init mount struct realtime fields
648 * - allocate inode hash table for fs
649 * - init directory manager
650 * - perform recovery and init the log manager
656 xfs_sb_t
*sbp
= &(mp
->m_sb
);
663 xfs_sb_mount_common(mp
, sbp
);
666 * Check for a mismatched features2 values. Older kernels
667 * read & wrote into the wrong sb offset for sb_features2
668 * on some platforms due to xfs_sb_t not being 64bit size aligned
669 * when sb_features2 was added, which made older superblock
670 * reading/writing routines swap it as a 64-bit value.
672 * For backwards compatibility, we make both slots equal.
674 * If we detect a mismatched field, we OR the set bits into the
675 * existing features2 field in case it has already been modified; we
676 * don't want to lose any features. We then update the bad location
677 * with the ORed value so that older kernels will see any features2
678 * flags, and mark the two fields as needing updates once the
679 * transaction subsystem is online.
681 if (xfs_sb_has_mismatched_features2(sbp
)) {
682 xfs_warn(mp
, "correcting sb_features alignment problem");
683 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
684 sbp
->sb_bad_features2
= sbp
->sb_features2
;
685 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
688 * Re-check for ATTR2 in case it was found in bad_features2
691 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
692 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
693 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
696 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
697 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
698 xfs_sb_version_removeattr2(&mp
->m_sb
);
699 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
701 /* update sb_versionnum for the clearing of the morebits */
702 if (!sbp
->sb_features2
)
703 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
706 /* always use v2 inodes by default now */
707 if (!(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_NLINKBIT
)) {
708 mp
->m_sb
.sb_versionnum
|= XFS_SB_VERSION_NLINKBIT
;
709 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
713 * Check if sb_agblocks is aligned at stripe boundary
714 * If sb_agblocks is NOT aligned turn off m_dalign since
715 * allocator alignment is within an ag, therefore ag has
716 * to be aligned at stripe boundary.
718 error
= xfs_update_alignment(mp
);
722 xfs_alloc_compute_maxlevels(mp
);
723 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
724 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
725 xfs_ialloc_compute_maxlevels(mp
);
727 xfs_set_maxicount(mp
);
729 mp
->m_kobj
.kobject
.kset
= xfs_kset
;
730 error
= xfs_sysfs_init(&mp
->m_kobj
, &xfs_mp_ktype
, NULL
, mp
->m_fsname
);
734 error
= xfs_uuid_mount(mp
);
736 goto out_remove_sysfs
;
739 * Set the minimum read and write sizes
741 xfs_set_rw_sizes(mp
);
743 /* set the low space thresholds for dynamic preallocation */
744 xfs_set_low_space_thresholds(mp
);
747 * Set the inode cluster size.
748 * This may still be overridden by the file system
749 * block size if it is larger than the chosen cluster size.
751 * For v5 filesystems, scale the cluster size with the inode size to
752 * keep a constant ratio of inode per cluster buffer, but only if mkfs
753 * has set the inode alignment value appropriately for larger cluster
756 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
757 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
758 int new_size
= mp
->m_inode_cluster_size
;
760 new_size
*= mp
->m_sb
.sb_inodesize
/ XFS_DINODE_MIN_SIZE
;
761 if (mp
->m_sb
.sb_inoalignmt
>= XFS_B_TO_FSBT(mp
, new_size
))
762 mp
->m_inode_cluster_size
= new_size
;
766 * Set inode alignment fields
768 xfs_set_inoalignment(mp
);
771 * Check that the data (and log if separate) is an ok size.
773 error
= xfs_check_sizes(mp
);
775 goto out_remove_uuid
;
778 * Initialize realtime fields in the mount structure
780 error
= xfs_rtmount_init(mp
);
782 xfs_warn(mp
, "RT mount failed");
783 goto out_remove_uuid
;
787 * Copies the low order bits of the timestamp and the randomly
788 * set "sequence" number out of a UUID.
790 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
792 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
794 error
= xfs_da_mount(mp
);
796 xfs_warn(mp
, "Failed dir/attr init: %d", error
);
797 goto out_remove_uuid
;
801 * Initialize the precomputed transaction reservations values.
806 * Allocate and initialize the per-ag data.
808 spin_lock_init(&mp
->m_perag_lock
);
809 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
810 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
812 xfs_warn(mp
, "Failed per-ag init: %d", error
);
816 if (!sbp
->sb_logblocks
) {
817 xfs_warn(mp
, "no log defined");
818 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
819 error
= -EFSCORRUPTED
;
824 * log's mount-time initialization. Perform 1st part recovery if needed
826 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
827 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
828 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
830 xfs_warn(mp
, "log mount failed");
835 * Now the log is mounted, we know if it was an unclean shutdown or
836 * not. If it was, with the first phase of recovery has completed, we
837 * have consistent AG blocks on disk. We have not recovered EFIs yet,
838 * but they are recovered transactionally in the second recovery phase
841 * Hence we can safely re-initialise incore superblock counters from
842 * the per-ag data. These may not be correct if the filesystem was not
843 * cleanly unmounted, so we need to wait for recovery to finish before
846 * If the filesystem was cleanly unmounted, then we can trust the
847 * values in the superblock to be correct and we don't need to do
850 * If we are currently making the filesystem, the initialisation will
851 * fail as the perag data is in an undefined state.
853 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
854 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
855 !mp
->m_sb
.sb_inprogress
) {
856 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
858 goto out_log_dealloc
;
862 * Get and sanity-check the root inode.
863 * Save the pointer to it in the mount structure.
865 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
867 xfs_warn(mp
, "failed to read root inode");
868 goto out_log_dealloc
;
873 if (unlikely(!S_ISDIR(rip
->i_d
.di_mode
))) {
874 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
875 (unsigned long long)rip
->i_ino
);
876 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
877 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
879 error
= -EFSCORRUPTED
;
882 mp
->m_rootip
= rip
; /* save it */
884 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
887 * Initialize realtime inode pointers in the mount structure
889 error
= xfs_rtmount_inodes(mp
);
892 * Free up the root inode.
894 xfs_warn(mp
, "failed to read RT inodes");
899 * If this is a read-only mount defer the superblock updates until
900 * the next remount into writeable mode. Otherwise we would never
901 * perform the update e.g. for the root filesystem.
903 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
904 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
906 xfs_warn(mp
, "failed to write sb changes");
912 * Initialise the XFS quota management subsystem for this mount
914 if (XFS_IS_QUOTA_RUNNING(mp
)) {
915 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
919 ASSERT(!XFS_IS_QUOTA_ON(mp
));
922 * If a file system had quotas running earlier, but decided to
923 * mount without -o uquota/pquota/gquota options, revoke the
924 * quotachecked license.
926 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
927 xfs_notice(mp
, "resetting quota flags");
928 error
= xfs_mount_reset_sbqflags(mp
);
935 * Finish recovering the file system. This part needed to be
936 * delayed until after the root and real-time bitmap inodes
937 * were consistently read in.
939 error
= xfs_log_mount_finish(mp
);
941 xfs_warn(mp
, "log mount finish failed");
946 * Complete the quota initialisation, post-log-replay component.
949 ASSERT(mp
->m_qflags
== 0);
950 mp
->m_qflags
= quotaflags
;
952 xfs_qm_mount_quotas(mp
);
956 * Now we are mounted, reserve a small amount of unused space for
957 * privileged transactions. This is needed so that transaction
958 * space required for critical operations can dip into this pool
959 * when at ENOSPC. This is needed for operations like create with
960 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
961 * are not allowed to use this reserved space.
963 * This may drive us straight to ENOSPC on mount, but that implies
964 * we were already there on the last unmount. Warn if this occurs.
966 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
967 resblks
= xfs_default_resblks(mp
);
968 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
971 "Unable to allocate reserve blocks. Continuing without reserve pool.");
977 xfs_rtunmount_inodes(mp
);
983 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
984 xfs_wait_buftarg(mp
->m_logdev_targp
);
985 xfs_wait_buftarg(mp
->m_ddev_targp
);
991 xfs_uuid_unmount(mp
);
993 xfs_sysfs_del(&mp
->m_kobj
);
999 * This flushes out the inodes,dquots and the superblock, unmounts the
1000 * log and makes sure that incore structures are freed.
1004 struct xfs_mount
*mp
)
1009 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
1011 xfs_qm_unmount_quotas(mp
);
1012 xfs_rtunmount_inodes(mp
);
1013 IRELE(mp
->m_rootip
);
1016 * We can potentially deadlock here if we have an inode cluster
1017 * that has been freed has its buffer still pinned in memory because
1018 * the transaction is still sitting in a iclog. The stale inodes
1019 * on that buffer will have their flush locks held until the
1020 * transaction hits the disk and the callbacks run. the inode
1021 * flush takes the flush lock unconditionally and with nothing to
1022 * push out the iclog we will never get that unlocked. hence we
1023 * need to force the log first.
1025 xfs_log_force(mp
, XFS_LOG_SYNC
);
1028 * Flush all pending changes from the AIL.
1030 xfs_ail_push_all_sync(mp
->m_ail
);
1033 * And reclaim all inodes. At this point there should be no dirty
1034 * inodes and none should be pinned or locked, but use synchronous
1035 * reclaim just to be sure. We can stop background inode reclaim
1036 * here as well if it is still running.
1038 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1039 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1044 * Unreserve any blocks we have so that when we unmount we don't account
1045 * the reserved free space as used. This is really only necessary for
1046 * lazy superblock counting because it trusts the incore superblock
1047 * counters to be absolutely correct on clean unmount.
1049 * We don't bother correcting this elsewhere for lazy superblock
1050 * counting because on mount of an unclean filesystem we reconstruct the
1051 * correct counter value and this is irrelevant.
1053 * For non-lazy counter filesystems, this doesn't matter at all because
1054 * we only every apply deltas to the superblock and hence the incore
1055 * value does not matter....
1058 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1060 xfs_warn(mp
, "Unable to free reserved block pool. "
1061 "Freespace may not be correct on next mount.");
1063 error
= xfs_log_sbcount(mp
);
1065 xfs_warn(mp
, "Unable to update superblock counters. "
1066 "Freespace may not be correct on next mount.");
1068 xfs_log_unmount(mp
);
1070 xfs_uuid_unmount(mp
);
1073 xfs_errortag_clearall(mp
, 0);
1077 xfs_sysfs_del(&mp
->m_kobj
);
1081 xfs_fs_writable(xfs_mount_t
*mp
)
1083 return !(mp
->m_super
->s_writers
.frozen
|| XFS_FORCED_SHUTDOWN(mp
) ||
1084 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1090 * Sync the superblock counters to disk.
1092 * Note this code can be called during the process of freezing, so
1093 * we may need to use the transaction allocator which does not
1094 * block when the transaction subsystem is in its frozen state.
1097 xfs_log_sbcount(xfs_mount_t
*mp
)
1102 if (!xfs_fs_writable(mp
))
1105 xfs_icsb_sync_counters(mp
, 0);
1108 * we don't need to do this if we are updating the superblock
1109 * counters on every modification.
1111 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1114 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1115 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_sb
, 0, 0);
1117 xfs_trans_cancel(tp
, 0);
1121 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1122 xfs_trans_set_sync(tp
);
1123 error
= xfs_trans_commit(tp
, 0);
1128 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1129 * a delta to a specified field in the in-core superblock. Simply
1130 * switch on the field indicated and apply the delta to that field.
1131 * Fields are not allowed to dip below zero, so if the delta would
1132 * do this do not apply it and return EINVAL.
1134 * The m_sb_lock must be held when this routine is called.
1137 xfs_mod_incore_sb_unlocked(
1139 xfs_sb_field_t field
,
1143 int scounter
; /* short counter for 32 bit fields */
1144 long long lcounter
; /* long counter for 64 bit fields */
1145 long long res_used
, rem
;
1148 * With the in-core superblock spin lock held, switch
1149 * on the indicated field. Apply the delta to the
1150 * proper field. If the fields value would dip below
1151 * 0, then do not apply the delta and return EINVAL.
1154 case XFS_SBS_ICOUNT
:
1155 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1161 mp
->m_sb
.sb_icount
= lcounter
;
1164 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1170 mp
->m_sb
.sb_ifree
= lcounter
;
1172 case XFS_SBS_FDBLOCKS
:
1173 lcounter
= (long long)
1174 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1175 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1177 if (delta
> 0) { /* Putting blocks back */
1178 if (res_used
> delta
) {
1179 mp
->m_resblks_avail
+= delta
;
1181 rem
= delta
- res_used
;
1182 mp
->m_resblks_avail
= mp
->m_resblks
;
1185 } else { /* Taking blocks away */
1187 if (lcounter
>= 0) {
1188 mp
->m_sb
.sb_fdblocks
= lcounter
+
1189 XFS_ALLOC_SET_ASIDE(mp
);
1194 * We are out of blocks, use any available reserved
1195 * blocks if were allowed to.
1200 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1201 if (lcounter
>= 0) {
1202 mp
->m_resblks_avail
= lcounter
;
1205 printk_once(KERN_WARNING
1206 "Filesystem \"%s\": reserve blocks depleted! "
1207 "Consider increasing reserve pool size.",
1212 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1214 case XFS_SBS_FREXTENTS
:
1215 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1220 mp
->m_sb
.sb_frextents
= lcounter
;
1222 case XFS_SBS_DBLOCKS
:
1223 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1229 mp
->m_sb
.sb_dblocks
= lcounter
;
1231 case XFS_SBS_AGCOUNT
:
1232 scounter
= mp
->m_sb
.sb_agcount
;
1238 mp
->m_sb
.sb_agcount
= scounter
;
1240 case XFS_SBS_IMAX_PCT
:
1241 scounter
= mp
->m_sb
.sb_imax_pct
;
1247 mp
->m_sb
.sb_imax_pct
= scounter
;
1249 case XFS_SBS_REXTSIZE
:
1250 scounter
= mp
->m_sb
.sb_rextsize
;
1256 mp
->m_sb
.sb_rextsize
= scounter
;
1258 case XFS_SBS_RBMBLOCKS
:
1259 scounter
= mp
->m_sb
.sb_rbmblocks
;
1265 mp
->m_sb
.sb_rbmblocks
= scounter
;
1267 case XFS_SBS_RBLOCKS
:
1268 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1274 mp
->m_sb
.sb_rblocks
= lcounter
;
1276 case XFS_SBS_REXTENTS
:
1277 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1283 mp
->m_sb
.sb_rextents
= lcounter
;
1285 case XFS_SBS_REXTSLOG
:
1286 scounter
= mp
->m_sb
.sb_rextslog
;
1292 mp
->m_sb
.sb_rextslog
= scounter
;
1301 * xfs_mod_incore_sb() is used to change a field in the in-core
1302 * superblock structure by the specified delta. This modification
1303 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1304 * routine to do the work.
1308 struct xfs_mount
*mp
,
1309 xfs_sb_field_t field
,
1315 #ifdef HAVE_PERCPU_SB
1316 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1318 spin_lock(&mp
->m_sb_lock
);
1319 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1320 spin_unlock(&mp
->m_sb_lock
);
1326 * Change more than one field in the in-core superblock structure at a time.
1328 * The fields and changes to those fields are specified in the array of
1329 * xfs_mod_sb structures passed in. Either all of the specified deltas
1330 * will be applied or none of them will. If any modified field dips below 0,
1331 * then all modifications will be backed out and EINVAL will be returned.
1333 * Note that this function may not be used for the superblock values that
1334 * are tracked with the in-memory per-cpu counters - a direct call to
1335 * xfs_icsb_modify_counters is required for these.
1338 xfs_mod_incore_sb_batch(
1339 struct xfs_mount
*mp
,
1348 * Loop through the array of mod structures and apply each individually.
1349 * If any fail, then back out all those which have already been applied.
1350 * Do all of this within the scope of the m_sb_lock so that all of the
1351 * changes will be atomic.
1353 spin_lock(&mp
->m_sb_lock
);
1354 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1355 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1356 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1358 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1359 msbp
->msb_delta
, rsvd
);
1363 spin_unlock(&mp
->m_sb_lock
);
1367 while (--msbp
>= msb
) {
1368 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1369 -msbp
->msb_delta
, rsvd
);
1372 spin_unlock(&mp
->m_sb_lock
);
1377 * xfs_getsb() is called to obtain the buffer for the superblock.
1378 * The buffer is returned locked and read in from disk.
1379 * The buffer should be released with a call to xfs_brelse().
1381 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1382 * the superblock buffer if it can be locked without sleeping.
1383 * If it can't then we'll return NULL.
1387 struct xfs_mount
*mp
,
1390 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1392 if (!xfs_buf_trylock(bp
)) {
1393 if (flags
& XBF_TRYLOCK
)
1399 ASSERT(XFS_BUF_ISDONE(bp
));
1404 * Used to free the superblock along various error paths.
1408 struct xfs_mount
*mp
)
1410 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1418 * Used to log changes to the superblock unit and width fields which could
1419 * be altered by the mount options, as well as any potential sb_features2
1420 * fixup. Only the first superblock is updated.
1430 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1431 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1432 XFS_SB_VERSIONNUM
));
1434 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1435 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_sb
, 0, 0);
1437 xfs_trans_cancel(tp
, 0);
1440 xfs_mod_sb(tp
, fields
);
1441 error
= xfs_trans_commit(tp
, 0);
1446 * If the underlying (data/log/rt) device is readonly, there are some
1447 * operations that cannot proceed.
1450 xfs_dev_is_read_only(
1451 struct xfs_mount
*mp
,
1454 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1455 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1456 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1457 xfs_notice(mp
, "%s required on read-only device.", message
);
1458 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1464 #ifdef HAVE_PERCPU_SB
1466 * Per-cpu incore superblock counters
1468 * Simple concept, difficult implementation
1470 * Basically, replace the incore superblock counters with a distributed per cpu
1471 * counter for contended fields (e.g. free block count).
1473 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1474 * hence needs to be accurately read when we are running low on space. Hence
1475 * there is a method to enable and disable the per-cpu counters based on how
1476 * much "stuff" is available in them.
1478 * Basically, a counter is enabled if there is enough free resource to justify
1479 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1480 * ENOSPC), then we disable the counters to synchronise all callers and
1481 * re-distribute the available resources.
1483 * If, once we redistributed the available resources, we still get a failure,
1484 * we disable the per-cpu counter and go through the slow path.
1486 * The slow path is the current xfs_mod_incore_sb() function. This means that
1487 * when we disable a per-cpu counter, we need to drain its resources back to
1488 * the global superblock. We do this after disabling the counter to prevent
1489 * more threads from queueing up on the counter.
1491 * Essentially, this means that we still need a lock in the fast path to enable
1492 * synchronisation between the global counters and the per-cpu counters. This
1493 * is not a problem because the lock will be local to a CPU almost all the time
1494 * and have little contention except when we get to ENOSPC conditions.
1496 * Basically, this lock becomes a barrier that enables us to lock out the fast
1497 * path while we do things like enabling and disabling counters and
1498 * synchronising the counters.
1502 * 1. m_sb_lock before picking up per-cpu locks
1503 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1504 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1505 * 4. modifying per-cpu counters requires holding per-cpu lock
1506 * 5. modifying global counters requires holding m_sb_lock
1507 * 6. enabling or disabling a counter requires holding the m_sb_lock
1508 * and _none_ of the per-cpu locks.
1510 * Disabled counters are only ever re-enabled by a balance operation
1511 * that results in more free resources per CPU than a given threshold.
1512 * To ensure counters don't remain disabled, they are rebalanced when
1513 * the global resource goes above a higher threshold (i.e. some hysteresis
1514 * is present to prevent thrashing).
1517 #ifdef CONFIG_HOTPLUG_CPU
1519 * hot-plug CPU notifier support.
1521 * We need a notifier per filesystem as we need to be able to identify
1522 * the filesystem to balance the counters out. This is achieved by
1523 * having a notifier block embedded in the xfs_mount_t and doing pointer
1524 * magic to get the mount pointer from the notifier block address.
1527 xfs_icsb_cpu_notify(
1528 struct notifier_block
*nfb
,
1529 unsigned long action
,
1532 xfs_icsb_cnts_t
*cntp
;
1535 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1536 cntp
= (xfs_icsb_cnts_t
*)
1537 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1539 case CPU_UP_PREPARE
:
1540 case CPU_UP_PREPARE_FROZEN
:
1541 /* Easy Case - initialize the area and locks, and
1542 * then rebalance when online does everything else for us. */
1543 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1546 case CPU_ONLINE_FROZEN
:
1548 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1549 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1550 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1551 xfs_icsb_unlock(mp
);
1554 case CPU_DEAD_FROZEN
:
1555 /* Disable all the counters, then fold the dead cpu's
1556 * count into the total on the global superblock and
1557 * re-enable the counters. */
1559 spin_lock(&mp
->m_sb_lock
);
1560 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1561 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1562 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1564 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1565 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1566 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1568 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1570 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
1571 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
1572 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
1573 spin_unlock(&mp
->m_sb_lock
);
1574 xfs_icsb_unlock(mp
);
1580 #endif /* CONFIG_HOTPLUG_CPU */
1583 xfs_icsb_init_counters(
1586 xfs_icsb_cnts_t
*cntp
;
1589 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
1590 if (mp
->m_sb_cnts
== NULL
)
1593 for_each_online_cpu(i
) {
1594 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1595 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1598 mutex_init(&mp
->m_icsb_mutex
);
1601 * start with all counters disabled so that the
1602 * initial balance kicks us off correctly
1604 mp
->m_icsb_counters
= -1;
1606 #ifdef CONFIG_HOTPLUG_CPU
1607 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
1608 mp
->m_icsb_notifier
.priority
= 0;
1609 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
1610 #endif /* CONFIG_HOTPLUG_CPU */
1616 xfs_icsb_reinit_counters(
1621 * start with all counters disabled so that the
1622 * initial balance kicks us off correctly
1624 mp
->m_icsb_counters
= -1;
1625 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1626 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1627 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1628 xfs_icsb_unlock(mp
);
1632 xfs_icsb_destroy_counters(
1635 if (mp
->m_sb_cnts
) {
1636 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
1637 free_percpu(mp
->m_sb_cnts
);
1639 mutex_destroy(&mp
->m_icsb_mutex
);
1644 xfs_icsb_cnts_t
*icsbp
)
1646 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
1652 xfs_icsb_unlock_cntr(
1653 xfs_icsb_cnts_t
*icsbp
)
1655 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
1660 xfs_icsb_lock_all_counters(
1663 xfs_icsb_cnts_t
*cntp
;
1666 for_each_online_cpu(i
) {
1667 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1668 xfs_icsb_lock_cntr(cntp
);
1673 xfs_icsb_unlock_all_counters(
1676 xfs_icsb_cnts_t
*cntp
;
1679 for_each_online_cpu(i
) {
1680 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1681 xfs_icsb_unlock_cntr(cntp
);
1688 xfs_icsb_cnts_t
*cnt
,
1691 xfs_icsb_cnts_t
*cntp
;
1694 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
1696 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1697 xfs_icsb_lock_all_counters(mp
);
1699 for_each_online_cpu(i
) {
1700 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1701 cnt
->icsb_icount
+= cntp
->icsb_icount
;
1702 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
1703 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
1706 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1707 xfs_icsb_unlock_all_counters(mp
);
1711 xfs_icsb_counter_disabled(
1713 xfs_sb_field_t field
)
1715 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1716 return test_bit(field
, &mp
->m_icsb_counters
);
1720 xfs_icsb_disable_counter(
1722 xfs_sb_field_t field
)
1724 xfs_icsb_cnts_t cnt
;
1726 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1729 * If we are already disabled, then there is nothing to do
1730 * here. We check before locking all the counters to avoid
1731 * the expensive lock operation when being called in the
1732 * slow path and the counter is already disabled. This is
1733 * safe because the only time we set or clear this state is under
1736 if (xfs_icsb_counter_disabled(mp
, field
))
1739 xfs_icsb_lock_all_counters(mp
);
1740 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
1741 /* drain back to superblock */
1743 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
1745 case XFS_SBS_ICOUNT
:
1746 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1749 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1751 case XFS_SBS_FDBLOCKS
:
1752 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1759 xfs_icsb_unlock_all_counters(mp
);
1763 xfs_icsb_enable_counter(
1765 xfs_sb_field_t field
,
1769 xfs_icsb_cnts_t
*cntp
;
1772 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1774 xfs_icsb_lock_all_counters(mp
);
1775 for_each_online_cpu(i
) {
1776 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
1778 case XFS_SBS_ICOUNT
:
1779 cntp
->icsb_icount
= count
+ resid
;
1782 cntp
->icsb_ifree
= count
+ resid
;
1784 case XFS_SBS_FDBLOCKS
:
1785 cntp
->icsb_fdblocks
= count
+ resid
;
1793 clear_bit(field
, &mp
->m_icsb_counters
);
1794 xfs_icsb_unlock_all_counters(mp
);
1798 xfs_icsb_sync_counters_locked(
1802 xfs_icsb_cnts_t cnt
;
1804 xfs_icsb_count(mp
, &cnt
, flags
);
1806 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
1807 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1808 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
1809 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1810 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
1811 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1815 * Accurate update of per-cpu counters to incore superblock
1818 xfs_icsb_sync_counters(
1822 spin_lock(&mp
->m_sb_lock
);
1823 xfs_icsb_sync_counters_locked(mp
, flags
);
1824 spin_unlock(&mp
->m_sb_lock
);
1828 * Balance and enable/disable counters as necessary.
1830 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1831 * chosen to be the same number as single on disk allocation chunk per CPU, and
1832 * free blocks is something far enough zero that we aren't going thrash when we
1833 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1834 * prevent looping endlessly when xfs_alloc_space asks for more than will
1835 * be distributed to a single CPU but each CPU has enough blocks to be
1838 * Note that we can be called when counters are already disabled.
1839 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1840 * prevent locking every per-cpu counter needlessly.
1843 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1844 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1845 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1847 xfs_icsb_balance_counter_locked(
1849 xfs_sb_field_t field
,
1852 uint64_t count
, resid
;
1853 int weight
= num_online_cpus();
1854 uint64_t min
= (uint64_t)min_per_cpu
;
1856 /* disable counter and sync counter */
1857 xfs_icsb_disable_counter(mp
, field
);
1859 /* update counters - first CPU gets residual*/
1861 case XFS_SBS_ICOUNT
:
1862 count
= mp
->m_sb
.sb_icount
;
1863 resid
= do_div(count
, weight
);
1864 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
1868 count
= mp
->m_sb
.sb_ifree
;
1869 resid
= do_div(count
, weight
);
1870 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
1873 case XFS_SBS_FDBLOCKS
:
1874 count
= mp
->m_sb
.sb_fdblocks
;
1875 resid
= do_div(count
, weight
);
1876 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
1881 count
= resid
= 0; /* quiet, gcc */
1885 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
1889 xfs_icsb_balance_counter(
1891 xfs_sb_field_t fields
,
1894 spin_lock(&mp
->m_sb_lock
);
1895 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
1896 spin_unlock(&mp
->m_sb_lock
);
1900 xfs_icsb_modify_counters(
1902 xfs_sb_field_t field
,
1906 xfs_icsb_cnts_t
*icsbp
;
1907 long long lcounter
; /* long counter for 64 bit fields */
1913 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
1916 * if the counter is disabled, go to slow path
1918 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
1920 xfs_icsb_lock_cntr(icsbp
);
1921 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
1922 xfs_icsb_unlock_cntr(icsbp
);
1927 case XFS_SBS_ICOUNT
:
1928 lcounter
= icsbp
->icsb_icount
;
1930 if (unlikely(lcounter
< 0))
1931 goto balance_counter
;
1932 icsbp
->icsb_icount
= lcounter
;
1936 lcounter
= icsbp
->icsb_ifree
;
1938 if (unlikely(lcounter
< 0))
1939 goto balance_counter
;
1940 icsbp
->icsb_ifree
= lcounter
;
1943 case XFS_SBS_FDBLOCKS
:
1944 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
1946 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1948 if (unlikely(lcounter
< 0))
1949 goto balance_counter
;
1950 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1956 xfs_icsb_unlock_cntr(icsbp
);
1964 * serialise with a mutex so we don't burn lots of cpu on
1965 * the superblock lock. We still need to hold the superblock
1966 * lock, however, when we modify the global structures.
1971 * Now running atomically.
1973 * If the counter is enabled, someone has beaten us to rebalancing.
1974 * Drop the lock and try again in the fast path....
1976 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
1977 xfs_icsb_unlock(mp
);
1982 * The counter is currently disabled. Because we are
1983 * running atomically here, we know a rebalance cannot
1984 * be in progress. Hence we can go straight to operating
1985 * on the global superblock. We do not call xfs_mod_incore_sb()
1986 * here even though we need to get the m_sb_lock. Doing so
1987 * will cause us to re-enter this function and deadlock.
1988 * Hence we get the m_sb_lock ourselves and then call
1989 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1990 * directly on the global counters.
1992 spin_lock(&mp
->m_sb_lock
);
1993 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1994 spin_unlock(&mp
->m_sb_lock
);
1997 * Now that we've modified the global superblock, we
1998 * may be able to re-enable the distributed counters
1999 * (e.g. lots of space just got freed). After that
2003 xfs_icsb_balance_counter(mp
, field
, 0);
2004 xfs_icsb_unlock(mp
);
2008 xfs_icsb_unlock_cntr(icsbp
);
2012 * We may have multiple threads here if multiple per-cpu
2013 * counters run dry at the same time. This will mean we can
2014 * do more balances than strictly necessary but it is not
2015 * the common slowpath case.
2020 * running atomically.
2022 * This will leave the counter in the correct state for future
2023 * accesses. After the rebalance, we simply try again and our retry
2024 * will either succeed through the fast path or slow path without
2025 * another balance operation being required.
2027 xfs_icsb_balance_counter(mp
, field
, delta
);
2028 xfs_icsb_unlock(mp
);