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_format.h"
21 #include "xfs_shared.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_dir2_format.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_alloc.h"
41 #include "xfs_rtalloc.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_fsops.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_cksum.h"
49 #include "xfs_buf_item.h"
53 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
55 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
57 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
60 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
61 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
64 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
65 static int xfs_uuid_table_size
;
66 static uuid_t
*xfs_uuid_table
;
69 * See if the UUID is unique among mounted XFS filesystems.
70 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
76 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
79 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
82 if (uuid_is_nil(uuid
)) {
83 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
84 return XFS_ERROR(EINVAL
);
87 mutex_lock(&xfs_uuid_table_mutex
);
88 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
89 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
93 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
98 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
99 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
100 xfs_uuid_table_size
* 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
);
112 return XFS_ERROR(EINVAL
);
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 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
170 * Check size of device based on the (data/realtime) block count.
171 * Note: this check is used by the growfs code as well as mount.
174 xfs_sb_validate_fsb_count(
178 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
179 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
181 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
182 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
184 #else /* Limited by UINT_MAX of sectors */
185 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
192 xfs_initialize_perag(
194 xfs_agnumber_t agcount
,
195 xfs_agnumber_t
*maxagi
)
197 xfs_agnumber_t index
;
198 xfs_agnumber_t first_initialised
= 0;
202 xfs_sb_t
*sbp
= &mp
->m_sb
;
206 * Walk the current per-ag tree so we don't try to initialise AGs
207 * that already exist (growfs case). Allocate and insert all the
208 * AGs we don't find ready for initialisation.
210 for (index
= 0; index
< agcount
; index
++) {
211 pag
= xfs_perag_get(mp
, index
);
216 if (!first_initialised
)
217 first_initialised
= index
;
219 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
222 pag
->pag_agno
= index
;
224 spin_lock_init(&pag
->pag_ici_lock
);
225 mutex_init(&pag
->pag_ici_reclaim_lock
);
226 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
227 spin_lock_init(&pag
->pag_buf_lock
);
228 pag
->pag_buf_tree
= RB_ROOT
;
230 if (radix_tree_preload(GFP_NOFS
))
233 spin_lock(&mp
->m_perag_lock
);
234 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
236 spin_unlock(&mp
->m_perag_lock
);
237 radix_tree_preload_end();
241 spin_unlock(&mp
->m_perag_lock
);
242 radix_tree_preload_end();
246 * If we mount with the inode64 option, or no inode overflows
247 * the legacy 32-bit address space clear the inode32 option.
249 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
250 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
252 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
253 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
255 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
257 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
)
258 index
= xfs_set_inode32(mp
);
260 index
= xfs_set_inode64(mp
);
268 for (; index
> first_initialised
; index
--) {
269 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
278 * Does the initial read of the superblock.
282 struct xfs_mount
*mp
,
285 unsigned int sector_size
;
287 struct xfs_sb
*sbp
= &mp
->m_sb
;
289 int loud
= !(flags
& XFS_MFSI_QUIET
);
291 ASSERT(mp
->m_sb_bp
== NULL
);
292 ASSERT(mp
->m_ddev_targp
!= NULL
);
295 * Allocate a (locked) buffer to hold the superblock.
296 * This will be kept around at all times to optimize
297 * access to the superblock.
299 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
302 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
303 BTOBB(sector_size
), 0,
304 loud
? &xfs_sb_buf_ops
305 : &xfs_sb_quiet_buf_ops
);
308 xfs_warn(mp
, "SB buffer read failed");
314 xfs_warn(mp
, "SB validate failed with error %d.", error
);
319 * Initialize the mount structure from the superblock.
321 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
322 xfs_sb_quota_from_disk(&mp
->m_sb
);
325 * We must be able to do sector-sized and sector-aligned IO.
327 if (sector_size
> sbp
->sb_sectsize
) {
329 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
330 sector_size
, sbp
->sb_sectsize
);
336 * If device sector size is smaller than the superblock size,
337 * re-read the superblock so the buffer is correctly sized.
339 if (sector_size
< sbp
->sb_sectsize
) {
341 sector_size
= sbp
->sb_sectsize
;
345 /* Initialize per-cpu counters */
346 xfs_icsb_reinit_counters(mp
);
348 /* no need to be quiet anymore, so reset the buf ops */
349 bp
->b_ops
= &xfs_sb_buf_ops
;
361 * Update alignment values based on mount options and sb values
364 xfs_update_alignment(xfs_mount_t
*mp
)
366 xfs_sb_t
*sbp
= &(mp
->m_sb
);
370 * If stripe unit and stripe width are not multiples
371 * of the fs blocksize turn off alignment.
373 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
374 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
376 "alignment check failed: sunit/swidth vs. blocksize(%d)",
378 return XFS_ERROR(EINVAL
);
381 * Convert the stripe unit and width to FSBs.
383 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
384 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
386 "alignment check failed: sunit/swidth vs. agsize(%d)",
388 return XFS_ERROR(EINVAL
);
389 } else if (mp
->m_dalign
) {
390 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
393 "alignment check failed: sunit(%d) less than bsize(%d)",
394 mp
->m_dalign
, sbp
->sb_blocksize
);
395 return XFS_ERROR(EINVAL
);
400 * Update superblock with new values
403 if (xfs_sb_version_hasdalign(sbp
)) {
404 if (sbp
->sb_unit
!= mp
->m_dalign
) {
405 sbp
->sb_unit
= mp
->m_dalign
;
406 mp
->m_update_flags
|= XFS_SB_UNIT
;
408 if (sbp
->sb_width
!= mp
->m_swidth
) {
409 sbp
->sb_width
= mp
->m_swidth
;
410 mp
->m_update_flags
|= XFS_SB_WIDTH
;
414 "cannot change alignment: superblock does not support data alignment");
415 return XFS_ERROR(EINVAL
);
417 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
418 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
419 mp
->m_dalign
= sbp
->sb_unit
;
420 mp
->m_swidth
= sbp
->sb_width
;
427 * Set the maximum inode count for this filesystem
430 xfs_set_maxicount(xfs_mount_t
*mp
)
432 xfs_sb_t
*sbp
= &(mp
->m_sb
);
435 if (sbp
->sb_imax_pct
) {
437 * Make sure the maximum inode count is a multiple
438 * of the units we allocate inodes in.
440 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
442 do_div(icount
, mp
->m_ialloc_blks
);
443 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
451 * Set the default minimum read and write sizes unless
452 * already specified in a mount option.
453 * We use smaller I/O sizes when the file system
454 * is being used for NFS service (wsync mount option).
457 xfs_set_rw_sizes(xfs_mount_t
*mp
)
459 xfs_sb_t
*sbp
= &(mp
->m_sb
);
460 int readio_log
, writeio_log
;
462 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
463 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
464 readio_log
= XFS_WSYNC_READIO_LOG
;
465 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
467 readio_log
= XFS_READIO_LOG_LARGE
;
468 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
471 readio_log
= mp
->m_readio_log
;
472 writeio_log
= mp
->m_writeio_log
;
475 if (sbp
->sb_blocklog
> readio_log
) {
476 mp
->m_readio_log
= sbp
->sb_blocklog
;
478 mp
->m_readio_log
= readio_log
;
480 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
481 if (sbp
->sb_blocklog
> writeio_log
) {
482 mp
->m_writeio_log
= sbp
->sb_blocklog
;
484 mp
->m_writeio_log
= writeio_log
;
486 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
490 * precalculate the low space thresholds for dynamic speculative preallocation.
493 xfs_set_low_space_thresholds(
494 struct xfs_mount
*mp
)
498 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
499 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
502 mp
->m_low_space
[i
] = space
* (i
+ 1);
508 * Set whether we're using inode alignment.
511 xfs_set_inoalignment(xfs_mount_t
*mp
)
513 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
514 mp
->m_sb
.sb_inoalignmt
>=
515 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
516 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
518 mp
->m_inoalign_mask
= 0;
520 * If we are using stripe alignment, check whether
521 * the stripe unit is a multiple of the inode alignment
523 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
524 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
525 mp
->m_sinoalign
= mp
->m_dalign
;
531 * Check that the data (and log if separate) is an ok size.
534 xfs_check_sizes(xfs_mount_t
*mp
)
539 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
540 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
541 xfs_warn(mp
, "filesystem size mismatch detected");
542 return XFS_ERROR(EFBIG
);
544 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
545 d
- XFS_FSS_TO_BB(mp
, 1),
546 XFS_FSS_TO_BB(mp
, 1), 0, NULL
);
548 xfs_warn(mp
, "last sector read failed");
553 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
554 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
555 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
556 xfs_warn(mp
, "log size mismatch detected");
557 return XFS_ERROR(EFBIG
);
559 bp
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
560 d
- XFS_FSB_TO_BB(mp
, 1),
561 XFS_FSB_TO_BB(mp
, 1), 0, NULL
);
563 xfs_warn(mp
, "log device read failed");
572 * Clear the quotaflags in memory and in the superblock.
575 xfs_mount_reset_sbqflags(
576 struct xfs_mount
*mp
)
579 struct xfs_trans
*tp
;
584 * It is OK to look at sb_qflags here in mount path,
587 if (mp
->m_sb
.sb_qflags
== 0)
589 spin_lock(&mp
->m_sb_lock
);
590 mp
->m_sb
.sb_qflags
= 0;
591 spin_unlock(&mp
->m_sb_lock
);
594 * If the fs is readonly, let the incore superblock run
595 * with quotas off but don't flush the update out to disk
597 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
600 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
601 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_qm_sbchange
, 0, 0);
603 xfs_trans_cancel(tp
, 0);
604 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
608 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
609 return xfs_trans_commit(tp
, 0);
613 xfs_default_resblks(xfs_mount_t
*mp
)
618 * We default to 5% or 8192 fsbs of space reserved, whichever is
619 * smaller. This is intended to cover concurrent allocation
620 * transactions when we initially hit enospc. These each require a 4
621 * block reservation. Hence by default we cover roughly 2000 concurrent
622 * allocation reservations.
624 resblks
= mp
->m_sb
.sb_dblocks
;
626 resblks
= min_t(__uint64_t
, resblks
, 8192);
631 * This function does the following on an initial mount of a file system:
632 * - reads the superblock from disk and init the mount struct
633 * - if we're a 32-bit kernel, do a size check on the superblock
634 * so we don't mount terabyte filesystems
635 * - init mount struct realtime fields
636 * - allocate inode hash table for fs
637 * - init directory manager
638 * - perform recovery and init the log manager
644 xfs_sb_t
*sbp
= &(mp
->m_sb
);
651 xfs_sb_mount_common(mp
, sbp
);
654 * Check for a mismatched features2 values. Older kernels
655 * read & wrote into the wrong sb offset for sb_features2
656 * on some platforms due to xfs_sb_t not being 64bit size aligned
657 * when sb_features2 was added, which made older superblock
658 * reading/writing routines swap it as a 64-bit value.
660 * For backwards compatibility, we make both slots equal.
662 * If we detect a mismatched field, we OR the set bits into the
663 * existing features2 field in case it has already been modified; we
664 * don't want to lose any features. We then update the bad location
665 * with the ORed value so that older kernels will see any features2
666 * flags, and mark the two fields as needing updates once the
667 * transaction subsystem is online.
669 if (xfs_sb_has_mismatched_features2(sbp
)) {
670 xfs_warn(mp
, "correcting sb_features alignment problem");
671 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
672 sbp
->sb_bad_features2
= sbp
->sb_features2
;
673 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
676 * Re-check for ATTR2 in case it was found in bad_features2
679 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
680 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
681 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
684 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
685 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
686 xfs_sb_version_removeattr2(&mp
->m_sb
);
687 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
689 /* update sb_versionnum for the clearing of the morebits */
690 if (!sbp
->sb_features2
)
691 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
695 * Check if sb_agblocks is aligned at stripe boundary
696 * If sb_agblocks is NOT aligned turn off m_dalign since
697 * allocator alignment is within an ag, therefore ag has
698 * to be aligned at stripe boundary.
700 error
= xfs_update_alignment(mp
);
704 xfs_alloc_compute_maxlevels(mp
);
705 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
706 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
707 xfs_ialloc_compute_maxlevels(mp
);
709 xfs_set_maxicount(mp
);
711 error
= xfs_uuid_mount(mp
);
716 * Set the minimum read and write sizes
718 xfs_set_rw_sizes(mp
);
720 /* set the low space thresholds for dynamic preallocation */
721 xfs_set_low_space_thresholds(mp
);
724 * Set the inode cluster size.
725 * This may still be overridden by the file system
726 * block size if it is larger than the chosen cluster size.
728 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
731 * Set inode alignment fields
733 xfs_set_inoalignment(mp
);
736 * Check that the data (and log if separate) is an ok size.
738 error
= xfs_check_sizes(mp
);
740 goto out_remove_uuid
;
743 * Initialize realtime fields in the mount structure
745 error
= xfs_rtmount_init(mp
);
747 xfs_warn(mp
, "RT mount failed");
748 goto out_remove_uuid
;
752 * Copies the low order bits of the timestamp and the randomly
753 * set "sequence" number out of a UUID.
755 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
757 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
762 * Initialize the attribute manager's entries.
764 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
767 * Initialize the precomputed transaction reservations values.
772 * Allocate and initialize the per-ag data.
774 spin_lock_init(&mp
->m_perag_lock
);
775 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
776 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
778 xfs_warn(mp
, "Failed per-ag init: %d", error
);
779 goto out_remove_uuid
;
782 if (!sbp
->sb_logblocks
) {
783 xfs_warn(mp
, "no log defined");
784 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
785 error
= XFS_ERROR(EFSCORRUPTED
);
790 * log's mount-time initialization. Perform 1st part recovery if needed
792 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
793 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
794 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
796 xfs_warn(mp
, "log mount failed");
801 * Now the log is mounted, we know if it was an unclean shutdown or
802 * not. If it was, with the first phase of recovery has completed, we
803 * have consistent AG blocks on disk. We have not recovered EFIs yet,
804 * but they are recovered transactionally in the second recovery phase
807 * Hence we can safely re-initialise incore superblock counters from
808 * the per-ag data. These may not be correct if the filesystem was not
809 * cleanly unmounted, so we need to wait for recovery to finish before
812 * If the filesystem was cleanly unmounted, then we can trust the
813 * values in the superblock to be correct and we don't need to do
816 * If we are currently making the filesystem, the initialisation will
817 * fail as the perag data is in an undefined state.
819 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
820 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
821 !mp
->m_sb
.sb_inprogress
) {
822 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
828 * Get and sanity-check the root inode.
829 * Save the pointer to it in the mount structure.
831 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
833 xfs_warn(mp
, "failed to read root inode");
834 goto out_log_dealloc
;
839 if (unlikely(!S_ISDIR(rip
->i_d
.di_mode
))) {
840 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
841 (unsigned long long)rip
->i_ino
);
842 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
843 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
845 error
= XFS_ERROR(EFSCORRUPTED
);
848 mp
->m_rootip
= rip
; /* save it */
850 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
853 * Initialize realtime inode pointers in the mount structure
855 error
= xfs_rtmount_inodes(mp
);
858 * Free up the root inode.
860 xfs_warn(mp
, "failed to read RT inodes");
865 * If this is a read-only mount defer the superblock updates until
866 * the next remount into writeable mode. Otherwise we would never
867 * perform the update e.g. for the root filesystem.
869 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
870 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
872 xfs_warn(mp
, "failed to write sb changes");
878 * Initialise the XFS quota management subsystem for this mount
880 if (XFS_IS_QUOTA_RUNNING(mp
)) {
881 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
885 ASSERT(!XFS_IS_QUOTA_ON(mp
));
888 * If a file system had quotas running earlier, but decided to
889 * mount without -o uquota/pquota/gquota options, revoke the
890 * quotachecked license.
892 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
893 xfs_notice(mp
, "resetting quota flags");
894 error
= xfs_mount_reset_sbqflags(mp
);
901 * Finish recovering the file system. This part needed to be
902 * delayed until after the root and real-time bitmap inodes
903 * were consistently read in.
905 error
= xfs_log_mount_finish(mp
);
907 xfs_warn(mp
, "log mount finish failed");
912 * Complete the quota initialisation, post-log-replay component.
915 ASSERT(mp
->m_qflags
== 0);
916 mp
->m_qflags
= quotaflags
;
918 xfs_qm_mount_quotas(mp
);
922 * Now we are mounted, reserve a small amount of unused space for
923 * privileged transactions. This is needed so that transaction
924 * space required for critical operations can dip into this pool
925 * when at ENOSPC. This is needed for operations like create with
926 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
927 * are not allowed to use this reserved space.
929 * This may drive us straight to ENOSPC on mount, but that implies
930 * we were already there on the last unmount. Warn if this occurs.
932 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
933 resblks
= xfs_default_resblks(mp
);
934 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
937 "Unable to allocate reserve blocks. Continuing without reserve pool.");
943 xfs_rtunmount_inodes(mp
);
949 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
950 xfs_wait_buftarg(mp
->m_logdev_targp
);
951 xfs_wait_buftarg(mp
->m_ddev_targp
);
955 xfs_uuid_unmount(mp
);
961 * This flushes out the inodes,dquots and the superblock, unmounts the
962 * log and makes sure that incore structures are freed.
966 struct xfs_mount
*mp
)
971 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
973 xfs_qm_unmount_quotas(mp
);
974 xfs_rtunmount_inodes(mp
);
978 * We can potentially deadlock here if we have an inode cluster
979 * that has been freed has its buffer still pinned in memory because
980 * the transaction is still sitting in a iclog. The stale inodes
981 * on that buffer will have their flush locks held until the
982 * transaction hits the disk and the callbacks run. the inode
983 * flush takes the flush lock unconditionally and with nothing to
984 * push out the iclog we will never get that unlocked. hence we
985 * need to force the log first.
987 xfs_log_force(mp
, XFS_LOG_SYNC
);
990 * Flush all pending changes from the AIL.
992 xfs_ail_push_all_sync(mp
->m_ail
);
995 * And reclaim all inodes. At this point there should be no dirty
996 * inodes and none should be pinned or locked, but use synchronous
997 * reclaim just to be sure. We can stop background inode reclaim
998 * here as well if it is still running.
1000 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1001 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1006 * Unreserve any blocks we have so that when we unmount we don't account
1007 * the reserved free space as used. This is really only necessary for
1008 * lazy superblock counting because it trusts the incore superblock
1009 * counters to be absolutely correct on clean unmount.
1011 * We don't bother correcting this elsewhere for lazy superblock
1012 * counting because on mount of an unclean filesystem we reconstruct the
1013 * correct counter value and this is irrelevant.
1015 * For non-lazy counter filesystems, this doesn't matter at all because
1016 * we only every apply deltas to the superblock and hence the incore
1017 * value does not matter....
1020 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1022 xfs_warn(mp
, "Unable to free reserved block pool. "
1023 "Freespace may not be correct on next mount.");
1025 error
= xfs_log_sbcount(mp
);
1027 xfs_warn(mp
, "Unable to update superblock counters. "
1028 "Freespace may not be correct on next mount.");
1030 xfs_log_unmount(mp
);
1031 xfs_uuid_unmount(mp
);
1034 xfs_errortag_clearall(mp
, 0);
1040 xfs_fs_writable(xfs_mount_t
*mp
)
1042 return !(mp
->m_super
->s_writers
.frozen
|| XFS_FORCED_SHUTDOWN(mp
) ||
1043 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1049 * Sync the superblock counters to disk.
1051 * Note this code can be called during the process of freezing, so
1052 * we may need to use the transaction allocator which does not
1053 * block when the transaction subsystem is in its frozen state.
1056 xfs_log_sbcount(xfs_mount_t
*mp
)
1061 if (!xfs_fs_writable(mp
))
1064 xfs_icsb_sync_counters(mp
, 0);
1067 * we don't need to do this if we are updating the superblock
1068 * counters on every modification.
1070 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1073 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1074 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_sb
, 0, 0);
1076 xfs_trans_cancel(tp
, 0);
1080 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1081 xfs_trans_set_sync(tp
);
1082 error
= xfs_trans_commit(tp
, 0);
1087 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1088 * a delta to a specified field in the in-core superblock. Simply
1089 * switch on the field indicated and apply the delta to that field.
1090 * Fields are not allowed to dip below zero, so if the delta would
1091 * do this do not apply it and return EINVAL.
1093 * The m_sb_lock must be held when this routine is called.
1096 xfs_mod_incore_sb_unlocked(
1098 xfs_sb_field_t field
,
1102 int scounter
; /* short counter for 32 bit fields */
1103 long long lcounter
; /* long counter for 64 bit fields */
1104 long long res_used
, rem
;
1107 * With the in-core superblock spin lock held, switch
1108 * on the indicated field. Apply the delta to the
1109 * proper field. If the fields value would dip below
1110 * 0, then do not apply the delta and return EINVAL.
1113 case XFS_SBS_ICOUNT
:
1114 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1118 return XFS_ERROR(EINVAL
);
1120 mp
->m_sb
.sb_icount
= lcounter
;
1123 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1127 return XFS_ERROR(EINVAL
);
1129 mp
->m_sb
.sb_ifree
= lcounter
;
1131 case XFS_SBS_FDBLOCKS
:
1132 lcounter
= (long long)
1133 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1134 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1136 if (delta
> 0) { /* Putting blocks back */
1137 if (res_used
> delta
) {
1138 mp
->m_resblks_avail
+= delta
;
1140 rem
= delta
- res_used
;
1141 mp
->m_resblks_avail
= mp
->m_resblks
;
1144 } else { /* Taking blocks away */
1146 if (lcounter
>= 0) {
1147 mp
->m_sb
.sb_fdblocks
= lcounter
+
1148 XFS_ALLOC_SET_ASIDE(mp
);
1153 * We are out of blocks, use any available reserved
1154 * blocks if were allowed to.
1157 return XFS_ERROR(ENOSPC
);
1159 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1160 if (lcounter
>= 0) {
1161 mp
->m_resblks_avail
= lcounter
;
1164 printk_once(KERN_WARNING
1165 "Filesystem \"%s\": reserve blocks depleted! "
1166 "Consider increasing reserve pool size.",
1168 return XFS_ERROR(ENOSPC
);
1171 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1173 case XFS_SBS_FREXTENTS
:
1174 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1177 return XFS_ERROR(ENOSPC
);
1179 mp
->m_sb
.sb_frextents
= lcounter
;
1181 case XFS_SBS_DBLOCKS
:
1182 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1186 return XFS_ERROR(EINVAL
);
1188 mp
->m_sb
.sb_dblocks
= lcounter
;
1190 case XFS_SBS_AGCOUNT
:
1191 scounter
= mp
->m_sb
.sb_agcount
;
1195 return XFS_ERROR(EINVAL
);
1197 mp
->m_sb
.sb_agcount
= scounter
;
1199 case XFS_SBS_IMAX_PCT
:
1200 scounter
= mp
->m_sb
.sb_imax_pct
;
1204 return XFS_ERROR(EINVAL
);
1206 mp
->m_sb
.sb_imax_pct
= scounter
;
1208 case XFS_SBS_REXTSIZE
:
1209 scounter
= mp
->m_sb
.sb_rextsize
;
1213 return XFS_ERROR(EINVAL
);
1215 mp
->m_sb
.sb_rextsize
= scounter
;
1217 case XFS_SBS_RBMBLOCKS
:
1218 scounter
= mp
->m_sb
.sb_rbmblocks
;
1222 return XFS_ERROR(EINVAL
);
1224 mp
->m_sb
.sb_rbmblocks
= scounter
;
1226 case XFS_SBS_RBLOCKS
:
1227 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1231 return XFS_ERROR(EINVAL
);
1233 mp
->m_sb
.sb_rblocks
= lcounter
;
1235 case XFS_SBS_REXTENTS
:
1236 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1240 return XFS_ERROR(EINVAL
);
1242 mp
->m_sb
.sb_rextents
= lcounter
;
1244 case XFS_SBS_REXTSLOG
:
1245 scounter
= mp
->m_sb
.sb_rextslog
;
1249 return XFS_ERROR(EINVAL
);
1251 mp
->m_sb
.sb_rextslog
= scounter
;
1255 return XFS_ERROR(EINVAL
);
1260 * xfs_mod_incore_sb() is used to change a field in the in-core
1261 * superblock structure by the specified delta. This modification
1262 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1263 * routine to do the work.
1267 struct xfs_mount
*mp
,
1268 xfs_sb_field_t field
,
1274 #ifdef HAVE_PERCPU_SB
1275 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1277 spin_lock(&mp
->m_sb_lock
);
1278 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1279 spin_unlock(&mp
->m_sb_lock
);
1285 * Change more than one field in the in-core superblock structure at a time.
1287 * The fields and changes to those fields are specified in the array of
1288 * xfs_mod_sb structures passed in. Either all of the specified deltas
1289 * will be applied or none of them will. If any modified field dips below 0,
1290 * then all modifications will be backed out and EINVAL will be returned.
1292 * Note that this function may not be used for the superblock values that
1293 * are tracked with the in-memory per-cpu counters - a direct call to
1294 * xfs_icsb_modify_counters is required for these.
1297 xfs_mod_incore_sb_batch(
1298 struct xfs_mount
*mp
,
1307 * Loop through the array of mod structures and apply each individually.
1308 * If any fail, then back out all those which have already been applied.
1309 * Do all of this within the scope of the m_sb_lock so that all of the
1310 * changes will be atomic.
1312 spin_lock(&mp
->m_sb_lock
);
1313 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1314 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1315 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1317 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1318 msbp
->msb_delta
, rsvd
);
1322 spin_unlock(&mp
->m_sb_lock
);
1326 while (--msbp
>= msb
) {
1327 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1328 -msbp
->msb_delta
, rsvd
);
1331 spin_unlock(&mp
->m_sb_lock
);
1336 * xfs_getsb() is called to obtain the buffer for the superblock.
1337 * The buffer is returned locked and read in from disk.
1338 * The buffer should be released with a call to xfs_brelse().
1340 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1341 * the superblock buffer if it can be locked without sleeping.
1342 * If it can't then we'll return NULL.
1346 struct xfs_mount
*mp
,
1349 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1351 if (!xfs_buf_trylock(bp
)) {
1352 if (flags
& XBF_TRYLOCK
)
1358 ASSERT(XFS_BUF_ISDONE(bp
));
1363 * Used to free the superblock along various error paths.
1367 struct xfs_mount
*mp
)
1369 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1377 * Used to log changes to the superblock unit and width fields which could
1378 * be altered by the mount options, as well as any potential sb_features2
1379 * fixup. Only the first superblock is updated.
1389 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1390 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1391 XFS_SB_VERSIONNUM
));
1393 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1394 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_sb
, 0, 0);
1396 xfs_trans_cancel(tp
, 0);
1399 xfs_mod_sb(tp
, fields
);
1400 error
= xfs_trans_commit(tp
, 0);
1405 * If the underlying (data/log/rt) device is readonly, there are some
1406 * operations that cannot proceed.
1409 xfs_dev_is_read_only(
1410 struct xfs_mount
*mp
,
1413 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1414 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1415 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1416 xfs_notice(mp
, "%s required on read-only device.", message
);
1417 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1423 #ifdef HAVE_PERCPU_SB
1425 * Per-cpu incore superblock counters
1427 * Simple concept, difficult implementation
1429 * Basically, replace the incore superblock counters with a distributed per cpu
1430 * counter for contended fields (e.g. free block count).
1432 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1433 * hence needs to be accurately read when we are running low on space. Hence
1434 * there is a method to enable and disable the per-cpu counters based on how
1435 * much "stuff" is available in them.
1437 * Basically, a counter is enabled if there is enough free resource to justify
1438 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1439 * ENOSPC), then we disable the counters to synchronise all callers and
1440 * re-distribute the available resources.
1442 * If, once we redistributed the available resources, we still get a failure,
1443 * we disable the per-cpu counter and go through the slow path.
1445 * The slow path is the current xfs_mod_incore_sb() function. This means that
1446 * when we disable a per-cpu counter, we need to drain its resources back to
1447 * the global superblock. We do this after disabling the counter to prevent
1448 * more threads from queueing up on the counter.
1450 * Essentially, this means that we still need a lock in the fast path to enable
1451 * synchronisation between the global counters and the per-cpu counters. This
1452 * is not a problem because the lock will be local to a CPU almost all the time
1453 * and have little contention except when we get to ENOSPC conditions.
1455 * Basically, this lock becomes a barrier that enables us to lock out the fast
1456 * path while we do things like enabling and disabling counters and
1457 * synchronising the counters.
1461 * 1. m_sb_lock before picking up per-cpu locks
1462 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1463 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1464 * 4. modifying per-cpu counters requires holding per-cpu lock
1465 * 5. modifying global counters requires holding m_sb_lock
1466 * 6. enabling or disabling a counter requires holding the m_sb_lock
1467 * and _none_ of the per-cpu locks.
1469 * Disabled counters are only ever re-enabled by a balance operation
1470 * that results in more free resources per CPU than a given threshold.
1471 * To ensure counters don't remain disabled, they are rebalanced when
1472 * the global resource goes above a higher threshold (i.e. some hysteresis
1473 * is present to prevent thrashing).
1476 #ifdef CONFIG_HOTPLUG_CPU
1478 * hot-plug CPU notifier support.
1480 * We need a notifier per filesystem as we need to be able to identify
1481 * the filesystem to balance the counters out. This is achieved by
1482 * having a notifier block embedded in the xfs_mount_t and doing pointer
1483 * magic to get the mount pointer from the notifier block address.
1486 xfs_icsb_cpu_notify(
1487 struct notifier_block
*nfb
,
1488 unsigned long action
,
1491 xfs_icsb_cnts_t
*cntp
;
1494 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1495 cntp
= (xfs_icsb_cnts_t
*)
1496 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1498 case CPU_UP_PREPARE
:
1499 case CPU_UP_PREPARE_FROZEN
:
1500 /* Easy Case - initialize the area and locks, and
1501 * then rebalance when online does everything else for us. */
1502 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1505 case CPU_ONLINE_FROZEN
:
1507 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1508 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1509 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1510 xfs_icsb_unlock(mp
);
1513 case CPU_DEAD_FROZEN
:
1514 /* Disable all the counters, then fold the dead cpu's
1515 * count into the total on the global superblock and
1516 * re-enable the counters. */
1518 spin_lock(&mp
->m_sb_lock
);
1519 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1520 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1521 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1523 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1524 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1525 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1527 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1529 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
1530 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
1531 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
1532 spin_unlock(&mp
->m_sb_lock
);
1533 xfs_icsb_unlock(mp
);
1539 #endif /* CONFIG_HOTPLUG_CPU */
1542 xfs_icsb_init_counters(
1545 xfs_icsb_cnts_t
*cntp
;
1548 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
1549 if (mp
->m_sb_cnts
== NULL
)
1552 for_each_online_cpu(i
) {
1553 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1554 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1557 mutex_init(&mp
->m_icsb_mutex
);
1560 * start with all counters disabled so that the
1561 * initial balance kicks us off correctly
1563 mp
->m_icsb_counters
= -1;
1565 #ifdef CONFIG_HOTPLUG_CPU
1566 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
1567 mp
->m_icsb_notifier
.priority
= 0;
1568 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
1569 #endif /* CONFIG_HOTPLUG_CPU */
1575 xfs_icsb_reinit_counters(
1580 * start with all counters disabled so that the
1581 * initial balance kicks us off correctly
1583 mp
->m_icsb_counters
= -1;
1584 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1585 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1586 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1587 xfs_icsb_unlock(mp
);
1591 xfs_icsb_destroy_counters(
1594 if (mp
->m_sb_cnts
) {
1595 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
1596 free_percpu(mp
->m_sb_cnts
);
1598 mutex_destroy(&mp
->m_icsb_mutex
);
1603 xfs_icsb_cnts_t
*icsbp
)
1605 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
1611 xfs_icsb_unlock_cntr(
1612 xfs_icsb_cnts_t
*icsbp
)
1614 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
1619 xfs_icsb_lock_all_counters(
1622 xfs_icsb_cnts_t
*cntp
;
1625 for_each_online_cpu(i
) {
1626 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1627 xfs_icsb_lock_cntr(cntp
);
1632 xfs_icsb_unlock_all_counters(
1635 xfs_icsb_cnts_t
*cntp
;
1638 for_each_online_cpu(i
) {
1639 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1640 xfs_icsb_unlock_cntr(cntp
);
1647 xfs_icsb_cnts_t
*cnt
,
1650 xfs_icsb_cnts_t
*cntp
;
1653 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
1655 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1656 xfs_icsb_lock_all_counters(mp
);
1658 for_each_online_cpu(i
) {
1659 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1660 cnt
->icsb_icount
+= cntp
->icsb_icount
;
1661 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
1662 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
1665 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1666 xfs_icsb_unlock_all_counters(mp
);
1670 xfs_icsb_counter_disabled(
1672 xfs_sb_field_t field
)
1674 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1675 return test_bit(field
, &mp
->m_icsb_counters
);
1679 xfs_icsb_disable_counter(
1681 xfs_sb_field_t field
)
1683 xfs_icsb_cnts_t cnt
;
1685 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1688 * If we are already disabled, then there is nothing to do
1689 * here. We check before locking all the counters to avoid
1690 * the expensive lock operation when being called in the
1691 * slow path and the counter is already disabled. This is
1692 * safe because the only time we set or clear this state is under
1695 if (xfs_icsb_counter_disabled(mp
, field
))
1698 xfs_icsb_lock_all_counters(mp
);
1699 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
1700 /* drain back to superblock */
1702 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
1704 case XFS_SBS_ICOUNT
:
1705 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1708 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1710 case XFS_SBS_FDBLOCKS
:
1711 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1718 xfs_icsb_unlock_all_counters(mp
);
1722 xfs_icsb_enable_counter(
1724 xfs_sb_field_t field
,
1728 xfs_icsb_cnts_t
*cntp
;
1731 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1733 xfs_icsb_lock_all_counters(mp
);
1734 for_each_online_cpu(i
) {
1735 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
1737 case XFS_SBS_ICOUNT
:
1738 cntp
->icsb_icount
= count
+ resid
;
1741 cntp
->icsb_ifree
= count
+ resid
;
1743 case XFS_SBS_FDBLOCKS
:
1744 cntp
->icsb_fdblocks
= count
+ resid
;
1752 clear_bit(field
, &mp
->m_icsb_counters
);
1753 xfs_icsb_unlock_all_counters(mp
);
1757 xfs_icsb_sync_counters_locked(
1761 xfs_icsb_cnts_t cnt
;
1763 xfs_icsb_count(mp
, &cnt
, flags
);
1765 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
1766 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1767 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
1768 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1769 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
1770 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1774 * Accurate update of per-cpu counters to incore superblock
1777 xfs_icsb_sync_counters(
1781 spin_lock(&mp
->m_sb_lock
);
1782 xfs_icsb_sync_counters_locked(mp
, flags
);
1783 spin_unlock(&mp
->m_sb_lock
);
1787 * Balance and enable/disable counters as necessary.
1789 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1790 * chosen to be the same number as single on disk allocation chunk per CPU, and
1791 * free blocks is something far enough zero that we aren't going thrash when we
1792 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1793 * prevent looping endlessly when xfs_alloc_space asks for more than will
1794 * be distributed to a single CPU but each CPU has enough blocks to be
1797 * Note that we can be called when counters are already disabled.
1798 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1799 * prevent locking every per-cpu counter needlessly.
1802 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1803 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1804 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1806 xfs_icsb_balance_counter_locked(
1808 xfs_sb_field_t field
,
1811 uint64_t count
, resid
;
1812 int weight
= num_online_cpus();
1813 uint64_t min
= (uint64_t)min_per_cpu
;
1815 /* disable counter and sync counter */
1816 xfs_icsb_disable_counter(mp
, field
);
1818 /* update counters - first CPU gets residual*/
1820 case XFS_SBS_ICOUNT
:
1821 count
= mp
->m_sb
.sb_icount
;
1822 resid
= do_div(count
, weight
);
1823 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
1827 count
= mp
->m_sb
.sb_ifree
;
1828 resid
= do_div(count
, weight
);
1829 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
1832 case XFS_SBS_FDBLOCKS
:
1833 count
= mp
->m_sb
.sb_fdblocks
;
1834 resid
= do_div(count
, weight
);
1835 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
1840 count
= resid
= 0; /* quiet, gcc */
1844 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
1848 xfs_icsb_balance_counter(
1850 xfs_sb_field_t fields
,
1853 spin_lock(&mp
->m_sb_lock
);
1854 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
1855 spin_unlock(&mp
->m_sb_lock
);
1859 xfs_icsb_modify_counters(
1861 xfs_sb_field_t field
,
1865 xfs_icsb_cnts_t
*icsbp
;
1866 long long lcounter
; /* long counter for 64 bit fields */
1872 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
1875 * if the counter is disabled, go to slow path
1877 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
1879 xfs_icsb_lock_cntr(icsbp
);
1880 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
1881 xfs_icsb_unlock_cntr(icsbp
);
1886 case XFS_SBS_ICOUNT
:
1887 lcounter
= icsbp
->icsb_icount
;
1889 if (unlikely(lcounter
< 0))
1890 goto balance_counter
;
1891 icsbp
->icsb_icount
= lcounter
;
1895 lcounter
= icsbp
->icsb_ifree
;
1897 if (unlikely(lcounter
< 0))
1898 goto balance_counter
;
1899 icsbp
->icsb_ifree
= lcounter
;
1902 case XFS_SBS_FDBLOCKS
:
1903 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
1905 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1907 if (unlikely(lcounter
< 0))
1908 goto balance_counter
;
1909 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1915 xfs_icsb_unlock_cntr(icsbp
);
1923 * serialise with a mutex so we don't burn lots of cpu on
1924 * the superblock lock. We still need to hold the superblock
1925 * lock, however, when we modify the global structures.
1930 * Now running atomically.
1932 * If the counter is enabled, someone has beaten us to rebalancing.
1933 * Drop the lock and try again in the fast path....
1935 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
1936 xfs_icsb_unlock(mp
);
1941 * The counter is currently disabled. Because we are
1942 * running atomically here, we know a rebalance cannot
1943 * be in progress. Hence we can go straight to operating
1944 * on the global superblock. We do not call xfs_mod_incore_sb()
1945 * here even though we need to get the m_sb_lock. Doing so
1946 * will cause us to re-enter this function and deadlock.
1947 * Hence we get the m_sb_lock ourselves and then call
1948 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1949 * directly on the global counters.
1951 spin_lock(&mp
->m_sb_lock
);
1952 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1953 spin_unlock(&mp
->m_sb_lock
);
1956 * Now that we've modified the global superblock, we
1957 * may be able to re-enable the distributed counters
1958 * (e.g. lots of space just got freed). After that
1962 xfs_icsb_balance_counter(mp
, field
, 0);
1963 xfs_icsb_unlock(mp
);
1967 xfs_icsb_unlock_cntr(icsbp
);
1971 * We may have multiple threads here if multiple per-cpu
1972 * counters run dry at the same time. This will mean we can
1973 * do more balances than strictly necessary but it is not
1974 * the common slowpath case.
1979 * running atomically.
1981 * This will leave the counter in the correct state for future
1982 * accesses. After the rebalance, we simply try again and our retry
1983 * will either succeed through the fast path or slow path without
1984 * another balance operation being required.
1986 xfs_icsb_balance_counter(mp
, field
, delta
);
1987 xfs_icsb_unlock(mp
);