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"
48 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
50 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
52 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
55 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
56 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
59 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
60 static int xfs_uuid_table_size
;
61 static uuid_t
*xfs_uuid_table
;
64 * See if the UUID is unique among mounted XFS filesystems.
65 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
71 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
74 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
77 if (uuid_is_nil(uuid
)) {
78 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
79 return XFS_ERROR(EINVAL
);
82 mutex_lock(&xfs_uuid_table_mutex
);
83 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
84 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
88 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
93 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
94 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
95 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
97 hole
= xfs_uuid_table_size
++;
99 xfs_uuid_table
[hole
] = *uuid
;
100 mutex_unlock(&xfs_uuid_table_mutex
);
105 mutex_unlock(&xfs_uuid_table_mutex
);
106 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
107 return XFS_ERROR(EINVAL
);
112 struct xfs_mount
*mp
)
114 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
117 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
120 mutex_lock(&xfs_uuid_table_mutex
);
121 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
122 if (uuid_is_nil(&xfs_uuid_table
[i
]))
124 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
126 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
129 ASSERT(i
< xfs_uuid_table_size
);
130 mutex_unlock(&xfs_uuid_table_mutex
);
136 struct rcu_head
*head
)
138 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
140 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
145 * Free up the per-ag resources associated with the mount structure.
152 struct xfs_perag
*pag
;
154 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
155 spin_lock(&mp
->m_perag_lock
);
156 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
157 spin_unlock(&mp
->m_perag_lock
);
159 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
160 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
165 * Check size of device based on the (data/realtime) block count.
166 * Note: this check is used by the growfs code as well as mount.
169 xfs_sb_validate_fsb_count(
173 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
174 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
176 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
177 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
179 #else /* Limited by UINT_MAX of sectors */
180 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
187 xfs_initialize_perag(
189 xfs_agnumber_t agcount
,
190 xfs_agnumber_t
*maxagi
)
192 xfs_agnumber_t index
;
193 xfs_agnumber_t first_initialised
= 0;
197 xfs_sb_t
*sbp
= &mp
->m_sb
;
201 * Walk the current per-ag tree so we don't try to initialise AGs
202 * that already exist (growfs case). Allocate and insert all the
203 * AGs we don't find ready for initialisation.
205 for (index
= 0; index
< agcount
; index
++) {
206 pag
= xfs_perag_get(mp
, index
);
211 if (!first_initialised
)
212 first_initialised
= index
;
214 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
217 pag
->pag_agno
= index
;
219 spin_lock_init(&pag
->pag_ici_lock
);
220 mutex_init(&pag
->pag_ici_reclaim_lock
);
221 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
222 spin_lock_init(&pag
->pag_buf_lock
);
223 pag
->pag_buf_tree
= RB_ROOT
;
225 if (radix_tree_preload(GFP_NOFS
))
228 spin_lock(&mp
->m_perag_lock
);
229 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
231 spin_unlock(&mp
->m_perag_lock
);
232 radix_tree_preload_end();
236 spin_unlock(&mp
->m_perag_lock
);
237 radix_tree_preload_end();
241 * If we mount with the inode64 option, or no inode overflows
242 * the legacy 32-bit address space clear the inode32 option.
244 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
245 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
247 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
248 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
250 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
252 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
)
253 index
= xfs_set_inode32(mp
);
255 index
= xfs_set_inode64(mp
);
263 for (; index
> first_initialised
; index
--) {
264 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
273 * Does the initial read of the superblock.
277 struct xfs_mount
*mp
,
280 unsigned int sector_size
;
282 struct xfs_sb
*sbp
= &mp
->m_sb
;
284 int loud
= !(flags
& XFS_MFSI_QUIET
);
286 ASSERT(mp
->m_sb_bp
== NULL
);
287 ASSERT(mp
->m_ddev_targp
!= NULL
);
290 * Allocate a (locked) buffer to hold the superblock.
291 * This will be kept around at all times to optimize
292 * access to the superblock.
294 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
297 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
298 BTOBB(sector_size
), 0,
299 loud
? &xfs_sb_buf_ops
300 : &xfs_sb_quiet_buf_ops
);
303 xfs_warn(mp
, "SB buffer read failed");
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(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
320 xfs_sb_quota_from_disk(&mp
->m_sb
);
323 * We must be able to do sector-sized and sector-aligned IO.
325 if (sector_size
> sbp
->sb_sectsize
) {
327 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
328 sector_size
, sbp
->sb_sectsize
);
334 * If device sector size is smaller than the superblock size,
335 * re-read the superblock so the buffer is correctly sized.
337 if (sector_size
< sbp
->sb_sectsize
) {
339 sector_size
= sbp
->sb_sectsize
;
343 /* Initialize per-cpu counters */
344 xfs_icsb_reinit_counters(mp
);
346 /* no need to be quiet anymore, so reset the buf ops */
347 bp
->b_ops
= &xfs_sb_buf_ops
;
359 * Update alignment values based on mount options and sb values
362 xfs_update_alignment(xfs_mount_t
*mp
)
364 xfs_sb_t
*sbp
= &(mp
->m_sb
);
368 * If stripe unit and stripe width are not multiples
369 * of the fs blocksize turn off alignment.
371 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
372 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
374 "alignment check failed: sunit/swidth vs. blocksize(%d)",
376 return XFS_ERROR(EINVAL
);
379 * Convert the stripe unit and width to FSBs.
381 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
382 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
384 "alignment check failed: sunit/swidth vs. agsize(%d)",
386 return XFS_ERROR(EINVAL
);
387 } else if (mp
->m_dalign
) {
388 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
391 "alignment check failed: sunit(%d) less than bsize(%d)",
392 mp
->m_dalign
, sbp
->sb_blocksize
);
393 return XFS_ERROR(EINVAL
);
398 * Update superblock with new values
401 if (xfs_sb_version_hasdalign(sbp
)) {
402 if (sbp
->sb_unit
!= mp
->m_dalign
) {
403 sbp
->sb_unit
= mp
->m_dalign
;
404 mp
->m_update_flags
|= XFS_SB_UNIT
;
406 if (sbp
->sb_width
!= mp
->m_swidth
) {
407 sbp
->sb_width
= mp
->m_swidth
;
408 mp
->m_update_flags
|= XFS_SB_WIDTH
;
412 "cannot change alignment: superblock does not support data alignment");
413 return XFS_ERROR(EINVAL
);
415 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
416 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
417 mp
->m_dalign
= sbp
->sb_unit
;
418 mp
->m_swidth
= sbp
->sb_width
;
425 * Set the maximum inode count for this filesystem
428 xfs_set_maxicount(xfs_mount_t
*mp
)
430 xfs_sb_t
*sbp
= &(mp
->m_sb
);
433 if (sbp
->sb_imax_pct
) {
435 * Make sure the maximum inode count is a multiple
436 * of the units we allocate inodes in.
438 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
440 do_div(icount
, mp
->m_ialloc_blks
);
441 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
449 * Set the default minimum read and write sizes unless
450 * already specified in a mount option.
451 * We use smaller I/O sizes when the file system
452 * is being used for NFS service (wsync mount option).
455 xfs_set_rw_sizes(xfs_mount_t
*mp
)
457 xfs_sb_t
*sbp
= &(mp
->m_sb
);
458 int readio_log
, writeio_log
;
460 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
461 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
462 readio_log
= XFS_WSYNC_READIO_LOG
;
463 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
465 readio_log
= XFS_READIO_LOG_LARGE
;
466 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
469 readio_log
= mp
->m_readio_log
;
470 writeio_log
= mp
->m_writeio_log
;
473 if (sbp
->sb_blocklog
> readio_log
) {
474 mp
->m_readio_log
= sbp
->sb_blocklog
;
476 mp
->m_readio_log
= readio_log
;
478 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
479 if (sbp
->sb_blocklog
> writeio_log
) {
480 mp
->m_writeio_log
= sbp
->sb_blocklog
;
482 mp
->m_writeio_log
= writeio_log
;
484 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
488 * precalculate the low space thresholds for dynamic speculative preallocation.
491 xfs_set_low_space_thresholds(
492 struct xfs_mount
*mp
)
496 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
497 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
500 mp
->m_low_space
[i
] = space
* (i
+ 1);
506 * Set whether we're using inode alignment.
509 xfs_set_inoalignment(xfs_mount_t
*mp
)
511 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
512 mp
->m_sb
.sb_inoalignmt
>=
513 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
514 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
516 mp
->m_inoalign_mask
= 0;
518 * If we are using stripe alignment, check whether
519 * the stripe unit is a multiple of the inode alignment
521 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
522 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
523 mp
->m_sinoalign
= mp
->m_dalign
;
529 * Check that the data (and log if separate) is an ok size.
532 xfs_check_sizes(xfs_mount_t
*mp
)
537 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
538 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
539 xfs_warn(mp
, "filesystem size mismatch detected");
540 return XFS_ERROR(EFBIG
);
542 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
543 d
- XFS_FSS_TO_BB(mp
, 1),
544 XFS_FSS_TO_BB(mp
, 1), 0, NULL
);
546 xfs_warn(mp
, "last sector read failed");
551 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
552 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
553 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
554 xfs_warn(mp
, "log size mismatch detected");
555 return XFS_ERROR(EFBIG
);
557 bp
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
558 d
- XFS_FSB_TO_BB(mp
, 1),
559 XFS_FSB_TO_BB(mp
, 1), 0, NULL
);
561 xfs_warn(mp
, "log device read failed");
570 * Clear the quotaflags in memory and in the superblock.
573 xfs_mount_reset_sbqflags(
574 struct xfs_mount
*mp
)
577 struct xfs_trans
*tp
;
582 * It is OK to look at sb_qflags here in mount path,
585 if (mp
->m_sb
.sb_qflags
== 0)
587 spin_lock(&mp
->m_sb_lock
);
588 mp
->m_sb
.sb_qflags
= 0;
589 spin_unlock(&mp
->m_sb_lock
);
592 * If the fs is readonly, let the incore superblock run
593 * with quotas off but don't flush the update out to disk
595 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
598 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
599 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_qm_sbchange
, 0, 0);
601 xfs_trans_cancel(tp
, 0);
602 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
606 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
607 return xfs_trans_commit(tp
, 0);
611 xfs_default_resblks(xfs_mount_t
*mp
)
616 * We default to 5% or 8192 fsbs of space reserved, whichever is
617 * smaller. This is intended to cover concurrent allocation
618 * transactions when we initially hit enospc. These each require a 4
619 * block reservation. Hence by default we cover roughly 2000 concurrent
620 * allocation reservations.
622 resblks
= mp
->m_sb
.sb_dblocks
;
624 resblks
= min_t(__uint64_t
, resblks
, 8192);
629 * This function does the following on an initial mount of a file system:
630 * - reads the superblock from disk and init the mount struct
631 * - if we're a 32-bit kernel, do a size check on the superblock
632 * so we don't mount terabyte filesystems
633 * - init mount struct realtime fields
634 * - allocate inode hash table for fs
635 * - init directory manager
636 * - perform recovery and init the log manager
642 xfs_sb_t
*sbp
= &(mp
->m_sb
);
649 xfs_sb_mount_common(mp
, sbp
);
652 * Check for a mismatched features2 values. Older kernels
653 * read & wrote into the wrong sb offset for sb_features2
654 * on some platforms due to xfs_sb_t not being 64bit size aligned
655 * when sb_features2 was added, which made older superblock
656 * reading/writing routines swap it as a 64-bit value.
658 * For backwards compatibility, we make both slots equal.
660 * If we detect a mismatched field, we OR the set bits into the
661 * existing features2 field in case it has already been modified; we
662 * don't want to lose any features. We then update the bad location
663 * with the ORed value so that older kernels will see any features2
664 * flags, and mark the two fields as needing updates once the
665 * transaction subsystem is online.
667 if (xfs_sb_has_mismatched_features2(sbp
)) {
668 xfs_warn(mp
, "correcting sb_features alignment problem");
669 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
670 sbp
->sb_bad_features2
= sbp
->sb_features2
;
671 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
674 * Re-check for ATTR2 in case it was found in bad_features2
677 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
678 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
679 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
682 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
683 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
684 xfs_sb_version_removeattr2(&mp
->m_sb
);
685 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
687 /* update sb_versionnum for the clearing of the morebits */
688 if (!sbp
->sb_features2
)
689 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
693 * Check if sb_agblocks is aligned at stripe boundary
694 * If sb_agblocks is NOT aligned turn off m_dalign since
695 * allocator alignment is within an ag, therefore ag has
696 * to be aligned at stripe boundary.
698 error
= xfs_update_alignment(mp
);
702 xfs_alloc_compute_maxlevels(mp
);
703 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
704 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
705 xfs_ialloc_compute_maxlevels(mp
);
707 xfs_set_maxicount(mp
);
709 error
= xfs_uuid_mount(mp
);
714 * Set the minimum read and write sizes
716 xfs_set_rw_sizes(mp
);
718 /* set the low space thresholds for dynamic preallocation */
719 xfs_set_low_space_thresholds(mp
);
722 * Set the inode cluster size.
723 * This may still be overridden by the file system
724 * block size if it is larger than the chosen cluster size.
726 * For v5 filesystems, scale the cluster size with the inode size to
727 * keep a constant ratio of inode per cluster buffer, but only if mkfs
728 * has set the inode alignment value appropriately for larger cluster
731 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
732 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
733 int new_size
= mp
->m_inode_cluster_size
;
735 new_size
*= mp
->m_sb
.sb_inodesize
/ XFS_DINODE_MIN_SIZE
;
736 if (mp
->m_sb
.sb_inoalignmt
>= XFS_B_TO_FSBT(mp
, new_size
))
737 mp
->m_inode_cluster_size
= new_size
;
738 xfs_info(mp
, "Using inode cluster size of %d bytes",
739 mp
->m_inode_cluster_size
);
743 * Set inode alignment fields
745 xfs_set_inoalignment(mp
);
748 * Check that the data (and log if separate) is an ok size.
750 error
= xfs_check_sizes(mp
);
752 goto out_remove_uuid
;
755 * Initialize realtime fields in the mount structure
757 error
= xfs_rtmount_init(mp
);
759 xfs_warn(mp
, "RT mount failed");
760 goto out_remove_uuid
;
764 * Copies the low order bits of the timestamp and the randomly
765 * set "sequence" number out of a UUID.
767 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
769 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
774 * Initialize the attribute manager's entries.
776 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
779 * Initialize the precomputed transaction reservations values.
784 * Allocate and initialize the per-ag data.
786 spin_lock_init(&mp
->m_perag_lock
);
787 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
788 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
790 xfs_warn(mp
, "Failed per-ag init: %d", error
);
791 goto out_remove_uuid
;
794 if (!sbp
->sb_logblocks
) {
795 xfs_warn(mp
, "no log defined");
796 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
797 error
= XFS_ERROR(EFSCORRUPTED
);
802 * log's mount-time initialization. Perform 1st part recovery if needed
804 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
805 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
806 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
808 xfs_warn(mp
, "log mount failed");
813 * Now the log is mounted, we know if it was an unclean shutdown or
814 * not. If it was, with the first phase of recovery has completed, we
815 * have consistent AG blocks on disk. We have not recovered EFIs yet,
816 * but they are recovered transactionally in the second recovery phase
819 * Hence we can safely re-initialise incore superblock counters from
820 * the per-ag data. These may not be correct if the filesystem was not
821 * cleanly unmounted, so we need to wait for recovery to finish before
824 * If the filesystem was cleanly unmounted, then we can trust the
825 * values in the superblock to be correct and we don't need to do
828 * If we are currently making the filesystem, the initialisation will
829 * fail as the perag data is in an undefined state.
831 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
832 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
833 !mp
->m_sb
.sb_inprogress
) {
834 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
840 * Get and sanity-check the root inode.
841 * Save the pointer to it in the mount structure.
843 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
845 xfs_warn(mp
, "failed to read root inode");
846 goto out_log_dealloc
;
851 if (unlikely(!S_ISDIR(rip
->i_d
.di_mode
))) {
852 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
853 (unsigned long long)rip
->i_ino
);
854 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
855 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
857 error
= XFS_ERROR(EFSCORRUPTED
);
860 mp
->m_rootip
= rip
; /* save it */
862 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
865 * Initialize realtime inode pointers in the mount structure
867 error
= xfs_rtmount_inodes(mp
);
870 * Free up the root inode.
872 xfs_warn(mp
, "failed to read RT inodes");
877 * If this is a read-only mount defer the superblock updates until
878 * the next remount into writeable mode. Otherwise we would never
879 * perform the update e.g. for the root filesystem.
881 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
882 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
884 xfs_warn(mp
, "failed to write sb changes");
890 * Initialise the XFS quota management subsystem for this mount
892 if (XFS_IS_QUOTA_RUNNING(mp
)) {
893 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
897 ASSERT(!XFS_IS_QUOTA_ON(mp
));
900 * If a file system had quotas running earlier, but decided to
901 * mount without -o uquota/pquota/gquota options, revoke the
902 * quotachecked license.
904 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
905 xfs_notice(mp
, "resetting quota flags");
906 error
= xfs_mount_reset_sbqflags(mp
);
913 * Finish recovering the file system. This part needed to be
914 * delayed until after the root and real-time bitmap inodes
915 * were consistently read in.
917 error
= xfs_log_mount_finish(mp
);
919 xfs_warn(mp
, "log mount finish failed");
924 * Complete the quota initialisation, post-log-replay component.
927 ASSERT(mp
->m_qflags
== 0);
928 mp
->m_qflags
= quotaflags
;
930 xfs_qm_mount_quotas(mp
);
934 * Now we are mounted, reserve a small amount of unused space for
935 * privileged transactions. This is needed so that transaction
936 * space required for critical operations can dip into this pool
937 * when at ENOSPC. This is needed for operations like create with
938 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
939 * are not allowed to use this reserved space.
941 * This may drive us straight to ENOSPC on mount, but that implies
942 * we were already there on the last unmount. Warn if this occurs.
944 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
945 resblks
= xfs_default_resblks(mp
);
946 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
949 "Unable to allocate reserve blocks. Continuing without reserve pool.");
955 xfs_rtunmount_inodes(mp
);
961 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
962 xfs_wait_buftarg(mp
->m_logdev_targp
);
963 xfs_wait_buftarg(mp
->m_ddev_targp
);
967 xfs_uuid_unmount(mp
);
973 * This flushes out the inodes,dquots and the superblock, unmounts the
974 * log and makes sure that incore structures are freed.
978 struct xfs_mount
*mp
)
983 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
985 xfs_qm_unmount_quotas(mp
);
986 xfs_rtunmount_inodes(mp
);
990 * We can potentially deadlock here if we have an inode cluster
991 * that has been freed has its buffer still pinned in memory because
992 * the transaction is still sitting in a iclog. The stale inodes
993 * on that buffer will have their flush locks held until the
994 * transaction hits the disk and the callbacks run. the inode
995 * flush takes the flush lock unconditionally and with nothing to
996 * push out the iclog we will never get that unlocked. hence we
997 * need to force the log first.
999 xfs_log_force(mp
, XFS_LOG_SYNC
);
1002 * Flush all pending changes from the AIL.
1004 xfs_ail_push_all_sync(mp
->m_ail
);
1007 * And reclaim all inodes. At this point there should be no dirty
1008 * inodes and none should be pinned or locked, but use synchronous
1009 * reclaim just to be sure. We can stop background inode reclaim
1010 * here as well if it is still running.
1012 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1013 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1018 * Unreserve any blocks we have so that when we unmount we don't account
1019 * the reserved free space as used. This is really only necessary for
1020 * lazy superblock counting because it trusts the incore superblock
1021 * counters to be absolutely correct on clean unmount.
1023 * We don't bother correcting this elsewhere for lazy superblock
1024 * counting because on mount of an unclean filesystem we reconstruct the
1025 * correct counter value and this is irrelevant.
1027 * For non-lazy counter filesystems, this doesn't matter at all because
1028 * we only every apply deltas to the superblock and hence the incore
1029 * value does not matter....
1032 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1034 xfs_warn(mp
, "Unable to free reserved block pool. "
1035 "Freespace may not be correct on next mount.");
1037 error
= xfs_log_sbcount(mp
);
1039 xfs_warn(mp
, "Unable to update superblock counters. "
1040 "Freespace may not be correct on next mount.");
1042 xfs_log_unmount(mp
);
1043 xfs_uuid_unmount(mp
);
1046 xfs_errortag_clearall(mp
, 0);
1052 xfs_fs_writable(xfs_mount_t
*mp
)
1054 return !(mp
->m_super
->s_writers
.frozen
|| XFS_FORCED_SHUTDOWN(mp
) ||
1055 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1061 * Sync the superblock counters to disk.
1063 * Note this code can be called during the process of freezing, so
1064 * we may need to use the transaction allocator which does not
1065 * block when the transaction subsystem is in its frozen state.
1068 xfs_log_sbcount(xfs_mount_t
*mp
)
1073 if (!xfs_fs_writable(mp
))
1076 xfs_icsb_sync_counters(mp
, 0);
1079 * we don't need to do this if we are updating the superblock
1080 * counters on every modification.
1082 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1085 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1086 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_sb
, 0, 0);
1088 xfs_trans_cancel(tp
, 0);
1092 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1093 xfs_trans_set_sync(tp
);
1094 error
= xfs_trans_commit(tp
, 0);
1099 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1100 * a delta to a specified field in the in-core superblock. Simply
1101 * switch on the field indicated and apply the delta to that field.
1102 * Fields are not allowed to dip below zero, so if the delta would
1103 * do this do not apply it and return EINVAL.
1105 * The m_sb_lock must be held when this routine is called.
1108 xfs_mod_incore_sb_unlocked(
1110 xfs_sb_field_t field
,
1114 int scounter
; /* short counter for 32 bit fields */
1115 long long lcounter
; /* long counter for 64 bit fields */
1116 long long res_used
, rem
;
1119 * With the in-core superblock spin lock held, switch
1120 * on the indicated field. Apply the delta to the
1121 * proper field. If the fields value would dip below
1122 * 0, then do not apply the delta and return EINVAL.
1125 case XFS_SBS_ICOUNT
:
1126 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1130 return XFS_ERROR(EINVAL
);
1132 mp
->m_sb
.sb_icount
= lcounter
;
1135 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1139 return XFS_ERROR(EINVAL
);
1141 mp
->m_sb
.sb_ifree
= lcounter
;
1143 case XFS_SBS_FDBLOCKS
:
1144 lcounter
= (long long)
1145 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1146 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1148 if (delta
> 0) { /* Putting blocks back */
1149 if (res_used
> delta
) {
1150 mp
->m_resblks_avail
+= delta
;
1152 rem
= delta
- res_used
;
1153 mp
->m_resblks_avail
= mp
->m_resblks
;
1156 } else { /* Taking blocks away */
1158 if (lcounter
>= 0) {
1159 mp
->m_sb
.sb_fdblocks
= lcounter
+
1160 XFS_ALLOC_SET_ASIDE(mp
);
1165 * We are out of blocks, use any available reserved
1166 * blocks if were allowed to.
1169 return XFS_ERROR(ENOSPC
);
1171 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1172 if (lcounter
>= 0) {
1173 mp
->m_resblks_avail
= lcounter
;
1176 printk_once(KERN_WARNING
1177 "Filesystem \"%s\": reserve blocks depleted! "
1178 "Consider increasing reserve pool size.",
1180 return XFS_ERROR(ENOSPC
);
1183 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1185 case XFS_SBS_FREXTENTS
:
1186 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1189 return XFS_ERROR(ENOSPC
);
1191 mp
->m_sb
.sb_frextents
= lcounter
;
1193 case XFS_SBS_DBLOCKS
:
1194 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1198 return XFS_ERROR(EINVAL
);
1200 mp
->m_sb
.sb_dblocks
= lcounter
;
1202 case XFS_SBS_AGCOUNT
:
1203 scounter
= mp
->m_sb
.sb_agcount
;
1207 return XFS_ERROR(EINVAL
);
1209 mp
->m_sb
.sb_agcount
= scounter
;
1211 case XFS_SBS_IMAX_PCT
:
1212 scounter
= mp
->m_sb
.sb_imax_pct
;
1216 return XFS_ERROR(EINVAL
);
1218 mp
->m_sb
.sb_imax_pct
= scounter
;
1220 case XFS_SBS_REXTSIZE
:
1221 scounter
= mp
->m_sb
.sb_rextsize
;
1225 return XFS_ERROR(EINVAL
);
1227 mp
->m_sb
.sb_rextsize
= scounter
;
1229 case XFS_SBS_RBMBLOCKS
:
1230 scounter
= mp
->m_sb
.sb_rbmblocks
;
1234 return XFS_ERROR(EINVAL
);
1236 mp
->m_sb
.sb_rbmblocks
= scounter
;
1238 case XFS_SBS_RBLOCKS
:
1239 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1243 return XFS_ERROR(EINVAL
);
1245 mp
->m_sb
.sb_rblocks
= lcounter
;
1247 case XFS_SBS_REXTENTS
:
1248 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1252 return XFS_ERROR(EINVAL
);
1254 mp
->m_sb
.sb_rextents
= lcounter
;
1256 case XFS_SBS_REXTSLOG
:
1257 scounter
= mp
->m_sb
.sb_rextslog
;
1261 return XFS_ERROR(EINVAL
);
1263 mp
->m_sb
.sb_rextslog
= scounter
;
1267 return XFS_ERROR(EINVAL
);
1272 * xfs_mod_incore_sb() is used to change a field in the in-core
1273 * superblock structure by the specified delta. This modification
1274 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1275 * routine to do the work.
1279 struct xfs_mount
*mp
,
1280 xfs_sb_field_t field
,
1286 #ifdef HAVE_PERCPU_SB
1287 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1289 spin_lock(&mp
->m_sb_lock
);
1290 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1291 spin_unlock(&mp
->m_sb_lock
);
1297 * Change more than one field in the in-core superblock structure at a time.
1299 * The fields and changes to those fields are specified in the array of
1300 * xfs_mod_sb structures passed in. Either all of the specified deltas
1301 * will be applied or none of them will. If any modified field dips below 0,
1302 * then all modifications will be backed out and EINVAL will be returned.
1304 * Note that this function may not be used for the superblock values that
1305 * are tracked with the in-memory per-cpu counters - a direct call to
1306 * xfs_icsb_modify_counters is required for these.
1309 xfs_mod_incore_sb_batch(
1310 struct xfs_mount
*mp
,
1319 * Loop through the array of mod structures and apply each individually.
1320 * If any fail, then back out all those which have already been applied.
1321 * Do all of this within the scope of the m_sb_lock so that all of the
1322 * changes will be atomic.
1324 spin_lock(&mp
->m_sb_lock
);
1325 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1326 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1327 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1329 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1330 msbp
->msb_delta
, rsvd
);
1334 spin_unlock(&mp
->m_sb_lock
);
1338 while (--msbp
>= msb
) {
1339 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1340 -msbp
->msb_delta
, rsvd
);
1343 spin_unlock(&mp
->m_sb_lock
);
1348 * xfs_getsb() is called to obtain the buffer for the superblock.
1349 * The buffer is returned locked and read in from disk.
1350 * The buffer should be released with a call to xfs_brelse().
1352 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1353 * the superblock buffer if it can be locked without sleeping.
1354 * If it can't then we'll return NULL.
1358 struct xfs_mount
*mp
,
1361 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1363 if (!xfs_buf_trylock(bp
)) {
1364 if (flags
& XBF_TRYLOCK
)
1370 ASSERT(XFS_BUF_ISDONE(bp
));
1375 * Used to free the superblock along various error paths.
1379 struct xfs_mount
*mp
)
1381 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1389 * Used to log changes to the superblock unit and width fields which could
1390 * be altered by the mount options, as well as any potential sb_features2
1391 * fixup. Only the first superblock is updated.
1401 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1402 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1403 XFS_SB_VERSIONNUM
));
1405 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1406 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_sb
, 0, 0);
1408 xfs_trans_cancel(tp
, 0);
1411 xfs_mod_sb(tp
, fields
);
1412 error
= xfs_trans_commit(tp
, 0);
1417 * If the underlying (data/log/rt) device is readonly, there are some
1418 * operations that cannot proceed.
1421 xfs_dev_is_read_only(
1422 struct xfs_mount
*mp
,
1425 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1426 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1427 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1428 xfs_notice(mp
, "%s required on read-only device.", message
);
1429 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1435 #ifdef HAVE_PERCPU_SB
1437 * Per-cpu incore superblock counters
1439 * Simple concept, difficult implementation
1441 * Basically, replace the incore superblock counters with a distributed per cpu
1442 * counter for contended fields (e.g. free block count).
1444 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1445 * hence needs to be accurately read when we are running low on space. Hence
1446 * there is a method to enable and disable the per-cpu counters based on how
1447 * much "stuff" is available in them.
1449 * Basically, a counter is enabled if there is enough free resource to justify
1450 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1451 * ENOSPC), then we disable the counters to synchronise all callers and
1452 * re-distribute the available resources.
1454 * If, once we redistributed the available resources, we still get a failure,
1455 * we disable the per-cpu counter and go through the slow path.
1457 * The slow path is the current xfs_mod_incore_sb() function. This means that
1458 * when we disable a per-cpu counter, we need to drain its resources back to
1459 * the global superblock. We do this after disabling the counter to prevent
1460 * more threads from queueing up on the counter.
1462 * Essentially, this means that we still need a lock in the fast path to enable
1463 * synchronisation between the global counters and the per-cpu counters. This
1464 * is not a problem because the lock will be local to a CPU almost all the time
1465 * and have little contention except when we get to ENOSPC conditions.
1467 * Basically, this lock becomes a barrier that enables us to lock out the fast
1468 * path while we do things like enabling and disabling counters and
1469 * synchronising the counters.
1473 * 1. m_sb_lock before picking up per-cpu locks
1474 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1475 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1476 * 4. modifying per-cpu counters requires holding per-cpu lock
1477 * 5. modifying global counters requires holding m_sb_lock
1478 * 6. enabling or disabling a counter requires holding the m_sb_lock
1479 * and _none_ of the per-cpu locks.
1481 * Disabled counters are only ever re-enabled by a balance operation
1482 * that results in more free resources per CPU than a given threshold.
1483 * To ensure counters don't remain disabled, they are rebalanced when
1484 * the global resource goes above a higher threshold (i.e. some hysteresis
1485 * is present to prevent thrashing).
1488 #ifdef CONFIG_HOTPLUG_CPU
1490 * hot-plug CPU notifier support.
1492 * We need a notifier per filesystem as we need to be able to identify
1493 * the filesystem to balance the counters out. This is achieved by
1494 * having a notifier block embedded in the xfs_mount_t and doing pointer
1495 * magic to get the mount pointer from the notifier block address.
1498 xfs_icsb_cpu_notify(
1499 struct notifier_block
*nfb
,
1500 unsigned long action
,
1503 xfs_icsb_cnts_t
*cntp
;
1506 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1507 cntp
= (xfs_icsb_cnts_t
*)
1508 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1510 case CPU_UP_PREPARE
:
1511 case CPU_UP_PREPARE_FROZEN
:
1512 /* Easy Case - initialize the area and locks, and
1513 * then rebalance when online does everything else for us. */
1514 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1517 case CPU_ONLINE_FROZEN
:
1519 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1520 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1521 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1522 xfs_icsb_unlock(mp
);
1525 case CPU_DEAD_FROZEN
:
1526 /* Disable all the counters, then fold the dead cpu's
1527 * count into the total on the global superblock and
1528 * re-enable the counters. */
1530 spin_lock(&mp
->m_sb_lock
);
1531 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1532 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1533 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1535 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1536 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1537 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1539 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1541 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
1542 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
1543 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
1544 spin_unlock(&mp
->m_sb_lock
);
1545 xfs_icsb_unlock(mp
);
1551 #endif /* CONFIG_HOTPLUG_CPU */
1554 xfs_icsb_init_counters(
1557 xfs_icsb_cnts_t
*cntp
;
1560 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
1561 if (mp
->m_sb_cnts
== NULL
)
1564 for_each_online_cpu(i
) {
1565 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1566 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1569 mutex_init(&mp
->m_icsb_mutex
);
1572 * start with all counters disabled so that the
1573 * initial balance kicks us off correctly
1575 mp
->m_icsb_counters
= -1;
1577 #ifdef CONFIG_HOTPLUG_CPU
1578 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
1579 mp
->m_icsb_notifier
.priority
= 0;
1580 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
1581 #endif /* CONFIG_HOTPLUG_CPU */
1587 xfs_icsb_reinit_counters(
1592 * start with all counters disabled so that the
1593 * initial balance kicks us off correctly
1595 mp
->m_icsb_counters
= -1;
1596 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1597 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1598 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1599 xfs_icsb_unlock(mp
);
1603 xfs_icsb_destroy_counters(
1606 if (mp
->m_sb_cnts
) {
1607 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
1608 free_percpu(mp
->m_sb_cnts
);
1610 mutex_destroy(&mp
->m_icsb_mutex
);
1615 xfs_icsb_cnts_t
*icsbp
)
1617 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
1623 xfs_icsb_unlock_cntr(
1624 xfs_icsb_cnts_t
*icsbp
)
1626 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
1631 xfs_icsb_lock_all_counters(
1634 xfs_icsb_cnts_t
*cntp
;
1637 for_each_online_cpu(i
) {
1638 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1639 xfs_icsb_lock_cntr(cntp
);
1644 xfs_icsb_unlock_all_counters(
1647 xfs_icsb_cnts_t
*cntp
;
1650 for_each_online_cpu(i
) {
1651 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1652 xfs_icsb_unlock_cntr(cntp
);
1659 xfs_icsb_cnts_t
*cnt
,
1662 xfs_icsb_cnts_t
*cntp
;
1665 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
1667 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1668 xfs_icsb_lock_all_counters(mp
);
1670 for_each_online_cpu(i
) {
1671 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1672 cnt
->icsb_icount
+= cntp
->icsb_icount
;
1673 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
1674 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
1677 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1678 xfs_icsb_unlock_all_counters(mp
);
1682 xfs_icsb_counter_disabled(
1684 xfs_sb_field_t field
)
1686 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1687 return test_bit(field
, &mp
->m_icsb_counters
);
1691 xfs_icsb_disable_counter(
1693 xfs_sb_field_t field
)
1695 xfs_icsb_cnts_t cnt
;
1697 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1700 * If we are already disabled, then there is nothing to do
1701 * here. We check before locking all the counters to avoid
1702 * the expensive lock operation when being called in the
1703 * slow path and the counter is already disabled. This is
1704 * safe because the only time we set or clear this state is under
1707 if (xfs_icsb_counter_disabled(mp
, field
))
1710 xfs_icsb_lock_all_counters(mp
);
1711 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
1712 /* drain back to superblock */
1714 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
1716 case XFS_SBS_ICOUNT
:
1717 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1720 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1722 case XFS_SBS_FDBLOCKS
:
1723 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1730 xfs_icsb_unlock_all_counters(mp
);
1734 xfs_icsb_enable_counter(
1736 xfs_sb_field_t field
,
1740 xfs_icsb_cnts_t
*cntp
;
1743 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1745 xfs_icsb_lock_all_counters(mp
);
1746 for_each_online_cpu(i
) {
1747 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
1749 case XFS_SBS_ICOUNT
:
1750 cntp
->icsb_icount
= count
+ resid
;
1753 cntp
->icsb_ifree
= count
+ resid
;
1755 case XFS_SBS_FDBLOCKS
:
1756 cntp
->icsb_fdblocks
= count
+ resid
;
1764 clear_bit(field
, &mp
->m_icsb_counters
);
1765 xfs_icsb_unlock_all_counters(mp
);
1769 xfs_icsb_sync_counters_locked(
1773 xfs_icsb_cnts_t cnt
;
1775 xfs_icsb_count(mp
, &cnt
, flags
);
1777 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
1778 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1779 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
1780 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1781 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
1782 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1786 * Accurate update of per-cpu counters to incore superblock
1789 xfs_icsb_sync_counters(
1793 spin_lock(&mp
->m_sb_lock
);
1794 xfs_icsb_sync_counters_locked(mp
, flags
);
1795 spin_unlock(&mp
->m_sb_lock
);
1799 * Balance and enable/disable counters as necessary.
1801 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1802 * chosen to be the same number as single on disk allocation chunk per CPU, and
1803 * free blocks is something far enough zero that we aren't going thrash when we
1804 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1805 * prevent looping endlessly when xfs_alloc_space asks for more than will
1806 * be distributed to a single CPU but each CPU has enough blocks to be
1809 * Note that we can be called when counters are already disabled.
1810 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1811 * prevent locking every per-cpu counter needlessly.
1814 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1815 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1816 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1818 xfs_icsb_balance_counter_locked(
1820 xfs_sb_field_t field
,
1823 uint64_t count
, resid
;
1824 int weight
= num_online_cpus();
1825 uint64_t min
= (uint64_t)min_per_cpu
;
1827 /* disable counter and sync counter */
1828 xfs_icsb_disable_counter(mp
, field
);
1830 /* update counters - first CPU gets residual*/
1832 case XFS_SBS_ICOUNT
:
1833 count
= mp
->m_sb
.sb_icount
;
1834 resid
= do_div(count
, weight
);
1835 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
1839 count
= mp
->m_sb
.sb_ifree
;
1840 resid
= do_div(count
, weight
);
1841 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
1844 case XFS_SBS_FDBLOCKS
:
1845 count
= mp
->m_sb
.sb_fdblocks
;
1846 resid
= do_div(count
, weight
);
1847 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
1852 count
= resid
= 0; /* quiet, gcc */
1856 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
1860 xfs_icsb_balance_counter(
1862 xfs_sb_field_t fields
,
1865 spin_lock(&mp
->m_sb_lock
);
1866 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
1867 spin_unlock(&mp
->m_sb_lock
);
1871 xfs_icsb_modify_counters(
1873 xfs_sb_field_t field
,
1877 xfs_icsb_cnts_t
*icsbp
;
1878 long long lcounter
; /* long counter for 64 bit fields */
1884 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
1887 * if the counter is disabled, go to slow path
1889 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
1891 xfs_icsb_lock_cntr(icsbp
);
1892 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
1893 xfs_icsb_unlock_cntr(icsbp
);
1898 case XFS_SBS_ICOUNT
:
1899 lcounter
= icsbp
->icsb_icount
;
1901 if (unlikely(lcounter
< 0))
1902 goto balance_counter
;
1903 icsbp
->icsb_icount
= lcounter
;
1907 lcounter
= icsbp
->icsb_ifree
;
1909 if (unlikely(lcounter
< 0))
1910 goto balance_counter
;
1911 icsbp
->icsb_ifree
= lcounter
;
1914 case XFS_SBS_FDBLOCKS
:
1915 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
1917 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1919 if (unlikely(lcounter
< 0))
1920 goto balance_counter
;
1921 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1927 xfs_icsb_unlock_cntr(icsbp
);
1935 * serialise with a mutex so we don't burn lots of cpu on
1936 * the superblock lock. We still need to hold the superblock
1937 * lock, however, when we modify the global structures.
1942 * Now running atomically.
1944 * If the counter is enabled, someone has beaten us to rebalancing.
1945 * Drop the lock and try again in the fast path....
1947 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
1948 xfs_icsb_unlock(mp
);
1953 * The counter is currently disabled. Because we are
1954 * running atomically here, we know a rebalance cannot
1955 * be in progress. Hence we can go straight to operating
1956 * on the global superblock. We do not call xfs_mod_incore_sb()
1957 * here even though we need to get the m_sb_lock. Doing so
1958 * will cause us to re-enter this function and deadlock.
1959 * Hence we get the m_sb_lock ourselves and then call
1960 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1961 * directly on the global counters.
1963 spin_lock(&mp
->m_sb_lock
);
1964 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1965 spin_unlock(&mp
->m_sb_lock
);
1968 * Now that we've modified the global superblock, we
1969 * may be able to re-enable the distributed counters
1970 * (e.g. lots of space just got freed). After that
1974 xfs_icsb_balance_counter(mp
, field
, 0);
1975 xfs_icsb_unlock(mp
);
1979 xfs_icsb_unlock_cntr(icsbp
);
1983 * We may have multiple threads here if multiple per-cpu
1984 * counters run dry at the same time. This will mean we can
1985 * do more balances than strictly necessary but it is not
1986 * the common slowpath case.
1991 * running atomically.
1993 * This will leave the counter in the correct state for future
1994 * accesses. After the rebalance, we simply try again and our retry
1995 * will either succeed through the fast path or slow path without
1996 * another balance operation being required.
1998 xfs_icsb_balance_counter(mp
, field
, delta
);
1999 xfs_icsb_unlock(mp
);