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_types.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_alloc.h"
38 #include "xfs_rtalloc.h"
40 #include "xfs_error.h"
41 #include "xfs_quota.h"
42 #include "xfs_fsops.h"
43 #include "xfs_utils.h"
44 #include "xfs_trace.h"
45 #include "xfs_icache.h"
49 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
51 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
53 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
56 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
57 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
62 short type
; /* 0 = integer
63 * 1 = binary / string (no translation)
66 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
67 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
68 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
69 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
70 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
71 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
72 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
73 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
74 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
75 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
76 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
77 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
78 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
79 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
80 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
81 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
82 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
83 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
84 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
85 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
86 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
87 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
88 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
89 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
90 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
91 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
92 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
93 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
94 { offsetof(xfs_sb_t
, sb_icount
), 0 },
95 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
96 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
97 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
98 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
99 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
100 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
101 { offsetof(xfs_sb_t
, sb_flags
), 0 },
102 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
103 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
104 { offsetof(xfs_sb_t
, sb_unit
), 0 },
105 { offsetof(xfs_sb_t
, sb_width
), 0 },
106 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
107 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
108 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
109 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
110 { offsetof(xfs_sb_t
, sb_features2
), 0 },
111 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
112 { sizeof(xfs_sb_t
), 0 }
115 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
116 static int xfs_uuid_table_size
;
117 static uuid_t
*xfs_uuid_table
;
120 * See if the UUID is unique among mounted XFS filesystems.
121 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
125 struct xfs_mount
*mp
)
127 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
130 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
133 if (uuid_is_nil(uuid
)) {
134 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
135 return XFS_ERROR(EINVAL
);
138 mutex_lock(&xfs_uuid_table_mutex
);
139 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
140 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
144 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
149 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
150 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
151 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
153 hole
= xfs_uuid_table_size
++;
155 xfs_uuid_table
[hole
] = *uuid
;
156 mutex_unlock(&xfs_uuid_table_mutex
);
161 mutex_unlock(&xfs_uuid_table_mutex
);
162 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
163 return XFS_ERROR(EINVAL
);
168 struct xfs_mount
*mp
)
170 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
173 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
176 mutex_lock(&xfs_uuid_table_mutex
);
177 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
178 if (uuid_is_nil(&xfs_uuid_table
[i
]))
180 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
182 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
185 ASSERT(i
< xfs_uuid_table_size
);
186 mutex_unlock(&xfs_uuid_table_mutex
);
191 * Reference counting access wrappers to the perag structures.
192 * Because we never free per-ag structures, the only thing we
193 * have to protect against changes is the tree structure itself.
196 xfs_perag_get(struct xfs_mount
*mp
, xfs_agnumber_t agno
)
198 struct xfs_perag
*pag
;
202 pag
= radix_tree_lookup(&mp
->m_perag_tree
, agno
);
204 ASSERT(atomic_read(&pag
->pag_ref
) >= 0);
205 ref
= atomic_inc_return(&pag
->pag_ref
);
208 trace_xfs_perag_get(mp
, agno
, ref
, _RET_IP_
);
213 * search from @first to find the next perag with the given tag set.
217 struct xfs_mount
*mp
,
218 xfs_agnumber_t first
,
221 struct xfs_perag
*pag
;
226 found
= radix_tree_gang_lookup_tag(&mp
->m_perag_tree
,
227 (void **)&pag
, first
, 1, tag
);
232 ref
= atomic_inc_return(&pag
->pag_ref
);
234 trace_xfs_perag_get_tag(mp
, pag
->pag_agno
, ref
, _RET_IP_
);
239 xfs_perag_put(struct xfs_perag
*pag
)
243 ASSERT(atomic_read(&pag
->pag_ref
) > 0);
244 ref
= atomic_dec_return(&pag
->pag_ref
);
245 trace_xfs_perag_put(pag
->pag_mount
, pag
->pag_agno
, ref
, _RET_IP_
);
250 struct rcu_head
*head
)
252 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
254 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
259 * Free up the per-ag resources associated with the mount structure.
266 struct xfs_perag
*pag
;
268 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
269 spin_lock(&mp
->m_perag_lock
);
270 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
271 spin_unlock(&mp
->m_perag_lock
);
273 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
274 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
279 * Check size of device based on the (data/realtime) block count.
280 * Note: this check is used by the growfs code as well as mount.
283 xfs_sb_validate_fsb_count(
287 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
288 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
290 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
291 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
293 #else /* Limited by UINT_MAX of sectors */
294 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
301 * Check the validity of the SB found.
304 xfs_mount_validate_sb(
307 bool check_inprogress
)
311 * If the log device and data device have the
312 * same device number, the log is internal.
313 * Consequently, the sb_logstart should be non-zero. If
314 * we have a zero sb_logstart in this case, we may be trying to mount
315 * a volume filesystem in a non-volume manner.
317 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
318 xfs_warn(mp
, "bad magic number");
319 return XFS_ERROR(EWRONGFS
);
322 if (!xfs_sb_good_version(sbp
)) {
323 xfs_warn(mp
, "bad version");
324 return XFS_ERROR(EWRONGFS
);
328 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
330 "filesystem is marked as having an external log; "
331 "specify logdev on the mount command line.");
332 return XFS_ERROR(EINVAL
);
336 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
338 "filesystem is marked as having an internal log; "
339 "do not specify logdev on the mount command line.");
340 return XFS_ERROR(EINVAL
);
344 * More sanity checking. Most of these were stolen directly from
348 sbp
->sb_agcount
<= 0 ||
349 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
350 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
351 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
352 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
353 sbp
->sb_sectsize
!= (1 << sbp
->sb_sectlog
) ||
354 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
355 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
356 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
357 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
358 sbp
->sb_blocksize
!= (1 << sbp
->sb_blocklog
) ||
359 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
360 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
361 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
362 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
363 sbp
->sb_inodesize
!= (1 << sbp
->sb_inodelog
) ||
364 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
365 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
366 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
367 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */) ||
368 sbp
->sb_dblocks
== 0 ||
369 sbp
->sb_dblocks
> XFS_MAX_DBLOCKS(sbp
) ||
370 sbp
->sb_dblocks
< XFS_MIN_DBLOCKS(sbp
))) {
371 XFS_CORRUPTION_ERROR("SB sanity check failed",
372 XFS_ERRLEVEL_LOW
, mp
, sbp
);
373 return XFS_ERROR(EFSCORRUPTED
);
377 * Until this is fixed only page-sized or smaller data blocks work.
379 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
381 "File system with blocksize %d bytes. "
382 "Only pagesize (%ld) or less will currently work.",
383 sbp
->sb_blocksize
, PAGE_SIZE
);
384 return XFS_ERROR(ENOSYS
);
388 * Currently only very few inode sizes are supported.
390 switch (sbp
->sb_inodesize
) {
397 xfs_warn(mp
, "inode size of %d bytes not supported",
399 return XFS_ERROR(ENOSYS
);
402 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
403 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
405 "file system too large to be mounted on this system.");
406 return XFS_ERROR(EFBIG
);
409 if (check_inprogress
&& sbp
->sb_inprogress
) {
410 xfs_warn(mp
, "Offline file system operation in progress!");
411 return XFS_ERROR(EFSCORRUPTED
);
415 * Version 1 directory format has never worked on Linux.
417 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
418 xfs_warn(mp
, "file system using version 1 directory format");
419 return XFS_ERROR(ENOSYS
);
426 xfs_initialize_perag(
428 xfs_agnumber_t agcount
,
429 xfs_agnumber_t
*maxagi
)
431 xfs_agnumber_t index
;
432 xfs_agnumber_t first_initialised
= 0;
436 xfs_sb_t
*sbp
= &mp
->m_sb
;
440 * Walk the current per-ag tree so we don't try to initialise AGs
441 * that already exist (growfs case). Allocate and insert all the
442 * AGs we don't find ready for initialisation.
444 for (index
= 0; index
< agcount
; index
++) {
445 pag
= xfs_perag_get(mp
, index
);
450 if (!first_initialised
)
451 first_initialised
= index
;
453 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
456 pag
->pag_agno
= index
;
458 spin_lock_init(&pag
->pag_ici_lock
);
459 mutex_init(&pag
->pag_ici_reclaim_lock
);
460 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
461 spin_lock_init(&pag
->pag_buf_lock
);
462 pag
->pag_buf_tree
= RB_ROOT
;
464 if (radix_tree_preload(GFP_NOFS
))
467 spin_lock(&mp
->m_perag_lock
);
468 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
470 spin_unlock(&mp
->m_perag_lock
);
471 radix_tree_preload_end();
475 spin_unlock(&mp
->m_perag_lock
);
476 radix_tree_preload_end();
480 * If we mount with the inode64 option, or no inode overflows
481 * the legacy 32-bit address space clear the inode32 option.
483 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
484 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
486 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
487 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
489 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
491 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
)
492 index
= xfs_set_inode32(mp
);
494 index
= xfs_set_inode64(mp
);
502 for (; index
> first_initialised
; index
--) {
503 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
514 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
515 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
516 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
517 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
518 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
519 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
520 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
521 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
522 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
523 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
524 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
525 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
526 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
527 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
528 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
529 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
530 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
531 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
532 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
533 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
534 to
->sb_blocklog
= from
->sb_blocklog
;
535 to
->sb_sectlog
= from
->sb_sectlog
;
536 to
->sb_inodelog
= from
->sb_inodelog
;
537 to
->sb_inopblog
= from
->sb_inopblog
;
538 to
->sb_agblklog
= from
->sb_agblklog
;
539 to
->sb_rextslog
= from
->sb_rextslog
;
540 to
->sb_inprogress
= from
->sb_inprogress
;
541 to
->sb_imax_pct
= from
->sb_imax_pct
;
542 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
543 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
544 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
545 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
546 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
547 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
548 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
549 to
->sb_flags
= from
->sb_flags
;
550 to
->sb_shared_vn
= from
->sb_shared_vn
;
551 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
552 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
553 to
->sb_width
= be32_to_cpu(from
->sb_width
);
554 to
->sb_dirblklog
= from
->sb_dirblklog
;
555 to
->sb_logsectlog
= from
->sb_logsectlog
;
556 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
557 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
558 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
559 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
563 * Copy in core superblock to ondisk one.
565 * The fields argument is mask of superblock fields to copy.
573 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
574 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
584 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
585 first
= xfs_sb_info
[f
].offset
;
586 size
= xfs_sb_info
[f
+ 1].offset
- first
;
588 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
590 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
591 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
595 *(__be16
*)(to_ptr
+ first
) =
596 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
599 *(__be32
*)(to_ptr
+ first
) =
600 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
603 *(__be64
*)(to_ptr
+ first
) =
604 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
611 fields
&= ~(1LL << f
);
619 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
623 xfs_sb_from_disk(&sb
, XFS_BUF_TO_SBP(bp
));
626 * Only check the in progress field for the primary superblock as
627 * mkfs.xfs doesn't clear it from secondary superblocks.
629 error
= xfs_mount_validate_sb(mp
, &sb
, bp
->b_bn
== XFS_SB_DADDR
);
631 xfs_buf_ioerror(bp
, error
);
646 bp
->b_pre_io
= xfs_sb_write_verify
;
648 xfs_buf_ioend(bp
, 0);
652 * We may be probed for a filesystem match, so we may not want to emit
653 * messages when the superblock buffer is not actually an XFS superblock.
654 * If we find an XFS superblock, the run a normal, noisy mount because we are
655 * really going to mount it and want to know about errors.
658 xfs_sb_quiet_read_verify(
663 xfs_sb_from_disk(&sb
, XFS_BUF_TO_SBP(bp
));
665 if (sb
.sb_magicnum
== XFS_SB_MAGIC
) {
666 /* XFS filesystem, verify noisily! */
667 xfs_sb_read_verify(bp
);
671 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
677 * Does the initial read of the superblock.
680 xfs_readsb(xfs_mount_t
*mp
, int flags
)
682 unsigned int sector_size
;
685 int loud
= !(flags
& XFS_MFSI_QUIET
);
687 ASSERT(mp
->m_sb_bp
== NULL
);
688 ASSERT(mp
->m_ddev_targp
!= NULL
);
691 * Allocate a (locked) buffer to hold the superblock.
692 * This will be kept around at all times to optimize
693 * access to the superblock.
695 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
698 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
699 BTOBB(sector_size
), 0,
700 loud
? xfs_sb_read_verify
701 : xfs_sb_quiet_read_verify
);
704 xfs_warn(mp
, "SB buffer read failed");
710 xfs_warn(mp
, "SB validate failed");
715 * Initialize the mount structure from the superblock.
717 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
720 * We must be able to do sector-sized and sector-aligned IO.
722 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
724 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
725 sector_size
, mp
->m_sb
.sb_sectsize
);
731 * If device sector size is smaller than the superblock size,
732 * re-read the superblock so the buffer is correctly sized.
734 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
736 sector_size
= mp
->m_sb
.sb_sectsize
;
740 /* Initialize per-cpu counters */
741 xfs_icsb_reinit_counters(mp
);
756 * Mount initialization code establishing various mount
757 * fields from the superblock associated with the given
761 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
763 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
764 spin_lock_init(&mp
->m_agirotor_lock
);
765 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
766 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
767 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
768 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
769 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
770 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
771 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
772 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
773 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
775 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
776 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
777 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
778 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
780 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
781 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
782 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
783 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
785 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
786 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
787 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
788 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
790 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
791 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
793 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
797 * xfs_initialize_perag_data
799 * Read in each per-ag structure so we can count up the number of
800 * allocated inodes, free inodes and used filesystem blocks as this
801 * information is no longer persistent in the superblock. Once we have
802 * this information, write it into the in-core superblock structure.
805 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
807 xfs_agnumber_t index
;
809 xfs_sb_t
*sbp
= &mp
->m_sb
;
813 uint64_t bfreelst
= 0;
817 for (index
= 0; index
< agcount
; index
++) {
819 * read the agf, then the agi. This gets us
820 * all the information we need and populates the
821 * per-ag structures for us.
823 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
827 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
830 pag
= xfs_perag_get(mp
, index
);
831 ifree
+= pag
->pagi_freecount
;
832 ialloc
+= pag
->pagi_count
;
833 bfree
+= pag
->pagf_freeblks
;
834 bfreelst
+= pag
->pagf_flcount
;
835 btree
+= pag
->pagf_btreeblks
;
839 * Overwrite incore superblock counters with just-read data
841 spin_lock(&mp
->m_sb_lock
);
842 sbp
->sb_ifree
= ifree
;
843 sbp
->sb_icount
= ialloc
;
844 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
845 spin_unlock(&mp
->m_sb_lock
);
847 /* Fixup the per-cpu counters as well. */
848 xfs_icsb_reinit_counters(mp
);
854 * Update alignment values based on mount options and sb values
857 xfs_update_alignment(xfs_mount_t
*mp
)
859 xfs_sb_t
*sbp
= &(mp
->m_sb
);
863 * If stripe unit and stripe width are not multiples
864 * of the fs blocksize turn off alignment.
866 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
867 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
868 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
869 xfs_warn(mp
, "alignment check failed: "
870 "(sunit/swidth vs. blocksize)");
871 return XFS_ERROR(EINVAL
);
873 mp
->m_dalign
= mp
->m_swidth
= 0;
876 * Convert the stripe unit and width to FSBs.
878 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
879 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
880 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
881 xfs_warn(mp
, "alignment check failed: "
882 "(sunit/swidth vs. ag size)");
883 return XFS_ERROR(EINVAL
);
886 "stripe alignment turned off: sunit(%d)/swidth(%d) "
887 "incompatible with agsize(%d)",
888 mp
->m_dalign
, mp
->m_swidth
,
893 } else if (mp
->m_dalign
) {
894 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
896 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
897 xfs_warn(mp
, "alignment check failed: "
898 "sunit(%d) less than bsize(%d)",
901 return XFS_ERROR(EINVAL
);
908 * Update superblock with new values
911 if (xfs_sb_version_hasdalign(sbp
)) {
912 if (sbp
->sb_unit
!= mp
->m_dalign
) {
913 sbp
->sb_unit
= mp
->m_dalign
;
914 mp
->m_update_flags
|= XFS_SB_UNIT
;
916 if (sbp
->sb_width
!= mp
->m_swidth
) {
917 sbp
->sb_width
= mp
->m_swidth
;
918 mp
->m_update_flags
|= XFS_SB_WIDTH
;
921 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
922 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
923 mp
->m_dalign
= sbp
->sb_unit
;
924 mp
->m_swidth
= sbp
->sb_width
;
931 * Set the maximum inode count for this filesystem
934 xfs_set_maxicount(xfs_mount_t
*mp
)
936 xfs_sb_t
*sbp
= &(mp
->m_sb
);
939 if (sbp
->sb_imax_pct
) {
941 * Make sure the maximum inode count is a multiple
942 * of the units we allocate inodes in.
944 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
946 do_div(icount
, mp
->m_ialloc_blks
);
947 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
955 * Set the default minimum read and write sizes unless
956 * already specified in a mount option.
957 * We use smaller I/O sizes when the file system
958 * is being used for NFS service (wsync mount option).
961 xfs_set_rw_sizes(xfs_mount_t
*mp
)
963 xfs_sb_t
*sbp
= &(mp
->m_sb
);
964 int readio_log
, writeio_log
;
966 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
967 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
968 readio_log
= XFS_WSYNC_READIO_LOG
;
969 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
971 readio_log
= XFS_READIO_LOG_LARGE
;
972 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
975 readio_log
= mp
->m_readio_log
;
976 writeio_log
= mp
->m_writeio_log
;
979 if (sbp
->sb_blocklog
> readio_log
) {
980 mp
->m_readio_log
= sbp
->sb_blocklog
;
982 mp
->m_readio_log
= readio_log
;
984 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
985 if (sbp
->sb_blocklog
> writeio_log
) {
986 mp
->m_writeio_log
= sbp
->sb_blocklog
;
988 mp
->m_writeio_log
= writeio_log
;
990 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
994 * precalculate the low space thresholds for dynamic speculative preallocation.
997 xfs_set_low_space_thresholds(
998 struct xfs_mount
*mp
)
1002 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
1003 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
1006 mp
->m_low_space
[i
] = space
* (i
+ 1);
1012 * Set whether we're using inode alignment.
1015 xfs_set_inoalignment(xfs_mount_t
*mp
)
1017 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
1018 mp
->m_sb
.sb_inoalignmt
>=
1019 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
1020 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
1022 mp
->m_inoalign_mask
= 0;
1024 * If we are using stripe alignment, check whether
1025 * the stripe unit is a multiple of the inode alignment
1027 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
1028 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
1029 mp
->m_sinoalign
= mp
->m_dalign
;
1031 mp
->m_sinoalign
= 0;
1035 * Check that the data (and log if separate) are an ok size.
1038 xfs_check_sizes(xfs_mount_t
*mp
)
1043 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
1044 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
1045 xfs_warn(mp
, "filesystem size mismatch detected");
1046 return XFS_ERROR(EFBIG
);
1048 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
1049 d
- XFS_FSS_TO_BB(mp
, 1),
1050 XFS_FSS_TO_BB(mp
, 1), 0, NULL
);
1052 xfs_warn(mp
, "last sector read failed");
1057 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
1058 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
1059 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
1060 xfs_warn(mp
, "log size mismatch detected");
1061 return XFS_ERROR(EFBIG
);
1063 bp
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
1064 d
- XFS_FSB_TO_BB(mp
, 1),
1065 XFS_FSB_TO_BB(mp
, 1), 0, NULL
);
1067 xfs_warn(mp
, "log device read failed");
1076 * Clear the quotaflags in memory and in the superblock.
1079 xfs_mount_reset_sbqflags(
1080 struct xfs_mount
*mp
)
1083 struct xfs_trans
*tp
;
1088 * It is OK to look at sb_qflags here in mount path,
1089 * without m_sb_lock.
1091 if (mp
->m_sb
.sb_qflags
== 0)
1093 spin_lock(&mp
->m_sb_lock
);
1094 mp
->m_sb
.sb_qflags
= 0;
1095 spin_unlock(&mp
->m_sb_lock
);
1098 * If the fs is readonly, let the incore superblock run
1099 * with quotas off but don't flush the update out to disk
1101 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1104 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1105 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1106 XFS_DEFAULT_LOG_COUNT
);
1108 xfs_trans_cancel(tp
, 0);
1109 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
1113 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
1114 return xfs_trans_commit(tp
, 0);
1118 xfs_default_resblks(xfs_mount_t
*mp
)
1123 * We default to 5% or 8192 fsbs of space reserved, whichever is
1124 * smaller. This is intended to cover concurrent allocation
1125 * transactions when we initially hit enospc. These each require a 4
1126 * block reservation. Hence by default we cover roughly 2000 concurrent
1127 * allocation reservations.
1129 resblks
= mp
->m_sb
.sb_dblocks
;
1130 do_div(resblks
, 20);
1131 resblks
= min_t(__uint64_t
, resblks
, 8192);
1136 * This function does the following on an initial mount of a file system:
1137 * - reads the superblock from disk and init the mount struct
1138 * - if we're a 32-bit kernel, do a size check on the superblock
1139 * so we don't mount terabyte filesystems
1140 * - init mount struct realtime fields
1141 * - allocate inode hash table for fs
1142 * - init directory manager
1143 * - perform recovery and init the log manager
1149 xfs_sb_t
*sbp
= &(mp
->m_sb
);
1152 uint quotamount
= 0;
1153 uint quotaflags
= 0;
1156 xfs_mount_common(mp
, sbp
);
1159 * Check for a mismatched features2 values. Older kernels
1160 * read & wrote into the wrong sb offset for sb_features2
1161 * on some platforms due to xfs_sb_t not being 64bit size aligned
1162 * when sb_features2 was added, which made older superblock
1163 * reading/writing routines swap it as a 64-bit value.
1165 * For backwards compatibility, we make both slots equal.
1167 * If we detect a mismatched field, we OR the set bits into the
1168 * existing features2 field in case it has already been modified; we
1169 * don't want to lose any features. We then update the bad location
1170 * with the ORed value so that older kernels will see any features2
1171 * flags, and mark the two fields as needing updates once the
1172 * transaction subsystem is online.
1174 if (xfs_sb_has_mismatched_features2(sbp
)) {
1175 xfs_warn(mp
, "correcting sb_features alignment problem");
1176 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
1177 sbp
->sb_bad_features2
= sbp
->sb_features2
;
1178 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
1181 * Re-check for ATTR2 in case it was found in bad_features2
1184 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1185 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
1186 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1189 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1190 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1191 xfs_sb_version_removeattr2(&mp
->m_sb
);
1192 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
1194 /* update sb_versionnum for the clearing of the morebits */
1195 if (!sbp
->sb_features2
)
1196 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
1200 * Check if sb_agblocks is aligned at stripe boundary
1201 * If sb_agblocks is NOT aligned turn off m_dalign since
1202 * allocator alignment is within an ag, therefore ag has
1203 * to be aligned at stripe boundary.
1205 error
= xfs_update_alignment(mp
);
1209 xfs_alloc_compute_maxlevels(mp
);
1210 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1211 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1212 xfs_ialloc_compute_maxlevels(mp
);
1214 xfs_set_maxicount(mp
);
1216 error
= xfs_uuid_mount(mp
);
1221 * Set the minimum read and write sizes
1223 xfs_set_rw_sizes(mp
);
1225 /* set the low space thresholds for dynamic preallocation */
1226 xfs_set_low_space_thresholds(mp
);
1229 * Set the inode cluster size.
1230 * This may still be overridden by the file system
1231 * block size if it is larger than the chosen cluster size.
1233 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1236 * Set inode alignment fields
1238 xfs_set_inoalignment(mp
);
1241 * Check that the data (and log if separate) are an ok size.
1243 error
= xfs_check_sizes(mp
);
1245 goto out_remove_uuid
;
1248 * Initialize realtime fields in the mount structure
1250 error
= xfs_rtmount_init(mp
);
1252 xfs_warn(mp
, "RT mount failed");
1253 goto out_remove_uuid
;
1257 * Copies the low order bits of the timestamp and the randomly
1258 * set "sequence" number out of a UUID.
1260 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1262 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1267 * Initialize the attribute manager's entries.
1269 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1272 * Initialize the precomputed transaction reservations values.
1277 * Allocate and initialize the per-ag data.
1279 spin_lock_init(&mp
->m_perag_lock
);
1280 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
1281 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
1283 xfs_warn(mp
, "Failed per-ag init: %d", error
);
1284 goto out_remove_uuid
;
1287 if (!sbp
->sb_logblocks
) {
1288 xfs_warn(mp
, "no log defined");
1289 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1290 error
= XFS_ERROR(EFSCORRUPTED
);
1291 goto out_free_perag
;
1295 * log's mount-time initialization. Perform 1st part recovery if needed
1297 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1298 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1299 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1301 xfs_warn(mp
, "log mount failed");
1306 * Now the log is mounted, we know if it was an unclean shutdown or
1307 * not. If it was, with the first phase of recovery has completed, we
1308 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1309 * but they are recovered transactionally in the second recovery phase
1312 * Hence we can safely re-initialise incore superblock counters from
1313 * the per-ag data. These may not be correct if the filesystem was not
1314 * cleanly unmounted, so we need to wait for recovery to finish before
1317 * If the filesystem was cleanly unmounted, then we can trust the
1318 * values in the superblock to be correct and we don't need to do
1321 * If we are currently making the filesystem, the initialisation will
1322 * fail as the perag data is in an undefined state.
1324 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1325 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1326 !mp
->m_sb
.sb_inprogress
) {
1327 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1333 * Get and sanity-check the root inode.
1334 * Save the pointer to it in the mount structure.
1336 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
1338 xfs_warn(mp
, "failed to read root inode");
1339 goto out_log_dealloc
;
1342 ASSERT(rip
!= NULL
);
1344 if (unlikely(!S_ISDIR(rip
->i_d
.di_mode
))) {
1345 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
1346 (unsigned long long)rip
->i_ino
);
1347 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1348 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1350 error
= XFS_ERROR(EFSCORRUPTED
);
1353 mp
->m_rootip
= rip
; /* save it */
1355 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1358 * Initialize realtime inode pointers in the mount structure
1360 error
= xfs_rtmount_inodes(mp
);
1363 * Free up the root inode.
1365 xfs_warn(mp
, "failed to read RT inodes");
1370 * If this is a read-only mount defer the superblock updates until
1371 * the next remount into writeable mode. Otherwise we would never
1372 * perform the update e.g. for the root filesystem.
1374 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1375 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1377 xfs_warn(mp
, "failed to write sb changes");
1383 * Initialise the XFS quota management subsystem for this mount
1385 if (XFS_IS_QUOTA_RUNNING(mp
)) {
1386 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
1390 ASSERT(!XFS_IS_QUOTA_ON(mp
));
1393 * If a file system had quotas running earlier, but decided to
1394 * mount without -o uquota/pquota/gquota options, revoke the
1395 * quotachecked license.
1397 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
1398 xfs_notice(mp
, "resetting quota flags");
1399 error
= xfs_mount_reset_sbqflags(mp
);
1406 * Finish recovering the file system. This part needed to be
1407 * delayed until after the root and real-time bitmap inodes
1408 * were consistently read in.
1410 error
= xfs_log_mount_finish(mp
);
1412 xfs_warn(mp
, "log mount finish failed");
1417 * Complete the quota initialisation, post-log-replay component.
1420 ASSERT(mp
->m_qflags
== 0);
1421 mp
->m_qflags
= quotaflags
;
1423 xfs_qm_mount_quotas(mp
);
1427 * Now we are mounted, reserve a small amount of unused space for
1428 * privileged transactions. This is needed so that transaction
1429 * space required for critical operations can dip into this pool
1430 * when at ENOSPC. This is needed for operations like create with
1431 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1432 * are not allowed to use this reserved space.
1434 * This may drive us straight to ENOSPC on mount, but that implies
1435 * we were already there on the last unmount. Warn if this occurs.
1437 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1438 resblks
= xfs_default_resblks(mp
);
1439 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1442 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1448 xfs_rtunmount_inodes(mp
);
1452 xfs_log_unmount(mp
);
1454 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1455 xfs_wait_buftarg(mp
->m_logdev_targp
);
1456 xfs_wait_buftarg(mp
->m_ddev_targp
);
1460 xfs_uuid_unmount(mp
);
1466 * This flushes out the inodes,dquots and the superblock, unmounts the
1467 * log and makes sure that incore structures are freed.
1471 struct xfs_mount
*mp
)
1476 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
1478 xfs_qm_unmount_quotas(mp
);
1479 xfs_rtunmount_inodes(mp
);
1480 IRELE(mp
->m_rootip
);
1483 * We can potentially deadlock here if we have an inode cluster
1484 * that has been freed has its buffer still pinned in memory because
1485 * the transaction is still sitting in a iclog. The stale inodes
1486 * on that buffer will have their flush locks held until the
1487 * transaction hits the disk and the callbacks run. the inode
1488 * flush takes the flush lock unconditionally and with nothing to
1489 * push out the iclog we will never get that unlocked. hence we
1490 * need to force the log first.
1492 xfs_log_force(mp
, XFS_LOG_SYNC
);
1495 * Flush all pending changes from the AIL.
1497 xfs_ail_push_all_sync(mp
->m_ail
);
1500 * And reclaim all inodes. At this point there should be no dirty
1501 * inodes and none should be pinned or locked, but use synchronous
1502 * reclaim just to be sure. We can stop background inode reclaim
1503 * here as well if it is still running.
1505 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
1506 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1511 * Unreserve any blocks we have so that when we unmount we don't account
1512 * the reserved free space as used. This is really only necessary for
1513 * lazy superblock counting because it trusts the incore superblock
1514 * counters to be absolutely correct on clean unmount.
1516 * We don't bother correcting this elsewhere for lazy superblock
1517 * counting because on mount of an unclean filesystem we reconstruct the
1518 * correct counter value and this is irrelevant.
1520 * For non-lazy counter filesystems, this doesn't matter at all because
1521 * we only every apply deltas to the superblock and hence the incore
1522 * value does not matter....
1525 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1527 xfs_warn(mp
, "Unable to free reserved block pool. "
1528 "Freespace may not be correct on next mount.");
1530 error
= xfs_log_sbcount(mp
);
1532 xfs_warn(mp
, "Unable to update superblock counters. "
1533 "Freespace may not be correct on next mount.");
1535 xfs_log_unmount(mp
);
1536 xfs_uuid_unmount(mp
);
1539 xfs_errortag_clearall(mp
, 0);
1545 xfs_fs_writable(xfs_mount_t
*mp
)
1547 return !(mp
->m_super
->s_writers
.frozen
|| XFS_FORCED_SHUTDOWN(mp
) ||
1548 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1554 * Sync the superblock counters to disk.
1556 * Note this code can be called during the process of freezing, so
1557 * we may need to use the transaction allocator which does not
1558 * block when the transaction subsystem is in its frozen state.
1561 xfs_log_sbcount(xfs_mount_t
*mp
)
1566 if (!xfs_fs_writable(mp
))
1569 xfs_icsb_sync_counters(mp
, 0);
1572 * we don't need to do this if we are updating the superblock
1573 * counters on every modification.
1575 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1578 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1579 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1580 XFS_DEFAULT_LOG_COUNT
);
1582 xfs_trans_cancel(tp
, 0);
1586 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1587 xfs_trans_set_sync(tp
);
1588 error
= xfs_trans_commit(tp
, 0);
1593 * xfs_mod_sb() can be used to copy arbitrary changes to the
1594 * in-core superblock into the superblock buffer to be logged.
1595 * It does not provide the higher level of locking that is
1596 * needed to protect the in-core superblock from concurrent
1600 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1612 bp
= xfs_trans_getsb(tp
, mp
, 0);
1613 first
= sizeof(xfs_sb_t
);
1616 /* translate/copy */
1618 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1620 /* find modified range */
1621 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1622 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1623 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1625 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1626 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1627 first
= xfs_sb_info
[f
].offset
;
1629 xfs_trans_log_buf(tp
, bp
, first
, last
);
1634 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1635 * a delta to a specified field in the in-core superblock. Simply
1636 * switch on the field indicated and apply the delta to that field.
1637 * Fields are not allowed to dip below zero, so if the delta would
1638 * do this do not apply it and return EINVAL.
1640 * The m_sb_lock must be held when this routine is called.
1643 xfs_mod_incore_sb_unlocked(
1645 xfs_sb_field_t field
,
1649 int scounter
; /* short counter for 32 bit fields */
1650 long long lcounter
; /* long counter for 64 bit fields */
1651 long long res_used
, rem
;
1654 * With the in-core superblock spin lock held, switch
1655 * on the indicated field. Apply the delta to the
1656 * proper field. If the fields value would dip below
1657 * 0, then do not apply the delta and return EINVAL.
1660 case XFS_SBS_ICOUNT
:
1661 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1665 return XFS_ERROR(EINVAL
);
1667 mp
->m_sb
.sb_icount
= lcounter
;
1670 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1674 return XFS_ERROR(EINVAL
);
1676 mp
->m_sb
.sb_ifree
= lcounter
;
1678 case XFS_SBS_FDBLOCKS
:
1679 lcounter
= (long long)
1680 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1681 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1683 if (delta
> 0) { /* Putting blocks back */
1684 if (res_used
> delta
) {
1685 mp
->m_resblks_avail
+= delta
;
1687 rem
= delta
- res_used
;
1688 mp
->m_resblks_avail
= mp
->m_resblks
;
1691 } else { /* Taking blocks away */
1693 if (lcounter
>= 0) {
1694 mp
->m_sb
.sb_fdblocks
= lcounter
+
1695 XFS_ALLOC_SET_ASIDE(mp
);
1700 * We are out of blocks, use any available reserved
1701 * blocks if were allowed to.
1704 return XFS_ERROR(ENOSPC
);
1706 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1707 if (lcounter
>= 0) {
1708 mp
->m_resblks_avail
= lcounter
;
1711 printk_once(KERN_WARNING
1712 "Filesystem \"%s\": reserve blocks depleted! "
1713 "Consider increasing reserve pool size.",
1715 return XFS_ERROR(ENOSPC
);
1718 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1720 case XFS_SBS_FREXTENTS
:
1721 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1724 return XFS_ERROR(ENOSPC
);
1726 mp
->m_sb
.sb_frextents
= lcounter
;
1728 case XFS_SBS_DBLOCKS
:
1729 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1733 return XFS_ERROR(EINVAL
);
1735 mp
->m_sb
.sb_dblocks
= lcounter
;
1737 case XFS_SBS_AGCOUNT
:
1738 scounter
= mp
->m_sb
.sb_agcount
;
1742 return XFS_ERROR(EINVAL
);
1744 mp
->m_sb
.sb_agcount
= scounter
;
1746 case XFS_SBS_IMAX_PCT
:
1747 scounter
= mp
->m_sb
.sb_imax_pct
;
1751 return XFS_ERROR(EINVAL
);
1753 mp
->m_sb
.sb_imax_pct
= scounter
;
1755 case XFS_SBS_REXTSIZE
:
1756 scounter
= mp
->m_sb
.sb_rextsize
;
1760 return XFS_ERROR(EINVAL
);
1762 mp
->m_sb
.sb_rextsize
= scounter
;
1764 case XFS_SBS_RBMBLOCKS
:
1765 scounter
= mp
->m_sb
.sb_rbmblocks
;
1769 return XFS_ERROR(EINVAL
);
1771 mp
->m_sb
.sb_rbmblocks
= scounter
;
1773 case XFS_SBS_RBLOCKS
:
1774 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1778 return XFS_ERROR(EINVAL
);
1780 mp
->m_sb
.sb_rblocks
= lcounter
;
1782 case XFS_SBS_REXTENTS
:
1783 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1787 return XFS_ERROR(EINVAL
);
1789 mp
->m_sb
.sb_rextents
= lcounter
;
1791 case XFS_SBS_REXTSLOG
:
1792 scounter
= mp
->m_sb
.sb_rextslog
;
1796 return XFS_ERROR(EINVAL
);
1798 mp
->m_sb
.sb_rextslog
= scounter
;
1802 return XFS_ERROR(EINVAL
);
1807 * xfs_mod_incore_sb() is used to change a field in the in-core
1808 * superblock structure by the specified delta. This modification
1809 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1810 * routine to do the work.
1814 struct xfs_mount
*mp
,
1815 xfs_sb_field_t field
,
1821 #ifdef HAVE_PERCPU_SB
1822 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1824 spin_lock(&mp
->m_sb_lock
);
1825 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1826 spin_unlock(&mp
->m_sb_lock
);
1832 * Change more than one field in the in-core superblock structure at a time.
1834 * The fields and changes to those fields are specified in the array of
1835 * xfs_mod_sb structures passed in. Either all of the specified deltas
1836 * will be applied or none of them will. If any modified field dips below 0,
1837 * then all modifications will be backed out and EINVAL will be returned.
1839 * Note that this function may not be used for the superblock values that
1840 * are tracked with the in-memory per-cpu counters - a direct call to
1841 * xfs_icsb_modify_counters is required for these.
1844 xfs_mod_incore_sb_batch(
1845 struct xfs_mount
*mp
,
1854 * Loop through the array of mod structures and apply each individually.
1855 * If any fail, then back out all those which have already been applied.
1856 * Do all of this within the scope of the m_sb_lock so that all of the
1857 * changes will be atomic.
1859 spin_lock(&mp
->m_sb_lock
);
1860 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1861 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1862 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1864 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1865 msbp
->msb_delta
, rsvd
);
1869 spin_unlock(&mp
->m_sb_lock
);
1873 while (--msbp
>= msb
) {
1874 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1875 -msbp
->msb_delta
, rsvd
);
1878 spin_unlock(&mp
->m_sb_lock
);
1883 * xfs_getsb() is called to obtain the buffer for the superblock.
1884 * The buffer is returned locked and read in from disk.
1885 * The buffer should be released with a call to xfs_brelse().
1887 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1888 * the superblock buffer if it can be locked without sleeping.
1889 * If it can't then we'll return NULL.
1893 struct xfs_mount
*mp
,
1896 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1898 if (!xfs_buf_trylock(bp
)) {
1899 if (flags
& XBF_TRYLOCK
)
1905 ASSERT(XFS_BUF_ISDONE(bp
));
1910 * Used to free the superblock along various error paths.
1914 struct xfs_mount
*mp
)
1916 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1924 * Used to log changes to the superblock unit and width fields which could
1925 * be altered by the mount options, as well as any potential sb_features2
1926 * fixup. Only the first superblock is updated.
1936 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1937 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1938 XFS_SB_VERSIONNUM
));
1940 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1941 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1942 XFS_DEFAULT_LOG_COUNT
);
1944 xfs_trans_cancel(tp
, 0);
1947 xfs_mod_sb(tp
, fields
);
1948 error
= xfs_trans_commit(tp
, 0);
1953 * If the underlying (data/log/rt) device is readonly, there are some
1954 * operations that cannot proceed.
1957 xfs_dev_is_read_only(
1958 struct xfs_mount
*mp
,
1961 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1962 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1963 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1964 xfs_notice(mp
, "%s required on read-only device.", message
);
1965 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1971 #ifdef HAVE_PERCPU_SB
1973 * Per-cpu incore superblock counters
1975 * Simple concept, difficult implementation
1977 * Basically, replace the incore superblock counters with a distributed per cpu
1978 * counter for contended fields (e.g. free block count).
1980 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1981 * hence needs to be accurately read when we are running low on space. Hence
1982 * there is a method to enable and disable the per-cpu counters based on how
1983 * much "stuff" is available in them.
1985 * Basically, a counter is enabled if there is enough free resource to justify
1986 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1987 * ENOSPC), then we disable the counters to synchronise all callers and
1988 * re-distribute the available resources.
1990 * If, once we redistributed the available resources, we still get a failure,
1991 * we disable the per-cpu counter and go through the slow path.
1993 * The slow path is the current xfs_mod_incore_sb() function. This means that
1994 * when we disable a per-cpu counter, we need to drain its resources back to
1995 * the global superblock. We do this after disabling the counter to prevent
1996 * more threads from queueing up on the counter.
1998 * Essentially, this means that we still need a lock in the fast path to enable
1999 * synchronisation between the global counters and the per-cpu counters. This
2000 * is not a problem because the lock will be local to a CPU almost all the time
2001 * and have little contention except when we get to ENOSPC conditions.
2003 * Basically, this lock becomes a barrier that enables us to lock out the fast
2004 * path while we do things like enabling and disabling counters and
2005 * synchronising the counters.
2009 * 1. m_sb_lock before picking up per-cpu locks
2010 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2011 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2012 * 4. modifying per-cpu counters requires holding per-cpu lock
2013 * 5. modifying global counters requires holding m_sb_lock
2014 * 6. enabling or disabling a counter requires holding the m_sb_lock
2015 * and _none_ of the per-cpu locks.
2017 * Disabled counters are only ever re-enabled by a balance operation
2018 * that results in more free resources per CPU than a given threshold.
2019 * To ensure counters don't remain disabled, they are rebalanced when
2020 * the global resource goes above a higher threshold (i.e. some hysteresis
2021 * is present to prevent thrashing).
2024 #ifdef CONFIG_HOTPLUG_CPU
2026 * hot-plug CPU notifier support.
2028 * We need a notifier per filesystem as we need to be able to identify
2029 * the filesystem to balance the counters out. This is achieved by
2030 * having a notifier block embedded in the xfs_mount_t and doing pointer
2031 * magic to get the mount pointer from the notifier block address.
2034 xfs_icsb_cpu_notify(
2035 struct notifier_block
*nfb
,
2036 unsigned long action
,
2039 xfs_icsb_cnts_t
*cntp
;
2042 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2043 cntp
= (xfs_icsb_cnts_t
*)
2044 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2046 case CPU_UP_PREPARE
:
2047 case CPU_UP_PREPARE_FROZEN
:
2048 /* Easy Case - initialize the area and locks, and
2049 * then rebalance when online does everything else for us. */
2050 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2053 case CPU_ONLINE_FROZEN
:
2055 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2056 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2057 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2058 xfs_icsb_unlock(mp
);
2061 case CPU_DEAD_FROZEN
:
2062 /* Disable all the counters, then fold the dead cpu's
2063 * count into the total on the global superblock and
2064 * re-enable the counters. */
2066 spin_lock(&mp
->m_sb_lock
);
2067 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2068 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2069 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2071 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2072 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2073 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2075 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2077 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
2078 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
2079 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
2080 spin_unlock(&mp
->m_sb_lock
);
2081 xfs_icsb_unlock(mp
);
2087 #endif /* CONFIG_HOTPLUG_CPU */
2090 xfs_icsb_init_counters(
2093 xfs_icsb_cnts_t
*cntp
;
2096 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2097 if (mp
->m_sb_cnts
== NULL
)
2100 #ifdef CONFIG_HOTPLUG_CPU
2101 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2102 mp
->m_icsb_notifier
.priority
= 0;
2103 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2104 #endif /* CONFIG_HOTPLUG_CPU */
2106 for_each_online_cpu(i
) {
2107 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2108 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2111 mutex_init(&mp
->m_icsb_mutex
);
2114 * start with all counters disabled so that the
2115 * initial balance kicks us off correctly
2117 mp
->m_icsb_counters
= -1;
2122 xfs_icsb_reinit_counters(
2127 * start with all counters disabled so that the
2128 * initial balance kicks us off correctly
2130 mp
->m_icsb_counters
= -1;
2131 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2132 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2133 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2134 xfs_icsb_unlock(mp
);
2138 xfs_icsb_destroy_counters(
2141 if (mp
->m_sb_cnts
) {
2142 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2143 free_percpu(mp
->m_sb_cnts
);
2145 mutex_destroy(&mp
->m_icsb_mutex
);
2150 xfs_icsb_cnts_t
*icsbp
)
2152 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2158 xfs_icsb_unlock_cntr(
2159 xfs_icsb_cnts_t
*icsbp
)
2161 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2166 xfs_icsb_lock_all_counters(
2169 xfs_icsb_cnts_t
*cntp
;
2172 for_each_online_cpu(i
) {
2173 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2174 xfs_icsb_lock_cntr(cntp
);
2179 xfs_icsb_unlock_all_counters(
2182 xfs_icsb_cnts_t
*cntp
;
2185 for_each_online_cpu(i
) {
2186 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2187 xfs_icsb_unlock_cntr(cntp
);
2194 xfs_icsb_cnts_t
*cnt
,
2197 xfs_icsb_cnts_t
*cntp
;
2200 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2202 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2203 xfs_icsb_lock_all_counters(mp
);
2205 for_each_online_cpu(i
) {
2206 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2207 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2208 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2209 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2212 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2213 xfs_icsb_unlock_all_counters(mp
);
2217 xfs_icsb_counter_disabled(
2219 xfs_sb_field_t field
)
2221 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2222 return test_bit(field
, &mp
->m_icsb_counters
);
2226 xfs_icsb_disable_counter(
2228 xfs_sb_field_t field
)
2230 xfs_icsb_cnts_t cnt
;
2232 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2235 * If we are already disabled, then there is nothing to do
2236 * here. We check before locking all the counters to avoid
2237 * the expensive lock operation when being called in the
2238 * slow path and the counter is already disabled. This is
2239 * safe because the only time we set or clear this state is under
2242 if (xfs_icsb_counter_disabled(mp
, field
))
2245 xfs_icsb_lock_all_counters(mp
);
2246 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2247 /* drain back to superblock */
2249 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2251 case XFS_SBS_ICOUNT
:
2252 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2255 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2257 case XFS_SBS_FDBLOCKS
:
2258 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2265 xfs_icsb_unlock_all_counters(mp
);
2269 xfs_icsb_enable_counter(
2271 xfs_sb_field_t field
,
2275 xfs_icsb_cnts_t
*cntp
;
2278 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2280 xfs_icsb_lock_all_counters(mp
);
2281 for_each_online_cpu(i
) {
2282 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2284 case XFS_SBS_ICOUNT
:
2285 cntp
->icsb_icount
= count
+ resid
;
2288 cntp
->icsb_ifree
= count
+ resid
;
2290 case XFS_SBS_FDBLOCKS
:
2291 cntp
->icsb_fdblocks
= count
+ resid
;
2299 clear_bit(field
, &mp
->m_icsb_counters
);
2300 xfs_icsb_unlock_all_counters(mp
);
2304 xfs_icsb_sync_counters_locked(
2308 xfs_icsb_cnts_t cnt
;
2310 xfs_icsb_count(mp
, &cnt
, flags
);
2312 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2313 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2314 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2315 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2316 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2317 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2321 * Accurate update of per-cpu counters to incore superblock
2324 xfs_icsb_sync_counters(
2328 spin_lock(&mp
->m_sb_lock
);
2329 xfs_icsb_sync_counters_locked(mp
, flags
);
2330 spin_unlock(&mp
->m_sb_lock
);
2334 * Balance and enable/disable counters as necessary.
2336 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2337 * chosen to be the same number as single on disk allocation chunk per CPU, and
2338 * free blocks is something far enough zero that we aren't going thrash when we
2339 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2340 * prevent looping endlessly when xfs_alloc_space asks for more than will
2341 * be distributed to a single CPU but each CPU has enough blocks to be
2344 * Note that we can be called when counters are already disabled.
2345 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2346 * prevent locking every per-cpu counter needlessly.
2349 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2350 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2351 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2353 xfs_icsb_balance_counter_locked(
2355 xfs_sb_field_t field
,
2358 uint64_t count
, resid
;
2359 int weight
= num_online_cpus();
2360 uint64_t min
= (uint64_t)min_per_cpu
;
2362 /* disable counter and sync counter */
2363 xfs_icsb_disable_counter(mp
, field
);
2365 /* update counters - first CPU gets residual*/
2367 case XFS_SBS_ICOUNT
:
2368 count
= mp
->m_sb
.sb_icount
;
2369 resid
= do_div(count
, weight
);
2370 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2374 count
= mp
->m_sb
.sb_ifree
;
2375 resid
= do_div(count
, weight
);
2376 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2379 case XFS_SBS_FDBLOCKS
:
2380 count
= mp
->m_sb
.sb_fdblocks
;
2381 resid
= do_div(count
, weight
);
2382 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2387 count
= resid
= 0; /* quiet, gcc */
2391 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2395 xfs_icsb_balance_counter(
2397 xfs_sb_field_t fields
,
2400 spin_lock(&mp
->m_sb_lock
);
2401 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2402 spin_unlock(&mp
->m_sb_lock
);
2406 xfs_icsb_modify_counters(
2408 xfs_sb_field_t field
,
2412 xfs_icsb_cnts_t
*icsbp
;
2413 long long lcounter
; /* long counter for 64 bit fields */
2419 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
2422 * if the counter is disabled, go to slow path
2424 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2426 xfs_icsb_lock_cntr(icsbp
);
2427 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2428 xfs_icsb_unlock_cntr(icsbp
);
2433 case XFS_SBS_ICOUNT
:
2434 lcounter
= icsbp
->icsb_icount
;
2436 if (unlikely(lcounter
< 0))
2437 goto balance_counter
;
2438 icsbp
->icsb_icount
= lcounter
;
2442 lcounter
= icsbp
->icsb_ifree
;
2444 if (unlikely(lcounter
< 0))
2445 goto balance_counter
;
2446 icsbp
->icsb_ifree
= lcounter
;
2449 case XFS_SBS_FDBLOCKS
:
2450 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2452 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2454 if (unlikely(lcounter
< 0))
2455 goto balance_counter
;
2456 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2462 xfs_icsb_unlock_cntr(icsbp
);
2470 * serialise with a mutex so we don't burn lots of cpu on
2471 * the superblock lock. We still need to hold the superblock
2472 * lock, however, when we modify the global structures.
2477 * Now running atomically.
2479 * If the counter is enabled, someone has beaten us to rebalancing.
2480 * Drop the lock and try again in the fast path....
2482 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2483 xfs_icsb_unlock(mp
);
2488 * The counter is currently disabled. Because we are
2489 * running atomically here, we know a rebalance cannot
2490 * be in progress. Hence we can go straight to operating
2491 * on the global superblock. We do not call xfs_mod_incore_sb()
2492 * here even though we need to get the m_sb_lock. Doing so
2493 * will cause us to re-enter this function and deadlock.
2494 * Hence we get the m_sb_lock ourselves and then call
2495 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2496 * directly on the global counters.
2498 spin_lock(&mp
->m_sb_lock
);
2499 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2500 spin_unlock(&mp
->m_sb_lock
);
2503 * Now that we've modified the global superblock, we
2504 * may be able to re-enable the distributed counters
2505 * (e.g. lots of space just got freed). After that
2509 xfs_icsb_balance_counter(mp
, field
, 0);
2510 xfs_icsb_unlock(mp
);
2514 xfs_icsb_unlock_cntr(icsbp
);
2518 * We may have multiple threads here if multiple per-cpu
2519 * counters run dry at the same time. This will mean we can
2520 * do more balances than strictly necessary but it is not
2521 * the common slowpath case.
2526 * running atomically.
2528 * This will leave the counter in the correct state for future
2529 * accesses. After the rebalance, we simply try again and our retry
2530 * will either succeed through the fast path or slow path without
2531 * another balance operation being required.
2533 xfs_icsb_balance_counter(mp
, field
, delta
);
2534 xfs_icsb_unlock(mp
);