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"
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)
61 short type
; /* 0 = integer
62 * 1 = binary / string (no translation)
65 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
66 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
67 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
68 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
69 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
70 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
71 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
72 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
73 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
74 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
75 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
76 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
77 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
78 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
79 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
80 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
81 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
82 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
83 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
84 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
85 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
86 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
87 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
88 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
89 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
90 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
91 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
92 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
93 { offsetof(xfs_sb_t
, sb_icount
), 0 },
94 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
95 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
96 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
97 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
98 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
99 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
100 { offsetof(xfs_sb_t
, sb_flags
), 0 },
101 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
102 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
103 { offsetof(xfs_sb_t
, sb_unit
), 0 },
104 { offsetof(xfs_sb_t
, sb_width
), 0 },
105 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
106 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
107 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
108 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
109 { offsetof(xfs_sb_t
, sb_features2
), 0 },
110 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
111 { sizeof(xfs_sb_t
), 0 }
114 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
115 static int xfs_uuid_table_size
;
116 static uuid_t
*xfs_uuid_table
;
119 * See if the UUID is unique among mounted XFS filesystems.
120 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
124 struct xfs_mount
*mp
)
126 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
129 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
132 if (uuid_is_nil(uuid
)) {
133 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
134 return XFS_ERROR(EINVAL
);
137 mutex_lock(&xfs_uuid_table_mutex
);
138 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
139 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
143 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
148 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
149 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
150 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
152 hole
= xfs_uuid_table_size
++;
154 xfs_uuid_table
[hole
] = *uuid
;
155 mutex_unlock(&xfs_uuid_table_mutex
);
160 mutex_unlock(&xfs_uuid_table_mutex
);
161 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
162 return XFS_ERROR(EINVAL
);
167 struct xfs_mount
*mp
)
169 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
172 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
175 mutex_lock(&xfs_uuid_table_mutex
);
176 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
177 if (uuid_is_nil(&xfs_uuid_table
[i
]))
179 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
181 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
184 ASSERT(i
< xfs_uuid_table_size
);
185 mutex_unlock(&xfs_uuid_table_mutex
);
190 * Reference counting access wrappers to the perag structures.
191 * Because we never free per-ag structures, the only thing we
192 * have to protect against changes is the tree structure itself.
195 xfs_perag_get(struct xfs_mount
*mp
, xfs_agnumber_t agno
)
197 struct xfs_perag
*pag
;
201 pag
= radix_tree_lookup(&mp
->m_perag_tree
, agno
);
203 ASSERT(atomic_read(&pag
->pag_ref
) >= 0);
204 ref
= atomic_inc_return(&pag
->pag_ref
);
207 trace_xfs_perag_get(mp
, agno
, ref
, _RET_IP_
);
212 * search from @first to find the next perag with the given tag set.
216 struct xfs_mount
*mp
,
217 xfs_agnumber_t first
,
220 struct xfs_perag
*pag
;
225 found
= radix_tree_gang_lookup_tag(&mp
->m_perag_tree
,
226 (void **)&pag
, first
, 1, tag
);
231 ref
= atomic_inc_return(&pag
->pag_ref
);
233 trace_xfs_perag_get_tag(mp
, pag
->pag_agno
, ref
, _RET_IP_
);
238 xfs_perag_put(struct xfs_perag
*pag
)
242 ASSERT(atomic_read(&pag
->pag_ref
) > 0);
243 ref
= atomic_dec_return(&pag
->pag_ref
);
244 trace_xfs_perag_put(pag
->pag_mount
, pag
->pag_agno
, ref
, _RET_IP_
);
249 struct rcu_head
*head
)
251 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
253 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
258 * Free up the per-ag resources associated with the mount structure.
265 struct xfs_perag
*pag
;
267 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
268 spin_lock(&mp
->m_perag_lock
);
269 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
270 spin_unlock(&mp
->m_perag_lock
);
272 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
273 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
278 * Check size of device based on the (data/realtime) block count.
279 * Note: this check is used by the growfs code as well as mount.
282 xfs_sb_validate_fsb_count(
286 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
287 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
289 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
290 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
292 #else /* Limited by UINT_MAX of sectors */
293 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
300 * Check the validity of the SB found.
303 xfs_mount_validate_sb(
308 int loud
= !(flags
& XFS_MFSI_QUIET
);
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
) {
319 xfs_warn(mp
, "bad magic number");
320 return XFS_ERROR(EWRONGFS
);
323 if (!xfs_sb_good_version(sbp
)) {
325 xfs_warn(mp
, "bad version");
326 return XFS_ERROR(EWRONGFS
);
330 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
333 "filesystem is marked as having an external log; "
334 "specify logdev on the mount command line.");
335 return XFS_ERROR(EINVAL
);
339 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
342 "filesystem is marked as having an internal log; "
343 "do not specify logdev on the mount command line.");
344 return XFS_ERROR(EINVAL
);
348 * More sanity checking. Most of these were stolen directly from
352 sbp
->sb_agcount
<= 0 ||
353 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
354 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
355 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
356 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
357 sbp
->sb_sectsize
!= (1 << sbp
->sb_sectlog
) ||
358 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
359 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
360 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
361 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
362 sbp
->sb_blocksize
!= (1 << sbp
->sb_blocklog
) ||
363 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
364 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
365 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
366 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
367 sbp
->sb_inodesize
!= (1 << sbp
->sb_inodelog
) ||
368 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
369 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
370 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
371 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */) ||
372 sbp
->sb_dblocks
== 0 ||
373 sbp
->sb_dblocks
> XFS_MAX_DBLOCKS(sbp
) ||
374 sbp
->sb_dblocks
< XFS_MIN_DBLOCKS(sbp
))) {
376 XFS_CORRUPTION_ERROR("SB sanity check failed",
377 XFS_ERRLEVEL_LOW
, mp
, sbp
);
378 return XFS_ERROR(EFSCORRUPTED
);
382 * Until this is fixed only page-sized or smaller data blocks work.
384 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
387 "File system with blocksize %d bytes. "
388 "Only pagesize (%ld) or less will currently work.",
389 sbp
->sb_blocksize
, PAGE_SIZE
);
391 return XFS_ERROR(ENOSYS
);
395 * Currently only very few inode sizes are supported.
397 switch (sbp
->sb_inodesize
) {
405 xfs_warn(mp
, "inode size of %d bytes not supported",
407 return XFS_ERROR(ENOSYS
);
410 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
411 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
414 "file system too large to be mounted on this system.");
415 return XFS_ERROR(EFBIG
);
418 if (unlikely(sbp
->sb_inprogress
)) {
420 xfs_warn(mp
, "file system busy");
421 return XFS_ERROR(EFSCORRUPTED
);
425 * Version 1 directory format has never worked on Linux.
427 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
430 "file system using version 1 directory format");
431 return XFS_ERROR(ENOSYS
);
438 xfs_initialize_perag(
440 xfs_agnumber_t agcount
,
441 xfs_agnumber_t
*maxagi
)
443 xfs_agnumber_t index
;
444 xfs_agnumber_t first_initialised
= 0;
448 xfs_sb_t
*sbp
= &mp
->m_sb
;
452 * Walk the current per-ag tree so we don't try to initialise AGs
453 * that already exist (growfs case). Allocate and insert all the
454 * AGs we don't find ready for initialisation.
456 for (index
= 0; index
< agcount
; index
++) {
457 pag
= xfs_perag_get(mp
, index
);
462 if (!first_initialised
)
463 first_initialised
= index
;
465 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
468 pag
->pag_agno
= index
;
470 spin_lock_init(&pag
->pag_ici_lock
);
471 mutex_init(&pag
->pag_ici_reclaim_lock
);
472 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
473 spin_lock_init(&pag
->pag_buf_lock
);
474 pag
->pag_buf_tree
= RB_ROOT
;
476 if (radix_tree_preload(GFP_NOFS
))
479 spin_lock(&mp
->m_perag_lock
);
480 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
482 spin_unlock(&mp
->m_perag_lock
);
483 radix_tree_preload_end();
487 spin_unlock(&mp
->m_perag_lock
);
488 radix_tree_preload_end();
492 * If we mount with the inode64 option, or no inode overflows
493 * the legacy 32-bit address space clear the inode32 option.
495 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
496 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
498 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
499 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
501 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
503 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
)
504 index
= xfs_set_inode32(mp
);
506 index
= xfs_set_inode64(mp
);
514 for (; index
> first_initialised
; index
--) {
515 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
523 struct xfs_mount
*mp
,
526 struct xfs_sb
*to
= &mp
->m_sb
;
528 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
529 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
530 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
531 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
532 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
533 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
534 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
535 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
536 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
537 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
538 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
539 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
540 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
541 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
542 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
543 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
544 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
545 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
546 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
547 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
548 to
->sb_blocklog
= from
->sb_blocklog
;
549 to
->sb_sectlog
= from
->sb_sectlog
;
550 to
->sb_inodelog
= from
->sb_inodelog
;
551 to
->sb_inopblog
= from
->sb_inopblog
;
552 to
->sb_agblklog
= from
->sb_agblklog
;
553 to
->sb_rextslog
= from
->sb_rextslog
;
554 to
->sb_inprogress
= from
->sb_inprogress
;
555 to
->sb_imax_pct
= from
->sb_imax_pct
;
556 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
557 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
558 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
559 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
560 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
561 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
562 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
563 to
->sb_flags
= from
->sb_flags
;
564 to
->sb_shared_vn
= from
->sb_shared_vn
;
565 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
566 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
567 to
->sb_width
= be32_to_cpu(from
->sb_width
);
568 to
->sb_dirblklog
= from
->sb_dirblklog
;
569 to
->sb_logsectlog
= from
->sb_logsectlog
;
570 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
571 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
572 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
573 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
577 * Copy in core superblock to ondisk one.
579 * The fields argument is mask of superblock fields to copy.
587 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
588 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
598 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
599 first
= xfs_sb_info
[f
].offset
;
600 size
= xfs_sb_info
[f
+ 1].offset
- first
;
602 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
604 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
605 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
609 *(__be16
*)(to_ptr
+ first
) =
610 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
613 *(__be32
*)(to_ptr
+ first
) =
614 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
617 *(__be64
*)(to_ptr
+ first
) =
618 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
625 fields
&= ~(1LL << f
);
632 * Does the initial read of the superblock.
635 xfs_readsb(xfs_mount_t
*mp
, int flags
)
637 unsigned int sector_size
;
640 int loud
= !(flags
& XFS_MFSI_QUIET
);
642 ASSERT(mp
->m_sb_bp
== NULL
);
643 ASSERT(mp
->m_ddev_targp
!= NULL
);
646 * Allocate a (locked) buffer to hold the superblock.
647 * This will be kept around at all times to optimize
648 * access to the superblock.
650 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
653 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
654 BTOBB(sector_size
), 0);
657 xfs_warn(mp
, "SB buffer read failed");
662 * Initialize the mount structure from the superblock.
663 * But first do some basic consistency checking.
665 xfs_sb_from_disk(mp
, XFS_BUF_TO_SBP(bp
));
666 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
669 xfs_warn(mp
, "SB validate failed");
674 * We must be able to do sector-sized and sector-aligned IO.
676 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
678 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
679 sector_size
, mp
->m_sb
.sb_sectsize
);
685 * If device sector size is smaller than the superblock size,
686 * re-read the superblock so the buffer is correctly sized.
688 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
690 sector_size
= mp
->m_sb
.sb_sectsize
;
694 /* Initialize per-cpu counters */
695 xfs_icsb_reinit_counters(mp
);
710 * Mount initialization code establishing various mount
711 * fields from the superblock associated with the given
715 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
717 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
718 spin_lock_init(&mp
->m_agirotor_lock
);
719 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
720 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
721 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
722 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
723 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
724 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
725 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
726 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
727 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
729 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
730 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
731 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
732 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
734 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
735 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
736 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
737 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
739 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
740 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
741 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
742 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
744 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
745 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
747 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
751 * xfs_initialize_perag_data
753 * Read in each per-ag structure so we can count up the number of
754 * allocated inodes, free inodes and used filesystem blocks as this
755 * information is no longer persistent in the superblock. Once we have
756 * this information, write it into the in-core superblock structure.
759 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
761 xfs_agnumber_t index
;
763 xfs_sb_t
*sbp
= &mp
->m_sb
;
767 uint64_t bfreelst
= 0;
771 for (index
= 0; index
< agcount
; index
++) {
773 * read the agf, then the agi. This gets us
774 * all the information we need and populates the
775 * per-ag structures for us.
777 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
781 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
784 pag
= xfs_perag_get(mp
, index
);
785 ifree
+= pag
->pagi_freecount
;
786 ialloc
+= pag
->pagi_count
;
787 bfree
+= pag
->pagf_freeblks
;
788 bfreelst
+= pag
->pagf_flcount
;
789 btree
+= pag
->pagf_btreeblks
;
793 * Overwrite incore superblock counters with just-read data
795 spin_lock(&mp
->m_sb_lock
);
796 sbp
->sb_ifree
= ifree
;
797 sbp
->sb_icount
= ialloc
;
798 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
799 spin_unlock(&mp
->m_sb_lock
);
801 /* Fixup the per-cpu counters as well. */
802 xfs_icsb_reinit_counters(mp
);
808 * Update alignment values based on mount options and sb values
811 xfs_update_alignment(xfs_mount_t
*mp
)
813 xfs_sb_t
*sbp
= &(mp
->m_sb
);
817 * If stripe unit and stripe width are not multiples
818 * of the fs blocksize turn off alignment.
820 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
821 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
822 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
823 xfs_warn(mp
, "alignment check failed: "
824 "(sunit/swidth vs. blocksize)");
825 return XFS_ERROR(EINVAL
);
827 mp
->m_dalign
= mp
->m_swidth
= 0;
830 * Convert the stripe unit and width to FSBs.
832 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
833 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
834 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
835 xfs_warn(mp
, "alignment check failed: "
836 "(sunit/swidth vs. ag size)");
837 return XFS_ERROR(EINVAL
);
840 "stripe alignment turned off: sunit(%d)/swidth(%d) "
841 "incompatible with agsize(%d)",
842 mp
->m_dalign
, mp
->m_swidth
,
847 } else if (mp
->m_dalign
) {
848 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
850 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
851 xfs_warn(mp
, "alignment check failed: "
852 "sunit(%d) less than bsize(%d)",
855 return XFS_ERROR(EINVAL
);
862 * Update superblock with new values
865 if (xfs_sb_version_hasdalign(sbp
)) {
866 if (sbp
->sb_unit
!= mp
->m_dalign
) {
867 sbp
->sb_unit
= mp
->m_dalign
;
868 mp
->m_update_flags
|= XFS_SB_UNIT
;
870 if (sbp
->sb_width
!= mp
->m_swidth
) {
871 sbp
->sb_width
= mp
->m_swidth
;
872 mp
->m_update_flags
|= XFS_SB_WIDTH
;
875 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
876 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
877 mp
->m_dalign
= sbp
->sb_unit
;
878 mp
->m_swidth
= sbp
->sb_width
;
885 * Set the maximum inode count for this filesystem
888 xfs_set_maxicount(xfs_mount_t
*mp
)
890 xfs_sb_t
*sbp
= &(mp
->m_sb
);
893 if (sbp
->sb_imax_pct
) {
895 * Make sure the maximum inode count is a multiple
896 * of the units we allocate inodes in.
898 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
900 do_div(icount
, mp
->m_ialloc_blks
);
901 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
909 * Set the default minimum read and write sizes unless
910 * already specified in a mount option.
911 * We use smaller I/O sizes when the file system
912 * is being used for NFS service (wsync mount option).
915 xfs_set_rw_sizes(xfs_mount_t
*mp
)
917 xfs_sb_t
*sbp
= &(mp
->m_sb
);
918 int readio_log
, writeio_log
;
920 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
921 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
922 readio_log
= XFS_WSYNC_READIO_LOG
;
923 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
925 readio_log
= XFS_READIO_LOG_LARGE
;
926 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
929 readio_log
= mp
->m_readio_log
;
930 writeio_log
= mp
->m_writeio_log
;
933 if (sbp
->sb_blocklog
> readio_log
) {
934 mp
->m_readio_log
= sbp
->sb_blocklog
;
936 mp
->m_readio_log
= readio_log
;
938 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
939 if (sbp
->sb_blocklog
> writeio_log
) {
940 mp
->m_writeio_log
= sbp
->sb_blocklog
;
942 mp
->m_writeio_log
= writeio_log
;
944 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
948 * precalculate the low space thresholds for dynamic speculative preallocation.
951 xfs_set_low_space_thresholds(
952 struct xfs_mount
*mp
)
956 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
957 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
960 mp
->m_low_space
[i
] = space
* (i
+ 1);
966 * Set whether we're using inode alignment.
969 xfs_set_inoalignment(xfs_mount_t
*mp
)
971 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
972 mp
->m_sb
.sb_inoalignmt
>=
973 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
974 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
976 mp
->m_inoalign_mask
= 0;
978 * If we are using stripe alignment, check whether
979 * the stripe unit is a multiple of the inode alignment
981 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
982 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
983 mp
->m_sinoalign
= mp
->m_dalign
;
989 * Check that the data (and log if separate) are an ok size.
992 xfs_check_sizes(xfs_mount_t
*mp
)
997 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
998 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
999 xfs_warn(mp
, "filesystem size mismatch detected");
1000 return XFS_ERROR(EFBIG
);
1002 bp
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
1003 d
- XFS_FSS_TO_BB(mp
, 1),
1004 XFS_FSS_TO_BB(mp
, 1), 0);
1006 xfs_warn(mp
, "last sector read failed");
1011 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
1012 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
1013 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
1014 xfs_warn(mp
, "log size mismatch detected");
1015 return XFS_ERROR(EFBIG
);
1017 bp
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
1018 d
- XFS_FSB_TO_BB(mp
, 1),
1019 XFS_FSB_TO_BB(mp
, 1), 0);
1021 xfs_warn(mp
, "log device read failed");
1030 * Clear the quotaflags in memory and in the superblock.
1033 xfs_mount_reset_sbqflags(
1034 struct xfs_mount
*mp
)
1037 struct xfs_trans
*tp
;
1042 * It is OK to look at sb_qflags here in mount path,
1043 * without m_sb_lock.
1045 if (mp
->m_sb
.sb_qflags
== 0)
1047 spin_lock(&mp
->m_sb_lock
);
1048 mp
->m_sb
.sb_qflags
= 0;
1049 spin_unlock(&mp
->m_sb_lock
);
1052 * If the fs is readonly, let the incore superblock run
1053 * with quotas off but don't flush the update out to disk
1055 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1058 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1059 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1060 XFS_DEFAULT_LOG_COUNT
);
1062 xfs_trans_cancel(tp
, 0);
1063 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
1067 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
1068 return xfs_trans_commit(tp
, 0);
1072 xfs_default_resblks(xfs_mount_t
*mp
)
1077 * We default to 5% or 8192 fsbs of space reserved, whichever is
1078 * smaller. This is intended to cover concurrent allocation
1079 * transactions when we initially hit enospc. These each require a 4
1080 * block reservation. Hence by default we cover roughly 2000 concurrent
1081 * allocation reservations.
1083 resblks
= mp
->m_sb
.sb_dblocks
;
1084 do_div(resblks
, 20);
1085 resblks
= min_t(__uint64_t
, resblks
, 8192);
1090 * This function does the following on an initial mount of a file system:
1091 * - reads the superblock from disk and init the mount struct
1092 * - if we're a 32-bit kernel, do a size check on the superblock
1093 * so we don't mount terabyte filesystems
1094 * - init mount struct realtime fields
1095 * - allocate inode hash table for fs
1096 * - init directory manager
1097 * - perform recovery and init the log manager
1103 xfs_sb_t
*sbp
= &(mp
->m_sb
);
1106 uint quotamount
= 0;
1107 uint quotaflags
= 0;
1110 xfs_mount_common(mp
, sbp
);
1113 * Check for a mismatched features2 values. Older kernels
1114 * read & wrote into the wrong sb offset for sb_features2
1115 * on some platforms due to xfs_sb_t not being 64bit size aligned
1116 * when sb_features2 was added, which made older superblock
1117 * reading/writing routines swap it as a 64-bit value.
1119 * For backwards compatibility, we make both slots equal.
1121 * If we detect a mismatched field, we OR the set bits into the
1122 * existing features2 field in case it has already been modified; we
1123 * don't want to lose any features. We then update the bad location
1124 * with the ORed value so that older kernels will see any features2
1125 * flags, and mark the two fields as needing updates once the
1126 * transaction subsystem is online.
1128 if (xfs_sb_has_mismatched_features2(sbp
)) {
1129 xfs_warn(mp
, "correcting sb_features alignment problem");
1130 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
1131 sbp
->sb_bad_features2
= sbp
->sb_features2
;
1132 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
1135 * Re-check for ATTR2 in case it was found in bad_features2
1138 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1139 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
1140 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1143 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1144 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1145 xfs_sb_version_removeattr2(&mp
->m_sb
);
1146 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
1148 /* update sb_versionnum for the clearing of the morebits */
1149 if (!sbp
->sb_features2
)
1150 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
1154 * Check if sb_agblocks is aligned at stripe boundary
1155 * If sb_agblocks is NOT aligned turn off m_dalign since
1156 * allocator alignment is within an ag, therefore ag has
1157 * to be aligned at stripe boundary.
1159 error
= xfs_update_alignment(mp
);
1163 xfs_alloc_compute_maxlevels(mp
);
1164 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1165 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1166 xfs_ialloc_compute_maxlevels(mp
);
1168 xfs_set_maxicount(mp
);
1170 error
= xfs_uuid_mount(mp
);
1175 * Set the minimum read and write sizes
1177 xfs_set_rw_sizes(mp
);
1179 /* set the low space thresholds for dynamic preallocation */
1180 xfs_set_low_space_thresholds(mp
);
1183 * Set the inode cluster size.
1184 * This may still be overridden by the file system
1185 * block size if it is larger than the chosen cluster size.
1187 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1190 * Set inode alignment fields
1192 xfs_set_inoalignment(mp
);
1195 * Check that the data (and log if separate) are an ok size.
1197 error
= xfs_check_sizes(mp
);
1199 goto out_remove_uuid
;
1202 * Initialize realtime fields in the mount structure
1204 error
= xfs_rtmount_init(mp
);
1206 xfs_warn(mp
, "RT mount failed");
1207 goto out_remove_uuid
;
1211 * Copies the low order bits of the timestamp and the randomly
1212 * set "sequence" number out of a UUID.
1214 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1216 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1221 * Initialize the attribute manager's entries.
1223 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1226 * Initialize the precomputed transaction reservations values.
1231 * Allocate and initialize the per-ag data.
1233 spin_lock_init(&mp
->m_perag_lock
);
1234 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
1235 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
1237 xfs_warn(mp
, "Failed per-ag init: %d", error
);
1238 goto out_remove_uuid
;
1241 if (!sbp
->sb_logblocks
) {
1242 xfs_warn(mp
, "no log defined");
1243 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1244 error
= XFS_ERROR(EFSCORRUPTED
);
1245 goto out_free_perag
;
1249 * log's mount-time initialization. Perform 1st part recovery if needed
1251 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1252 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1253 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1255 xfs_warn(mp
, "log mount failed");
1260 * Now the log is mounted, we know if it was an unclean shutdown or
1261 * not. If it was, with the first phase of recovery has completed, we
1262 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1263 * but they are recovered transactionally in the second recovery phase
1266 * Hence we can safely re-initialise incore superblock counters from
1267 * the per-ag data. These may not be correct if the filesystem was not
1268 * cleanly unmounted, so we need to wait for recovery to finish before
1271 * If the filesystem was cleanly unmounted, then we can trust the
1272 * values in the superblock to be correct and we don't need to do
1275 * If we are currently making the filesystem, the initialisation will
1276 * fail as the perag data is in an undefined state.
1278 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1279 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1280 !mp
->m_sb
.sb_inprogress
) {
1281 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1287 * Get and sanity-check the root inode.
1288 * Save the pointer to it in the mount structure.
1290 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
1292 xfs_warn(mp
, "failed to read root inode");
1293 goto out_log_dealloc
;
1296 ASSERT(rip
!= NULL
);
1298 if (unlikely(!S_ISDIR(rip
->i_d
.di_mode
))) {
1299 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
1300 (unsigned long long)rip
->i_ino
);
1301 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1302 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1304 error
= XFS_ERROR(EFSCORRUPTED
);
1307 mp
->m_rootip
= rip
; /* save it */
1309 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1312 * Initialize realtime inode pointers in the mount structure
1314 error
= xfs_rtmount_inodes(mp
);
1317 * Free up the root inode.
1319 xfs_warn(mp
, "failed to read RT inodes");
1324 * If this is a read-only mount defer the superblock updates until
1325 * the next remount into writeable mode. Otherwise we would never
1326 * perform the update e.g. for the root filesystem.
1328 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1329 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1331 xfs_warn(mp
, "failed to write sb changes");
1337 * Initialise the XFS quota management subsystem for this mount
1339 if (XFS_IS_QUOTA_RUNNING(mp
)) {
1340 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
1344 ASSERT(!XFS_IS_QUOTA_ON(mp
));
1347 * If a file system had quotas running earlier, but decided to
1348 * mount without -o uquota/pquota/gquota options, revoke the
1349 * quotachecked license.
1351 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
1352 xfs_notice(mp
, "resetting quota flags");
1353 error
= xfs_mount_reset_sbqflags(mp
);
1360 * Finish recovering the file system. This part needed to be
1361 * delayed until after the root and real-time bitmap inodes
1362 * were consistently read in.
1364 error
= xfs_log_mount_finish(mp
);
1366 xfs_warn(mp
, "log mount finish failed");
1371 * Complete the quota initialisation, post-log-replay component.
1374 ASSERT(mp
->m_qflags
== 0);
1375 mp
->m_qflags
= quotaflags
;
1377 xfs_qm_mount_quotas(mp
);
1381 * Now we are mounted, reserve a small amount of unused space for
1382 * privileged transactions. This is needed so that transaction
1383 * space required for critical operations can dip into this pool
1384 * when at ENOSPC. This is needed for operations like create with
1385 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1386 * are not allowed to use this reserved space.
1388 * This may drive us straight to ENOSPC on mount, but that implies
1389 * we were already there on the last unmount. Warn if this occurs.
1391 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1392 resblks
= xfs_default_resblks(mp
);
1393 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1396 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1402 xfs_rtunmount_inodes(mp
);
1406 xfs_log_unmount(mp
);
1408 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1409 xfs_wait_buftarg(mp
->m_logdev_targp
);
1410 xfs_wait_buftarg(mp
->m_ddev_targp
);
1414 xfs_uuid_unmount(mp
);
1420 * This flushes out the inodes,dquots and the superblock, unmounts the
1421 * log and makes sure that incore structures are freed.
1425 struct xfs_mount
*mp
)
1430 xfs_qm_unmount_quotas(mp
);
1431 xfs_rtunmount_inodes(mp
);
1432 IRELE(mp
->m_rootip
);
1435 * We can potentially deadlock here if we have an inode cluster
1436 * that has been freed has its buffer still pinned in memory because
1437 * the transaction is still sitting in a iclog. The stale inodes
1438 * on that buffer will have their flush locks held until the
1439 * transaction hits the disk and the callbacks run. the inode
1440 * flush takes the flush lock unconditionally and with nothing to
1441 * push out the iclog we will never get that unlocked. hence we
1442 * need to force the log first.
1444 xfs_log_force(mp
, XFS_LOG_SYNC
);
1447 * Flush all pending changes from the AIL.
1449 xfs_ail_push_all_sync(mp
->m_ail
);
1452 * And reclaim all inodes. At this point there should be no dirty
1453 * inode, and none should be pinned or locked, but use synchronous
1454 * reclaim just to be sure.
1456 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1461 * Flush out the log synchronously so that we know for sure
1462 * that nothing is pinned. This is important because bflush()
1463 * will skip pinned buffers.
1465 xfs_log_force(mp
, XFS_LOG_SYNC
);
1468 * Unreserve any blocks we have so that when we unmount we don't account
1469 * the reserved free space as used. This is really only necessary for
1470 * lazy superblock counting because it trusts the incore superblock
1471 * counters to be absolutely correct on clean unmount.
1473 * We don't bother correcting this elsewhere for lazy superblock
1474 * counting because on mount of an unclean filesystem we reconstruct the
1475 * correct counter value and this is irrelevant.
1477 * For non-lazy counter filesystems, this doesn't matter at all because
1478 * we only every apply deltas to the superblock and hence the incore
1479 * value does not matter....
1482 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1484 xfs_warn(mp
, "Unable to free reserved block pool. "
1485 "Freespace may not be correct on next mount.");
1487 error
= xfs_log_sbcount(mp
);
1489 xfs_warn(mp
, "Unable to update superblock counters. "
1490 "Freespace may not be correct on next mount.");
1493 * At this point we might have modified the superblock again and thus
1494 * added an item to the AIL, thus flush it again.
1496 xfs_ail_push_all_sync(mp
->m_ail
);
1497 xfs_wait_buftarg(mp
->m_ddev_targp
);
1500 * The superblock buffer is uncached and xfsaild_push() will lock and
1501 * set the XBF_ASYNC flag on the buffer. We cannot do xfs_buf_iowait()
1502 * here but a lock on the superblock buffer will block until iodone()
1505 xfs_buf_lock(mp
->m_sb_bp
);
1506 xfs_buf_unlock(mp
->m_sb_bp
);
1508 xfs_log_unmount_write(mp
);
1509 xfs_log_unmount(mp
);
1510 xfs_uuid_unmount(mp
);
1513 xfs_errortag_clearall(mp
, 0);
1519 xfs_fs_writable(xfs_mount_t
*mp
)
1521 return !(mp
->m_super
->s_writers
.frozen
|| XFS_FORCED_SHUTDOWN(mp
) ||
1522 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1528 * Sync the superblock counters to disk.
1530 * Note this code can be called during the process of freezing, so
1531 * we may need to use the transaction allocator which does not
1532 * block when the transaction subsystem is in its frozen state.
1535 xfs_log_sbcount(xfs_mount_t
*mp
)
1540 if (!xfs_fs_writable(mp
))
1543 xfs_icsb_sync_counters(mp
, 0);
1546 * we don't need to do this if we are updating the superblock
1547 * counters on every modification.
1549 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1552 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1553 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1554 XFS_DEFAULT_LOG_COUNT
);
1556 xfs_trans_cancel(tp
, 0);
1560 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1561 xfs_trans_set_sync(tp
);
1562 error
= xfs_trans_commit(tp
, 0);
1567 * xfs_mod_sb() can be used to copy arbitrary changes to the
1568 * in-core superblock into the superblock buffer to be logged.
1569 * It does not provide the higher level of locking that is
1570 * needed to protect the in-core superblock from concurrent
1574 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1586 bp
= xfs_trans_getsb(tp
, mp
, 0);
1587 first
= sizeof(xfs_sb_t
);
1590 /* translate/copy */
1592 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1594 /* find modified range */
1595 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1596 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1597 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1599 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1600 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1601 first
= xfs_sb_info
[f
].offset
;
1603 xfs_trans_log_buf(tp
, bp
, first
, last
);
1608 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1609 * a delta to a specified field in the in-core superblock. Simply
1610 * switch on the field indicated and apply the delta to that field.
1611 * Fields are not allowed to dip below zero, so if the delta would
1612 * do this do not apply it and return EINVAL.
1614 * The m_sb_lock must be held when this routine is called.
1617 xfs_mod_incore_sb_unlocked(
1619 xfs_sb_field_t field
,
1623 int scounter
; /* short counter for 32 bit fields */
1624 long long lcounter
; /* long counter for 64 bit fields */
1625 long long res_used
, rem
;
1628 * With the in-core superblock spin lock held, switch
1629 * on the indicated field. Apply the delta to the
1630 * proper field. If the fields value would dip below
1631 * 0, then do not apply the delta and return EINVAL.
1634 case XFS_SBS_ICOUNT
:
1635 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1639 return XFS_ERROR(EINVAL
);
1641 mp
->m_sb
.sb_icount
= lcounter
;
1644 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1648 return XFS_ERROR(EINVAL
);
1650 mp
->m_sb
.sb_ifree
= lcounter
;
1652 case XFS_SBS_FDBLOCKS
:
1653 lcounter
= (long long)
1654 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1655 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1657 if (delta
> 0) { /* Putting blocks back */
1658 if (res_used
> delta
) {
1659 mp
->m_resblks_avail
+= delta
;
1661 rem
= delta
- res_used
;
1662 mp
->m_resblks_avail
= mp
->m_resblks
;
1665 } else { /* Taking blocks away */
1667 if (lcounter
>= 0) {
1668 mp
->m_sb
.sb_fdblocks
= lcounter
+
1669 XFS_ALLOC_SET_ASIDE(mp
);
1674 * We are out of blocks, use any available reserved
1675 * blocks if were allowed to.
1678 return XFS_ERROR(ENOSPC
);
1680 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1681 if (lcounter
>= 0) {
1682 mp
->m_resblks_avail
= lcounter
;
1685 printk_once(KERN_WARNING
1686 "Filesystem \"%s\": reserve blocks depleted! "
1687 "Consider increasing reserve pool size.",
1689 return XFS_ERROR(ENOSPC
);
1692 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1694 case XFS_SBS_FREXTENTS
:
1695 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1698 return XFS_ERROR(ENOSPC
);
1700 mp
->m_sb
.sb_frextents
= lcounter
;
1702 case XFS_SBS_DBLOCKS
:
1703 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1707 return XFS_ERROR(EINVAL
);
1709 mp
->m_sb
.sb_dblocks
= lcounter
;
1711 case XFS_SBS_AGCOUNT
:
1712 scounter
= mp
->m_sb
.sb_agcount
;
1716 return XFS_ERROR(EINVAL
);
1718 mp
->m_sb
.sb_agcount
= scounter
;
1720 case XFS_SBS_IMAX_PCT
:
1721 scounter
= mp
->m_sb
.sb_imax_pct
;
1725 return XFS_ERROR(EINVAL
);
1727 mp
->m_sb
.sb_imax_pct
= scounter
;
1729 case XFS_SBS_REXTSIZE
:
1730 scounter
= mp
->m_sb
.sb_rextsize
;
1734 return XFS_ERROR(EINVAL
);
1736 mp
->m_sb
.sb_rextsize
= scounter
;
1738 case XFS_SBS_RBMBLOCKS
:
1739 scounter
= mp
->m_sb
.sb_rbmblocks
;
1743 return XFS_ERROR(EINVAL
);
1745 mp
->m_sb
.sb_rbmblocks
= scounter
;
1747 case XFS_SBS_RBLOCKS
:
1748 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1752 return XFS_ERROR(EINVAL
);
1754 mp
->m_sb
.sb_rblocks
= lcounter
;
1756 case XFS_SBS_REXTENTS
:
1757 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1761 return XFS_ERROR(EINVAL
);
1763 mp
->m_sb
.sb_rextents
= lcounter
;
1765 case XFS_SBS_REXTSLOG
:
1766 scounter
= mp
->m_sb
.sb_rextslog
;
1770 return XFS_ERROR(EINVAL
);
1772 mp
->m_sb
.sb_rextslog
= scounter
;
1776 return XFS_ERROR(EINVAL
);
1781 * xfs_mod_incore_sb() is used to change a field in the in-core
1782 * superblock structure by the specified delta. This modification
1783 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1784 * routine to do the work.
1788 struct xfs_mount
*mp
,
1789 xfs_sb_field_t field
,
1795 #ifdef HAVE_PERCPU_SB
1796 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1798 spin_lock(&mp
->m_sb_lock
);
1799 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1800 spin_unlock(&mp
->m_sb_lock
);
1806 * Change more than one field in the in-core superblock structure at a time.
1808 * The fields and changes to those fields are specified in the array of
1809 * xfs_mod_sb structures passed in. Either all of the specified deltas
1810 * will be applied or none of them will. If any modified field dips below 0,
1811 * then all modifications will be backed out and EINVAL will be returned.
1813 * Note that this function may not be used for the superblock values that
1814 * are tracked with the in-memory per-cpu counters - a direct call to
1815 * xfs_icsb_modify_counters is required for these.
1818 xfs_mod_incore_sb_batch(
1819 struct xfs_mount
*mp
,
1828 * Loop through the array of mod structures and apply each individually.
1829 * If any fail, then back out all those which have already been applied.
1830 * Do all of this within the scope of the m_sb_lock so that all of the
1831 * changes will be atomic.
1833 spin_lock(&mp
->m_sb_lock
);
1834 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1835 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1836 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1838 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1839 msbp
->msb_delta
, rsvd
);
1843 spin_unlock(&mp
->m_sb_lock
);
1847 while (--msbp
>= msb
) {
1848 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1849 -msbp
->msb_delta
, rsvd
);
1852 spin_unlock(&mp
->m_sb_lock
);
1857 * xfs_getsb() is called to obtain the buffer for the superblock.
1858 * The buffer is returned locked and read in from disk.
1859 * The buffer should be released with a call to xfs_brelse().
1861 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1862 * the superblock buffer if it can be locked without sleeping.
1863 * If it can't then we'll return NULL.
1867 struct xfs_mount
*mp
,
1870 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1872 if (!xfs_buf_trylock(bp
)) {
1873 if (flags
& XBF_TRYLOCK
)
1879 ASSERT(XFS_BUF_ISDONE(bp
));
1884 * Used to free the superblock along various error paths.
1888 struct xfs_mount
*mp
)
1890 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1898 * Used to log changes to the superblock unit and width fields which could
1899 * be altered by the mount options, as well as any potential sb_features2
1900 * fixup. Only the first superblock is updated.
1910 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1911 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1912 XFS_SB_VERSIONNUM
));
1914 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1915 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1916 XFS_DEFAULT_LOG_COUNT
);
1918 xfs_trans_cancel(tp
, 0);
1921 xfs_mod_sb(tp
, fields
);
1922 error
= xfs_trans_commit(tp
, 0);
1927 * If the underlying (data/log/rt) device is readonly, there are some
1928 * operations that cannot proceed.
1931 xfs_dev_is_read_only(
1932 struct xfs_mount
*mp
,
1935 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1936 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1937 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1938 xfs_notice(mp
, "%s required on read-only device.", message
);
1939 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1945 #ifdef HAVE_PERCPU_SB
1947 * Per-cpu incore superblock counters
1949 * Simple concept, difficult implementation
1951 * Basically, replace the incore superblock counters with a distributed per cpu
1952 * counter for contended fields (e.g. free block count).
1954 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1955 * hence needs to be accurately read when we are running low on space. Hence
1956 * there is a method to enable and disable the per-cpu counters based on how
1957 * much "stuff" is available in them.
1959 * Basically, a counter is enabled if there is enough free resource to justify
1960 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1961 * ENOSPC), then we disable the counters to synchronise all callers and
1962 * re-distribute the available resources.
1964 * If, once we redistributed the available resources, we still get a failure,
1965 * we disable the per-cpu counter and go through the slow path.
1967 * The slow path is the current xfs_mod_incore_sb() function. This means that
1968 * when we disable a per-cpu counter, we need to drain its resources back to
1969 * the global superblock. We do this after disabling the counter to prevent
1970 * more threads from queueing up on the counter.
1972 * Essentially, this means that we still need a lock in the fast path to enable
1973 * synchronisation between the global counters and the per-cpu counters. This
1974 * is not a problem because the lock will be local to a CPU almost all the time
1975 * and have little contention except when we get to ENOSPC conditions.
1977 * Basically, this lock becomes a barrier that enables us to lock out the fast
1978 * path while we do things like enabling and disabling counters and
1979 * synchronising the counters.
1983 * 1. m_sb_lock before picking up per-cpu locks
1984 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1985 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1986 * 4. modifying per-cpu counters requires holding per-cpu lock
1987 * 5. modifying global counters requires holding m_sb_lock
1988 * 6. enabling or disabling a counter requires holding the m_sb_lock
1989 * and _none_ of the per-cpu locks.
1991 * Disabled counters are only ever re-enabled by a balance operation
1992 * that results in more free resources per CPU than a given threshold.
1993 * To ensure counters don't remain disabled, they are rebalanced when
1994 * the global resource goes above a higher threshold (i.e. some hysteresis
1995 * is present to prevent thrashing).
1998 #ifdef CONFIG_HOTPLUG_CPU
2000 * hot-plug CPU notifier support.
2002 * We need a notifier per filesystem as we need to be able to identify
2003 * the filesystem to balance the counters out. This is achieved by
2004 * having a notifier block embedded in the xfs_mount_t and doing pointer
2005 * magic to get the mount pointer from the notifier block address.
2008 xfs_icsb_cpu_notify(
2009 struct notifier_block
*nfb
,
2010 unsigned long action
,
2013 xfs_icsb_cnts_t
*cntp
;
2016 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2017 cntp
= (xfs_icsb_cnts_t
*)
2018 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2020 case CPU_UP_PREPARE
:
2021 case CPU_UP_PREPARE_FROZEN
:
2022 /* Easy Case - initialize the area and locks, and
2023 * then rebalance when online does everything else for us. */
2024 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2027 case CPU_ONLINE_FROZEN
:
2029 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2030 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2031 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2032 xfs_icsb_unlock(mp
);
2035 case CPU_DEAD_FROZEN
:
2036 /* Disable all the counters, then fold the dead cpu's
2037 * count into the total on the global superblock and
2038 * re-enable the counters. */
2040 spin_lock(&mp
->m_sb_lock
);
2041 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2042 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2043 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2045 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2046 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2047 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2049 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2051 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
2052 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
2053 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
2054 spin_unlock(&mp
->m_sb_lock
);
2055 xfs_icsb_unlock(mp
);
2061 #endif /* CONFIG_HOTPLUG_CPU */
2064 xfs_icsb_init_counters(
2067 xfs_icsb_cnts_t
*cntp
;
2070 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2071 if (mp
->m_sb_cnts
== NULL
)
2074 #ifdef CONFIG_HOTPLUG_CPU
2075 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2076 mp
->m_icsb_notifier
.priority
= 0;
2077 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2078 #endif /* CONFIG_HOTPLUG_CPU */
2080 for_each_online_cpu(i
) {
2081 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2082 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2085 mutex_init(&mp
->m_icsb_mutex
);
2088 * start with all counters disabled so that the
2089 * initial balance kicks us off correctly
2091 mp
->m_icsb_counters
= -1;
2096 xfs_icsb_reinit_counters(
2101 * start with all counters disabled so that the
2102 * initial balance kicks us off correctly
2104 mp
->m_icsb_counters
= -1;
2105 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2106 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2107 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2108 xfs_icsb_unlock(mp
);
2112 xfs_icsb_destroy_counters(
2115 if (mp
->m_sb_cnts
) {
2116 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2117 free_percpu(mp
->m_sb_cnts
);
2119 mutex_destroy(&mp
->m_icsb_mutex
);
2124 xfs_icsb_cnts_t
*icsbp
)
2126 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2132 xfs_icsb_unlock_cntr(
2133 xfs_icsb_cnts_t
*icsbp
)
2135 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2140 xfs_icsb_lock_all_counters(
2143 xfs_icsb_cnts_t
*cntp
;
2146 for_each_online_cpu(i
) {
2147 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2148 xfs_icsb_lock_cntr(cntp
);
2153 xfs_icsb_unlock_all_counters(
2156 xfs_icsb_cnts_t
*cntp
;
2159 for_each_online_cpu(i
) {
2160 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2161 xfs_icsb_unlock_cntr(cntp
);
2168 xfs_icsb_cnts_t
*cnt
,
2171 xfs_icsb_cnts_t
*cntp
;
2174 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2176 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2177 xfs_icsb_lock_all_counters(mp
);
2179 for_each_online_cpu(i
) {
2180 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2181 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2182 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2183 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2186 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2187 xfs_icsb_unlock_all_counters(mp
);
2191 xfs_icsb_counter_disabled(
2193 xfs_sb_field_t field
)
2195 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2196 return test_bit(field
, &mp
->m_icsb_counters
);
2200 xfs_icsb_disable_counter(
2202 xfs_sb_field_t field
)
2204 xfs_icsb_cnts_t cnt
;
2206 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2209 * If we are already disabled, then there is nothing to do
2210 * here. We check before locking all the counters to avoid
2211 * the expensive lock operation when being called in the
2212 * slow path and the counter is already disabled. This is
2213 * safe because the only time we set or clear this state is under
2216 if (xfs_icsb_counter_disabled(mp
, field
))
2219 xfs_icsb_lock_all_counters(mp
);
2220 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2221 /* drain back to superblock */
2223 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2225 case XFS_SBS_ICOUNT
:
2226 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2229 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2231 case XFS_SBS_FDBLOCKS
:
2232 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2239 xfs_icsb_unlock_all_counters(mp
);
2243 xfs_icsb_enable_counter(
2245 xfs_sb_field_t field
,
2249 xfs_icsb_cnts_t
*cntp
;
2252 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2254 xfs_icsb_lock_all_counters(mp
);
2255 for_each_online_cpu(i
) {
2256 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2258 case XFS_SBS_ICOUNT
:
2259 cntp
->icsb_icount
= count
+ resid
;
2262 cntp
->icsb_ifree
= count
+ resid
;
2264 case XFS_SBS_FDBLOCKS
:
2265 cntp
->icsb_fdblocks
= count
+ resid
;
2273 clear_bit(field
, &mp
->m_icsb_counters
);
2274 xfs_icsb_unlock_all_counters(mp
);
2278 xfs_icsb_sync_counters_locked(
2282 xfs_icsb_cnts_t cnt
;
2284 xfs_icsb_count(mp
, &cnt
, flags
);
2286 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2287 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2288 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2289 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2290 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2291 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2295 * Accurate update of per-cpu counters to incore superblock
2298 xfs_icsb_sync_counters(
2302 spin_lock(&mp
->m_sb_lock
);
2303 xfs_icsb_sync_counters_locked(mp
, flags
);
2304 spin_unlock(&mp
->m_sb_lock
);
2308 * Balance and enable/disable counters as necessary.
2310 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2311 * chosen to be the same number as single on disk allocation chunk per CPU, and
2312 * free blocks is something far enough zero that we aren't going thrash when we
2313 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2314 * prevent looping endlessly when xfs_alloc_space asks for more than will
2315 * be distributed to a single CPU but each CPU has enough blocks to be
2318 * Note that we can be called when counters are already disabled.
2319 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2320 * prevent locking every per-cpu counter needlessly.
2323 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2324 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2325 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2327 xfs_icsb_balance_counter_locked(
2329 xfs_sb_field_t field
,
2332 uint64_t count
, resid
;
2333 int weight
= num_online_cpus();
2334 uint64_t min
= (uint64_t)min_per_cpu
;
2336 /* disable counter and sync counter */
2337 xfs_icsb_disable_counter(mp
, field
);
2339 /* update counters - first CPU gets residual*/
2341 case XFS_SBS_ICOUNT
:
2342 count
= mp
->m_sb
.sb_icount
;
2343 resid
= do_div(count
, weight
);
2344 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2348 count
= mp
->m_sb
.sb_ifree
;
2349 resid
= do_div(count
, weight
);
2350 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2353 case XFS_SBS_FDBLOCKS
:
2354 count
= mp
->m_sb
.sb_fdblocks
;
2355 resid
= do_div(count
, weight
);
2356 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2361 count
= resid
= 0; /* quiet, gcc */
2365 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2369 xfs_icsb_balance_counter(
2371 xfs_sb_field_t fields
,
2374 spin_lock(&mp
->m_sb_lock
);
2375 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2376 spin_unlock(&mp
->m_sb_lock
);
2380 xfs_icsb_modify_counters(
2382 xfs_sb_field_t field
,
2386 xfs_icsb_cnts_t
*icsbp
;
2387 long long lcounter
; /* long counter for 64 bit fields */
2393 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
2396 * if the counter is disabled, go to slow path
2398 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2400 xfs_icsb_lock_cntr(icsbp
);
2401 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2402 xfs_icsb_unlock_cntr(icsbp
);
2407 case XFS_SBS_ICOUNT
:
2408 lcounter
= icsbp
->icsb_icount
;
2410 if (unlikely(lcounter
< 0))
2411 goto balance_counter
;
2412 icsbp
->icsb_icount
= lcounter
;
2416 lcounter
= icsbp
->icsb_ifree
;
2418 if (unlikely(lcounter
< 0))
2419 goto balance_counter
;
2420 icsbp
->icsb_ifree
= lcounter
;
2423 case XFS_SBS_FDBLOCKS
:
2424 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2426 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2428 if (unlikely(lcounter
< 0))
2429 goto balance_counter
;
2430 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2436 xfs_icsb_unlock_cntr(icsbp
);
2444 * serialise with a mutex so we don't burn lots of cpu on
2445 * the superblock lock. We still need to hold the superblock
2446 * lock, however, when we modify the global structures.
2451 * Now running atomically.
2453 * If the counter is enabled, someone has beaten us to rebalancing.
2454 * Drop the lock and try again in the fast path....
2456 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2457 xfs_icsb_unlock(mp
);
2462 * The counter is currently disabled. Because we are
2463 * running atomically here, we know a rebalance cannot
2464 * be in progress. Hence we can go straight to operating
2465 * on the global superblock. We do not call xfs_mod_incore_sb()
2466 * here even though we need to get the m_sb_lock. Doing so
2467 * will cause us to re-enter this function and deadlock.
2468 * Hence we get the m_sb_lock ourselves and then call
2469 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2470 * directly on the global counters.
2472 spin_lock(&mp
->m_sb_lock
);
2473 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2474 spin_unlock(&mp
->m_sb_lock
);
2477 * Now that we've modified the global superblock, we
2478 * may be able to re-enable the distributed counters
2479 * (e.g. lots of space just got freed). After that
2483 xfs_icsb_balance_counter(mp
, field
, 0);
2484 xfs_icsb_unlock(mp
);
2488 xfs_icsb_unlock_cntr(icsbp
);
2492 * We may have multiple threads here if multiple per-cpu
2493 * counters run dry at the same time. This will mean we can
2494 * do more balances than strictly necessary but it is not
2495 * the common slowpath case.
2500 * running atomically.
2502 * This will leave the counter in the correct state for future
2503 * accesses. After the rebalance, we simply try again and our retry
2504 * will either succeed through the fast path or slow path without
2505 * another balance operation being required.
2507 xfs_icsb_balance_counter(mp
, field
, delta
);
2508 xfs_icsb_unlock(mp
);