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"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_alloc.h"
37 #include "xfs_rtalloc.h"
39 #include "xfs_error.h"
41 #include "xfs_quota.h"
42 #include "xfs_fsops.h"
43 #include "xfs_utils.h"
44 #include "xfs_trace.h"
47 STATIC
void xfs_unmountfs_wait(xfs_mount_t
*);
51 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
53 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
55 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
58 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
59 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
64 short type
; /* 0 = integer
65 * 1 = binary / string (no translation)
68 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
69 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
70 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
71 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
72 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
73 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
74 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
75 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
76 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
77 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
78 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
79 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
80 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
81 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
82 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
83 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
84 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
85 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
86 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
87 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
88 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
89 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
90 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
91 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
92 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
93 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
94 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
95 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
96 { offsetof(xfs_sb_t
, sb_icount
), 0 },
97 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
98 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
99 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
100 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
101 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
102 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
103 { offsetof(xfs_sb_t
, sb_flags
), 0 },
104 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
105 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
106 { offsetof(xfs_sb_t
, sb_unit
), 0 },
107 { offsetof(xfs_sb_t
, sb_width
), 0 },
108 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
109 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
110 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
111 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
112 { offsetof(xfs_sb_t
, sb_features2
), 0 },
113 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
114 { sizeof(xfs_sb_t
), 0 }
117 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
118 static int xfs_uuid_table_size
;
119 static uuid_t
*xfs_uuid_table
;
122 * See if the UUID is unique among mounted XFS filesystems.
123 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
127 struct xfs_mount
*mp
)
129 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
132 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
135 if (uuid_is_nil(uuid
)) {
136 xfs_warn(mp
, "Filesystem has nil UUID - can't mount");
137 return XFS_ERROR(EINVAL
);
140 mutex_lock(&xfs_uuid_table_mutex
);
141 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
142 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
146 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
151 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
152 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
153 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
155 hole
= xfs_uuid_table_size
++;
157 xfs_uuid_table
[hole
] = *uuid
;
158 mutex_unlock(&xfs_uuid_table_mutex
);
163 mutex_unlock(&xfs_uuid_table_mutex
);
164 xfs_warn(mp
, "Filesystem has duplicate UUID - can't mount");
165 return XFS_ERROR(EINVAL
);
170 struct xfs_mount
*mp
)
172 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
175 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
178 mutex_lock(&xfs_uuid_table_mutex
);
179 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
180 if (uuid_is_nil(&xfs_uuid_table
[i
]))
182 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
184 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
187 ASSERT(i
< xfs_uuid_table_size
);
188 mutex_unlock(&xfs_uuid_table_mutex
);
193 * Reference counting access wrappers to the perag structures.
194 * Because we never free per-ag structures, the only thing we
195 * have to protect against changes is the tree structure itself.
198 xfs_perag_get(struct xfs_mount
*mp
, xfs_agnumber_t agno
)
200 struct xfs_perag
*pag
;
204 pag
= radix_tree_lookup(&mp
->m_perag_tree
, agno
);
206 ASSERT(atomic_read(&pag
->pag_ref
) >= 0);
207 ref
= atomic_inc_return(&pag
->pag_ref
);
210 trace_xfs_perag_get(mp
, agno
, ref
, _RET_IP_
);
215 * search from @first to find the next perag with the given tag set.
219 struct xfs_mount
*mp
,
220 xfs_agnumber_t first
,
223 struct xfs_perag
*pag
;
228 found
= radix_tree_gang_lookup_tag(&mp
->m_perag_tree
,
229 (void **)&pag
, first
, 1, tag
);
234 ref
= atomic_inc_return(&pag
->pag_ref
);
236 trace_xfs_perag_get_tag(mp
, pag
->pag_agno
, ref
, _RET_IP_
);
241 xfs_perag_put(struct xfs_perag
*pag
)
245 ASSERT(atomic_read(&pag
->pag_ref
) > 0);
246 ref
= atomic_dec_return(&pag
->pag_ref
);
247 trace_xfs_perag_put(pag
->pag_mount
, pag
->pag_agno
, ref
, _RET_IP_
);
252 struct rcu_head
*head
)
254 struct xfs_perag
*pag
= container_of(head
, struct xfs_perag
, rcu_head
);
256 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
261 * Free up the per-ag resources associated with the mount structure.
268 struct xfs_perag
*pag
;
270 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
271 spin_lock(&mp
->m_perag_lock
);
272 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
273 spin_unlock(&mp
->m_perag_lock
);
275 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
276 call_rcu(&pag
->rcu_head
, __xfs_free_perag
);
281 * Check size of device based on the (data/realtime) block count.
282 * Note: this check is used by the growfs code as well as mount.
285 xfs_sb_validate_fsb_count(
289 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
290 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
292 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
293 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
295 #else /* Limited by UINT_MAX of sectors */
296 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
303 * Check the validity of the SB found.
306 xfs_mount_validate_sb(
311 int loud
= !(flags
& XFS_MFSI_QUIET
);
314 * If the log device and data device have the
315 * same device number, the log is internal.
316 * Consequently, the sb_logstart should be non-zero. If
317 * we have a zero sb_logstart in this case, we may be trying to mount
318 * a volume filesystem in a non-volume manner.
320 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
322 xfs_warn(mp
, "bad magic number");
323 return XFS_ERROR(EWRONGFS
);
326 if (!xfs_sb_good_version(sbp
)) {
328 xfs_warn(mp
, "bad version");
329 return XFS_ERROR(EWRONGFS
);
333 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
336 "filesystem is marked as having an external log; "
337 "specify logdev on the mount command line.");
338 return XFS_ERROR(EINVAL
);
342 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
345 "filesystem is marked as having an internal log; "
346 "do not specify logdev on the mount command line.");
347 return XFS_ERROR(EINVAL
);
351 * More sanity checking. Most of these were stolen directly from
355 sbp
->sb_agcount
<= 0 ||
356 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
357 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
358 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
359 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
360 sbp
->sb_sectsize
!= (1 << sbp
->sb_sectlog
) ||
361 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
362 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
363 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
364 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
365 sbp
->sb_blocksize
!= (1 << sbp
->sb_blocklog
) ||
366 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
367 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
368 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
369 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
370 sbp
->sb_inodesize
!= (1 << sbp
->sb_inodelog
) ||
371 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
372 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
373 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
374 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */) ||
375 sbp
->sb_dblocks
== 0 ||
376 sbp
->sb_dblocks
> XFS_MAX_DBLOCKS(sbp
) ||
377 sbp
->sb_dblocks
< XFS_MIN_DBLOCKS(sbp
))) {
379 XFS_CORRUPTION_ERROR("SB sanity check failed",
380 XFS_ERRLEVEL_LOW
, mp
, sbp
);
381 return XFS_ERROR(EFSCORRUPTED
);
385 * Until this is fixed only page-sized or smaller data blocks work.
387 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
390 "File system with blocksize %d bytes. "
391 "Only pagesize (%ld) or less will currently work.",
392 sbp
->sb_blocksize
, PAGE_SIZE
);
394 return XFS_ERROR(ENOSYS
);
398 * Currently only very few inode sizes are supported.
400 switch (sbp
->sb_inodesize
) {
408 xfs_warn(mp
, "inode size of %d bytes not supported",
410 return XFS_ERROR(ENOSYS
);
413 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
414 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
417 "file system too large to be mounted on this system.");
418 return XFS_ERROR(EFBIG
);
421 if (unlikely(sbp
->sb_inprogress
)) {
423 xfs_warn(mp
, "file system busy");
424 return XFS_ERROR(EFSCORRUPTED
);
428 * Version 1 directory format has never worked on Linux.
430 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
433 "file system using version 1 directory format");
434 return XFS_ERROR(ENOSYS
);
441 xfs_initialize_perag(
443 xfs_agnumber_t agcount
,
444 xfs_agnumber_t
*maxagi
)
446 xfs_agnumber_t index
, max_metadata
;
447 xfs_agnumber_t first_initialised
= 0;
451 xfs_sb_t
*sbp
= &mp
->m_sb
;
455 * Walk the current per-ag tree so we don't try to initialise AGs
456 * that already exist (growfs case). Allocate and insert all the
457 * AGs we don't find ready for initialisation.
459 for (index
= 0; index
< agcount
; index
++) {
460 pag
= xfs_perag_get(mp
, index
);
465 if (!first_initialised
)
466 first_initialised
= index
;
468 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
471 pag
->pag_agno
= index
;
473 spin_lock_init(&pag
->pag_ici_lock
);
474 mutex_init(&pag
->pag_ici_reclaim_lock
);
475 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
476 spin_lock_init(&pag
->pag_buf_lock
);
477 pag
->pag_buf_tree
= RB_ROOT
;
479 if (radix_tree_preload(GFP_NOFS
))
482 spin_lock(&mp
->m_perag_lock
);
483 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
485 spin_unlock(&mp
->m_perag_lock
);
486 radix_tree_preload_end();
490 spin_unlock(&mp
->m_perag_lock
);
491 radix_tree_preload_end();
495 * If we mount with the inode64 option, or no inode overflows
496 * the legacy 32-bit address space clear the inode32 option.
498 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
499 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
501 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
502 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
504 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
506 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
508 * Calculate how much should be reserved for inodes to meet
509 * the max inode percentage.
511 if (mp
->m_maxicount
) {
514 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
516 icount
+= sbp
->sb_agblocks
- 1;
517 do_div(icount
, sbp
->sb_agblocks
);
518 max_metadata
= icount
;
520 max_metadata
= agcount
;
523 for (index
= 0; index
< agcount
; index
++) {
524 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
525 if (ino
> XFS_MAXINUMBER_32
) {
530 pag
= xfs_perag_get(mp
, index
);
531 pag
->pagi_inodeok
= 1;
532 if (index
< max_metadata
)
533 pag
->pagf_metadata
= 1;
537 for (index
= 0; index
< agcount
; index
++) {
538 pag
= xfs_perag_get(mp
, index
);
539 pag
->pagi_inodeok
= 1;
550 for (; index
> first_initialised
; index
--) {
551 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
562 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
563 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
564 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
565 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
566 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
567 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
568 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
569 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
570 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
571 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
572 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
573 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
574 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
575 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
576 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
577 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
578 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
579 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
580 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
581 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
582 to
->sb_blocklog
= from
->sb_blocklog
;
583 to
->sb_sectlog
= from
->sb_sectlog
;
584 to
->sb_inodelog
= from
->sb_inodelog
;
585 to
->sb_inopblog
= from
->sb_inopblog
;
586 to
->sb_agblklog
= from
->sb_agblklog
;
587 to
->sb_rextslog
= from
->sb_rextslog
;
588 to
->sb_inprogress
= from
->sb_inprogress
;
589 to
->sb_imax_pct
= from
->sb_imax_pct
;
590 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
591 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
592 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
593 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
594 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
595 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
596 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
597 to
->sb_flags
= from
->sb_flags
;
598 to
->sb_shared_vn
= from
->sb_shared_vn
;
599 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
600 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
601 to
->sb_width
= be32_to_cpu(from
->sb_width
);
602 to
->sb_dirblklog
= from
->sb_dirblklog
;
603 to
->sb_logsectlog
= from
->sb_logsectlog
;
604 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
605 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
606 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
607 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
611 * Copy in core superblock to ondisk one.
613 * The fields argument is mask of superblock fields to copy.
621 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
622 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
632 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
633 first
= xfs_sb_info
[f
].offset
;
634 size
= xfs_sb_info
[f
+ 1].offset
- first
;
636 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
638 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
639 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
643 *(__be16
*)(to_ptr
+ first
) =
644 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
647 *(__be32
*)(to_ptr
+ first
) =
648 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
651 *(__be64
*)(to_ptr
+ first
) =
652 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
659 fields
&= ~(1LL << f
);
666 * Does the initial read of the superblock.
669 xfs_readsb(xfs_mount_t
*mp
, int flags
)
671 unsigned int sector_size
;
674 int loud
= !(flags
& XFS_MFSI_QUIET
);
676 ASSERT(mp
->m_sb_bp
== NULL
);
677 ASSERT(mp
->m_ddev_targp
!= NULL
);
680 * Allocate a (locked) buffer to hold the superblock.
681 * This will be kept around at all times to optimize
682 * access to the superblock.
684 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
687 bp
= xfs_buf_read_uncached(mp
, mp
->m_ddev_targp
,
688 XFS_SB_DADDR
, sector_size
, 0);
691 xfs_warn(mp
, "SB buffer read failed");
696 * Initialize the mount structure from the superblock.
697 * But first do some basic consistency checking.
699 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
700 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
703 xfs_warn(mp
, "SB validate failed");
708 * We must be able to do sector-sized and sector-aligned IO.
710 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
712 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
713 sector_size
, mp
->m_sb
.sb_sectsize
);
719 * If device sector size is smaller than the superblock size,
720 * re-read the superblock so the buffer is correctly sized.
722 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
724 sector_size
= mp
->m_sb
.sb_sectsize
;
728 /* Initialize per-cpu counters */
729 xfs_icsb_reinit_counters(mp
);
744 * Mount initialization code establishing various mount
745 * fields from the superblock associated with the given
749 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
751 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
752 spin_lock_init(&mp
->m_agirotor_lock
);
753 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
754 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
755 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
756 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
757 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
758 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
759 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
760 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
761 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
763 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
764 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
765 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
766 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
768 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
769 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
770 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
771 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
773 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
774 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
775 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
776 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
778 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
779 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
781 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
785 * xfs_initialize_perag_data
787 * Read in each per-ag structure so we can count up the number of
788 * allocated inodes, free inodes and used filesystem blocks as this
789 * information is no longer persistent in the superblock. Once we have
790 * this information, write it into the in-core superblock structure.
793 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
795 xfs_agnumber_t index
;
797 xfs_sb_t
*sbp
= &mp
->m_sb
;
801 uint64_t bfreelst
= 0;
805 for (index
= 0; index
< agcount
; index
++) {
807 * read the agf, then the agi. This gets us
808 * all the information we need and populates the
809 * per-ag structures for us.
811 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
815 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
818 pag
= xfs_perag_get(mp
, index
);
819 ifree
+= pag
->pagi_freecount
;
820 ialloc
+= pag
->pagi_count
;
821 bfree
+= pag
->pagf_freeblks
;
822 bfreelst
+= pag
->pagf_flcount
;
823 btree
+= pag
->pagf_btreeblks
;
827 * Overwrite incore superblock counters with just-read data
829 spin_lock(&mp
->m_sb_lock
);
830 sbp
->sb_ifree
= ifree
;
831 sbp
->sb_icount
= ialloc
;
832 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
833 spin_unlock(&mp
->m_sb_lock
);
835 /* Fixup the per-cpu counters as well. */
836 xfs_icsb_reinit_counters(mp
);
842 * Update alignment values based on mount options and sb values
845 xfs_update_alignment(xfs_mount_t
*mp
)
847 xfs_sb_t
*sbp
= &(mp
->m_sb
);
851 * If stripe unit and stripe width are not multiples
852 * of the fs blocksize turn off alignment.
854 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
855 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
856 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
857 xfs_warn(mp
, "alignment check failed: "
858 "(sunit/swidth vs. blocksize)");
859 return XFS_ERROR(EINVAL
);
861 mp
->m_dalign
= mp
->m_swidth
= 0;
864 * Convert the stripe unit and width to FSBs.
866 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
867 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
868 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
869 xfs_warn(mp
, "alignment check failed: "
870 "(sunit/swidth vs. ag size)");
871 return XFS_ERROR(EINVAL
);
874 "stripe alignment turned off: sunit(%d)/swidth(%d) "
875 "incompatible with agsize(%d)",
876 mp
->m_dalign
, mp
->m_swidth
,
881 } else if (mp
->m_dalign
) {
882 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
884 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
885 xfs_warn(mp
, "alignment check failed: "
886 "sunit(%d) less than bsize(%d)",
889 return XFS_ERROR(EINVAL
);
896 * Update superblock with new values
899 if (xfs_sb_version_hasdalign(sbp
)) {
900 if (sbp
->sb_unit
!= mp
->m_dalign
) {
901 sbp
->sb_unit
= mp
->m_dalign
;
902 mp
->m_update_flags
|= XFS_SB_UNIT
;
904 if (sbp
->sb_width
!= mp
->m_swidth
) {
905 sbp
->sb_width
= mp
->m_swidth
;
906 mp
->m_update_flags
|= XFS_SB_WIDTH
;
909 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
910 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
911 mp
->m_dalign
= sbp
->sb_unit
;
912 mp
->m_swidth
= sbp
->sb_width
;
919 * Set the maximum inode count for this filesystem
922 xfs_set_maxicount(xfs_mount_t
*mp
)
924 xfs_sb_t
*sbp
= &(mp
->m_sb
);
927 if (sbp
->sb_imax_pct
) {
929 * Make sure the maximum inode count is a multiple
930 * of the units we allocate inodes in.
932 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
934 do_div(icount
, mp
->m_ialloc_blks
);
935 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
943 * Set the default minimum read and write sizes unless
944 * already specified in a mount option.
945 * We use smaller I/O sizes when the file system
946 * is being used for NFS service (wsync mount option).
949 xfs_set_rw_sizes(xfs_mount_t
*mp
)
951 xfs_sb_t
*sbp
= &(mp
->m_sb
);
952 int readio_log
, writeio_log
;
954 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
955 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
956 readio_log
= XFS_WSYNC_READIO_LOG
;
957 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
959 readio_log
= XFS_READIO_LOG_LARGE
;
960 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
963 readio_log
= mp
->m_readio_log
;
964 writeio_log
= mp
->m_writeio_log
;
967 if (sbp
->sb_blocklog
> readio_log
) {
968 mp
->m_readio_log
= sbp
->sb_blocklog
;
970 mp
->m_readio_log
= readio_log
;
972 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
973 if (sbp
->sb_blocklog
> writeio_log
) {
974 mp
->m_writeio_log
= sbp
->sb_blocklog
;
976 mp
->m_writeio_log
= writeio_log
;
978 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
982 * precalculate the low space thresholds for dynamic speculative preallocation.
985 xfs_set_low_space_thresholds(
986 struct xfs_mount
*mp
)
990 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
991 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
994 mp
->m_low_space
[i
] = space
* (i
+ 1);
1000 * Set whether we're using inode alignment.
1003 xfs_set_inoalignment(xfs_mount_t
*mp
)
1005 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
1006 mp
->m_sb
.sb_inoalignmt
>=
1007 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
1008 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
1010 mp
->m_inoalign_mask
= 0;
1012 * If we are using stripe alignment, check whether
1013 * the stripe unit is a multiple of the inode alignment
1015 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
1016 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
1017 mp
->m_sinoalign
= mp
->m_dalign
;
1019 mp
->m_sinoalign
= 0;
1023 * Check that the data (and log if separate) are an ok size.
1026 xfs_check_sizes(xfs_mount_t
*mp
)
1031 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
1032 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
1033 xfs_warn(mp
, "filesystem size mismatch detected");
1034 return XFS_ERROR(EFBIG
);
1036 bp
= xfs_buf_read_uncached(mp
, mp
->m_ddev_targp
,
1037 d
- XFS_FSS_TO_BB(mp
, 1),
1038 BBTOB(XFS_FSS_TO_BB(mp
, 1)), 0);
1040 xfs_warn(mp
, "last sector read failed");
1045 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
1046 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
1047 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
1048 xfs_warn(mp
, "log size mismatch detected");
1049 return XFS_ERROR(EFBIG
);
1051 bp
= xfs_buf_read_uncached(mp
, mp
->m_logdev_targp
,
1052 d
- XFS_FSB_TO_BB(mp
, 1),
1053 XFS_FSB_TO_B(mp
, 1), 0);
1055 xfs_warn(mp
, "log device read failed");
1064 * Clear the quotaflags in memory and in the superblock.
1067 xfs_mount_reset_sbqflags(
1068 struct xfs_mount
*mp
)
1071 struct xfs_trans
*tp
;
1076 * It is OK to look at sb_qflags here in mount path,
1077 * without m_sb_lock.
1079 if (mp
->m_sb
.sb_qflags
== 0)
1081 spin_lock(&mp
->m_sb_lock
);
1082 mp
->m_sb
.sb_qflags
= 0;
1083 spin_unlock(&mp
->m_sb_lock
);
1086 * If the fs is readonly, let the incore superblock run
1087 * with quotas off but don't flush the update out to disk
1089 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1093 xfs_notice(mp
, "Writing superblock quota changes");
1096 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1097 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1098 XFS_DEFAULT_LOG_COUNT
);
1100 xfs_trans_cancel(tp
, 0);
1101 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
1105 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
1106 return xfs_trans_commit(tp
, 0);
1110 xfs_default_resblks(xfs_mount_t
*mp
)
1115 * We default to 5% or 8192 fsbs of space reserved, whichever is
1116 * smaller. This is intended to cover concurrent allocation
1117 * transactions when we initially hit enospc. These each require a 4
1118 * block reservation. Hence by default we cover roughly 2000 concurrent
1119 * allocation reservations.
1121 resblks
= mp
->m_sb
.sb_dblocks
;
1122 do_div(resblks
, 20);
1123 resblks
= min_t(__uint64_t
, resblks
, 8192);
1128 * This function does the following on an initial mount of a file system:
1129 * - reads the superblock from disk and init the mount struct
1130 * - if we're a 32-bit kernel, do a size check on the superblock
1131 * so we don't mount terabyte filesystems
1132 * - init mount struct realtime fields
1133 * - allocate inode hash table for fs
1134 * - init directory manager
1135 * - perform recovery and init the log manager
1141 xfs_sb_t
*sbp
= &(mp
->m_sb
);
1144 uint quotamount
= 0;
1145 uint quotaflags
= 0;
1148 xfs_mount_common(mp
, sbp
);
1151 * Check for a mismatched features2 values. Older kernels
1152 * read & wrote into the wrong sb offset for sb_features2
1153 * on some platforms due to xfs_sb_t not being 64bit size aligned
1154 * when sb_features2 was added, which made older superblock
1155 * reading/writing routines swap it as a 64-bit value.
1157 * For backwards compatibility, we make both slots equal.
1159 * If we detect a mismatched field, we OR the set bits into the
1160 * existing features2 field in case it has already been modified; we
1161 * don't want to lose any features. We then update the bad location
1162 * with the ORed value so that older kernels will see any features2
1163 * flags, and mark the two fields as needing updates once the
1164 * transaction subsystem is online.
1166 if (xfs_sb_has_mismatched_features2(sbp
)) {
1167 xfs_warn(mp
, "correcting sb_features alignment problem");
1168 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
1169 sbp
->sb_bad_features2
= sbp
->sb_features2
;
1170 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
1173 * Re-check for ATTR2 in case it was found in bad_features2
1176 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1177 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
1178 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1181 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1182 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1183 xfs_sb_version_removeattr2(&mp
->m_sb
);
1184 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
1186 /* update sb_versionnum for the clearing of the morebits */
1187 if (!sbp
->sb_features2
)
1188 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
1192 * Check if sb_agblocks is aligned at stripe boundary
1193 * If sb_agblocks is NOT aligned turn off m_dalign since
1194 * allocator alignment is within an ag, therefore ag has
1195 * to be aligned at stripe boundary.
1197 error
= xfs_update_alignment(mp
);
1201 xfs_alloc_compute_maxlevels(mp
);
1202 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1203 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1204 xfs_ialloc_compute_maxlevels(mp
);
1206 xfs_set_maxicount(mp
);
1208 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
1210 error
= xfs_uuid_mount(mp
);
1215 * Set the minimum read and write sizes
1217 xfs_set_rw_sizes(mp
);
1219 /* set the low space thresholds for dynamic preallocation */
1220 xfs_set_low_space_thresholds(mp
);
1223 * Set the inode cluster size.
1224 * This may still be overridden by the file system
1225 * block size if it is larger than the chosen cluster size.
1227 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1230 * Set inode alignment fields
1232 xfs_set_inoalignment(mp
);
1235 * Check that the data (and log if separate) are an ok size.
1237 error
= xfs_check_sizes(mp
);
1239 goto out_remove_uuid
;
1242 * Initialize realtime fields in the mount structure
1244 error
= xfs_rtmount_init(mp
);
1246 xfs_warn(mp
, "RT mount failed");
1247 goto out_remove_uuid
;
1251 * Copies the low order bits of the timestamp and the randomly
1252 * set "sequence" number out of a UUID.
1254 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1256 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1261 * Initialize the attribute manager's entries.
1263 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1266 * Initialize the precomputed transaction reservations values.
1271 * Allocate and initialize the per-ag data.
1273 spin_lock_init(&mp
->m_perag_lock
);
1274 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
1275 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
1277 xfs_warn(mp
, "Failed per-ag init: %d", error
);
1278 goto out_remove_uuid
;
1281 if (!sbp
->sb_logblocks
) {
1282 xfs_warn(mp
, "no log defined");
1283 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1284 error
= XFS_ERROR(EFSCORRUPTED
);
1285 goto out_free_perag
;
1289 * log's mount-time initialization. Perform 1st part recovery if needed
1291 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1292 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1293 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1295 xfs_warn(mp
, "log mount failed");
1296 goto out_free_perag
;
1300 * Now the log is mounted, we know if it was an unclean shutdown or
1301 * not. If it was, with the first phase of recovery has completed, we
1302 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1303 * but they are recovered transactionally in the second recovery phase
1306 * Hence we can safely re-initialise incore superblock counters from
1307 * the per-ag data. These may not be correct if the filesystem was not
1308 * cleanly unmounted, so we need to wait for recovery to finish before
1311 * If the filesystem was cleanly unmounted, then we can trust the
1312 * values in the superblock to be correct and we don't need to do
1315 * If we are currently making the filesystem, the initialisation will
1316 * fail as the perag data is in an undefined state.
1318 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1319 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1320 !mp
->m_sb
.sb_inprogress
) {
1321 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1323 goto out_free_perag
;
1327 * Get and sanity-check the root inode.
1328 * Save the pointer to it in the mount structure.
1330 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
1332 xfs_warn(mp
, "failed to read root inode");
1333 goto out_log_dealloc
;
1336 ASSERT(rip
!= NULL
);
1338 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1339 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
1340 (unsigned long long)rip
->i_ino
);
1341 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1342 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1344 error
= XFS_ERROR(EFSCORRUPTED
);
1347 mp
->m_rootip
= rip
; /* save it */
1349 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1352 * Initialize realtime inode pointers in the mount structure
1354 error
= xfs_rtmount_inodes(mp
);
1357 * Free up the root inode.
1359 xfs_warn(mp
, "failed to read RT inodes");
1364 * If this is a read-only mount defer the superblock updates until
1365 * the next remount into writeable mode. Otherwise we would never
1366 * perform the update e.g. for the root filesystem.
1368 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1369 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1371 xfs_warn(mp
, "failed to write sb changes");
1377 * Initialise the XFS quota management subsystem for this mount
1379 if (XFS_IS_QUOTA_RUNNING(mp
)) {
1380 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
1384 ASSERT(!XFS_IS_QUOTA_ON(mp
));
1387 * If a file system had quotas running earlier, but decided to
1388 * mount without -o uquota/pquota/gquota options, revoke the
1389 * quotachecked license.
1391 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
1392 xfs_notice(mp
, "resetting quota flags");
1393 error
= xfs_mount_reset_sbqflags(mp
);
1400 * Finish recovering the file system. This part needed to be
1401 * delayed until after the root and real-time bitmap inodes
1402 * were consistently read in.
1404 error
= xfs_log_mount_finish(mp
);
1406 xfs_warn(mp
, "log mount finish failed");
1411 * Complete the quota initialisation, post-log-replay component.
1414 ASSERT(mp
->m_qflags
== 0);
1415 mp
->m_qflags
= quotaflags
;
1417 xfs_qm_mount_quotas(mp
);
1421 * Now we are mounted, reserve a small amount of unused space for
1422 * privileged transactions. This is needed so that transaction
1423 * space required for critical operations can dip into this pool
1424 * when at ENOSPC. This is needed for operations like create with
1425 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1426 * are not allowed to use this reserved space.
1428 * This may drive us straight to ENOSPC on mount, but that implies
1429 * we were already there on the last unmount. Warn if this occurs.
1431 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1432 resblks
= xfs_default_resblks(mp
);
1433 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1436 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1442 xfs_rtunmount_inodes(mp
);
1446 xfs_log_unmount(mp
);
1450 xfs_uuid_unmount(mp
);
1456 * This flushes out the inodes,dquots and the superblock, unmounts the
1457 * log and makes sure that incore structures are freed.
1461 struct xfs_mount
*mp
)
1466 xfs_qm_unmount_quotas(mp
);
1467 xfs_rtunmount_inodes(mp
);
1468 IRELE(mp
->m_rootip
);
1471 * We can potentially deadlock here if we have an inode cluster
1472 * that has been freed has its buffer still pinned in memory because
1473 * the transaction is still sitting in a iclog. The stale inodes
1474 * on that buffer will have their flush locks held until the
1475 * transaction hits the disk and the callbacks run. the inode
1476 * flush takes the flush lock unconditionally and with nothing to
1477 * push out the iclog we will never get that unlocked. hence we
1478 * need to force the log first.
1480 xfs_log_force(mp
, XFS_LOG_SYNC
);
1483 * Do a delwri reclaim pass first so that as many dirty inodes are
1484 * queued up for IO as possible. Then flush the buffers before making
1485 * a synchronous path to catch all the remaining inodes are reclaimed.
1486 * This makes the reclaim process as quick as possible by avoiding
1487 * synchronous writeout and blocking on inodes already in the delwri
1488 * state as much as possible.
1490 xfs_reclaim_inodes(mp
, 0);
1491 XFS_bflush(mp
->m_ddev_targp
);
1492 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1497 * Flush out the log synchronously so that we know for sure
1498 * that nothing is pinned. This is important because bflush()
1499 * will skip pinned buffers.
1501 xfs_log_force(mp
, XFS_LOG_SYNC
);
1503 xfs_binval(mp
->m_ddev_targp
);
1504 if (mp
->m_rtdev_targp
) {
1505 xfs_binval(mp
->m_rtdev_targp
);
1509 * Unreserve any blocks we have so that when we unmount we don't account
1510 * the reserved free space as used. This is really only necessary for
1511 * lazy superblock counting because it trusts the incore superblock
1512 * counters to be absolutely correct on clean unmount.
1514 * We don't bother correcting this elsewhere for lazy superblock
1515 * counting because on mount of an unclean filesystem we reconstruct the
1516 * correct counter value and this is irrelevant.
1518 * For non-lazy counter filesystems, this doesn't matter at all because
1519 * we only every apply deltas to the superblock and hence the incore
1520 * value does not matter....
1523 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1525 xfs_warn(mp
, "Unable to free reserved block pool. "
1526 "Freespace may not be correct on next mount.");
1528 error
= xfs_log_sbcount(mp
, 1);
1530 xfs_warn(mp
, "Unable to update superblock counters. "
1531 "Freespace may not be correct on next mount.");
1532 xfs_unmountfs_writesb(mp
);
1533 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1534 xfs_log_unmount_write(mp
);
1535 xfs_log_unmount(mp
);
1536 xfs_uuid_unmount(mp
);
1539 xfs_errortag_clearall(mp
, 0);
1545 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1547 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1548 xfs_wait_buftarg(mp
->m_logdev_targp
);
1549 if (mp
->m_rtdev_targp
)
1550 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1551 xfs_wait_buftarg(mp
->m_ddev_targp
);
1555 xfs_fs_writable(xfs_mount_t
*mp
)
1557 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1558 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1564 * Called either periodically to keep the on disk superblock values
1565 * roughly up to date or from unmount to make sure the values are
1566 * correct on a clean unmount.
1568 * Note this code can be called during the process of freezing, so
1569 * we may need to use the transaction allocator which does not not
1570 * block when the transaction subsystem is in its frozen state.
1580 if (!xfs_fs_writable(mp
))
1583 xfs_icsb_sync_counters(mp
, 0);
1586 * we don't need to do this if we are updating the superblock
1587 * counters on every modification.
1589 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1592 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1593 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1594 XFS_DEFAULT_LOG_COUNT
);
1596 xfs_trans_cancel(tp
, 0);
1600 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1602 xfs_trans_set_sync(tp
);
1603 error
= xfs_trans_commit(tp
, 0);
1608 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1614 * skip superblock write if fs is read-only, or
1615 * if we are doing a forced umount.
1617 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1618 XFS_FORCED_SHUTDOWN(mp
))) {
1620 sbp
= xfs_getsb(mp
, 0);
1622 XFS_BUF_UNDONE(sbp
);
1623 XFS_BUF_UNREAD(sbp
);
1624 XFS_BUF_UNDELAYWRITE(sbp
);
1626 XFS_BUF_UNASYNC(sbp
);
1627 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1628 xfsbdstrat(mp
, sbp
);
1629 error
= xfs_buf_iowait(sbp
);
1631 xfs_ioerror_alert("xfs_unmountfs_writesb",
1632 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1639 * xfs_mod_sb() can be used to copy arbitrary changes to the
1640 * in-core superblock into the superblock buffer to be logged.
1641 * It does not provide the higher level of locking that is
1642 * needed to protect the in-core superblock from concurrent
1646 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1658 bp
= xfs_trans_getsb(tp
, mp
, 0);
1659 first
= sizeof(xfs_sb_t
);
1662 /* translate/copy */
1664 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1666 /* find modified range */
1667 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1668 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1669 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1671 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1672 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1673 first
= xfs_sb_info
[f
].offset
;
1675 xfs_trans_log_buf(tp
, bp
, first
, last
);
1680 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1681 * a delta to a specified field in the in-core superblock. Simply
1682 * switch on the field indicated and apply the delta to that field.
1683 * Fields are not allowed to dip below zero, so if the delta would
1684 * do this do not apply it and return EINVAL.
1686 * The m_sb_lock must be held when this routine is called.
1689 xfs_mod_incore_sb_unlocked(
1691 xfs_sb_field_t field
,
1695 int scounter
; /* short counter for 32 bit fields */
1696 long long lcounter
; /* long counter for 64 bit fields */
1697 long long res_used
, rem
;
1700 * With the in-core superblock spin lock held, switch
1701 * on the indicated field. Apply the delta to the
1702 * proper field. If the fields value would dip below
1703 * 0, then do not apply the delta and return EINVAL.
1706 case XFS_SBS_ICOUNT
:
1707 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1711 return XFS_ERROR(EINVAL
);
1713 mp
->m_sb
.sb_icount
= lcounter
;
1716 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1720 return XFS_ERROR(EINVAL
);
1722 mp
->m_sb
.sb_ifree
= lcounter
;
1724 case XFS_SBS_FDBLOCKS
:
1725 lcounter
= (long long)
1726 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1727 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1729 if (delta
> 0) { /* Putting blocks back */
1730 if (res_used
> delta
) {
1731 mp
->m_resblks_avail
+= delta
;
1733 rem
= delta
- res_used
;
1734 mp
->m_resblks_avail
= mp
->m_resblks
;
1737 } else { /* Taking blocks away */
1739 if (lcounter
>= 0) {
1740 mp
->m_sb
.sb_fdblocks
= lcounter
+
1741 XFS_ALLOC_SET_ASIDE(mp
);
1746 * We are out of blocks, use any available reserved
1747 * blocks if were allowed to.
1750 return XFS_ERROR(ENOSPC
);
1752 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1753 if (lcounter
>= 0) {
1754 mp
->m_resblks_avail
= lcounter
;
1757 printk_once(KERN_WARNING
1758 "Filesystem \"%s\": reserve blocks depleted! "
1759 "Consider increasing reserve pool size.",
1761 return XFS_ERROR(ENOSPC
);
1764 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1766 case XFS_SBS_FREXTENTS
:
1767 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1770 return XFS_ERROR(ENOSPC
);
1772 mp
->m_sb
.sb_frextents
= lcounter
;
1774 case XFS_SBS_DBLOCKS
:
1775 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1779 return XFS_ERROR(EINVAL
);
1781 mp
->m_sb
.sb_dblocks
= lcounter
;
1783 case XFS_SBS_AGCOUNT
:
1784 scounter
= mp
->m_sb
.sb_agcount
;
1788 return XFS_ERROR(EINVAL
);
1790 mp
->m_sb
.sb_agcount
= scounter
;
1792 case XFS_SBS_IMAX_PCT
:
1793 scounter
= mp
->m_sb
.sb_imax_pct
;
1797 return XFS_ERROR(EINVAL
);
1799 mp
->m_sb
.sb_imax_pct
= scounter
;
1801 case XFS_SBS_REXTSIZE
:
1802 scounter
= mp
->m_sb
.sb_rextsize
;
1806 return XFS_ERROR(EINVAL
);
1808 mp
->m_sb
.sb_rextsize
= scounter
;
1810 case XFS_SBS_RBMBLOCKS
:
1811 scounter
= mp
->m_sb
.sb_rbmblocks
;
1815 return XFS_ERROR(EINVAL
);
1817 mp
->m_sb
.sb_rbmblocks
= scounter
;
1819 case XFS_SBS_RBLOCKS
:
1820 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1824 return XFS_ERROR(EINVAL
);
1826 mp
->m_sb
.sb_rblocks
= lcounter
;
1828 case XFS_SBS_REXTENTS
:
1829 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1833 return XFS_ERROR(EINVAL
);
1835 mp
->m_sb
.sb_rextents
= lcounter
;
1837 case XFS_SBS_REXTSLOG
:
1838 scounter
= mp
->m_sb
.sb_rextslog
;
1842 return XFS_ERROR(EINVAL
);
1844 mp
->m_sb
.sb_rextslog
= scounter
;
1848 return XFS_ERROR(EINVAL
);
1853 * xfs_mod_incore_sb() is used to change a field in the in-core
1854 * superblock structure by the specified delta. This modification
1855 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1856 * routine to do the work.
1860 struct xfs_mount
*mp
,
1861 xfs_sb_field_t field
,
1867 #ifdef HAVE_PERCPU_SB
1868 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1870 spin_lock(&mp
->m_sb_lock
);
1871 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1872 spin_unlock(&mp
->m_sb_lock
);
1878 * Change more than one field in the in-core superblock structure at a time.
1880 * The fields and changes to those fields are specified in the array of
1881 * xfs_mod_sb structures passed in. Either all of the specified deltas
1882 * will be applied or none of them will. If any modified field dips below 0,
1883 * then all modifications will be backed out and EINVAL will be returned.
1885 * Note that this function may not be used for the superblock values that
1886 * are tracked with the in-memory per-cpu counters - a direct call to
1887 * xfs_icsb_modify_counters is required for these.
1890 xfs_mod_incore_sb_batch(
1891 struct xfs_mount
*mp
,
1900 * Loop through the array of mod structures and apply each individually.
1901 * If any fail, then back out all those which have already been applied.
1902 * Do all of this within the scope of the m_sb_lock so that all of the
1903 * changes will be atomic.
1905 spin_lock(&mp
->m_sb_lock
);
1906 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1907 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1908 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1910 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1911 msbp
->msb_delta
, rsvd
);
1915 spin_unlock(&mp
->m_sb_lock
);
1919 while (--msbp
>= msb
) {
1920 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1921 -msbp
->msb_delta
, rsvd
);
1924 spin_unlock(&mp
->m_sb_lock
);
1929 * xfs_getsb() is called to obtain the buffer for the superblock.
1930 * The buffer is returned locked and read in from disk.
1931 * The buffer should be released with a call to xfs_brelse().
1933 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1934 * the superblock buffer if it can be locked without sleeping.
1935 * If it can't then we'll return NULL.
1939 struct xfs_mount
*mp
,
1942 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1944 if (!xfs_buf_trylock(bp
)) {
1945 if (flags
& XBF_TRYLOCK
)
1951 ASSERT(XFS_BUF_ISDONE(bp
));
1956 * Used to free the superblock along various error paths.
1960 struct xfs_mount
*mp
)
1962 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1970 * Used to log changes to the superblock unit and width fields which could
1971 * be altered by the mount options, as well as any potential sb_features2
1972 * fixup. Only the first superblock is updated.
1982 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1983 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1984 XFS_SB_VERSIONNUM
));
1986 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1987 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1988 XFS_DEFAULT_LOG_COUNT
);
1990 xfs_trans_cancel(tp
, 0);
1993 xfs_mod_sb(tp
, fields
);
1994 error
= xfs_trans_commit(tp
, 0);
1999 * If the underlying (data/log/rt) device is readonly, there are some
2000 * operations that cannot proceed.
2003 xfs_dev_is_read_only(
2004 struct xfs_mount
*mp
,
2007 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
2008 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
2009 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
2010 xfs_notice(mp
, "%s required on read-only device.", message
);
2011 xfs_notice(mp
, "write access unavailable, cannot proceed.");
2017 #ifdef HAVE_PERCPU_SB
2019 * Per-cpu incore superblock counters
2021 * Simple concept, difficult implementation
2023 * Basically, replace the incore superblock counters with a distributed per cpu
2024 * counter for contended fields (e.g. free block count).
2026 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2027 * hence needs to be accurately read when we are running low on space. Hence
2028 * there is a method to enable and disable the per-cpu counters based on how
2029 * much "stuff" is available in them.
2031 * Basically, a counter is enabled if there is enough free resource to justify
2032 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2033 * ENOSPC), then we disable the counters to synchronise all callers and
2034 * re-distribute the available resources.
2036 * If, once we redistributed the available resources, we still get a failure,
2037 * we disable the per-cpu counter and go through the slow path.
2039 * The slow path is the current xfs_mod_incore_sb() function. This means that
2040 * when we disable a per-cpu counter, we need to drain its resources back to
2041 * the global superblock. We do this after disabling the counter to prevent
2042 * more threads from queueing up on the counter.
2044 * Essentially, this means that we still need a lock in the fast path to enable
2045 * synchronisation between the global counters and the per-cpu counters. This
2046 * is not a problem because the lock will be local to a CPU almost all the time
2047 * and have little contention except when we get to ENOSPC conditions.
2049 * Basically, this lock becomes a barrier that enables us to lock out the fast
2050 * path while we do things like enabling and disabling counters and
2051 * synchronising the counters.
2055 * 1. m_sb_lock before picking up per-cpu locks
2056 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2057 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2058 * 4. modifying per-cpu counters requires holding per-cpu lock
2059 * 5. modifying global counters requires holding m_sb_lock
2060 * 6. enabling or disabling a counter requires holding the m_sb_lock
2061 * and _none_ of the per-cpu locks.
2063 * Disabled counters are only ever re-enabled by a balance operation
2064 * that results in more free resources per CPU than a given threshold.
2065 * To ensure counters don't remain disabled, they are rebalanced when
2066 * the global resource goes above a higher threshold (i.e. some hysteresis
2067 * is present to prevent thrashing).
2070 #ifdef CONFIG_HOTPLUG_CPU
2072 * hot-plug CPU notifier support.
2074 * We need a notifier per filesystem as we need to be able to identify
2075 * the filesystem to balance the counters out. This is achieved by
2076 * having a notifier block embedded in the xfs_mount_t and doing pointer
2077 * magic to get the mount pointer from the notifier block address.
2080 xfs_icsb_cpu_notify(
2081 struct notifier_block
*nfb
,
2082 unsigned long action
,
2085 xfs_icsb_cnts_t
*cntp
;
2088 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2089 cntp
= (xfs_icsb_cnts_t
*)
2090 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2092 case CPU_UP_PREPARE
:
2093 case CPU_UP_PREPARE_FROZEN
:
2094 /* Easy Case - initialize the area and locks, and
2095 * then rebalance when online does everything else for us. */
2096 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2099 case CPU_ONLINE_FROZEN
:
2101 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2102 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2103 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2104 xfs_icsb_unlock(mp
);
2107 case CPU_DEAD_FROZEN
:
2108 /* Disable all the counters, then fold the dead cpu's
2109 * count into the total on the global superblock and
2110 * re-enable the counters. */
2112 spin_lock(&mp
->m_sb_lock
);
2113 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2114 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2115 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2117 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2118 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2119 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2121 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2123 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
2124 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
2125 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
2126 spin_unlock(&mp
->m_sb_lock
);
2127 xfs_icsb_unlock(mp
);
2133 #endif /* CONFIG_HOTPLUG_CPU */
2136 xfs_icsb_init_counters(
2139 xfs_icsb_cnts_t
*cntp
;
2142 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2143 if (mp
->m_sb_cnts
== NULL
)
2146 #ifdef CONFIG_HOTPLUG_CPU
2147 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2148 mp
->m_icsb_notifier
.priority
= 0;
2149 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2150 #endif /* CONFIG_HOTPLUG_CPU */
2152 for_each_online_cpu(i
) {
2153 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2154 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2157 mutex_init(&mp
->m_icsb_mutex
);
2160 * start with all counters disabled so that the
2161 * initial balance kicks us off correctly
2163 mp
->m_icsb_counters
= -1;
2168 xfs_icsb_reinit_counters(
2173 * start with all counters disabled so that the
2174 * initial balance kicks us off correctly
2176 mp
->m_icsb_counters
= -1;
2177 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2178 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2179 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2180 xfs_icsb_unlock(mp
);
2184 xfs_icsb_destroy_counters(
2187 if (mp
->m_sb_cnts
) {
2188 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2189 free_percpu(mp
->m_sb_cnts
);
2191 mutex_destroy(&mp
->m_icsb_mutex
);
2196 xfs_icsb_cnts_t
*icsbp
)
2198 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2204 xfs_icsb_unlock_cntr(
2205 xfs_icsb_cnts_t
*icsbp
)
2207 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2212 xfs_icsb_lock_all_counters(
2215 xfs_icsb_cnts_t
*cntp
;
2218 for_each_online_cpu(i
) {
2219 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2220 xfs_icsb_lock_cntr(cntp
);
2225 xfs_icsb_unlock_all_counters(
2228 xfs_icsb_cnts_t
*cntp
;
2231 for_each_online_cpu(i
) {
2232 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2233 xfs_icsb_unlock_cntr(cntp
);
2240 xfs_icsb_cnts_t
*cnt
,
2243 xfs_icsb_cnts_t
*cntp
;
2246 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2248 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2249 xfs_icsb_lock_all_counters(mp
);
2251 for_each_online_cpu(i
) {
2252 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2253 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2254 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2255 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2258 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2259 xfs_icsb_unlock_all_counters(mp
);
2263 xfs_icsb_counter_disabled(
2265 xfs_sb_field_t field
)
2267 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2268 return test_bit(field
, &mp
->m_icsb_counters
);
2272 xfs_icsb_disable_counter(
2274 xfs_sb_field_t field
)
2276 xfs_icsb_cnts_t cnt
;
2278 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2281 * If we are already disabled, then there is nothing to do
2282 * here. We check before locking all the counters to avoid
2283 * the expensive lock operation when being called in the
2284 * slow path and the counter is already disabled. This is
2285 * safe because the only time we set or clear this state is under
2288 if (xfs_icsb_counter_disabled(mp
, field
))
2291 xfs_icsb_lock_all_counters(mp
);
2292 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2293 /* drain back to superblock */
2295 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2297 case XFS_SBS_ICOUNT
:
2298 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2301 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2303 case XFS_SBS_FDBLOCKS
:
2304 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2311 xfs_icsb_unlock_all_counters(mp
);
2315 xfs_icsb_enable_counter(
2317 xfs_sb_field_t field
,
2321 xfs_icsb_cnts_t
*cntp
;
2324 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2326 xfs_icsb_lock_all_counters(mp
);
2327 for_each_online_cpu(i
) {
2328 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2330 case XFS_SBS_ICOUNT
:
2331 cntp
->icsb_icount
= count
+ resid
;
2334 cntp
->icsb_ifree
= count
+ resid
;
2336 case XFS_SBS_FDBLOCKS
:
2337 cntp
->icsb_fdblocks
= count
+ resid
;
2345 clear_bit(field
, &mp
->m_icsb_counters
);
2346 xfs_icsb_unlock_all_counters(mp
);
2350 xfs_icsb_sync_counters_locked(
2354 xfs_icsb_cnts_t cnt
;
2356 xfs_icsb_count(mp
, &cnt
, flags
);
2358 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2359 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2360 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2361 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2362 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2363 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2367 * Accurate update of per-cpu counters to incore superblock
2370 xfs_icsb_sync_counters(
2374 spin_lock(&mp
->m_sb_lock
);
2375 xfs_icsb_sync_counters_locked(mp
, flags
);
2376 spin_unlock(&mp
->m_sb_lock
);
2380 * Balance and enable/disable counters as necessary.
2382 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2383 * chosen to be the same number as single on disk allocation chunk per CPU, and
2384 * free blocks is something far enough zero that we aren't going thrash when we
2385 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2386 * prevent looping endlessly when xfs_alloc_space asks for more than will
2387 * be distributed to a single CPU but each CPU has enough blocks to be
2390 * Note that we can be called when counters are already disabled.
2391 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2392 * prevent locking every per-cpu counter needlessly.
2395 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2396 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2397 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2399 xfs_icsb_balance_counter_locked(
2401 xfs_sb_field_t field
,
2404 uint64_t count
, resid
;
2405 int weight
= num_online_cpus();
2406 uint64_t min
= (uint64_t)min_per_cpu
;
2408 /* disable counter and sync counter */
2409 xfs_icsb_disable_counter(mp
, field
);
2411 /* update counters - first CPU gets residual*/
2413 case XFS_SBS_ICOUNT
:
2414 count
= mp
->m_sb
.sb_icount
;
2415 resid
= do_div(count
, weight
);
2416 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2420 count
= mp
->m_sb
.sb_ifree
;
2421 resid
= do_div(count
, weight
);
2422 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2425 case XFS_SBS_FDBLOCKS
:
2426 count
= mp
->m_sb
.sb_fdblocks
;
2427 resid
= do_div(count
, weight
);
2428 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2433 count
= resid
= 0; /* quiet, gcc */
2437 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2441 xfs_icsb_balance_counter(
2443 xfs_sb_field_t fields
,
2446 spin_lock(&mp
->m_sb_lock
);
2447 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2448 spin_unlock(&mp
->m_sb_lock
);
2452 xfs_icsb_modify_counters(
2454 xfs_sb_field_t field
,
2458 xfs_icsb_cnts_t
*icsbp
;
2459 long long lcounter
; /* long counter for 64 bit fields */
2465 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
2468 * if the counter is disabled, go to slow path
2470 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2472 xfs_icsb_lock_cntr(icsbp
);
2473 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2474 xfs_icsb_unlock_cntr(icsbp
);
2479 case XFS_SBS_ICOUNT
:
2480 lcounter
= icsbp
->icsb_icount
;
2482 if (unlikely(lcounter
< 0))
2483 goto balance_counter
;
2484 icsbp
->icsb_icount
= lcounter
;
2488 lcounter
= icsbp
->icsb_ifree
;
2490 if (unlikely(lcounter
< 0))
2491 goto balance_counter
;
2492 icsbp
->icsb_ifree
= lcounter
;
2495 case XFS_SBS_FDBLOCKS
:
2496 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2498 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2500 if (unlikely(lcounter
< 0))
2501 goto balance_counter
;
2502 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2508 xfs_icsb_unlock_cntr(icsbp
);
2516 * serialise with a mutex so we don't burn lots of cpu on
2517 * the superblock lock. We still need to hold the superblock
2518 * lock, however, when we modify the global structures.
2523 * Now running atomically.
2525 * If the counter is enabled, someone has beaten us to rebalancing.
2526 * Drop the lock and try again in the fast path....
2528 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2529 xfs_icsb_unlock(mp
);
2534 * The counter is currently disabled. Because we are
2535 * running atomically here, we know a rebalance cannot
2536 * be in progress. Hence we can go straight to operating
2537 * on the global superblock. We do not call xfs_mod_incore_sb()
2538 * here even though we need to get the m_sb_lock. Doing so
2539 * will cause us to re-enter this function and deadlock.
2540 * Hence we get the m_sb_lock ourselves and then call
2541 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2542 * directly on the global counters.
2544 spin_lock(&mp
->m_sb_lock
);
2545 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2546 spin_unlock(&mp
->m_sb_lock
);
2549 * Now that we've modified the global superblock, we
2550 * may be able to re-enable the distributed counters
2551 * (e.g. lots of space just got freed). After that
2555 xfs_icsb_balance_counter(mp
, field
, 0);
2556 xfs_icsb_unlock(mp
);
2560 xfs_icsb_unlock_cntr(icsbp
);
2564 * We may have multiple threads here if multiple per-cpu
2565 * counters run dry at the same time. This will mean we can
2566 * do more balances than strictly necessary but it is not
2567 * the common slowpath case.
2572 * running atomically.
2574 * This will leave the counter in the correct state for future
2575 * accesses. After the rebalance, we simply try again and our retry
2576 * will either succeed through the fast path or slow path without
2577 * another balance operation being required.
2579 xfs_icsb_balance_counter(mp
, field
, delta
);
2580 xfs_icsb_unlock(mp
);