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"
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 - can't mount");
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. 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 */))) {
373 xfs_warn(mp
, "SB sanity check 1 failed");
374 return XFS_ERROR(EFSCORRUPTED
);
378 * Sanity check AG count, size fields against data size field
381 sbp
->sb_dblocks
== 0 ||
383 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
384 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
385 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
387 xfs_warn(mp
, "SB sanity check 2 failed");
388 return XFS_ERROR(EFSCORRUPTED
);
392 * Until this is fixed only page-sized or smaller data blocks work.
394 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
397 "File system with blocksize %d bytes. "
398 "Only pagesize (%ld) or less will currently work.",
399 sbp
->sb_blocksize
, PAGE_SIZE
);
401 return XFS_ERROR(ENOSYS
);
405 * Currently only very few inode sizes are supported.
407 switch (sbp
->sb_inodesize
) {
415 xfs_warn(mp
, "inode size of %d bytes not supported",
417 return XFS_ERROR(ENOSYS
);
420 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
421 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
424 "file system too large to be mounted on this system.");
425 return XFS_ERROR(EFBIG
);
428 if (unlikely(sbp
->sb_inprogress
)) {
430 xfs_warn(mp
, "file system busy");
431 return XFS_ERROR(EFSCORRUPTED
);
435 * Version 1 directory format has never worked on Linux.
437 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
440 "file system using version 1 directory format");
441 return XFS_ERROR(ENOSYS
);
448 xfs_initialize_perag(
450 xfs_agnumber_t agcount
,
451 xfs_agnumber_t
*maxagi
)
453 xfs_agnumber_t index
, max_metadata
;
454 xfs_agnumber_t first_initialised
= 0;
458 xfs_sb_t
*sbp
= &mp
->m_sb
;
462 * Walk the current per-ag tree so we don't try to initialise AGs
463 * that already exist (growfs case). Allocate and insert all the
464 * AGs we don't find ready for initialisation.
466 for (index
= 0; index
< agcount
; index
++) {
467 pag
= xfs_perag_get(mp
, index
);
472 if (!first_initialised
)
473 first_initialised
= index
;
475 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
478 pag
->pag_agno
= index
;
480 spin_lock_init(&pag
->pag_ici_lock
);
481 mutex_init(&pag
->pag_ici_reclaim_lock
);
482 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
483 spin_lock_init(&pag
->pag_buf_lock
);
484 pag
->pag_buf_tree
= RB_ROOT
;
486 if (radix_tree_preload(GFP_NOFS
))
489 spin_lock(&mp
->m_perag_lock
);
490 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
492 spin_unlock(&mp
->m_perag_lock
);
493 radix_tree_preload_end();
497 spin_unlock(&mp
->m_perag_lock
);
498 radix_tree_preload_end();
502 * If we mount with the inode64 option, or no inode overflows
503 * the legacy 32-bit address space clear the inode32 option.
505 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
506 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
508 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
509 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
511 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
513 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
515 * Calculate how much should be reserved for inodes to meet
516 * the max inode percentage.
518 if (mp
->m_maxicount
) {
521 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
523 icount
+= sbp
->sb_agblocks
- 1;
524 do_div(icount
, sbp
->sb_agblocks
);
525 max_metadata
= icount
;
527 max_metadata
= agcount
;
530 for (index
= 0; index
< agcount
; index
++) {
531 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
532 if (ino
> XFS_MAXINUMBER_32
) {
537 pag
= xfs_perag_get(mp
, index
);
538 pag
->pagi_inodeok
= 1;
539 if (index
< max_metadata
)
540 pag
->pagf_metadata
= 1;
544 for (index
= 0; index
< agcount
; index
++) {
545 pag
= xfs_perag_get(mp
, index
);
546 pag
->pagi_inodeok
= 1;
557 for (; index
> first_initialised
; index
--) {
558 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
569 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
570 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
571 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
572 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
573 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
574 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
575 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
576 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
577 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
578 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
579 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
580 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
581 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
582 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
583 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
584 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
585 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
586 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
587 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
588 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
589 to
->sb_blocklog
= from
->sb_blocklog
;
590 to
->sb_sectlog
= from
->sb_sectlog
;
591 to
->sb_inodelog
= from
->sb_inodelog
;
592 to
->sb_inopblog
= from
->sb_inopblog
;
593 to
->sb_agblklog
= from
->sb_agblklog
;
594 to
->sb_rextslog
= from
->sb_rextslog
;
595 to
->sb_inprogress
= from
->sb_inprogress
;
596 to
->sb_imax_pct
= from
->sb_imax_pct
;
597 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
598 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
599 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
600 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
601 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
602 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
603 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
604 to
->sb_flags
= from
->sb_flags
;
605 to
->sb_shared_vn
= from
->sb_shared_vn
;
606 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
607 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
608 to
->sb_width
= be32_to_cpu(from
->sb_width
);
609 to
->sb_dirblklog
= from
->sb_dirblklog
;
610 to
->sb_logsectlog
= from
->sb_logsectlog
;
611 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
612 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
613 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
614 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
618 * Copy in core superblock to ondisk one.
620 * The fields argument is mask of superblock fields to copy.
628 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
629 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
639 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
640 first
= xfs_sb_info
[f
].offset
;
641 size
= xfs_sb_info
[f
+ 1].offset
- first
;
643 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
645 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
646 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
650 *(__be16
*)(to_ptr
+ first
) =
651 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
654 *(__be32
*)(to_ptr
+ first
) =
655 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
658 *(__be64
*)(to_ptr
+ first
) =
659 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
666 fields
&= ~(1LL << f
);
673 * Does the initial read of the superblock.
676 xfs_readsb(xfs_mount_t
*mp
, int flags
)
678 unsigned int sector_size
;
681 int loud
= !(flags
& XFS_MFSI_QUIET
);
683 ASSERT(mp
->m_sb_bp
== NULL
);
684 ASSERT(mp
->m_ddev_targp
!= NULL
);
687 * Allocate a (locked) buffer to hold the superblock.
688 * This will be kept around at all times to optimize
689 * access to the superblock.
691 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
694 bp
= xfs_buf_read_uncached(mp
, mp
->m_ddev_targp
,
695 XFS_SB_DADDR
, sector_size
, 0);
698 xfs_warn(mp
, "SB buffer read failed");
703 * Initialize the mount structure from the superblock.
704 * But first do some basic consistency checking.
706 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
707 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
710 xfs_warn(mp
, "SB validate failed");
715 * We must be able to do sector-sized and sector-aligned IO.
717 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
719 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
720 sector_size
, mp
->m_sb
.sb_sectsize
);
726 * If device sector size is smaller than the superblock size,
727 * re-read the superblock so the buffer is correctly sized.
729 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
731 sector_size
= mp
->m_sb
.sb_sectsize
;
735 /* Initialize per-cpu counters */
736 xfs_icsb_reinit_counters(mp
);
751 * Mount initialization code establishing various mount
752 * fields from the superblock associated with the given
756 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
758 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
759 spin_lock_init(&mp
->m_agirotor_lock
);
760 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
761 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
762 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
763 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
764 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
765 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
766 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
767 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
768 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
770 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
771 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
772 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
773 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
775 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
776 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
777 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
778 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
780 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
781 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
782 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
783 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
785 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
786 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
788 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
792 * xfs_initialize_perag_data
794 * Read in each per-ag structure so we can count up the number of
795 * allocated inodes, free inodes and used filesystem blocks as this
796 * information is no longer persistent in the superblock. Once we have
797 * this information, write it into the in-core superblock structure.
800 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
802 xfs_agnumber_t index
;
804 xfs_sb_t
*sbp
= &mp
->m_sb
;
808 uint64_t bfreelst
= 0;
812 for (index
= 0; index
< agcount
; index
++) {
814 * read the agf, then the agi. This gets us
815 * all the information we need and populates the
816 * per-ag structures for us.
818 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
822 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
825 pag
= xfs_perag_get(mp
, index
);
826 ifree
+= pag
->pagi_freecount
;
827 ialloc
+= pag
->pagi_count
;
828 bfree
+= pag
->pagf_freeblks
;
829 bfreelst
+= pag
->pagf_flcount
;
830 btree
+= pag
->pagf_btreeblks
;
834 * Overwrite incore superblock counters with just-read data
836 spin_lock(&mp
->m_sb_lock
);
837 sbp
->sb_ifree
= ifree
;
838 sbp
->sb_icount
= ialloc
;
839 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
840 spin_unlock(&mp
->m_sb_lock
);
842 /* Fixup the per-cpu counters as well. */
843 xfs_icsb_reinit_counters(mp
);
849 * Update alignment values based on mount options and sb values
852 xfs_update_alignment(xfs_mount_t
*mp
)
854 xfs_sb_t
*sbp
= &(mp
->m_sb
);
858 * If stripe unit and stripe width are not multiples
859 * of the fs blocksize turn off alignment.
861 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
862 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
863 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
864 xfs_warn(mp
, "alignment check 1 failed");
865 return XFS_ERROR(EINVAL
);
867 mp
->m_dalign
= mp
->m_swidth
= 0;
870 * Convert the stripe unit and width to FSBs.
872 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
873 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
874 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
875 return XFS_ERROR(EINVAL
);
878 "stripe alignment turned off: sunit(%d)/swidth(%d) "
879 "incompatible with agsize(%d)",
880 mp
->m_dalign
, mp
->m_swidth
,
885 } else if (mp
->m_dalign
) {
886 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
888 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
890 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
893 return XFS_ERROR(EINVAL
);
900 * Update superblock with new values
903 if (xfs_sb_version_hasdalign(sbp
)) {
904 if (sbp
->sb_unit
!= mp
->m_dalign
) {
905 sbp
->sb_unit
= mp
->m_dalign
;
906 mp
->m_update_flags
|= XFS_SB_UNIT
;
908 if (sbp
->sb_width
!= mp
->m_swidth
) {
909 sbp
->sb_width
= mp
->m_swidth
;
910 mp
->m_update_flags
|= XFS_SB_WIDTH
;
913 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
914 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
915 mp
->m_dalign
= sbp
->sb_unit
;
916 mp
->m_swidth
= sbp
->sb_width
;
923 * Set the maximum inode count for this filesystem
926 xfs_set_maxicount(xfs_mount_t
*mp
)
928 xfs_sb_t
*sbp
= &(mp
->m_sb
);
931 if (sbp
->sb_imax_pct
) {
933 * Make sure the maximum inode count is a multiple
934 * of the units we allocate inodes in.
936 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
938 do_div(icount
, mp
->m_ialloc_blks
);
939 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
947 * Set the default minimum read and write sizes unless
948 * already specified in a mount option.
949 * We use smaller I/O sizes when the file system
950 * is being used for NFS service (wsync mount option).
953 xfs_set_rw_sizes(xfs_mount_t
*mp
)
955 xfs_sb_t
*sbp
= &(mp
->m_sb
);
956 int readio_log
, writeio_log
;
958 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
959 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
960 readio_log
= XFS_WSYNC_READIO_LOG
;
961 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
963 readio_log
= XFS_READIO_LOG_LARGE
;
964 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
967 readio_log
= mp
->m_readio_log
;
968 writeio_log
= mp
->m_writeio_log
;
971 if (sbp
->sb_blocklog
> readio_log
) {
972 mp
->m_readio_log
= sbp
->sb_blocklog
;
974 mp
->m_readio_log
= readio_log
;
976 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
977 if (sbp
->sb_blocklog
> writeio_log
) {
978 mp
->m_writeio_log
= sbp
->sb_blocklog
;
980 mp
->m_writeio_log
= writeio_log
;
982 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
986 * precalculate the low space thresholds for dynamic speculative preallocation.
989 xfs_set_low_space_thresholds(
990 struct xfs_mount
*mp
)
994 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
995 __uint64_t space
= mp
->m_sb
.sb_dblocks
;
998 mp
->m_low_space
[i
] = space
* (i
+ 1);
1004 * Set whether we're using inode alignment.
1007 xfs_set_inoalignment(xfs_mount_t
*mp
)
1009 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
1010 mp
->m_sb
.sb_inoalignmt
>=
1011 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
1012 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
1014 mp
->m_inoalign_mask
= 0;
1016 * If we are using stripe alignment, check whether
1017 * the stripe unit is a multiple of the inode alignment
1019 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
1020 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
1021 mp
->m_sinoalign
= mp
->m_dalign
;
1023 mp
->m_sinoalign
= 0;
1027 * Check that the data (and log if separate) are an ok size.
1030 xfs_check_sizes(xfs_mount_t
*mp
)
1035 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
1036 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
1037 xfs_warn(mp
, "filesystem size mismatch detected");
1038 return XFS_ERROR(EFBIG
);
1040 bp
= xfs_buf_read_uncached(mp
, mp
->m_ddev_targp
,
1041 d
- XFS_FSS_TO_BB(mp
, 1),
1042 BBTOB(XFS_FSS_TO_BB(mp
, 1)), 0);
1044 xfs_warn(mp
, "last sector read failed");
1049 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
1050 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
1051 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
1052 xfs_warn(mp
, "log size mismatch detected");
1053 return XFS_ERROR(EFBIG
);
1055 bp
= xfs_buf_read_uncached(mp
, mp
->m_logdev_targp
,
1056 d
- XFS_FSB_TO_BB(mp
, 1),
1057 XFS_FSB_TO_B(mp
, 1), 0);
1059 xfs_warn(mp
, "log device read failed");
1068 * Clear the quotaflags in memory and in the superblock.
1071 xfs_mount_reset_sbqflags(
1072 struct xfs_mount
*mp
)
1075 struct xfs_trans
*tp
;
1080 * It is OK to look at sb_qflags here in mount path,
1081 * without m_sb_lock.
1083 if (mp
->m_sb
.sb_qflags
== 0)
1085 spin_lock(&mp
->m_sb_lock
);
1086 mp
->m_sb
.sb_qflags
= 0;
1087 spin_unlock(&mp
->m_sb_lock
);
1090 * If the fs is readonly, let the incore superblock run
1091 * with quotas off but don't flush the update out to disk
1093 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1097 xfs_notice(mp
, "Writing superblock quota changes");
1100 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1101 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1102 XFS_DEFAULT_LOG_COUNT
);
1104 xfs_trans_cancel(tp
, 0);
1105 xfs_alert(mp
, "%s: Superblock update failed!", __func__
);
1109 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
1110 return xfs_trans_commit(tp
, 0);
1114 xfs_default_resblks(xfs_mount_t
*mp
)
1119 * We default to 5% or 8192 fsbs of space reserved, whichever is
1120 * smaller. This is intended to cover concurrent allocation
1121 * transactions when we initially hit enospc. These each require a 4
1122 * block reservation. Hence by default we cover roughly 2000 concurrent
1123 * allocation reservations.
1125 resblks
= mp
->m_sb
.sb_dblocks
;
1126 do_div(resblks
, 20);
1127 resblks
= min_t(__uint64_t
, resblks
, 8192);
1132 * This function does the following on an initial mount of a file system:
1133 * - reads the superblock from disk and init the mount struct
1134 * - if we're a 32-bit kernel, do a size check on the superblock
1135 * so we don't mount terabyte filesystems
1136 * - init mount struct realtime fields
1137 * - allocate inode hash table for fs
1138 * - init directory manager
1139 * - perform recovery and init the log manager
1145 xfs_sb_t
*sbp
= &(mp
->m_sb
);
1148 uint quotamount
= 0;
1149 uint quotaflags
= 0;
1152 xfs_mount_common(mp
, sbp
);
1155 * Check for a mismatched features2 values. Older kernels
1156 * read & wrote into the wrong sb offset for sb_features2
1157 * on some platforms due to xfs_sb_t not being 64bit size aligned
1158 * when sb_features2 was added, which made older superblock
1159 * reading/writing routines swap it as a 64-bit value.
1161 * For backwards compatibility, we make both slots equal.
1163 * If we detect a mismatched field, we OR the set bits into the
1164 * existing features2 field in case it has already been modified; we
1165 * don't want to lose any features. We then update the bad location
1166 * with the ORed value so that older kernels will see any features2
1167 * flags, and mark the two fields as needing updates once the
1168 * transaction subsystem is online.
1170 if (xfs_sb_has_mismatched_features2(sbp
)) {
1171 xfs_warn(mp
, "correcting sb_features alignment problem");
1172 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
1173 sbp
->sb_bad_features2
= sbp
->sb_features2
;
1174 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
1177 * Re-check for ATTR2 in case it was found in bad_features2
1180 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1181 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
1182 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1185 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1186 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1187 xfs_sb_version_removeattr2(&mp
->m_sb
);
1188 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
1190 /* update sb_versionnum for the clearing of the morebits */
1191 if (!sbp
->sb_features2
)
1192 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
1196 * Check if sb_agblocks is aligned at stripe boundary
1197 * If sb_agblocks is NOT aligned turn off m_dalign since
1198 * allocator alignment is within an ag, therefore ag has
1199 * to be aligned at stripe boundary.
1201 error
= xfs_update_alignment(mp
);
1205 xfs_alloc_compute_maxlevels(mp
);
1206 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1207 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1208 xfs_ialloc_compute_maxlevels(mp
);
1210 xfs_set_maxicount(mp
);
1212 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
1214 error
= xfs_uuid_mount(mp
);
1219 * Set the minimum read and write sizes
1221 xfs_set_rw_sizes(mp
);
1223 /* set the low space thresholds for dynamic preallocation */
1224 xfs_set_low_space_thresholds(mp
);
1227 * Set the inode cluster size.
1228 * This may still be overridden by the file system
1229 * block size if it is larger than the chosen cluster size.
1231 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1234 * Set inode alignment fields
1236 xfs_set_inoalignment(mp
);
1239 * Check that the data (and log if separate) are an ok size.
1241 error
= xfs_check_sizes(mp
);
1243 goto out_remove_uuid
;
1246 * Initialize realtime fields in the mount structure
1248 error
= xfs_rtmount_init(mp
);
1250 xfs_warn(mp
, "RT mount failed");
1251 goto out_remove_uuid
;
1255 * Copies the low order bits of the timestamp and the randomly
1256 * set "sequence" number out of a UUID.
1258 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1260 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1265 * Initialize the attribute manager's entries.
1267 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1270 * Initialize the precomputed transaction reservations values.
1275 * Allocate and initialize the per-ag data.
1277 spin_lock_init(&mp
->m_perag_lock
);
1278 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
1279 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
1281 xfs_warn(mp
, "Failed per-ag init: %d", error
);
1282 goto out_remove_uuid
;
1285 if (!sbp
->sb_logblocks
) {
1286 xfs_warn(mp
, "no log defined");
1287 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1288 error
= XFS_ERROR(EFSCORRUPTED
);
1289 goto out_free_perag
;
1293 * log's mount-time initialization. Perform 1st part recovery if needed
1295 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1296 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1297 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1299 xfs_warn(mp
, "log mount failed");
1300 goto out_free_perag
;
1304 * Now the log is mounted, we know if it was an unclean shutdown or
1305 * not. If it was, with the first phase of recovery has completed, we
1306 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1307 * but they are recovered transactionally in the second recovery phase
1310 * Hence we can safely re-initialise incore superblock counters from
1311 * the per-ag data. These may not be correct if the filesystem was not
1312 * cleanly unmounted, so we need to wait for recovery to finish before
1315 * If the filesystem was cleanly unmounted, then we can trust the
1316 * values in the superblock to be correct and we don't need to do
1319 * If we are currently making the filesystem, the initialisation will
1320 * fail as the perag data is in an undefined state.
1322 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1323 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1324 !mp
->m_sb
.sb_inprogress
) {
1325 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1327 goto out_free_perag
;
1331 * Get and sanity-check the root inode.
1332 * Save the pointer to it in the mount structure.
1334 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
);
1336 xfs_warn(mp
, "failed to read root inode");
1337 goto out_log_dealloc
;
1340 ASSERT(rip
!= NULL
);
1342 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1343 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
1344 (unsigned long long)rip
->i_ino
);
1345 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1346 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1348 error
= XFS_ERROR(EFSCORRUPTED
);
1351 mp
->m_rootip
= rip
; /* save it */
1353 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1356 * Initialize realtime inode pointers in the mount structure
1358 error
= xfs_rtmount_inodes(mp
);
1361 * Free up the root inode.
1363 xfs_warn(mp
, "failed to read RT inodes");
1368 * If this is a read-only mount defer the superblock updates until
1369 * the next remount into writeable mode. Otherwise we would never
1370 * perform the update e.g. for the root filesystem.
1372 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1373 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1375 xfs_warn(mp
, "failed to write sb changes");
1381 * Initialise the XFS quota management subsystem for this mount
1383 if (XFS_IS_QUOTA_RUNNING(mp
)) {
1384 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
1388 ASSERT(!XFS_IS_QUOTA_ON(mp
));
1391 * If a file system had quotas running earlier, but decided to
1392 * mount without -o uquota/pquota/gquota options, revoke the
1393 * quotachecked license.
1395 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
1396 xfs_notice(mp
, "resetting quota flags");
1397 error
= xfs_mount_reset_sbqflags(mp
);
1404 * Finish recovering the file system. This part needed to be
1405 * delayed until after the root and real-time bitmap inodes
1406 * were consistently read in.
1408 error
= xfs_log_mount_finish(mp
);
1410 xfs_warn(mp
, "log mount finish failed");
1415 * Complete the quota initialisation, post-log-replay component.
1418 ASSERT(mp
->m_qflags
== 0);
1419 mp
->m_qflags
= quotaflags
;
1421 xfs_qm_mount_quotas(mp
);
1425 * Now we are mounted, reserve a small amount of unused space for
1426 * privileged transactions. This is needed so that transaction
1427 * space required for critical operations can dip into this pool
1428 * when at ENOSPC. This is needed for operations like create with
1429 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1430 * are not allowed to use this reserved space.
1432 * This may drive us straight to ENOSPC on mount, but that implies
1433 * we were already there on the last unmount. Warn if this occurs.
1435 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1436 resblks
= xfs_default_resblks(mp
);
1437 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1440 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1446 xfs_rtunmount_inodes(mp
);
1450 xfs_log_unmount(mp
);
1454 xfs_uuid_unmount(mp
);
1460 * This flushes out the inodes,dquots and the superblock, unmounts the
1461 * log and makes sure that incore structures are freed.
1465 struct xfs_mount
*mp
)
1470 xfs_qm_unmount_quotas(mp
);
1471 xfs_rtunmount_inodes(mp
);
1472 IRELE(mp
->m_rootip
);
1475 * We can potentially deadlock here if we have an inode cluster
1476 * that has been freed has its buffer still pinned in memory because
1477 * the transaction is still sitting in a iclog. The stale inodes
1478 * on that buffer will have their flush locks held until the
1479 * transaction hits the disk and the callbacks run. the inode
1480 * flush takes the flush lock unconditionally and with nothing to
1481 * push out the iclog we will never get that unlocked. hence we
1482 * need to force the log first.
1484 xfs_log_force(mp
, XFS_LOG_SYNC
);
1487 * Do a delwri reclaim pass first so that as many dirty inodes are
1488 * queued up for IO as possible. Then flush the buffers before making
1489 * a synchronous path to catch all the remaining inodes are reclaimed.
1490 * This makes the reclaim process as quick as possible by avoiding
1491 * synchronous writeout and blocking on inodes already in the delwri
1492 * state as much as possible.
1494 xfs_reclaim_inodes(mp
, 0);
1495 XFS_bflush(mp
->m_ddev_targp
);
1496 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
1501 * Flush out the log synchronously so that we know for sure
1502 * that nothing is pinned. This is important because bflush()
1503 * will skip pinned buffers.
1505 xfs_log_force(mp
, XFS_LOG_SYNC
);
1508 * Unreserve any blocks we have so that when we unmount we don't account
1509 * the reserved free space as used. This is really only necessary for
1510 * lazy superblock counting because it trusts the incore superblock
1511 * counters to be absolutely correct on clean unmount.
1513 * We don't bother correcting this elsewhere for lazy superblock
1514 * counting because on mount of an unclean filesystem we reconstruct the
1515 * correct counter value and this is irrelevant.
1517 * For non-lazy counter filesystems, this doesn't matter at all because
1518 * we only every apply deltas to the superblock and hence the incore
1519 * value does not matter....
1522 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1524 xfs_warn(mp
, "Unable to free reserved block pool. "
1525 "Freespace may not be correct on next mount.");
1527 error
= xfs_log_sbcount(mp
, 1);
1529 xfs_warn(mp
, "Unable to update superblock counters. "
1530 "Freespace may not be correct on next mount.");
1531 xfs_unmountfs_writesb(mp
);
1534 * Make sure all buffers have been flushed and completed before
1535 * unmounting the log.
1537 error
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
1539 xfs_warn(mp
, "%d busy buffers during unmount.", error
);
1540 xfs_wait_buftarg(mp
->m_ddev_targp
);
1542 xfs_log_unmount_write(mp
);
1543 xfs_log_unmount(mp
);
1544 xfs_uuid_unmount(mp
);
1547 xfs_errortag_clearall(mp
, 0);
1553 xfs_fs_writable(xfs_mount_t
*mp
)
1555 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1556 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1562 * Called either periodically to keep the on disk superblock values
1563 * roughly up to date or from unmount to make sure the values are
1564 * correct on a clean unmount.
1566 * Note this code can be called during the process of freezing, so
1567 * we may need to use the transaction allocator which does not not
1568 * block when the transaction subsystem is in its frozen state.
1578 if (!xfs_fs_writable(mp
))
1581 xfs_icsb_sync_counters(mp
, 0);
1584 * we don't need to do this if we are updating the superblock
1585 * counters on every modification.
1587 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1590 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1591 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1592 XFS_DEFAULT_LOG_COUNT
);
1594 xfs_trans_cancel(tp
, 0);
1598 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1600 xfs_trans_set_sync(tp
);
1601 error
= xfs_trans_commit(tp
, 0);
1606 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1612 * skip superblock write if fs is read-only, or
1613 * if we are doing a forced umount.
1615 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1616 XFS_FORCED_SHUTDOWN(mp
))) {
1618 sbp
= xfs_getsb(mp
, 0);
1620 XFS_BUF_UNDONE(sbp
);
1621 XFS_BUF_UNREAD(sbp
);
1622 XFS_BUF_UNDELAYWRITE(sbp
);
1624 XFS_BUF_UNASYNC(sbp
);
1625 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1626 xfsbdstrat(mp
, sbp
);
1627 error
= xfs_buf_iowait(sbp
);
1629 xfs_ioerror_alert("xfs_unmountfs_writesb",
1630 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1637 * xfs_mod_sb() can be used to copy arbitrary changes to the
1638 * in-core superblock into the superblock buffer to be logged.
1639 * It does not provide the higher level of locking that is
1640 * needed to protect the in-core superblock from concurrent
1644 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1656 bp
= xfs_trans_getsb(tp
, mp
, 0);
1657 first
= sizeof(xfs_sb_t
);
1660 /* translate/copy */
1662 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1664 /* find modified range */
1665 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1666 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1667 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1669 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1670 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1671 first
= xfs_sb_info
[f
].offset
;
1673 xfs_trans_log_buf(tp
, bp
, first
, last
);
1678 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1679 * a delta to a specified field in the in-core superblock. Simply
1680 * switch on the field indicated and apply the delta to that field.
1681 * Fields are not allowed to dip below zero, so if the delta would
1682 * do this do not apply it and return EINVAL.
1684 * The m_sb_lock must be held when this routine is called.
1687 xfs_mod_incore_sb_unlocked(
1689 xfs_sb_field_t field
,
1693 int scounter
; /* short counter for 32 bit fields */
1694 long long lcounter
; /* long counter for 64 bit fields */
1695 long long res_used
, rem
;
1698 * With the in-core superblock spin lock held, switch
1699 * on the indicated field. Apply the delta to the
1700 * proper field. If the fields value would dip below
1701 * 0, then do not apply the delta and return EINVAL.
1704 case XFS_SBS_ICOUNT
:
1705 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1709 return XFS_ERROR(EINVAL
);
1711 mp
->m_sb
.sb_icount
= lcounter
;
1714 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1718 return XFS_ERROR(EINVAL
);
1720 mp
->m_sb
.sb_ifree
= lcounter
;
1722 case XFS_SBS_FDBLOCKS
:
1723 lcounter
= (long long)
1724 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1725 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1727 if (delta
> 0) { /* Putting blocks back */
1728 if (res_used
> delta
) {
1729 mp
->m_resblks_avail
+= delta
;
1731 rem
= delta
- res_used
;
1732 mp
->m_resblks_avail
= mp
->m_resblks
;
1735 } else { /* Taking blocks away */
1737 if (lcounter
>= 0) {
1738 mp
->m_sb
.sb_fdblocks
= lcounter
+
1739 XFS_ALLOC_SET_ASIDE(mp
);
1744 * We are out of blocks, use any available reserved
1745 * blocks if were allowed to.
1748 return XFS_ERROR(ENOSPC
);
1750 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1751 if (lcounter
>= 0) {
1752 mp
->m_resblks_avail
= lcounter
;
1755 printk_once(KERN_WARNING
1756 "Filesystem \"%s\": reserve blocks depleted! "
1757 "Consider increasing reserve pool size.",
1759 return XFS_ERROR(ENOSPC
);
1762 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1764 case XFS_SBS_FREXTENTS
:
1765 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1768 return XFS_ERROR(ENOSPC
);
1770 mp
->m_sb
.sb_frextents
= lcounter
;
1772 case XFS_SBS_DBLOCKS
:
1773 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1777 return XFS_ERROR(EINVAL
);
1779 mp
->m_sb
.sb_dblocks
= lcounter
;
1781 case XFS_SBS_AGCOUNT
:
1782 scounter
= mp
->m_sb
.sb_agcount
;
1786 return XFS_ERROR(EINVAL
);
1788 mp
->m_sb
.sb_agcount
= scounter
;
1790 case XFS_SBS_IMAX_PCT
:
1791 scounter
= mp
->m_sb
.sb_imax_pct
;
1795 return XFS_ERROR(EINVAL
);
1797 mp
->m_sb
.sb_imax_pct
= scounter
;
1799 case XFS_SBS_REXTSIZE
:
1800 scounter
= mp
->m_sb
.sb_rextsize
;
1804 return XFS_ERROR(EINVAL
);
1806 mp
->m_sb
.sb_rextsize
= scounter
;
1808 case XFS_SBS_RBMBLOCKS
:
1809 scounter
= mp
->m_sb
.sb_rbmblocks
;
1813 return XFS_ERROR(EINVAL
);
1815 mp
->m_sb
.sb_rbmblocks
= scounter
;
1817 case XFS_SBS_RBLOCKS
:
1818 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1822 return XFS_ERROR(EINVAL
);
1824 mp
->m_sb
.sb_rblocks
= lcounter
;
1826 case XFS_SBS_REXTENTS
:
1827 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1831 return XFS_ERROR(EINVAL
);
1833 mp
->m_sb
.sb_rextents
= lcounter
;
1835 case XFS_SBS_REXTSLOG
:
1836 scounter
= mp
->m_sb
.sb_rextslog
;
1840 return XFS_ERROR(EINVAL
);
1842 mp
->m_sb
.sb_rextslog
= scounter
;
1846 return XFS_ERROR(EINVAL
);
1851 * xfs_mod_incore_sb() is used to change a field in the in-core
1852 * superblock structure by the specified delta. This modification
1853 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1854 * routine to do the work.
1858 struct xfs_mount
*mp
,
1859 xfs_sb_field_t field
,
1865 #ifdef HAVE_PERCPU_SB
1866 ASSERT(field
< XFS_SBS_ICOUNT
|| field
> XFS_SBS_FDBLOCKS
);
1868 spin_lock(&mp
->m_sb_lock
);
1869 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1870 spin_unlock(&mp
->m_sb_lock
);
1876 * Change more than one field in the in-core superblock structure at a time.
1878 * The fields and changes to those fields are specified in the array of
1879 * xfs_mod_sb structures passed in. Either all of the specified deltas
1880 * will be applied or none of them will. If any modified field dips below 0,
1881 * then all modifications will be backed out and EINVAL will be returned.
1883 * Note that this function may not be used for the superblock values that
1884 * are tracked with the in-memory per-cpu counters - a direct call to
1885 * xfs_icsb_modify_counters is required for these.
1888 xfs_mod_incore_sb_batch(
1889 struct xfs_mount
*mp
,
1898 * Loop through the array of mod structures and apply each individually.
1899 * If any fail, then back out all those which have already been applied.
1900 * Do all of this within the scope of the m_sb_lock so that all of the
1901 * changes will be atomic.
1903 spin_lock(&mp
->m_sb_lock
);
1904 for (msbp
= msb
; msbp
< (msb
+ nmsb
); msbp
++) {
1905 ASSERT(msbp
->msb_field
< XFS_SBS_ICOUNT
||
1906 msbp
->msb_field
> XFS_SBS_FDBLOCKS
);
1908 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1909 msbp
->msb_delta
, rsvd
);
1913 spin_unlock(&mp
->m_sb_lock
);
1917 while (--msbp
>= msb
) {
1918 error
= xfs_mod_incore_sb_unlocked(mp
, msbp
->msb_field
,
1919 -msbp
->msb_delta
, rsvd
);
1922 spin_unlock(&mp
->m_sb_lock
);
1927 * xfs_getsb() is called to obtain the buffer for the superblock.
1928 * The buffer is returned locked and read in from disk.
1929 * The buffer should be released with a call to xfs_brelse().
1931 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1932 * the superblock buffer if it can be locked without sleeping.
1933 * If it can't then we'll return NULL.
1942 ASSERT(mp
->m_sb_bp
!= NULL
);
1944 if (flags
& XBF_TRYLOCK
) {
1945 if (!XFS_BUF_CPSEMA(bp
)) {
1949 XFS_BUF_PSEMA(bp
, PRIBIO
);
1952 ASSERT(XFS_BUF_ISDONE(bp
));
1957 * Used to free the superblock along various error paths.
1961 struct xfs_mount
*mp
)
1963 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1971 * Used to log changes to the superblock unit and width fields which could
1972 * be altered by the mount options, as well as any potential sb_features2
1973 * fixup. Only the first superblock is updated.
1983 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1984 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1985 XFS_SB_VERSIONNUM
));
1987 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1988 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1989 XFS_DEFAULT_LOG_COUNT
);
1991 xfs_trans_cancel(tp
, 0);
1994 xfs_mod_sb(tp
, fields
);
1995 error
= xfs_trans_commit(tp
, 0);
2000 * If the underlying (data/log/rt) device is readonly, there are some
2001 * operations that cannot proceed.
2004 xfs_dev_is_read_only(
2005 struct xfs_mount
*mp
,
2008 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
2009 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
2010 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
2011 xfs_notice(mp
, "%s required on read-only device.", message
);
2012 xfs_notice(mp
, "write access unavailable, cannot proceed.");
2018 #ifdef HAVE_PERCPU_SB
2020 * Per-cpu incore superblock counters
2022 * Simple concept, difficult implementation
2024 * Basically, replace the incore superblock counters with a distributed per cpu
2025 * counter for contended fields (e.g. free block count).
2027 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2028 * hence needs to be accurately read when we are running low on space. Hence
2029 * there is a method to enable and disable the per-cpu counters based on how
2030 * much "stuff" is available in them.
2032 * Basically, a counter is enabled if there is enough free resource to justify
2033 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2034 * ENOSPC), then we disable the counters to synchronise all callers and
2035 * re-distribute the available resources.
2037 * If, once we redistributed the available resources, we still get a failure,
2038 * we disable the per-cpu counter and go through the slow path.
2040 * The slow path is the current xfs_mod_incore_sb() function. This means that
2041 * when we disable a per-cpu counter, we need to drain its resources back to
2042 * the global superblock. We do this after disabling the counter to prevent
2043 * more threads from queueing up on the counter.
2045 * Essentially, this means that we still need a lock in the fast path to enable
2046 * synchronisation between the global counters and the per-cpu counters. This
2047 * is not a problem because the lock will be local to a CPU almost all the time
2048 * and have little contention except when we get to ENOSPC conditions.
2050 * Basically, this lock becomes a barrier that enables us to lock out the fast
2051 * path while we do things like enabling and disabling counters and
2052 * synchronising the counters.
2056 * 1. m_sb_lock before picking up per-cpu locks
2057 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2058 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2059 * 4. modifying per-cpu counters requires holding per-cpu lock
2060 * 5. modifying global counters requires holding m_sb_lock
2061 * 6. enabling or disabling a counter requires holding the m_sb_lock
2062 * and _none_ of the per-cpu locks.
2064 * Disabled counters are only ever re-enabled by a balance operation
2065 * that results in more free resources per CPU than a given threshold.
2066 * To ensure counters don't remain disabled, they are rebalanced when
2067 * the global resource goes above a higher threshold (i.e. some hysteresis
2068 * is present to prevent thrashing).
2071 #ifdef CONFIG_HOTPLUG_CPU
2073 * hot-plug CPU notifier support.
2075 * We need a notifier per filesystem as we need to be able to identify
2076 * the filesystem to balance the counters out. This is achieved by
2077 * having a notifier block embedded in the xfs_mount_t and doing pointer
2078 * magic to get the mount pointer from the notifier block address.
2081 xfs_icsb_cpu_notify(
2082 struct notifier_block
*nfb
,
2083 unsigned long action
,
2086 xfs_icsb_cnts_t
*cntp
;
2089 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2090 cntp
= (xfs_icsb_cnts_t
*)
2091 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2093 case CPU_UP_PREPARE
:
2094 case CPU_UP_PREPARE_FROZEN
:
2095 /* Easy Case - initialize the area and locks, and
2096 * then rebalance when online does everything else for us. */
2097 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2100 case CPU_ONLINE_FROZEN
:
2102 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2103 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2104 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2105 xfs_icsb_unlock(mp
);
2108 case CPU_DEAD_FROZEN
:
2109 /* Disable all the counters, then fold the dead cpu's
2110 * count into the total on the global superblock and
2111 * re-enable the counters. */
2113 spin_lock(&mp
->m_sb_lock
);
2114 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2115 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2116 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2118 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2119 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2120 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2122 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2124 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
2125 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
2126 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
2127 spin_unlock(&mp
->m_sb_lock
);
2128 xfs_icsb_unlock(mp
);
2134 #endif /* CONFIG_HOTPLUG_CPU */
2137 xfs_icsb_init_counters(
2140 xfs_icsb_cnts_t
*cntp
;
2143 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2144 if (mp
->m_sb_cnts
== NULL
)
2147 #ifdef CONFIG_HOTPLUG_CPU
2148 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2149 mp
->m_icsb_notifier
.priority
= 0;
2150 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2151 #endif /* CONFIG_HOTPLUG_CPU */
2153 for_each_online_cpu(i
) {
2154 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2155 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2158 mutex_init(&mp
->m_icsb_mutex
);
2161 * start with all counters disabled so that the
2162 * initial balance kicks us off correctly
2164 mp
->m_icsb_counters
= -1;
2169 xfs_icsb_reinit_counters(
2174 * start with all counters disabled so that the
2175 * initial balance kicks us off correctly
2177 mp
->m_icsb_counters
= -1;
2178 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2179 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2180 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2181 xfs_icsb_unlock(mp
);
2185 xfs_icsb_destroy_counters(
2188 if (mp
->m_sb_cnts
) {
2189 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2190 free_percpu(mp
->m_sb_cnts
);
2192 mutex_destroy(&mp
->m_icsb_mutex
);
2197 xfs_icsb_cnts_t
*icsbp
)
2199 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2205 xfs_icsb_unlock_cntr(
2206 xfs_icsb_cnts_t
*icsbp
)
2208 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2213 xfs_icsb_lock_all_counters(
2216 xfs_icsb_cnts_t
*cntp
;
2219 for_each_online_cpu(i
) {
2220 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2221 xfs_icsb_lock_cntr(cntp
);
2226 xfs_icsb_unlock_all_counters(
2229 xfs_icsb_cnts_t
*cntp
;
2232 for_each_online_cpu(i
) {
2233 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2234 xfs_icsb_unlock_cntr(cntp
);
2241 xfs_icsb_cnts_t
*cnt
,
2244 xfs_icsb_cnts_t
*cntp
;
2247 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2249 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2250 xfs_icsb_lock_all_counters(mp
);
2252 for_each_online_cpu(i
) {
2253 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2254 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2255 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2256 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2259 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2260 xfs_icsb_unlock_all_counters(mp
);
2264 xfs_icsb_counter_disabled(
2266 xfs_sb_field_t field
)
2268 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2269 return test_bit(field
, &mp
->m_icsb_counters
);
2273 xfs_icsb_disable_counter(
2275 xfs_sb_field_t field
)
2277 xfs_icsb_cnts_t cnt
;
2279 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2282 * If we are already disabled, then there is nothing to do
2283 * here. We check before locking all the counters to avoid
2284 * the expensive lock operation when being called in the
2285 * slow path and the counter is already disabled. This is
2286 * safe because the only time we set or clear this state is under
2289 if (xfs_icsb_counter_disabled(mp
, field
))
2292 xfs_icsb_lock_all_counters(mp
);
2293 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2294 /* drain back to superblock */
2296 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2298 case XFS_SBS_ICOUNT
:
2299 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2302 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2304 case XFS_SBS_FDBLOCKS
:
2305 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2312 xfs_icsb_unlock_all_counters(mp
);
2316 xfs_icsb_enable_counter(
2318 xfs_sb_field_t field
,
2322 xfs_icsb_cnts_t
*cntp
;
2325 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2327 xfs_icsb_lock_all_counters(mp
);
2328 for_each_online_cpu(i
) {
2329 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2331 case XFS_SBS_ICOUNT
:
2332 cntp
->icsb_icount
= count
+ resid
;
2335 cntp
->icsb_ifree
= count
+ resid
;
2337 case XFS_SBS_FDBLOCKS
:
2338 cntp
->icsb_fdblocks
= count
+ resid
;
2346 clear_bit(field
, &mp
->m_icsb_counters
);
2347 xfs_icsb_unlock_all_counters(mp
);
2351 xfs_icsb_sync_counters_locked(
2355 xfs_icsb_cnts_t cnt
;
2357 xfs_icsb_count(mp
, &cnt
, flags
);
2359 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2360 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2361 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2362 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2363 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2364 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2368 * Accurate update of per-cpu counters to incore superblock
2371 xfs_icsb_sync_counters(
2375 spin_lock(&mp
->m_sb_lock
);
2376 xfs_icsb_sync_counters_locked(mp
, flags
);
2377 spin_unlock(&mp
->m_sb_lock
);
2381 * Balance and enable/disable counters as necessary.
2383 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2384 * chosen to be the same number as single on disk allocation chunk per CPU, and
2385 * free blocks is something far enough zero that we aren't going thrash when we
2386 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2387 * prevent looping endlessly when xfs_alloc_space asks for more than will
2388 * be distributed to a single CPU but each CPU has enough blocks to be
2391 * Note that we can be called when counters are already disabled.
2392 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2393 * prevent locking every per-cpu counter needlessly.
2396 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2397 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2398 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2400 xfs_icsb_balance_counter_locked(
2402 xfs_sb_field_t field
,
2405 uint64_t count
, resid
;
2406 int weight
= num_online_cpus();
2407 uint64_t min
= (uint64_t)min_per_cpu
;
2409 /* disable counter and sync counter */
2410 xfs_icsb_disable_counter(mp
, field
);
2412 /* update counters - first CPU gets residual*/
2414 case XFS_SBS_ICOUNT
:
2415 count
= mp
->m_sb
.sb_icount
;
2416 resid
= do_div(count
, weight
);
2417 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2421 count
= mp
->m_sb
.sb_ifree
;
2422 resid
= do_div(count
, weight
);
2423 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2426 case XFS_SBS_FDBLOCKS
:
2427 count
= mp
->m_sb
.sb_fdblocks
;
2428 resid
= do_div(count
, weight
);
2429 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2434 count
= resid
= 0; /* quiet, gcc */
2438 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2442 xfs_icsb_balance_counter(
2444 xfs_sb_field_t fields
,
2447 spin_lock(&mp
->m_sb_lock
);
2448 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2449 spin_unlock(&mp
->m_sb_lock
);
2453 xfs_icsb_modify_counters(
2455 xfs_sb_field_t field
,
2459 xfs_icsb_cnts_t
*icsbp
;
2460 long long lcounter
; /* long counter for 64 bit fields */
2466 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
2469 * if the counter is disabled, go to slow path
2471 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2473 xfs_icsb_lock_cntr(icsbp
);
2474 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2475 xfs_icsb_unlock_cntr(icsbp
);
2480 case XFS_SBS_ICOUNT
:
2481 lcounter
= icsbp
->icsb_icount
;
2483 if (unlikely(lcounter
< 0))
2484 goto balance_counter
;
2485 icsbp
->icsb_icount
= lcounter
;
2489 lcounter
= icsbp
->icsb_ifree
;
2491 if (unlikely(lcounter
< 0))
2492 goto balance_counter
;
2493 icsbp
->icsb_ifree
= lcounter
;
2496 case XFS_SBS_FDBLOCKS
:
2497 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2499 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2501 if (unlikely(lcounter
< 0))
2502 goto balance_counter
;
2503 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2509 xfs_icsb_unlock_cntr(icsbp
);
2517 * serialise with a mutex so we don't burn lots of cpu on
2518 * the superblock lock. We still need to hold the superblock
2519 * lock, however, when we modify the global structures.
2524 * Now running atomically.
2526 * If the counter is enabled, someone has beaten us to rebalancing.
2527 * Drop the lock and try again in the fast path....
2529 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2530 xfs_icsb_unlock(mp
);
2535 * The counter is currently disabled. Because we are
2536 * running atomically here, we know a rebalance cannot
2537 * be in progress. Hence we can go straight to operating
2538 * on the global superblock. We do not call xfs_mod_incore_sb()
2539 * here even though we need to get the m_sb_lock. Doing so
2540 * will cause us to re-enter this function and deadlock.
2541 * Hence we get the m_sb_lock ourselves and then call
2542 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2543 * directly on the global counters.
2545 spin_lock(&mp
->m_sb_lock
);
2546 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2547 spin_unlock(&mp
->m_sb_lock
);
2550 * Now that we've modified the global superblock, we
2551 * may be able to re-enable the distributed counters
2552 * (e.g. lots of space just got freed). After that
2556 xfs_icsb_balance_counter(mp
, field
, 0);
2557 xfs_icsb_unlock(mp
);
2561 xfs_icsb_unlock_cntr(icsbp
);
2565 * We may have multiple threads here if multiple per-cpu
2566 * counters run dry at the same time. This will mean we can
2567 * do more balances than strictly necessary but it is not
2568 * the common slowpath case.
2573 * running atomically.
2575 * This will leave the counter in the correct state for future
2576 * accesses. After the rebalance, we simply try again and our retry
2577 * will either succeed through the fast path or slow path without
2578 * another balance operation being required.
2580 xfs_icsb_balance_counter(mp
, field
, delta
);
2581 xfs_icsb_unlock(mp
);