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_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_fsops.h"
46 #include "xfs_utils.h"
47 #include "xfs_trace.h"
50 STATIC
void xfs_unmountfs_wait(xfs_mount_t
*);
54 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
56 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
58 STATIC
int xfs_icsb_modify_counters(xfs_mount_t
*, xfs_sb_field_t
,
60 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
64 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
65 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
66 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
72 short type
; /* 0 = integer
73 * 1 = binary / string (no translation)
76 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
77 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
78 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
79 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
80 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
81 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
82 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
83 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
84 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
85 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
86 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
87 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
88 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
89 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
90 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
91 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
92 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
93 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
94 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
95 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
96 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
97 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
98 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
99 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
100 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
101 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
102 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
103 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
104 { offsetof(xfs_sb_t
, sb_icount
), 0 },
105 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
106 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
107 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
108 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
109 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
110 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
111 { offsetof(xfs_sb_t
, sb_flags
), 0 },
112 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
113 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
114 { offsetof(xfs_sb_t
, sb_unit
), 0 },
115 { offsetof(xfs_sb_t
, sb_width
), 0 },
116 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
117 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
118 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
119 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
120 { offsetof(xfs_sb_t
, sb_features2
), 0 },
121 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
122 { sizeof(xfs_sb_t
), 0 }
125 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
126 static int xfs_uuid_table_size
;
127 static uuid_t
*xfs_uuid_table
;
130 * See if the UUID is unique among mounted XFS filesystems.
131 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
135 struct xfs_mount
*mp
)
137 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
140 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
143 if (uuid_is_nil(uuid
)) {
145 "XFS: Filesystem %s has nil UUID - can't mount",
147 return XFS_ERROR(EINVAL
);
150 mutex_lock(&xfs_uuid_table_mutex
);
151 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
152 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
156 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
161 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
162 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
163 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
165 hole
= xfs_uuid_table_size
++;
167 xfs_uuid_table
[hole
] = *uuid
;
168 mutex_unlock(&xfs_uuid_table_mutex
);
173 mutex_unlock(&xfs_uuid_table_mutex
);
174 cmn_err(CE_WARN
, "XFS: Filesystem %s has duplicate UUID - can't mount",
176 return XFS_ERROR(EINVAL
);
181 struct xfs_mount
*mp
)
183 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
186 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
189 mutex_lock(&xfs_uuid_table_mutex
);
190 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
191 if (uuid_is_nil(&xfs_uuid_table
[i
]))
193 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
195 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
198 ASSERT(i
< xfs_uuid_table_size
);
199 mutex_unlock(&xfs_uuid_table_mutex
);
204 * Reference counting access wrappers to the perag structures.
207 xfs_perag_get(struct xfs_mount
*mp
, xfs_agnumber_t agno
)
209 struct xfs_perag
*pag
;
212 spin_lock(&mp
->m_perag_lock
);
213 pag
= radix_tree_lookup(&mp
->m_perag_tree
, agno
);
215 ASSERT(atomic_read(&pag
->pag_ref
) >= 0);
216 /* catch leaks in the positive direction during testing */
217 ASSERT(atomic_read(&pag
->pag_ref
) < 1000);
218 ref
= atomic_inc_return(&pag
->pag_ref
);
220 spin_unlock(&mp
->m_perag_lock
);
221 trace_xfs_perag_get(mp
, agno
, ref
, _RET_IP_
);
226 xfs_perag_put(struct xfs_perag
*pag
)
230 ASSERT(atomic_read(&pag
->pag_ref
) > 0);
231 ref
= atomic_dec_return(&pag
->pag_ref
);
232 trace_xfs_perag_put(pag
->pag_mount
, pag
->pag_agno
, ref
, _RET_IP_
);
236 * Free up the resources associated with a mount structure. Assume that
237 * the structure was initially zeroed, so we can tell which fields got
245 struct xfs_perag
*pag
;
247 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
248 spin_lock(&mp
->m_perag_lock
);
249 pag
= radix_tree_delete(&mp
->m_perag_tree
, agno
);
251 ASSERT(atomic_read(&pag
->pag_ref
) == 0);
252 spin_unlock(&mp
->m_perag_lock
);
258 * Check size of device based on the (data/realtime) block count.
259 * Note: this check is used by the growfs code as well as mount.
262 xfs_sb_validate_fsb_count(
266 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
267 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
269 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
270 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
272 #else /* Limited by UINT_MAX of sectors */
273 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
280 * Check the validity of the SB found.
283 xfs_mount_validate_sb(
289 * If the log device and data device have the
290 * same device number, the log is internal.
291 * Consequently, the sb_logstart should be non-zero. If
292 * we have a zero sb_logstart in this case, we may be trying to mount
293 * a volume filesystem in a non-volume manner.
295 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
296 xfs_fs_mount_cmn_err(flags
, "bad magic number");
297 return XFS_ERROR(EWRONGFS
);
300 if (!xfs_sb_good_version(sbp
)) {
301 xfs_fs_mount_cmn_err(flags
, "bad version");
302 return XFS_ERROR(EWRONGFS
);
306 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
307 xfs_fs_mount_cmn_err(flags
,
308 "filesystem is marked as having an external log; "
309 "specify logdev on the\nmount command line.");
310 return XFS_ERROR(EINVAL
);
314 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
315 xfs_fs_mount_cmn_err(flags
,
316 "filesystem is marked as having an internal log; "
317 "do not specify logdev on\nthe mount command line.");
318 return XFS_ERROR(EINVAL
);
322 * More sanity checking. These were stolen directly from
326 sbp
->sb_agcount
<= 0 ||
327 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
328 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
329 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
330 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
331 sbp
->sb_sectsize
!= (1 << sbp
->sb_sectlog
) ||
332 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
333 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
334 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
335 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
336 sbp
->sb_blocksize
!= (1 << sbp
->sb_blocklog
) ||
337 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
338 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
339 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
340 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
341 sbp
->sb_inodesize
!= (1 << sbp
->sb_inodelog
) ||
342 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
343 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
344 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
345 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
346 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
347 return XFS_ERROR(EFSCORRUPTED
);
351 * Sanity check AG count, size fields against data size field
354 sbp
->sb_dblocks
== 0 ||
356 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
357 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
358 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
359 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
360 return XFS_ERROR(EFSCORRUPTED
);
364 * Until this is fixed only page-sized or smaller data blocks work.
366 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
367 xfs_fs_mount_cmn_err(flags
,
368 "file system with blocksize %d bytes",
370 xfs_fs_mount_cmn_err(flags
,
371 "only pagesize (%ld) or less will currently work.",
373 return XFS_ERROR(ENOSYS
);
377 * Currently only very few inode sizes are supported.
379 switch (sbp
->sb_inodesize
) {
386 xfs_fs_mount_cmn_err(flags
,
387 "inode size of %d bytes not supported",
389 return XFS_ERROR(ENOSYS
);
392 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
393 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
394 xfs_fs_mount_cmn_err(flags
,
395 "file system too large to be mounted on this system.");
396 return XFS_ERROR(E2BIG
);
399 if (unlikely(sbp
->sb_inprogress
)) {
400 xfs_fs_mount_cmn_err(flags
, "file system busy");
401 return XFS_ERROR(EFSCORRUPTED
);
405 * Version 1 directory format has never worked on Linux.
407 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
408 xfs_fs_mount_cmn_err(flags
,
409 "file system using version 1 directory format");
410 return XFS_ERROR(ENOSYS
);
417 xfs_initialize_perag_icache(
420 if (!pag
->pag_ici_init
) {
421 rwlock_init(&pag
->pag_ici_lock
);
422 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
423 pag
->pag_ici_init
= 1;
428 xfs_initialize_perag(
430 xfs_agnumber_t agcount
,
431 xfs_agnumber_t
*maxagi
)
433 xfs_agnumber_t index
, max_metadata
;
434 xfs_agnumber_t first_initialised
= 0;
438 xfs_sb_t
*sbp
= &mp
->m_sb
;
439 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
442 /* Check to see if the filesystem can overflow 32 bit inodes */
443 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
444 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
447 * Walk the current per-ag tree so we don't try to initialise AGs
448 * that already exist (growfs case). Allocate and insert all the
449 * AGs we don't find ready for initialisation.
451 for (index
= 0; index
< agcount
; index
++) {
452 pag
= xfs_perag_get(mp
, index
);
457 if (!first_initialised
)
458 first_initialised
= index
;
459 pag
= kmem_zalloc(sizeof(*pag
), KM_MAYFAIL
);
462 if (radix_tree_preload(GFP_NOFS
))
464 spin_lock(&mp
->m_perag_lock
);
465 if (radix_tree_insert(&mp
->m_perag_tree
, index
, pag
)) {
467 spin_unlock(&mp
->m_perag_lock
);
468 radix_tree_preload_end();
472 pag
->pag_agno
= index
;
474 spin_unlock(&mp
->m_perag_lock
);
475 radix_tree_preload_end();
478 /* Clear the mount flag if no inode can overflow 32 bits
479 * on this filesystem, or if specifically requested..
481 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> max_inum
) {
482 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
484 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
487 /* If we can overflow then setup the ag headers accordingly */
488 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
489 /* Calculate how much should be reserved for inodes to
490 * meet the max inode percentage.
492 if (mp
->m_maxicount
) {
495 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
497 icount
+= sbp
->sb_agblocks
- 1;
498 do_div(icount
, sbp
->sb_agblocks
);
499 max_metadata
= icount
;
501 max_metadata
= agcount
;
503 for (index
= 0; index
< agcount
; index
++) {
504 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
505 if (ino
> max_inum
) {
510 /* This ag is preferred for inodes */
511 pag
= xfs_perag_get(mp
, index
);
512 pag
->pagi_inodeok
= 1;
513 if (index
< max_metadata
)
514 pag
->pagf_metadata
= 1;
515 xfs_initialize_perag_icache(pag
);
519 /* Setup default behavior for smaller filesystems */
520 for (index
= 0; index
< agcount
; index
++) {
521 pag
= xfs_perag_get(mp
, index
);
522 pag
->pagi_inodeok
= 1;
523 xfs_initialize_perag_icache(pag
);
533 for (; index
> first_initialised
; index
--) {
534 pag
= radix_tree_delete(&mp
->m_perag_tree
, index
);
545 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
546 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
547 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
548 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
549 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
550 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
551 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
552 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
553 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
554 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
555 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
556 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
557 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
558 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
559 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
560 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
561 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
562 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
563 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
564 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
565 to
->sb_blocklog
= from
->sb_blocklog
;
566 to
->sb_sectlog
= from
->sb_sectlog
;
567 to
->sb_inodelog
= from
->sb_inodelog
;
568 to
->sb_inopblog
= from
->sb_inopblog
;
569 to
->sb_agblklog
= from
->sb_agblklog
;
570 to
->sb_rextslog
= from
->sb_rextslog
;
571 to
->sb_inprogress
= from
->sb_inprogress
;
572 to
->sb_imax_pct
= from
->sb_imax_pct
;
573 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
574 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
575 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
576 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
577 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
578 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
579 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
580 to
->sb_flags
= from
->sb_flags
;
581 to
->sb_shared_vn
= from
->sb_shared_vn
;
582 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
583 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
584 to
->sb_width
= be32_to_cpu(from
->sb_width
);
585 to
->sb_dirblklog
= from
->sb_dirblklog
;
586 to
->sb_logsectlog
= from
->sb_logsectlog
;
587 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
588 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
589 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
590 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
594 * Copy in core superblock to ondisk one.
596 * The fields argument is mask of superblock fields to copy.
604 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
605 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
615 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
616 first
= xfs_sb_info
[f
].offset
;
617 size
= xfs_sb_info
[f
+ 1].offset
- first
;
619 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
621 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
622 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
626 *(__be16
*)(to_ptr
+ first
) =
627 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
630 *(__be32
*)(to_ptr
+ first
) =
631 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
634 *(__be64
*)(to_ptr
+ first
) =
635 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
642 fields
&= ~(1LL << f
);
649 * Does the initial read of the superblock.
652 xfs_readsb(xfs_mount_t
*mp
, int flags
)
654 unsigned int sector_size
;
655 unsigned int extra_flags
;
659 ASSERT(mp
->m_sb_bp
== NULL
);
660 ASSERT(mp
->m_ddev_targp
!= NULL
);
663 * Allocate a (locked) buffer to hold the superblock.
664 * This will be kept around at all times to optimize
665 * access to the superblock.
667 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
668 extra_flags
= XBF_LOCK
| XBF_FS_MANAGED
| XBF_MAPPED
;
670 bp
= xfs_buf_read(mp
->m_ddev_targp
, XFS_SB_DADDR
, BTOBB(sector_size
),
672 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
673 xfs_fs_mount_cmn_err(flags
, "SB read failed");
674 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
677 ASSERT(XFS_BUF_ISBUSY(bp
));
678 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
681 * Initialize the mount structure from the superblock.
682 * But first do some basic consistency checking.
684 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
686 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
688 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
693 * We must be able to do sector-sized and sector-aligned IO.
695 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
696 xfs_fs_mount_cmn_err(flags
,
697 "device supports only %u byte sectors (not %u)",
698 sector_size
, mp
->m_sb
.sb_sectsize
);
704 * If device sector size is smaller than the superblock size,
705 * re-read the superblock so the buffer is correctly sized.
707 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
708 XFS_BUF_UNMANAGE(bp
);
710 sector_size
= mp
->m_sb
.sb_sectsize
;
711 bp
= xfs_buf_read(mp
->m_ddev_targp
, XFS_SB_DADDR
,
712 BTOBB(sector_size
), extra_flags
);
713 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
714 xfs_fs_mount_cmn_err(flags
, "SB re-read failed");
715 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
718 ASSERT(XFS_BUF_ISBUSY(bp
));
719 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
722 /* Initialize per-cpu counters */
723 xfs_icsb_reinit_counters(mp
);
727 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
732 XFS_BUF_UNMANAGE(bp
);
742 * Mount initialization code establishing various mount
743 * fields from the superblock associated with the given
747 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
749 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
750 spin_lock_init(&mp
->m_agirotor_lock
);
751 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
752 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
753 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
754 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
755 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
756 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
757 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
758 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
759 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
761 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
762 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
763 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
764 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
766 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
767 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
768 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
769 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
771 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
772 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
773 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
774 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
776 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
777 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
779 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
783 * xfs_initialize_perag_data
785 * Read in each per-ag structure so we can count up the number of
786 * allocated inodes, free inodes and used filesystem blocks as this
787 * information is no longer persistent in the superblock. Once we have
788 * this information, write it into the in-core superblock structure.
791 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
793 xfs_agnumber_t index
;
795 xfs_sb_t
*sbp
= &mp
->m_sb
;
799 uint64_t bfreelst
= 0;
803 for (index
= 0; index
< agcount
; index
++) {
805 * read the agf, then the agi. This gets us
806 * all the information we need and populates the
807 * per-ag structures for us.
809 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
813 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
816 pag
= xfs_perag_get(mp
, index
);
817 ifree
+= pag
->pagi_freecount
;
818 ialloc
+= pag
->pagi_count
;
819 bfree
+= pag
->pagf_freeblks
;
820 bfreelst
+= pag
->pagf_flcount
;
821 btree
+= pag
->pagf_btreeblks
;
825 * Overwrite incore superblock counters with just-read data
827 spin_lock(&mp
->m_sb_lock
);
828 sbp
->sb_ifree
= ifree
;
829 sbp
->sb_icount
= ialloc
;
830 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
831 spin_unlock(&mp
->m_sb_lock
);
833 /* Fixup the per-cpu counters as well. */
834 xfs_icsb_reinit_counters(mp
);
840 * Update alignment values based on mount options and sb values
843 xfs_update_alignment(xfs_mount_t
*mp
)
845 xfs_sb_t
*sbp
= &(mp
->m_sb
);
849 * If stripe unit and stripe width are not multiples
850 * of the fs blocksize turn off alignment.
852 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
853 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
854 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
856 "XFS: alignment check 1 failed");
857 return XFS_ERROR(EINVAL
);
859 mp
->m_dalign
= mp
->m_swidth
= 0;
862 * Convert the stripe unit and width to FSBs.
864 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
865 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
866 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
867 return XFS_ERROR(EINVAL
);
869 xfs_fs_cmn_err(CE_WARN
, mp
,
870 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
871 mp
->m_dalign
, mp
->m_swidth
,
876 } else if (mp
->m_dalign
) {
877 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
879 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
880 xfs_fs_cmn_err(CE_WARN
, mp
,
881 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
884 return XFS_ERROR(EINVAL
);
891 * Update superblock with new values
894 if (xfs_sb_version_hasdalign(sbp
)) {
895 if (sbp
->sb_unit
!= mp
->m_dalign
) {
896 sbp
->sb_unit
= mp
->m_dalign
;
897 mp
->m_update_flags
|= XFS_SB_UNIT
;
899 if (sbp
->sb_width
!= mp
->m_swidth
) {
900 sbp
->sb_width
= mp
->m_swidth
;
901 mp
->m_update_flags
|= XFS_SB_WIDTH
;
904 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
905 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
906 mp
->m_dalign
= sbp
->sb_unit
;
907 mp
->m_swidth
= sbp
->sb_width
;
914 * Set the maximum inode count for this filesystem
917 xfs_set_maxicount(xfs_mount_t
*mp
)
919 xfs_sb_t
*sbp
= &(mp
->m_sb
);
922 if (sbp
->sb_imax_pct
) {
924 * Make sure the maximum inode count is a multiple
925 * of the units we allocate inodes in.
927 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
929 do_div(icount
, mp
->m_ialloc_blks
);
930 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
938 * Set the default minimum read and write sizes unless
939 * already specified in a mount option.
940 * We use smaller I/O sizes when the file system
941 * is being used for NFS service (wsync mount option).
944 xfs_set_rw_sizes(xfs_mount_t
*mp
)
946 xfs_sb_t
*sbp
= &(mp
->m_sb
);
947 int readio_log
, writeio_log
;
949 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
950 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
951 readio_log
= XFS_WSYNC_READIO_LOG
;
952 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
954 readio_log
= XFS_READIO_LOG_LARGE
;
955 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
958 readio_log
= mp
->m_readio_log
;
959 writeio_log
= mp
->m_writeio_log
;
962 if (sbp
->sb_blocklog
> readio_log
) {
963 mp
->m_readio_log
= sbp
->sb_blocklog
;
965 mp
->m_readio_log
= readio_log
;
967 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
968 if (sbp
->sb_blocklog
> writeio_log
) {
969 mp
->m_writeio_log
= sbp
->sb_blocklog
;
971 mp
->m_writeio_log
= writeio_log
;
973 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
977 * Set whether we're using inode alignment.
980 xfs_set_inoalignment(xfs_mount_t
*mp
)
982 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
983 mp
->m_sb
.sb_inoalignmt
>=
984 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
985 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
987 mp
->m_inoalign_mask
= 0;
989 * If we are using stripe alignment, check whether
990 * the stripe unit is a multiple of the inode alignment
992 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
993 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
994 mp
->m_sinoalign
= mp
->m_dalign
;
1000 * Check that the data (and log if separate) are an ok size.
1003 xfs_check_sizes(xfs_mount_t
*mp
)
1009 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
1010 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
1011 cmn_err(CE_WARN
, "XFS: size check 1 failed");
1012 return XFS_ERROR(E2BIG
);
1014 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
1015 d
- XFS_FSS_TO_BB(mp
, 1),
1016 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
1020 cmn_err(CE_WARN
, "XFS: size check 2 failed");
1021 if (error
== ENOSPC
)
1022 error
= XFS_ERROR(E2BIG
);
1026 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
1027 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
1028 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
1029 cmn_err(CE_WARN
, "XFS: size check 3 failed");
1030 return XFS_ERROR(E2BIG
);
1032 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
1033 d
- XFS_FSB_TO_BB(mp
, 1),
1034 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
1038 cmn_err(CE_WARN
, "XFS: size check 3 failed");
1039 if (error
== ENOSPC
)
1040 error
= XFS_ERROR(E2BIG
);
1048 * Clear the quotaflags in memory and in the superblock.
1051 xfs_mount_reset_sbqflags(
1052 struct xfs_mount
*mp
)
1055 struct xfs_trans
*tp
;
1060 * It is OK to look at sb_qflags here in mount path,
1061 * without m_sb_lock.
1063 if (mp
->m_sb
.sb_qflags
== 0)
1065 spin_lock(&mp
->m_sb_lock
);
1066 mp
->m_sb
.sb_qflags
= 0;
1067 spin_unlock(&mp
->m_sb_lock
);
1070 * If the fs is readonly, let the incore superblock run
1071 * with quotas off but don't flush the update out to disk
1073 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1077 xfs_fs_cmn_err(CE_NOTE
, mp
, "Writing superblock quota changes");
1080 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1081 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1082 XFS_DEFAULT_LOG_COUNT
);
1084 xfs_trans_cancel(tp
, 0);
1085 xfs_fs_cmn_err(CE_ALERT
, mp
,
1086 "xfs_mount_reset_sbqflags: Superblock update failed!");
1090 xfs_mod_sb(tp
, XFS_SB_QFLAGS
);
1091 return xfs_trans_commit(tp
, 0);
1095 * This function does the following on an initial mount of a file system:
1096 * - reads the superblock from disk and init the mount struct
1097 * - if we're a 32-bit kernel, do a size check on the superblock
1098 * so we don't mount terabyte filesystems
1099 * - init mount struct realtime fields
1100 * - allocate inode hash table for fs
1101 * - init directory manager
1102 * - perform recovery and init the log manager
1108 xfs_sb_t
*sbp
= &(mp
->m_sb
);
1111 uint quotamount
= 0;
1112 uint quotaflags
= 0;
1115 xfs_mount_common(mp
, sbp
);
1118 * Check for a mismatched features2 values. Older kernels
1119 * read & wrote into the wrong sb offset for sb_features2
1120 * on some platforms due to xfs_sb_t not being 64bit size aligned
1121 * when sb_features2 was added, which made older superblock
1122 * reading/writing routines swap it as a 64-bit value.
1124 * For backwards compatibility, we make both slots equal.
1126 * If we detect a mismatched field, we OR the set bits into the
1127 * existing features2 field in case it has already been modified; we
1128 * don't want to lose any features. We then update the bad location
1129 * with the ORed value so that older kernels will see any features2
1130 * flags, and mark the two fields as needing updates once the
1131 * transaction subsystem is online.
1133 if (xfs_sb_has_mismatched_features2(sbp
)) {
1135 "XFS: correcting sb_features alignment problem");
1136 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
1137 sbp
->sb_bad_features2
= sbp
->sb_features2
;
1138 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
1141 * Re-check for ATTR2 in case it was found in bad_features2
1144 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1145 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
1146 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1149 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1150 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1151 xfs_sb_version_removeattr2(&mp
->m_sb
);
1152 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
1154 /* update sb_versionnum for the clearing of the morebits */
1155 if (!sbp
->sb_features2
)
1156 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
1160 * Check if sb_agblocks is aligned at stripe boundary
1161 * If sb_agblocks is NOT aligned turn off m_dalign since
1162 * allocator alignment is within an ag, therefore ag has
1163 * to be aligned at stripe boundary.
1165 error
= xfs_update_alignment(mp
);
1169 xfs_alloc_compute_maxlevels(mp
);
1170 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1171 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1172 xfs_ialloc_compute_maxlevels(mp
);
1174 xfs_set_maxicount(mp
);
1176 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
1178 error
= xfs_uuid_mount(mp
);
1183 * Set the minimum read and write sizes
1185 xfs_set_rw_sizes(mp
);
1188 * Set the inode cluster size.
1189 * This may still be overridden by the file system
1190 * block size if it is larger than the chosen cluster size.
1192 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1195 * Set inode alignment fields
1197 xfs_set_inoalignment(mp
);
1200 * Check that the data (and log if separate) are an ok size.
1202 error
= xfs_check_sizes(mp
);
1204 goto out_remove_uuid
;
1207 * Initialize realtime fields in the mount structure
1209 error
= xfs_rtmount_init(mp
);
1211 cmn_err(CE_WARN
, "XFS: RT mount failed");
1212 goto out_remove_uuid
;
1216 * Copies the low order bits of the timestamp and the randomly
1217 * set "sequence" number out of a UUID.
1219 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1221 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1226 * Initialize the attribute manager's entries.
1228 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1231 * Initialize the precomputed transaction reservations values.
1236 * Allocate and initialize the per-ag data.
1238 spin_lock_init(&mp
->m_perag_lock
);
1239 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_NOFS
);
1240 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
1242 cmn_err(CE_WARN
, "XFS: Failed per-ag init: %d", error
);
1243 goto out_remove_uuid
;
1246 if (!sbp
->sb_logblocks
) {
1247 cmn_err(CE_WARN
, "XFS: no log defined");
1248 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1249 error
= XFS_ERROR(EFSCORRUPTED
);
1250 goto out_free_perag
;
1254 * log's mount-time initialization. Perform 1st part recovery if needed
1256 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1257 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1258 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1260 cmn_err(CE_WARN
, "XFS: log mount failed");
1261 goto out_free_perag
;
1265 * Now the log is mounted, we know if it was an unclean shutdown or
1266 * not. If it was, with the first phase of recovery has completed, we
1267 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1268 * but they are recovered transactionally in the second recovery phase
1271 * Hence we can safely re-initialise incore superblock counters from
1272 * the per-ag data. These may not be correct if the filesystem was not
1273 * cleanly unmounted, so we need to wait for recovery to finish before
1276 * If the filesystem was cleanly unmounted, then we can trust the
1277 * values in the superblock to be correct and we don't need to do
1280 * If we are currently making the filesystem, the initialisation will
1281 * fail as the perag data is in an undefined state.
1283 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1284 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1285 !mp
->m_sb
.sb_inprogress
) {
1286 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1288 goto out_free_perag
;
1292 * Get and sanity-check the root inode.
1293 * Save the pointer to it in the mount structure.
1295 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1297 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1298 goto out_log_dealloc
;
1301 ASSERT(rip
!= NULL
);
1303 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1304 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1305 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1306 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1307 (unsigned long long)rip
->i_ino
);
1308 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1309 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1311 error
= XFS_ERROR(EFSCORRUPTED
);
1314 mp
->m_rootip
= rip
; /* save it */
1316 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1319 * Initialize realtime inode pointers in the mount structure
1321 error
= xfs_rtmount_inodes(mp
);
1324 * Free up the root inode.
1326 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1331 * If this is a read-only mount defer the superblock updates until
1332 * the next remount into writeable mode. Otherwise we would never
1333 * perform the update e.g. for the root filesystem.
1335 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1336 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1338 cmn_err(CE_WARN
, "XFS: failed to write sb changes");
1344 * Initialise the XFS quota management subsystem for this mount
1346 if (XFS_IS_QUOTA_RUNNING(mp
)) {
1347 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
1351 ASSERT(!XFS_IS_QUOTA_ON(mp
));
1354 * If a file system had quotas running earlier, but decided to
1355 * mount without -o uquota/pquota/gquota options, revoke the
1356 * quotachecked license.
1358 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
1360 "XFS: resetting qflags for filesystem %s",
1363 error
= xfs_mount_reset_sbqflags(mp
);
1370 * Finish recovering the file system. This part needed to be
1371 * delayed until after the root and real-time bitmap inodes
1372 * were consistently read in.
1374 error
= xfs_log_mount_finish(mp
);
1376 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1381 * Complete the quota initialisation, post-log-replay component.
1384 ASSERT(mp
->m_qflags
== 0);
1385 mp
->m_qflags
= quotaflags
;
1387 xfs_qm_mount_quotas(mp
);
1390 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1391 if (XFS_IS_QUOTA_ON(mp
))
1392 xfs_fs_cmn_err(CE_NOTE
, mp
, "Disk quotas turned on");
1394 xfs_fs_cmn_err(CE_NOTE
, mp
, "Disk quotas not turned on");
1398 * Now we are mounted, reserve a small amount of unused space for
1399 * privileged transactions. This is needed so that transaction
1400 * space required for critical operations can dip into this pool
1401 * when at ENOSPC. This is needed for operations like create with
1402 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1403 * are not allowed to use this reserved space.
1405 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1406 * This may drive us straight to ENOSPC on mount, but that implies
1407 * we were already there on the last unmount. Warn if this occurs.
1409 resblks
= mp
->m_sb
.sb_dblocks
;
1410 do_div(resblks
, 20);
1411 resblks
= min_t(__uint64_t
, resblks
, 1024);
1412 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1414 cmn_err(CE_WARN
, "XFS: Unable to allocate reserve blocks. "
1415 "Continuing without a reserve pool.");
1420 xfs_rtunmount_inodes(mp
);
1424 xfs_log_unmount(mp
);
1428 xfs_uuid_unmount(mp
);
1434 * This flushes out the inodes,dquots and the superblock, unmounts the
1435 * log and makes sure that incore structures are freed.
1439 struct xfs_mount
*mp
)
1444 xfs_qm_unmount_quotas(mp
);
1445 xfs_rtunmount_inodes(mp
);
1446 IRELE(mp
->m_rootip
);
1449 * We can potentially deadlock here if we have an inode cluster
1450 * that has been freed has its buffer still pinned in memory because
1451 * the transaction is still sitting in a iclog. The stale inodes
1452 * on that buffer will have their flush locks held until the
1453 * transaction hits the disk and the callbacks run. the inode
1454 * flush takes the flush lock unconditionally and with nothing to
1455 * push out the iclog we will never get that unlocked. hence we
1456 * need to force the log first.
1458 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1459 xfs_reclaim_inodes(mp
, XFS_IFLUSH_ASYNC
);
1464 * Flush out the log synchronously so that we know for sure
1465 * that nothing is pinned. This is important because bflush()
1466 * will skip pinned buffers.
1468 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1470 xfs_binval(mp
->m_ddev_targp
);
1471 if (mp
->m_rtdev_targp
) {
1472 xfs_binval(mp
->m_rtdev_targp
);
1476 * Unreserve any blocks we have so that when we unmount we don't account
1477 * the reserved free space as used. This is really only necessary for
1478 * lazy superblock counting because it trusts the incore superblock
1479 * counters to be absolutely correct on clean unmount.
1481 * We don't bother correcting this elsewhere for lazy superblock
1482 * counting because on mount of an unclean filesystem we reconstruct the
1483 * correct counter value and this is irrelevant.
1485 * For non-lazy counter filesystems, this doesn't matter at all because
1486 * we only every apply deltas to the superblock and hence the incore
1487 * value does not matter....
1490 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1492 cmn_err(CE_WARN
, "XFS: Unable to free reserved block pool. "
1493 "Freespace may not be correct on next mount.");
1495 error
= xfs_log_sbcount(mp
, 1);
1497 cmn_err(CE_WARN
, "XFS: Unable to update superblock counters. "
1498 "Freespace may not be correct on next mount.");
1499 xfs_unmountfs_writesb(mp
);
1500 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1501 xfs_log_unmount_write(mp
);
1502 xfs_log_unmount(mp
);
1503 xfs_uuid_unmount(mp
);
1506 xfs_errortag_clearall(mp
, 0);
1512 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1514 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1515 xfs_wait_buftarg(mp
->m_logdev_targp
);
1516 if (mp
->m_rtdev_targp
)
1517 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1518 xfs_wait_buftarg(mp
->m_ddev_targp
);
1522 xfs_fs_writable(xfs_mount_t
*mp
)
1524 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1525 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1531 * Called either periodically to keep the on disk superblock values
1532 * roughly up to date or from unmount to make sure the values are
1533 * correct on a clean unmount.
1535 * Note this code can be called during the process of freezing, so
1536 * we may need to use the transaction allocator which does not not
1537 * block when the transaction subsystem is in its frozen state.
1547 if (!xfs_fs_writable(mp
))
1550 xfs_icsb_sync_counters(mp
, 0);
1553 * we don't need to do this if we are updating the superblock
1554 * counters on every modification.
1556 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1559 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
, KM_SLEEP
);
1560 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1561 XFS_DEFAULT_LOG_COUNT
);
1563 xfs_trans_cancel(tp
, 0);
1567 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1569 xfs_trans_set_sync(tp
);
1570 error
= xfs_trans_commit(tp
, 0);
1575 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1581 * skip superblock write if fs is read-only, or
1582 * if we are doing a forced umount.
1584 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1585 XFS_FORCED_SHUTDOWN(mp
))) {
1587 sbp
= xfs_getsb(mp
, 0);
1589 XFS_BUF_UNDONE(sbp
);
1590 XFS_BUF_UNREAD(sbp
);
1591 XFS_BUF_UNDELAYWRITE(sbp
);
1593 XFS_BUF_UNASYNC(sbp
);
1594 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1595 xfsbdstrat(mp
, sbp
);
1596 error
= xfs_iowait(sbp
);
1598 xfs_ioerror_alert("xfs_unmountfs_writesb",
1599 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1606 * xfs_mod_sb() can be used to copy arbitrary changes to the
1607 * in-core superblock into the superblock buffer to be logged.
1608 * It does not provide the higher level of locking that is
1609 * needed to protect the in-core superblock from concurrent
1613 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1625 bp
= xfs_trans_getsb(tp
, mp
, 0);
1626 first
= sizeof(xfs_sb_t
);
1629 /* translate/copy */
1631 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1633 /* find modified range */
1634 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1635 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1636 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1638 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1639 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1640 first
= xfs_sb_info
[f
].offset
;
1642 xfs_trans_log_buf(tp
, bp
, first
, last
);
1647 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1648 * a delta to a specified field in the in-core superblock. Simply
1649 * switch on the field indicated and apply the delta to that field.
1650 * Fields are not allowed to dip below zero, so if the delta would
1651 * do this do not apply it and return EINVAL.
1653 * The m_sb_lock must be held when this routine is called.
1656 xfs_mod_incore_sb_unlocked(
1658 xfs_sb_field_t field
,
1662 int scounter
; /* short counter for 32 bit fields */
1663 long long lcounter
; /* long counter for 64 bit fields */
1664 long long res_used
, rem
;
1667 * With the in-core superblock spin lock held, switch
1668 * on the indicated field. Apply the delta to the
1669 * proper field. If the fields value would dip below
1670 * 0, then do not apply the delta and return EINVAL.
1673 case XFS_SBS_ICOUNT
:
1674 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1678 return XFS_ERROR(EINVAL
);
1680 mp
->m_sb
.sb_icount
= lcounter
;
1683 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1687 return XFS_ERROR(EINVAL
);
1689 mp
->m_sb
.sb_ifree
= lcounter
;
1691 case XFS_SBS_FDBLOCKS
:
1692 lcounter
= (long long)
1693 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1694 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1696 if (delta
> 0) { /* Putting blocks back */
1697 if (res_used
> delta
) {
1698 mp
->m_resblks_avail
+= delta
;
1700 rem
= delta
- res_used
;
1701 mp
->m_resblks_avail
= mp
->m_resblks
;
1704 } else { /* Taking blocks away */
1709 * If were out of blocks, use any available reserved blocks if
1715 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1717 return XFS_ERROR(ENOSPC
);
1719 mp
->m_resblks_avail
= lcounter
;
1721 } else { /* not reserved */
1722 return XFS_ERROR(ENOSPC
);
1727 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1729 case XFS_SBS_FREXTENTS
:
1730 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1733 return XFS_ERROR(ENOSPC
);
1735 mp
->m_sb
.sb_frextents
= lcounter
;
1737 case XFS_SBS_DBLOCKS
:
1738 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1742 return XFS_ERROR(EINVAL
);
1744 mp
->m_sb
.sb_dblocks
= lcounter
;
1746 case XFS_SBS_AGCOUNT
:
1747 scounter
= mp
->m_sb
.sb_agcount
;
1751 return XFS_ERROR(EINVAL
);
1753 mp
->m_sb
.sb_agcount
= scounter
;
1755 case XFS_SBS_IMAX_PCT
:
1756 scounter
= mp
->m_sb
.sb_imax_pct
;
1760 return XFS_ERROR(EINVAL
);
1762 mp
->m_sb
.sb_imax_pct
= scounter
;
1764 case XFS_SBS_REXTSIZE
:
1765 scounter
= mp
->m_sb
.sb_rextsize
;
1769 return XFS_ERROR(EINVAL
);
1771 mp
->m_sb
.sb_rextsize
= scounter
;
1773 case XFS_SBS_RBMBLOCKS
:
1774 scounter
= mp
->m_sb
.sb_rbmblocks
;
1778 return XFS_ERROR(EINVAL
);
1780 mp
->m_sb
.sb_rbmblocks
= scounter
;
1782 case XFS_SBS_RBLOCKS
:
1783 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1787 return XFS_ERROR(EINVAL
);
1789 mp
->m_sb
.sb_rblocks
= lcounter
;
1791 case XFS_SBS_REXTENTS
:
1792 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1796 return XFS_ERROR(EINVAL
);
1798 mp
->m_sb
.sb_rextents
= lcounter
;
1800 case XFS_SBS_REXTSLOG
:
1801 scounter
= mp
->m_sb
.sb_rextslog
;
1805 return XFS_ERROR(EINVAL
);
1807 mp
->m_sb
.sb_rextslog
= scounter
;
1811 return XFS_ERROR(EINVAL
);
1816 * xfs_mod_incore_sb() is used to change a field in the in-core
1817 * superblock structure by the specified delta. This modification
1818 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1819 * routine to do the work.
1824 xfs_sb_field_t field
,
1830 /* check for per-cpu counters */
1832 #ifdef HAVE_PERCPU_SB
1833 case XFS_SBS_ICOUNT
:
1835 case XFS_SBS_FDBLOCKS
:
1836 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1837 status
= xfs_icsb_modify_counters(mp
, field
,
1844 spin_lock(&mp
->m_sb_lock
);
1845 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1846 spin_unlock(&mp
->m_sb_lock
);
1854 * xfs_mod_incore_sb_batch() is used to change more than one field
1855 * in the in-core superblock structure at a time. This modification
1856 * is protected by a lock internal to this module. The fields and
1857 * changes to those fields are specified in the array of xfs_mod_sb
1858 * structures passed in.
1860 * Either all of the specified deltas will be applied or none of
1861 * them will. If any modified field dips below 0, then all modifications
1862 * will be backed out and EINVAL will be returned.
1865 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1871 * Loop through the array of mod structures and apply each
1872 * individually. If any fail, then back out all those
1873 * which have already been applied. Do all of this within
1874 * the scope of the m_sb_lock so that all of the changes will
1877 spin_lock(&mp
->m_sb_lock
);
1879 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1881 * Apply the delta at index n. If it fails, break
1882 * from the loop so we'll fall into the undo loop
1885 switch (msbp
->msb_field
) {
1886 #ifdef HAVE_PERCPU_SB
1887 case XFS_SBS_ICOUNT
:
1889 case XFS_SBS_FDBLOCKS
:
1890 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1891 spin_unlock(&mp
->m_sb_lock
);
1892 status
= xfs_icsb_modify_counters(mp
,
1894 msbp
->msb_delta
, rsvd
);
1895 spin_lock(&mp
->m_sb_lock
);
1901 status
= xfs_mod_incore_sb_unlocked(mp
,
1903 msbp
->msb_delta
, rsvd
);
1913 * If we didn't complete the loop above, then back out
1914 * any changes made to the superblock. If you add code
1915 * between the loop above and here, make sure that you
1916 * preserve the value of status. Loop back until
1917 * we step below the beginning of the array. Make sure
1918 * we don't touch anything back there.
1922 while (msbp
>= msb
) {
1923 switch (msbp
->msb_field
) {
1924 #ifdef HAVE_PERCPU_SB
1925 case XFS_SBS_ICOUNT
:
1927 case XFS_SBS_FDBLOCKS
:
1928 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1929 spin_unlock(&mp
->m_sb_lock
);
1930 status
= xfs_icsb_modify_counters(mp
,
1934 spin_lock(&mp
->m_sb_lock
);
1940 status
= xfs_mod_incore_sb_unlocked(mp
,
1946 ASSERT(status
== 0);
1950 spin_unlock(&mp
->m_sb_lock
);
1955 * xfs_getsb() is called to obtain the buffer for the superblock.
1956 * The buffer is returned locked and read in from disk.
1957 * The buffer should be released with a call to xfs_brelse().
1959 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1960 * the superblock buffer if it can be locked without sleeping.
1961 * If it can't then we'll return NULL.
1970 ASSERT(mp
->m_sb_bp
!= NULL
);
1972 if (flags
& XBF_TRYLOCK
) {
1973 if (!XFS_BUF_CPSEMA(bp
)) {
1977 XFS_BUF_PSEMA(bp
, PRIBIO
);
1980 ASSERT(XFS_BUF_ISDONE(bp
));
1985 * Used to free the superblock along various error paths.
1994 * Use xfs_getsb() so that the buffer will be locked
1995 * when we call xfs_buf_relse().
1997 bp
= xfs_getsb(mp
, 0);
1998 XFS_BUF_UNMANAGE(bp
);
2004 * Used to log changes to the superblock unit and width fields which could
2005 * be altered by the mount options, as well as any potential sb_features2
2006 * fixup. Only the first superblock is updated.
2016 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
2017 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
2018 XFS_SB_VERSIONNUM
));
2020 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
2021 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
2022 XFS_DEFAULT_LOG_COUNT
);
2024 xfs_trans_cancel(tp
, 0);
2027 xfs_mod_sb(tp
, fields
);
2028 error
= xfs_trans_commit(tp
, 0);
2033 #ifdef HAVE_PERCPU_SB
2035 * Per-cpu incore superblock counters
2037 * Simple concept, difficult implementation
2039 * Basically, replace the incore superblock counters with a distributed per cpu
2040 * counter for contended fields (e.g. free block count).
2042 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2043 * hence needs to be accurately read when we are running low on space. Hence
2044 * there is a method to enable and disable the per-cpu counters based on how
2045 * much "stuff" is available in them.
2047 * Basically, a counter is enabled if there is enough free resource to justify
2048 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2049 * ENOSPC), then we disable the counters to synchronise all callers and
2050 * re-distribute the available resources.
2052 * If, once we redistributed the available resources, we still get a failure,
2053 * we disable the per-cpu counter and go through the slow path.
2055 * The slow path is the current xfs_mod_incore_sb() function. This means that
2056 * when we disable a per-cpu counter, we need to drain its resources back to
2057 * the global superblock. We do this after disabling the counter to prevent
2058 * more threads from queueing up on the counter.
2060 * Essentially, this means that we still need a lock in the fast path to enable
2061 * synchronisation between the global counters and the per-cpu counters. This
2062 * is not a problem because the lock will be local to a CPU almost all the time
2063 * and have little contention except when we get to ENOSPC conditions.
2065 * Basically, this lock becomes a barrier that enables us to lock out the fast
2066 * path while we do things like enabling and disabling counters and
2067 * synchronising the counters.
2071 * 1. m_sb_lock before picking up per-cpu locks
2072 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2073 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2074 * 4. modifying per-cpu counters requires holding per-cpu lock
2075 * 5. modifying global counters requires holding m_sb_lock
2076 * 6. enabling or disabling a counter requires holding the m_sb_lock
2077 * and _none_ of the per-cpu locks.
2079 * Disabled counters are only ever re-enabled by a balance operation
2080 * that results in more free resources per CPU than a given threshold.
2081 * To ensure counters don't remain disabled, they are rebalanced when
2082 * the global resource goes above a higher threshold (i.e. some hysteresis
2083 * is present to prevent thrashing).
2086 #ifdef CONFIG_HOTPLUG_CPU
2088 * hot-plug CPU notifier support.
2090 * We need a notifier per filesystem as we need to be able to identify
2091 * the filesystem to balance the counters out. This is achieved by
2092 * having a notifier block embedded in the xfs_mount_t and doing pointer
2093 * magic to get the mount pointer from the notifier block address.
2096 xfs_icsb_cpu_notify(
2097 struct notifier_block
*nfb
,
2098 unsigned long action
,
2101 xfs_icsb_cnts_t
*cntp
;
2104 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2105 cntp
= (xfs_icsb_cnts_t
*)
2106 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2108 case CPU_UP_PREPARE
:
2109 case CPU_UP_PREPARE_FROZEN
:
2110 /* Easy Case - initialize the area and locks, and
2111 * then rebalance when online does everything else for us. */
2112 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2115 case CPU_ONLINE_FROZEN
:
2117 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2118 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2119 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2120 xfs_icsb_unlock(mp
);
2123 case CPU_DEAD_FROZEN
:
2124 /* Disable all the counters, then fold the dead cpu's
2125 * count into the total on the global superblock and
2126 * re-enable the counters. */
2128 spin_lock(&mp
->m_sb_lock
);
2129 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2130 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2131 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2133 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2134 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2135 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2137 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2139 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
2140 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
2141 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
2142 spin_unlock(&mp
->m_sb_lock
);
2143 xfs_icsb_unlock(mp
);
2149 #endif /* CONFIG_HOTPLUG_CPU */
2152 xfs_icsb_init_counters(
2155 xfs_icsb_cnts_t
*cntp
;
2158 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2159 if (mp
->m_sb_cnts
== NULL
)
2162 #ifdef CONFIG_HOTPLUG_CPU
2163 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2164 mp
->m_icsb_notifier
.priority
= 0;
2165 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2166 #endif /* CONFIG_HOTPLUG_CPU */
2168 for_each_online_cpu(i
) {
2169 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2170 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2173 mutex_init(&mp
->m_icsb_mutex
);
2176 * start with all counters disabled so that the
2177 * initial balance kicks us off correctly
2179 mp
->m_icsb_counters
= -1;
2184 xfs_icsb_reinit_counters(
2189 * start with all counters disabled so that the
2190 * initial balance kicks us off correctly
2192 mp
->m_icsb_counters
= -1;
2193 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2194 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2195 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2196 xfs_icsb_unlock(mp
);
2200 xfs_icsb_destroy_counters(
2203 if (mp
->m_sb_cnts
) {
2204 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2205 free_percpu(mp
->m_sb_cnts
);
2207 mutex_destroy(&mp
->m_icsb_mutex
);
2212 xfs_icsb_cnts_t
*icsbp
)
2214 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2220 xfs_icsb_unlock_cntr(
2221 xfs_icsb_cnts_t
*icsbp
)
2223 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2228 xfs_icsb_lock_all_counters(
2231 xfs_icsb_cnts_t
*cntp
;
2234 for_each_online_cpu(i
) {
2235 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2236 xfs_icsb_lock_cntr(cntp
);
2241 xfs_icsb_unlock_all_counters(
2244 xfs_icsb_cnts_t
*cntp
;
2247 for_each_online_cpu(i
) {
2248 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2249 xfs_icsb_unlock_cntr(cntp
);
2256 xfs_icsb_cnts_t
*cnt
,
2259 xfs_icsb_cnts_t
*cntp
;
2262 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2264 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2265 xfs_icsb_lock_all_counters(mp
);
2267 for_each_online_cpu(i
) {
2268 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2269 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2270 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2271 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2274 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2275 xfs_icsb_unlock_all_counters(mp
);
2279 xfs_icsb_counter_disabled(
2281 xfs_sb_field_t field
)
2283 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2284 return test_bit(field
, &mp
->m_icsb_counters
);
2288 xfs_icsb_disable_counter(
2290 xfs_sb_field_t field
)
2292 xfs_icsb_cnts_t cnt
;
2294 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2297 * If we are already disabled, then there is nothing to do
2298 * here. We check before locking all the counters to avoid
2299 * the expensive lock operation when being called in the
2300 * slow path and the counter is already disabled. This is
2301 * safe because the only time we set or clear this state is under
2304 if (xfs_icsb_counter_disabled(mp
, field
))
2307 xfs_icsb_lock_all_counters(mp
);
2308 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2309 /* drain back to superblock */
2311 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2313 case XFS_SBS_ICOUNT
:
2314 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2317 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2319 case XFS_SBS_FDBLOCKS
:
2320 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2327 xfs_icsb_unlock_all_counters(mp
);
2331 xfs_icsb_enable_counter(
2333 xfs_sb_field_t field
,
2337 xfs_icsb_cnts_t
*cntp
;
2340 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2342 xfs_icsb_lock_all_counters(mp
);
2343 for_each_online_cpu(i
) {
2344 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2346 case XFS_SBS_ICOUNT
:
2347 cntp
->icsb_icount
= count
+ resid
;
2350 cntp
->icsb_ifree
= count
+ resid
;
2352 case XFS_SBS_FDBLOCKS
:
2353 cntp
->icsb_fdblocks
= count
+ resid
;
2361 clear_bit(field
, &mp
->m_icsb_counters
);
2362 xfs_icsb_unlock_all_counters(mp
);
2366 xfs_icsb_sync_counters_locked(
2370 xfs_icsb_cnts_t cnt
;
2372 xfs_icsb_count(mp
, &cnt
, flags
);
2374 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2375 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2376 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2377 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2378 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2379 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2383 * Accurate update of per-cpu counters to incore superblock
2386 xfs_icsb_sync_counters(
2390 spin_lock(&mp
->m_sb_lock
);
2391 xfs_icsb_sync_counters_locked(mp
, flags
);
2392 spin_unlock(&mp
->m_sb_lock
);
2396 * Balance and enable/disable counters as necessary.
2398 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2399 * chosen to be the same number as single on disk allocation chunk per CPU, and
2400 * free blocks is something far enough zero that we aren't going thrash when we
2401 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2402 * prevent looping endlessly when xfs_alloc_space asks for more than will
2403 * be distributed to a single CPU but each CPU has enough blocks to be
2406 * Note that we can be called when counters are already disabled.
2407 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2408 * prevent locking every per-cpu counter needlessly.
2411 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2412 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2413 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2415 xfs_icsb_balance_counter_locked(
2417 xfs_sb_field_t field
,
2420 uint64_t count
, resid
;
2421 int weight
= num_online_cpus();
2422 uint64_t min
= (uint64_t)min_per_cpu
;
2424 /* disable counter and sync counter */
2425 xfs_icsb_disable_counter(mp
, field
);
2427 /* update counters - first CPU gets residual*/
2429 case XFS_SBS_ICOUNT
:
2430 count
= mp
->m_sb
.sb_icount
;
2431 resid
= do_div(count
, weight
);
2432 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2436 count
= mp
->m_sb
.sb_ifree
;
2437 resid
= do_div(count
, weight
);
2438 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2441 case XFS_SBS_FDBLOCKS
:
2442 count
= mp
->m_sb
.sb_fdblocks
;
2443 resid
= do_div(count
, weight
);
2444 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2449 count
= resid
= 0; /* quiet, gcc */
2453 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2457 xfs_icsb_balance_counter(
2459 xfs_sb_field_t fields
,
2462 spin_lock(&mp
->m_sb_lock
);
2463 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2464 spin_unlock(&mp
->m_sb_lock
);
2468 xfs_icsb_modify_counters(
2470 xfs_sb_field_t field
,
2474 xfs_icsb_cnts_t
*icsbp
;
2475 long long lcounter
; /* long counter for 64 bit fields */
2481 icsbp
= this_cpu_ptr(mp
->m_sb_cnts
);
2484 * if the counter is disabled, go to slow path
2486 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2488 xfs_icsb_lock_cntr(icsbp
);
2489 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2490 xfs_icsb_unlock_cntr(icsbp
);
2495 case XFS_SBS_ICOUNT
:
2496 lcounter
= icsbp
->icsb_icount
;
2498 if (unlikely(lcounter
< 0))
2499 goto balance_counter
;
2500 icsbp
->icsb_icount
= lcounter
;
2504 lcounter
= icsbp
->icsb_ifree
;
2506 if (unlikely(lcounter
< 0))
2507 goto balance_counter
;
2508 icsbp
->icsb_ifree
= lcounter
;
2511 case XFS_SBS_FDBLOCKS
:
2512 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2514 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2516 if (unlikely(lcounter
< 0))
2517 goto balance_counter
;
2518 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2524 xfs_icsb_unlock_cntr(icsbp
);
2532 * serialise with a mutex so we don't burn lots of cpu on
2533 * the superblock lock. We still need to hold the superblock
2534 * lock, however, when we modify the global structures.
2539 * Now running atomically.
2541 * If the counter is enabled, someone has beaten us to rebalancing.
2542 * Drop the lock and try again in the fast path....
2544 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2545 xfs_icsb_unlock(mp
);
2550 * The counter is currently disabled. Because we are
2551 * running atomically here, we know a rebalance cannot
2552 * be in progress. Hence we can go straight to operating
2553 * on the global superblock. We do not call xfs_mod_incore_sb()
2554 * here even though we need to get the m_sb_lock. Doing so
2555 * will cause us to re-enter this function and deadlock.
2556 * Hence we get the m_sb_lock ourselves and then call
2557 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2558 * directly on the global counters.
2560 spin_lock(&mp
->m_sb_lock
);
2561 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2562 spin_unlock(&mp
->m_sb_lock
);
2565 * Now that we've modified the global superblock, we
2566 * may be able to re-enable the distributed counters
2567 * (e.g. lots of space just got freed). After that
2571 xfs_icsb_balance_counter(mp
, field
, 0);
2572 xfs_icsb_unlock(mp
);
2576 xfs_icsb_unlock_cntr(icsbp
);
2580 * We may have multiple threads here if multiple per-cpu
2581 * counters run dry at the same time. This will mean we can
2582 * do more balances than strictly necessary but it is not
2583 * the common slowpath case.
2588 * running atomically.
2590 * This will leave the counter in the correct state for future
2591 * accesses. After the rebalance, we simply try again and our retry
2592 * will either succeed through the fast path or slow path without
2593 * another balance operation being required.
2595 xfs_icsb_balance_counter(mp
, field
, delta
);
2596 xfs_icsb_unlock(mp
);