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"
48 STATIC
void xfs_unmountfs_wait(xfs_mount_t
*);
52 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
54 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
56 STATIC
int xfs_icsb_modify_counters(xfs_mount_t
*, xfs_sb_field_t
,
58 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
62 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
63 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
64 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
70 short type
; /* 0 = integer
71 * 1 = binary / string (no translation)
74 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
75 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
76 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
77 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
78 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
79 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
80 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
81 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
82 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
83 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
84 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
85 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
86 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
87 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
88 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
89 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
90 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
91 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
92 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
93 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
94 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
95 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
96 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
97 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
98 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
99 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
100 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
101 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
102 { offsetof(xfs_sb_t
, sb_icount
), 0 },
103 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
104 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
105 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
106 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
107 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
108 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
109 { offsetof(xfs_sb_t
, sb_flags
), 0 },
110 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
111 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
112 { offsetof(xfs_sb_t
, sb_unit
), 0 },
113 { offsetof(xfs_sb_t
, sb_width
), 0 },
114 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
115 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
116 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
117 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
118 { offsetof(xfs_sb_t
, sb_features2
), 0 },
119 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
120 { sizeof(xfs_sb_t
), 0 }
123 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
124 static int xfs_uuid_table_size
;
125 static uuid_t
*xfs_uuid_table
;
128 * See if the UUID is unique among mounted XFS filesystems.
129 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
133 struct xfs_mount
*mp
)
135 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
138 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
141 if (uuid_is_nil(uuid
)) {
143 "XFS: Filesystem %s has nil UUID - can't mount",
145 return XFS_ERROR(EINVAL
);
148 mutex_lock(&xfs_uuid_table_mutex
);
149 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
150 if (uuid_is_nil(&xfs_uuid_table
[i
])) {
154 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
159 xfs_uuid_table
= kmem_realloc(xfs_uuid_table
,
160 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
161 xfs_uuid_table_size
* sizeof(*xfs_uuid_table
),
163 hole
= xfs_uuid_table_size
++;
165 xfs_uuid_table
[hole
] = *uuid
;
166 mutex_unlock(&xfs_uuid_table_mutex
);
171 mutex_unlock(&xfs_uuid_table_mutex
);
172 cmn_err(CE_WARN
, "XFS: Filesystem %s has duplicate UUID - can't mount",
174 return XFS_ERROR(EINVAL
);
179 struct xfs_mount
*mp
)
181 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
184 if (mp
->m_flags
& XFS_MOUNT_NOUUID
)
187 mutex_lock(&xfs_uuid_table_mutex
);
188 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
189 if (uuid_is_nil(&xfs_uuid_table
[i
]))
191 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
193 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
196 ASSERT(i
< xfs_uuid_table_size
);
197 mutex_unlock(&xfs_uuid_table_mutex
);
202 * Free up the resources associated with a mount structure. Assume that
203 * the structure was initially zeroed, so we can tell which fields got
213 for (agno
= 0; agno
< mp
->m_maxagi
; agno
++)
214 if (mp
->m_perag
[agno
].pagb_list
)
215 kmem_free(mp
->m_perag
[agno
].pagb_list
);
216 kmem_free(mp
->m_perag
);
221 * Check size of device based on the (data/realtime) block count.
222 * Note: this check is used by the growfs code as well as mount.
225 xfs_sb_validate_fsb_count(
229 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
230 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
232 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
233 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
235 #else /* Limited by UINT_MAX of sectors */
236 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
243 * Check the validity of the SB found.
246 xfs_mount_validate_sb(
252 * If the log device and data device have the
253 * same device number, the log is internal.
254 * Consequently, the sb_logstart should be non-zero. If
255 * we have a zero sb_logstart in this case, we may be trying to mount
256 * a volume filesystem in a non-volume manner.
258 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
259 xfs_fs_mount_cmn_err(flags
, "bad magic number");
260 return XFS_ERROR(EWRONGFS
);
263 if (!xfs_sb_good_version(sbp
)) {
264 xfs_fs_mount_cmn_err(flags
, "bad version");
265 return XFS_ERROR(EWRONGFS
);
269 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
270 xfs_fs_mount_cmn_err(flags
,
271 "filesystem is marked as having an external log; "
272 "specify logdev on the\nmount command line.");
273 return XFS_ERROR(EINVAL
);
277 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
278 xfs_fs_mount_cmn_err(flags
,
279 "filesystem is marked as having an internal log; "
280 "do not specify logdev on\nthe mount command line.");
281 return XFS_ERROR(EINVAL
);
285 * More sanity checking. These were stolen directly from
289 sbp
->sb_agcount
<= 0 ||
290 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
291 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
292 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
293 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
294 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
295 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
296 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
297 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
298 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
299 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
300 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
301 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
302 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
303 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
304 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
305 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
306 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
307 return XFS_ERROR(EFSCORRUPTED
);
311 * Sanity check AG count, size fields against data size field
314 sbp
->sb_dblocks
== 0 ||
316 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
317 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
318 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
319 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
320 return XFS_ERROR(EFSCORRUPTED
);
324 * Until this is fixed only page-sized or smaller data blocks work.
326 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
327 xfs_fs_mount_cmn_err(flags
,
328 "file system with blocksize %d bytes",
330 xfs_fs_mount_cmn_err(flags
,
331 "only pagesize (%ld) or less will currently work.",
333 return XFS_ERROR(ENOSYS
);
337 * Currently only very few inode sizes are supported.
339 switch (sbp
->sb_inodesize
) {
346 xfs_fs_mount_cmn_err(flags
,
347 "inode size of %d bytes not supported",
349 return XFS_ERROR(ENOSYS
);
352 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
353 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
354 xfs_fs_mount_cmn_err(flags
,
355 "file system too large to be mounted on this system.");
356 return XFS_ERROR(E2BIG
);
359 if (unlikely(sbp
->sb_inprogress
)) {
360 xfs_fs_mount_cmn_err(flags
, "file system busy");
361 return XFS_ERROR(EFSCORRUPTED
);
365 * Version 1 directory format has never worked on Linux.
367 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
368 xfs_fs_mount_cmn_err(flags
,
369 "file system using version 1 directory format");
370 return XFS_ERROR(ENOSYS
);
377 xfs_initialize_perag_icache(
380 if (!pag
->pag_ici_init
) {
381 rwlock_init(&pag
->pag_ici_lock
);
382 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
383 pag
->pag_ici_init
= 1;
388 xfs_initialize_perag(
390 xfs_agnumber_t agcount
)
392 xfs_agnumber_t index
, max_metadata
;
396 xfs_sb_t
*sbp
= &mp
->m_sb
;
397 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
399 /* Check to see if the filesystem can overflow 32 bit inodes */
400 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
401 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
403 /* Clear the mount flag if no inode can overflow 32 bits
404 * on this filesystem, or if specifically requested..
406 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> max_inum
) {
407 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
409 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
412 /* If we can overflow then setup the ag headers accordingly */
413 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
414 /* Calculate how much should be reserved for inodes to
415 * meet the max inode percentage.
417 if (mp
->m_maxicount
) {
420 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
422 icount
+= sbp
->sb_agblocks
- 1;
423 do_div(icount
, sbp
->sb_agblocks
);
424 max_metadata
= icount
;
426 max_metadata
= agcount
;
428 for (index
= 0; index
< agcount
; index
++) {
429 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
430 if (ino
> max_inum
) {
435 /* This ag is preferred for inodes */
436 pag
= &mp
->m_perag
[index
];
437 pag
->pagi_inodeok
= 1;
438 if (index
< max_metadata
)
439 pag
->pagf_metadata
= 1;
440 xfs_initialize_perag_icache(pag
);
443 /* Setup default behavior for smaller filesystems */
444 for (index
= 0; index
< agcount
; index
++) {
445 pag
= &mp
->m_perag
[index
];
446 pag
->pagi_inodeok
= 1;
447 xfs_initialize_perag_icache(pag
);
458 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
459 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
460 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
461 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
462 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
463 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
464 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
465 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
466 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
467 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
468 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
469 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
470 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
471 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
472 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
473 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
474 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
475 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
476 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
477 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
478 to
->sb_blocklog
= from
->sb_blocklog
;
479 to
->sb_sectlog
= from
->sb_sectlog
;
480 to
->sb_inodelog
= from
->sb_inodelog
;
481 to
->sb_inopblog
= from
->sb_inopblog
;
482 to
->sb_agblklog
= from
->sb_agblklog
;
483 to
->sb_rextslog
= from
->sb_rextslog
;
484 to
->sb_inprogress
= from
->sb_inprogress
;
485 to
->sb_imax_pct
= from
->sb_imax_pct
;
486 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
487 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
488 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
489 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
490 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
491 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
492 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
493 to
->sb_flags
= from
->sb_flags
;
494 to
->sb_shared_vn
= from
->sb_shared_vn
;
495 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
496 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
497 to
->sb_width
= be32_to_cpu(from
->sb_width
);
498 to
->sb_dirblklog
= from
->sb_dirblklog
;
499 to
->sb_logsectlog
= from
->sb_logsectlog
;
500 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
501 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
502 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
503 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
507 * Copy in core superblock to ondisk one.
509 * The fields argument is mask of superblock fields to copy.
517 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
518 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
528 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
529 first
= xfs_sb_info
[f
].offset
;
530 size
= xfs_sb_info
[f
+ 1].offset
- first
;
532 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
534 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
535 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
539 *(__be16
*)(to_ptr
+ first
) =
540 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
543 *(__be32
*)(to_ptr
+ first
) =
544 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
547 *(__be64
*)(to_ptr
+ first
) =
548 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
555 fields
&= ~(1LL << f
);
562 * Does the initial read of the superblock.
565 xfs_readsb(xfs_mount_t
*mp
, int flags
)
567 unsigned int sector_size
;
568 unsigned int extra_flags
;
572 ASSERT(mp
->m_sb_bp
== NULL
);
573 ASSERT(mp
->m_ddev_targp
!= NULL
);
576 * Allocate a (locked) buffer to hold the superblock.
577 * This will be kept around at all times to optimize
578 * access to the superblock.
580 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
581 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
583 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
584 BTOBB(sector_size
), extra_flags
);
585 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
586 xfs_fs_mount_cmn_err(flags
, "SB read failed");
587 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
590 ASSERT(XFS_BUF_ISBUSY(bp
));
591 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
594 * Initialize the mount structure from the superblock.
595 * But first do some basic consistency checking.
597 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
599 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
601 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
606 * We must be able to do sector-sized and sector-aligned IO.
608 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
609 xfs_fs_mount_cmn_err(flags
,
610 "device supports only %u byte sectors (not %u)",
611 sector_size
, mp
->m_sb
.sb_sectsize
);
617 * If device sector size is smaller than the superblock size,
618 * re-read the superblock so the buffer is correctly sized.
620 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
621 XFS_BUF_UNMANAGE(bp
);
623 sector_size
= mp
->m_sb
.sb_sectsize
;
624 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
625 BTOBB(sector_size
), extra_flags
);
626 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
627 xfs_fs_mount_cmn_err(flags
, "SB re-read failed");
628 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
631 ASSERT(XFS_BUF_ISBUSY(bp
));
632 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
635 /* Initialize per-cpu counters */
636 xfs_icsb_reinit_counters(mp
);
640 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
645 XFS_BUF_UNMANAGE(bp
);
655 * Mount initialization code establishing various mount
656 * fields from the superblock associated with the given
660 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
662 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
663 spin_lock_init(&mp
->m_agirotor_lock
);
664 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
665 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
666 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
667 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
668 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
669 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
670 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
671 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
672 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
674 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
675 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
676 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
677 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
679 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
680 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
681 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
682 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
684 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
685 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
686 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
687 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
689 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
690 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
692 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
696 * xfs_initialize_perag_data
698 * Read in each per-ag structure so we can count up the number of
699 * allocated inodes, free inodes and used filesystem blocks as this
700 * information is no longer persistent in the superblock. Once we have
701 * this information, write it into the in-core superblock structure.
704 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
706 xfs_agnumber_t index
;
708 xfs_sb_t
*sbp
= &mp
->m_sb
;
712 uint64_t bfreelst
= 0;
716 for (index
= 0; index
< agcount
; index
++) {
718 * read the agf, then the agi. This gets us
719 * all the information we need and populates the
720 * per-ag structures for us.
722 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
726 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
729 pag
= &mp
->m_perag
[index
];
730 ifree
+= pag
->pagi_freecount
;
731 ialloc
+= pag
->pagi_count
;
732 bfree
+= pag
->pagf_freeblks
;
733 bfreelst
+= pag
->pagf_flcount
;
734 btree
+= pag
->pagf_btreeblks
;
737 * Overwrite incore superblock counters with just-read data
739 spin_lock(&mp
->m_sb_lock
);
740 sbp
->sb_ifree
= ifree
;
741 sbp
->sb_icount
= ialloc
;
742 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
743 spin_unlock(&mp
->m_sb_lock
);
745 /* Fixup the per-cpu counters as well. */
746 xfs_icsb_reinit_counters(mp
);
752 * Update alignment values based on mount options and sb values
755 xfs_update_alignment(xfs_mount_t
*mp
)
757 xfs_sb_t
*sbp
= &(mp
->m_sb
);
761 * If stripe unit and stripe width are not multiples
762 * of the fs blocksize turn off alignment.
764 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
765 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
766 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
768 "XFS: alignment check 1 failed");
769 return XFS_ERROR(EINVAL
);
771 mp
->m_dalign
= mp
->m_swidth
= 0;
774 * Convert the stripe unit and width to FSBs.
776 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
777 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
778 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
779 return XFS_ERROR(EINVAL
);
781 xfs_fs_cmn_err(CE_WARN
, mp
,
782 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
783 mp
->m_dalign
, mp
->m_swidth
,
788 } else if (mp
->m_dalign
) {
789 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
791 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
792 xfs_fs_cmn_err(CE_WARN
, mp
,
793 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
796 return XFS_ERROR(EINVAL
);
803 * Update superblock with new values
806 if (xfs_sb_version_hasdalign(sbp
)) {
807 if (sbp
->sb_unit
!= mp
->m_dalign
) {
808 sbp
->sb_unit
= mp
->m_dalign
;
809 mp
->m_update_flags
|= XFS_SB_UNIT
;
811 if (sbp
->sb_width
!= mp
->m_swidth
) {
812 sbp
->sb_width
= mp
->m_swidth
;
813 mp
->m_update_flags
|= XFS_SB_WIDTH
;
816 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
817 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
818 mp
->m_dalign
= sbp
->sb_unit
;
819 mp
->m_swidth
= sbp
->sb_width
;
826 * Set the maximum inode count for this filesystem
829 xfs_set_maxicount(xfs_mount_t
*mp
)
831 xfs_sb_t
*sbp
= &(mp
->m_sb
);
834 if (sbp
->sb_imax_pct
) {
836 * Make sure the maximum inode count is a multiple
837 * of the units we allocate inodes in.
839 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
841 do_div(icount
, mp
->m_ialloc_blks
);
842 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
850 * Set the default minimum read and write sizes unless
851 * already specified in a mount option.
852 * We use smaller I/O sizes when the file system
853 * is being used for NFS service (wsync mount option).
856 xfs_set_rw_sizes(xfs_mount_t
*mp
)
858 xfs_sb_t
*sbp
= &(mp
->m_sb
);
859 int readio_log
, writeio_log
;
861 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
862 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
863 readio_log
= XFS_WSYNC_READIO_LOG
;
864 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
866 readio_log
= XFS_READIO_LOG_LARGE
;
867 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
870 readio_log
= mp
->m_readio_log
;
871 writeio_log
= mp
->m_writeio_log
;
874 if (sbp
->sb_blocklog
> readio_log
) {
875 mp
->m_readio_log
= sbp
->sb_blocklog
;
877 mp
->m_readio_log
= readio_log
;
879 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
880 if (sbp
->sb_blocklog
> writeio_log
) {
881 mp
->m_writeio_log
= sbp
->sb_blocklog
;
883 mp
->m_writeio_log
= writeio_log
;
885 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
889 * Set whether we're using inode alignment.
892 xfs_set_inoalignment(xfs_mount_t
*mp
)
894 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
895 mp
->m_sb
.sb_inoalignmt
>=
896 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
897 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
899 mp
->m_inoalign_mask
= 0;
901 * If we are using stripe alignment, check whether
902 * the stripe unit is a multiple of the inode alignment
904 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
905 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
906 mp
->m_sinoalign
= mp
->m_dalign
;
912 * Check that the data (and log if separate) are an ok size.
915 xfs_check_sizes(xfs_mount_t
*mp
)
921 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
922 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
923 cmn_err(CE_WARN
, "XFS: size check 1 failed");
924 return XFS_ERROR(E2BIG
);
926 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
927 d
- XFS_FSS_TO_BB(mp
, 1),
928 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
932 cmn_err(CE_WARN
, "XFS: size check 2 failed");
934 error
= XFS_ERROR(E2BIG
);
938 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
939 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
940 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
941 cmn_err(CE_WARN
, "XFS: size check 3 failed");
942 return XFS_ERROR(E2BIG
);
944 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
945 d
- XFS_FSB_TO_BB(mp
, 1),
946 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
950 cmn_err(CE_WARN
, "XFS: size check 3 failed");
952 error
= XFS_ERROR(E2BIG
);
960 * This function does the following on an initial mount of a file system:
961 * - reads the superblock from disk and init the mount struct
962 * - if we're a 32-bit kernel, do a size check on the superblock
963 * so we don't mount terabyte filesystems
964 * - init mount struct realtime fields
965 * - allocate inode hash table for fs
966 * - init directory manager
967 * - perform recovery and init the log manager
973 xfs_sb_t
*sbp
= &(mp
->m_sb
);
976 uint quotamount
, quotaflags
;
979 xfs_mount_common(mp
, sbp
);
982 * Check for a mismatched features2 values. Older kernels
983 * read & wrote into the wrong sb offset for sb_features2
984 * on some platforms due to xfs_sb_t not being 64bit size aligned
985 * when sb_features2 was added, which made older superblock
986 * reading/writing routines swap it as a 64-bit value.
988 * For backwards compatibility, we make both slots equal.
990 * If we detect a mismatched field, we OR the set bits into the
991 * existing features2 field in case it has already been modified; we
992 * don't want to lose any features. We then update the bad location
993 * with the ORed value so that older kernels will see any features2
994 * flags, and mark the two fields as needing updates once the
995 * transaction subsystem is online.
997 if (xfs_sb_has_mismatched_features2(sbp
)) {
999 "XFS: correcting sb_features alignment problem");
1000 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
1001 sbp
->sb_bad_features2
= sbp
->sb_features2
;
1002 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
1005 * Re-check for ATTR2 in case it was found in bad_features2
1008 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1009 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
1010 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1013 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
1014 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1015 xfs_sb_version_removeattr2(&mp
->m_sb
);
1016 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
1018 /* update sb_versionnum for the clearing of the morebits */
1019 if (!sbp
->sb_features2
)
1020 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
1024 * Check if sb_agblocks is aligned at stripe boundary
1025 * If sb_agblocks is NOT aligned turn off m_dalign since
1026 * allocator alignment is within an ag, therefore ag has
1027 * to be aligned at stripe boundary.
1029 error
= xfs_update_alignment(mp
);
1033 xfs_alloc_compute_maxlevels(mp
);
1034 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1035 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1036 xfs_ialloc_compute_maxlevels(mp
);
1038 xfs_set_maxicount(mp
);
1040 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
1042 error
= xfs_uuid_mount(mp
);
1047 * Set the minimum read and write sizes
1049 xfs_set_rw_sizes(mp
);
1052 * Set the inode cluster size.
1053 * This may still be overridden by the file system
1054 * block size if it is larger than the chosen cluster size.
1056 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1059 * Set inode alignment fields
1061 xfs_set_inoalignment(mp
);
1064 * Check that the data (and log if separate) are an ok size.
1066 error
= xfs_check_sizes(mp
);
1068 goto out_remove_uuid
;
1071 * Initialize realtime fields in the mount structure
1073 error
= xfs_rtmount_init(mp
);
1075 cmn_err(CE_WARN
, "XFS: RT mount failed");
1076 goto out_remove_uuid
;
1080 * Copies the low order bits of the timestamp and the randomly
1081 * set "sequence" number out of a UUID.
1083 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1085 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1090 * Initialize the attribute manager's entries.
1092 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1095 * Initialize the precomputed transaction reservations values.
1100 * Allocate and initialize the per-ag data.
1102 init_rwsem(&mp
->m_peraglock
);
1103 mp
->m_perag
= kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
),
1106 goto out_remove_uuid
;
1108 mp
->m_maxagi
= xfs_initialize_perag(mp
, sbp
->sb_agcount
);
1110 if (!sbp
->sb_logblocks
) {
1111 cmn_err(CE_WARN
, "XFS: no log defined");
1112 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1113 error
= XFS_ERROR(EFSCORRUPTED
);
1114 goto out_free_perag
;
1118 * log's mount-time initialization. Perform 1st part recovery if needed
1120 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1121 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1122 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1124 cmn_err(CE_WARN
, "XFS: log mount failed");
1125 goto out_free_perag
;
1129 * Now the log is mounted, we know if it was an unclean shutdown or
1130 * not. If it was, with the first phase of recovery has completed, we
1131 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1132 * but they are recovered transactionally in the second recovery phase
1135 * Hence we can safely re-initialise incore superblock counters from
1136 * the per-ag data. These may not be correct if the filesystem was not
1137 * cleanly unmounted, so we need to wait for recovery to finish before
1140 * If the filesystem was cleanly unmounted, then we can trust the
1141 * values in the superblock to be correct and we don't need to do
1144 * If we are currently making the filesystem, the initialisation will
1145 * fail as the perag data is in an undefined state.
1147 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1148 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1149 !mp
->m_sb
.sb_inprogress
) {
1150 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1152 goto out_free_perag
;
1156 * Get and sanity-check the root inode.
1157 * Save the pointer to it in the mount structure.
1159 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1161 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1162 goto out_log_dealloc
;
1165 ASSERT(rip
!= NULL
);
1167 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1168 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1169 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1170 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1171 (unsigned long long)rip
->i_ino
);
1172 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1173 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1175 error
= XFS_ERROR(EFSCORRUPTED
);
1178 mp
->m_rootip
= rip
; /* save it */
1180 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1183 * Initialize realtime inode pointers in the mount structure
1185 error
= xfs_rtmount_inodes(mp
);
1188 * Free up the root inode.
1190 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1195 * If this is a read-only mount defer the superblock updates until
1196 * the next remount into writeable mode. Otherwise we would never
1197 * perform the update e.g. for the root filesystem.
1199 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1200 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1202 cmn_err(CE_WARN
, "XFS: failed to write sb changes");
1208 * Initialise the XFS quota management subsystem for this mount
1210 error
= XFS_QM_INIT(mp
, "amount
, "aflags
);
1215 * Finish recovering the file system. This part needed to be
1216 * delayed until after the root and real-time bitmap inodes
1217 * were consistently read in.
1219 error
= xfs_log_mount_finish(mp
);
1221 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1226 * Complete the quota initialisation, post-log-replay component.
1228 error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
);
1233 * Now we are mounted, reserve a small amount of unused space for
1234 * privileged transactions. This is needed so that transaction
1235 * space required for critical operations can dip into this pool
1236 * when at ENOSPC. This is needed for operations like create with
1237 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1238 * are not allowed to use this reserved space.
1240 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1241 * This may drive us straight to ENOSPC on mount, but that implies
1242 * we were already there on the last unmount. Warn if this occurs.
1244 resblks
= mp
->m_sb
.sb_dblocks
;
1245 do_div(resblks
, 20);
1246 resblks
= min_t(__uint64_t
, resblks
, 1024);
1247 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1249 cmn_err(CE_WARN
, "XFS: Unable to allocate reserve blocks. "
1250 "Continuing without a reserve pool.");
1255 xfs_rtunmount_inodes(mp
);
1259 xfs_log_unmount(mp
);
1263 xfs_uuid_unmount(mp
);
1269 * This flushes out the inodes,dquots and the superblock, unmounts the
1270 * log and makes sure that incore structures are freed.
1274 struct xfs_mount
*mp
)
1280 * Release dquot that rootinode, rbmino and rsumino might be holding,
1281 * and release the quota inodes.
1285 xfs_rtunmount_inodes(mp
);
1286 IRELE(mp
->m_rootip
);
1289 * We can potentially deadlock here if we have an inode cluster
1290 * that has been freed has its buffer still pinned in memory because
1291 * the transaction is still sitting in a iclog. The stale inodes
1292 * on that buffer will have their flush locks held until the
1293 * transaction hits the disk and the callbacks run. the inode
1294 * flush takes the flush lock unconditionally and with nothing to
1295 * push out the iclog we will never get that unlocked. hence we
1296 * need to force the log first.
1298 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1299 xfs_reclaim_inodes(mp
, 0, XFS_IFLUSH_ASYNC
);
1301 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1303 if (mp
->m_quotainfo
)
1307 * Flush out the log synchronously so that we know for sure
1308 * that nothing is pinned. This is important because bflush()
1309 * will skip pinned buffers.
1311 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1313 xfs_binval(mp
->m_ddev_targp
);
1314 if (mp
->m_rtdev_targp
) {
1315 xfs_binval(mp
->m_rtdev_targp
);
1319 * Unreserve any blocks we have so that when we unmount we don't account
1320 * the reserved free space as used. This is really only necessary for
1321 * lazy superblock counting because it trusts the incore superblock
1322 * counters to be absolutely correct on clean unmount.
1324 * We don't bother correcting this elsewhere for lazy superblock
1325 * counting because on mount of an unclean filesystem we reconstruct the
1326 * correct counter value and this is irrelevant.
1328 * For non-lazy counter filesystems, this doesn't matter at all because
1329 * we only every apply deltas to the superblock and hence the incore
1330 * value does not matter....
1333 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1335 cmn_err(CE_WARN
, "XFS: Unable to free reserved block pool. "
1336 "Freespace may not be correct on next mount.");
1338 error
= xfs_log_sbcount(mp
, 1);
1340 cmn_err(CE_WARN
, "XFS: Unable to update superblock counters. "
1341 "Freespace may not be correct on next mount.");
1342 xfs_unmountfs_writesb(mp
);
1343 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1344 xfs_log_unmount_write(mp
);
1345 xfs_log_unmount(mp
);
1346 xfs_uuid_unmount(mp
);
1349 xfs_errortag_clearall(mp
, 0);
1355 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1357 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1358 xfs_wait_buftarg(mp
->m_logdev_targp
);
1359 if (mp
->m_rtdev_targp
)
1360 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1361 xfs_wait_buftarg(mp
->m_ddev_targp
);
1365 xfs_fs_writable(xfs_mount_t
*mp
)
1367 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1368 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1374 * Called either periodically to keep the on disk superblock values
1375 * roughly up to date or from unmount to make sure the values are
1376 * correct on a clean unmount.
1378 * Note this code can be called during the process of freezing, so
1379 * we may need to use the transaction allocator which does not not
1380 * block when the transaction subsystem is in its frozen state.
1390 if (!xfs_fs_writable(mp
))
1393 xfs_icsb_sync_counters(mp
, 0);
1396 * we don't need to do this if we are updating the superblock
1397 * counters on every modification.
1399 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1402 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
);
1403 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1404 XFS_DEFAULT_LOG_COUNT
);
1406 xfs_trans_cancel(tp
, 0);
1410 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1412 xfs_trans_set_sync(tp
);
1413 error
= xfs_trans_commit(tp
, 0);
1418 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1424 * skip superblock write if fs is read-only, or
1425 * if we are doing a forced umount.
1427 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1428 XFS_FORCED_SHUTDOWN(mp
))) {
1430 sbp
= xfs_getsb(mp
, 0);
1432 XFS_BUF_UNDONE(sbp
);
1433 XFS_BUF_UNREAD(sbp
);
1434 XFS_BUF_UNDELAYWRITE(sbp
);
1436 XFS_BUF_UNASYNC(sbp
);
1437 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1438 xfsbdstrat(mp
, sbp
);
1439 error
= xfs_iowait(sbp
);
1441 xfs_ioerror_alert("xfs_unmountfs_writesb",
1442 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1449 * xfs_mod_sb() can be used to copy arbitrary changes to the
1450 * in-core superblock into the superblock buffer to be logged.
1451 * It does not provide the higher level of locking that is
1452 * needed to protect the in-core superblock from concurrent
1456 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1468 bp
= xfs_trans_getsb(tp
, mp
, 0);
1469 first
= sizeof(xfs_sb_t
);
1472 /* translate/copy */
1474 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1476 /* find modified range */
1478 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1479 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1480 first
= xfs_sb_info
[f
].offset
;
1482 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1483 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1484 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1486 xfs_trans_log_buf(tp
, bp
, first
, last
);
1491 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1492 * a delta to a specified field in the in-core superblock. Simply
1493 * switch on the field indicated and apply the delta to that field.
1494 * Fields are not allowed to dip below zero, so if the delta would
1495 * do this do not apply it and return EINVAL.
1497 * The m_sb_lock must be held when this routine is called.
1500 xfs_mod_incore_sb_unlocked(
1502 xfs_sb_field_t field
,
1506 int scounter
; /* short counter for 32 bit fields */
1507 long long lcounter
; /* long counter for 64 bit fields */
1508 long long res_used
, rem
;
1511 * With the in-core superblock spin lock held, switch
1512 * on the indicated field. Apply the delta to the
1513 * proper field. If the fields value would dip below
1514 * 0, then do not apply the delta and return EINVAL.
1517 case XFS_SBS_ICOUNT
:
1518 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1522 return XFS_ERROR(EINVAL
);
1524 mp
->m_sb
.sb_icount
= lcounter
;
1527 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1531 return XFS_ERROR(EINVAL
);
1533 mp
->m_sb
.sb_ifree
= lcounter
;
1535 case XFS_SBS_FDBLOCKS
:
1536 lcounter
= (long long)
1537 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1538 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1540 if (delta
> 0) { /* Putting blocks back */
1541 if (res_used
> delta
) {
1542 mp
->m_resblks_avail
+= delta
;
1544 rem
= delta
- res_used
;
1545 mp
->m_resblks_avail
= mp
->m_resblks
;
1548 } else { /* Taking blocks away */
1553 * If were out of blocks, use any available reserved blocks if
1559 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1561 return XFS_ERROR(ENOSPC
);
1563 mp
->m_resblks_avail
= lcounter
;
1565 } else { /* not reserved */
1566 return XFS_ERROR(ENOSPC
);
1571 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1573 case XFS_SBS_FREXTENTS
:
1574 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1577 return XFS_ERROR(ENOSPC
);
1579 mp
->m_sb
.sb_frextents
= lcounter
;
1581 case XFS_SBS_DBLOCKS
:
1582 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1586 return XFS_ERROR(EINVAL
);
1588 mp
->m_sb
.sb_dblocks
= lcounter
;
1590 case XFS_SBS_AGCOUNT
:
1591 scounter
= mp
->m_sb
.sb_agcount
;
1595 return XFS_ERROR(EINVAL
);
1597 mp
->m_sb
.sb_agcount
= scounter
;
1599 case XFS_SBS_IMAX_PCT
:
1600 scounter
= mp
->m_sb
.sb_imax_pct
;
1604 return XFS_ERROR(EINVAL
);
1606 mp
->m_sb
.sb_imax_pct
= scounter
;
1608 case XFS_SBS_REXTSIZE
:
1609 scounter
= mp
->m_sb
.sb_rextsize
;
1613 return XFS_ERROR(EINVAL
);
1615 mp
->m_sb
.sb_rextsize
= scounter
;
1617 case XFS_SBS_RBMBLOCKS
:
1618 scounter
= mp
->m_sb
.sb_rbmblocks
;
1622 return XFS_ERROR(EINVAL
);
1624 mp
->m_sb
.sb_rbmblocks
= scounter
;
1626 case XFS_SBS_RBLOCKS
:
1627 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1631 return XFS_ERROR(EINVAL
);
1633 mp
->m_sb
.sb_rblocks
= lcounter
;
1635 case XFS_SBS_REXTENTS
:
1636 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1640 return XFS_ERROR(EINVAL
);
1642 mp
->m_sb
.sb_rextents
= lcounter
;
1644 case XFS_SBS_REXTSLOG
:
1645 scounter
= mp
->m_sb
.sb_rextslog
;
1649 return XFS_ERROR(EINVAL
);
1651 mp
->m_sb
.sb_rextslog
= scounter
;
1655 return XFS_ERROR(EINVAL
);
1660 * xfs_mod_incore_sb() is used to change a field in the in-core
1661 * superblock structure by the specified delta. This modification
1662 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1663 * routine to do the work.
1668 xfs_sb_field_t field
,
1674 /* check for per-cpu counters */
1676 #ifdef HAVE_PERCPU_SB
1677 case XFS_SBS_ICOUNT
:
1679 case XFS_SBS_FDBLOCKS
:
1680 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1681 status
= xfs_icsb_modify_counters(mp
, field
,
1688 spin_lock(&mp
->m_sb_lock
);
1689 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1690 spin_unlock(&mp
->m_sb_lock
);
1698 * xfs_mod_incore_sb_batch() is used to change more than one field
1699 * in the in-core superblock structure at a time. This modification
1700 * is protected by a lock internal to this module. The fields and
1701 * changes to those fields are specified in the array of xfs_mod_sb
1702 * structures passed in.
1704 * Either all of the specified deltas will be applied or none of
1705 * them will. If any modified field dips below 0, then all modifications
1706 * will be backed out and EINVAL will be returned.
1709 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1715 * Loop through the array of mod structures and apply each
1716 * individually. If any fail, then back out all those
1717 * which have already been applied. Do all of this within
1718 * the scope of the m_sb_lock so that all of the changes will
1721 spin_lock(&mp
->m_sb_lock
);
1723 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1725 * Apply the delta at index n. If it fails, break
1726 * from the loop so we'll fall into the undo loop
1729 switch (msbp
->msb_field
) {
1730 #ifdef HAVE_PERCPU_SB
1731 case XFS_SBS_ICOUNT
:
1733 case XFS_SBS_FDBLOCKS
:
1734 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1735 spin_unlock(&mp
->m_sb_lock
);
1736 status
= xfs_icsb_modify_counters(mp
,
1738 msbp
->msb_delta
, rsvd
);
1739 spin_lock(&mp
->m_sb_lock
);
1745 status
= xfs_mod_incore_sb_unlocked(mp
,
1747 msbp
->msb_delta
, rsvd
);
1757 * If we didn't complete the loop above, then back out
1758 * any changes made to the superblock. If you add code
1759 * between the loop above and here, make sure that you
1760 * preserve the value of status. Loop back until
1761 * we step below the beginning of the array. Make sure
1762 * we don't touch anything back there.
1766 while (msbp
>= msb
) {
1767 switch (msbp
->msb_field
) {
1768 #ifdef HAVE_PERCPU_SB
1769 case XFS_SBS_ICOUNT
:
1771 case XFS_SBS_FDBLOCKS
:
1772 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1773 spin_unlock(&mp
->m_sb_lock
);
1774 status
= xfs_icsb_modify_counters(mp
,
1778 spin_lock(&mp
->m_sb_lock
);
1784 status
= xfs_mod_incore_sb_unlocked(mp
,
1790 ASSERT(status
== 0);
1794 spin_unlock(&mp
->m_sb_lock
);
1799 * xfs_getsb() is called to obtain the buffer for the superblock.
1800 * The buffer is returned locked and read in from disk.
1801 * The buffer should be released with a call to xfs_brelse().
1803 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1804 * the superblock buffer if it can be locked without sleeping.
1805 * If it can't then we'll return NULL.
1814 ASSERT(mp
->m_sb_bp
!= NULL
);
1816 if (flags
& XFS_BUF_TRYLOCK
) {
1817 if (!XFS_BUF_CPSEMA(bp
)) {
1821 XFS_BUF_PSEMA(bp
, PRIBIO
);
1824 ASSERT(XFS_BUF_ISDONE(bp
));
1829 * Used to free the superblock along various error paths.
1838 * Use xfs_getsb() so that the buffer will be locked
1839 * when we call xfs_buf_relse().
1841 bp
= xfs_getsb(mp
, 0);
1842 XFS_BUF_UNMANAGE(bp
);
1848 * Used to log changes to the superblock unit and width fields which could
1849 * be altered by the mount options, as well as any potential sb_features2
1850 * fixup. Only the first superblock is updated.
1860 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1861 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1862 XFS_SB_VERSIONNUM
));
1864 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1865 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1866 XFS_DEFAULT_LOG_COUNT
);
1868 xfs_trans_cancel(tp
, 0);
1871 xfs_mod_sb(tp
, fields
);
1872 error
= xfs_trans_commit(tp
, 0);
1877 #ifdef HAVE_PERCPU_SB
1879 * Per-cpu incore superblock counters
1881 * Simple concept, difficult implementation
1883 * Basically, replace the incore superblock counters with a distributed per cpu
1884 * counter for contended fields (e.g. free block count).
1886 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1887 * hence needs to be accurately read when we are running low on space. Hence
1888 * there is a method to enable and disable the per-cpu counters based on how
1889 * much "stuff" is available in them.
1891 * Basically, a counter is enabled if there is enough free resource to justify
1892 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1893 * ENOSPC), then we disable the counters to synchronise all callers and
1894 * re-distribute the available resources.
1896 * If, once we redistributed the available resources, we still get a failure,
1897 * we disable the per-cpu counter and go through the slow path.
1899 * The slow path is the current xfs_mod_incore_sb() function. This means that
1900 * when we disable a per-cpu counter, we need to drain its resources back to
1901 * the global superblock. We do this after disabling the counter to prevent
1902 * more threads from queueing up on the counter.
1904 * Essentially, this means that we still need a lock in the fast path to enable
1905 * synchronisation between the global counters and the per-cpu counters. This
1906 * is not a problem because the lock will be local to a CPU almost all the time
1907 * and have little contention except when we get to ENOSPC conditions.
1909 * Basically, this lock becomes a barrier that enables us to lock out the fast
1910 * path while we do things like enabling and disabling counters and
1911 * synchronising the counters.
1915 * 1. m_sb_lock before picking up per-cpu locks
1916 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1917 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1918 * 4. modifying per-cpu counters requires holding per-cpu lock
1919 * 5. modifying global counters requires holding m_sb_lock
1920 * 6. enabling or disabling a counter requires holding the m_sb_lock
1921 * and _none_ of the per-cpu locks.
1923 * Disabled counters are only ever re-enabled by a balance operation
1924 * that results in more free resources per CPU than a given threshold.
1925 * To ensure counters don't remain disabled, they are rebalanced when
1926 * the global resource goes above a higher threshold (i.e. some hysteresis
1927 * is present to prevent thrashing).
1930 #ifdef CONFIG_HOTPLUG_CPU
1932 * hot-plug CPU notifier support.
1934 * We need a notifier per filesystem as we need to be able to identify
1935 * the filesystem to balance the counters out. This is achieved by
1936 * having a notifier block embedded in the xfs_mount_t and doing pointer
1937 * magic to get the mount pointer from the notifier block address.
1940 xfs_icsb_cpu_notify(
1941 struct notifier_block
*nfb
,
1942 unsigned long action
,
1945 xfs_icsb_cnts_t
*cntp
;
1948 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1949 cntp
= (xfs_icsb_cnts_t
*)
1950 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1952 case CPU_UP_PREPARE
:
1953 case CPU_UP_PREPARE_FROZEN
:
1954 /* Easy Case - initialize the area and locks, and
1955 * then rebalance when online does everything else for us. */
1956 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1959 case CPU_ONLINE_FROZEN
:
1961 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1962 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1963 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1964 xfs_icsb_unlock(mp
);
1967 case CPU_DEAD_FROZEN
:
1968 /* Disable all the counters, then fold the dead cpu's
1969 * count into the total on the global superblock and
1970 * re-enable the counters. */
1972 spin_lock(&mp
->m_sb_lock
);
1973 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1974 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1975 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1977 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1978 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1979 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1981 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1983 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
1984 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
1985 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
1986 spin_unlock(&mp
->m_sb_lock
);
1987 xfs_icsb_unlock(mp
);
1993 #endif /* CONFIG_HOTPLUG_CPU */
1996 xfs_icsb_init_counters(
1999 xfs_icsb_cnts_t
*cntp
;
2002 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2003 if (mp
->m_sb_cnts
== NULL
)
2006 #ifdef CONFIG_HOTPLUG_CPU
2007 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2008 mp
->m_icsb_notifier
.priority
= 0;
2009 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2010 #endif /* CONFIG_HOTPLUG_CPU */
2012 for_each_online_cpu(i
) {
2013 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2014 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2017 mutex_init(&mp
->m_icsb_mutex
);
2020 * start with all counters disabled so that the
2021 * initial balance kicks us off correctly
2023 mp
->m_icsb_counters
= -1;
2028 xfs_icsb_reinit_counters(
2033 * start with all counters disabled so that the
2034 * initial balance kicks us off correctly
2036 mp
->m_icsb_counters
= -1;
2037 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2038 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2039 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2040 xfs_icsb_unlock(mp
);
2044 xfs_icsb_destroy_counters(
2047 if (mp
->m_sb_cnts
) {
2048 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2049 free_percpu(mp
->m_sb_cnts
);
2051 mutex_destroy(&mp
->m_icsb_mutex
);
2056 xfs_icsb_cnts_t
*icsbp
)
2058 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2064 xfs_icsb_unlock_cntr(
2065 xfs_icsb_cnts_t
*icsbp
)
2067 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2072 xfs_icsb_lock_all_counters(
2075 xfs_icsb_cnts_t
*cntp
;
2078 for_each_online_cpu(i
) {
2079 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2080 xfs_icsb_lock_cntr(cntp
);
2085 xfs_icsb_unlock_all_counters(
2088 xfs_icsb_cnts_t
*cntp
;
2091 for_each_online_cpu(i
) {
2092 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2093 xfs_icsb_unlock_cntr(cntp
);
2100 xfs_icsb_cnts_t
*cnt
,
2103 xfs_icsb_cnts_t
*cntp
;
2106 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2108 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2109 xfs_icsb_lock_all_counters(mp
);
2111 for_each_online_cpu(i
) {
2112 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2113 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2114 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2115 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2118 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2119 xfs_icsb_unlock_all_counters(mp
);
2123 xfs_icsb_counter_disabled(
2125 xfs_sb_field_t field
)
2127 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2128 return test_bit(field
, &mp
->m_icsb_counters
);
2132 xfs_icsb_disable_counter(
2134 xfs_sb_field_t field
)
2136 xfs_icsb_cnts_t cnt
;
2138 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2141 * If we are already disabled, then there is nothing to do
2142 * here. We check before locking all the counters to avoid
2143 * the expensive lock operation when being called in the
2144 * slow path and the counter is already disabled. This is
2145 * safe because the only time we set or clear this state is under
2148 if (xfs_icsb_counter_disabled(mp
, field
))
2151 xfs_icsb_lock_all_counters(mp
);
2152 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2153 /* drain back to superblock */
2155 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2157 case XFS_SBS_ICOUNT
:
2158 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2161 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2163 case XFS_SBS_FDBLOCKS
:
2164 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2171 xfs_icsb_unlock_all_counters(mp
);
2175 xfs_icsb_enable_counter(
2177 xfs_sb_field_t field
,
2181 xfs_icsb_cnts_t
*cntp
;
2184 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2186 xfs_icsb_lock_all_counters(mp
);
2187 for_each_online_cpu(i
) {
2188 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2190 case XFS_SBS_ICOUNT
:
2191 cntp
->icsb_icount
= count
+ resid
;
2194 cntp
->icsb_ifree
= count
+ resid
;
2196 case XFS_SBS_FDBLOCKS
:
2197 cntp
->icsb_fdblocks
= count
+ resid
;
2205 clear_bit(field
, &mp
->m_icsb_counters
);
2206 xfs_icsb_unlock_all_counters(mp
);
2210 xfs_icsb_sync_counters_locked(
2214 xfs_icsb_cnts_t cnt
;
2216 xfs_icsb_count(mp
, &cnt
, flags
);
2218 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2219 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2220 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2221 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2222 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2223 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2227 * Accurate update of per-cpu counters to incore superblock
2230 xfs_icsb_sync_counters(
2234 spin_lock(&mp
->m_sb_lock
);
2235 xfs_icsb_sync_counters_locked(mp
, flags
);
2236 spin_unlock(&mp
->m_sb_lock
);
2240 * Balance and enable/disable counters as necessary.
2242 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2243 * chosen to be the same number as single on disk allocation chunk per CPU, and
2244 * free blocks is something far enough zero that we aren't going thrash when we
2245 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2246 * prevent looping endlessly when xfs_alloc_space asks for more than will
2247 * be distributed to a single CPU but each CPU has enough blocks to be
2250 * Note that we can be called when counters are already disabled.
2251 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2252 * prevent locking every per-cpu counter needlessly.
2255 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2256 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2257 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2259 xfs_icsb_balance_counter_locked(
2261 xfs_sb_field_t field
,
2264 uint64_t count
, resid
;
2265 int weight
= num_online_cpus();
2266 uint64_t min
= (uint64_t)min_per_cpu
;
2268 /* disable counter and sync counter */
2269 xfs_icsb_disable_counter(mp
, field
);
2271 /* update counters - first CPU gets residual*/
2273 case XFS_SBS_ICOUNT
:
2274 count
= mp
->m_sb
.sb_icount
;
2275 resid
= do_div(count
, weight
);
2276 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2280 count
= mp
->m_sb
.sb_ifree
;
2281 resid
= do_div(count
, weight
);
2282 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2285 case XFS_SBS_FDBLOCKS
:
2286 count
= mp
->m_sb
.sb_fdblocks
;
2287 resid
= do_div(count
, weight
);
2288 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2293 count
= resid
= 0; /* quiet, gcc */
2297 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2301 xfs_icsb_balance_counter(
2303 xfs_sb_field_t fields
,
2306 spin_lock(&mp
->m_sb_lock
);
2307 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2308 spin_unlock(&mp
->m_sb_lock
);
2312 xfs_icsb_modify_counters(
2314 xfs_sb_field_t field
,
2318 xfs_icsb_cnts_t
*icsbp
;
2319 long long lcounter
; /* long counter for 64 bit fields */
2325 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
);
2328 * if the counter is disabled, go to slow path
2330 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2332 xfs_icsb_lock_cntr(icsbp
);
2333 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2334 xfs_icsb_unlock_cntr(icsbp
);
2339 case XFS_SBS_ICOUNT
:
2340 lcounter
= icsbp
->icsb_icount
;
2342 if (unlikely(lcounter
< 0))
2343 goto balance_counter
;
2344 icsbp
->icsb_icount
= lcounter
;
2348 lcounter
= icsbp
->icsb_ifree
;
2350 if (unlikely(lcounter
< 0))
2351 goto balance_counter
;
2352 icsbp
->icsb_ifree
= lcounter
;
2355 case XFS_SBS_FDBLOCKS
:
2356 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2358 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2360 if (unlikely(lcounter
< 0))
2361 goto balance_counter
;
2362 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2368 xfs_icsb_unlock_cntr(icsbp
);
2376 * serialise with a mutex so we don't burn lots of cpu on
2377 * the superblock lock. We still need to hold the superblock
2378 * lock, however, when we modify the global structures.
2383 * Now running atomically.
2385 * If the counter is enabled, someone has beaten us to rebalancing.
2386 * Drop the lock and try again in the fast path....
2388 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2389 xfs_icsb_unlock(mp
);
2394 * The counter is currently disabled. Because we are
2395 * running atomically here, we know a rebalance cannot
2396 * be in progress. Hence we can go straight to operating
2397 * on the global superblock. We do not call xfs_mod_incore_sb()
2398 * here even though we need to get the m_sb_lock. Doing so
2399 * will cause us to re-enter this function and deadlock.
2400 * Hence we get the m_sb_lock ourselves and then call
2401 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2402 * directly on the global counters.
2404 spin_lock(&mp
->m_sb_lock
);
2405 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2406 spin_unlock(&mp
->m_sb_lock
);
2409 * Now that we've modified the global superblock, we
2410 * may be able to re-enable the distributed counters
2411 * (e.g. lots of space just got freed). After that
2415 xfs_icsb_balance_counter(mp
, field
, 0);
2416 xfs_icsb_unlock(mp
);
2420 xfs_icsb_unlock_cntr(icsbp
);
2424 * We may have multiple threads here if multiple per-cpu
2425 * counters run dry at the same time. This will mean we can
2426 * do more balances than strictly necessary but it is not
2427 * the common slowpath case.
2432 * running atomically.
2434 * This will leave the counter in the correct state for future
2435 * accesses. After the rebalance, we simply try again and our retry
2436 * will either succeed through the fast path or slow path without
2437 * another balance operation being required.
2439 xfs_icsb_balance_counter(mp
, field
, delta
);
2440 xfs_icsb_unlock(mp
);