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
int xfs_mount_log_sb(xfs_mount_t
*, __int64_t
);
49 STATIC
int xfs_uuid_mount(xfs_mount_t
*);
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 }
126 * Free up the resources associated with a mount structure. Assume that
127 * the structure was initially zeroed, so we can tell which fields got
137 for (agno
= 0; agno
< mp
->m_maxagi
; agno
++)
138 if (mp
->m_perag
[agno
].pagb_list
)
139 kmem_free(mp
->m_perag
[agno
].pagb_list
);
140 kmem_free(mp
->m_perag
);
143 spinlock_destroy(&mp
->m_ail_lock
);
144 spinlock_destroy(&mp
->m_sb_lock
);
145 mutex_destroy(&mp
->m_ilock
);
146 mutex_destroy(&mp
->m_growlock
);
152 * Check size of device based on the (data/realtime) block count.
153 * Note: this check is used by the growfs code as well as mount.
156 xfs_sb_validate_fsb_count(
160 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
161 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
163 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
164 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
166 #else /* Limited by UINT_MAX of sectors */
167 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
174 * Check the validity of the SB found.
177 xfs_mount_validate_sb(
183 * If the log device and data device have the
184 * same device number, the log is internal.
185 * Consequently, the sb_logstart should be non-zero. If
186 * we have a zero sb_logstart in this case, we may be trying to mount
187 * a volume filesystem in a non-volume manner.
189 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
190 xfs_fs_mount_cmn_err(flags
, "bad magic number");
191 return XFS_ERROR(EWRONGFS
);
194 if (!xfs_sb_good_version(sbp
)) {
195 xfs_fs_mount_cmn_err(flags
, "bad version");
196 return XFS_ERROR(EWRONGFS
);
200 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
201 xfs_fs_mount_cmn_err(flags
,
202 "filesystem is marked as having an external log; "
203 "specify logdev on the\nmount command line.");
204 return XFS_ERROR(EINVAL
);
208 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
209 xfs_fs_mount_cmn_err(flags
,
210 "filesystem is marked as having an internal log; "
211 "do not specify logdev on\nthe mount command line.");
212 return XFS_ERROR(EINVAL
);
216 * More sanity checking. These were stolen directly from
220 sbp
->sb_agcount
<= 0 ||
221 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
222 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
223 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
224 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
225 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
226 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
227 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
228 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
229 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
230 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
231 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
232 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
233 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
234 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
235 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
236 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
237 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
238 return XFS_ERROR(EFSCORRUPTED
);
242 * Sanity check AG count, size fields against data size field
245 sbp
->sb_dblocks
== 0 ||
247 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
248 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
249 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
250 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
251 return XFS_ERROR(EFSCORRUPTED
);
255 * Until this is fixed only page-sized or smaller data blocks work.
257 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
258 xfs_fs_mount_cmn_err(flags
,
259 "file system with blocksize %d bytes",
261 xfs_fs_mount_cmn_err(flags
,
262 "only pagesize (%ld) or less will currently work.",
264 return XFS_ERROR(ENOSYS
);
267 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
268 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
269 xfs_fs_mount_cmn_err(flags
,
270 "file system too large to be mounted on this system.");
271 return XFS_ERROR(E2BIG
);
274 if (unlikely(sbp
->sb_inprogress
)) {
275 xfs_fs_mount_cmn_err(flags
, "file system busy");
276 return XFS_ERROR(EFSCORRUPTED
);
280 * Version 1 directory format has never worked on Linux.
282 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
283 xfs_fs_mount_cmn_err(flags
,
284 "file system using version 1 directory format");
285 return XFS_ERROR(ENOSYS
);
292 xfs_initialize_perag_icache(
295 if (!pag
->pag_ici_init
) {
296 rwlock_init(&pag
->pag_ici_lock
);
297 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
298 pag
->pag_ici_init
= 1;
303 xfs_initialize_perag(
305 xfs_agnumber_t agcount
)
307 xfs_agnumber_t index
, max_metadata
;
311 xfs_sb_t
*sbp
= &mp
->m_sb
;
312 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
314 /* Check to see if the filesystem can overflow 32 bit inodes */
315 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
316 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
318 /* Clear the mount flag if no inode can overflow 32 bits
319 * on this filesystem, or if specifically requested..
321 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> max_inum
) {
322 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
324 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
327 /* If we can overflow then setup the ag headers accordingly */
328 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
329 /* Calculate how much should be reserved for inodes to
330 * meet the max inode percentage.
332 if (mp
->m_maxicount
) {
335 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
337 icount
+= sbp
->sb_agblocks
- 1;
338 do_div(icount
, sbp
->sb_agblocks
);
339 max_metadata
= icount
;
341 max_metadata
= agcount
;
343 for (index
= 0; index
< agcount
; index
++) {
344 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
345 if (ino
> max_inum
) {
350 /* This ag is preferred for inodes */
351 pag
= &mp
->m_perag
[index
];
352 pag
->pagi_inodeok
= 1;
353 if (index
< max_metadata
)
354 pag
->pagf_metadata
= 1;
355 xfs_initialize_perag_icache(pag
);
358 /* Setup default behavior for smaller filesystems */
359 for (index
= 0; index
< agcount
; index
++) {
360 pag
= &mp
->m_perag
[index
];
361 pag
->pagi_inodeok
= 1;
362 xfs_initialize_perag_icache(pag
);
373 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
374 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
375 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
376 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
377 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
378 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
379 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
380 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
381 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
382 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
383 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
384 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
385 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
386 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
387 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
388 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
389 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
390 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
391 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
392 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
393 to
->sb_blocklog
= from
->sb_blocklog
;
394 to
->sb_sectlog
= from
->sb_sectlog
;
395 to
->sb_inodelog
= from
->sb_inodelog
;
396 to
->sb_inopblog
= from
->sb_inopblog
;
397 to
->sb_agblklog
= from
->sb_agblklog
;
398 to
->sb_rextslog
= from
->sb_rextslog
;
399 to
->sb_inprogress
= from
->sb_inprogress
;
400 to
->sb_imax_pct
= from
->sb_imax_pct
;
401 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
402 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
403 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
404 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
405 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
406 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
407 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
408 to
->sb_flags
= from
->sb_flags
;
409 to
->sb_shared_vn
= from
->sb_shared_vn
;
410 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
411 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
412 to
->sb_width
= be32_to_cpu(from
->sb_width
);
413 to
->sb_dirblklog
= from
->sb_dirblklog
;
414 to
->sb_logsectlog
= from
->sb_logsectlog
;
415 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
416 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
417 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
418 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
422 * Copy in core superblock to ondisk one.
424 * The fields argument is mask of superblock fields to copy.
432 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
433 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
443 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
444 first
= xfs_sb_info
[f
].offset
;
445 size
= xfs_sb_info
[f
+ 1].offset
- first
;
447 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
449 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
450 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
454 *(__be16
*)(to_ptr
+ first
) =
455 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
458 *(__be32
*)(to_ptr
+ first
) =
459 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
462 *(__be64
*)(to_ptr
+ first
) =
463 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
470 fields
&= ~(1LL << f
);
477 * Does the initial read of the superblock.
480 xfs_readsb(xfs_mount_t
*mp
, int flags
)
482 unsigned int sector_size
;
483 unsigned int extra_flags
;
487 ASSERT(mp
->m_sb_bp
== NULL
);
488 ASSERT(mp
->m_ddev_targp
!= NULL
);
491 * Allocate a (locked) buffer to hold the superblock.
492 * This will be kept around at all times to optimize
493 * access to the superblock.
495 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
496 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
498 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
499 BTOBB(sector_size
), extra_flags
);
500 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
501 xfs_fs_mount_cmn_err(flags
, "SB read failed");
502 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
505 ASSERT(XFS_BUF_ISBUSY(bp
));
506 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
509 * Initialize the mount structure from the superblock.
510 * But first do some basic consistency checking.
512 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
514 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
516 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
521 * We must be able to do sector-sized and sector-aligned IO.
523 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
524 xfs_fs_mount_cmn_err(flags
,
525 "device supports only %u byte sectors (not %u)",
526 sector_size
, mp
->m_sb
.sb_sectsize
);
532 * If device sector size is smaller than the superblock size,
533 * re-read the superblock so the buffer is correctly sized.
535 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
536 XFS_BUF_UNMANAGE(bp
);
538 sector_size
= mp
->m_sb
.sb_sectsize
;
539 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
540 BTOBB(sector_size
), extra_flags
);
541 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
542 xfs_fs_mount_cmn_err(flags
, "SB re-read failed");
543 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
546 ASSERT(XFS_BUF_ISBUSY(bp
));
547 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
550 /* Initialize per-cpu counters */
551 xfs_icsb_reinit_counters(mp
);
555 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
560 XFS_BUF_UNMANAGE(bp
);
570 * Mount initialization code establishing various mount
571 * fields from the superblock associated with the given
575 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
579 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
580 spin_lock_init(&mp
->m_agirotor_lock
);
581 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
582 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
583 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
584 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
585 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
586 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
587 mp
->m_litino
= sbp
->sb_inodesize
-
588 ((uint
)sizeof(xfs_dinode_core_t
) + (uint
)sizeof(xfs_agino_t
));
589 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
590 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
591 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
592 INIT_LIST_HEAD(&mp
->m_del_inodes
);
595 * Setup for attributes, in case they get created.
596 * This value is for inodes getting attributes for the first time,
597 * the per-inode value is for old attribute values.
599 ASSERT(sbp
->sb_inodesize
>= 256 && sbp
->sb_inodesize
<= 2048);
600 switch (sbp
->sb_inodesize
) {
602 mp
->m_attroffset
= XFS_LITINO(mp
) -
603 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
608 mp
->m_attroffset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
613 ASSERT(mp
->m_attroffset
< XFS_LITINO(mp
));
615 for (i
= 0; i
< 2; i
++) {
616 mp
->m_alloc_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
618 mp
->m_alloc_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
621 for (i
= 0; i
< 2; i
++) {
622 mp
->m_bmap_dmxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
624 mp
->m_bmap_dmnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
627 for (i
= 0; i
< 2; i
++) {
628 mp
->m_inobt_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
630 mp
->m_inobt_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
634 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
635 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
637 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
641 * xfs_initialize_perag_data
643 * Read in each per-ag structure so we can count up the number of
644 * allocated inodes, free inodes and used filesystem blocks as this
645 * information is no longer persistent in the superblock. Once we have
646 * this information, write it into the in-core superblock structure.
649 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
651 xfs_agnumber_t index
;
653 xfs_sb_t
*sbp
= &mp
->m_sb
;
657 uint64_t bfreelst
= 0;
661 for (index
= 0; index
< agcount
; index
++) {
663 * read the agf, then the agi. This gets us
664 * all the inforamtion we need and populates the
665 * per-ag structures for us.
667 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
671 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
674 pag
= &mp
->m_perag
[index
];
675 ifree
+= pag
->pagi_freecount
;
676 ialloc
+= pag
->pagi_count
;
677 bfree
+= pag
->pagf_freeblks
;
678 bfreelst
+= pag
->pagf_flcount
;
679 btree
+= pag
->pagf_btreeblks
;
682 * Overwrite incore superblock counters with just-read data
684 spin_lock(&mp
->m_sb_lock
);
685 sbp
->sb_ifree
= ifree
;
686 sbp
->sb_icount
= ialloc
;
687 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
688 spin_unlock(&mp
->m_sb_lock
);
690 /* Fixup the per-cpu counters as well. */
691 xfs_icsb_reinit_counters(mp
);
697 * Update alignment values based on mount options and sb values
700 xfs_update_alignment(xfs_mount_t
*mp
, __uint64_t
*update_flags
)
702 xfs_sb_t
*sbp
= &(mp
->m_sb
);
706 * If stripe unit and stripe width are not multiples
707 * of the fs blocksize turn off alignment.
709 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
710 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
711 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
713 "XFS: alignment check 1 failed");
714 return XFS_ERROR(EINVAL
);
716 mp
->m_dalign
= mp
->m_swidth
= 0;
719 * Convert the stripe unit and width to FSBs.
721 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
722 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
723 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
724 return XFS_ERROR(EINVAL
);
726 xfs_fs_cmn_err(CE_WARN
, mp
,
727 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
728 mp
->m_dalign
, mp
->m_swidth
,
733 } else if (mp
->m_dalign
) {
734 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
736 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
737 xfs_fs_cmn_err(CE_WARN
, mp
,
738 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
741 return XFS_ERROR(EINVAL
);
748 * Update superblock with new values
751 if (xfs_sb_version_hasdalign(sbp
)) {
752 if (sbp
->sb_unit
!= mp
->m_dalign
) {
753 sbp
->sb_unit
= mp
->m_dalign
;
754 *update_flags
|= XFS_SB_UNIT
;
756 if (sbp
->sb_width
!= mp
->m_swidth
) {
757 sbp
->sb_width
= mp
->m_swidth
;
758 *update_flags
|= XFS_SB_WIDTH
;
761 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
762 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
763 mp
->m_dalign
= sbp
->sb_unit
;
764 mp
->m_swidth
= sbp
->sb_width
;
771 * Set the maximum inode count for this filesystem
774 xfs_set_maxicount(xfs_mount_t
*mp
)
776 xfs_sb_t
*sbp
= &(mp
->m_sb
);
779 if (sbp
->sb_imax_pct
) {
781 * Make sure the maximum inode count is a multiple
782 * of the units we allocate inodes in.
784 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
786 do_div(icount
, mp
->m_ialloc_blks
);
787 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
795 * Set the default minimum read and write sizes unless
796 * already specified in a mount option.
797 * We use smaller I/O sizes when the file system
798 * is being used for NFS service (wsync mount option).
801 xfs_set_rw_sizes(xfs_mount_t
*mp
)
803 xfs_sb_t
*sbp
= &(mp
->m_sb
);
804 int readio_log
, writeio_log
;
806 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
807 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
808 readio_log
= XFS_WSYNC_READIO_LOG
;
809 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
811 readio_log
= XFS_READIO_LOG_LARGE
;
812 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
815 readio_log
= mp
->m_readio_log
;
816 writeio_log
= mp
->m_writeio_log
;
819 if (sbp
->sb_blocklog
> readio_log
) {
820 mp
->m_readio_log
= sbp
->sb_blocklog
;
822 mp
->m_readio_log
= readio_log
;
824 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
825 if (sbp
->sb_blocklog
> writeio_log
) {
826 mp
->m_writeio_log
= sbp
->sb_blocklog
;
828 mp
->m_writeio_log
= writeio_log
;
830 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
834 * Set whether we're using inode alignment.
837 xfs_set_inoalignment(xfs_mount_t
*mp
)
839 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
840 mp
->m_sb
.sb_inoalignmt
>=
841 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
842 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
844 mp
->m_inoalign_mask
= 0;
846 * If we are using stripe alignment, check whether
847 * the stripe unit is a multiple of the inode alignment
849 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
850 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
851 mp
->m_sinoalign
= mp
->m_dalign
;
857 * Check that the data (and log if separate) are an ok size.
860 xfs_check_sizes(xfs_mount_t
*mp
)
866 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
867 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
868 cmn_err(CE_WARN
, "XFS: size check 1 failed");
869 return XFS_ERROR(E2BIG
);
871 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
872 d
- XFS_FSS_TO_BB(mp
, 1),
873 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
877 cmn_err(CE_WARN
, "XFS: size check 2 failed");
879 error
= XFS_ERROR(E2BIG
);
883 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
884 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
885 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
886 cmn_err(CE_WARN
, "XFS: size check 3 failed");
887 return XFS_ERROR(E2BIG
);
889 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
890 d
- XFS_FSB_TO_BB(mp
, 1),
891 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
895 cmn_err(CE_WARN
, "XFS: size check 3 failed");
897 error
= XFS_ERROR(E2BIG
);
907 * This function does the following on an initial mount of a file system:
908 * - reads the superblock from disk and init the mount struct
909 * - if we're a 32-bit kernel, do a size check on the superblock
910 * so we don't mount terabyte filesystems
911 * - init mount struct realtime fields
912 * - allocate inode hash table for fs
913 * - init directory manager
914 * - perform recovery and init the log manager
920 xfs_sb_t
*sbp
= &(mp
->m_sb
);
923 __int64_t update_flags
= 0LL;
924 uint quotamount
, quotaflags
;
926 int uuid_mounted
= 0;
929 xfs_mount_common(mp
, sbp
);
932 * Check for a mismatched features2 values. Older kernels
933 * read & wrote into the wrong sb offset for sb_features2
934 * on some platforms due to xfs_sb_t not being 64bit size aligned
935 * when sb_features2 was added, which made older superblock
936 * reading/writing routines swap it as a 64-bit value.
938 * For backwards compatibility, we make both slots equal.
940 * If we detect a mismatched field, we OR the set bits into the
941 * existing features2 field in case it has already been modified; we
942 * don't want to lose any features. We then update the bad location
943 * with the ORed value so that older kernels will see any features2
944 * flags, and mark the two fields as needing updates once the
945 * transaction subsystem is online.
947 if (xfs_sb_has_mismatched_features2(sbp
)) {
949 "XFS: correcting sb_features alignment problem");
950 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
951 sbp
->sb_bad_features2
= sbp
->sb_features2
;
952 update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
955 * Re-check for ATTR2 in case it was found in bad_features2
958 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
959 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
960 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
963 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
964 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
965 xfs_sb_version_removeattr2(&mp
->m_sb
);
966 update_flags
|= XFS_SB_FEATURES2
;
968 /* update sb_versionnum for the clearing of the morebits */
969 if (!sbp
->sb_features2
)
970 update_flags
|= XFS_SB_VERSIONNUM
;
974 * Check if sb_agblocks is aligned at stripe boundary
975 * If sb_agblocks is NOT aligned turn off m_dalign since
976 * allocator alignment is within an ag, therefore ag has
977 * to be aligned at stripe boundary.
979 error
= xfs_update_alignment(mp
, &update_flags
);
983 xfs_alloc_compute_maxlevels(mp
);
984 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
985 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
986 xfs_ialloc_compute_maxlevels(mp
);
988 xfs_set_maxicount(mp
);
990 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
993 * XFS uses the uuid from the superblock as the unique
994 * identifier for fsid. We can not use the uuid from the volume
995 * since a single partition filesystem is identical to a single
996 * partition volume/filesystem.
998 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0) {
999 if (xfs_uuid_mount(mp
)) {
1000 error
= XFS_ERROR(EINVAL
);
1007 * Set the minimum read and write sizes
1009 xfs_set_rw_sizes(mp
);
1012 * Set the inode cluster size.
1013 * This may still be overridden by the file system
1014 * block size if it is larger than the chosen cluster size.
1016 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1019 * Set inode alignment fields
1021 xfs_set_inoalignment(mp
);
1024 * Check that the data (and log if separate) are an ok size.
1026 error
= xfs_check_sizes(mp
);
1031 * Initialize realtime fields in the mount structure
1033 error
= xfs_rtmount_init(mp
);
1035 cmn_err(CE_WARN
, "XFS: RT mount failed");
1040 * Copies the low order bits of the timestamp and the randomly
1041 * set "sequence" number out of a UUID.
1043 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1045 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1050 * Initialize the attribute manager's entries.
1052 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1055 * Initialize the precomputed transaction reservations values.
1060 * Allocate and initialize the per-ag data.
1062 init_rwsem(&mp
->m_peraglock
);
1064 kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
), KM_SLEEP
);
1066 mp
->m_maxagi
= xfs_initialize_perag(mp
, sbp
->sb_agcount
);
1069 * log's mount-time initialization. Perform 1st part recovery if needed
1071 if (likely(sbp
->sb_logblocks
> 0)) { /* check for volume case */
1072 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1073 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1074 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1076 cmn_err(CE_WARN
, "XFS: log mount failed");
1079 } else { /* No log has been defined */
1080 cmn_err(CE_WARN
, "XFS: no log defined");
1081 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW
, mp
);
1082 error
= XFS_ERROR(EFSCORRUPTED
);
1087 * Now the log is mounted, we know if it was an unclean shutdown or
1088 * not. If it was, with the first phase of recovery has completed, we
1089 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1090 * but they are recovered transactionally in the second recovery phase
1093 * Hence we can safely re-initialise incore superblock counters from
1094 * the per-ag data. These may not be correct if the filesystem was not
1095 * cleanly unmounted, so we need to wait for recovery to finish before
1098 * If the filesystem was cleanly unmounted, then we can trust the
1099 * values in the superblock to be correct and we don't need to do
1102 * If we are currently making the filesystem, the initialisation will
1103 * fail as the perag data is in an undefined state.
1106 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1107 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1108 !mp
->m_sb
.sb_inprogress
) {
1109 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1115 * Get and sanity-check the root inode.
1116 * Save the pointer to it in the mount structure.
1118 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1120 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1124 ASSERT(rip
!= NULL
);
1126 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1127 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1128 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1129 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1130 (unsigned long long)rip
->i_ino
);
1131 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1132 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1134 error
= XFS_ERROR(EFSCORRUPTED
);
1137 mp
->m_rootip
= rip
; /* save it */
1139 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1142 * Initialize realtime inode pointers in the mount structure
1144 error
= xfs_rtmount_inodes(mp
);
1147 * Free up the root inode.
1149 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1154 * If fs is not mounted readonly, then update the superblock changes.
1156 if (update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1157 error
= xfs_mount_log_sb(mp
, update_flags
);
1159 cmn_err(CE_WARN
, "XFS: failed to write sb changes");
1165 * Initialise the XFS quota management subsystem for this mount
1167 error
= XFS_QM_INIT(mp
, "amount
, "aflags
);
1172 * Finish recovering the file system. This part needed to be
1173 * delayed until after the root and real-time bitmap inodes
1174 * were consistently read in.
1176 error
= xfs_log_mount_finish(mp
);
1178 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1183 * Complete the quota initialisation, post-log-replay component.
1185 error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
);
1190 * Now we are mounted, reserve a small amount of unused space for
1191 * privileged transactions. This is needed so that transaction
1192 * space required for critical operations can dip into this pool
1193 * when at ENOSPC. This is needed for operations like create with
1194 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1195 * are not allowed to use this reserved space.
1197 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1198 * This may drive us straight to ENOSPC on mount, but that implies
1199 * we were already there on the last unmount. Warn if this occurs.
1201 resblks
= mp
->m_sb
.sb_dblocks
;
1202 do_div(resblks
, 20);
1203 resblks
= min_t(__uint64_t
, resblks
, 1024);
1204 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1206 cmn_err(CE_WARN
, "XFS: Unable to allocate reserve blocks. "
1207 "Continuing without a reserve pool.");
1213 * Free up the root inode.
1217 xfs_log_unmount_dealloc(mp
);
1219 for (agno
= 0; agno
< sbp
->sb_agcount
; agno
++)
1220 if (mp
->m_perag
[agno
].pagb_list
)
1221 kmem_free(mp
->m_perag
[agno
].pagb_list
);
1222 kmem_free(mp
->m_perag
);
1227 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1232 * This flushes out the inodes,dquots and the superblock, unmounts the
1233 * log and makes sure that incore structures are freed.
1237 struct xfs_mount
*mp
)
1242 IRELE(mp
->m_rootip
);
1245 * We can potentially deadlock here if we have an inode cluster
1246 * that has been freed has it's buffer still pinned in memory because
1247 * the transaction is still sitting in a iclog. The stale inodes
1248 * on that buffer will have their flush locks held until the
1249 * transaction hits the disk and the callbacks run. the inode
1250 * flush takes the flush lock unconditionally and with nothing to
1251 * push out the iclog we will never get that unlocked. hence we
1252 * need to force the log first.
1254 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1257 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1260 * Flush out the log synchronously so that we know for sure
1261 * that nothing is pinned. This is important because bflush()
1262 * will skip pinned buffers.
1264 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1266 xfs_binval(mp
->m_ddev_targp
);
1267 if (mp
->m_rtdev_targp
) {
1268 xfs_binval(mp
->m_rtdev_targp
);
1272 * Unreserve any blocks we have so that when we unmount we don't account
1273 * the reserved free space as used. This is really only necessary for
1274 * lazy superblock counting because it trusts the incore superblock
1275 * counters to be aboslutely correct on clean unmount.
1277 * We don't bother correcting this elsewhere for lazy superblock
1278 * counting because on mount of an unclean filesystem we reconstruct the
1279 * correct counter value and this is irrelevant.
1281 * For non-lazy counter filesystems, this doesn't matter at all because
1282 * we only every apply deltas to the superblock and hence the incore
1283 * value does not matter....
1286 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1288 cmn_err(CE_WARN
, "XFS: Unable to free reserved block pool. "
1289 "Freespace may not be correct on next mount.");
1291 error
= xfs_log_sbcount(mp
, 1);
1293 cmn_err(CE_WARN
, "XFS: Unable to update superblock counters. "
1294 "Freespace may not be correct on next mount.");
1295 xfs_unmountfs_writesb(mp
);
1296 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1297 xfs_log_unmount(mp
); /* Done! No more fs ops. */
1302 * All inodes from this mount point should be freed.
1304 ASSERT(mp
->m_inodes
== NULL
);
1306 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0)
1307 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1310 xfs_errortag_clearall(mp
, 0);
1316 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1318 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1319 xfs_wait_buftarg(mp
->m_logdev_targp
);
1320 if (mp
->m_rtdev_targp
)
1321 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1322 xfs_wait_buftarg(mp
->m_ddev_targp
);
1326 xfs_fs_writable(xfs_mount_t
*mp
)
1328 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1329 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1335 * Called either periodically to keep the on disk superblock values
1336 * roughly up to date or from unmount to make sure the values are
1337 * correct on a clean unmount.
1339 * Note this code can be called during the process of freezing, so
1340 * we may need to use the transaction allocator which does not not
1341 * block when the transaction subsystem is in its frozen state.
1351 if (!xfs_fs_writable(mp
))
1354 xfs_icsb_sync_counters(mp
, 0);
1357 * we don't need to do this if we are updating the superblock
1358 * counters on every modification.
1360 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1363 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
);
1364 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1365 XFS_DEFAULT_LOG_COUNT
);
1367 xfs_trans_cancel(tp
, 0);
1371 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1373 xfs_trans_set_sync(tp
);
1374 error
= xfs_trans_commit(tp
, 0);
1383 xfs_dsb_t
*sb
= XFS_BUF_TO_SBP(bp
);
1386 if (!(sb
->sb_flags
& XFS_SBF_READONLY
))
1387 sb
->sb_flags
|= XFS_SBF_READONLY
;
1389 version
= be16_to_cpu(sb
->sb_versionnum
);
1390 if ((version
& XFS_SB_VERSION_NUMBITS
) != XFS_SB_VERSION_4
||
1391 !(version
& XFS_SB_VERSION_SHAREDBIT
))
1392 version
|= XFS_SB_VERSION_SHAREDBIT
;
1393 sb
->sb_versionnum
= cpu_to_be16(version
);
1397 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1403 * skip superblock write if fs is read-only, or
1404 * if we are doing a forced umount.
1406 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1407 XFS_FORCED_SHUTDOWN(mp
))) {
1409 sbp
= xfs_getsb(mp
, 0);
1412 * mark shared-readonly if desired
1414 if (mp
->m_mk_sharedro
)
1415 xfs_mark_shared_ro(mp
, sbp
);
1417 XFS_BUF_UNDONE(sbp
);
1418 XFS_BUF_UNREAD(sbp
);
1419 XFS_BUF_UNDELAYWRITE(sbp
);
1421 XFS_BUF_UNASYNC(sbp
);
1422 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1423 xfsbdstrat(mp
, sbp
);
1424 error
= xfs_iowait(sbp
);
1426 xfs_ioerror_alert("xfs_unmountfs_writesb",
1427 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1428 if (error
&& mp
->m_mk_sharedro
)
1429 xfs_fs_cmn_err(CE_ALERT
, mp
, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1436 * xfs_mod_sb() can be used to copy arbitrary changes to the
1437 * in-core superblock into the superblock buffer to be logged.
1438 * It does not provide the higher level of locking that is
1439 * needed to protect the in-core superblock from concurrent
1443 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1455 bp
= xfs_trans_getsb(tp
, mp
, 0);
1456 first
= sizeof(xfs_sb_t
);
1459 /* translate/copy */
1461 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1463 /* find modified range */
1465 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1466 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1467 first
= xfs_sb_info
[f
].offset
;
1469 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1470 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1471 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1473 xfs_trans_log_buf(tp
, bp
, first
, last
);
1478 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1479 * a delta to a specified field in the in-core superblock. Simply
1480 * switch on the field indicated and apply the delta to that field.
1481 * Fields are not allowed to dip below zero, so if the delta would
1482 * do this do not apply it and return EINVAL.
1484 * The m_sb_lock must be held when this routine is called.
1487 xfs_mod_incore_sb_unlocked(
1489 xfs_sb_field_t field
,
1493 int scounter
; /* short counter for 32 bit fields */
1494 long long lcounter
; /* long counter for 64 bit fields */
1495 long long res_used
, rem
;
1498 * With the in-core superblock spin lock held, switch
1499 * on the indicated field. Apply the delta to the
1500 * proper field. If the fields value would dip below
1501 * 0, then do not apply the delta and return EINVAL.
1504 case XFS_SBS_ICOUNT
:
1505 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1509 return XFS_ERROR(EINVAL
);
1511 mp
->m_sb
.sb_icount
= lcounter
;
1514 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1518 return XFS_ERROR(EINVAL
);
1520 mp
->m_sb
.sb_ifree
= lcounter
;
1522 case XFS_SBS_FDBLOCKS
:
1523 lcounter
= (long long)
1524 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1525 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1527 if (delta
> 0) { /* Putting blocks back */
1528 if (res_used
> delta
) {
1529 mp
->m_resblks_avail
+= delta
;
1531 rem
= delta
- res_used
;
1532 mp
->m_resblks_avail
= mp
->m_resblks
;
1535 } else { /* Taking blocks away */
1540 * If were out of blocks, use any available reserved blocks if
1546 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1548 return XFS_ERROR(ENOSPC
);
1550 mp
->m_resblks_avail
= lcounter
;
1552 } else { /* not reserved */
1553 return XFS_ERROR(ENOSPC
);
1558 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1560 case XFS_SBS_FREXTENTS
:
1561 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1564 return XFS_ERROR(ENOSPC
);
1566 mp
->m_sb
.sb_frextents
= lcounter
;
1568 case XFS_SBS_DBLOCKS
:
1569 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1573 return XFS_ERROR(EINVAL
);
1575 mp
->m_sb
.sb_dblocks
= lcounter
;
1577 case XFS_SBS_AGCOUNT
:
1578 scounter
= mp
->m_sb
.sb_agcount
;
1582 return XFS_ERROR(EINVAL
);
1584 mp
->m_sb
.sb_agcount
= scounter
;
1586 case XFS_SBS_IMAX_PCT
:
1587 scounter
= mp
->m_sb
.sb_imax_pct
;
1591 return XFS_ERROR(EINVAL
);
1593 mp
->m_sb
.sb_imax_pct
= scounter
;
1595 case XFS_SBS_REXTSIZE
:
1596 scounter
= mp
->m_sb
.sb_rextsize
;
1600 return XFS_ERROR(EINVAL
);
1602 mp
->m_sb
.sb_rextsize
= scounter
;
1604 case XFS_SBS_RBMBLOCKS
:
1605 scounter
= mp
->m_sb
.sb_rbmblocks
;
1609 return XFS_ERROR(EINVAL
);
1611 mp
->m_sb
.sb_rbmblocks
= scounter
;
1613 case XFS_SBS_RBLOCKS
:
1614 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1618 return XFS_ERROR(EINVAL
);
1620 mp
->m_sb
.sb_rblocks
= lcounter
;
1622 case XFS_SBS_REXTENTS
:
1623 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1627 return XFS_ERROR(EINVAL
);
1629 mp
->m_sb
.sb_rextents
= lcounter
;
1631 case XFS_SBS_REXTSLOG
:
1632 scounter
= mp
->m_sb
.sb_rextslog
;
1636 return XFS_ERROR(EINVAL
);
1638 mp
->m_sb
.sb_rextslog
= scounter
;
1642 return XFS_ERROR(EINVAL
);
1647 * xfs_mod_incore_sb() is used to change a field in the in-core
1648 * superblock structure by the specified delta. This modification
1649 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1650 * routine to do the work.
1655 xfs_sb_field_t field
,
1661 /* check for per-cpu counters */
1663 #ifdef HAVE_PERCPU_SB
1664 case XFS_SBS_ICOUNT
:
1666 case XFS_SBS_FDBLOCKS
:
1667 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1668 status
= xfs_icsb_modify_counters(mp
, field
,
1675 spin_lock(&mp
->m_sb_lock
);
1676 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1677 spin_unlock(&mp
->m_sb_lock
);
1685 * xfs_mod_incore_sb_batch() is used to change more than one field
1686 * in the in-core superblock structure at a time. This modification
1687 * is protected by a lock internal to this module. The fields and
1688 * changes to those fields are specified in the array of xfs_mod_sb
1689 * structures passed in.
1691 * Either all of the specified deltas will be applied or none of
1692 * them will. If any modified field dips below 0, then all modifications
1693 * will be backed out and EINVAL will be returned.
1696 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1702 * Loop through the array of mod structures and apply each
1703 * individually. If any fail, then back out all those
1704 * which have already been applied. Do all of this within
1705 * the scope of the m_sb_lock so that all of the changes will
1708 spin_lock(&mp
->m_sb_lock
);
1710 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1712 * Apply the delta at index n. If it fails, break
1713 * from the loop so we'll fall into the undo loop
1716 switch (msbp
->msb_field
) {
1717 #ifdef HAVE_PERCPU_SB
1718 case XFS_SBS_ICOUNT
:
1720 case XFS_SBS_FDBLOCKS
:
1721 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1722 spin_unlock(&mp
->m_sb_lock
);
1723 status
= xfs_icsb_modify_counters(mp
,
1725 msbp
->msb_delta
, rsvd
);
1726 spin_lock(&mp
->m_sb_lock
);
1732 status
= xfs_mod_incore_sb_unlocked(mp
,
1734 msbp
->msb_delta
, rsvd
);
1744 * If we didn't complete the loop above, then back out
1745 * any changes made to the superblock. If you add code
1746 * between the loop above and here, make sure that you
1747 * preserve the value of status. Loop back until
1748 * we step below the beginning of the array. Make sure
1749 * we don't touch anything back there.
1753 while (msbp
>= msb
) {
1754 switch (msbp
->msb_field
) {
1755 #ifdef HAVE_PERCPU_SB
1756 case XFS_SBS_ICOUNT
:
1758 case XFS_SBS_FDBLOCKS
:
1759 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1760 spin_unlock(&mp
->m_sb_lock
);
1761 status
= xfs_icsb_modify_counters(mp
,
1765 spin_lock(&mp
->m_sb_lock
);
1771 status
= xfs_mod_incore_sb_unlocked(mp
,
1777 ASSERT(status
== 0);
1781 spin_unlock(&mp
->m_sb_lock
);
1786 * xfs_getsb() is called to obtain the buffer for the superblock.
1787 * The buffer is returned locked and read in from disk.
1788 * The buffer should be released with a call to xfs_brelse().
1790 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1791 * the superblock buffer if it can be locked without sleeping.
1792 * If it can't then we'll return NULL.
1801 ASSERT(mp
->m_sb_bp
!= NULL
);
1803 if (flags
& XFS_BUF_TRYLOCK
) {
1804 if (!XFS_BUF_CPSEMA(bp
)) {
1808 XFS_BUF_PSEMA(bp
, PRIBIO
);
1811 ASSERT(XFS_BUF_ISDONE(bp
));
1816 * Used to free the superblock along various error paths.
1825 * Use xfs_getsb() so that the buffer will be locked
1826 * when we call xfs_buf_relse().
1828 bp
= xfs_getsb(mp
, 0);
1829 XFS_BUF_UNMANAGE(bp
);
1835 * See if the UUID is unique among mounted XFS filesystems.
1836 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1842 if (uuid_is_nil(&mp
->m_sb
.sb_uuid
)) {
1844 "XFS: Filesystem %s has nil UUID - can't mount",
1848 if (!uuid_table_insert(&mp
->m_sb
.sb_uuid
)) {
1850 "XFS: Filesystem %s has duplicate UUID - can't mount",
1858 * Used to log changes to the superblock unit and width fields which could
1859 * be altered by the mount options, as well as any potential sb_features2
1860 * fixup. Only the first superblock is updated.
1870 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1871 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1872 XFS_SB_VERSIONNUM
));
1874 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1875 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1876 XFS_DEFAULT_LOG_COUNT
);
1878 xfs_trans_cancel(tp
, 0);
1881 xfs_mod_sb(tp
, fields
);
1882 error
= xfs_trans_commit(tp
, 0);
1887 #ifdef HAVE_PERCPU_SB
1889 * Per-cpu incore superblock counters
1891 * Simple concept, difficult implementation
1893 * Basically, replace the incore superblock counters with a distributed per cpu
1894 * counter for contended fields (e.g. free block count).
1896 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1897 * hence needs to be accurately read when we are running low on space. Hence
1898 * there is a method to enable and disable the per-cpu counters based on how
1899 * much "stuff" is available in them.
1901 * Basically, a counter is enabled if there is enough free resource to justify
1902 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1903 * ENOSPC), then we disable the counters to synchronise all callers and
1904 * re-distribute the available resources.
1906 * If, once we redistributed the available resources, we still get a failure,
1907 * we disable the per-cpu counter and go through the slow path.
1909 * The slow path is the current xfs_mod_incore_sb() function. This means that
1910 * when we disable a per-cpu counter, we need to drain it's resources back to
1911 * the global superblock. We do this after disabling the counter to prevent
1912 * more threads from queueing up on the counter.
1914 * Essentially, this means that we still need a lock in the fast path to enable
1915 * synchronisation between the global counters and the per-cpu counters. This
1916 * is not a problem because the lock will be local to a CPU almost all the time
1917 * and have little contention except when we get to ENOSPC conditions.
1919 * Basically, this lock becomes a barrier that enables us to lock out the fast
1920 * path while we do things like enabling and disabling counters and
1921 * synchronising the counters.
1925 * 1. m_sb_lock before picking up per-cpu locks
1926 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1927 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1928 * 4. modifying per-cpu counters requires holding per-cpu lock
1929 * 5. modifying global counters requires holding m_sb_lock
1930 * 6. enabling or disabling a counter requires holding the m_sb_lock
1931 * and _none_ of the per-cpu locks.
1933 * Disabled counters are only ever re-enabled by a balance operation
1934 * that results in more free resources per CPU than a given threshold.
1935 * To ensure counters don't remain disabled, they are rebalanced when
1936 * the global resource goes above a higher threshold (i.e. some hysteresis
1937 * is present to prevent thrashing).
1940 #ifdef CONFIG_HOTPLUG_CPU
1942 * hot-plug CPU notifier support.
1944 * We need a notifier per filesystem as we need to be able to identify
1945 * the filesystem to balance the counters out. This is achieved by
1946 * having a notifier block embedded in the xfs_mount_t and doing pointer
1947 * magic to get the mount pointer from the notifier block address.
1950 xfs_icsb_cpu_notify(
1951 struct notifier_block
*nfb
,
1952 unsigned long action
,
1955 xfs_icsb_cnts_t
*cntp
;
1958 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1959 cntp
= (xfs_icsb_cnts_t
*)
1960 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1962 case CPU_UP_PREPARE
:
1963 case CPU_UP_PREPARE_FROZEN
:
1964 /* Easy Case - initialize the area and locks, and
1965 * then rebalance when online does everything else for us. */
1966 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1969 case CPU_ONLINE_FROZEN
:
1971 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1972 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1973 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1974 xfs_icsb_unlock(mp
);
1977 case CPU_DEAD_FROZEN
:
1978 /* Disable all the counters, then fold the dead cpu's
1979 * count into the total on the global superblock and
1980 * re-enable the counters. */
1982 spin_lock(&mp
->m_sb_lock
);
1983 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1984 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1985 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1987 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1988 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1989 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1991 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1993 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
1994 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
1995 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
1996 spin_unlock(&mp
->m_sb_lock
);
1997 xfs_icsb_unlock(mp
);
2003 #endif /* CONFIG_HOTPLUG_CPU */
2006 xfs_icsb_init_counters(
2009 xfs_icsb_cnts_t
*cntp
;
2012 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2013 if (mp
->m_sb_cnts
== NULL
)
2016 #ifdef CONFIG_HOTPLUG_CPU
2017 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2018 mp
->m_icsb_notifier
.priority
= 0;
2019 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2020 #endif /* CONFIG_HOTPLUG_CPU */
2022 for_each_online_cpu(i
) {
2023 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2024 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2027 mutex_init(&mp
->m_icsb_mutex
);
2030 * start with all counters disabled so that the
2031 * initial balance kicks us off correctly
2033 mp
->m_icsb_counters
= -1;
2038 xfs_icsb_reinit_counters(
2043 * start with all counters disabled so that the
2044 * initial balance kicks us off correctly
2046 mp
->m_icsb_counters
= -1;
2047 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2048 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2049 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2050 xfs_icsb_unlock(mp
);
2054 xfs_icsb_destroy_counters(
2057 if (mp
->m_sb_cnts
) {
2058 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2059 free_percpu(mp
->m_sb_cnts
);
2061 mutex_destroy(&mp
->m_icsb_mutex
);
2066 xfs_icsb_cnts_t
*icsbp
)
2068 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2074 xfs_icsb_unlock_cntr(
2075 xfs_icsb_cnts_t
*icsbp
)
2077 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2082 xfs_icsb_lock_all_counters(
2085 xfs_icsb_cnts_t
*cntp
;
2088 for_each_online_cpu(i
) {
2089 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2090 xfs_icsb_lock_cntr(cntp
);
2095 xfs_icsb_unlock_all_counters(
2098 xfs_icsb_cnts_t
*cntp
;
2101 for_each_online_cpu(i
) {
2102 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2103 xfs_icsb_unlock_cntr(cntp
);
2110 xfs_icsb_cnts_t
*cnt
,
2113 xfs_icsb_cnts_t
*cntp
;
2116 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2118 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2119 xfs_icsb_lock_all_counters(mp
);
2121 for_each_online_cpu(i
) {
2122 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2123 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2124 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2125 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2128 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2129 xfs_icsb_unlock_all_counters(mp
);
2133 xfs_icsb_counter_disabled(
2135 xfs_sb_field_t field
)
2137 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2138 return test_bit(field
, &mp
->m_icsb_counters
);
2142 xfs_icsb_disable_counter(
2144 xfs_sb_field_t field
)
2146 xfs_icsb_cnts_t cnt
;
2148 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2151 * If we are already disabled, then there is nothing to do
2152 * here. We check before locking all the counters to avoid
2153 * the expensive lock operation when being called in the
2154 * slow path and the counter is already disabled. This is
2155 * safe because the only time we set or clear this state is under
2158 if (xfs_icsb_counter_disabled(mp
, field
))
2161 xfs_icsb_lock_all_counters(mp
);
2162 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2163 /* drain back to superblock */
2165 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2167 case XFS_SBS_ICOUNT
:
2168 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2171 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2173 case XFS_SBS_FDBLOCKS
:
2174 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2181 xfs_icsb_unlock_all_counters(mp
);
2185 xfs_icsb_enable_counter(
2187 xfs_sb_field_t field
,
2191 xfs_icsb_cnts_t
*cntp
;
2194 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2196 xfs_icsb_lock_all_counters(mp
);
2197 for_each_online_cpu(i
) {
2198 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2200 case XFS_SBS_ICOUNT
:
2201 cntp
->icsb_icount
= count
+ resid
;
2204 cntp
->icsb_ifree
= count
+ resid
;
2206 case XFS_SBS_FDBLOCKS
:
2207 cntp
->icsb_fdblocks
= count
+ resid
;
2215 clear_bit(field
, &mp
->m_icsb_counters
);
2216 xfs_icsb_unlock_all_counters(mp
);
2220 xfs_icsb_sync_counters_locked(
2224 xfs_icsb_cnts_t cnt
;
2226 xfs_icsb_count(mp
, &cnt
, flags
);
2228 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2229 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2230 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2231 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2232 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2233 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2237 * Accurate update of per-cpu counters to incore superblock
2240 xfs_icsb_sync_counters(
2244 spin_lock(&mp
->m_sb_lock
);
2245 xfs_icsb_sync_counters_locked(mp
, flags
);
2246 spin_unlock(&mp
->m_sb_lock
);
2250 * Balance and enable/disable counters as necessary.
2252 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2253 * chosen to be the same number as single on disk allocation chunk per CPU, and
2254 * free blocks is something far enough zero that we aren't going thrash when we
2255 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2256 * prevent looping endlessly when xfs_alloc_space asks for more than will
2257 * be distributed to a single CPU but each CPU has enough blocks to be
2260 * Note that we can be called when counters are already disabled.
2261 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2262 * prevent locking every per-cpu counter needlessly.
2265 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2266 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2267 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2269 xfs_icsb_balance_counter_locked(
2271 xfs_sb_field_t field
,
2274 uint64_t count
, resid
;
2275 int weight
= num_online_cpus();
2276 uint64_t min
= (uint64_t)min_per_cpu
;
2278 /* disable counter and sync counter */
2279 xfs_icsb_disable_counter(mp
, field
);
2281 /* update counters - first CPU gets residual*/
2283 case XFS_SBS_ICOUNT
:
2284 count
= mp
->m_sb
.sb_icount
;
2285 resid
= do_div(count
, weight
);
2286 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2290 count
= mp
->m_sb
.sb_ifree
;
2291 resid
= do_div(count
, weight
);
2292 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2295 case XFS_SBS_FDBLOCKS
:
2296 count
= mp
->m_sb
.sb_fdblocks
;
2297 resid
= do_div(count
, weight
);
2298 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2303 count
= resid
= 0; /* quiet, gcc */
2307 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2311 xfs_icsb_balance_counter(
2313 xfs_sb_field_t fields
,
2316 spin_lock(&mp
->m_sb_lock
);
2317 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2318 spin_unlock(&mp
->m_sb_lock
);
2322 xfs_icsb_modify_counters(
2324 xfs_sb_field_t field
,
2328 xfs_icsb_cnts_t
*icsbp
;
2329 long long lcounter
; /* long counter for 64 bit fields */
2335 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
);
2338 * if the counter is disabled, go to slow path
2340 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2342 xfs_icsb_lock_cntr(icsbp
);
2343 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2344 xfs_icsb_unlock_cntr(icsbp
);
2349 case XFS_SBS_ICOUNT
:
2350 lcounter
= icsbp
->icsb_icount
;
2352 if (unlikely(lcounter
< 0))
2353 goto balance_counter
;
2354 icsbp
->icsb_icount
= lcounter
;
2358 lcounter
= icsbp
->icsb_ifree
;
2360 if (unlikely(lcounter
< 0))
2361 goto balance_counter
;
2362 icsbp
->icsb_ifree
= lcounter
;
2365 case XFS_SBS_FDBLOCKS
:
2366 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2368 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2370 if (unlikely(lcounter
< 0))
2371 goto balance_counter
;
2372 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2378 xfs_icsb_unlock_cntr(icsbp
);
2386 * serialise with a mutex so we don't burn lots of cpu on
2387 * the superblock lock. We still need to hold the superblock
2388 * lock, however, when we modify the global structures.
2393 * Now running atomically.
2395 * If the counter is enabled, someone has beaten us to rebalancing.
2396 * Drop the lock and try again in the fast path....
2398 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2399 xfs_icsb_unlock(mp
);
2404 * The counter is currently disabled. Because we are
2405 * running atomically here, we know a rebalance cannot
2406 * be in progress. Hence we can go straight to operating
2407 * on the global superblock. We do not call xfs_mod_incore_sb()
2408 * here even though we need to get the m_sb_lock. Doing so
2409 * will cause us to re-enter this function and deadlock.
2410 * Hence we get the m_sb_lock ourselves and then call
2411 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2412 * directly on the global counters.
2414 spin_lock(&mp
->m_sb_lock
);
2415 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2416 spin_unlock(&mp
->m_sb_lock
);
2419 * Now that we've modified the global superblock, we
2420 * may be able to re-enable the distributed counters
2421 * (e.g. lots of space just got freed). After that
2425 xfs_icsb_balance_counter(mp
, field
, 0);
2426 xfs_icsb_unlock(mp
);
2430 xfs_icsb_unlock_cntr(icsbp
);
2434 * We may have multiple threads here if multiple per-cpu
2435 * counters run dry at the same time. This will mean we can
2436 * do more balances than strictly necessary but it is not
2437 * the common slowpath case.
2442 * running atomically.
2444 * This will leave the counter in the correct state for future
2445 * accesses. After the rebalance, we simply try again and our retry
2446 * will either succeed through the fast path or slow path without
2447 * another balance operation being required.
2449 xfs_icsb_balance_counter(mp
, field
, delta
);
2450 xfs_icsb_unlock(mp
);