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_uuid_unmount(xfs_mount_t
*mp
);
51 STATIC
void xfs_unmountfs_wait(xfs_mount_t
*);
55 STATIC
void xfs_icsb_destroy_counters(xfs_mount_t
*);
56 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
58 STATIC
void xfs_icsb_sync_counters(xfs_mount_t
*);
59 STATIC
int xfs_icsb_modify_counters(xfs_mount_t
*, xfs_sb_field_t
,
61 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
65 #define xfs_icsb_destroy_counters(mp) do { } while (0)
66 #define xfs_icsb_balance_counter(mp, a, b, c) do { } while (0)
67 #define xfs_icsb_sync_counters(mp) do { } while (0)
68 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
74 short type
; /* 0 = integer
75 * 1 = binary / string (no translation)
78 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
79 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
80 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
81 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
82 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
83 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
84 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
85 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
86 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
87 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
88 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
89 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
90 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
91 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
92 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
93 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
94 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
95 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
96 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
97 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
98 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
99 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
100 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
101 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
102 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
103 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
104 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
105 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
106 { offsetof(xfs_sb_t
, sb_icount
), 0 },
107 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
108 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
109 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
110 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
111 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
112 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
113 { offsetof(xfs_sb_t
, sb_flags
), 0 },
114 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
115 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
116 { offsetof(xfs_sb_t
, sb_unit
), 0 },
117 { offsetof(xfs_sb_t
, sb_width
), 0 },
118 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
119 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
120 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
121 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
122 { offsetof(xfs_sb_t
, sb_features2
), 0 },
123 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
124 { sizeof(xfs_sb_t
), 0 }
128 * Return a pointer to an initialized xfs_mount structure.
135 mp
= kmem_zalloc(sizeof(xfs_mount_t
), KM_SLEEP
);
137 if (xfs_icsb_init_counters(mp
)) {
138 mp
->m_flags
|= XFS_MOUNT_NO_PERCPU_SB
;
141 spin_lock_init(&mp
->m_sb_lock
);
142 mutex_init(&mp
->m_ilock
);
143 mutex_init(&mp
->m_growlock
);
144 atomic_set(&mp
->m_active_trans
, 0);
150 * Free up the resources associated with a mount structure. Assume that
151 * the structure was initially zeroed, so we can tell which fields got
161 for (agno
= 0; agno
< mp
->m_maxagi
; agno
++)
162 if (mp
->m_perag
[agno
].pagb_list
)
163 kmem_free(mp
->m_perag
[agno
].pagb_list
,
164 sizeof(xfs_perag_busy_t
) *
166 kmem_free(mp
->m_perag
,
167 sizeof(xfs_perag_t
) * mp
->m_sb
.sb_agcount
);
170 spinlock_destroy(&mp
->m_ail_lock
);
171 spinlock_destroy(&mp
->m_sb_lock
);
172 mutex_destroy(&mp
->m_ilock
);
173 mutex_destroy(&mp
->m_growlock
);
177 if (mp
->m_fsname
!= NULL
)
178 kmem_free(mp
->m_fsname
, mp
->m_fsname_len
);
179 if (mp
->m_rtname
!= NULL
)
180 kmem_free(mp
->m_rtname
, strlen(mp
->m_rtname
) + 1);
181 if (mp
->m_logname
!= NULL
)
182 kmem_free(mp
->m_logname
, strlen(mp
->m_logname
) + 1);
184 xfs_icsb_destroy_counters(mp
);
188 * Check size of device based on the (data/realtime) block count.
189 * Note: this check is used by the growfs code as well as mount.
192 xfs_sb_validate_fsb_count(
196 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
197 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
199 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
200 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
202 #else /* Limited by UINT_MAX of sectors */
203 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
210 * Check the validity of the SB found.
213 xfs_mount_validate_sb(
219 * If the log device and data device have the
220 * same device number, the log is internal.
221 * Consequently, the sb_logstart should be non-zero. If
222 * we have a zero sb_logstart in this case, we may be trying to mount
223 * a volume filesystem in a non-volume manner.
225 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
226 xfs_fs_mount_cmn_err(flags
, "bad magic number");
227 return XFS_ERROR(EWRONGFS
);
230 if (!xfs_sb_good_version(sbp
)) {
231 xfs_fs_mount_cmn_err(flags
, "bad version");
232 return XFS_ERROR(EWRONGFS
);
236 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
237 xfs_fs_mount_cmn_err(flags
,
238 "filesystem is marked as having an external log; "
239 "specify logdev on the\nmount command line.");
240 return XFS_ERROR(EINVAL
);
244 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
245 xfs_fs_mount_cmn_err(flags
,
246 "filesystem is marked as having an internal log; "
247 "do not specify logdev on\nthe mount command line.");
248 return XFS_ERROR(EINVAL
);
252 * More sanity checking. These were stolen directly from
256 sbp
->sb_agcount
<= 0 ||
257 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
258 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
259 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
260 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
261 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
262 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
263 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
264 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
265 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
266 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
267 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
268 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
269 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
270 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
271 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
272 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
273 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
274 return XFS_ERROR(EFSCORRUPTED
);
278 * Sanity check AG count, size fields against data size field
281 sbp
->sb_dblocks
== 0 ||
283 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
284 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
285 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
286 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
287 return XFS_ERROR(EFSCORRUPTED
);
290 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
291 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
292 xfs_fs_mount_cmn_err(flags
,
293 "file system too large to be mounted on this system.");
294 return XFS_ERROR(E2BIG
);
297 if (unlikely(sbp
->sb_inprogress
)) {
298 xfs_fs_mount_cmn_err(flags
, "file system busy");
299 return XFS_ERROR(EFSCORRUPTED
);
303 * Version 1 directory format has never worked on Linux.
305 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
306 xfs_fs_mount_cmn_err(flags
,
307 "file system using version 1 directory format");
308 return XFS_ERROR(ENOSYS
);
312 * Until this is fixed only page-sized or smaller data blocks work.
314 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
315 xfs_fs_mount_cmn_err(flags
,
316 "file system with blocksize %d bytes",
318 xfs_fs_mount_cmn_err(flags
,
319 "only pagesize (%ld) or less will currently work.",
321 return XFS_ERROR(ENOSYS
);
328 xfs_initialize_perag_icache(
331 if (!pag
->pag_ici_init
) {
332 rwlock_init(&pag
->pag_ici_lock
);
333 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
334 pag
->pag_ici_init
= 1;
339 xfs_initialize_perag(
341 xfs_agnumber_t agcount
)
343 xfs_agnumber_t index
, max_metadata
;
347 xfs_sb_t
*sbp
= &mp
->m_sb
;
348 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
350 /* Check to see if the filesystem can overflow 32 bit inodes */
351 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
352 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
354 /* Clear the mount flag if no inode can overflow 32 bits
355 * on this filesystem, or if specifically requested..
357 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> max_inum
) {
358 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
360 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
363 /* If we can overflow then setup the ag headers accordingly */
364 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
365 /* Calculate how much should be reserved for inodes to
366 * meet the max inode percentage.
368 if (mp
->m_maxicount
) {
371 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
373 icount
+= sbp
->sb_agblocks
- 1;
374 do_div(icount
, sbp
->sb_agblocks
);
375 max_metadata
= icount
;
377 max_metadata
= agcount
;
379 for (index
= 0; index
< agcount
; index
++) {
380 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
381 if (ino
> max_inum
) {
386 /* This ag is preferred for inodes */
387 pag
= &mp
->m_perag
[index
];
388 pag
->pagi_inodeok
= 1;
389 if (index
< max_metadata
)
390 pag
->pagf_metadata
= 1;
391 xfs_initialize_perag_icache(pag
);
394 /* Setup default behavior for smaller filesystems */
395 for (index
= 0; index
< agcount
; index
++) {
396 pag
= &mp
->m_perag
[index
];
397 pag
->pagi_inodeok
= 1;
398 xfs_initialize_perag_icache(pag
);
409 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
410 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
411 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
412 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
413 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
414 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
415 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
416 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
417 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
418 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
419 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
420 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
421 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
422 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
423 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
424 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
425 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
426 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
427 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
428 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
429 to
->sb_blocklog
= from
->sb_blocklog
;
430 to
->sb_sectlog
= from
->sb_sectlog
;
431 to
->sb_inodelog
= from
->sb_inodelog
;
432 to
->sb_inopblog
= from
->sb_inopblog
;
433 to
->sb_agblklog
= from
->sb_agblklog
;
434 to
->sb_rextslog
= from
->sb_rextslog
;
435 to
->sb_inprogress
= from
->sb_inprogress
;
436 to
->sb_imax_pct
= from
->sb_imax_pct
;
437 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
438 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
439 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
440 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
441 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
442 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
443 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
444 to
->sb_flags
= from
->sb_flags
;
445 to
->sb_shared_vn
= from
->sb_shared_vn
;
446 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
447 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
448 to
->sb_width
= be32_to_cpu(from
->sb_width
);
449 to
->sb_dirblklog
= from
->sb_dirblklog
;
450 to
->sb_logsectlog
= from
->sb_logsectlog
;
451 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
452 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
453 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
454 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
458 * Copy in core superblock to ondisk one.
460 * The fields argument is mask of superblock fields to copy.
468 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
469 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
479 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
480 first
= xfs_sb_info
[f
].offset
;
481 size
= xfs_sb_info
[f
+ 1].offset
- first
;
483 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
485 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
486 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
490 *(__be16
*)(to_ptr
+ first
) =
491 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
494 *(__be32
*)(to_ptr
+ first
) =
495 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
498 *(__be64
*)(to_ptr
+ first
) =
499 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
506 fields
&= ~(1LL << f
);
513 * Does the initial read of the superblock.
516 xfs_readsb(xfs_mount_t
*mp
, int flags
)
518 unsigned int sector_size
;
519 unsigned int extra_flags
;
523 ASSERT(mp
->m_sb_bp
== NULL
);
524 ASSERT(mp
->m_ddev_targp
!= NULL
);
527 * Allocate a (locked) buffer to hold the superblock.
528 * This will be kept around at all times to optimize
529 * access to the superblock.
531 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
532 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
534 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
535 BTOBB(sector_size
), extra_flags
);
536 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
537 xfs_fs_mount_cmn_err(flags
, "SB read failed");
538 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
541 ASSERT(XFS_BUF_ISBUSY(bp
));
542 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
545 * Initialize the mount structure from the superblock.
546 * But first do some basic consistency checking.
548 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
550 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
552 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
557 * We must be able to do sector-sized and sector-aligned IO.
559 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
560 xfs_fs_mount_cmn_err(flags
,
561 "device supports only %u byte sectors (not %u)",
562 sector_size
, mp
->m_sb
.sb_sectsize
);
568 * If device sector size is smaller than the superblock size,
569 * re-read the superblock so the buffer is correctly sized.
571 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
572 XFS_BUF_UNMANAGE(bp
);
574 sector_size
= mp
->m_sb
.sb_sectsize
;
575 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
576 BTOBB(sector_size
), extra_flags
);
577 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
578 xfs_fs_mount_cmn_err(flags
, "SB re-read failed");
579 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
582 ASSERT(XFS_BUF_ISBUSY(bp
));
583 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
586 /* Initialize per-cpu counters */
587 xfs_icsb_reinit_counters(mp
);
591 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
596 XFS_BUF_UNMANAGE(bp
);
606 * Mount initialization code establishing various mount
607 * fields from the superblock associated with the given
611 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
615 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
616 spin_lock_init(&mp
->m_agirotor_lock
);
617 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
618 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
619 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
620 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
621 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
622 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
623 mp
->m_litino
= sbp
->sb_inodesize
-
624 ((uint
)sizeof(xfs_dinode_core_t
) + (uint
)sizeof(xfs_agino_t
));
625 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
626 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
627 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
628 INIT_LIST_HEAD(&mp
->m_del_inodes
);
631 * Setup for attributes, in case they get created.
632 * This value is for inodes getting attributes for the first time,
633 * the per-inode value is for old attribute values.
635 ASSERT(sbp
->sb_inodesize
>= 256 && sbp
->sb_inodesize
<= 2048);
636 switch (sbp
->sb_inodesize
) {
638 mp
->m_attroffset
= XFS_LITINO(mp
) -
639 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
644 mp
->m_attroffset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
649 ASSERT(mp
->m_attroffset
< XFS_LITINO(mp
));
651 for (i
= 0; i
< 2; i
++) {
652 mp
->m_alloc_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
654 mp
->m_alloc_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
657 for (i
= 0; i
< 2; i
++) {
658 mp
->m_bmap_dmxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
660 mp
->m_bmap_dmnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
663 for (i
= 0; i
< 2; i
++) {
664 mp
->m_inobt_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
666 mp
->m_inobt_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
670 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
671 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
673 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
677 * xfs_initialize_perag_data
679 * Read in each per-ag structure so we can count up the number of
680 * allocated inodes, free inodes and used filesystem blocks as this
681 * information is no longer persistent in the superblock. Once we have
682 * this information, write it into the in-core superblock structure.
685 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
687 xfs_agnumber_t index
;
689 xfs_sb_t
*sbp
= &mp
->m_sb
;
693 uint64_t bfreelst
= 0;
697 for (index
= 0; index
< agcount
; index
++) {
699 * read the agf, then the agi. This gets us
700 * all the inforamtion we need and populates the
701 * per-ag structures for us.
703 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
707 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
710 pag
= &mp
->m_perag
[index
];
711 ifree
+= pag
->pagi_freecount
;
712 ialloc
+= pag
->pagi_count
;
713 bfree
+= pag
->pagf_freeblks
;
714 bfreelst
+= pag
->pagf_flcount
;
715 btree
+= pag
->pagf_btreeblks
;
718 * Overwrite incore superblock counters with just-read data
720 spin_lock(&mp
->m_sb_lock
);
721 sbp
->sb_ifree
= ifree
;
722 sbp
->sb_icount
= ialloc
;
723 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
724 spin_unlock(&mp
->m_sb_lock
);
726 /* Fixup the per-cpu counters as well. */
727 xfs_icsb_reinit_counters(mp
);
733 * Update alignment values based on mount options and sb values
736 xfs_update_alignment(xfs_mount_t
*mp
, int mfsi_flags
, __uint64_t
*update_flags
)
738 xfs_sb_t
*sbp
= &(mp
->m_sb
);
740 if (mp
->m_dalign
&& !(mfsi_flags
& XFS_MFSI_SECOND
)) {
742 * If stripe unit and stripe width are not multiples
743 * of the fs blocksize turn off alignment.
745 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
746 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
747 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
749 "XFS: alignment check 1 failed");
750 return XFS_ERROR(EINVAL
);
752 mp
->m_dalign
= mp
->m_swidth
= 0;
755 * Convert the stripe unit and width to FSBs.
757 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
758 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
759 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
760 return XFS_ERROR(EINVAL
);
762 xfs_fs_cmn_err(CE_WARN
, mp
,
763 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
764 mp
->m_dalign
, mp
->m_swidth
,
769 } else if (mp
->m_dalign
) {
770 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
772 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
773 xfs_fs_cmn_err(CE_WARN
, mp
,
774 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
777 return XFS_ERROR(EINVAL
);
784 * Update superblock with new values
787 if (xfs_sb_version_hasdalign(sbp
)) {
788 if (sbp
->sb_unit
!= mp
->m_dalign
) {
789 sbp
->sb_unit
= mp
->m_dalign
;
790 *update_flags
|= XFS_SB_UNIT
;
792 if (sbp
->sb_width
!= mp
->m_swidth
) {
793 sbp
->sb_width
= mp
->m_swidth
;
794 *update_flags
|= XFS_SB_WIDTH
;
797 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
798 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
799 mp
->m_dalign
= sbp
->sb_unit
;
800 mp
->m_swidth
= sbp
->sb_width
;
807 * Set the maximum inode count for this filesystem
810 xfs_set_maxicount(xfs_mount_t
*mp
)
812 xfs_sb_t
*sbp
= &(mp
->m_sb
);
815 if (sbp
->sb_imax_pct
) {
817 * Make sure the maximum inode count is a multiple
818 * of the units we allocate inodes in.
820 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
822 do_div(icount
, mp
->m_ialloc_blks
);
823 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
831 * Set the default minimum read and write sizes unless
832 * already specified in a mount option.
833 * We use smaller I/O sizes when the file system
834 * is being used for NFS service (wsync mount option).
837 xfs_set_rw_sizes(xfs_mount_t
*mp
)
839 xfs_sb_t
*sbp
= &(mp
->m_sb
);
840 int readio_log
, writeio_log
;
842 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
843 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
844 readio_log
= XFS_WSYNC_READIO_LOG
;
845 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
847 readio_log
= XFS_READIO_LOG_LARGE
;
848 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
851 readio_log
= mp
->m_readio_log
;
852 writeio_log
= mp
->m_writeio_log
;
855 if (sbp
->sb_blocklog
> readio_log
) {
856 mp
->m_readio_log
= sbp
->sb_blocklog
;
858 mp
->m_readio_log
= readio_log
;
860 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
861 if (sbp
->sb_blocklog
> writeio_log
) {
862 mp
->m_writeio_log
= sbp
->sb_blocklog
;
864 mp
->m_writeio_log
= writeio_log
;
866 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
870 * Set whether we're using inode alignment.
873 xfs_set_inoalignment(xfs_mount_t
*mp
)
875 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
876 mp
->m_sb
.sb_inoalignmt
>=
877 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
878 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
880 mp
->m_inoalign_mask
= 0;
882 * If we are using stripe alignment, check whether
883 * the stripe unit is a multiple of the inode alignment
885 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
886 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
887 mp
->m_sinoalign
= mp
->m_dalign
;
893 * Check that the data (and log if separate) are an ok size.
896 xfs_check_sizes(xfs_mount_t
*mp
, int mfsi_flags
)
902 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
903 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
904 cmn_err(CE_WARN
, "XFS: size check 1 failed");
905 return XFS_ERROR(E2BIG
);
907 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
908 d
- XFS_FSS_TO_BB(mp
, 1),
909 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
913 cmn_err(CE_WARN
, "XFS: size check 2 failed");
915 error
= XFS_ERROR(E2BIG
);
919 if (((mfsi_flags
& XFS_MFSI_CLIENT
) == 0) &&
920 mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
921 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
922 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
923 cmn_err(CE_WARN
, "XFS: size check 3 failed");
924 return XFS_ERROR(E2BIG
);
926 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
927 d
- XFS_FSB_TO_BB(mp
, 1),
928 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
932 cmn_err(CE_WARN
, "XFS: size check 3 failed");
934 error
= XFS_ERROR(E2BIG
);
944 * This function does the following on an initial mount of a file system:
945 * - reads the superblock from disk and init the mount struct
946 * - if we're a 32-bit kernel, do a size check on the superblock
947 * so we don't mount terabyte filesystems
948 * - init mount struct realtime fields
949 * - allocate inode hash table for fs
950 * - init directory manager
951 * - perform recovery and init the log manager
958 xfs_sb_t
*sbp
= &(mp
->m_sb
);
961 __int64_t update_flags
= 0LL;
962 uint quotamount
, quotaflags
;
964 int uuid_mounted
= 0;
967 xfs_mount_common(mp
, sbp
);
970 * Check for a mismatched features2 values. Older kernels
971 * read & wrote into the wrong sb offset for sb_features2
972 * on some platforms due to xfs_sb_t not being 64bit size aligned
973 * when sb_features2 was added, which made older superblock
974 * reading/writing routines swap it as a 64-bit value.
976 * For backwards compatibility, we make both slots equal.
978 * If we detect a mismatched field, we OR the set bits into the
979 * existing features2 field in case it has already been modified; we
980 * don't want to lose any features. We then update the bad location
981 * with the ORed value so that older kernels will see any features2
982 * flags, and mark the two fields as needing updates once the
983 * transaction subsystem is online.
985 if (xfs_sb_has_mismatched_features2(sbp
)) {
987 "XFS: correcting sb_features alignment problem");
988 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
989 sbp
->sb_bad_features2
= sbp
->sb_features2
;
990 update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
993 * Re-check for ATTR2 in case it was found in bad_features2
996 if (xfs_sb_version_hasattr2(&mp
->m_sb
))
997 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1002 * Check if sb_agblocks is aligned at stripe boundary
1003 * If sb_agblocks is NOT aligned turn off m_dalign since
1004 * allocator alignment is within an ag, therefore ag has
1005 * to be aligned at stripe boundary.
1007 error
= xfs_update_alignment(mp
, mfsi_flags
, &update_flags
);
1011 xfs_alloc_compute_maxlevels(mp
);
1012 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1013 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1014 xfs_ialloc_compute_maxlevels(mp
);
1016 xfs_set_maxicount(mp
);
1018 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
1021 * XFS uses the uuid from the superblock as the unique
1022 * identifier for fsid. We can not use the uuid from the volume
1023 * since a single partition filesystem is identical to a single
1024 * partition volume/filesystem.
1026 if ((mfsi_flags
& XFS_MFSI_SECOND
) == 0 &&
1027 (mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0) {
1028 if (xfs_uuid_mount(mp
)) {
1029 error
= XFS_ERROR(EINVAL
);
1036 * Set the minimum read and write sizes
1038 xfs_set_rw_sizes(mp
);
1041 * Set the inode cluster size.
1042 * This may still be overridden by the file system
1043 * block size if it is larger than the chosen cluster size.
1045 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1048 * Set inode alignment fields
1050 xfs_set_inoalignment(mp
);
1053 * Check that the data (and log if separate) are an ok size.
1055 error
= xfs_check_sizes(mp
, mfsi_flags
);
1060 * Initialize realtime fields in the mount structure
1062 error
= xfs_rtmount_init(mp
);
1064 cmn_err(CE_WARN
, "XFS: RT mount failed");
1069 * For client case we are done now
1071 if (mfsi_flags
& XFS_MFSI_CLIENT
) {
1076 * Copies the low order bits of the timestamp and the randomly
1077 * set "sequence" number out of a UUID.
1079 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1081 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1086 * Initialize the attribute manager's entries.
1088 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1091 * Initialize the precomputed transaction reservations values.
1096 * Allocate and initialize the per-ag data.
1098 init_rwsem(&mp
->m_peraglock
);
1100 kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
), KM_SLEEP
);
1102 mp
->m_maxagi
= xfs_initialize_perag(mp
, sbp
->sb_agcount
);
1105 * log's mount-time initialization. Perform 1st part recovery if needed
1107 if (likely(sbp
->sb_logblocks
> 0)) { /* check for volume case */
1108 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1109 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1110 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1112 cmn_err(CE_WARN
, "XFS: log mount failed");
1115 } else { /* No log has been defined */
1116 cmn_err(CE_WARN
, "XFS: no log defined");
1117 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW
, mp
);
1118 error
= XFS_ERROR(EFSCORRUPTED
);
1123 * Now the log is mounted, we know if it was an unclean shutdown or
1124 * not. If it was, with the first phase of recovery has completed, we
1125 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1126 * but they are recovered transactionally in the second recovery phase
1129 * Hence we can safely re-initialise incore superblock counters from
1130 * the per-ag data. These may not be correct if the filesystem was not
1131 * cleanly unmounted, so we need to wait for recovery to finish before
1134 * If the filesystem was cleanly unmounted, then we can trust the
1135 * values in the superblock to be correct and we don't need to do
1138 * If we are currently making the filesystem, the initialisation will
1139 * fail as the perag data is in an undefined state.
1142 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1143 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1144 !mp
->m_sb
.sb_inprogress
) {
1145 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1151 * Get and sanity-check the root inode.
1152 * Save the pointer to it in the mount structure.
1154 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1156 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1160 ASSERT(rip
!= NULL
);
1162 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1163 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1164 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1165 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1166 (unsigned long long)rip
->i_ino
);
1167 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1168 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1170 error
= XFS_ERROR(EFSCORRUPTED
);
1173 mp
->m_rootip
= rip
; /* save it */
1175 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1178 * Initialize realtime inode pointers in the mount structure
1180 error
= xfs_rtmount_inodes(mp
);
1183 * Free up the root inode.
1185 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1190 * If fs is not mounted readonly, then update the superblock changes.
1192 if (update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1193 error
= xfs_mount_log_sb(mp
, update_flags
);
1195 cmn_err(CE_WARN
, "XFS: failed to write sb changes");
1201 * Initialise the XFS quota management subsystem for this mount
1203 error
= XFS_QM_INIT(mp
, "amount
, "aflags
);
1208 * Finish recovering the file system. This part needed to be
1209 * delayed until after the root and real-time bitmap inodes
1210 * were consistently read in.
1212 error
= xfs_log_mount_finish(mp
, mfsi_flags
);
1214 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1219 * Complete the quota initialisation, post-log-replay component.
1221 error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
, mfsi_flags
);
1226 * Now we are mounted, reserve a small amount of unused space for
1227 * privileged transactions. This is needed so that transaction
1228 * space required for critical operations can dip into this pool
1229 * when at ENOSPC. This is needed for operations like create with
1230 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1231 * are not allowed to use this reserved space.
1233 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1234 * This may drive us straight to ENOSPC on mount, but that implies
1235 * we were already there on the last unmount. Warn if this occurs.
1237 resblks
= mp
->m_sb
.sb_dblocks
;
1238 do_div(resblks
, 20);
1239 resblks
= min_t(__uint64_t
, resblks
, 1024);
1240 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1242 cmn_err(CE_WARN
, "XFS: Unable to allocate reserve blocks. "
1243 "Continuing without a reserve pool.");
1249 * Free up the root inode.
1253 xfs_log_unmount_dealloc(mp
);
1255 for (agno
= 0; agno
< sbp
->sb_agcount
; agno
++)
1256 if (mp
->m_perag
[agno
].pagb_list
)
1257 kmem_free(mp
->m_perag
[agno
].pagb_list
,
1258 sizeof(xfs_perag_busy_t
) * XFS_PAGB_NUM_SLOTS
);
1259 kmem_free(mp
->m_perag
, sbp
->sb_agcount
* sizeof(xfs_perag_t
));
1264 xfs_uuid_unmount(mp
);
1272 * This flushes out the inodes,dquots and the superblock, unmounts the
1273 * log and makes sure that incore structures are freed.
1276 xfs_unmountfs(xfs_mount_t
*mp
, struct cred
*cr
)
1282 * We can potentially deadlock here if we have an inode cluster
1283 * that has been freed has it's buffer still pinned in memory because
1284 * the transaction is still sitting in a iclog. The stale inodes
1285 * on that buffer will have their flush locks held until the
1286 * transaction hits the disk and the callbacks run. the inode
1287 * flush takes the flush lock unconditionally and with nothing to
1288 * push out the iclog we will never get that unlocked. hence we
1289 * need to force the log first.
1291 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1294 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1297 * Flush out the log synchronously so that we know for sure
1298 * that nothing is pinned. This is important because bflush()
1299 * will skip pinned buffers.
1301 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1303 xfs_binval(mp
->m_ddev_targp
);
1304 if (mp
->m_rtdev_targp
) {
1305 xfs_binval(mp
->m_rtdev_targp
);
1309 * Unreserve any blocks we have so that when we unmount we don't account
1310 * the reserved free space as used. This is really only necessary for
1311 * lazy superblock counting because it trusts the incore superblock
1312 * counters to be aboslutely correct on clean unmount.
1314 * We don't bother correcting this elsewhere for lazy superblock
1315 * counting because on mount of an unclean filesystem we reconstruct the
1316 * correct counter value and this is irrelevant.
1318 * For non-lazy counter filesystems, this doesn't matter at all because
1319 * we only every apply deltas to the superblock and hence the incore
1320 * value does not matter....
1323 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1325 cmn_err(CE_WARN
, "XFS: Unable to free reserved block pool. "
1326 "Freespace may not be correct on next mount.");
1328 error
= xfs_log_sbcount(mp
, 1);
1330 cmn_err(CE_WARN
, "XFS: Unable to update superblock counters. "
1331 "Freespace may not be correct on next mount.");
1332 xfs_unmountfs_writesb(mp
);
1333 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1334 xfs_log_unmount(mp
); /* Done! No more fs ops. */
1339 * All inodes from this mount point should be freed.
1341 ASSERT(mp
->m_inodes
== NULL
);
1343 xfs_unmountfs_close(mp
, cr
);
1344 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0)
1345 xfs_uuid_unmount(mp
);
1347 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1348 xfs_errortag_clearall(mp
, 0);
1355 xfs_unmountfs_close(xfs_mount_t
*mp
, struct cred
*cr
)
1357 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1358 xfs_free_buftarg(mp
->m_logdev_targp
, 1);
1359 if (mp
->m_rtdev_targp
)
1360 xfs_free_buftarg(mp
->m_rtdev_targp
, 1);
1361 xfs_free_buftarg(mp
->m_ddev_targp
, 0);
1365 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1367 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1368 xfs_wait_buftarg(mp
->m_logdev_targp
);
1369 if (mp
->m_rtdev_targp
)
1370 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1371 xfs_wait_buftarg(mp
->m_ddev_targp
);
1375 xfs_fs_writable(xfs_mount_t
*mp
)
1377 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1378 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1384 * Called either periodically to keep the on disk superblock values
1385 * roughly up to date or from unmount to make sure the values are
1386 * correct on a clean unmount.
1388 * Note this code can be called during the process of freezing, so
1389 * we may need to use the transaction allocator which does not not
1390 * block when the transaction subsystem is in its frozen state.
1400 if (!xfs_fs_writable(mp
))
1403 xfs_icsb_sync_counters(mp
);
1406 * we don't need to do this if we are updating the superblock
1407 * counters on every modification.
1409 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1412 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
);
1413 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1414 XFS_DEFAULT_LOG_COUNT
);
1416 xfs_trans_cancel(tp
, 0);
1420 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1422 xfs_trans_set_sync(tp
);
1423 error
= xfs_trans_commit(tp
, 0);
1432 xfs_dsb_t
*sb
= XFS_BUF_TO_SBP(bp
);
1435 if (!(sb
->sb_flags
& XFS_SBF_READONLY
))
1436 sb
->sb_flags
|= XFS_SBF_READONLY
;
1438 version
= be16_to_cpu(sb
->sb_versionnum
);
1439 if ((version
& XFS_SB_VERSION_NUMBITS
) != XFS_SB_VERSION_4
||
1440 !(version
& XFS_SB_VERSION_SHAREDBIT
))
1441 version
|= XFS_SB_VERSION_SHAREDBIT
;
1442 sb
->sb_versionnum
= cpu_to_be16(version
);
1446 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1452 * skip superblock write if fs is read-only, or
1453 * if we are doing a forced umount.
1455 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1456 XFS_FORCED_SHUTDOWN(mp
))) {
1458 sbp
= xfs_getsb(mp
, 0);
1461 * mark shared-readonly if desired
1463 if (mp
->m_mk_sharedro
)
1464 xfs_mark_shared_ro(mp
, sbp
);
1466 XFS_BUF_UNDONE(sbp
);
1467 XFS_BUF_UNREAD(sbp
);
1468 XFS_BUF_UNDELAYWRITE(sbp
);
1470 XFS_BUF_UNASYNC(sbp
);
1471 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1472 xfsbdstrat(mp
, sbp
);
1473 error
= xfs_iowait(sbp
);
1475 xfs_ioerror_alert("xfs_unmountfs_writesb",
1476 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1477 if (error
&& mp
->m_mk_sharedro
)
1478 xfs_fs_cmn_err(CE_ALERT
, mp
, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1485 * xfs_mod_sb() can be used to copy arbitrary changes to the
1486 * in-core superblock into the superblock buffer to be logged.
1487 * It does not provide the higher level of locking that is
1488 * needed to protect the in-core superblock from concurrent
1492 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1504 bp
= xfs_trans_getsb(tp
, mp
, 0);
1505 first
= sizeof(xfs_sb_t
);
1508 /* translate/copy */
1510 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1512 /* find modified range */
1514 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1515 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1516 first
= xfs_sb_info
[f
].offset
;
1518 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1519 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1520 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1522 xfs_trans_log_buf(tp
, bp
, first
, last
);
1527 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1528 * a delta to a specified field in the in-core superblock. Simply
1529 * switch on the field indicated and apply the delta to that field.
1530 * Fields are not allowed to dip below zero, so if the delta would
1531 * do this do not apply it and return EINVAL.
1533 * The m_sb_lock must be held when this routine is called.
1536 xfs_mod_incore_sb_unlocked(
1538 xfs_sb_field_t field
,
1542 int scounter
; /* short counter for 32 bit fields */
1543 long long lcounter
; /* long counter for 64 bit fields */
1544 long long res_used
, rem
;
1547 * With the in-core superblock spin lock held, switch
1548 * on the indicated field. Apply the delta to the
1549 * proper field. If the fields value would dip below
1550 * 0, then do not apply the delta and return EINVAL.
1553 case XFS_SBS_ICOUNT
:
1554 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1558 return XFS_ERROR(EINVAL
);
1560 mp
->m_sb
.sb_icount
= lcounter
;
1563 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1567 return XFS_ERROR(EINVAL
);
1569 mp
->m_sb
.sb_ifree
= lcounter
;
1571 case XFS_SBS_FDBLOCKS
:
1572 lcounter
= (long long)
1573 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1574 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1576 if (delta
> 0) { /* Putting blocks back */
1577 if (res_used
> delta
) {
1578 mp
->m_resblks_avail
+= delta
;
1580 rem
= delta
- res_used
;
1581 mp
->m_resblks_avail
= mp
->m_resblks
;
1584 } else { /* Taking blocks away */
1589 * If were out of blocks, use any available reserved blocks if
1595 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1597 return XFS_ERROR(ENOSPC
);
1599 mp
->m_resblks_avail
= lcounter
;
1601 } else { /* not reserved */
1602 return XFS_ERROR(ENOSPC
);
1607 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1609 case XFS_SBS_FREXTENTS
:
1610 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1613 return XFS_ERROR(ENOSPC
);
1615 mp
->m_sb
.sb_frextents
= lcounter
;
1617 case XFS_SBS_DBLOCKS
:
1618 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1622 return XFS_ERROR(EINVAL
);
1624 mp
->m_sb
.sb_dblocks
= lcounter
;
1626 case XFS_SBS_AGCOUNT
:
1627 scounter
= mp
->m_sb
.sb_agcount
;
1631 return XFS_ERROR(EINVAL
);
1633 mp
->m_sb
.sb_agcount
= scounter
;
1635 case XFS_SBS_IMAX_PCT
:
1636 scounter
= mp
->m_sb
.sb_imax_pct
;
1640 return XFS_ERROR(EINVAL
);
1642 mp
->m_sb
.sb_imax_pct
= scounter
;
1644 case XFS_SBS_REXTSIZE
:
1645 scounter
= mp
->m_sb
.sb_rextsize
;
1649 return XFS_ERROR(EINVAL
);
1651 mp
->m_sb
.sb_rextsize
= scounter
;
1653 case XFS_SBS_RBMBLOCKS
:
1654 scounter
= mp
->m_sb
.sb_rbmblocks
;
1658 return XFS_ERROR(EINVAL
);
1660 mp
->m_sb
.sb_rbmblocks
= scounter
;
1662 case XFS_SBS_RBLOCKS
:
1663 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1667 return XFS_ERROR(EINVAL
);
1669 mp
->m_sb
.sb_rblocks
= lcounter
;
1671 case XFS_SBS_REXTENTS
:
1672 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1676 return XFS_ERROR(EINVAL
);
1678 mp
->m_sb
.sb_rextents
= lcounter
;
1680 case XFS_SBS_REXTSLOG
:
1681 scounter
= mp
->m_sb
.sb_rextslog
;
1685 return XFS_ERROR(EINVAL
);
1687 mp
->m_sb
.sb_rextslog
= scounter
;
1691 return XFS_ERROR(EINVAL
);
1696 * xfs_mod_incore_sb() is used to change a field in the in-core
1697 * superblock structure by the specified delta. This modification
1698 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1699 * routine to do the work.
1704 xfs_sb_field_t field
,
1710 /* check for per-cpu counters */
1712 #ifdef HAVE_PERCPU_SB
1713 case XFS_SBS_ICOUNT
:
1715 case XFS_SBS_FDBLOCKS
:
1716 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1717 status
= xfs_icsb_modify_counters(mp
, field
,
1724 spin_lock(&mp
->m_sb_lock
);
1725 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1726 spin_unlock(&mp
->m_sb_lock
);
1734 * xfs_mod_incore_sb_batch() is used to change more than one field
1735 * in the in-core superblock structure at a time. This modification
1736 * is protected by a lock internal to this module. The fields and
1737 * changes to those fields are specified in the array of xfs_mod_sb
1738 * structures passed in.
1740 * Either all of the specified deltas will be applied or none of
1741 * them will. If any modified field dips below 0, then all modifications
1742 * will be backed out and EINVAL will be returned.
1745 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1751 * Loop through the array of mod structures and apply each
1752 * individually. If any fail, then back out all those
1753 * which have already been applied. Do all of this within
1754 * the scope of the m_sb_lock so that all of the changes will
1757 spin_lock(&mp
->m_sb_lock
);
1759 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1761 * Apply the delta at index n. If it fails, break
1762 * from the loop so we'll fall into the undo loop
1765 switch (msbp
->msb_field
) {
1766 #ifdef HAVE_PERCPU_SB
1767 case XFS_SBS_ICOUNT
:
1769 case XFS_SBS_FDBLOCKS
:
1770 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1771 spin_unlock(&mp
->m_sb_lock
);
1772 status
= xfs_icsb_modify_counters(mp
,
1774 msbp
->msb_delta
, rsvd
);
1775 spin_lock(&mp
->m_sb_lock
);
1781 status
= xfs_mod_incore_sb_unlocked(mp
,
1783 msbp
->msb_delta
, rsvd
);
1793 * If we didn't complete the loop above, then back out
1794 * any changes made to the superblock. If you add code
1795 * between the loop above and here, make sure that you
1796 * preserve the value of status. Loop back until
1797 * we step below the beginning of the array. Make sure
1798 * we don't touch anything back there.
1802 while (msbp
>= msb
) {
1803 switch (msbp
->msb_field
) {
1804 #ifdef HAVE_PERCPU_SB
1805 case XFS_SBS_ICOUNT
:
1807 case XFS_SBS_FDBLOCKS
:
1808 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1809 spin_unlock(&mp
->m_sb_lock
);
1810 status
= xfs_icsb_modify_counters(mp
,
1814 spin_lock(&mp
->m_sb_lock
);
1820 status
= xfs_mod_incore_sb_unlocked(mp
,
1826 ASSERT(status
== 0);
1830 spin_unlock(&mp
->m_sb_lock
);
1835 * xfs_getsb() is called to obtain the buffer for the superblock.
1836 * The buffer is returned locked and read in from disk.
1837 * The buffer should be released with a call to xfs_brelse().
1839 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1840 * the superblock buffer if it can be locked without sleeping.
1841 * If it can't then we'll return NULL.
1850 ASSERT(mp
->m_sb_bp
!= NULL
);
1852 if (flags
& XFS_BUF_TRYLOCK
) {
1853 if (!XFS_BUF_CPSEMA(bp
)) {
1857 XFS_BUF_PSEMA(bp
, PRIBIO
);
1860 ASSERT(XFS_BUF_ISDONE(bp
));
1865 * Used to free the superblock along various error paths.
1874 * Use xfs_getsb() so that the buffer will be locked
1875 * when we call xfs_buf_relse().
1877 bp
= xfs_getsb(mp
, 0);
1878 XFS_BUF_UNMANAGE(bp
);
1884 * See if the UUID is unique among mounted XFS filesystems.
1885 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1891 if (uuid_is_nil(&mp
->m_sb
.sb_uuid
)) {
1893 "XFS: Filesystem %s has nil UUID - can't mount",
1897 if (!uuid_table_insert(&mp
->m_sb
.sb_uuid
)) {
1899 "XFS: Filesystem %s has duplicate UUID - can't mount",
1907 * Remove filesystem from the UUID table.
1913 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1917 * Used to log changes to the superblock unit and width fields which could
1918 * be altered by the mount options, as well as any potential sb_features2
1919 * fixup. Only the first superblock is updated.
1929 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1930 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
));
1932 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1933 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1934 XFS_DEFAULT_LOG_COUNT
);
1936 xfs_trans_cancel(tp
, 0);
1939 xfs_mod_sb(tp
, fields
);
1940 error
= xfs_trans_commit(tp
, 0);
1945 #ifdef HAVE_PERCPU_SB
1947 * Per-cpu incore superblock counters
1949 * Simple concept, difficult implementation
1951 * Basically, replace the incore superblock counters with a distributed per cpu
1952 * counter for contended fields (e.g. free block count).
1954 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1955 * hence needs to be accurately read when we are running low on space. Hence
1956 * there is a method to enable and disable the per-cpu counters based on how
1957 * much "stuff" is available in them.
1959 * Basically, a counter is enabled if there is enough free resource to justify
1960 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1961 * ENOSPC), then we disable the counters to synchronise all callers and
1962 * re-distribute the available resources.
1964 * If, once we redistributed the available resources, we still get a failure,
1965 * we disable the per-cpu counter and go through the slow path.
1967 * The slow path is the current xfs_mod_incore_sb() function. This means that
1968 * when we disable a per-cpu counter, we need to drain it's resources back to
1969 * the global superblock. We do this after disabling the counter to prevent
1970 * more threads from queueing up on the counter.
1972 * Essentially, this means that we still need a lock in the fast path to enable
1973 * synchronisation between the global counters and the per-cpu counters. This
1974 * is not a problem because the lock will be local to a CPU almost all the time
1975 * and have little contention except when we get to ENOSPC conditions.
1977 * Basically, this lock becomes a barrier that enables us to lock out the fast
1978 * path while we do things like enabling and disabling counters and
1979 * synchronising the counters.
1983 * 1. m_sb_lock before picking up per-cpu locks
1984 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1985 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1986 * 4. modifying per-cpu counters requires holding per-cpu lock
1987 * 5. modifying global counters requires holding m_sb_lock
1988 * 6. enabling or disabling a counter requires holding the m_sb_lock
1989 * and _none_ of the per-cpu locks.
1991 * Disabled counters are only ever re-enabled by a balance operation
1992 * that results in more free resources per CPU than a given threshold.
1993 * To ensure counters don't remain disabled, they are rebalanced when
1994 * the global resource goes above a higher threshold (i.e. some hysteresis
1995 * is present to prevent thrashing).
1998 #ifdef CONFIG_HOTPLUG_CPU
2000 * hot-plug CPU notifier support.
2002 * We need a notifier per filesystem as we need to be able to identify
2003 * the filesystem to balance the counters out. This is achieved by
2004 * having a notifier block embedded in the xfs_mount_t and doing pointer
2005 * magic to get the mount pointer from the notifier block address.
2008 xfs_icsb_cpu_notify(
2009 struct notifier_block
*nfb
,
2010 unsigned long action
,
2013 xfs_icsb_cnts_t
*cntp
;
2016 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2017 cntp
= (xfs_icsb_cnts_t
*)
2018 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2020 case CPU_UP_PREPARE
:
2021 case CPU_UP_PREPARE_FROZEN
:
2022 /* Easy Case - initialize the area and locks, and
2023 * then rebalance when online does everything else for us. */
2024 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2027 case CPU_ONLINE_FROZEN
:
2029 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0, 0);
2030 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0, 0);
2031 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0, 0);
2032 xfs_icsb_unlock(mp
);
2035 case CPU_DEAD_FROZEN
:
2036 /* Disable all the counters, then fold the dead cpu's
2037 * count into the total on the global superblock and
2038 * re-enable the counters. */
2040 spin_lock(&mp
->m_sb_lock
);
2041 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2042 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2043 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2045 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2046 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2047 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2049 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2051 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
,
2052 XFS_ICSB_SB_LOCKED
, 0);
2053 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
,
2054 XFS_ICSB_SB_LOCKED
, 0);
2055 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
,
2056 XFS_ICSB_SB_LOCKED
, 0);
2057 spin_unlock(&mp
->m_sb_lock
);
2058 xfs_icsb_unlock(mp
);
2064 #endif /* CONFIG_HOTPLUG_CPU */
2067 xfs_icsb_init_counters(
2070 xfs_icsb_cnts_t
*cntp
;
2073 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2074 if (mp
->m_sb_cnts
== NULL
)
2077 #ifdef CONFIG_HOTPLUG_CPU
2078 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2079 mp
->m_icsb_notifier
.priority
= 0;
2080 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2081 #endif /* CONFIG_HOTPLUG_CPU */
2083 for_each_online_cpu(i
) {
2084 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2085 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2088 mutex_init(&mp
->m_icsb_mutex
);
2091 * start with all counters disabled so that the
2092 * initial balance kicks us off correctly
2094 mp
->m_icsb_counters
= -1;
2099 xfs_icsb_reinit_counters(
2104 * start with all counters disabled so that the
2105 * initial balance kicks us off correctly
2107 mp
->m_icsb_counters
= -1;
2108 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0, 0);
2109 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0, 0);
2110 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0, 0);
2111 xfs_icsb_unlock(mp
);
2115 xfs_icsb_destroy_counters(
2118 if (mp
->m_sb_cnts
) {
2119 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2120 free_percpu(mp
->m_sb_cnts
);
2122 mutex_destroy(&mp
->m_icsb_mutex
);
2127 xfs_icsb_cnts_t
*icsbp
)
2129 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2135 xfs_icsb_unlock_cntr(
2136 xfs_icsb_cnts_t
*icsbp
)
2138 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2143 xfs_icsb_lock_all_counters(
2146 xfs_icsb_cnts_t
*cntp
;
2149 for_each_online_cpu(i
) {
2150 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2151 xfs_icsb_lock_cntr(cntp
);
2156 xfs_icsb_unlock_all_counters(
2159 xfs_icsb_cnts_t
*cntp
;
2162 for_each_online_cpu(i
) {
2163 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2164 xfs_icsb_unlock_cntr(cntp
);
2171 xfs_icsb_cnts_t
*cnt
,
2174 xfs_icsb_cnts_t
*cntp
;
2177 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2179 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2180 xfs_icsb_lock_all_counters(mp
);
2182 for_each_online_cpu(i
) {
2183 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2184 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2185 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2186 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2189 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2190 xfs_icsb_unlock_all_counters(mp
);
2194 xfs_icsb_counter_disabled(
2196 xfs_sb_field_t field
)
2198 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2199 return test_bit(field
, &mp
->m_icsb_counters
);
2203 xfs_icsb_disable_counter(
2205 xfs_sb_field_t field
)
2207 xfs_icsb_cnts_t cnt
;
2209 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2212 * If we are already disabled, then there is nothing to do
2213 * here. We check before locking all the counters to avoid
2214 * the expensive lock operation when being called in the
2215 * slow path and the counter is already disabled. This is
2216 * safe because the only time we set or clear this state is under
2219 if (xfs_icsb_counter_disabled(mp
, field
))
2222 xfs_icsb_lock_all_counters(mp
);
2223 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2224 /* drain back to superblock */
2226 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_SB_LOCKED
|XFS_ICSB_LAZY_COUNT
);
2228 case XFS_SBS_ICOUNT
:
2229 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2232 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2234 case XFS_SBS_FDBLOCKS
:
2235 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2242 xfs_icsb_unlock_all_counters(mp
);
2246 xfs_icsb_enable_counter(
2248 xfs_sb_field_t field
,
2252 xfs_icsb_cnts_t
*cntp
;
2255 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2257 xfs_icsb_lock_all_counters(mp
);
2258 for_each_online_cpu(i
) {
2259 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2261 case XFS_SBS_ICOUNT
:
2262 cntp
->icsb_icount
= count
+ resid
;
2265 cntp
->icsb_ifree
= count
+ resid
;
2267 case XFS_SBS_FDBLOCKS
:
2268 cntp
->icsb_fdblocks
= count
+ resid
;
2276 clear_bit(field
, &mp
->m_icsb_counters
);
2277 xfs_icsb_unlock_all_counters(mp
);
2281 xfs_icsb_sync_counters_flags(
2285 xfs_icsb_cnts_t cnt
;
2287 /* Pass 1: lock all counters */
2288 if ((flags
& XFS_ICSB_SB_LOCKED
) == 0)
2289 spin_lock(&mp
->m_sb_lock
);
2291 xfs_icsb_count(mp
, &cnt
, flags
);
2293 /* Step 3: update mp->m_sb fields */
2294 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2295 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2296 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2297 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2298 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2299 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2301 if ((flags
& XFS_ICSB_SB_LOCKED
) == 0)
2302 spin_unlock(&mp
->m_sb_lock
);
2306 * Accurate update of per-cpu counters to incore superblock
2309 xfs_icsb_sync_counters(
2312 xfs_icsb_sync_counters_flags(mp
, 0);
2316 * Balance and enable/disable counters as necessary.
2318 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2319 * chosen to be the same number as single on disk allocation chunk per CPU, and
2320 * free blocks is something far enough zero that we aren't going thrash when we
2321 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2322 * prevent looping endlessly when xfs_alloc_space asks for more than will
2323 * be distributed to a single CPU but each CPU has enough blocks to be
2326 * Note that we can be called when counters are already disabled.
2327 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2328 * prevent locking every per-cpu counter needlessly.
2331 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2332 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2333 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2335 xfs_icsb_balance_counter(
2337 xfs_sb_field_t field
,
2341 uint64_t count
, resid
;
2342 int weight
= num_online_cpus();
2343 uint64_t min
= (uint64_t)min_per_cpu
;
2345 if (!(flags
& XFS_ICSB_SB_LOCKED
))
2346 spin_lock(&mp
->m_sb_lock
);
2348 /* disable counter and sync counter */
2349 xfs_icsb_disable_counter(mp
, field
);
2351 /* update counters - first CPU gets residual*/
2353 case XFS_SBS_ICOUNT
:
2354 count
= mp
->m_sb
.sb_icount
;
2355 resid
= do_div(count
, weight
);
2356 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2360 count
= mp
->m_sb
.sb_ifree
;
2361 resid
= do_div(count
, weight
);
2362 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2365 case XFS_SBS_FDBLOCKS
:
2366 count
= mp
->m_sb
.sb_fdblocks
;
2367 resid
= do_div(count
, weight
);
2368 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2373 count
= resid
= 0; /* quiet, gcc */
2377 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2379 if (!(flags
& XFS_ICSB_SB_LOCKED
))
2380 spin_unlock(&mp
->m_sb_lock
);
2384 xfs_icsb_modify_counters(
2386 xfs_sb_field_t field
,
2390 xfs_icsb_cnts_t
*icsbp
;
2391 long long lcounter
; /* long counter for 64 bit fields */
2397 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
);
2400 * if the counter is disabled, go to slow path
2402 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2404 xfs_icsb_lock_cntr(icsbp
);
2405 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2406 xfs_icsb_unlock_cntr(icsbp
);
2411 case XFS_SBS_ICOUNT
:
2412 lcounter
= icsbp
->icsb_icount
;
2414 if (unlikely(lcounter
< 0))
2415 goto balance_counter
;
2416 icsbp
->icsb_icount
= lcounter
;
2420 lcounter
= icsbp
->icsb_ifree
;
2422 if (unlikely(lcounter
< 0))
2423 goto balance_counter
;
2424 icsbp
->icsb_ifree
= lcounter
;
2427 case XFS_SBS_FDBLOCKS
:
2428 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2430 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2432 if (unlikely(lcounter
< 0))
2433 goto balance_counter
;
2434 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2440 xfs_icsb_unlock_cntr(icsbp
);
2448 * serialise with a mutex so we don't burn lots of cpu on
2449 * the superblock lock. We still need to hold the superblock
2450 * lock, however, when we modify the global structures.
2455 * Now running atomically.
2457 * If the counter is enabled, someone has beaten us to rebalancing.
2458 * Drop the lock and try again in the fast path....
2460 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2461 xfs_icsb_unlock(mp
);
2466 * The counter is currently disabled. Because we are
2467 * running atomically here, we know a rebalance cannot
2468 * be in progress. Hence we can go straight to operating
2469 * on the global superblock. We do not call xfs_mod_incore_sb()
2470 * here even though we need to get the m_sb_lock. Doing so
2471 * will cause us to re-enter this function and deadlock.
2472 * Hence we get the m_sb_lock ourselves and then call
2473 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2474 * directly on the global counters.
2476 spin_lock(&mp
->m_sb_lock
);
2477 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2478 spin_unlock(&mp
->m_sb_lock
);
2481 * Now that we've modified the global superblock, we
2482 * may be able to re-enable the distributed counters
2483 * (e.g. lots of space just got freed). After that
2487 xfs_icsb_balance_counter(mp
, field
, 0, 0);
2488 xfs_icsb_unlock(mp
);
2492 xfs_icsb_unlock_cntr(icsbp
);
2496 * We may have multiple threads here if multiple per-cpu
2497 * counters run dry at the same time. This will mean we can
2498 * do more balances than strictly necessary but it is not
2499 * the common slowpath case.
2504 * running atomically.
2506 * This will leave the counter in the correct state for future
2507 * accesses. After the rebalance, we simply try again and our retry
2508 * will either succeed through the fast path or slow path without
2509 * another balance operation being required.
2511 xfs_icsb_balance_counter(mp
, field
, 0, delta
);
2512 xfs_icsb_unlock(mp
);