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_destroy_counters(xfs_mount_t
*);
55 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
57 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_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) do { } while (0)
67 #define xfs_icsb_balance_counter_locked(mp, a, b) 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 kmem_free(mp
->m_perag
);
167 spinlock_destroy(&mp
->m_ail_lock
);
168 spinlock_destroy(&mp
->m_sb_lock
);
169 mutex_destroy(&mp
->m_ilock
);
170 mutex_destroy(&mp
->m_growlock
);
174 if (mp
->m_fsname
!= NULL
)
175 kmem_free(mp
->m_fsname
);
176 if (mp
->m_rtname
!= NULL
)
177 kmem_free(mp
->m_rtname
);
178 if (mp
->m_logname
!= NULL
)
179 kmem_free(mp
->m_logname
);
181 xfs_icsb_destroy_counters(mp
);
185 * Check size of device based on the (data/realtime) block count.
186 * Note: this check is used by the growfs code as well as mount.
189 xfs_sb_validate_fsb_count(
193 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
194 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
196 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
197 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
199 #else /* Limited by UINT_MAX of sectors */
200 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
207 * Check the validity of the SB found.
210 xfs_mount_validate_sb(
216 * If the log device and data device have the
217 * same device number, the log is internal.
218 * Consequently, the sb_logstart should be non-zero. If
219 * we have a zero sb_logstart in this case, we may be trying to mount
220 * a volume filesystem in a non-volume manner.
222 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
223 xfs_fs_mount_cmn_err(flags
, "bad magic number");
224 return XFS_ERROR(EWRONGFS
);
227 if (!xfs_sb_good_version(sbp
)) {
228 xfs_fs_mount_cmn_err(flags
, "bad version");
229 return XFS_ERROR(EWRONGFS
);
233 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
234 xfs_fs_mount_cmn_err(flags
,
235 "filesystem is marked as having an external log; "
236 "specify logdev on the\nmount command line.");
237 return XFS_ERROR(EINVAL
);
241 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
242 xfs_fs_mount_cmn_err(flags
,
243 "filesystem is marked as having an internal log; "
244 "do not specify logdev on\nthe mount command line.");
245 return XFS_ERROR(EINVAL
);
249 * More sanity checking. These were stolen directly from
253 sbp
->sb_agcount
<= 0 ||
254 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
255 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
256 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
257 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
258 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
259 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
260 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
261 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
262 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
263 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
264 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
265 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
266 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
267 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
268 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
269 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
270 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
271 return XFS_ERROR(EFSCORRUPTED
);
275 * Sanity check AG count, size fields against data size field
278 sbp
->sb_dblocks
== 0 ||
280 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
281 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
282 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
283 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
284 return XFS_ERROR(EFSCORRUPTED
);
287 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
288 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
289 xfs_fs_mount_cmn_err(flags
,
290 "file system too large to be mounted on this system.");
291 return XFS_ERROR(E2BIG
);
294 if (unlikely(sbp
->sb_inprogress
)) {
295 xfs_fs_mount_cmn_err(flags
, "file system busy");
296 return XFS_ERROR(EFSCORRUPTED
);
300 * Version 1 directory format has never worked on Linux.
302 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
303 xfs_fs_mount_cmn_err(flags
,
304 "file system using version 1 directory format");
305 return XFS_ERROR(ENOSYS
);
309 * Until this is fixed only page-sized or smaller data blocks work.
311 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
312 xfs_fs_mount_cmn_err(flags
,
313 "file system with blocksize %d bytes",
315 xfs_fs_mount_cmn_err(flags
,
316 "only pagesize (%ld) or less will currently work.",
318 return XFS_ERROR(ENOSYS
);
325 xfs_initialize_perag_icache(
328 if (!pag
->pag_ici_init
) {
329 rwlock_init(&pag
->pag_ici_lock
);
330 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
331 pag
->pag_ici_init
= 1;
336 xfs_initialize_perag(
338 xfs_agnumber_t agcount
)
340 xfs_agnumber_t index
, max_metadata
;
344 xfs_sb_t
*sbp
= &mp
->m_sb
;
345 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
347 /* Check to see if the filesystem can overflow 32 bit inodes */
348 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
349 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
351 /* Clear the mount flag if no inode can overflow 32 bits
352 * on this filesystem, or if specifically requested..
354 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> max_inum
) {
355 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
357 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
360 /* If we can overflow then setup the ag headers accordingly */
361 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
362 /* Calculate how much should be reserved for inodes to
363 * meet the max inode percentage.
365 if (mp
->m_maxicount
) {
368 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
370 icount
+= sbp
->sb_agblocks
- 1;
371 do_div(icount
, sbp
->sb_agblocks
);
372 max_metadata
= icount
;
374 max_metadata
= agcount
;
376 for (index
= 0; index
< agcount
; index
++) {
377 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
378 if (ino
> max_inum
) {
383 /* This ag is preferred for inodes */
384 pag
= &mp
->m_perag
[index
];
385 pag
->pagi_inodeok
= 1;
386 if (index
< max_metadata
)
387 pag
->pagf_metadata
= 1;
388 xfs_initialize_perag_icache(pag
);
391 /* Setup default behavior for smaller filesystems */
392 for (index
= 0; index
< agcount
; index
++) {
393 pag
= &mp
->m_perag
[index
];
394 pag
->pagi_inodeok
= 1;
395 xfs_initialize_perag_icache(pag
);
406 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
407 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
408 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
409 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
410 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
411 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
412 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
413 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
414 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
415 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
416 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
417 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
418 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
419 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
420 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
421 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
422 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
423 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
424 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
425 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
426 to
->sb_blocklog
= from
->sb_blocklog
;
427 to
->sb_sectlog
= from
->sb_sectlog
;
428 to
->sb_inodelog
= from
->sb_inodelog
;
429 to
->sb_inopblog
= from
->sb_inopblog
;
430 to
->sb_agblklog
= from
->sb_agblklog
;
431 to
->sb_rextslog
= from
->sb_rextslog
;
432 to
->sb_inprogress
= from
->sb_inprogress
;
433 to
->sb_imax_pct
= from
->sb_imax_pct
;
434 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
435 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
436 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
437 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
438 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
439 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
440 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
441 to
->sb_flags
= from
->sb_flags
;
442 to
->sb_shared_vn
= from
->sb_shared_vn
;
443 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
444 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
445 to
->sb_width
= be32_to_cpu(from
->sb_width
);
446 to
->sb_dirblklog
= from
->sb_dirblklog
;
447 to
->sb_logsectlog
= from
->sb_logsectlog
;
448 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
449 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
450 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
451 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
455 * Copy in core superblock to ondisk one.
457 * The fields argument is mask of superblock fields to copy.
465 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
466 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
476 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
477 first
= xfs_sb_info
[f
].offset
;
478 size
= xfs_sb_info
[f
+ 1].offset
- first
;
480 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
482 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
483 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
487 *(__be16
*)(to_ptr
+ first
) =
488 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
491 *(__be32
*)(to_ptr
+ first
) =
492 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
495 *(__be64
*)(to_ptr
+ first
) =
496 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
503 fields
&= ~(1LL << f
);
510 * Does the initial read of the superblock.
513 xfs_readsb(xfs_mount_t
*mp
, int flags
)
515 unsigned int sector_size
;
516 unsigned int extra_flags
;
520 ASSERT(mp
->m_sb_bp
== NULL
);
521 ASSERT(mp
->m_ddev_targp
!= NULL
);
524 * Allocate a (locked) buffer to hold the superblock.
525 * This will be kept around at all times to optimize
526 * access to the superblock.
528 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
529 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
531 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
532 BTOBB(sector_size
), extra_flags
);
533 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
534 xfs_fs_mount_cmn_err(flags
, "SB read failed");
535 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
538 ASSERT(XFS_BUF_ISBUSY(bp
));
539 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
542 * Initialize the mount structure from the superblock.
543 * But first do some basic consistency checking.
545 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
547 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
549 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
554 * We must be able to do sector-sized and sector-aligned IO.
556 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
557 xfs_fs_mount_cmn_err(flags
,
558 "device supports only %u byte sectors (not %u)",
559 sector_size
, mp
->m_sb
.sb_sectsize
);
565 * If device sector size is smaller than the superblock size,
566 * re-read the superblock so the buffer is correctly sized.
568 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
569 XFS_BUF_UNMANAGE(bp
);
571 sector_size
= mp
->m_sb
.sb_sectsize
;
572 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
573 BTOBB(sector_size
), extra_flags
);
574 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
575 xfs_fs_mount_cmn_err(flags
, "SB re-read failed");
576 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
579 ASSERT(XFS_BUF_ISBUSY(bp
));
580 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
583 /* Initialize per-cpu counters */
584 xfs_icsb_reinit_counters(mp
);
588 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
593 XFS_BUF_UNMANAGE(bp
);
603 * Mount initialization code establishing various mount
604 * fields from the superblock associated with the given
608 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
612 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
613 spin_lock_init(&mp
->m_agirotor_lock
);
614 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
615 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
616 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
617 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
618 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
619 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
620 mp
->m_litino
= sbp
->sb_inodesize
-
621 ((uint
)sizeof(xfs_dinode_core_t
) + (uint
)sizeof(xfs_agino_t
));
622 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
623 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
624 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
625 INIT_LIST_HEAD(&mp
->m_del_inodes
);
628 * Setup for attributes, in case they get created.
629 * This value is for inodes getting attributes for the first time,
630 * the per-inode value is for old attribute values.
632 ASSERT(sbp
->sb_inodesize
>= 256 && sbp
->sb_inodesize
<= 2048);
633 switch (sbp
->sb_inodesize
) {
635 mp
->m_attroffset
= XFS_LITINO(mp
) -
636 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
641 mp
->m_attroffset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
646 ASSERT(mp
->m_attroffset
< XFS_LITINO(mp
));
648 for (i
= 0; i
< 2; i
++) {
649 mp
->m_alloc_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
651 mp
->m_alloc_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
654 for (i
= 0; i
< 2; i
++) {
655 mp
->m_bmap_dmxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
657 mp
->m_bmap_dmnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
660 for (i
= 0; i
< 2; i
++) {
661 mp
->m_inobt_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
663 mp
->m_inobt_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
667 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
668 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
670 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
674 * xfs_initialize_perag_data
676 * Read in each per-ag structure so we can count up the number of
677 * allocated inodes, free inodes and used filesystem blocks as this
678 * information is no longer persistent in the superblock. Once we have
679 * this information, write it into the in-core superblock structure.
682 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
684 xfs_agnumber_t index
;
686 xfs_sb_t
*sbp
= &mp
->m_sb
;
690 uint64_t bfreelst
= 0;
694 for (index
= 0; index
< agcount
; index
++) {
696 * read the agf, then the agi. This gets us
697 * all the inforamtion we need and populates the
698 * per-ag structures for us.
700 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
704 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
707 pag
= &mp
->m_perag
[index
];
708 ifree
+= pag
->pagi_freecount
;
709 ialloc
+= pag
->pagi_count
;
710 bfree
+= pag
->pagf_freeblks
;
711 bfreelst
+= pag
->pagf_flcount
;
712 btree
+= pag
->pagf_btreeblks
;
715 * Overwrite incore superblock counters with just-read data
717 spin_lock(&mp
->m_sb_lock
);
718 sbp
->sb_ifree
= ifree
;
719 sbp
->sb_icount
= ialloc
;
720 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
721 spin_unlock(&mp
->m_sb_lock
);
723 /* Fixup the per-cpu counters as well. */
724 xfs_icsb_reinit_counters(mp
);
730 * Update alignment values based on mount options and sb values
733 xfs_update_alignment(xfs_mount_t
*mp
, int mfsi_flags
, __uint64_t
*update_flags
)
735 xfs_sb_t
*sbp
= &(mp
->m_sb
);
737 if (mp
->m_dalign
&& !(mfsi_flags
& XFS_MFSI_SECOND
)) {
739 * If stripe unit and stripe width are not multiples
740 * of the fs blocksize turn off alignment.
742 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
743 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
744 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
746 "XFS: alignment check 1 failed");
747 return XFS_ERROR(EINVAL
);
749 mp
->m_dalign
= mp
->m_swidth
= 0;
752 * Convert the stripe unit and width to FSBs.
754 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
755 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
756 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
757 return XFS_ERROR(EINVAL
);
759 xfs_fs_cmn_err(CE_WARN
, mp
,
760 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
761 mp
->m_dalign
, mp
->m_swidth
,
766 } else if (mp
->m_dalign
) {
767 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
769 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
770 xfs_fs_cmn_err(CE_WARN
, mp
,
771 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
774 return XFS_ERROR(EINVAL
);
781 * Update superblock with new values
784 if (xfs_sb_version_hasdalign(sbp
)) {
785 if (sbp
->sb_unit
!= mp
->m_dalign
) {
786 sbp
->sb_unit
= mp
->m_dalign
;
787 *update_flags
|= XFS_SB_UNIT
;
789 if (sbp
->sb_width
!= mp
->m_swidth
) {
790 sbp
->sb_width
= mp
->m_swidth
;
791 *update_flags
|= XFS_SB_WIDTH
;
794 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
795 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
796 mp
->m_dalign
= sbp
->sb_unit
;
797 mp
->m_swidth
= sbp
->sb_width
;
804 * Set the maximum inode count for this filesystem
807 xfs_set_maxicount(xfs_mount_t
*mp
)
809 xfs_sb_t
*sbp
= &(mp
->m_sb
);
812 if (sbp
->sb_imax_pct
) {
814 * Make sure the maximum inode count is a multiple
815 * of the units we allocate inodes in.
817 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
819 do_div(icount
, mp
->m_ialloc_blks
);
820 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
828 * Set the default minimum read and write sizes unless
829 * already specified in a mount option.
830 * We use smaller I/O sizes when the file system
831 * is being used for NFS service (wsync mount option).
834 xfs_set_rw_sizes(xfs_mount_t
*mp
)
836 xfs_sb_t
*sbp
= &(mp
->m_sb
);
837 int readio_log
, writeio_log
;
839 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
840 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
841 readio_log
= XFS_WSYNC_READIO_LOG
;
842 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
844 readio_log
= XFS_READIO_LOG_LARGE
;
845 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
848 readio_log
= mp
->m_readio_log
;
849 writeio_log
= mp
->m_writeio_log
;
852 if (sbp
->sb_blocklog
> readio_log
) {
853 mp
->m_readio_log
= sbp
->sb_blocklog
;
855 mp
->m_readio_log
= readio_log
;
857 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
858 if (sbp
->sb_blocklog
> writeio_log
) {
859 mp
->m_writeio_log
= sbp
->sb_blocklog
;
861 mp
->m_writeio_log
= writeio_log
;
863 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
867 * Set whether we're using inode alignment.
870 xfs_set_inoalignment(xfs_mount_t
*mp
)
872 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
873 mp
->m_sb
.sb_inoalignmt
>=
874 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
875 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
877 mp
->m_inoalign_mask
= 0;
879 * If we are using stripe alignment, check whether
880 * the stripe unit is a multiple of the inode alignment
882 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
883 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
884 mp
->m_sinoalign
= mp
->m_dalign
;
890 * Check that the data (and log if separate) are an ok size.
893 xfs_check_sizes(xfs_mount_t
*mp
, int mfsi_flags
)
899 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
900 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
901 cmn_err(CE_WARN
, "XFS: size check 1 failed");
902 return XFS_ERROR(E2BIG
);
904 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
905 d
- XFS_FSS_TO_BB(mp
, 1),
906 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
910 cmn_err(CE_WARN
, "XFS: size check 2 failed");
912 error
= XFS_ERROR(E2BIG
);
916 if (((mfsi_flags
& XFS_MFSI_CLIENT
) == 0) &&
917 mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
918 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
919 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
920 cmn_err(CE_WARN
, "XFS: size check 3 failed");
921 return XFS_ERROR(E2BIG
);
923 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
924 d
- XFS_FSB_TO_BB(mp
, 1),
925 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
929 cmn_err(CE_WARN
, "XFS: size check 3 failed");
931 error
= XFS_ERROR(E2BIG
);
941 * This function does the following on an initial mount of a file system:
942 * - reads the superblock from disk and init the mount struct
943 * - if we're a 32-bit kernel, do a size check on the superblock
944 * so we don't mount terabyte filesystems
945 * - init mount struct realtime fields
946 * - allocate inode hash table for fs
947 * - init directory manager
948 * - perform recovery and init the log manager
955 xfs_sb_t
*sbp
= &(mp
->m_sb
);
958 __int64_t update_flags
= 0LL;
959 uint quotamount
, quotaflags
;
961 int uuid_mounted
= 0;
964 xfs_mount_common(mp
, sbp
);
967 * Check for a mismatched features2 values. Older kernels
968 * read & wrote into the wrong sb offset for sb_features2
969 * on some platforms due to xfs_sb_t not being 64bit size aligned
970 * when sb_features2 was added, which made older superblock
971 * reading/writing routines swap it as a 64-bit value.
973 * For backwards compatibility, we make both slots equal.
975 * If we detect a mismatched field, we OR the set bits into the
976 * existing features2 field in case it has already been modified; we
977 * don't want to lose any features. We then update the bad location
978 * with the ORed value so that older kernels will see any features2
979 * flags, and mark the two fields as needing updates once the
980 * transaction subsystem is online.
982 if (xfs_sb_has_mismatched_features2(sbp
)) {
984 "XFS: correcting sb_features alignment problem");
985 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
986 sbp
->sb_bad_features2
= sbp
->sb_features2
;
987 update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
990 * Re-check for ATTR2 in case it was found in bad_features2
993 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
994 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
995 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
998 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
999 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
1000 xfs_sb_version_removeattr2(&mp
->m_sb
);
1001 update_flags
|= XFS_SB_FEATURES2
;
1003 /* update sb_versionnum for the clearing of the morebits */
1004 if (!sbp
->sb_features2
)
1005 update_flags
|= XFS_SB_VERSIONNUM
;
1009 * Check if sb_agblocks is aligned at stripe boundary
1010 * If sb_agblocks is NOT aligned turn off m_dalign since
1011 * allocator alignment is within an ag, therefore ag has
1012 * to be aligned at stripe boundary.
1014 error
= xfs_update_alignment(mp
, mfsi_flags
, &update_flags
);
1018 xfs_alloc_compute_maxlevels(mp
);
1019 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
1020 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
1021 xfs_ialloc_compute_maxlevels(mp
);
1023 xfs_set_maxicount(mp
);
1025 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
1028 * XFS uses the uuid from the superblock as the unique
1029 * identifier for fsid. We can not use the uuid from the volume
1030 * since a single partition filesystem is identical to a single
1031 * partition volume/filesystem.
1033 if ((mfsi_flags
& XFS_MFSI_SECOND
) == 0 &&
1034 (mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0) {
1035 if (xfs_uuid_mount(mp
)) {
1036 error
= XFS_ERROR(EINVAL
);
1043 * Set the minimum read and write sizes
1045 xfs_set_rw_sizes(mp
);
1048 * Set the inode cluster size.
1049 * This may still be overridden by the file system
1050 * block size if it is larger than the chosen cluster size.
1052 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
1055 * Set inode alignment fields
1057 xfs_set_inoalignment(mp
);
1060 * Check that the data (and log if separate) are an ok size.
1062 error
= xfs_check_sizes(mp
, mfsi_flags
);
1067 * Initialize realtime fields in the mount structure
1069 error
= xfs_rtmount_init(mp
);
1071 cmn_err(CE_WARN
, "XFS: RT mount failed");
1076 * For client case we are done now
1078 if (mfsi_flags
& XFS_MFSI_CLIENT
) {
1083 * Copies the low order bits of the timestamp and the randomly
1084 * set "sequence" number out of a UUID.
1086 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1088 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1093 * Initialize the attribute manager's entries.
1095 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1098 * Initialize the precomputed transaction reservations values.
1103 * Allocate and initialize the per-ag data.
1105 init_rwsem(&mp
->m_peraglock
);
1107 kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
), KM_SLEEP
);
1109 mp
->m_maxagi
= xfs_initialize_perag(mp
, sbp
->sb_agcount
);
1112 * log's mount-time initialization. Perform 1st part recovery if needed
1114 if (likely(sbp
->sb_logblocks
> 0)) { /* check for volume case */
1115 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1116 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1117 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1119 cmn_err(CE_WARN
, "XFS: log mount failed");
1122 } else { /* No log has been defined */
1123 cmn_err(CE_WARN
, "XFS: no log defined");
1124 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW
, mp
);
1125 error
= XFS_ERROR(EFSCORRUPTED
);
1130 * Now the log is mounted, we know if it was an unclean shutdown or
1131 * not. If it was, with the first phase of recovery has completed, we
1132 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1133 * but they are recovered transactionally in the second recovery phase
1136 * Hence we can safely re-initialise incore superblock counters from
1137 * the per-ag data. These may not be correct if the filesystem was not
1138 * cleanly unmounted, so we need to wait for recovery to finish before
1141 * If the filesystem was cleanly unmounted, then we can trust the
1142 * values in the superblock to be correct and we don't need to do
1145 * If we are currently making the filesystem, the initialisation will
1146 * fail as the perag data is in an undefined state.
1149 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1150 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1151 !mp
->m_sb
.sb_inprogress
) {
1152 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1158 * Get and sanity-check the root inode.
1159 * Save the pointer to it in the mount structure.
1161 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1163 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1167 ASSERT(rip
!= NULL
);
1169 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1170 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1171 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1172 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1173 (unsigned long long)rip
->i_ino
);
1174 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1175 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1177 error
= XFS_ERROR(EFSCORRUPTED
);
1180 mp
->m_rootip
= rip
; /* save it */
1182 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1185 * Initialize realtime inode pointers in the mount structure
1187 error
= xfs_rtmount_inodes(mp
);
1190 * Free up the root inode.
1192 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1197 * If fs is not mounted readonly, then update the superblock changes.
1199 if (update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1200 error
= xfs_mount_log_sb(mp
, update_flags
);
1202 cmn_err(CE_WARN
, "XFS: failed to write sb changes");
1208 * Initialise the XFS quota management subsystem for this mount
1210 error
= XFS_QM_INIT(mp
, "amount
, "aflags
);
1215 * Finish recovering the file system. This part needed to be
1216 * delayed until after the root and real-time bitmap inodes
1217 * were consistently read in.
1219 error
= xfs_log_mount_finish(mp
, mfsi_flags
);
1221 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1226 * Complete the quota initialisation, post-log-replay component.
1228 error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
, mfsi_flags
);
1233 * Now we are mounted, reserve a small amount of unused space for
1234 * privileged transactions. This is needed so that transaction
1235 * space required for critical operations can dip into this pool
1236 * when at ENOSPC. This is needed for operations like create with
1237 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1238 * are not allowed to use this reserved space.
1240 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1241 * This may drive us straight to ENOSPC on mount, but that implies
1242 * we were already there on the last unmount. Warn if this occurs.
1244 resblks
= mp
->m_sb
.sb_dblocks
;
1245 do_div(resblks
, 20);
1246 resblks
= min_t(__uint64_t
, resblks
, 1024);
1247 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1249 cmn_err(CE_WARN
, "XFS: Unable to allocate reserve blocks. "
1250 "Continuing without a reserve pool.");
1256 * Free up the root inode.
1260 xfs_log_unmount_dealloc(mp
);
1262 for (agno
= 0; agno
< sbp
->sb_agcount
; agno
++)
1263 if (mp
->m_perag
[agno
].pagb_list
)
1264 kmem_free(mp
->m_perag
[agno
].pagb_list
);
1265 kmem_free(mp
->m_perag
);
1270 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1278 * This flushes out the inodes,dquots and the superblock, unmounts the
1279 * log and makes sure that incore structures are freed.
1282 xfs_unmountfs(xfs_mount_t
*mp
, struct cred
*cr
)
1288 * We can potentially deadlock here if we have an inode cluster
1289 * that has been freed has it's buffer still pinned in memory because
1290 * the transaction is still sitting in a iclog. The stale inodes
1291 * on that buffer will have their flush locks held until the
1292 * transaction hits the disk and the callbacks run. the inode
1293 * flush takes the flush lock unconditionally and with nothing to
1294 * push out the iclog we will never get that unlocked. hence we
1295 * need to force the log first.
1297 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1300 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1303 * Flush out the log synchronously so that we know for sure
1304 * that nothing is pinned. This is important because bflush()
1305 * will skip pinned buffers.
1307 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1309 xfs_binval(mp
->m_ddev_targp
);
1310 if (mp
->m_rtdev_targp
) {
1311 xfs_binval(mp
->m_rtdev_targp
);
1315 * Unreserve any blocks we have so that when we unmount we don't account
1316 * the reserved free space as used. This is really only necessary for
1317 * lazy superblock counting because it trusts the incore superblock
1318 * counters to be aboslutely correct on clean unmount.
1320 * We don't bother correcting this elsewhere for lazy superblock
1321 * counting because on mount of an unclean filesystem we reconstruct the
1322 * correct counter value and this is irrelevant.
1324 * For non-lazy counter filesystems, this doesn't matter at all because
1325 * we only every apply deltas to the superblock and hence the incore
1326 * value does not matter....
1329 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1331 cmn_err(CE_WARN
, "XFS: Unable to free reserved block pool. "
1332 "Freespace may not be correct on next mount.");
1334 error
= xfs_log_sbcount(mp
, 1);
1336 cmn_err(CE_WARN
, "XFS: Unable to update superblock counters. "
1337 "Freespace may not be correct on next mount.");
1338 xfs_unmountfs_writesb(mp
);
1339 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1340 xfs_log_unmount(mp
); /* Done! No more fs ops. */
1345 * All inodes from this mount point should be freed.
1347 ASSERT(mp
->m_inodes
== NULL
);
1349 xfs_unmountfs_close(mp
, cr
);
1350 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0)
1351 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1353 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1354 xfs_errortag_clearall(mp
, 0);
1361 xfs_unmountfs_close(xfs_mount_t
*mp
, struct cred
*cr
)
1363 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1364 xfs_free_buftarg(mp
->m_logdev_targp
, 1);
1365 if (mp
->m_rtdev_targp
)
1366 xfs_free_buftarg(mp
->m_rtdev_targp
, 1);
1367 xfs_free_buftarg(mp
->m_ddev_targp
, 0);
1371 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1373 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1374 xfs_wait_buftarg(mp
->m_logdev_targp
);
1375 if (mp
->m_rtdev_targp
)
1376 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1377 xfs_wait_buftarg(mp
->m_ddev_targp
);
1381 xfs_fs_writable(xfs_mount_t
*mp
)
1383 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1384 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1390 * Called either periodically to keep the on disk superblock values
1391 * roughly up to date or from unmount to make sure the values are
1392 * correct on a clean unmount.
1394 * Note this code can be called during the process of freezing, so
1395 * we may need to use the transaction allocator which does not not
1396 * block when the transaction subsystem is in its frozen state.
1406 if (!xfs_fs_writable(mp
))
1409 xfs_icsb_sync_counters(mp
, 0);
1412 * we don't need to do this if we are updating the superblock
1413 * counters on every modification.
1415 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1418 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
);
1419 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1420 XFS_DEFAULT_LOG_COUNT
);
1422 xfs_trans_cancel(tp
, 0);
1426 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1428 xfs_trans_set_sync(tp
);
1429 error
= xfs_trans_commit(tp
, 0);
1438 xfs_dsb_t
*sb
= XFS_BUF_TO_SBP(bp
);
1441 if (!(sb
->sb_flags
& XFS_SBF_READONLY
))
1442 sb
->sb_flags
|= XFS_SBF_READONLY
;
1444 version
= be16_to_cpu(sb
->sb_versionnum
);
1445 if ((version
& XFS_SB_VERSION_NUMBITS
) != XFS_SB_VERSION_4
||
1446 !(version
& XFS_SB_VERSION_SHAREDBIT
))
1447 version
|= XFS_SB_VERSION_SHAREDBIT
;
1448 sb
->sb_versionnum
= cpu_to_be16(version
);
1452 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1458 * skip superblock write if fs is read-only, or
1459 * if we are doing a forced umount.
1461 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1462 XFS_FORCED_SHUTDOWN(mp
))) {
1464 sbp
= xfs_getsb(mp
, 0);
1467 * mark shared-readonly if desired
1469 if (mp
->m_mk_sharedro
)
1470 xfs_mark_shared_ro(mp
, sbp
);
1472 XFS_BUF_UNDONE(sbp
);
1473 XFS_BUF_UNREAD(sbp
);
1474 XFS_BUF_UNDELAYWRITE(sbp
);
1476 XFS_BUF_UNASYNC(sbp
);
1477 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1478 xfsbdstrat(mp
, sbp
);
1479 error
= xfs_iowait(sbp
);
1481 xfs_ioerror_alert("xfs_unmountfs_writesb",
1482 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1483 if (error
&& mp
->m_mk_sharedro
)
1484 xfs_fs_cmn_err(CE_ALERT
, mp
, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1491 * xfs_mod_sb() can be used to copy arbitrary changes to the
1492 * in-core superblock into the superblock buffer to be logged.
1493 * It does not provide the higher level of locking that is
1494 * needed to protect the in-core superblock from concurrent
1498 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1510 bp
= xfs_trans_getsb(tp
, mp
, 0);
1511 first
= sizeof(xfs_sb_t
);
1514 /* translate/copy */
1516 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1518 /* find modified range */
1520 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1521 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1522 first
= xfs_sb_info
[f
].offset
;
1524 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1525 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1526 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1528 xfs_trans_log_buf(tp
, bp
, first
, last
);
1533 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1534 * a delta to a specified field in the in-core superblock. Simply
1535 * switch on the field indicated and apply the delta to that field.
1536 * Fields are not allowed to dip below zero, so if the delta would
1537 * do this do not apply it and return EINVAL.
1539 * The m_sb_lock must be held when this routine is called.
1542 xfs_mod_incore_sb_unlocked(
1544 xfs_sb_field_t field
,
1548 int scounter
; /* short counter for 32 bit fields */
1549 long long lcounter
; /* long counter for 64 bit fields */
1550 long long res_used
, rem
;
1553 * With the in-core superblock spin lock held, switch
1554 * on the indicated field. Apply the delta to the
1555 * proper field. If the fields value would dip below
1556 * 0, then do not apply the delta and return EINVAL.
1559 case XFS_SBS_ICOUNT
:
1560 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1564 return XFS_ERROR(EINVAL
);
1566 mp
->m_sb
.sb_icount
= lcounter
;
1569 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1573 return XFS_ERROR(EINVAL
);
1575 mp
->m_sb
.sb_ifree
= lcounter
;
1577 case XFS_SBS_FDBLOCKS
:
1578 lcounter
= (long long)
1579 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1580 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1582 if (delta
> 0) { /* Putting blocks back */
1583 if (res_used
> delta
) {
1584 mp
->m_resblks_avail
+= delta
;
1586 rem
= delta
- res_used
;
1587 mp
->m_resblks_avail
= mp
->m_resblks
;
1590 } else { /* Taking blocks away */
1595 * If were out of blocks, use any available reserved blocks if
1601 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1603 return XFS_ERROR(ENOSPC
);
1605 mp
->m_resblks_avail
= lcounter
;
1607 } else { /* not reserved */
1608 return XFS_ERROR(ENOSPC
);
1613 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1615 case XFS_SBS_FREXTENTS
:
1616 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1619 return XFS_ERROR(ENOSPC
);
1621 mp
->m_sb
.sb_frextents
= lcounter
;
1623 case XFS_SBS_DBLOCKS
:
1624 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1628 return XFS_ERROR(EINVAL
);
1630 mp
->m_sb
.sb_dblocks
= lcounter
;
1632 case XFS_SBS_AGCOUNT
:
1633 scounter
= mp
->m_sb
.sb_agcount
;
1637 return XFS_ERROR(EINVAL
);
1639 mp
->m_sb
.sb_agcount
= scounter
;
1641 case XFS_SBS_IMAX_PCT
:
1642 scounter
= mp
->m_sb
.sb_imax_pct
;
1646 return XFS_ERROR(EINVAL
);
1648 mp
->m_sb
.sb_imax_pct
= scounter
;
1650 case XFS_SBS_REXTSIZE
:
1651 scounter
= mp
->m_sb
.sb_rextsize
;
1655 return XFS_ERROR(EINVAL
);
1657 mp
->m_sb
.sb_rextsize
= scounter
;
1659 case XFS_SBS_RBMBLOCKS
:
1660 scounter
= mp
->m_sb
.sb_rbmblocks
;
1664 return XFS_ERROR(EINVAL
);
1666 mp
->m_sb
.sb_rbmblocks
= scounter
;
1668 case XFS_SBS_RBLOCKS
:
1669 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1673 return XFS_ERROR(EINVAL
);
1675 mp
->m_sb
.sb_rblocks
= lcounter
;
1677 case XFS_SBS_REXTENTS
:
1678 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1682 return XFS_ERROR(EINVAL
);
1684 mp
->m_sb
.sb_rextents
= lcounter
;
1686 case XFS_SBS_REXTSLOG
:
1687 scounter
= mp
->m_sb
.sb_rextslog
;
1691 return XFS_ERROR(EINVAL
);
1693 mp
->m_sb
.sb_rextslog
= scounter
;
1697 return XFS_ERROR(EINVAL
);
1702 * xfs_mod_incore_sb() is used to change a field in the in-core
1703 * superblock structure by the specified delta. This modification
1704 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1705 * routine to do the work.
1710 xfs_sb_field_t field
,
1716 /* check for per-cpu counters */
1718 #ifdef HAVE_PERCPU_SB
1719 case XFS_SBS_ICOUNT
:
1721 case XFS_SBS_FDBLOCKS
:
1722 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1723 status
= xfs_icsb_modify_counters(mp
, field
,
1730 spin_lock(&mp
->m_sb_lock
);
1731 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1732 spin_unlock(&mp
->m_sb_lock
);
1740 * xfs_mod_incore_sb_batch() is used to change more than one field
1741 * in the in-core superblock structure at a time. This modification
1742 * is protected by a lock internal to this module. The fields and
1743 * changes to those fields are specified in the array of xfs_mod_sb
1744 * structures passed in.
1746 * Either all of the specified deltas will be applied or none of
1747 * them will. If any modified field dips below 0, then all modifications
1748 * will be backed out and EINVAL will be returned.
1751 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1757 * Loop through the array of mod structures and apply each
1758 * individually. If any fail, then back out all those
1759 * which have already been applied. Do all of this within
1760 * the scope of the m_sb_lock so that all of the changes will
1763 spin_lock(&mp
->m_sb_lock
);
1765 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1767 * Apply the delta at index n. If it fails, break
1768 * from the loop so we'll fall into the undo loop
1771 switch (msbp
->msb_field
) {
1772 #ifdef HAVE_PERCPU_SB
1773 case XFS_SBS_ICOUNT
:
1775 case XFS_SBS_FDBLOCKS
:
1776 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1777 spin_unlock(&mp
->m_sb_lock
);
1778 status
= xfs_icsb_modify_counters(mp
,
1780 msbp
->msb_delta
, rsvd
);
1781 spin_lock(&mp
->m_sb_lock
);
1787 status
= xfs_mod_incore_sb_unlocked(mp
,
1789 msbp
->msb_delta
, rsvd
);
1799 * If we didn't complete the loop above, then back out
1800 * any changes made to the superblock. If you add code
1801 * between the loop above and here, make sure that you
1802 * preserve the value of status. Loop back until
1803 * we step below the beginning of the array. Make sure
1804 * we don't touch anything back there.
1808 while (msbp
>= msb
) {
1809 switch (msbp
->msb_field
) {
1810 #ifdef HAVE_PERCPU_SB
1811 case XFS_SBS_ICOUNT
:
1813 case XFS_SBS_FDBLOCKS
:
1814 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1815 spin_unlock(&mp
->m_sb_lock
);
1816 status
= xfs_icsb_modify_counters(mp
,
1820 spin_lock(&mp
->m_sb_lock
);
1826 status
= xfs_mod_incore_sb_unlocked(mp
,
1832 ASSERT(status
== 0);
1836 spin_unlock(&mp
->m_sb_lock
);
1841 * xfs_getsb() is called to obtain the buffer for the superblock.
1842 * The buffer is returned locked and read in from disk.
1843 * The buffer should be released with a call to xfs_brelse().
1845 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1846 * the superblock buffer if it can be locked without sleeping.
1847 * If it can't then we'll return NULL.
1856 ASSERT(mp
->m_sb_bp
!= NULL
);
1858 if (flags
& XFS_BUF_TRYLOCK
) {
1859 if (!XFS_BUF_CPSEMA(bp
)) {
1863 XFS_BUF_PSEMA(bp
, PRIBIO
);
1866 ASSERT(XFS_BUF_ISDONE(bp
));
1871 * Used to free the superblock along various error paths.
1880 * Use xfs_getsb() so that the buffer will be locked
1881 * when we call xfs_buf_relse().
1883 bp
= xfs_getsb(mp
, 0);
1884 XFS_BUF_UNMANAGE(bp
);
1890 * See if the UUID is unique among mounted XFS filesystems.
1891 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1897 if (uuid_is_nil(&mp
->m_sb
.sb_uuid
)) {
1899 "XFS: Filesystem %s has nil UUID - can't mount",
1903 if (!uuid_table_insert(&mp
->m_sb
.sb_uuid
)) {
1905 "XFS: Filesystem %s has duplicate UUID - can't mount",
1913 * Used to log changes to the superblock unit and width fields which could
1914 * be altered by the mount options, as well as any potential sb_features2
1915 * fixup. Only the first superblock is updated.
1925 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1926 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1927 XFS_SB_VERSIONNUM
));
1929 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1930 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1931 XFS_DEFAULT_LOG_COUNT
);
1933 xfs_trans_cancel(tp
, 0);
1936 xfs_mod_sb(tp
, fields
);
1937 error
= xfs_trans_commit(tp
, 0);
1942 #ifdef HAVE_PERCPU_SB
1944 * Per-cpu incore superblock counters
1946 * Simple concept, difficult implementation
1948 * Basically, replace the incore superblock counters with a distributed per cpu
1949 * counter for contended fields (e.g. free block count).
1951 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1952 * hence needs to be accurately read when we are running low on space. Hence
1953 * there is a method to enable and disable the per-cpu counters based on how
1954 * much "stuff" is available in them.
1956 * Basically, a counter is enabled if there is enough free resource to justify
1957 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1958 * ENOSPC), then we disable the counters to synchronise all callers and
1959 * re-distribute the available resources.
1961 * If, once we redistributed the available resources, we still get a failure,
1962 * we disable the per-cpu counter and go through the slow path.
1964 * The slow path is the current xfs_mod_incore_sb() function. This means that
1965 * when we disable a per-cpu counter, we need to drain it's resources back to
1966 * the global superblock. We do this after disabling the counter to prevent
1967 * more threads from queueing up on the counter.
1969 * Essentially, this means that we still need a lock in the fast path to enable
1970 * synchronisation between the global counters and the per-cpu counters. This
1971 * is not a problem because the lock will be local to a CPU almost all the time
1972 * and have little contention except when we get to ENOSPC conditions.
1974 * Basically, this lock becomes a barrier that enables us to lock out the fast
1975 * path while we do things like enabling and disabling counters and
1976 * synchronising the counters.
1980 * 1. m_sb_lock before picking up per-cpu locks
1981 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1982 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1983 * 4. modifying per-cpu counters requires holding per-cpu lock
1984 * 5. modifying global counters requires holding m_sb_lock
1985 * 6. enabling or disabling a counter requires holding the m_sb_lock
1986 * and _none_ of the per-cpu locks.
1988 * Disabled counters are only ever re-enabled by a balance operation
1989 * that results in more free resources per CPU than a given threshold.
1990 * To ensure counters don't remain disabled, they are rebalanced when
1991 * the global resource goes above a higher threshold (i.e. some hysteresis
1992 * is present to prevent thrashing).
1995 #ifdef CONFIG_HOTPLUG_CPU
1997 * hot-plug CPU notifier support.
1999 * We need a notifier per filesystem as we need to be able to identify
2000 * the filesystem to balance the counters out. This is achieved by
2001 * having a notifier block embedded in the xfs_mount_t and doing pointer
2002 * magic to get the mount pointer from the notifier block address.
2005 xfs_icsb_cpu_notify(
2006 struct notifier_block
*nfb
,
2007 unsigned long action
,
2010 xfs_icsb_cnts_t
*cntp
;
2013 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
2014 cntp
= (xfs_icsb_cnts_t
*)
2015 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
2017 case CPU_UP_PREPARE
:
2018 case CPU_UP_PREPARE_FROZEN
:
2019 /* Easy Case - initialize the area and locks, and
2020 * then rebalance when online does everything else for us. */
2021 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2024 case CPU_ONLINE_FROZEN
:
2026 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2027 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2028 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2029 xfs_icsb_unlock(mp
);
2032 case CPU_DEAD_FROZEN
:
2033 /* Disable all the counters, then fold the dead cpu's
2034 * count into the total on the global superblock and
2035 * re-enable the counters. */
2037 spin_lock(&mp
->m_sb_lock
);
2038 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
2039 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
2040 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
2042 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
2043 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
2044 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
2046 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2048 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
2049 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
2050 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
2051 spin_unlock(&mp
->m_sb_lock
);
2052 xfs_icsb_unlock(mp
);
2058 #endif /* CONFIG_HOTPLUG_CPU */
2061 xfs_icsb_init_counters(
2064 xfs_icsb_cnts_t
*cntp
;
2067 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
2068 if (mp
->m_sb_cnts
== NULL
)
2071 #ifdef CONFIG_HOTPLUG_CPU
2072 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
2073 mp
->m_icsb_notifier
.priority
= 0;
2074 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
2075 #endif /* CONFIG_HOTPLUG_CPU */
2077 for_each_online_cpu(i
) {
2078 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2079 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2082 mutex_init(&mp
->m_icsb_mutex
);
2085 * start with all counters disabled so that the
2086 * initial balance kicks us off correctly
2088 mp
->m_icsb_counters
= -1;
2093 xfs_icsb_reinit_counters(
2098 * start with all counters disabled so that the
2099 * initial balance kicks us off correctly
2101 mp
->m_icsb_counters
= -1;
2102 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2103 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2104 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2105 xfs_icsb_unlock(mp
);
2109 xfs_icsb_destroy_counters(
2112 if (mp
->m_sb_cnts
) {
2113 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2114 free_percpu(mp
->m_sb_cnts
);
2116 mutex_destroy(&mp
->m_icsb_mutex
);
2121 xfs_icsb_cnts_t
*icsbp
)
2123 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2129 xfs_icsb_unlock_cntr(
2130 xfs_icsb_cnts_t
*icsbp
)
2132 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2137 xfs_icsb_lock_all_counters(
2140 xfs_icsb_cnts_t
*cntp
;
2143 for_each_online_cpu(i
) {
2144 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2145 xfs_icsb_lock_cntr(cntp
);
2150 xfs_icsb_unlock_all_counters(
2153 xfs_icsb_cnts_t
*cntp
;
2156 for_each_online_cpu(i
) {
2157 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2158 xfs_icsb_unlock_cntr(cntp
);
2165 xfs_icsb_cnts_t
*cnt
,
2168 xfs_icsb_cnts_t
*cntp
;
2171 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2173 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2174 xfs_icsb_lock_all_counters(mp
);
2176 for_each_online_cpu(i
) {
2177 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2178 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2179 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2180 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2183 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2184 xfs_icsb_unlock_all_counters(mp
);
2188 xfs_icsb_counter_disabled(
2190 xfs_sb_field_t field
)
2192 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2193 return test_bit(field
, &mp
->m_icsb_counters
);
2197 xfs_icsb_disable_counter(
2199 xfs_sb_field_t field
)
2201 xfs_icsb_cnts_t cnt
;
2203 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2206 * If we are already disabled, then there is nothing to do
2207 * here. We check before locking all the counters to avoid
2208 * the expensive lock operation when being called in the
2209 * slow path and the counter is already disabled. This is
2210 * safe because the only time we set or clear this state is under
2213 if (xfs_icsb_counter_disabled(mp
, field
))
2216 xfs_icsb_lock_all_counters(mp
);
2217 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2218 /* drain back to superblock */
2220 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2222 case XFS_SBS_ICOUNT
:
2223 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2226 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2228 case XFS_SBS_FDBLOCKS
:
2229 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2236 xfs_icsb_unlock_all_counters(mp
);
2240 xfs_icsb_enable_counter(
2242 xfs_sb_field_t field
,
2246 xfs_icsb_cnts_t
*cntp
;
2249 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2251 xfs_icsb_lock_all_counters(mp
);
2252 for_each_online_cpu(i
) {
2253 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2255 case XFS_SBS_ICOUNT
:
2256 cntp
->icsb_icount
= count
+ resid
;
2259 cntp
->icsb_ifree
= count
+ resid
;
2261 case XFS_SBS_FDBLOCKS
:
2262 cntp
->icsb_fdblocks
= count
+ resid
;
2270 clear_bit(field
, &mp
->m_icsb_counters
);
2271 xfs_icsb_unlock_all_counters(mp
);
2275 xfs_icsb_sync_counters_locked(
2279 xfs_icsb_cnts_t cnt
;
2281 xfs_icsb_count(mp
, &cnt
, flags
);
2283 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2284 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2285 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2286 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2287 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2288 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2292 * Accurate update of per-cpu counters to incore superblock
2295 xfs_icsb_sync_counters(
2299 spin_lock(&mp
->m_sb_lock
);
2300 xfs_icsb_sync_counters_locked(mp
, flags
);
2301 spin_unlock(&mp
->m_sb_lock
);
2305 * Balance and enable/disable counters as necessary.
2307 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2308 * chosen to be the same number as single on disk allocation chunk per CPU, and
2309 * free blocks is something far enough zero that we aren't going thrash when we
2310 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2311 * prevent looping endlessly when xfs_alloc_space asks for more than will
2312 * be distributed to a single CPU but each CPU has enough blocks to be
2315 * Note that we can be called when counters are already disabled.
2316 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2317 * prevent locking every per-cpu counter needlessly.
2320 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2321 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2322 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2324 xfs_icsb_balance_counter_locked(
2326 xfs_sb_field_t field
,
2329 uint64_t count
, resid
;
2330 int weight
= num_online_cpus();
2331 uint64_t min
= (uint64_t)min_per_cpu
;
2333 /* disable counter and sync counter */
2334 xfs_icsb_disable_counter(mp
, field
);
2336 /* update counters - first CPU gets residual*/
2338 case XFS_SBS_ICOUNT
:
2339 count
= mp
->m_sb
.sb_icount
;
2340 resid
= do_div(count
, weight
);
2341 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2345 count
= mp
->m_sb
.sb_ifree
;
2346 resid
= do_div(count
, weight
);
2347 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2350 case XFS_SBS_FDBLOCKS
:
2351 count
= mp
->m_sb
.sb_fdblocks
;
2352 resid
= do_div(count
, weight
);
2353 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2358 count
= resid
= 0; /* quiet, gcc */
2362 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2366 xfs_icsb_balance_counter(
2368 xfs_sb_field_t fields
,
2371 spin_lock(&mp
->m_sb_lock
);
2372 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2373 spin_unlock(&mp
->m_sb_lock
);
2377 xfs_icsb_modify_counters(
2379 xfs_sb_field_t field
,
2383 xfs_icsb_cnts_t
*icsbp
;
2384 long long lcounter
; /* long counter for 64 bit fields */
2390 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
);
2393 * if the counter is disabled, go to slow path
2395 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2397 xfs_icsb_lock_cntr(icsbp
);
2398 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2399 xfs_icsb_unlock_cntr(icsbp
);
2404 case XFS_SBS_ICOUNT
:
2405 lcounter
= icsbp
->icsb_icount
;
2407 if (unlikely(lcounter
< 0))
2408 goto balance_counter
;
2409 icsbp
->icsb_icount
= lcounter
;
2413 lcounter
= icsbp
->icsb_ifree
;
2415 if (unlikely(lcounter
< 0))
2416 goto balance_counter
;
2417 icsbp
->icsb_ifree
= lcounter
;
2420 case XFS_SBS_FDBLOCKS
:
2421 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2423 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2425 if (unlikely(lcounter
< 0))
2426 goto balance_counter
;
2427 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2433 xfs_icsb_unlock_cntr(icsbp
);
2441 * serialise with a mutex so we don't burn lots of cpu on
2442 * the superblock lock. We still need to hold the superblock
2443 * lock, however, when we modify the global structures.
2448 * Now running atomically.
2450 * If the counter is enabled, someone has beaten us to rebalancing.
2451 * Drop the lock and try again in the fast path....
2453 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2454 xfs_icsb_unlock(mp
);
2459 * The counter is currently disabled. Because we are
2460 * running atomically here, we know a rebalance cannot
2461 * be in progress. Hence we can go straight to operating
2462 * on the global superblock. We do not call xfs_mod_incore_sb()
2463 * here even though we need to get the m_sb_lock. Doing so
2464 * will cause us to re-enter this function and deadlock.
2465 * Hence we get the m_sb_lock ourselves and then call
2466 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2467 * directly on the global counters.
2469 spin_lock(&mp
->m_sb_lock
);
2470 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2471 spin_unlock(&mp
->m_sb_lock
);
2474 * Now that we've modified the global superblock, we
2475 * may be able to re-enable the distributed counters
2476 * (e.g. lots of space just got freed). After that
2480 xfs_icsb_balance_counter(mp
, field
, 0);
2481 xfs_icsb_unlock(mp
);
2485 xfs_icsb_unlock_cntr(icsbp
);
2489 * We may have multiple threads here if multiple per-cpu
2490 * counters run dry at the same time. This will mean we can
2491 * do more balances than strictly necessary but it is not
2492 * the common slowpath case.
2497 * running atomically.
2499 * This will leave the counter in the correct state for future
2500 * accesses. After the rebalance, we simply try again and our retry
2501 * will either succeed through the fast path or slow path without
2502 * another balance operation being required.
2504 xfs_icsb_balance_counter(mp
, field
, delta
);
2505 xfs_icsb_unlock(mp
);