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_uuid_mount(xfs_mount_t
*);
49 STATIC
void xfs_unmountfs_wait(xfs_mount_t
*);
53 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
,
55 STATIC
void xfs_icsb_balance_counter_locked(xfs_mount_t
*, xfs_sb_field_t
,
57 STATIC
int xfs_icsb_modify_counters(xfs_mount_t
*, xfs_sb_field_t
,
59 STATIC
void xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
63 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
64 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
65 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
71 short type
; /* 0 = integer
72 * 1 = binary / string (no translation)
75 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
76 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
77 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
78 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
79 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
80 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
81 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
82 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
83 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
84 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
85 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
86 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
87 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
88 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
89 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
90 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
91 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
92 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
93 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
94 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
95 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
96 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
97 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
98 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
99 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
100 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
101 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
102 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
103 { offsetof(xfs_sb_t
, sb_icount
), 0 },
104 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
105 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
106 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
107 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
108 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
109 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
110 { offsetof(xfs_sb_t
, sb_flags
), 0 },
111 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
112 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
113 { offsetof(xfs_sb_t
, sb_unit
), 0 },
114 { offsetof(xfs_sb_t
, sb_width
), 0 },
115 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
116 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
117 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
118 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
119 { offsetof(xfs_sb_t
, sb_features2
), 0 },
120 { offsetof(xfs_sb_t
, sb_bad_features2
), 0 },
121 { sizeof(xfs_sb_t
), 0 }
125 * Free up the resources associated with a mount structure. Assume that
126 * the structure was initially zeroed, so we can tell which fields got
136 for (agno
= 0; agno
< mp
->m_maxagi
; agno
++)
137 if (mp
->m_perag
[agno
].pagb_list
)
138 kmem_free(mp
->m_perag
[agno
].pagb_list
);
139 kmem_free(mp
->m_perag
);
144 * Check size of device based on the (data/realtime) block count.
145 * Note: this check is used by the growfs code as well as mount.
148 xfs_sb_validate_fsb_count(
152 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
153 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
155 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
156 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
158 #else /* Limited by UINT_MAX of sectors */
159 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
166 * Check the validity of the SB found.
169 xfs_mount_validate_sb(
175 * If the log device and data device have the
176 * same device number, the log is internal.
177 * Consequently, the sb_logstart should be non-zero. If
178 * we have a zero sb_logstart in this case, we may be trying to mount
179 * a volume filesystem in a non-volume manner.
181 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
182 xfs_fs_mount_cmn_err(flags
, "bad magic number");
183 return XFS_ERROR(EWRONGFS
);
186 if (!xfs_sb_good_version(sbp
)) {
187 xfs_fs_mount_cmn_err(flags
, "bad version");
188 return XFS_ERROR(EWRONGFS
);
192 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
193 xfs_fs_mount_cmn_err(flags
,
194 "filesystem is marked as having an external log; "
195 "specify logdev on the\nmount command line.");
196 return XFS_ERROR(EINVAL
);
200 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
201 xfs_fs_mount_cmn_err(flags
,
202 "filesystem is marked as having an internal log; "
203 "do not specify logdev on\nthe mount command line.");
204 return XFS_ERROR(EINVAL
);
208 * More sanity checking. These were stolen directly from
212 sbp
->sb_agcount
<= 0 ||
213 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
214 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
215 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
216 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
217 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
218 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
219 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
220 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
221 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
222 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
223 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
224 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
225 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
226 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
227 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
228 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
229 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
230 return XFS_ERROR(EFSCORRUPTED
);
234 * Sanity check AG count, size fields against data size field
237 sbp
->sb_dblocks
== 0 ||
239 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
240 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
241 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
242 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
243 return XFS_ERROR(EFSCORRUPTED
);
247 * Until this is fixed only page-sized or smaller data blocks work.
249 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
250 xfs_fs_mount_cmn_err(flags
,
251 "file system with blocksize %d bytes",
253 xfs_fs_mount_cmn_err(flags
,
254 "only pagesize (%ld) or less will currently work.",
256 return XFS_ERROR(ENOSYS
);
259 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
260 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
261 xfs_fs_mount_cmn_err(flags
,
262 "file system too large to be mounted on this system.");
263 return XFS_ERROR(E2BIG
);
266 if (unlikely(sbp
->sb_inprogress
)) {
267 xfs_fs_mount_cmn_err(flags
, "file system busy");
268 return XFS_ERROR(EFSCORRUPTED
);
272 * Version 1 directory format has never worked on Linux.
274 if (unlikely(!xfs_sb_version_hasdirv2(sbp
))) {
275 xfs_fs_mount_cmn_err(flags
,
276 "file system using version 1 directory format");
277 return XFS_ERROR(ENOSYS
);
284 xfs_initialize_perag_icache(
287 if (!pag
->pag_ici_init
) {
288 rwlock_init(&pag
->pag_ici_lock
);
289 INIT_RADIX_TREE(&pag
->pag_ici_root
, GFP_ATOMIC
);
290 pag
->pag_ici_init
= 1;
295 xfs_initialize_perag(
297 xfs_agnumber_t agcount
)
299 xfs_agnumber_t index
, max_metadata
;
303 xfs_sb_t
*sbp
= &mp
->m_sb
;
304 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
306 /* Check to see if the filesystem can overflow 32 bit inodes */
307 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
308 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
310 /* Clear the mount flag if no inode can overflow 32 bits
311 * on this filesystem, or if specifically requested..
313 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> max_inum
) {
314 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
316 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
319 /* If we can overflow then setup the ag headers accordingly */
320 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
321 /* Calculate how much should be reserved for inodes to
322 * meet the max inode percentage.
324 if (mp
->m_maxicount
) {
327 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
329 icount
+= sbp
->sb_agblocks
- 1;
330 do_div(icount
, sbp
->sb_agblocks
);
331 max_metadata
= icount
;
333 max_metadata
= agcount
;
335 for (index
= 0; index
< agcount
; index
++) {
336 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
337 if (ino
> max_inum
) {
342 /* This ag is preferred for inodes */
343 pag
= &mp
->m_perag
[index
];
344 pag
->pagi_inodeok
= 1;
345 if (index
< max_metadata
)
346 pag
->pagf_metadata
= 1;
347 xfs_initialize_perag_icache(pag
);
350 /* Setup default behavior for smaller filesystems */
351 for (index
= 0; index
< agcount
; index
++) {
352 pag
= &mp
->m_perag
[index
];
353 pag
->pagi_inodeok
= 1;
354 xfs_initialize_perag_icache(pag
);
365 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
366 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
367 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
368 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
369 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
370 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
371 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
372 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
373 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
374 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
375 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
376 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
377 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
378 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
379 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
380 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
381 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
382 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
383 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
384 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
385 to
->sb_blocklog
= from
->sb_blocklog
;
386 to
->sb_sectlog
= from
->sb_sectlog
;
387 to
->sb_inodelog
= from
->sb_inodelog
;
388 to
->sb_inopblog
= from
->sb_inopblog
;
389 to
->sb_agblklog
= from
->sb_agblklog
;
390 to
->sb_rextslog
= from
->sb_rextslog
;
391 to
->sb_inprogress
= from
->sb_inprogress
;
392 to
->sb_imax_pct
= from
->sb_imax_pct
;
393 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
394 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
395 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
396 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
397 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
398 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
399 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
400 to
->sb_flags
= from
->sb_flags
;
401 to
->sb_shared_vn
= from
->sb_shared_vn
;
402 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
403 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
404 to
->sb_width
= be32_to_cpu(from
->sb_width
);
405 to
->sb_dirblklog
= from
->sb_dirblklog
;
406 to
->sb_logsectlog
= from
->sb_logsectlog
;
407 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
408 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
409 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
410 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
414 * Copy in core superblock to ondisk one.
416 * The fields argument is mask of superblock fields to copy.
424 xfs_caddr_t to_ptr
= (xfs_caddr_t
)to
;
425 xfs_caddr_t from_ptr
= (xfs_caddr_t
)from
;
435 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
436 first
= xfs_sb_info
[f
].offset
;
437 size
= xfs_sb_info
[f
+ 1].offset
- first
;
439 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
441 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
442 memcpy(to_ptr
+ first
, from_ptr
+ first
, size
);
446 *(__be16
*)(to_ptr
+ first
) =
447 cpu_to_be16(*(__u16
*)(from_ptr
+ first
));
450 *(__be32
*)(to_ptr
+ first
) =
451 cpu_to_be32(*(__u32
*)(from_ptr
+ first
));
454 *(__be64
*)(to_ptr
+ first
) =
455 cpu_to_be64(*(__u64
*)(from_ptr
+ first
));
462 fields
&= ~(1LL << f
);
469 * Does the initial read of the superblock.
472 xfs_readsb(xfs_mount_t
*mp
, int flags
)
474 unsigned int sector_size
;
475 unsigned int extra_flags
;
479 ASSERT(mp
->m_sb_bp
== NULL
);
480 ASSERT(mp
->m_ddev_targp
!= NULL
);
483 * Allocate a (locked) buffer to hold the superblock.
484 * This will be kept around at all times to optimize
485 * access to the superblock.
487 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
488 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
490 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
491 BTOBB(sector_size
), extra_flags
);
492 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
493 xfs_fs_mount_cmn_err(flags
, "SB read failed");
494 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
497 ASSERT(XFS_BUF_ISBUSY(bp
));
498 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
501 * Initialize the mount structure from the superblock.
502 * But first do some basic consistency checking.
504 xfs_sb_from_disk(&mp
->m_sb
, XFS_BUF_TO_SBP(bp
));
506 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
508 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
513 * We must be able to do sector-sized and sector-aligned IO.
515 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
516 xfs_fs_mount_cmn_err(flags
,
517 "device supports only %u byte sectors (not %u)",
518 sector_size
, mp
->m_sb
.sb_sectsize
);
524 * If device sector size is smaller than the superblock size,
525 * re-read the superblock so the buffer is correctly sized.
527 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
528 XFS_BUF_UNMANAGE(bp
);
530 sector_size
= mp
->m_sb
.sb_sectsize
;
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 re-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 per-cpu counters */
543 xfs_icsb_reinit_counters(mp
);
547 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
552 XFS_BUF_UNMANAGE(bp
);
562 * Mount initialization code establishing various mount
563 * fields from the superblock associated with the given
567 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
569 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
570 spin_lock_init(&mp
->m_agirotor_lock
);
571 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
572 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
573 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
574 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
575 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
576 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
577 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
578 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
579 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
582 * Setup for attributes, in case they get created.
583 * This value is for inodes getting attributes for the first time,
584 * the per-inode value is for old attribute values.
586 ASSERT(sbp
->sb_inodesize
>= 256 && sbp
->sb_inodesize
<= 2048);
587 switch (sbp
->sb_inodesize
) {
589 mp
->m_attroffset
= XFS_LITINO(mp
) -
590 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
595 mp
->m_attroffset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
600 ASSERT(mp
->m_attroffset
< XFS_LITINO(mp
));
602 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
603 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
604 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
605 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
607 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
608 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
609 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
610 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
612 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
613 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
614 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
615 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
617 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
618 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
620 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
624 * xfs_initialize_perag_data
626 * Read in each per-ag structure so we can count up the number of
627 * allocated inodes, free inodes and used filesystem blocks as this
628 * information is no longer persistent in the superblock. Once we have
629 * this information, write it into the in-core superblock structure.
632 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
634 xfs_agnumber_t index
;
636 xfs_sb_t
*sbp
= &mp
->m_sb
;
640 uint64_t bfreelst
= 0;
644 for (index
= 0; index
< agcount
; index
++) {
646 * read the agf, then the agi. This gets us
647 * all the inforamtion we need and populates the
648 * per-ag structures for us.
650 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
654 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
657 pag
= &mp
->m_perag
[index
];
658 ifree
+= pag
->pagi_freecount
;
659 ialloc
+= pag
->pagi_count
;
660 bfree
+= pag
->pagf_freeblks
;
661 bfreelst
+= pag
->pagf_flcount
;
662 btree
+= pag
->pagf_btreeblks
;
665 * Overwrite incore superblock counters with just-read data
667 spin_lock(&mp
->m_sb_lock
);
668 sbp
->sb_ifree
= ifree
;
669 sbp
->sb_icount
= ialloc
;
670 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
671 spin_unlock(&mp
->m_sb_lock
);
673 /* Fixup the per-cpu counters as well. */
674 xfs_icsb_reinit_counters(mp
);
680 * Update alignment values based on mount options and sb values
683 xfs_update_alignment(xfs_mount_t
*mp
)
685 xfs_sb_t
*sbp
= &(mp
->m_sb
);
689 * If stripe unit and stripe width are not multiples
690 * of the fs blocksize turn off alignment.
692 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
693 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
694 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
696 "XFS: alignment check 1 failed");
697 return XFS_ERROR(EINVAL
);
699 mp
->m_dalign
= mp
->m_swidth
= 0;
702 * Convert the stripe unit and width to FSBs.
704 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
705 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
706 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
707 return XFS_ERROR(EINVAL
);
709 xfs_fs_cmn_err(CE_WARN
, mp
,
710 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
711 mp
->m_dalign
, mp
->m_swidth
,
716 } else if (mp
->m_dalign
) {
717 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
719 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
720 xfs_fs_cmn_err(CE_WARN
, mp
,
721 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
724 return XFS_ERROR(EINVAL
);
731 * Update superblock with new values
734 if (xfs_sb_version_hasdalign(sbp
)) {
735 if (sbp
->sb_unit
!= mp
->m_dalign
) {
736 sbp
->sb_unit
= mp
->m_dalign
;
737 mp
->m_update_flags
|= XFS_SB_UNIT
;
739 if (sbp
->sb_width
!= mp
->m_swidth
) {
740 sbp
->sb_width
= mp
->m_swidth
;
741 mp
->m_update_flags
|= XFS_SB_WIDTH
;
744 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
745 xfs_sb_version_hasdalign(&mp
->m_sb
)) {
746 mp
->m_dalign
= sbp
->sb_unit
;
747 mp
->m_swidth
= sbp
->sb_width
;
754 * Set the maximum inode count for this filesystem
757 xfs_set_maxicount(xfs_mount_t
*mp
)
759 xfs_sb_t
*sbp
= &(mp
->m_sb
);
762 if (sbp
->sb_imax_pct
) {
764 * Make sure the maximum inode count is a multiple
765 * of the units we allocate inodes in.
767 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
769 do_div(icount
, mp
->m_ialloc_blks
);
770 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
778 * Set the default minimum read and write sizes unless
779 * already specified in a mount option.
780 * We use smaller I/O sizes when the file system
781 * is being used for NFS service (wsync mount option).
784 xfs_set_rw_sizes(xfs_mount_t
*mp
)
786 xfs_sb_t
*sbp
= &(mp
->m_sb
);
787 int readio_log
, writeio_log
;
789 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
790 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
791 readio_log
= XFS_WSYNC_READIO_LOG
;
792 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
794 readio_log
= XFS_READIO_LOG_LARGE
;
795 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
798 readio_log
= mp
->m_readio_log
;
799 writeio_log
= mp
->m_writeio_log
;
802 if (sbp
->sb_blocklog
> readio_log
) {
803 mp
->m_readio_log
= sbp
->sb_blocklog
;
805 mp
->m_readio_log
= readio_log
;
807 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
808 if (sbp
->sb_blocklog
> writeio_log
) {
809 mp
->m_writeio_log
= sbp
->sb_blocklog
;
811 mp
->m_writeio_log
= writeio_log
;
813 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
817 * Set whether we're using inode alignment.
820 xfs_set_inoalignment(xfs_mount_t
*mp
)
822 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
823 mp
->m_sb
.sb_inoalignmt
>=
824 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
825 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
827 mp
->m_inoalign_mask
= 0;
829 * If we are using stripe alignment, check whether
830 * the stripe unit is a multiple of the inode alignment
832 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
833 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
834 mp
->m_sinoalign
= mp
->m_dalign
;
840 * Check that the data (and log if separate) are an ok size.
843 xfs_check_sizes(xfs_mount_t
*mp
)
849 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
850 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
851 cmn_err(CE_WARN
, "XFS: size check 1 failed");
852 return XFS_ERROR(E2BIG
);
854 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
855 d
- XFS_FSS_TO_BB(mp
, 1),
856 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
860 cmn_err(CE_WARN
, "XFS: size check 2 failed");
862 error
= XFS_ERROR(E2BIG
);
866 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
867 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
868 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
869 cmn_err(CE_WARN
, "XFS: size check 3 failed");
870 return XFS_ERROR(E2BIG
);
872 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
873 d
- XFS_FSB_TO_BB(mp
, 1),
874 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
878 cmn_err(CE_WARN
, "XFS: size check 3 failed");
880 error
= XFS_ERROR(E2BIG
);
888 * This function does the following on an initial mount of a file system:
889 * - reads the superblock from disk and init the mount struct
890 * - if we're a 32-bit kernel, do a size check on the superblock
891 * so we don't mount terabyte filesystems
892 * - init mount struct realtime fields
893 * - allocate inode hash table for fs
894 * - init directory manager
895 * - perform recovery and init the log manager
901 xfs_sb_t
*sbp
= &(mp
->m_sb
);
904 uint quotamount
, quotaflags
;
907 xfs_mount_common(mp
, sbp
);
910 * Check for a mismatched features2 values. Older kernels
911 * read & wrote into the wrong sb offset for sb_features2
912 * on some platforms due to xfs_sb_t not being 64bit size aligned
913 * when sb_features2 was added, which made older superblock
914 * reading/writing routines swap it as a 64-bit value.
916 * For backwards compatibility, we make both slots equal.
918 * If we detect a mismatched field, we OR the set bits into the
919 * existing features2 field in case it has already been modified; we
920 * don't want to lose any features. We then update the bad location
921 * with the ORed value so that older kernels will see any features2
922 * flags, and mark the two fields as needing updates once the
923 * transaction subsystem is online.
925 if (xfs_sb_has_mismatched_features2(sbp
)) {
927 "XFS: correcting sb_features alignment problem");
928 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
929 sbp
->sb_bad_features2
= sbp
->sb_features2
;
930 mp
->m_update_flags
|= XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
;
933 * Re-check for ATTR2 in case it was found in bad_features2
936 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
937 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
938 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
941 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
942 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
943 xfs_sb_version_removeattr2(&mp
->m_sb
);
944 mp
->m_update_flags
|= XFS_SB_FEATURES2
;
946 /* update sb_versionnum for the clearing of the morebits */
947 if (!sbp
->sb_features2
)
948 mp
->m_update_flags
|= XFS_SB_VERSIONNUM
;
952 * Check if sb_agblocks is aligned at stripe boundary
953 * If sb_agblocks is NOT aligned turn off m_dalign since
954 * allocator alignment is within an ag, therefore ag has
955 * to be aligned at stripe boundary.
957 error
= xfs_update_alignment(mp
);
961 xfs_alloc_compute_maxlevels(mp
);
962 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
963 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
964 xfs_ialloc_compute_maxlevels(mp
);
966 xfs_set_maxicount(mp
);
968 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
971 * XFS uses the uuid from the superblock as the unique
972 * identifier for fsid. We can not use the uuid from the volume
973 * since a single partition filesystem is identical to a single
974 * partition volume/filesystem.
976 if (!(mp
->m_flags
& XFS_MOUNT_NOUUID
)) {
977 if (xfs_uuid_mount(mp
)) {
978 error
= XFS_ERROR(EINVAL
);
984 * Set the minimum read and write sizes
986 xfs_set_rw_sizes(mp
);
989 * Set the inode cluster size.
990 * This may still be overridden by the file system
991 * block size if it is larger than the chosen cluster size.
993 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
996 * Set inode alignment fields
998 xfs_set_inoalignment(mp
);
1001 * Check that the data (and log if separate) are an ok size.
1003 error
= xfs_check_sizes(mp
);
1005 goto out_remove_uuid
;
1008 * Initialize realtime fields in the mount structure
1010 error
= xfs_rtmount_init(mp
);
1012 cmn_err(CE_WARN
, "XFS: RT mount failed");
1013 goto out_remove_uuid
;
1017 * Copies the low order bits of the timestamp and the randomly
1018 * set "sequence" number out of a UUID.
1020 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
1022 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1027 * Initialize the attribute manager's entries.
1029 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1032 * Initialize the precomputed transaction reservations values.
1037 * Allocate and initialize the per-ag data.
1039 init_rwsem(&mp
->m_peraglock
);
1040 mp
->m_perag
= kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
),
1043 goto out_remove_uuid
;
1045 mp
->m_maxagi
= xfs_initialize_perag(mp
, sbp
->sb_agcount
);
1047 if (!sbp
->sb_logblocks
) {
1048 cmn_err(CE_WARN
, "XFS: no log defined");
1049 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW
, mp
);
1050 error
= XFS_ERROR(EFSCORRUPTED
);
1051 goto out_free_perag
;
1055 * log's mount-time initialization. Perform 1st part recovery if needed
1057 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1058 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1059 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1061 cmn_err(CE_WARN
, "XFS: log mount failed");
1062 goto out_free_perag
;
1066 * Now the log is mounted, we know if it was an unclean shutdown or
1067 * not. If it was, with the first phase of recovery has completed, we
1068 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1069 * but they are recovered transactionally in the second recovery phase
1072 * Hence we can safely re-initialise incore superblock counters from
1073 * the per-ag data. These may not be correct if the filesystem was not
1074 * cleanly unmounted, so we need to wait for recovery to finish before
1077 * If the filesystem was cleanly unmounted, then we can trust the
1078 * values in the superblock to be correct and we don't need to do
1081 * If we are currently making the filesystem, the initialisation will
1082 * fail as the perag data is in an undefined state.
1084 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1085 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1086 !mp
->m_sb
.sb_inprogress
) {
1087 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1089 goto out_free_perag
;
1093 * Get and sanity-check the root inode.
1094 * Save the pointer to it in the mount structure.
1096 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1098 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1099 goto out_log_dealloc
;
1102 ASSERT(rip
!= NULL
);
1104 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1105 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1106 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1107 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1108 (unsigned long long)rip
->i_ino
);
1109 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1110 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1112 error
= XFS_ERROR(EFSCORRUPTED
);
1115 mp
->m_rootip
= rip
; /* save it */
1117 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1120 * Initialize realtime inode pointers in the mount structure
1122 error
= xfs_rtmount_inodes(mp
);
1125 * Free up the root inode.
1127 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1132 * If this is a read-only mount defer the superblock updates until
1133 * the next remount into writeable mode. Otherwise we would never
1134 * perform the update e.g. for the root filesystem.
1136 if (mp
->m_update_flags
&& !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1137 error
= xfs_mount_log_sb(mp
, mp
->m_update_flags
);
1139 cmn_err(CE_WARN
, "XFS: failed to write sb changes");
1145 * Initialise the XFS quota management subsystem for this mount
1147 error
= XFS_QM_INIT(mp
, "amount
, "aflags
);
1152 * Finish recovering the file system. This part needed to be
1153 * delayed until after the root and real-time bitmap inodes
1154 * were consistently read in.
1156 error
= xfs_log_mount_finish(mp
);
1158 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1163 * Complete the quota initialisation, post-log-replay component.
1165 error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
);
1170 * Now we are mounted, reserve a small amount of unused space for
1171 * privileged transactions. This is needed so that transaction
1172 * space required for critical operations can dip into this pool
1173 * when at ENOSPC. This is needed for operations like create with
1174 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1175 * are not allowed to use this reserved space.
1177 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1178 * This may drive us straight to ENOSPC on mount, but that implies
1179 * we were already there on the last unmount. Warn if this occurs.
1181 resblks
= mp
->m_sb
.sb_dblocks
;
1182 do_div(resblks
, 20);
1183 resblks
= min_t(__uint64_t
, resblks
, 1024);
1184 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1186 cmn_err(CE_WARN
, "XFS: Unable to allocate reserve blocks. "
1187 "Continuing without a reserve pool.");
1192 xfs_rtunmount_inodes(mp
);
1196 xfs_log_unmount(mp
);
1200 if (!(mp
->m_flags
& XFS_MOUNT_NOUUID
))
1201 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1207 * This flushes out the inodes,dquots and the superblock, unmounts the
1208 * log and makes sure that incore structures are freed.
1212 struct xfs_mount
*mp
)
1218 * Release dquot that rootinode, rbmino and rsumino might be holding,
1219 * and release the quota inodes.
1223 xfs_rtunmount_inodes(mp
);
1224 IRELE(mp
->m_rootip
);
1227 * We can potentially deadlock here if we have an inode cluster
1228 * that has been freed has it's buffer still pinned in memory because
1229 * the transaction is still sitting in a iclog. The stale inodes
1230 * on that buffer will have their flush locks held until the
1231 * transaction hits the disk and the callbacks run. the inode
1232 * flush takes the flush lock unconditionally and with nothing to
1233 * push out the iclog we will never get that unlocked. hence we
1234 * need to force the log first.
1236 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1237 xfs_reclaim_inodes(mp
, 0, XFS_IFLUSH_ASYNC
);
1239 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1241 if (mp
->m_quotainfo
)
1245 * Flush out the log synchronously so that we know for sure
1246 * that nothing is pinned. This is important because bflush()
1247 * will skip pinned buffers.
1249 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1251 xfs_binval(mp
->m_ddev_targp
);
1252 if (mp
->m_rtdev_targp
) {
1253 xfs_binval(mp
->m_rtdev_targp
);
1257 * Unreserve any blocks we have so that when we unmount we don't account
1258 * the reserved free space as used. This is really only necessary for
1259 * lazy superblock counting because it trusts the incore superblock
1260 * counters to be aboslutely correct on clean unmount.
1262 * We don't bother correcting this elsewhere for lazy superblock
1263 * counting because on mount of an unclean filesystem we reconstruct the
1264 * correct counter value and this is irrelevant.
1266 * For non-lazy counter filesystems, this doesn't matter at all because
1267 * we only every apply deltas to the superblock and hence the incore
1268 * value does not matter....
1271 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1273 cmn_err(CE_WARN
, "XFS: Unable to free reserved block pool. "
1274 "Freespace may not be correct on next mount.");
1276 error
= xfs_log_sbcount(mp
, 1);
1278 cmn_err(CE_WARN
, "XFS: Unable to update superblock counters. "
1279 "Freespace may not be correct on next mount.");
1280 xfs_unmountfs_writesb(mp
);
1281 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1282 xfs_log_unmount_write(mp
);
1283 xfs_log_unmount(mp
);
1285 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0)
1286 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1289 xfs_errortag_clearall(mp
, 0);
1295 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1297 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1298 xfs_wait_buftarg(mp
->m_logdev_targp
);
1299 if (mp
->m_rtdev_targp
)
1300 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1301 xfs_wait_buftarg(mp
->m_ddev_targp
);
1305 xfs_fs_writable(xfs_mount_t
*mp
)
1307 return !(xfs_test_for_freeze(mp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1308 (mp
->m_flags
& XFS_MOUNT_RDONLY
));
1314 * Called either periodically to keep the on disk superblock values
1315 * roughly up to date or from unmount to make sure the values are
1316 * correct on a clean unmount.
1318 * Note this code can be called during the process of freezing, so
1319 * we may need to use the transaction allocator which does not not
1320 * block when the transaction subsystem is in its frozen state.
1330 if (!xfs_fs_writable(mp
))
1333 xfs_icsb_sync_counters(mp
, 0);
1336 * we don't need to do this if we are updating the superblock
1337 * counters on every modification.
1339 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1342 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
);
1343 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1344 XFS_DEFAULT_LOG_COUNT
);
1346 xfs_trans_cancel(tp
, 0);
1350 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1352 xfs_trans_set_sync(tp
);
1353 error
= xfs_trans_commit(tp
, 0);
1358 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1364 * skip superblock write if fs is read-only, or
1365 * if we are doing a forced umount.
1367 if (!((mp
->m_flags
& XFS_MOUNT_RDONLY
) ||
1368 XFS_FORCED_SHUTDOWN(mp
))) {
1370 sbp
= xfs_getsb(mp
, 0);
1372 XFS_BUF_UNDONE(sbp
);
1373 XFS_BUF_UNREAD(sbp
);
1374 XFS_BUF_UNDELAYWRITE(sbp
);
1376 XFS_BUF_UNASYNC(sbp
);
1377 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1378 xfsbdstrat(mp
, sbp
);
1379 error
= xfs_iowait(sbp
);
1381 xfs_ioerror_alert("xfs_unmountfs_writesb",
1382 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1389 * xfs_mod_sb() can be used to copy arbitrary changes to the
1390 * in-core superblock into the superblock buffer to be logged.
1391 * It does not provide the higher level of locking that is
1392 * needed to protect the in-core superblock from concurrent
1396 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1408 bp
= xfs_trans_getsb(tp
, mp
, 0);
1409 first
= sizeof(xfs_sb_t
);
1412 /* translate/copy */
1414 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
, fields
);
1416 /* find modified range */
1418 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1419 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1420 first
= xfs_sb_info
[f
].offset
;
1422 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1423 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1424 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1426 xfs_trans_log_buf(tp
, bp
, first
, last
);
1431 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1432 * a delta to a specified field in the in-core superblock. Simply
1433 * switch on the field indicated and apply the delta to that field.
1434 * Fields are not allowed to dip below zero, so if the delta would
1435 * do this do not apply it and return EINVAL.
1437 * The m_sb_lock must be held when this routine is called.
1440 xfs_mod_incore_sb_unlocked(
1442 xfs_sb_field_t field
,
1446 int scounter
; /* short counter for 32 bit fields */
1447 long long lcounter
; /* long counter for 64 bit fields */
1448 long long res_used
, rem
;
1451 * With the in-core superblock spin lock held, switch
1452 * on the indicated field. Apply the delta to the
1453 * proper field. If the fields value would dip below
1454 * 0, then do not apply the delta and return EINVAL.
1457 case XFS_SBS_ICOUNT
:
1458 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1462 return XFS_ERROR(EINVAL
);
1464 mp
->m_sb
.sb_icount
= lcounter
;
1467 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1471 return XFS_ERROR(EINVAL
);
1473 mp
->m_sb
.sb_ifree
= lcounter
;
1475 case XFS_SBS_FDBLOCKS
:
1476 lcounter
= (long long)
1477 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1478 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1480 if (delta
> 0) { /* Putting blocks back */
1481 if (res_used
> delta
) {
1482 mp
->m_resblks_avail
+= delta
;
1484 rem
= delta
- res_used
;
1485 mp
->m_resblks_avail
= mp
->m_resblks
;
1488 } else { /* Taking blocks away */
1493 * If were out of blocks, use any available reserved blocks if
1499 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1501 return XFS_ERROR(ENOSPC
);
1503 mp
->m_resblks_avail
= lcounter
;
1505 } else { /* not reserved */
1506 return XFS_ERROR(ENOSPC
);
1511 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1513 case XFS_SBS_FREXTENTS
:
1514 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1517 return XFS_ERROR(ENOSPC
);
1519 mp
->m_sb
.sb_frextents
= lcounter
;
1521 case XFS_SBS_DBLOCKS
:
1522 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1526 return XFS_ERROR(EINVAL
);
1528 mp
->m_sb
.sb_dblocks
= lcounter
;
1530 case XFS_SBS_AGCOUNT
:
1531 scounter
= mp
->m_sb
.sb_agcount
;
1535 return XFS_ERROR(EINVAL
);
1537 mp
->m_sb
.sb_agcount
= scounter
;
1539 case XFS_SBS_IMAX_PCT
:
1540 scounter
= mp
->m_sb
.sb_imax_pct
;
1544 return XFS_ERROR(EINVAL
);
1546 mp
->m_sb
.sb_imax_pct
= scounter
;
1548 case XFS_SBS_REXTSIZE
:
1549 scounter
= mp
->m_sb
.sb_rextsize
;
1553 return XFS_ERROR(EINVAL
);
1555 mp
->m_sb
.sb_rextsize
= scounter
;
1557 case XFS_SBS_RBMBLOCKS
:
1558 scounter
= mp
->m_sb
.sb_rbmblocks
;
1562 return XFS_ERROR(EINVAL
);
1564 mp
->m_sb
.sb_rbmblocks
= scounter
;
1566 case XFS_SBS_RBLOCKS
:
1567 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1571 return XFS_ERROR(EINVAL
);
1573 mp
->m_sb
.sb_rblocks
= lcounter
;
1575 case XFS_SBS_REXTENTS
:
1576 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1580 return XFS_ERROR(EINVAL
);
1582 mp
->m_sb
.sb_rextents
= lcounter
;
1584 case XFS_SBS_REXTSLOG
:
1585 scounter
= mp
->m_sb
.sb_rextslog
;
1589 return XFS_ERROR(EINVAL
);
1591 mp
->m_sb
.sb_rextslog
= scounter
;
1595 return XFS_ERROR(EINVAL
);
1600 * xfs_mod_incore_sb() is used to change a field in the in-core
1601 * superblock structure by the specified delta. This modification
1602 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1603 * routine to do the work.
1608 xfs_sb_field_t field
,
1614 /* check for per-cpu counters */
1616 #ifdef HAVE_PERCPU_SB
1617 case XFS_SBS_ICOUNT
:
1619 case XFS_SBS_FDBLOCKS
:
1620 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1621 status
= xfs_icsb_modify_counters(mp
, field
,
1628 spin_lock(&mp
->m_sb_lock
);
1629 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1630 spin_unlock(&mp
->m_sb_lock
);
1638 * xfs_mod_incore_sb_batch() is used to change more than one field
1639 * in the in-core superblock structure at a time. This modification
1640 * is protected by a lock internal to this module. The fields and
1641 * changes to those fields are specified in the array of xfs_mod_sb
1642 * structures passed in.
1644 * Either all of the specified deltas will be applied or none of
1645 * them will. If any modified field dips below 0, then all modifications
1646 * will be backed out and EINVAL will be returned.
1649 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1655 * Loop through the array of mod structures and apply each
1656 * individually. If any fail, then back out all those
1657 * which have already been applied. Do all of this within
1658 * the scope of the m_sb_lock so that all of the changes will
1661 spin_lock(&mp
->m_sb_lock
);
1663 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1665 * Apply the delta at index n. If it fails, break
1666 * from the loop so we'll fall into the undo loop
1669 switch (msbp
->msb_field
) {
1670 #ifdef HAVE_PERCPU_SB
1671 case XFS_SBS_ICOUNT
:
1673 case XFS_SBS_FDBLOCKS
:
1674 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1675 spin_unlock(&mp
->m_sb_lock
);
1676 status
= xfs_icsb_modify_counters(mp
,
1678 msbp
->msb_delta
, rsvd
);
1679 spin_lock(&mp
->m_sb_lock
);
1685 status
= xfs_mod_incore_sb_unlocked(mp
,
1687 msbp
->msb_delta
, rsvd
);
1697 * If we didn't complete the loop above, then back out
1698 * any changes made to the superblock. If you add code
1699 * between the loop above and here, make sure that you
1700 * preserve the value of status. Loop back until
1701 * we step below the beginning of the array. Make sure
1702 * we don't touch anything back there.
1706 while (msbp
>= msb
) {
1707 switch (msbp
->msb_field
) {
1708 #ifdef HAVE_PERCPU_SB
1709 case XFS_SBS_ICOUNT
:
1711 case XFS_SBS_FDBLOCKS
:
1712 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1713 spin_unlock(&mp
->m_sb_lock
);
1714 status
= xfs_icsb_modify_counters(mp
,
1718 spin_lock(&mp
->m_sb_lock
);
1724 status
= xfs_mod_incore_sb_unlocked(mp
,
1730 ASSERT(status
== 0);
1734 spin_unlock(&mp
->m_sb_lock
);
1739 * xfs_getsb() is called to obtain the buffer for the superblock.
1740 * The buffer is returned locked and read in from disk.
1741 * The buffer should be released with a call to xfs_brelse().
1743 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1744 * the superblock buffer if it can be locked without sleeping.
1745 * If it can't then we'll return NULL.
1754 ASSERT(mp
->m_sb_bp
!= NULL
);
1756 if (flags
& XFS_BUF_TRYLOCK
) {
1757 if (!XFS_BUF_CPSEMA(bp
)) {
1761 XFS_BUF_PSEMA(bp
, PRIBIO
);
1764 ASSERT(XFS_BUF_ISDONE(bp
));
1769 * Used to free the superblock along various error paths.
1778 * Use xfs_getsb() so that the buffer will be locked
1779 * when we call xfs_buf_relse().
1781 bp
= xfs_getsb(mp
, 0);
1782 XFS_BUF_UNMANAGE(bp
);
1788 * See if the UUID is unique among mounted XFS filesystems.
1789 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1795 if (uuid_is_nil(&mp
->m_sb
.sb_uuid
)) {
1797 "XFS: Filesystem %s has nil UUID - can't mount",
1801 if (!uuid_table_insert(&mp
->m_sb
.sb_uuid
)) {
1803 "XFS: Filesystem %s has duplicate UUID - can't mount",
1811 * Used to log changes to the superblock unit and width fields which could
1812 * be altered by the mount options, as well as any potential sb_features2
1813 * fixup. Only the first superblock is updated.
1823 ASSERT(fields
& (XFS_SB_UNIT
| XFS_SB_WIDTH
| XFS_SB_UUID
|
1824 XFS_SB_FEATURES2
| XFS_SB_BAD_FEATURES2
|
1825 XFS_SB_VERSIONNUM
));
1827 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1828 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1829 XFS_DEFAULT_LOG_COUNT
);
1831 xfs_trans_cancel(tp
, 0);
1834 xfs_mod_sb(tp
, fields
);
1835 error
= xfs_trans_commit(tp
, 0);
1840 #ifdef HAVE_PERCPU_SB
1842 * Per-cpu incore superblock counters
1844 * Simple concept, difficult implementation
1846 * Basically, replace the incore superblock counters with a distributed per cpu
1847 * counter for contended fields (e.g. free block count).
1849 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1850 * hence needs to be accurately read when we are running low on space. Hence
1851 * there is a method to enable and disable the per-cpu counters based on how
1852 * much "stuff" is available in them.
1854 * Basically, a counter is enabled if there is enough free resource to justify
1855 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1856 * ENOSPC), then we disable the counters to synchronise all callers and
1857 * re-distribute the available resources.
1859 * If, once we redistributed the available resources, we still get a failure,
1860 * we disable the per-cpu counter and go through the slow path.
1862 * The slow path is the current xfs_mod_incore_sb() function. This means that
1863 * when we disable a per-cpu counter, we need to drain it's resources back to
1864 * the global superblock. We do this after disabling the counter to prevent
1865 * more threads from queueing up on the counter.
1867 * Essentially, this means that we still need a lock in the fast path to enable
1868 * synchronisation between the global counters and the per-cpu counters. This
1869 * is not a problem because the lock will be local to a CPU almost all the time
1870 * and have little contention except when we get to ENOSPC conditions.
1872 * Basically, this lock becomes a barrier that enables us to lock out the fast
1873 * path while we do things like enabling and disabling counters and
1874 * synchronising the counters.
1878 * 1. m_sb_lock before picking up per-cpu locks
1879 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1880 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1881 * 4. modifying per-cpu counters requires holding per-cpu lock
1882 * 5. modifying global counters requires holding m_sb_lock
1883 * 6. enabling or disabling a counter requires holding the m_sb_lock
1884 * and _none_ of the per-cpu locks.
1886 * Disabled counters are only ever re-enabled by a balance operation
1887 * that results in more free resources per CPU than a given threshold.
1888 * To ensure counters don't remain disabled, they are rebalanced when
1889 * the global resource goes above a higher threshold (i.e. some hysteresis
1890 * is present to prevent thrashing).
1893 #ifdef CONFIG_HOTPLUG_CPU
1895 * hot-plug CPU notifier support.
1897 * We need a notifier per filesystem as we need to be able to identify
1898 * the filesystem to balance the counters out. This is achieved by
1899 * having a notifier block embedded in the xfs_mount_t and doing pointer
1900 * magic to get the mount pointer from the notifier block address.
1903 xfs_icsb_cpu_notify(
1904 struct notifier_block
*nfb
,
1905 unsigned long action
,
1908 xfs_icsb_cnts_t
*cntp
;
1911 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1912 cntp
= (xfs_icsb_cnts_t
*)
1913 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1915 case CPU_UP_PREPARE
:
1916 case CPU_UP_PREPARE_FROZEN
:
1917 /* Easy Case - initialize the area and locks, and
1918 * then rebalance when online does everything else for us. */
1919 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1922 case CPU_ONLINE_FROZEN
:
1924 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1925 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1926 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1927 xfs_icsb_unlock(mp
);
1930 case CPU_DEAD_FROZEN
:
1931 /* Disable all the counters, then fold the dead cpu's
1932 * count into the total on the global superblock and
1933 * re-enable the counters. */
1935 spin_lock(&mp
->m_sb_lock
);
1936 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1937 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1938 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1940 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1941 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1942 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1944 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1946 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_ICOUNT
, 0);
1947 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_IFREE
, 0);
1948 xfs_icsb_balance_counter_locked(mp
, XFS_SBS_FDBLOCKS
, 0);
1949 spin_unlock(&mp
->m_sb_lock
);
1950 xfs_icsb_unlock(mp
);
1956 #endif /* CONFIG_HOTPLUG_CPU */
1959 xfs_icsb_init_counters(
1962 xfs_icsb_cnts_t
*cntp
;
1965 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
1966 if (mp
->m_sb_cnts
== NULL
)
1969 #ifdef CONFIG_HOTPLUG_CPU
1970 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
1971 mp
->m_icsb_notifier
.priority
= 0;
1972 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
1973 #endif /* CONFIG_HOTPLUG_CPU */
1975 for_each_online_cpu(i
) {
1976 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1977 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1980 mutex_init(&mp
->m_icsb_mutex
);
1983 * start with all counters disabled so that the
1984 * initial balance kicks us off correctly
1986 mp
->m_icsb_counters
= -1;
1991 xfs_icsb_reinit_counters(
1996 * start with all counters disabled so that the
1997 * initial balance kicks us off correctly
1999 mp
->m_icsb_counters
= -1;
2000 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
2001 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
2002 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
2003 xfs_icsb_unlock(mp
);
2007 xfs_icsb_destroy_counters(
2010 if (mp
->m_sb_cnts
) {
2011 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2012 free_percpu(mp
->m_sb_cnts
);
2014 mutex_destroy(&mp
->m_icsb_mutex
);
2019 xfs_icsb_cnts_t
*icsbp
)
2021 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2027 xfs_icsb_unlock_cntr(
2028 xfs_icsb_cnts_t
*icsbp
)
2030 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2035 xfs_icsb_lock_all_counters(
2038 xfs_icsb_cnts_t
*cntp
;
2041 for_each_online_cpu(i
) {
2042 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2043 xfs_icsb_lock_cntr(cntp
);
2048 xfs_icsb_unlock_all_counters(
2051 xfs_icsb_cnts_t
*cntp
;
2054 for_each_online_cpu(i
) {
2055 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2056 xfs_icsb_unlock_cntr(cntp
);
2063 xfs_icsb_cnts_t
*cnt
,
2066 xfs_icsb_cnts_t
*cntp
;
2069 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2071 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2072 xfs_icsb_lock_all_counters(mp
);
2074 for_each_online_cpu(i
) {
2075 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2076 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2077 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2078 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2081 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2082 xfs_icsb_unlock_all_counters(mp
);
2086 xfs_icsb_counter_disabled(
2088 xfs_sb_field_t field
)
2090 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2091 return test_bit(field
, &mp
->m_icsb_counters
);
2095 xfs_icsb_disable_counter(
2097 xfs_sb_field_t field
)
2099 xfs_icsb_cnts_t cnt
;
2101 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2104 * If we are already disabled, then there is nothing to do
2105 * here. We check before locking all the counters to avoid
2106 * the expensive lock operation when being called in the
2107 * slow path and the counter is already disabled. This is
2108 * safe because the only time we set or clear this state is under
2111 if (xfs_icsb_counter_disabled(mp
, field
))
2114 xfs_icsb_lock_all_counters(mp
);
2115 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2116 /* drain back to superblock */
2118 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_LAZY_COUNT
);
2120 case XFS_SBS_ICOUNT
:
2121 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2124 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2126 case XFS_SBS_FDBLOCKS
:
2127 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2134 xfs_icsb_unlock_all_counters(mp
);
2138 xfs_icsb_enable_counter(
2140 xfs_sb_field_t field
,
2144 xfs_icsb_cnts_t
*cntp
;
2147 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2149 xfs_icsb_lock_all_counters(mp
);
2150 for_each_online_cpu(i
) {
2151 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2153 case XFS_SBS_ICOUNT
:
2154 cntp
->icsb_icount
= count
+ resid
;
2157 cntp
->icsb_ifree
= count
+ resid
;
2159 case XFS_SBS_FDBLOCKS
:
2160 cntp
->icsb_fdblocks
= count
+ resid
;
2168 clear_bit(field
, &mp
->m_icsb_counters
);
2169 xfs_icsb_unlock_all_counters(mp
);
2173 xfs_icsb_sync_counters_locked(
2177 xfs_icsb_cnts_t cnt
;
2179 xfs_icsb_count(mp
, &cnt
, flags
);
2181 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2182 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2183 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2184 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2185 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2186 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2190 * Accurate update of per-cpu counters to incore superblock
2193 xfs_icsb_sync_counters(
2197 spin_lock(&mp
->m_sb_lock
);
2198 xfs_icsb_sync_counters_locked(mp
, flags
);
2199 spin_unlock(&mp
->m_sb_lock
);
2203 * Balance and enable/disable counters as necessary.
2205 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2206 * chosen to be the same number as single on disk allocation chunk per CPU, and
2207 * free blocks is something far enough zero that we aren't going thrash when we
2208 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2209 * prevent looping endlessly when xfs_alloc_space asks for more than will
2210 * be distributed to a single CPU but each CPU has enough blocks to be
2213 * Note that we can be called when counters are already disabled.
2214 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2215 * prevent locking every per-cpu counter needlessly.
2218 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2219 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2220 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2222 xfs_icsb_balance_counter_locked(
2224 xfs_sb_field_t field
,
2227 uint64_t count
, resid
;
2228 int weight
= num_online_cpus();
2229 uint64_t min
= (uint64_t)min_per_cpu
;
2231 /* disable counter and sync counter */
2232 xfs_icsb_disable_counter(mp
, field
);
2234 /* update counters - first CPU gets residual*/
2236 case XFS_SBS_ICOUNT
:
2237 count
= mp
->m_sb
.sb_icount
;
2238 resid
= do_div(count
, weight
);
2239 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2243 count
= mp
->m_sb
.sb_ifree
;
2244 resid
= do_div(count
, weight
);
2245 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2248 case XFS_SBS_FDBLOCKS
:
2249 count
= mp
->m_sb
.sb_fdblocks
;
2250 resid
= do_div(count
, weight
);
2251 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2256 count
= resid
= 0; /* quiet, gcc */
2260 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2264 xfs_icsb_balance_counter(
2266 xfs_sb_field_t fields
,
2269 spin_lock(&mp
->m_sb_lock
);
2270 xfs_icsb_balance_counter_locked(mp
, fields
, min_per_cpu
);
2271 spin_unlock(&mp
->m_sb_lock
);
2275 xfs_icsb_modify_counters(
2277 xfs_sb_field_t field
,
2281 xfs_icsb_cnts_t
*icsbp
;
2282 long long lcounter
; /* long counter for 64 bit fields */
2288 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
);
2291 * if the counter is disabled, go to slow path
2293 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2295 xfs_icsb_lock_cntr(icsbp
);
2296 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2297 xfs_icsb_unlock_cntr(icsbp
);
2302 case XFS_SBS_ICOUNT
:
2303 lcounter
= icsbp
->icsb_icount
;
2305 if (unlikely(lcounter
< 0))
2306 goto balance_counter
;
2307 icsbp
->icsb_icount
= lcounter
;
2311 lcounter
= icsbp
->icsb_ifree
;
2313 if (unlikely(lcounter
< 0))
2314 goto balance_counter
;
2315 icsbp
->icsb_ifree
= lcounter
;
2318 case XFS_SBS_FDBLOCKS
:
2319 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2321 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2323 if (unlikely(lcounter
< 0))
2324 goto balance_counter
;
2325 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2331 xfs_icsb_unlock_cntr(icsbp
);
2339 * serialise with a mutex so we don't burn lots of cpu on
2340 * the superblock lock. We still need to hold the superblock
2341 * lock, however, when we modify the global structures.
2346 * Now running atomically.
2348 * If the counter is enabled, someone has beaten us to rebalancing.
2349 * Drop the lock and try again in the fast path....
2351 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2352 xfs_icsb_unlock(mp
);
2357 * The counter is currently disabled. Because we are
2358 * running atomically here, we know a rebalance cannot
2359 * be in progress. Hence we can go straight to operating
2360 * on the global superblock. We do not call xfs_mod_incore_sb()
2361 * here even though we need to get the m_sb_lock. Doing so
2362 * will cause us to re-enter this function and deadlock.
2363 * Hence we get the m_sb_lock ourselves and then call
2364 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2365 * directly on the global counters.
2367 spin_lock(&mp
->m_sb_lock
);
2368 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2369 spin_unlock(&mp
->m_sb_lock
);
2372 * Now that we've modified the global superblock, we
2373 * may be able to re-enable the distributed counters
2374 * (e.g. lots of space just got freed). After that
2378 xfs_icsb_balance_counter(mp
, field
, 0);
2379 xfs_icsb_unlock(mp
);
2383 xfs_icsb_unlock_cntr(icsbp
);
2387 * We may have multiple threads here if multiple per-cpu
2388 * counters run dry at the same time. This will mean we can
2389 * do more balances than strictly necessary but it is not
2390 * the common slowpath case.
2395 * running atomically.
2397 * This will leave the counter in the correct state for future
2398 * accesses. After the rebalance, we simply try again and our retry
2399 * will either succeed through the fast path or slow path without
2400 * another balance operation being required.
2402 xfs_icsb_balance_counter(mp
, field
, delta
);
2403 xfs_icsb_unlock(mp
);