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"
47 STATIC
void xfs_mount_log_sbunit(xfs_mount_t
*, __int64_t
);
48 STATIC
int xfs_uuid_mount(xfs_mount_t
*);
49 STATIC
void xfs_uuid_unmount(xfs_mount_t
*mp
);
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_sync_counters(xfs_mount_t
*);
58 STATIC
int xfs_icsb_modify_counters(xfs_mount_t
*, xfs_sb_field_t
,
60 STATIC
int xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
64 #define xfs_icsb_destroy_counters(mp) do { } while (0)
65 #define xfs_icsb_balance_counter(mp, a, b, c) do { } while (0)
66 #define xfs_icsb_sync_counters(mp) do { } while (0)
67 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
73 short type
; /* 0 = integer
74 * 1 = binary / string (no translation)
77 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
78 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
79 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
80 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
81 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
82 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
83 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
84 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
85 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
86 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
87 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
88 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
89 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
90 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
91 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
92 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
93 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
94 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
95 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
96 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
97 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
98 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
99 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
100 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
101 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
102 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
103 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
104 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
105 { offsetof(xfs_sb_t
, sb_icount
), 0 },
106 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
107 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
108 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
109 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
110 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
111 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
112 { offsetof(xfs_sb_t
, sb_flags
), 0 },
113 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
114 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
115 { offsetof(xfs_sb_t
, sb_unit
), 0 },
116 { offsetof(xfs_sb_t
, sb_width
), 0 },
117 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
118 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
119 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
120 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
121 { offsetof(xfs_sb_t
, sb_features2
), 0 },
122 { sizeof(xfs_sb_t
), 0 }
126 * Return a pointer to an initialized xfs_mount structure.
133 mp
= kmem_zalloc(sizeof(xfs_mount_t
), KM_SLEEP
);
135 if (xfs_icsb_init_counters(mp
)) {
136 mp
->m_flags
|= XFS_MOUNT_NO_PERCPU_SB
;
139 AIL_LOCKINIT(&mp
->m_ail_lock
, "xfs_ail");
140 spinlock_init(&mp
->m_sb_lock
, "xfs_sb");
141 mutex_init(&mp
->m_ilock
);
142 initnsema(&mp
->m_growlock
, 1, "xfs_grow");
144 * Initialize the AIL.
146 xfs_trans_ail_init(mp
);
148 atomic_set(&mp
->m_active_trans
, 0);
154 * Free up the resources associated with a mount structure. Assume that
155 * the structure was initially zeroed, so we can tell which fields got
171 for (agno
= 0; agno
< mp
->m_maxagi
; agno
++)
172 if (mp
->m_perag
[agno
].pagb_list
)
173 kmem_free(mp
->m_perag
[agno
].pagb_list
,
174 sizeof(xfs_perag_busy_t
) *
176 kmem_free(mp
->m_perag
,
177 sizeof(xfs_perag_t
) * mp
->m_sb
.sb_agcount
);
180 AIL_LOCK_DESTROY(&mp
->m_ail_lock
);
181 spinlock_destroy(&mp
->m_sb_lock
);
182 mutex_destroy(&mp
->m_ilock
);
183 freesema(&mp
->m_growlock
);
187 if (mp
->m_fsname
!= NULL
)
188 kmem_free(mp
->m_fsname
, mp
->m_fsname_len
);
189 if (mp
->m_rtname
!= NULL
)
190 kmem_free(mp
->m_rtname
, strlen(mp
->m_rtname
) + 1);
191 if (mp
->m_logname
!= NULL
)
192 kmem_free(mp
->m_logname
, strlen(mp
->m_logname
) + 1);
195 struct bhv_vfs
*vfsp
= XFS_MTOVFS(mp
);
197 bhv_remove_all_vfsops(vfsp
, 0);
198 VFS_REMOVEBHV(vfsp
, &mp
->m_bhv
);
201 xfs_icsb_destroy_counters(mp
);
202 kmem_free(mp
, sizeof(xfs_mount_t
));
206 * Check size of device based on the (data/realtime) block count.
207 * Note: this check is used by the growfs code as well as mount.
210 xfs_sb_validate_fsb_count(
214 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
215 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
217 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
218 if (nblocks
>> (PAGE_CACHE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
220 #else /* Limited by UINT_MAX of sectors */
221 if (nblocks
<< (sbp
->sb_blocklog
- BBSHIFT
) > UINT_MAX
)
228 * Check the validity of the SB found.
231 xfs_mount_validate_sb(
237 * If the log device and data device have the
238 * same device number, the log is internal.
239 * Consequently, the sb_logstart should be non-zero. If
240 * we have a zero sb_logstart in this case, we may be trying to mount
241 * a volume filesystem in a non-volume manner.
243 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
244 xfs_fs_mount_cmn_err(flags
, "bad magic number");
245 return XFS_ERROR(EWRONGFS
);
248 if (!XFS_SB_GOOD_VERSION(sbp
)) {
249 xfs_fs_mount_cmn_err(flags
, "bad version");
250 return XFS_ERROR(EWRONGFS
);
254 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
255 xfs_fs_mount_cmn_err(flags
,
256 "filesystem is marked as having an external log; "
257 "specify logdev on the\nmount command line.");
258 return XFS_ERROR(EINVAL
);
262 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
263 xfs_fs_mount_cmn_err(flags
,
264 "filesystem is marked as having an internal log; "
265 "do not specify logdev on\nthe mount command line.");
266 return XFS_ERROR(EINVAL
);
270 * More sanity checking. These were stolen directly from
274 sbp
->sb_agcount
<= 0 ||
275 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
276 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
277 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
278 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
279 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
280 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
281 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
282 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
283 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
284 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
285 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
286 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
287 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
288 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
289 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
290 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */))) {
291 xfs_fs_mount_cmn_err(flags
, "SB sanity check 1 failed");
292 return XFS_ERROR(EFSCORRUPTED
);
296 * Sanity check AG count, size fields against data size field
299 sbp
->sb_dblocks
== 0 ||
301 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
302 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
303 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
304 xfs_fs_mount_cmn_err(flags
, "SB sanity check 2 failed");
305 return XFS_ERROR(EFSCORRUPTED
);
308 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
309 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
310 xfs_fs_mount_cmn_err(flags
,
311 "file system too large to be mounted on this system.");
312 return XFS_ERROR(E2BIG
);
315 if (unlikely(sbp
->sb_inprogress
)) {
316 xfs_fs_mount_cmn_err(flags
, "file system busy");
317 return XFS_ERROR(EFSCORRUPTED
);
321 * Version 1 directory format has never worked on Linux.
323 if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp
))) {
324 xfs_fs_mount_cmn_err(flags
,
325 "file system using version 1 directory format");
326 return XFS_ERROR(ENOSYS
);
330 * Until this is fixed only page-sized or smaller data blocks work.
332 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
333 xfs_fs_mount_cmn_err(flags
,
334 "file system with blocksize %d bytes",
336 xfs_fs_mount_cmn_err(flags
,
337 "only pagesize (%ld) or less will currently work.",
339 return XFS_ERROR(ENOSYS
);
346 xfs_initialize_perag(
349 xfs_agnumber_t agcount
)
351 xfs_agnumber_t index
, max_metadata
;
355 xfs_sb_t
*sbp
= &mp
->m_sb
;
356 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
358 /* Check to see if the filesystem can overflow 32 bit inodes */
359 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
360 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
362 /* Clear the mount flag if no inode can overflow 32 bits
363 * on this filesystem, or if specifically requested..
365 if ((vfs
->vfs_flag
& VFS_32BITINODES
) && ino
> max_inum
) {
366 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
368 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
371 /* If we can overflow then setup the ag headers accordingly */
372 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
373 /* Calculate how much should be reserved for inodes to
374 * meet the max inode percentage.
376 if (mp
->m_maxicount
) {
379 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
381 icount
+= sbp
->sb_agblocks
- 1;
382 do_div(icount
, sbp
->sb_agblocks
);
383 max_metadata
= icount
;
385 max_metadata
= agcount
;
387 for (index
= 0; index
< agcount
; index
++) {
388 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
389 if (ino
> max_inum
) {
394 /* This ag is preferred for inodes */
395 pag
= &mp
->m_perag
[index
];
396 pag
->pagi_inodeok
= 1;
397 if (index
< max_metadata
)
398 pag
->pagf_metadata
= 1;
401 /* Setup default behavior for smaller filesystems */
402 for (index
= 0; index
< agcount
; index
++) {
403 pag
= &mp
->m_perag
[index
];
404 pag
->pagi_inodeok
= 1;
413 * data - on disk version of sb
415 * dir - conversion direction: <0 - convert sb to buf
416 * >0 - convert buf to sb
417 * fields - which fields to copy (bitmask)
438 buf_ptr
= (xfs_caddr_t
)data
;
439 mem_ptr
= (xfs_caddr_t
)sb
;
442 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
443 first
= xfs_sb_info
[f
].offset
;
444 size
= xfs_sb_info
[f
+ 1].offset
- first
;
446 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
448 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
450 memcpy(mem_ptr
+ first
, buf_ptr
+ first
, size
);
452 memcpy(buf_ptr
+ first
, mem_ptr
+ first
, size
);
457 INT_XLATE(*(__uint16_t
*)(buf_ptr
+first
),
458 *(__uint16_t
*)(mem_ptr
+first
),
462 INT_XLATE(*(__uint32_t
*)(buf_ptr
+first
),
463 *(__uint32_t
*)(mem_ptr
+first
),
467 INT_XLATE(*(__uint64_t
*)(buf_ptr
+first
),
468 *(__uint64_t
*)(mem_ptr
+first
), dir
, ARCH_CONVERT
);
475 fields
&= ~(1LL << f
);
482 * Does the initial read of the superblock.
485 xfs_readsb(xfs_mount_t
*mp
, int flags
)
487 unsigned int sector_size
;
488 unsigned int extra_flags
;
493 ASSERT(mp
->m_sb_bp
== NULL
);
494 ASSERT(mp
->m_ddev_targp
!= NULL
);
497 * Allocate a (locked) buffer to hold the superblock.
498 * This will be kept around at all times to optimize
499 * access to the superblock.
501 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
502 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
504 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
505 BTOBB(sector_size
), extra_flags
);
506 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
507 xfs_fs_mount_cmn_err(flags
, "SB read failed");
508 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
511 ASSERT(XFS_BUF_ISBUSY(bp
));
512 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
515 * Initialize the mount structure from the superblock.
516 * But first do some basic consistency checking.
518 sbp
= XFS_BUF_TO_SBP(bp
);
519 xfs_xlatesb(XFS_BUF_PTR(bp
), &(mp
->m_sb
), 1, XFS_SB_ALL_BITS
);
521 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
), flags
);
523 xfs_fs_mount_cmn_err(flags
, "SB validate failed");
528 * We must be able to do sector-sized and sector-aligned IO.
530 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
531 xfs_fs_mount_cmn_err(flags
,
532 "device supports only %u byte sectors (not %u)",
533 sector_size
, mp
->m_sb
.sb_sectsize
);
539 * If device sector size is smaller than the superblock size,
540 * re-read the superblock so the buffer is correctly sized.
542 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
543 XFS_BUF_UNMANAGE(bp
);
545 sector_size
= mp
->m_sb
.sb_sectsize
;
546 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
547 BTOBB(sector_size
), extra_flags
);
548 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
549 xfs_fs_mount_cmn_err(flags
, "SB re-read failed");
550 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
553 ASSERT(XFS_BUF_ISBUSY(bp
));
554 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
557 /* Initialize per-cpu counters */
558 xfs_icsb_reinit_counters(mp
);
562 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
567 XFS_BUF_UNMANAGE(bp
);
577 * Mount initialization code establishing various mount
578 * fields from the superblock associated with the given
582 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
586 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
587 spinlock_init(&mp
->m_agirotor_lock
, "m_agirotor_lock");
588 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
589 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
590 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
591 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
592 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
593 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
594 mp
->m_litino
= sbp
->sb_inodesize
-
595 ((uint
)sizeof(xfs_dinode_core_t
) + (uint
)sizeof(xfs_agino_t
));
596 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
597 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
598 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
599 INIT_LIST_HEAD(&mp
->m_del_inodes
);
602 * Setup for attributes, in case they get created.
603 * This value is for inodes getting attributes for the first time,
604 * the per-inode value is for old attribute values.
606 ASSERT(sbp
->sb_inodesize
>= 256 && sbp
->sb_inodesize
<= 2048);
607 switch (sbp
->sb_inodesize
) {
609 mp
->m_attroffset
= XFS_LITINO(mp
) -
610 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
615 mp
->m_attroffset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
620 ASSERT(mp
->m_attroffset
< XFS_LITINO(mp
));
622 for (i
= 0; i
< 2; i
++) {
623 mp
->m_alloc_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
625 mp
->m_alloc_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
628 for (i
= 0; i
< 2; i
++) {
629 mp
->m_bmap_dmxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
631 mp
->m_bmap_dmnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
634 for (i
= 0; i
< 2; i
++) {
635 mp
->m_inobt_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
637 mp
->m_inobt_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
641 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
642 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
644 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
648 * xfs_initialize_perag_data
650 * Read in each per-ag structure so we can count up the number of
651 * allocated inodes, free inodes and used filesystem blocks as this
652 * information is no longer persistent in the superblock. Once we have
653 * this information, write it into the in-core superblock structure.
656 xfs_initialize_perag_data(xfs_mount_t
*mp
, xfs_agnumber_t agcount
)
658 xfs_agnumber_t index
;
660 xfs_sb_t
*sbp
= &mp
->m_sb
;
664 uint64_t bfreelst
= 0;
669 for (index
= 0; index
< agcount
; index
++) {
671 * read the agf, then the agi. This gets us
672 * all the inforamtion we need and populates the
673 * per-ag structures for us.
675 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
679 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
682 pag
= &mp
->m_perag
[index
];
683 ifree
+= pag
->pagi_freecount
;
684 ialloc
+= pag
->pagi_count
;
685 bfree
+= pag
->pagf_freeblks
;
686 bfreelst
+= pag
->pagf_flcount
;
687 btree
+= pag
->pagf_btreeblks
;
690 * Overwrite incore superblock counters with just-read data
693 sbp
->sb_ifree
= ifree
;
694 sbp
->sb_icount
= ialloc
;
695 sbp
->sb_fdblocks
= bfree
+ bfreelst
+ btree
;
696 XFS_SB_UNLOCK(mp
, s
);
698 /* Fixup the per-cpu counters as well. */
699 xfs_icsb_reinit_counters(mp
);
707 * This function does the following on an initial mount of a file system:
708 * - reads the superblock from disk and init the mount struct
709 * - if we're a 32-bit kernel, do a size check on the superblock
710 * so we don't mount terabyte filesystems
711 * - init mount struct realtime fields
712 * - allocate inode hash table for fs
713 * - init directory manager
714 * - perform recovery and init the log manager
723 xfs_sb_t
*sbp
= &(mp
->m_sb
);
725 bhv_vnode_t
*rvp
= NULL
;
726 int readio_log
, writeio_log
;
729 __int64_t update_flags
;
730 uint quotamount
, quotaflags
;
732 int uuid_mounted
= 0;
735 if (mp
->m_sb_bp
== NULL
) {
736 if ((error
= xfs_readsb(mp
, mfsi_flags
))) {
740 xfs_mount_common(mp
, sbp
);
743 * Check if sb_agblocks is aligned at stripe boundary
744 * If sb_agblocks is NOT aligned turn off m_dalign since
745 * allocator alignment is within an ag, therefore ag has
746 * to be aligned at stripe boundary.
749 if (mp
->m_dalign
&& !(mfsi_flags
& XFS_MFSI_SECOND
)) {
751 * If stripe unit and stripe width are not multiples
752 * of the fs blocksize turn off alignment.
754 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
755 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
756 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
758 "XFS: alignment check 1 failed");
759 error
= XFS_ERROR(EINVAL
);
762 mp
->m_dalign
= mp
->m_swidth
= 0;
765 * Convert the stripe unit and width to FSBs.
767 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
768 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
769 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
770 error
= XFS_ERROR(EINVAL
);
773 xfs_fs_cmn_err(CE_WARN
, mp
,
774 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
775 mp
->m_dalign
, mp
->m_swidth
,
780 } else if (mp
->m_dalign
) {
781 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
783 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
784 xfs_fs_cmn_err(CE_WARN
, mp
,
785 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
788 error
= XFS_ERROR(EINVAL
);
796 * Update superblock with new values
799 if (XFS_SB_VERSION_HASDALIGN(sbp
)) {
800 if (sbp
->sb_unit
!= mp
->m_dalign
) {
801 sbp
->sb_unit
= mp
->m_dalign
;
802 update_flags
|= XFS_SB_UNIT
;
804 if (sbp
->sb_width
!= mp
->m_swidth
) {
805 sbp
->sb_width
= mp
->m_swidth
;
806 update_flags
|= XFS_SB_WIDTH
;
809 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
810 XFS_SB_VERSION_HASDALIGN(&mp
->m_sb
)) {
811 mp
->m_dalign
= sbp
->sb_unit
;
812 mp
->m_swidth
= sbp
->sb_width
;
815 xfs_alloc_compute_maxlevels(mp
);
816 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
817 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
818 xfs_ialloc_compute_maxlevels(mp
);
820 if (sbp
->sb_imax_pct
) {
823 /* Make sure the maximum inode count is a multiple of the
824 * units we allocate inodes in.
827 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
829 do_div(icount
, mp
->m_ialloc_blks
);
830 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
835 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
838 * XFS uses the uuid from the superblock as the unique
839 * identifier for fsid. We can not use the uuid from the volume
840 * since a single partition filesystem is identical to a single
841 * partition volume/filesystem.
843 if ((mfsi_flags
& XFS_MFSI_SECOND
) == 0 &&
844 (mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0) {
846 if (xfs_uuid_mount(mp
)) {
847 error
= XFS_ERROR(EINVAL
);
851 ret64
= uuid_hash64(&sbp
->sb_uuid
);
852 memcpy(&vfsp
->vfs_fsid
, &ret64
, sizeof(ret64
));
856 * Set the default minimum read and write sizes unless
857 * already specified in a mount option.
858 * We use smaller I/O sizes when the file system
859 * is being used for NFS service (wsync mount option).
861 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
862 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
863 readio_log
= XFS_WSYNC_READIO_LOG
;
864 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
866 readio_log
= XFS_READIO_LOG_LARGE
;
867 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
870 readio_log
= mp
->m_readio_log
;
871 writeio_log
= mp
->m_writeio_log
;
875 * Set the number of readahead buffers to use based on
876 * physical memory size.
878 if (xfs_physmem
<= 4096) /* <= 16MB */
879 mp
->m_nreadaheads
= XFS_RW_NREADAHEAD_16MB
;
880 else if (xfs_physmem
<= 8192) /* <= 32MB */
881 mp
->m_nreadaheads
= XFS_RW_NREADAHEAD_32MB
;
883 mp
->m_nreadaheads
= XFS_RW_NREADAHEAD_K32
;
884 if (sbp
->sb_blocklog
> readio_log
) {
885 mp
->m_readio_log
= sbp
->sb_blocklog
;
887 mp
->m_readio_log
= readio_log
;
889 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
890 if (sbp
->sb_blocklog
> writeio_log
) {
891 mp
->m_writeio_log
= sbp
->sb_blocklog
;
893 mp
->m_writeio_log
= writeio_log
;
895 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
898 * Set the inode cluster size based on the physical memory
899 * size. This may still be overridden by the file system
900 * block size if it is larger than the chosen cluster size.
902 if (xfs_physmem
<= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
903 mp
->m_inode_cluster_size
= XFS_INODE_SMALL_CLUSTER_SIZE
;
905 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
908 * Set whether we're using inode alignment.
910 if (XFS_SB_VERSION_HASALIGN(&mp
->m_sb
) &&
911 mp
->m_sb
.sb_inoalignmt
>=
912 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
913 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
915 mp
->m_inoalign_mask
= 0;
917 * If we are using stripe alignment, check whether
918 * the stripe unit is a multiple of the inode alignment
920 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
921 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
922 mp
->m_sinoalign
= mp
->m_dalign
;
926 * Check that the data (and log if separate) are an ok size.
928 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
929 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
930 cmn_err(CE_WARN
, "XFS: size check 1 failed");
931 error
= XFS_ERROR(E2BIG
);
934 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
935 d
- XFS_FSS_TO_BB(mp
, 1),
936 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
940 cmn_err(CE_WARN
, "XFS: size check 2 failed");
941 if (error
== ENOSPC
) {
942 error
= XFS_ERROR(E2BIG
);
947 if (((mfsi_flags
& XFS_MFSI_CLIENT
) == 0) &&
948 mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
949 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
950 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
951 cmn_err(CE_WARN
, "XFS: size check 3 failed");
952 error
= XFS_ERROR(E2BIG
);
955 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
956 d
- XFS_FSB_TO_BB(mp
, 1),
957 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
961 cmn_err(CE_WARN
, "XFS: size check 3 failed");
962 if (error
== ENOSPC
) {
963 error
= XFS_ERROR(E2BIG
);
970 * Initialize realtime fields in the mount structure
972 if ((error
= xfs_rtmount_init(mp
))) {
973 cmn_err(CE_WARN
, "XFS: RT mount failed");
978 * For client case we are done now
980 if (mfsi_flags
& XFS_MFSI_CLIENT
) {
985 * Copies the low order bits of the timestamp and the randomly
986 * set "sequence" number out of a UUID.
988 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
991 * The vfs structure needs to have a file system independent
992 * way of checking for the invariant file system ID. Since it
993 * can't look at mount structures it has a pointer to the data
994 * in the mount structure.
996 * File systems that don't support user level file handles (i.e.
997 * all of them except for XFS) will leave vfs_altfsid as NULL.
999 vfsp
->vfs_altfsid
= (xfs_fsid_t
*)mp
->m_fixedfsid
;
1000 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
1005 * Initialize the attribute manager's entries.
1007 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
1010 * Initialize the precomputed transaction reservations values.
1015 * Allocate and initialize the inode hash table for this
1022 * Allocate and initialize the per-ag data.
1024 init_rwsem(&mp
->m_peraglock
);
1026 kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
), KM_SLEEP
);
1028 mp
->m_maxagi
= xfs_initialize_perag(vfsp
, mp
, sbp
->sb_agcount
);
1031 * log's mount-time initialization. Perform 1st part recovery if needed
1033 if (likely(sbp
->sb_logblocks
> 0)) { /* check for volume case */
1034 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
1035 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
1036 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
1038 cmn_err(CE_WARN
, "XFS: log mount failed");
1041 } else { /* No log has been defined */
1042 cmn_err(CE_WARN
, "XFS: no log defined");
1043 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW
, mp
);
1044 error
= XFS_ERROR(EFSCORRUPTED
);
1049 * Now the log is mounted, we know if it was an unclean shutdown or
1050 * not. If it was, with the first phase of recovery has completed, we
1051 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1052 * but they are recovered transactionally in the second recovery phase
1055 * Hence we can safely re-initialise incore superblock counters from
1056 * the per-ag data. These may not be correct if the filesystem was not
1057 * cleanly unmounted, so we need to wait for recovery to finish before
1060 * If the filesystem was cleanly unmounted, then we can trust the
1061 * values in the superblock to be correct and we don't need to do
1064 * If we are currently making the filesystem, the initialisation will
1065 * fail as the perag data is in an undefined state.
1068 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
1069 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp
) &&
1070 !mp
->m_sb
.sb_inprogress
) {
1071 error
= xfs_initialize_perag_data(mp
, sbp
->sb_agcount
);
1077 * Get and sanity-check the root inode.
1078 * Save the pointer to it in the mount structure.
1080 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1082 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1086 ASSERT(rip
!= NULL
);
1087 rvp
= XFS_ITOV(rip
);
1089 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1090 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1091 cmn_err(CE_WARN
, "Device %s - root %llu is not a directory",
1092 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
1093 (unsigned long long)rip
->i_ino
);
1094 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1095 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1097 error
= XFS_ERROR(EFSCORRUPTED
);
1100 mp
->m_rootip
= rip
; /* save it */
1102 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1105 * Initialize realtime inode pointers in the mount structure
1107 if ((error
= xfs_rtmount_inodes(mp
))) {
1109 * Free up the root inode.
1111 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1116 * If fs is not mounted readonly, then update the superblock
1117 * unit and width changes.
1119 if (update_flags
&& !(vfsp
->vfs_flag
& VFS_RDONLY
))
1120 xfs_mount_log_sbunit(mp
, update_flags
);
1123 * Initialise the XFS quota management subsystem for this mount
1125 if ((error
= XFS_QM_INIT(mp
, "amount
, "aflags
)))
1129 * Finish recovering the file system. This part needed to be
1130 * delayed until after the root and real-time bitmap inodes
1131 * were consistently read in.
1133 error
= xfs_log_mount_finish(mp
, mfsi_flags
);
1135 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1140 * Complete the quota initialisation, post-log-replay component.
1142 if ((error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
, mfsi_flags
)))
1146 * Now we are mounted, reserve a small amount of unused space for
1147 * privileged transactions. This is needed so that transaction
1148 * space required for critical operations can dip into this pool
1149 * when at ENOSPC. This is needed for operations like create with
1150 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1151 * are not allowed to use this reserved space.
1153 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1154 * This may drive us straight to ENOSPC on mount, but that implies
1155 * we were already there on the last unmount.
1157 resblks
= min_t(__uint64_t
, mp
->m_sb
.sb_dblocks
/ 20, 1024);
1158 xfs_reserve_blocks(mp
, &resblks
, NULL
);
1164 * Free up the root inode.
1168 xfs_log_unmount_dealloc(mp
);
1172 for (agno
= 0; agno
< sbp
->sb_agcount
; agno
++)
1173 if (mp
->m_perag
[agno
].pagb_list
)
1174 kmem_free(mp
->m_perag
[agno
].pagb_list
,
1175 sizeof(xfs_perag_busy_t
) * XFS_PAGB_NUM_SLOTS
);
1176 kmem_free(mp
->m_perag
, sbp
->sb_agcount
* sizeof(xfs_perag_t
));
1181 xfs_uuid_unmount(mp
);
1189 * This flushes out the inodes,dquots and the superblock, unmounts the
1190 * log and makes sure that incore structures are freed.
1193 xfs_unmountfs(xfs_mount_t
*mp
, struct cred
*cr
)
1195 struct bhv_vfs
*vfsp
= XFS_MTOVFS(mp
);
1196 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1202 * We can potentially deadlock here if we have an inode cluster
1203 * that has been freed has it's buffer still pinned in memory because
1204 * the transaction is still sitting in a iclog. The stale inodes
1205 * on that buffer will have their flush locks held until the
1206 * transaction hits the disk and the callbacks run. the inode
1207 * flush takes the flush lock unconditionally and with nothing to
1208 * push out the iclog we will never get that unlocked. hence we
1209 * need to force the log first.
1211 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1214 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1217 * Flush out the log synchronously so that we know for sure
1218 * that nothing is pinned. This is important because bflush()
1219 * will skip pinned buffers.
1221 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1223 xfs_binval(mp
->m_ddev_targp
);
1224 if (mp
->m_rtdev_targp
) {
1225 xfs_binval(mp
->m_rtdev_targp
);
1229 * Unreserve any blocks we have so that when we unmount we don't account
1230 * the reserved free space as used. This is really only necessary for
1231 * lazy superblock counting because it trusts the incore superblock
1232 * counters to be aboslutely correct on clean unmount.
1234 * We don't bother correcting this elsewhere for lazy superblock
1235 * counting because on mount of an unclean filesystem we reconstruct the
1236 * correct counter value and this is irrelevant.
1238 * For non-lazy counter filesystems, this doesn't matter at all because
1239 * we only every apply deltas to the superblock and hence the incore
1240 * value does not matter....
1243 xfs_reserve_blocks(mp
, &resblks
, NULL
);
1245 xfs_log_sbcount(mp
, 1);
1246 xfs_unmountfs_writesb(mp
);
1247 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1248 xfs_log_unmount(mp
); /* Done! No more fs ops. */
1253 * All inodes from this mount point should be freed.
1255 ASSERT(mp
->m_inodes
== NULL
);
1257 xfs_unmountfs_close(mp
, cr
);
1258 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0)
1259 xfs_uuid_unmount(mp
);
1261 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1263 * clear all error tags on this filesystem
1265 memcpy(&fsid
, &vfsp
->vfs_fsid
, sizeof(int64_t));
1266 xfs_errortag_clearall_umount(fsid
, mp
->m_fsname
, 0);
1269 xfs_mount_free(mp
, 1);
1274 xfs_unmountfs_close(xfs_mount_t
*mp
, struct cred
*cr
)
1276 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1277 xfs_free_buftarg(mp
->m_logdev_targp
, 1);
1278 if (mp
->m_rtdev_targp
)
1279 xfs_free_buftarg(mp
->m_rtdev_targp
, 1);
1280 xfs_free_buftarg(mp
->m_ddev_targp
, 0);
1284 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1286 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1287 xfs_wait_buftarg(mp
->m_logdev_targp
);
1288 if (mp
->m_rtdev_targp
)
1289 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1290 xfs_wait_buftarg(mp
->m_ddev_targp
);
1294 xfs_fs_writable(xfs_mount_t
*mp
)
1296 bhv_vfs_t
*vfsp
= XFS_MTOVFS(mp
);
1298 return !(vfs_test_for_freeze(vfsp
) || XFS_FORCED_SHUTDOWN(mp
) ||
1299 (vfsp
->vfs_flag
& VFS_RDONLY
));
1305 * Called either periodically to keep the on disk superblock values
1306 * roughly up to date or from unmount to make sure the values are
1307 * correct on a clean unmount.
1309 * Note this code can be called during the process of freezing, so
1310 * we may need to use the transaction allocator which does not not
1311 * block when the transaction subsystem is in its frozen state.
1321 if (!xfs_fs_writable(mp
))
1324 xfs_icsb_sync_counters(mp
);
1327 * we don't need to do this if we are updating the superblock
1328 * counters on every modification.
1330 if (!xfs_sb_version_haslazysbcount(&mp
->m_sb
))
1333 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_SB_COUNT
);
1334 error
= xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1335 XFS_DEFAULT_LOG_COUNT
);
1337 xfs_trans_cancel(tp
, 0);
1341 xfs_mod_sb(tp
, XFS_SB_IFREE
| XFS_SB_ICOUNT
| XFS_SB_FDBLOCKS
);
1343 xfs_trans_set_sync(tp
);
1344 xfs_trans_commit(tp
, 0);
1350 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1357 * skip superblock write if fs is read-only, or
1358 * if we are doing a forced umount.
1360 if (!(XFS_MTOVFS(mp
)->vfs_flag
& VFS_RDONLY
||
1361 XFS_FORCED_SHUTDOWN(mp
))) {
1363 sbp
= xfs_getsb(mp
, 0);
1364 sb
= XFS_BUF_TO_SBP(sbp
);
1367 * mark shared-readonly if desired
1369 if (mp
->m_mk_sharedro
) {
1370 if (!(sb
->sb_flags
& XFS_SBF_READONLY
))
1371 sb
->sb_flags
|= XFS_SBF_READONLY
;
1372 if (!XFS_SB_VERSION_HASSHARED(sb
))
1373 XFS_SB_VERSION_ADDSHARED(sb
);
1374 xfs_fs_cmn_err(CE_NOTE
, mp
,
1375 "Unmounting, marking shared read-only");
1378 XFS_BUF_UNDONE(sbp
);
1379 XFS_BUF_UNREAD(sbp
);
1380 XFS_BUF_UNDELAYWRITE(sbp
);
1382 XFS_BUF_UNASYNC(sbp
);
1383 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1384 xfsbdstrat(mp
, sbp
);
1385 /* Nevermind errors we might get here. */
1386 error
= xfs_iowait(sbp
);
1388 xfs_ioerror_alert("xfs_unmountfs_writesb",
1389 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1390 if (error
&& mp
->m_mk_sharedro
)
1391 xfs_fs_cmn_err(CE_ALERT
, mp
, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1398 * xfs_mod_sb() can be used to copy arbitrary changes to the
1399 * in-core superblock into the superblock buffer to be logged.
1400 * It does not provide the higher level of locking that is
1401 * needed to protect the in-core superblock from concurrent
1405 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1418 bp
= xfs_trans_getsb(tp
, mp
, 0);
1419 sbp
= XFS_BUF_TO_SBP(bp
);
1420 first
= sizeof(xfs_sb_t
);
1423 /* translate/copy */
1425 xfs_xlatesb(XFS_BUF_PTR(bp
), &(mp
->m_sb
), -1, fields
);
1427 /* find modified range */
1429 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1430 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1431 first
= xfs_sb_info
[f
].offset
;
1433 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1434 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1435 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1437 xfs_trans_log_buf(tp
, bp
, first
, last
);
1442 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1443 * a delta to a specified field in the in-core superblock. Simply
1444 * switch on the field indicated and apply the delta to that field.
1445 * Fields are not allowed to dip below zero, so if the delta would
1446 * do this do not apply it and return EINVAL.
1448 * The SB_LOCK must be held when this routine is called.
1451 xfs_mod_incore_sb_unlocked(
1453 xfs_sb_field_t field
,
1457 int scounter
; /* short counter for 32 bit fields */
1458 long long lcounter
; /* long counter for 64 bit fields */
1459 long long res_used
, rem
;
1462 * With the in-core superblock spin lock held, switch
1463 * on the indicated field. Apply the delta to the
1464 * proper field. If the fields value would dip below
1465 * 0, then do not apply the delta and return EINVAL.
1468 case XFS_SBS_ICOUNT
:
1469 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1473 return XFS_ERROR(EINVAL
);
1475 mp
->m_sb
.sb_icount
= lcounter
;
1478 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1482 return XFS_ERROR(EINVAL
);
1484 mp
->m_sb
.sb_ifree
= lcounter
;
1486 case XFS_SBS_FDBLOCKS
:
1487 lcounter
= (long long)
1488 mp
->m_sb
.sb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
1489 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1491 if (delta
> 0) { /* Putting blocks back */
1492 if (res_used
> delta
) {
1493 mp
->m_resblks_avail
+= delta
;
1495 rem
= delta
- res_used
;
1496 mp
->m_resblks_avail
= mp
->m_resblks
;
1499 } else { /* Taking blocks away */
1504 * If were out of blocks, use any available reserved blocks if
1510 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1512 return XFS_ERROR(ENOSPC
);
1514 mp
->m_resblks_avail
= lcounter
;
1516 } else { /* not reserved */
1517 return XFS_ERROR(ENOSPC
);
1522 mp
->m_sb
.sb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
1524 case XFS_SBS_FREXTENTS
:
1525 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1528 return XFS_ERROR(ENOSPC
);
1530 mp
->m_sb
.sb_frextents
= lcounter
;
1532 case XFS_SBS_DBLOCKS
:
1533 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1537 return XFS_ERROR(EINVAL
);
1539 mp
->m_sb
.sb_dblocks
= lcounter
;
1541 case XFS_SBS_AGCOUNT
:
1542 scounter
= mp
->m_sb
.sb_agcount
;
1546 return XFS_ERROR(EINVAL
);
1548 mp
->m_sb
.sb_agcount
= scounter
;
1550 case XFS_SBS_IMAX_PCT
:
1551 scounter
= mp
->m_sb
.sb_imax_pct
;
1555 return XFS_ERROR(EINVAL
);
1557 mp
->m_sb
.sb_imax_pct
= scounter
;
1559 case XFS_SBS_REXTSIZE
:
1560 scounter
= mp
->m_sb
.sb_rextsize
;
1564 return XFS_ERROR(EINVAL
);
1566 mp
->m_sb
.sb_rextsize
= scounter
;
1568 case XFS_SBS_RBMBLOCKS
:
1569 scounter
= mp
->m_sb
.sb_rbmblocks
;
1573 return XFS_ERROR(EINVAL
);
1575 mp
->m_sb
.sb_rbmblocks
= scounter
;
1577 case XFS_SBS_RBLOCKS
:
1578 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1582 return XFS_ERROR(EINVAL
);
1584 mp
->m_sb
.sb_rblocks
= lcounter
;
1586 case XFS_SBS_REXTENTS
:
1587 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1591 return XFS_ERROR(EINVAL
);
1593 mp
->m_sb
.sb_rextents
= lcounter
;
1595 case XFS_SBS_REXTSLOG
:
1596 scounter
= mp
->m_sb
.sb_rextslog
;
1600 return XFS_ERROR(EINVAL
);
1602 mp
->m_sb
.sb_rextslog
= scounter
;
1606 return XFS_ERROR(EINVAL
);
1611 * xfs_mod_incore_sb() is used to change a field in the in-core
1612 * superblock structure by the specified delta. This modification
1613 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1614 * routine to do the work.
1619 xfs_sb_field_t field
,
1626 /* check for per-cpu counters */
1628 #ifdef HAVE_PERCPU_SB
1629 case XFS_SBS_ICOUNT
:
1631 case XFS_SBS_FDBLOCKS
:
1632 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1633 status
= xfs_icsb_modify_counters(mp
, field
,
1640 s
= XFS_SB_LOCK(mp
);
1641 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1642 XFS_SB_UNLOCK(mp
, s
);
1650 * xfs_mod_incore_sb_batch() is used to change more than one field
1651 * in the in-core superblock structure at a time. This modification
1652 * is protected by a lock internal to this module. The fields and
1653 * changes to those fields are specified in the array of xfs_mod_sb
1654 * structures passed in.
1656 * Either all of the specified deltas will be applied or none of
1657 * them will. If any modified field dips below 0, then all modifications
1658 * will be backed out and EINVAL will be returned.
1661 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1668 * Loop through the array of mod structures and apply each
1669 * individually. If any fail, then back out all those
1670 * which have already been applied. Do all of this within
1671 * the scope of the SB_LOCK so that all of the changes will
1674 s
= XFS_SB_LOCK(mp
);
1676 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1678 * Apply the delta at index n. If it fails, break
1679 * from the loop so we'll fall into the undo loop
1682 switch (msbp
->msb_field
) {
1683 #ifdef HAVE_PERCPU_SB
1684 case XFS_SBS_ICOUNT
:
1686 case XFS_SBS_FDBLOCKS
:
1687 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1688 XFS_SB_UNLOCK(mp
, s
);
1689 status
= xfs_icsb_modify_counters(mp
,
1691 msbp
->msb_delta
, rsvd
);
1692 s
= XFS_SB_LOCK(mp
);
1698 status
= xfs_mod_incore_sb_unlocked(mp
,
1700 msbp
->msb_delta
, rsvd
);
1710 * If we didn't complete the loop above, then back out
1711 * any changes made to the superblock. If you add code
1712 * between the loop above and here, make sure that you
1713 * preserve the value of status. Loop back until
1714 * we step below the beginning of the array. Make sure
1715 * we don't touch anything back there.
1719 while (msbp
>= msb
) {
1720 switch (msbp
->msb_field
) {
1721 #ifdef HAVE_PERCPU_SB
1722 case XFS_SBS_ICOUNT
:
1724 case XFS_SBS_FDBLOCKS
:
1725 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1726 XFS_SB_UNLOCK(mp
, s
);
1727 status
= xfs_icsb_modify_counters(mp
,
1731 s
= XFS_SB_LOCK(mp
);
1737 status
= xfs_mod_incore_sb_unlocked(mp
,
1743 ASSERT(status
== 0);
1747 XFS_SB_UNLOCK(mp
, s
);
1752 * xfs_getsb() is called to obtain the buffer for the superblock.
1753 * The buffer is returned locked and read in from disk.
1754 * The buffer should be released with a call to xfs_brelse().
1756 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1757 * the superblock buffer if it can be locked without sleeping.
1758 * If it can't then we'll return NULL.
1767 ASSERT(mp
->m_sb_bp
!= NULL
);
1769 if (flags
& XFS_BUF_TRYLOCK
) {
1770 if (!XFS_BUF_CPSEMA(bp
)) {
1774 XFS_BUF_PSEMA(bp
, PRIBIO
);
1777 ASSERT(XFS_BUF_ISDONE(bp
));
1782 * Used to free the superblock along various error paths.
1791 * Use xfs_getsb() so that the buffer will be locked
1792 * when we call xfs_buf_relse().
1794 bp
= xfs_getsb(mp
, 0);
1795 XFS_BUF_UNMANAGE(bp
);
1801 * See if the UUID is unique among mounted XFS filesystems.
1802 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1808 if (uuid_is_nil(&mp
->m_sb
.sb_uuid
)) {
1810 "XFS: Filesystem %s has nil UUID - can't mount",
1814 if (!uuid_table_insert(&mp
->m_sb
.sb_uuid
)) {
1816 "XFS: Filesystem %s has duplicate UUID - can't mount",
1824 * Remove filesystem from the UUID table.
1830 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1834 * Used to log changes to the superblock unit and width fields which could
1835 * be altered by the mount options. Only the first superblock is updated.
1838 xfs_mount_log_sbunit(
1844 ASSERT(fields
& (XFS_SB_UNIT
|XFS_SB_WIDTH
|XFS_SB_UUID
));
1846 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1847 if (xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1848 XFS_DEFAULT_LOG_COUNT
)) {
1849 xfs_trans_cancel(tp
, 0);
1852 xfs_mod_sb(tp
, fields
);
1853 xfs_trans_commit(tp
, 0);
1857 #ifdef HAVE_PERCPU_SB
1859 * Per-cpu incore superblock counters
1861 * Simple concept, difficult implementation
1863 * Basically, replace the incore superblock counters with a distributed per cpu
1864 * counter for contended fields (e.g. free block count).
1866 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1867 * hence needs to be accurately read when we are running low on space. Hence
1868 * there is a method to enable and disable the per-cpu counters based on how
1869 * much "stuff" is available in them.
1871 * Basically, a counter is enabled if there is enough free resource to justify
1872 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1873 * ENOSPC), then we disable the counters to synchronise all callers and
1874 * re-distribute the available resources.
1876 * If, once we redistributed the available resources, we still get a failure,
1877 * we disable the per-cpu counter and go through the slow path.
1879 * The slow path is the current xfs_mod_incore_sb() function. This means that
1880 * when we disable a per-cpu counter, we need to drain it's resources back to
1881 * the global superblock. We do this after disabling the counter to prevent
1882 * more threads from queueing up on the counter.
1884 * Essentially, this means that we still need a lock in the fast path to enable
1885 * synchronisation between the global counters and the per-cpu counters. This
1886 * is not a problem because the lock will be local to a CPU almost all the time
1887 * and have little contention except when we get to ENOSPC conditions.
1889 * Basically, this lock becomes a barrier that enables us to lock out the fast
1890 * path while we do things like enabling and disabling counters and
1891 * synchronising the counters.
1895 * 1. XFS_SB_LOCK() before picking up per-cpu locks
1896 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1897 * 3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
1898 * 4. modifying per-cpu counters requires holding per-cpu lock
1899 * 5. modifying global counters requires holding XFS_SB_LOCK
1900 * 6. enabling or disabling a counter requires holding the XFS_SB_LOCK
1901 * and _none_ of the per-cpu locks.
1903 * Disabled counters are only ever re-enabled by a balance operation
1904 * that results in more free resources per CPU than a given threshold.
1905 * To ensure counters don't remain disabled, they are rebalanced when
1906 * the global resource goes above a higher threshold (i.e. some hysteresis
1907 * is present to prevent thrashing).
1910 #ifdef CONFIG_HOTPLUG_CPU
1912 * hot-plug CPU notifier support.
1914 * We need a notifier per filesystem as we need to be able to identify
1915 * the filesystem to balance the counters out. This is achieved by
1916 * having a notifier block embedded in the xfs_mount_t and doing pointer
1917 * magic to get the mount pointer from the notifier block address.
1920 xfs_icsb_cpu_notify(
1921 struct notifier_block
*nfb
,
1922 unsigned long action
,
1925 xfs_icsb_cnts_t
*cntp
;
1929 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1930 cntp
= (xfs_icsb_cnts_t
*)
1931 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1933 case CPU_UP_PREPARE
:
1934 case CPU_UP_PREPARE_FROZEN
:
1935 /* Easy Case - initialize the area and locks, and
1936 * then rebalance when online does everything else for us. */
1937 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1940 case CPU_ONLINE_FROZEN
:
1942 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0, 0);
1943 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0, 0);
1944 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0, 0);
1945 xfs_icsb_unlock(mp
);
1948 case CPU_DEAD_FROZEN
:
1949 /* Disable all the counters, then fold the dead cpu's
1950 * count into the total on the global superblock and
1951 * re-enable the counters. */
1953 s
= XFS_SB_LOCK(mp
);
1954 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1955 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1956 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1958 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1959 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1960 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1962 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1964 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
,
1965 XFS_ICSB_SB_LOCKED
, 0);
1966 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
,
1967 XFS_ICSB_SB_LOCKED
, 0);
1968 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
,
1969 XFS_ICSB_SB_LOCKED
, 0);
1970 XFS_SB_UNLOCK(mp
, s
);
1971 xfs_icsb_unlock(mp
);
1977 #endif /* CONFIG_HOTPLUG_CPU */
1980 xfs_icsb_init_counters(
1983 xfs_icsb_cnts_t
*cntp
;
1986 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
1987 if (mp
->m_sb_cnts
== NULL
)
1990 #ifdef CONFIG_HOTPLUG_CPU
1991 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
1992 mp
->m_icsb_notifier
.priority
= 0;
1993 register_hotcpu_notifier(&mp
->m_icsb_notifier
);
1994 #endif /* CONFIG_HOTPLUG_CPU */
1996 for_each_online_cpu(i
) {
1997 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1998 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
2001 mutex_init(&mp
->m_icsb_mutex
);
2004 * start with all counters disabled so that the
2005 * initial balance kicks us off correctly
2007 mp
->m_icsb_counters
= -1;
2012 xfs_icsb_reinit_counters(
2017 * start with all counters disabled so that the
2018 * initial balance kicks us off correctly
2020 mp
->m_icsb_counters
= -1;
2021 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0, 0);
2022 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0, 0);
2023 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0, 0);
2024 xfs_icsb_unlock(mp
);
2028 xfs_icsb_destroy_counters(
2031 if (mp
->m_sb_cnts
) {
2032 unregister_hotcpu_notifier(&mp
->m_icsb_notifier
);
2033 free_percpu(mp
->m_sb_cnts
);
2035 mutex_destroy(&mp
->m_icsb_mutex
);
2040 xfs_icsb_cnts_t
*icsbp
)
2042 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
2048 xfs_icsb_unlock_cntr(
2049 xfs_icsb_cnts_t
*icsbp
)
2051 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
2056 xfs_icsb_lock_all_counters(
2059 xfs_icsb_cnts_t
*cntp
;
2062 for_each_online_cpu(i
) {
2063 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2064 xfs_icsb_lock_cntr(cntp
);
2069 xfs_icsb_unlock_all_counters(
2072 xfs_icsb_cnts_t
*cntp
;
2075 for_each_online_cpu(i
) {
2076 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2077 xfs_icsb_unlock_cntr(cntp
);
2084 xfs_icsb_cnts_t
*cnt
,
2087 xfs_icsb_cnts_t
*cntp
;
2090 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
2092 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2093 xfs_icsb_lock_all_counters(mp
);
2095 for_each_online_cpu(i
) {
2096 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
2097 cnt
->icsb_icount
+= cntp
->icsb_icount
;
2098 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
2099 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
2102 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
2103 xfs_icsb_unlock_all_counters(mp
);
2107 xfs_icsb_counter_disabled(
2109 xfs_sb_field_t field
)
2111 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2112 return test_bit(field
, &mp
->m_icsb_counters
);
2116 xfs_icsb_disable_counter(
2118 xfs_sb_field_t field
)
2120 xfs_icsb_cnts_t cnt
;
2122 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2125 * If we are already disabled, then there is nothing to do
2126 * here. We check before locking all the counters to avoid
2127 * the expensive lock operation when being called in the
2128 * slow path and the counter is already disabled. This is
2129 * safe because the only time we set or clear this state is under
2132 if (xfs_icsb_counter_disabled(mp
, field
))
2135 xfs_icsb_lock_all_counters(mp
);
2136 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
2137 /* drain back to superblock */
2139 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_SB_LOCKED
|XFS_ICSB_LAZY_COUNT
);
2141 case XFS_SBS_ICOUNT
:
2142 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2145 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2147 case XFS_SBS_FDBLOCKS
:
2148 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2155 xfs_icsb_unlock_all_counters(mp
);
2161 xfs_icsb_enable_counter(
2163 xfs_sb_field_t field
,
2167 xfs_icsb_cnts_t
*cntp
;
2170 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
2172 xfs_icsb_lock_all_counters(mp
);
2173 for_each_online_cpu(i
) {
2174 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
2176 case XFS_SBS_ICOUNT
:
2177 cntp
->icsb_icount
= count
+ resid
;
2180 cntp
->icsb_ifree
= count
+ resid
;
2182 case XFS_SBS_FDBLOCKS
:
2183 cntp
->icsb_fdblocks
= count
+ resid
;
2191 clear_bit(field
, &mp
->m_icsb_counters
);
2192 xfs_icsb_unlock_all_counters(mp
);
2196 xfs_icsb_sync_counters_flags(
2200 xfs_icsb_cnts_t cnt
;
2203 /* Pass 1: lock all counters */
2204 if ((flags
& XFS_ICSB_SB_LOCKED
) == 0)
2205 s
= XFS_SB_LOCK(mp
);
2207 xfs_icsb_count(mp
, &cnt
, flags
);
2209 /* Step 3: update mp->m_sb fields */
2210 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
2211 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
2212 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
2213 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
2214 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
2215 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
2217 if ((flags
& XFS_ICSB_SB_LOCKED
) == 0)
2218 XFS_SB_UNLOCK(mp
, s
);
2222 * Accurate update of per-cpu counters to incore superblock
2225 xfs_icsb_sync_counters(
2228 xfs_icsb_sync_counters_flags(mp
, 0);
2232 * Balance and enable/disable counters as necessary.
2234 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2235 * chosen to be the same number as single on disk allocation chunk per CPU, and
2236 * free blocks is something far enough zero that we aren't going thrash when we
2237 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2238 * prevent looping endlessly when xfs_alloc_space asks for more than will
2239 * be distributed to a single CPU but each CPU has enough blocks to be
2242 * Note that we can be called when counters are already disabled.
2243 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2244 * prevent locking every per-cpu counter needlessly.
2247 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2248 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2249 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2251 xfs_icsb_balance_counter(
2253 xfs_sb_field_t field
,
2257 uint64_t count
, resid
;
2258 int weight
= num_online_cpus();
2260 uint64_t min
= (uint64_t)min_per_cpu
;
2262 if (!(flags
& XFS_ICSB_SB_LOCKED
))
2263 s
= XFS_SB_LOCK(mp
);
2265 /* disable counter and sync counter */
2266 xfs_icsb_disable_counter(mp
, field
);
2268 /* update counters - first CPU gets residual*/
2270 case XFS_SBS_ICOUNT
:
2271 count
= mp
->m_sb
.sb_icount
;
2272 resid
= do_div(count
, weight
);
2273 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2277 count
= mp
->m_sb
.sb_ifree
;
2278 resid
= do_div(count
, weight
);
2279 if (count
< max(min
, XFS_ICSB_INO_CNTR_REENABLE
))
2282 case XFS_SBS_FDBLOCKS
:
2283 count
= mp
->m_sb
.sb_fdblocks
;
2284 resid
= do_div(count
, weight
);
2285 if (count
< max(min
, XFS_ICSB_FDBLK_CNTR_REENABLE(mp
)))
2290 count
= resid
= 0; /* quiet, gcc */
2294 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2296 if (!(flags
& XFS_ICSB_SB_LOCKED
))
2297 XFS_SB_UNLOCK(mp
, s
);
2301 xfs_icsb_modify_counters(
2303 xfs_sb_field_t field
,
2307 xfs_icsb_cnts_t
*icsbp
;
2308 long long lcounter
; /* long counter for 64 bit fields */
2309 int cpu
, ret
= 0, s
;
2314 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
);
2317 * if the counter is disabled, go to slow path
2319 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2321 xfs_icsb_lock_cntr(icsbp
);
2322 if (unlikely(xfs_icsb_counter_disabled(mp
, field
))) {
2323 xfs_icsb_unlock_cntr(icsbp
);
2328 case XFS_SBS_ICOUNT
:
2329 lcounter
= icsbp
->icsb_icount
;
2331 if (unlikely(lcounter
< 0))
2332 goto balance_counter
;
2333 icsbp
->icsb_icount
= lcounter
;
2337 lcounter
= icsbp
->icsb_ifree
;
2339 if (unlikely(lcounter
< 0))
2340 goto balance_counter
;
2341 icsbp
->icsb_ifree
= lcounter
;
2344 case XFS_SBS_FDBLOCKS
:
2345 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2347 lcounter
= icsbp
->icsb_fdblocks
- XFS_ALLOC_SET_ASIDE(mp
);
2349 if (unlikely(lcounter
< 0))
2350 goto balance_counter
;
2351 icsbp
->icsb_fdblocks
= lcounter
+ XFS_ALLOC_SET_ASIDE(mp
);
2357 xfs_icsb_unlock_cntr(icsbp
);
2365 * serialise with a mutex so we don't burn lots of cpu on
2366 * the superblock lock. We still need to hold the superblock
2367 * lock, however, when we modify the global structures.
2372 * Now running atomically.
2374 * If the counter is enabled, someone has beaten us to rebalancing.
2375 * Drop the lock and try again in the fast path....
2377 if (!(xfs_icsb_counter_disabled(mp
, field
))) {
2378 xfs_icsb_unlock(mp
);
2383 * The counter is currently disabled. Because we are
2384 * running atomically here, we know a rebalance cannot
2385 * be in progress. Hence we can go straight to operating
2386 * on the global superblock. We do not call xfs_mod_incore_sb()
2387 * here even though we need to get the SB_LOCK. Doing so
2388 * will cause us to re-enter this function and deadlock.
2389 * Hence we get the SB_LOCK ourselves and then call
2390 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2391 * directly on the global counters.
2393 s
= XFS_SB_LOCK(mp
);
2394 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2395 XFS_SB_UNLOCK(mp
, s
);
2398 * Now that we've modified the global superblock, we
2399 * may be able to re-enable the distributed counters
2400 * (e.g. lots of space just got freed). After that
2404 xfs_icsb_balance_counter(mp
, field
, 0, 0);
2405 xfs_icsb_unlock(mp
);
2409 xfs_icsb_unlock_cntr(icsbp
);
2413 * We may have multiple threads here if multiple per-cpu
2414 * counters run dry at the same time. This will mean we can
2415 * do more balances than strictly necessary but it is not
2416 * the common slowpath case.
2421 * running atomically.
2423 * This will leave the counter in the correct state for future
2424 * accesses. After the rebalance, we simply try again and our retry
2425 * will either succeed through the fast path or slow path without
2426 * another balance operation being required.
2428 xfs_icsb_balance_counter(mp
, field
, 0, delta
);
2429 xfs_icsb_unlock(mp
);