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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_alloc.h"
42 #include "xfs_rtalloc.h"
44 #include "xfs_error.h"
46 #include "xfs_quota.h"
47 #include "xfs_fsops.h"
49 STATIC
void xfs_mount_log_sbunit(xfs_mount_t
*, __int64_t
);
50 STATIC
int xfs_uuid_mount(xfs_mount_t
*);
51 STATIC
void xfs_uuid_unmount(xfs_mount_t
*mp
);
52 STATIC
void xfs_unmountfs_wait(xfs_mount_t
*);
56 STATIC
void xfs_icsb_destroy_counters(xfs_mount_t
*);
57 STATIC
void xfs_icsb_balance_counter(xfs_mount_t
*, xfs_sb_field_t
, int);
58 STATIC
void xfs_icsb_sync_counters(xfs_mount_t
*);
59 STATIC
int xfs_icsb_modify_counters(xfs_mount_t
*, xfs_sb_field_t
,
61 STATIC
int xfs_icsb_modify_counters_locked(xfs_mount_t
*, xfs_sb_field_t
,
63 STATIC
int xfs_icsb_disable_counter(xfs_mount_t
*, xfs_sb_field_t
);
67 #define xfs_icsb_destroy_counters(mp) do { } while (0)
68 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
69 #define xfs_icsb_sync_counters(mp) do { } while (0)
70 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
71 #define xfs_icsb_modify_counters_locked(mp, a, b, c) do { } while (0)
77 short type
; /* 0 = integer
78 * 1 = binary / string (no translation)
81 { offsetof(xfs_sb_t
, sb_magicnum
), 0 },
82 { offsetof(xfs_sb_t
, sb_blocksize
), 0 },
83 { offsetof(xfs_sb_t
, sb_dblocks
), 0 },
84 { offsetof(xfs_sb_t
, sb_rblocks
), 0 },
85 { offsetof(xfs_sb_t
, sb_rextents
), 0 },
86 { offsetof(xfs_sb_t
, sb_uuid
), 1 },
87 { offsetof(xfs_sb_t
, sb_logstart
), 0 },
88 { offsetof(xfs_sb_t
, sb_rootino
), 0 },
89 { offsetof(xfs_sb_t
, sb_rbmino
), 0 },
90 { offsetof(xfs_sb_t
, sb_rsumino
), 0 },
91 { offsetof(xfs_sb_t
, sb_rextsize
), 0 },
92 { offsetof(xfs_sb_t
, sb_agblocks
), 0 },
93 { offsetof(xfs_sb_t
, sb_agcount
), 0 },
94 { offsetof(xfs_sb_t
, sb_rbmblocks
), 0 },
95 { offsetof(xfs_sb_t
, sb_logblocks
), 0 },
96 { offsetof(xfs_sb_t
, sb_versionnum
), 0 },
97 { offsetof(xfs_sb_t
, sb_sectsize
), 0 },
98 { offsetof(xfs_sb_t
, sb_inodesize
), 0 },
99 { offsetof(xfs_sb_t
, sb_inopblock
), 0 },
100 { offsetof(xfs_sb_t
, sb_fname
[0]), 1 },
101 { offsetof(xfs_sb_t
, sb_blocklog
), 0 },
102 { offsetof(xfs_sb_t
, sb_sectlog
), 0 },
103 { offsetof(xfs_sb_t
, sb_inodelog
), 0 },
104 { offsetof(xfs_sb_t
, sb_inopblog
), 0 },
105 { offsetof(xfs_sb_t
, sb_agblklog
), 0 },
106 { offsetof(xfs_sb_t
, sb_rextslog
), 0 },
107 { offsetof(xfs_sb_t
, sb_inprogress
), 0 },
108 { offsetof(xfs_sb_t
, sb_imax_pct
), 0 },
109 { offsetof(xfs_sb_t
, sb_icount
), 0 },
110 { offsetof(xfs_sb_t
, sb_ifree
), 0 },
111 { offsetof(xfs_sb_t
, sb_fdblocks
), 0 },
112 { offsetof(xfs_sb_t
, sb_frextents
), 0 },
113 { offsetof(xfs_sb_t
, sb_uquotino
), 0 },
114 { offsetof(xfs_sb_t
, sb_gquotino
), 0 },
115 { offsetof(xfs_sb_t
, sb_qflags
), 0 },
116 { offsetof(xfs_sb_t
, sb_flags
), 0 },
117 { offsetof(xfs_sb_t
, sb_shared_vn
), 0 },
118 { offsetof(xfs_sb_t
, sb_inoalignmt
), 0 },
119 { offsetof(xfs_sb_t
, sb_unit
), 0 },
120 { offsetof(xfs_sb_t
, sb_width
), 0 },
121 { offsetof(xfs_sb_t
, sb_dirblklog
), 0 },
122 { offsetof(xfs_sb_t
, sb_logsectlog
), 0 },
123 { offsetof(xfs_sb_t
, sb_logsectsize
),0 },
124 { offsetof(xfs_sb_t
, sb_logsunit
), 0 },
125 { offsetof(xfs_sb_t
, sb_features2
), 0 },
126 { sizeof(xfs_sb_t
), 0 }
130 * Return a pointer to an initialized xfs_mount structure.
137 mp
= kmem_zalloc(sizeof(xfs_mount_t
), KM_SLEEP
);
139 if (xfs_icsb_init_counters(mp
)) {
140 mp
->m_flags
|= XFS_MOUNT_NO_PERCPU_SB
;
143 AIL_LOCKINIT(&mp
->m_ail_lock
, "xfs_ail");
144 spinlock_init(&mp
->m_sb_lock
, "xfs_sb");
145 mutex_init(&mp
->m_ilock
);
146 initnsema(&mp
->m_growlock
, 1, "xfs_grow");
148 * Initialize the AIL.
150 xfs_trans_ail_init(mp
);
152 atomic_set(&mp
->m_active_trans
, 0);
158 * Free up the resources associated with a mount structure. Assume that
159 * the structure was initially zeroed, so we can tell which fields got
175 for (agno
= 0; agno
< mp
->m_maxagi
; agno
++)
176 if (mp
->m_perag
[agno
].pagb_list
)
177 kmem_free(mp
->m_perag
[agno
].pagb_list
,
178 sizeof(xfs_perag_busy_t
) *
180 kmem_free(mp
->m_perag
,
181 sizeof(xfs_perag_t
) * mp
->m_sb
.sb_agcount
);
184 AIL_LOCK_DESTROY(&mp
->m_ail_lock
);
185 spinlock_destroy(&mp
->m_sb_lock
);
186 mutex_destroy(&mp
->m_ilock
);
187 freesema(&mp
->m_growlock
);
191 if (mp
->m_fsname
!= NULL
)
192 kmem_free(mp
->m_fsname
, mp
->m_fsname_len
);
193 if (mp
->m_rtname
!= NULL
)
194 kmem_free(mp
->m_rtname
, strlen(mp
->m_rtname
) + 1);
195 if (mp
->m_logname
!= NULL
)
196 kmem_free(mp
->m_logname
, strlen(mp
->m_logname
) + 1);
199 struct vfs
*vfsp
= XFS_MTOVFS(mp
);
201 bhv_remove_all_vfsops(vfsp
, 0);
202 VFS_REMOVEBHV(vfsp
, &mp
->m_bhv
);
205 xfs_icsb_destroy_counters(mp
);
206 kmem_free(mp
, sizeof(xfs_mount_t
));
211 * Check the validity of the SB found.
214 xfs_mount_validate_sb(
219 * If the log device and data device have the
220 * same device number, the log is internal.
221 * Consequently, the sb_logstart should be non-zero. If
222 * we have a zero sb_logstart in this case, we may be trying to mount
223 * a volume filesystem in a non-volume manner.
225 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
226 cmn_err(CE_WARN
, "XFS: bad magic number");
227 return XFS_ERROR(EWRONGFS
);
230 if (!XFS_SB_GOOD_VERSION(sbp
)) {
231 cmn_err(CE_WARN
, "XFS: bad version");
232 return XFS_ERROR(EWRONGFS
);
236 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
238 "XFS: filesystem is marked as having an external log; "
239 "specify logdev on the\nmount command line.");
240 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(1)",
241 XFS_ERRLEVEL_HIGH
, mp
, sbp
);
242 return XFS_ERROR(EFSCORRUPTED
);
246 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
248 "XFS: filesystem is marked as having an internal log; "
249 "don't specify logdev on\nthe mount command line.");
250 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(2)",
251 XFS_ERRLEVEL_HIGH
, mp
, sbp
);
252 return XFS_ERROR(EFSCORRUPTED
);
256 * More sanity checking. These were stolen directly from
260 sbp
->sb_agcount
<= 0 ||
261 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
262 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
263 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
264 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
265 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
266 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
267 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
268 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
269 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
270 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
271 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
272 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
273 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
274 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
275 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
276 (sbp
->sb_imax_pct
> 100 || sbp
->sb_imax_pct
< 1))) {
277 cmn_err(CE_WARN
, "XFS: SB sanity check 1 failed");
278 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(3)",
279 XFS_ERRLEVEL_LOW
, mp
, sbp
);
280 return XFS_ERROR(EFSCORRUPTED
);
284 * Sanity check AG count, size fields against data size field
287 sbp
->sb_dblocks
== 0 ||
289 (xfs_drfsbno_t
)sbp
->sb_agcount
* sbp
->sb_agblocks
||
290 sbp
->sb_dblocks
< (xfs_drfsbno_t
)(sbp
->sb_agcount
- 1) *
291 sbp
->sb_agblocks
+ XFS_MIN_AG_BLOCKS
)) {
292 cmn_err(CE_WARN
, "XFS: SB sanity check 2 failed");
293 XFS_ERROR_REPORT("xfs_mount_validate_sb(4)",
294 XFS_ERRLEVEL_LOW
, mp
);
295 return XFS_ERROR(EFSCORRUPTED
);
298 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
299 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
301 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
303 (sbp
->sb_dblocks
>> (PAGE_SHIFT
- sbp
->sb_blocklog
)) > ULONG_MAX
||
304 (sbp
->sb_rblocks
>> (PAGE_SHIFT
- sbp
->sb_blocklog
)) > ULONG_MAX
)) {
305 #else /* Limited by UINT_MAX of sectors */
307 (sbp
->sb_dblocks
<< (sbp
->sb_blocklog
- BBSHIFT
)) > UINT_MAX
||
308 (sbp
->sb_rblocks
<< (sbp
->sb_blocklog
- BBSHIFT
)) > UINT_MAX
)) {
311 "XFS: File system is too large to be mounted on this system.");
312 return XFS_ERROR(E2BIG
);
315 if (unlikely(sbp
->sb_inprogress
)) {
316 cmn_err(CE_WARN
, "XFS: file system busy");
317 XFS_ERROR_REPORT("xfs_mount_validate_sb(5)",
318 XFS_ERRLEVEL_LOW
, mp
);
319 return XFS_ERROR(EFSCORRUPTED
);
323 * Version 1 directory format has never worked on Linux.
325 if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp
))) {
327 "XFS: Attempted to mount file system using version 1 directory format");
328 return XFS_ERROR(ENOSYS
);
332 * Until this is fixed only page-sized or smaller data blocks work.
334 if (unlikely(sbp
->sb_blocksize
> PAGE_SIZE
)) {
336 "XFS: Attempted to mount file system with blocksize %d bytes",
339 "XFS: Only page-sized (%ld) or less blocksizes currently work.",
341 return XFS_ERROR(ENOSYS
);
348 xfs_initialize_perag(
351 xfs_agnumber_t agcount
)
353 xfs_agnumber_t index
, max_metadata
;
357 xfs_sb_t
*sbp
= &mp
->m_sb
;
358 xfs_ino_t max_inum
= XFS_MAXINUMBER_32
;
360 /* Check to see if the filesystem can overflow 32 bit inodes */
361 agino
= XFS_OFFBNO_TO_AGINO(mp
, sbp
->sb_agblocks
- 1, 0);
362 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
364 /* Clear the mount flag if no inode can overflow 32 bits
365 * on this filesystem, or if specifically requested..
367 if ((vfs
->vfs_flag
& VFS_32BITINODES
) && ino
> max_inum
) {
368 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
370 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
373 /* If we can overflow then setup the ag headers accordingly */
374 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
375 /* Calculate how much should be reserved for inodes to
376 * meet the max inode percentage.
378 if (mp
->m_maxicount
) {
381 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
383 icount
+= sbp
->sb_agblocks
- 1;
384 do_div(icount
, sbp
->sb_agblocks
);
385 max_metadata
= icount
;
387 max_metadata
= agcount
;
389 for (index
= 0; index
< agcount
; index
++) {
390 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
391 if (ino
> max_inum
) {
396 /* This ag is preferred for inodes */
397 pag
= &mp
->m_perag
[index
];
398 pag
->pagi_inodeok
= 1;
399 if (index
< max_metadata
)
400 pag
->pagf_metadata
= 1;
403 /* Setup default behavior for smaller filesystems */
404 for (index
= 0; index
< agcount
; index
++) {
405 pag
= &mp
->m_perag
[index
];
406 pag
->pagi_inodeok
= 1;
415 * data - on disk version of sb
417 * dir - conversion direction: <0 - convert sb to buf
418 * >0 - convert buf to sb
419 * fields - which fields to copy (bitmask)
440 buf_ptr
= (xfs_caddr_t
)data
;
441 mem_ptr
= (xfs_caddr_t
)sb
;
444 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
445 first
= xfs_sb_info
[f
].offset
;
446 size
= xfs_sb_info
[f
+ 1].offset
- first
;
448 ASSERT(xfs_sb_info
[f
].type
== 0 || xfs_sb_info
[f
].type
== 1);
450 if (size
== 1 || xfs_sb_info
[f
].type
== 1) {
452 memcpy(mem_ptr
+ first
, buf_ptr
+ first
, size
);
454 memcpy(buf_ptr
+ first
, mem_ptr
+ first
, size
);
459 INT_XLATE(*(__uint16_t
*)(buf_ptr
+first
),
460 *(__uint16_t
*)(mem_ptr
+first
),
464 INT_XLATE(*(__uint32_t
*)(buf_ptr
+first
),
465 *(__uint32_t
*)(mem_ptr
+first
),
469 INT_XLATE(*(__uint64_t
*)(buf_ptr
+first
),
470 *(__uint64_t
*)(mem_ptr
+first
), dir
, ARCH_CONVERT
);
477 fields
&= ~(1LL << f
);
484 * Does the initial read of the superblock.
487 xfs_readsb(xfs_mount_t
*mp
)
489 unsigned int sector_size
;
490 unsigned int extra_flags
;
495 ASSERT(mp
->m_sb_bp
== NULL
);
496 ASSERT(mp
->m_ddev_targp
!= NULL
);
499 * Allocate a (locked) buffer to hold the superblock.
500 * This will be kept around at all times to optimize
501 * access to the superblock.
503 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
504 extra_flags
= XFS_BUF_LOCK
| XFS_BUF_MANAGE
| XFS_BUF_MAPPED
;
506 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
507 BTOBB(sector_size
), extra_flags
);
508 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
509 cmn_err(CE_WARN
, "XFS: SB read failed");
510 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
513 ASSERT(XFS_BUF_ISBUSY(bp
));
514 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
517 * Initialize the mount structure from the superblock.
518 * But first do some basic consistency checking.
520 sbp
= XFS_BUF_TO_SBP(bp
);
521 xfs_xlatesb(XFS_BUF_PTR(bp
), &(mp
->m_sb
), 1, XFS_SB_ALL_BITS
);
523 error
= xfs_mount_validate_sb(mp
, &(mp
->m_sb
));
525 cmn_err(CE_WARN
, "XFS: SB validate failed");
530 * We must be able to do sector-sized and sector-aligned IO.
532 if (sector_size
> mp
->m_sb
.sb_sectsize
) {
534 "XFS: device supports only %u byte sectors (not %u)",
535 sector_size
, mp
->m_sb
.sb_sectsize
);
541 * If device sector size is smaller than the superblock size,
542 * re-read the superblock so the buffer is correctly sized.
544 if (sector_size
< mp
->m_sb
.sb_sectsize
) {
545 XFS_BUF_UNMANAGE(bp
);
547 sector_size
= mp
->m_sb
.sb_sectsize
;
548 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, XFS_SB_DADDR
,
549 BTOBB(sector_size
), extra_flags
);
550 if (!bp
|| XFS_BUF_ISERROR(bp
)) {
551 cmn_err(CE_WARN
, "XFS: SB re-read failed");
552 error
= bp
? XFS_BUF_GETERROR(bp
) : ENOMEM
;
555 ASSERT(XFS_BUF_ISBUSY(bp
));
556 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
559 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
560 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
561 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
565 ASSERT(XFS_BUF_VALUSEMA(bp
) > 0);
570 XFS_BUF_UNMANAGE(bp
);
580 * Mount initialization code establishing various mount
581 * fields from the superblock associated with the given
585 xfs_mount_common(xfs_mount_t
*mp
, xfs_sb_t
*sbp
)
589 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
590 spinlock_init(&mp
->m_agirotor_lock
, "m_agirotor_lock");
591 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
592 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
593 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
594 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
595 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
596 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
597 mp
->m_litino
= sbp
->sb_inodesize
-
598 ((uint
)sizeof(xfs_dinode_core_t
) + (uint
)sizeof(xfs_agino_t
));
599 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
600 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
601 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
602 INIT_LIST_HEAD(&mp
->m_del_inodes
);
605 * Setup for attributes, in case they get created.
606 * This value is for inodes getting attributes for the first time,
607 * the per-inode value is for old attribute values.
609 ASSERT(sbp
->sb_inodesize
>= 256 && sbp
->sb_inodesize
<= 2048);
610 switch (sbp
->sb_inodesize
) {
612 mp
->m_attroffset
= XFS_LITINO(mp
) -
613 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
618 mp
->m_attroffset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
623 ASSERT(mp
->m_attroffset
< XFS_LITINO(mp
));
625 for (i
= 0; i
< 2; i
++) {
626 mp
->m_alloc_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
628 mp
->m_alloc_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
631 for (i
= 0; i
< 2; i
++) {
632 mp
->m_bmap_dmxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
634 mp
->m_bmap_dmnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
637 for (i
= 0; i
< 2; i
++) {
638 mp
->m_inobt_mxr
[i
] = XFS_BTREE_BLOCK_MAXRECS(sbp
->sb_blocksize
,
640 mp
->m_inobt_mnr
[i
] = XFS_BTREE_BLOCK_MINRECS(sbp
->sb_blocksize
,
644 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
645 mp
->m_ialloc_inos
= (int)MAX((__uint16_t
)XFS_INODES_PER_CHUNK
,
647 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
652 * This function does the following on an initial mount of a file system:
653 * - reads the superblock from disk and init the mount struct
654 * - if we're a 32-bit kernel, do a size check on the superblock
655 * so we don't mount terabyte filesystems
656 * - init mount struct realtime fields
657 * - allocate inode hash table for fs
658 * - init directory manager
659 * - perform recovery and init the log manager
668 xfs_sb_t
*sbp
= &(mp
->m_sb
);
671 int readio_log
, writeio_log
;
674 __int64_t update_flags
;
675 uint quotamount
, quotaflags
;
677 int uuid_mounted
= 0;
680 if (mp
->m_sb_bp
== NULL
) {
681 if ((error
= xfs_readsb(mp
))) {
685 xfs_mount_common(mp
, sbp
);
688 * Check if sb_agblocks is aligned at stripe boundary
689 * If sb_agblocks is NOT aligned turn off m_dalign since
690 * allocator alignment is within an ag, therefore ag has
691 * to be aligned at stripe boundary.
694 if (mp
->m_dalign
&& !(mfsi_flags
& XFS_MFSI_SECOND
)) {
696 * If stripe unit and stripe width are not multiples
697 * of the fs blocksize turn off alignment.
699 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
700 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
701 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
703 "XFS: alignment check 1 failed");
704 error
= XFS_ERROR(EINVAL
);
707 mp
->m_dalign
= mp
->m_swidth
= 0;
710 * Convert the stripe unit and width to FSBs.
712 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
713 if (mp
->m_dalign
&& (sbp
->sb_agblocks
% mp
->m_dalign
)) {
714 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
715 error
= XFS_ERROR(EINVAL
);
718 xfs_fs_cmn_err(CE_WARN
, mp
,
719 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
720 mp
->m_dalign
, mp
->m_swidth
,
725 } else if (mp
->m_dalign
) {
726 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
728 if (mp
->m_flags
& XFS_MOUNT_RETERR
) {
729 xfs_fs_cmn_err(CE_WARN
, mp
,
730 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
733 error
= XFS_ERROR(EINVAL
);
741 * Update superblock with new values
744 if (XFS_SB_VERSION_HASDALIGN(sbp
)) {
745 if (sbp
->sb_unit
!= mp
->m_dalign
) {
746 sbp
->sb_unit
= mp
->m_dalign
;
747 update_flags
|= XFS_SB_UNIT
;
749 if (sbp
->sb_width
!= mp
->m_swidth
) {
750 sbp
->sb_width
= mp
->m_swidth
;
751 update_flags
|= XFS_SB_WIDTH
;
754 } else if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) != XFS_MOUNT_NOALIGN
&&
755 XFS_SB_VERSION_HASDALIGN(&mp
->m_sb
)) {
756 mp
->m_dalign
= sbp
->sb_unit
;
757 mp
->m_swidth
= sbp
->sb_width
;
760 xfs_alloc_compute_maxlevels(mp
);
761 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
762 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
763 xfs_ialloc_compute_maxlevels(mp
);
765 if (sbp
->sb_imax_pct
) {
768 /* Make sure the maximum inode count is a multiple of the
769 * units we allocate inodes in.
772 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
774 do_div(icount
, mp
->m_ialloc_blks
);
775 mp
->m_maxicount
= (icount
* mp
->m_ialloc_blks
) <<
780 mp
->m_maxioffset
= xfs_max_file_offset(sbp
->sb_blocklog
);
783 * XFS uses the uuid from the superblock as the unique
784 * identifier for fsid. We can not use the uuid from the volume
785 * since a single partition filesystem is identical to a single
786 * partition volume/filesystem.
788 if ((mfsi_flags
& XFS_MFSI_SECOND
) == 0 &&
789 (mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0) {
790 if (xfs_uuid_mount(mp
)) {
791 error
= XFS_ERROR(EINVAL
);
795 ret64
= uuid_hash64(&sbp
->sb_uuid
);
796 memcpy(&vfsp
->vfs_fsid
, &ret64
, sizeof(ret64
));
800 * Set the default minimum read and write sizes unless
801 * already specified in a mount option.
802 * We use smaller I/O sizes when the file system
803 * is being used for NFS service (wsync mount option).
805 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
806 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
807 readio_log
= XFS_WSYNC_READIO_LOG
;
808 writeio_log
= XFS_WSYNC_WRITEIO_LOG
;
810 readio_log
= XFS_READIO_LOG_LARGE
;
811 writeio_log
= XFS_WRITEIO_LOG_LARGE
;
814 readio_log
= mp
->m_readio_log
;
815 writeio_log
= mp
->m_writeio_log
;
819 * Set the number of readahead buffers to use based on
820 * physical memory size.
822 if (xfs_physmem
<= 4096) /* <= 16MB */
823 mp
->m_nreadaheads
= XFS_RW_NREADAHEAD_16MB
;
824 else if (xfs_physmem
<= 8192) /* <= 32MB */
825 mp
->m_nreadaheads
= XFS_RW_NREADAHEAD_32MB
;
827 mp
->m_nreadaheads
= XFS_RW_NREADAHEAD_K32
;
828 if (sbp
->sb_blocklog
> readio_log
) {
829 mp
->m_readio_log
= sbp
->sb_blocklog
;
831 mp
->m_readio_log
= readio_log
;
833 mp
->m_readio_blocks
= 1 << (mp
->m_readio_log
- sbp
->sb_blocklog
);
834 if (sbp
->sb_blocklog
> writeio_log
) {
835 mp
->m_writeio_log
= sbp
->sb_blocklog
;
837 mp
->m_writeio_log
= writeio_log
;
839 mp
->m_writeio_blocks
= 1 << (mp
->m_writeio_log
- sbp
->sb_blocklog
);
842 * Set the inode cluster size based on the physical memory
843 * size. This may still be overridden by the file system
844 * block size if it is larger than the chosen cluster size.
846 if (xfs_physmem
<= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
847 mp
->m_inode_cluster_size
= XFS_INODE_SMALL_CLUSTER_SIZE
;
849 mp
->m_inode_cluster_size
= XFS_INODE_BIG_CLUSTER_SIZE
;
852 * Set whether we're using inode alignment.
854 if (XFS_SB_VERSION_HASALIGN(&mp
->m_sb
) &&
855 mp
->m_sb
.sb_inoalignmt
>=
856 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
857 mp
->m_inoalign_mask
= mp
->m_sb
.sb_inoalignmt
- 1;
859 mp
->m_inoalign_mask
= 0;
861 * If we are using stripe alignment, check whether
862 * the stripe unit is a multiple of the inode alignment
864 if (mp
->m_dalign
&& mp
->m_inoalign_mask
&&
865 !(mp
->m_dalign
& mp
->m_inoalign_mask
))
866 mp
->m_sinoalign
= mp
->m_dalign
;
870 * Check that the data (and log if separate) are an ok size.
872 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
873 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
874 cmn_err(CE_WARN
, "XFS: size check 1 failed");
875 error
= XFS_ERROR(E2BIG
);
878 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
879 d
- XFS_FSS_TO_BB(mp
, 1),
880 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
884 cmn_err(CE_WARN
, "XFS: size check 2 failed");
885 if (error
== ENOSPC
) {
886 error
= XFS_ERROR(E2BIG
);
891 if (((mfsi_flags
& XFS_MFSI_CLIENT
) == 0) &&
892 mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
893 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
894 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
895 cmn_err(CE_WARN
, "XFS: size check 3 failed");
896 error
= XFS_ERROR(E2BIG
);
899 error
= xfs_read_buf(mp
, mp
->m_logdev_targp
,
900 d
- XFS_FSB_TO_BB(mp
, 1),
901 XFS_FSB_TO_BB(mp
, 1), 0, &bp
);
905 cmn_err(CE_WARN
, "XFS: size check 3 failed");
906 if (error
== ENOSPC
) {
907 error
= XFS_ERROR(E2BIG
);
914 * Initialize realtime fields in the mount structure
916 if ((error
= xfs_rtmount_init(mp
))) {
917 cmn_err(CE_WARN
, "XFS: RT mount failed");
922 * For client case we are done now
924 if (mfsi_flags
& XFS_MFSI_CLIENT
) {
929 * Copies the low order bits of the timestamp and the randomly
930 * set "sequence" number out of a UUID.
932 uuid_getnodeuniq(&sbp
->sb_uuid
, mp
->m_fixedfsid
);
935 * The vfs structure needs to have a file system independent
936 * way of checking for the invariant file system ID. Since it
937 * can't look at mount structures it has a pointer to the data
938 * in the mount structure.
940 * File systems that don't support user level file handles (i.e.
941 * all of them except for XFS) will leave vfs_altfsid as NULL.
943 vfsp
->vfs_altfsid
= (xfs_fsid_t
*)mp
->m_fixedfsid
;
944 mp
->m_dmevmask
= 0; /* not persistent; set after each mount */
947 * Select the right directory manager.
950 XFS_SB_VERSION_HASDIRV2(&mp
->m_sb
) ?
955 * Initialize directory manager's entries.
960 * Initialize the attribute manager's entries.
962 mp
->m_attr_magicpct
= (mp
->m_sb
.sb_blocksize
* 37) / 100;
965 * Initialize the precomputed transaction reservations values.
970 * Allocate and initialize the inode hash table for this
977 * Allocate and initialize the per-ag data.
979 init_rwsem(&mp
->m_peraglock
);
981 kmem_zalloc(sbp
->sb_agcount
* sizeof(xfs_perag_t
), KM_SLEEP
);
983 mp
->m_maxagi
= xfs_initialize_perag(vfsp
, mp
, sbp
->sb_agcount
);
986 * log's mount-time initialization. Perform 1st part recovery if needed
988 if (likely(sbp
->sb_logblocks
> 0)) { /* check for volume case */
989 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
990 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
991 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
993 cmn_err(CE_WARN
, "XFS: log mount failed");
996 } else { /* No log has been defined */
997 cmn_err(CE_WARN
, "XFS: no log defined");
998 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW
, mp
);
999 error
= XFS_ERROR(EFSCORRUPTED
);
1004 * Get and sanity-check the root inode.
1005 * Save the pointer to it in the mount structure.
1007 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, 0, XFS_ILOCK_EXCL
, &rip
, 0);
1009 cmn_err(CE_WARN
, "XFS: failed to read root inode");
1013 ASSERT(rip
!= NULL
);
1014 rvp
= XFS_ITOV(rip
);
1016 if (unlikely((rip
->i_d
.di_mode
& S_IFMT
) != S_IFDIR
)) {
1017 cmn_err(CE_WARN
, "XFS: corrupted root inode");
1018 prdev("Root inode %llu is not a directory",
1019 mp
->m_ddev_targp
, (unsigned long long)rip
->i_ino
);
1020 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1021 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW
,
1023 error
= XFS_ERROR(EFSCORRUPTED
);
1026 mp
->m_rootip
= rip
; /* save it */
1028 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
1031 * Initialize realtime inode pointers in the mount structure
1033 if ((error
= xfs_rtmount_inodes(mp
))) {
1035 * Free up the root inode.
1037 cmn_err(CE_WARN
, "XFS: failed to read RT inodes");
1042 * If fs is not mounted readonly, then update the superblock
1043 * unit and width changes.
1045 if (update_flags
&& !(vfsp
->vfs_flag
& VFS_RDONLY
))
1046 xfs_mount_log_sbunit(mp
, update_flags
);
1049 * Initialise the XFS quota management subsystem for this mount
1051 if ((error
= XFS_QM_INIT(mp
, "amount
, "aflags
)))
1055 * Finish recovering the file system. This part needed to be
1056 * delayed until after the root and real-time bitmap inodes
1057 * were consistently read in.
1059 error
= xfs_log_mount_finish(mp
, mfsi_flags
);
1061 cmn_err(CE_WARN
, "XFS: log mount finish failed");
1066 * Complete the quota initialisation, post-log-replay component.
1068 if ((error
= XFS_QM_MOUNT(mp
, quotamount
, quotaflags
, mfsi_flags
)))
1075 * Free up the root inode.
1079 xfs_log_unmount_dealloc(mp
);
1083 for (agno
= 0; agno
< sbp
->sb_agcount
; agno
++)
1084 if (mp
->m_perag
[agno
].pagb_list
)
1085 kmem_free(mp
->m_perag
[agno
].pagb_list
,
1086 sizeof(xfs_perag_busy_t
) * XFS_PAGB_NUM_SLOTS
);
1087 kmem_free(mp
->m_perag
, sbp
->sb_agcount
* sizeof(xfs_perag_t
));
1092 xfs_uuid_unmount(mp
);
1100 * This flushes out the inodes,dquots and the superblock, unmounts the
1101 * log and makes sure that incore structures are freed.
1104 xfs_unmountfs(xfs_mount_t
*mp
, struct cred
*cr
)
1106 struct vfs
*vfsp
= XFS_MTOVFS(mp
);
1107 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1113 XFS_QM_DQPURGEALL(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
1116 * Flush out the log synchronously so that we know for sure
1117 * that nothing is pinned. This is important because bflush()
1118 * will skip pinned buffers.
1120 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
| XFS_LOG_SYNC
);
1122 xfs_binval(mp
->m_ddev_targp
);
1123 if (mp
->m_rtdev_targp
) {
1124 xfs_binval(mp
->m_rtdev_targp
);
1127 xfs_unmountfs_writesb(mp
);
1129 xfs_unmountfs_wait(mp
); /* wait for async bufs */
1131 xfs_log_unmount(mp
); /* Done! No more fs ops. */
1136 * All inodes from this mount point should be freed.
1138 ASSERT(mp
->m_inodes
== NULL
);
1140 xfs_unmountfs_close(mp
, cr
);
1141 if ((mp
->m_flags
& XFS_MOUNT_NOUUID
) == 0)
1142 xfs_uuid_unmount(mp
);
1144 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1146 * clear all error tags on this filesystem
1148 memcpy(&fsid
, &vfsp
->vfs_fsid
, sizeof(int64_t));
1149 xfs_errortag_clearall_umount(fsid
, mp
->m_fsname
, 0);
1152 xfs_mount_free(mp
, 1);
1157 xfs_unmountfs_close(xfs_mount_t
*mp
, struct cred
*cr
)
1159 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1160 xfs_free_buftarg(mp
->m_logdev_targp
, 1);
1161 if (mp
->m_rtdev_targp
)
1162 xfs_free_buftarg(mp
->m_rtdev_targp
, 1);
1163 xfs_free_buftarg(mp
->m_ddev_targp
, 0);
1167 xfs_unmountfs_wait(xfs_mount_t
*mp
)
1169 if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1170 xfs_wait_buftarg(mp
->m_logdev_targp
);
1171 if (mp
->m_rtdev_targp
)
1172 xfs_wait_buftarg(mp
->m_rtdev_targp
);
1173 xfs_wait_buftarg(mp
->m_ddev_targp
);
1177 xfs_unmountfs_writesb(xfs_mount_t
*mp
)
1184 * skip superblock write if fs is read-only, or
1185 * if we are doing a forced umount.
1187 sbp
= xfs_getsb(mp
, 0);
1188 if (!(XFS_MTOVFS(mp
)->vfs_flag
& VFS_RDONLY
||
1189 XFS_FORCED_SHUTDOWN(mp
))) {
1191 xfs_icsb_sync_counters(mp
);
1194 * mark shared-readonly if desired
1196 sb
= XFS_BUF_TO_SBP(sbp
);
1197 if (mp
->m_mk_sharedro
) {
1198 if (!(sb
->sb_flags
& XFS_SBF_READONLY
))
1199 sb
->sb_flags
|= XFS_SBF_READONLY
;
1200 if (!XFS_SB_VERSION_HASSHARED(sb
))
1201 XFS_SB_VERSION_ADDSHARED(sb
);
1202 xfs_fs_cmn_err(CE_NOTE
, mp
,
1203 "Unmounting, marking shared read-only");
1205 XFS_BUF_UNDONE(sbp
);
1206 XFS_BUF_UNREAD(sbp
);
1207 XFS_BUF_UNDELAYWRITE(sbp
);
1209 XFS_BUF_UNASYNC(sbp
);
1210 ASSERT(XFS_BUF_TARGET(sbp
) == mp
->m_ddev_targp
);
1211 xfsbdstrat(mp
, sbp
);
1212 /* Nevermind errors we might get here. */
1213 error
= xfs_iowait(sbp
);
1215 xfs_ioerror_alert("xfs_unmountfs_writesb",
1216 mp
, sbp
, XFS_BUF_ADDR(sbp
));
1217 if (error
&& mp
->m_mk_sharedro
)
1218 xfs_fs_cmn_err(CE_ALERT
, mp
, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1225 * xfs_mod_sb() can be used to copy arbitrary changes to the
1226 * in-core superblock into the superblock buffer to be logged.
1227 * It does not provide the higher level of locking that is
1228 * needed to protect the in-core superblock from concurrent
1232 xfs_mod_sb(xfs_trans_t
*tp
, __int64_t fields
)
1245 bp
= xfs_trans_getsb(tp
, mp
, 0);
1246 sbp
= XFS_BUF_TO_SBP(bp
);
1247 first
= sizeof(xfs_sb_t
);
1250 /* translate/copy */
1252 xfs_xlatesb(XFS_BUF_PTR(bp
), &(mp
->m_sb
), -1, fields
);
1254 /* find modified range */
1256 f
= (xfs_sb_field_t
)xfs_lowbit64((__uint64_t
)fields
);
1257 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1258 first
= xfs_sb_info
[f
].offset
;
1260 f
= (xfs_sb_field_t
)xfs_highbit64((__uint64_t
)fields
);
1261 ASSERT((1LL << f
) & XFS_SB_MOD_BITS
);
1262 last
= xfs_sb_info
[f
+ 1].offset
- 1;
1264 xfs_trans_log_buf(tp
, bp
, first
, last
);
1267 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1268 * a delta to a specified field in the in-core superblock. Simply
1269 * switch on the field indicated and apply the delta to that field.
1270 * Fields are not allowed to dip below zero, so if the delta would
1271 * do this do not apply it and return EINVAL.
1273 * The SB_LOCK must be held when this routine is called.
1276 xfs_mod_incore_sb_unlocked(xfs_mount_t
*mp
, xfs_sb_field_t field
,
1277 int delta
, int rsvd
)
1279 int scounter
; /* short counter for 32 bit fields */
1280 long long lcounter
; /* long counter for 64 bit fields */
1281 long long res_used
, rem
;
1284 * With the in-core superblock spin lock held, switch
1285 * on the indicated field. Apply the delta to the
1286 * proper field. If the fields value would dip below
1287 * 0, then do not apply the delta and return EINVAL.
1290 case XFS_SBS_ICOUNT
:
1291 lcounter
= (long long)mp
->m_sb
.sb_icount
;
1295 return XFS_ERROR(EINVAL
);
1297 mp
->m_sb
.sb_icount
= lcounter
;
1300 lcounter
= (long long)mp
->m_sb
.sb_ifree
;
1304 return XFS_ERROR(EINVAL
);
1306 mp
->m_sb
.sb_ifree
= lcounter
;
1308 case XFS_SBS_FDBLOCKS
:
1310 lcounter
= (long long)mp
->m_sb
.sb_fdblocks
;
1311 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1313 if (delta
> 0) { /* Putting blocks back */
1314 if (res_used
> delta
) {
1315 mp
->m_resblks_avail
+= delta
;
1317 rem
= delta
- res_used
;
1318 mp
->m_resblks_avail
= mp
->m_resblks
;
1321 } else { /* Taking blocks away */
1326 * If were out of blocks, use any available reserved blocks if
1332 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1334 return XFS_ERROR(ENOSPC
);
1336 mp
->m_resblks_avail
= lcounter
;
1338 } else { /* not reserved */
1339 return XFS_ERROR(ENOSPC
);
1344 mp
->m_sb
.sb_fdblocks
= lcounter
;
1346 case XFS_SBS_FREXTENTS
:
1347 lcounter
= (long long)mp
->m_sb
.sb_frextents
;
1350 return XFS_ERROR(ENOSPC
);
1352 mp
->m_sb
.sb_frextents
= lcounter
;
1354 case XFS_SBS_DBLOCKS
:
1355 lcounter
= (long long)mp
->m_sb
.sb_dblocks
;
1359 return XFS_ERROR(EINVAL
);
1361 mp
->m_sb
.sb_dblocks
= lcounter
;
1363 case XFS_SBS_AGCOUNT
:
1364 scounter
= mp
->m_sb
.sb_agcount
;
1368 return XFS_ERROR(EINVAL
);
1370 mp
->m_sb
.sb_agcount
= scounter
;
1372 case XFS_SBS_IMAX_PCT
:
1373 scounter
= mp
->m_sb
.sb_imax_pct
;
1377 return XFS_ERROR(EINVAL
);
1379 mp
->m_sb
.sb_imax_pct
= scounter
;
1381 case XFS_SBS_REXTSIZE
:
1382 scounter
= mp
->m_sb
.sb_rextsize
;
1386 return XFS_ERROR(EINVAL
);
1388 mp
->m_sb
.sb_rextsize
= scounter
;
1390 case XFS_SBS_RBMBLOCKS
:
1391 scounter
= mp
->m_sb
.sb_rbmblocks
;
1395 return XFS_ERROR(EINVAL
);
1397 mp
->m_sb
.sb_rbmblocks
= scounter
;
1399 case XFS_SBS_RBLOCKS
:
1400 lcounter
= (long long)mp
->m_sb
.sb_rblocks
;
1404 return XFS_ERROR(EINVAL
);
1406 mp
->m_sb
.sb_rblocks
= lcounter
;
1408 case XFS_SBS_REXTENTS
:
1409 lcounter
= (long long)mp
->m_sb
.sb_rextents
;
1413 return XFS_ERROR(EINVAL
);
1415 mp
->m_sb
.sb_rextents
= lcounter
;
1417 case XFS_SBS_REXTSLOG
:
1418 scounter
= mp
->m_sb
.sb_rextslog
;
1422 return XFS_ERROR(EINVAL
);
1424 mp
->m_sb
.sb_rextslog
= scounter
;
1428 return XFS_ERROR(EINVAL
);
1433 * xfs_mod_incore_sb() is used to change a field in the in-core
1434 * superblock structure by the specified delta. This modification
1435 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1436 * routine to do the work.
1439 xfs_mod_incore_sb(xfs_mount_t
*mp
, xfs_sb_field_t field
, int delta
, int rsvd
)
1444 /* check for per-cpu counters */
1446 #ifdef HAVE_PERCPU_SB
1447 case XFS_SBS_ICOUNT
:
1449 case XFS_SBS_FDBLOCKS
:
1450 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1451 status
= xfs_icsb_modify_counters(mp
, field
,
1458 s
= XFS_SB_LOCK(mp
);
1459 status
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
1460 XFS_SB_UNLOCK(mp
, s
);
1468 * xfs_mod_incore_sb_batch() is used to change more than one field
1469 * in the in-core superblock structure at a time. This modification
1470 * is protected by a lock internal to this module. The fields and
1471 * changes to those fields are specified in the array of xfs_mod_sb
1472 * structures passed in.
1474 * Either all of the specified deltas will be applied or none of
1475 * them will. If any modified field dips below 0, then all modifications
1476 * will be backed out and EINVAL will be returned.
1479 xfs_mod_incore_sb_batch(xfs_mount_t
*mp
, xfs_mod_sb_t
*msb
, uint nmsb
, int rsvd
)
1486 * Loop through the array of mod structures and apply each
1487 * individually. If any fail, then back out all those
1488 * which have already been applied. Do all of this within
1489 * the scope of the SB_LOCK so that all of the changes will
1492 s
= XFS_SB_LOCK(mp
);
1494 for (msbp
= &msbp
[0]; msbp
< (msb
+ nmsb
); msbp
++) {
1496 * Apply the delta at index n. If it fails, break
1497 * from the loop so we'll fall into the undo loop
1500 switch (msbp
->msb_field
) {
1501 #ifdef HAVE_PERCPU_SB
1502 case XFS_SBS_ICOUNT
:
1504 case XFS_SBS_FDBLOCKS
:
1505 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1506 status
= xfs_icsb_modify_counters_locked(mp
,
1508 msbp
->msb_delta
, rsvd
);
1514 status
= xfs_mod_incore_sb_unlocked(mp
,
1516 msbp
->msb_delta
, rsvd
);
1526 * If we didn't complete the loop above, then back out
1527 * any changes made to the superblock. If you add code
1528 * between the loop above and here, make sure that you
1529 * preserve the value of status. Loop back until
1530 * we step below the beginning of the array. Make sure
1531 * we don't touch anything back there.
1535 while (msbp
>= msb
) {
1536 switch (msbp
->msb_field
) {
1537 #ifdef HAVE_PERCPU_SB
1538 case XFS_SBS_ICOUNT
:
1540 case XFS_SBS_FDBLOCKS
:
1541 if (!(mp
->m_flags
& XFS_MOUNT_NO_PERCPU_SB
)) {
1543 xfs_icsb_modify_counters_locked(mp
,
1552 status
= xfs_mod_incore_sb_unlocked(mp
,
1558 ASSERT(status
== 0);
1562 XFS_SB_UNLOCK(mp
, s
);
1567 * xfs_getsb() is called to obtain the buffer for the superblock.
1568 * The buffer is returned locked and read in from disk.
1569 * The buffer should be released with a call to xfs_brelse().
1571 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1572 * the superblock buffer if it can be locked without sleeping.
1573 * If it can't then we'll return NULL.
1582 ASSERT(mp
->m_sb_bp
!= NULL
);
1584 if (flags
& XFS_BUF_TRYLOCK
) {
1585 if (!XFS_BUF_CPSEMA(bp
)) {
1589 XFS_BUF_PSEMA(bp
, PRIBIO
);
1592 ASSERT(XFS_BUF_ISDONE(bp
));
1597 * Used to free the superblock along various error paths.
1606 * Use xfs_getsb() so that the buffer will be locked
1607 * when we call xfs_buf_relse().
1609 bp
= xfs_getsb(mp
, 0);
1610 XFS_BUF_UNMANAGE(bp
);
1616 * See if the UUID is unique among mounted XFS filesystems.
1617 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1623 if (uuid_is_nil(&mp
->m_sb
.sb_uuid
)) {
1625 "XFS: Filesystem %s has nil UUID - can't mount",
1629 if (!uuid_table_insert(&mp
->m_sb
.sb_uuid
)) {
1631 "XFS: Filesystem %s has duplicate UUID - can't mount",
1639 * Remove filesystem from the UUID table.
1645 uuid_table_remove(&mp
->m_sb
.sb_uuid
);
1649 * Used to log changes to the superblock unit and width fields which could
1650 * be altered by the mount options. Only the first superblock is updated.
1653 xfs_mount_log_sbunit(
1659 ASSERT(fields
& (XFS_SB_UNIT
|XFS_SB_WIDTH
|XFS_SB_UUID
));
1661 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SB_UNIT
);
1662 if (xfs_trans_reserve(tp
, 0, mp
->m_sb
.sb_sectsize
+ 128, 0, 0,
1663 XFS_DEFAULT_LOG_COUNT
)) {
1664 xfs_trans_cancel(tp
, 0);
1667 xfs_mod_sb(tp
, fields
);
1668 xfs_trans_commit(tp
, 0, NULL
);
1672 #ifdef HAVE_PERCPU_SB
1674 * Per-cpu incore superblock counters
1676 * Simple concept, difficult implementation
1678 * Basically, replace the incore superblock counters with a distributed per cpu
1679 * counter for contended fields (e.g. free block count).
1681 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1682 * hence needs to be accurately read when we are running low on space. Hence
1683 * there is a method to enable and disable the per-cpu counters based on how
1684 * much "stuff" is available in them.
1686 * Basically, a counter is enabled if there is enough free resource to justify
1687 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1688 * ENOSPC), then we disable the counters to synchronise all callers and
1689 * re-distribute the available resources.
1691 * If, once we redistributed the available resources, we still get a failure,
1692 * we disable the per-cpu counter and go through the slow path.
1694 * The slow path is the current xfs_mod_incore_sb() function. This means that
1695 * when we disable a per-cpu counter, we need to drain it's resources back to
1696 * the global superblock. We do this after disabling the counter to prevent
1697 * more threads from queueing up on the counter.
1699 * Essentially, this means that we still need a lock in the fast path to enable
1700 * synchronisation between the global counters and the per-cpu counters. This
1701 * is not a problem because the lock will be local to a CPU almost all the time
1702 * and have little contention except when we get to ENOSPC conditions.
1704 * Basically, this lock becomes a barrier that enables us to lock out the fast
1705 * path while we do things like enabling and disabling counters and
1706 * synchronising the counters.
1710 * 1. XFS_SB_LOCK() before picking up per-cpu locks
1711 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1712 * 3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
1713 * 4. modifying per-cpu counters requires holding per-cpu lock
1714 * 5. modifying global counters requires holding XFS_SB_LOCK
1715 * 6. enabling or disabling a counter requires holding the XFS_SB_LOCK
1716 * and _none_ of the per-cpu locks.
1718 * Disabled counters are only ever re-enabled by a balance operation
1719 * that results in more free resources per CPU than a given threshold.
1720 * To ensure counters don't remain disabled, they are rebalanced when
1721 * the global resource goes above a higher threshold (i.e. some hysteresis
1722 * is present to prevent thrashing).
1726 * hot-plug CPU notifier support.
1728 * We cannot use the hotcpu_register() function because it does
1729 * not allow notifier instances. We need a notifier per filesystem
1730 * as we need to be able to identify the filesystem to balance
1731 * the counters out. This is achieved by having a notifier block
1732 * embedded in the xfs_mount_t and doing pointer magic to get the
1733 * mount pointer from the notifier block address.
1736 xfs_icsb_cpu_notify(
1737 struct notifier_block
*nfb
,
1738 unsigned long action
,
1741 xfs_icsb_cnts_t
*cntp
;
1745 mp
= (xfs_mount_t
*)container_of(nfb
, xfs_mount_t
, m_icsb_notifier
);
1746 cntp
= (xfs_icsb_cnts_t
*)
1747 per_cpu_ptr(mp
->m_sb_cnts
, (unsigned long)hcpu
);
1749 case CPU_UP_PREPARE
:
1750 /* Easy Case - initialize the area and locks, and
1751 * then rebalance when online does everything else for us. */
1752 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1755 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, 0);
1756 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, 0);
1757 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, 0);
1760 /* Disable all the counters, then fold the dead cpu's
1761 * count into the total on the global superblock and
1762 * re-enable the counters. */
1763 s
= XFS_SB_LOCK(mp
);
1764 xfs_icsb_disable_counter(mp
, XFS_SBS_ICOUNT
);
1765 xfs_icsb_disable_counter(mp
, XFS_SBS_IFREE
);
1766 xfs_icsb_disable_counter(mp
, XFS_SBS_FDBLOCKS
);
1768 mp
->m_sb
.sb_icount
+= cntp
->icsb_icount
;
1769 mp
->m_sb
.sb_ifree
+= cntp
->icsb_ifree
;
1770 mp
->m_sb
.sb_fdblocks
+= cntp
->icsb_fdblocks
;
1772 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1774 xfs_icsb_balance_counter(mp
, XFS_SBS_ICOUNT
, XFS_ICSB_SB_LOCKED
);
1775 xfs_icsb_balance_counter(mp
, XFS_SBS_IFREE
, XFS_ICSB_SB_LOCKED
);
1776 xfs_icsb_balance_counter(mp
, XFS_SBS_FDBLOCKS
, XFS_ICSB_SB_LOCKED
);
1777 XFS_SB_UNLOCK(mp
, s
);
1785 xfs_icsb_init_counters(
1788 xfs_icsb_cnts_t
*cntp
;
1791 mp
->m_sb_cnts
= alloc_percpu(xfs_icsb_cnts_t
);
1792 if (mp
->m_sb_cnts
== NULL
)
1795 mp
->m_icsb_notifier
.notifier_call
= xfs_icsb_cpu_notify
;
1796 mp
->m_icsb_notifier
.priority
= 0;
1797 register_cpu_notifier(&mp
->m_icsb_notifier
);
1799 for_each_online_cpu(i
) {
1800 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1801 memset(cntp
, 0, sizeof(xfs_icsb_cnts_t
));
1804 * start with all counters disabled so that the
1805 * initial balance kicks us off correctly
1807 mp
->m_icsb_counters
= -1;
1812 xfs_icsb_destroy_counters(
1815 if (mp
->m_sb_cnts
) {
1816 unregister_cpu_notifier(&mp
->m_icsb_notifier
);
1817 free_percpu(mp
->m_sb_cnts
);
1823 xfs_icsb_cnts_t
*icsbp
)
1825 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
)) {
1831 xfs_icsb_unlock_cntr(
1832 xfs_icsb_cnts_t
*icsbp
)
1834 clear_bit(XFS_ICSB_FLAG_LOCK
, &icsbp
->icsb_flags
);
1839 xfs_icsb_lock_all_counters(
1842 xfs_icsb_cnts_t
*cntp
;
1845 for_each_online_cpu(i
) {
1846 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1847 xfs_icsb_lock_cntr(cntp
);
1852 xfs_icsb_unlock_all_counters(
1855 xfs_icsb_cnts_t
*cntp
;
1858 for_each_online_cpu(i
) {
1859 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1860 xfs_icsb_unlock_cntr(cntp
);
1867 xfs_icsb_cnts_t
*cnt
,
1870 xfs_icsb_cnts_t
*cntp
;
1873 memset(cnt
, 0, sizeof(xfs_icsb_cnts_t
));
1875 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1876 xfs_icsb_lock_all_counters(mp
);
1878 for_each_online_cpu(i
) {
1879 cntp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, i
);
1880 cnt
->icsb_icount
+= cntp
->icsb_icount
;
1881 cnt
->icsb_ifree
+= cntp
->icsb_ifree
;
1882 cnt
->icsb_fdblocks
+= cntp
->icsb_fdblocks
;
1885 if (!(flags
& XFS_ICSB_LAZY_COUNT
))
1886 xfs_icsb_unlock_all_counters(mp
);
1890 xfs_icsb_counter_disabled(
1892 xfs_sb_field_t field
)
1894 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1895 return test_bit(field
, &mp
->m_icsb_counters
);
1899 xfs_icsb_disable_counter(
1901 xfs_sb_field_t field
)
1903 xfs_icsb_cnts_t cnt
;
1905 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1907 xfs_icsb_lock_all_counters(mp
);
1908 if (!test_and_set_bit(field
, &mp
->m_icsb_counters
)) {
1909 /* drain back to superblock */
1911 xfs_icsb_count(mp
, &cnt
, XFS_ICSB_SB_LOCKED
|XFS_ICSB_LAZY_COUNT
);
1913 case XFS_SBS_ICOUNT
:
1914 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1917 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1919 case XFS_SBS_FDBLOCKS
:
1920 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1927 xfs_icsb_unlock_all_counters(mp
);
1933 xfs_icsb_enable_counter(
1935 xfs_sb_field_t field
,
1939 xfs_icsb_cnts_t
*cntp
;
1942 ASSERT((field
>= XFS_SBS_ICOUNT
) && (field
<= XFS_SBS_FDBLOCKS
));
1944 xfs_icsb_lock_all_counters(mp
);
1945 for_each_online_cpu(i
) {
1946 cntp
= per_cpu_ptr(mp
->m_sb_cnts
, i
);
1948 case XFS_SBS_ICOUNT
:
1949 cntp
->icsb_icount
= count
+ resid
;
1952 cntp
->icsb_ifree
= count
+ resid
;
1954 case XFS_SBS_FDBLOCKS
:
1955 cntp
->icsb_fdblocks
= count
+ resid
;
1963 clear_bit(field
, &mp
->m_icsb_counters
);
1964 xfs_icsb_unlock_all_counters(mp
);
1968 xfs_icsb_sync_counters_int(
1972 xfs_icsb_cnts_t cnt
;
1975 /* Pass 1: lock all counters */
1976 if ((flags
& XFS_ICSB_SB_LOCKED
) == 0)
1977 s
= XFS_SB_LOCK(mp
);
1979 xfs_icsb_count(mp
, &cnt
, flags
);
1981 /* Step 3: update mp->m_sb fields */
1982 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_ICOUNT
))
1983 mp
->m_sb
.sb_icount
= cnt
.icsb_icount
;
1984 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_IFREE
))
1985 mp
->m_sb
.sb_ifree
= cnt
.icsb_ifree
;
1986 if (!xfs_icsb_counter_disabled(mp
, XFS_SBS_FDBLOCKS
))
1987 mp
->m_sb
.sb_fdblocks
= cnt
.icsb_fdblocks
;
1989 if ((flags
& XFS_ICSB_SB_LOCKED
) == 0)
1990 XFS_SB_UNLOCK(mp
, s
);
1994 * Accurate update of per-cpu counters to incore superblock
1997 xfs_icsb_sync_counters(
2000 xfs_icsb_sync_counters_int(mp
, 0);
2004 * lazy addition used for things like df, background sb syncs, etc
2007 xfs_icsb_sync_counters_lazy(
2010 xfs_icsb_sync_counters_int(mp
, XFS_ICSB_LAZY_COUNT
);
2014 * Balance and enable/disable counters as necessary.
2016 * Thresholds for re-enabling counters are somewhat magic.
2017 * inode counts are chosen to be the same number as single
2018 * on disk allocation chunk per CPU, and free blocks is
2019 * something far enough zero that we aren't going thrash
2020 * when we get near ENOSPC.
2022 #define XFS_ICSB_INO_CNTR_REENABLE 64
2023 #define XFS_ICSB_FDBLK_CNTR_REENABLE 512
2025 xfs_icsb_balance_counter(
2027 xfs_sb_field_t field
,
2030 uint64_t count
, resid
= 0;
2031 int weight
= num_online_cpus();
2034 if (!(flags
& XFS_ICSB_SB_LOCKED
))
2035 s
= XFS_SB_LOCK(mp
);
2037 /* disable counter and sync counter */
2038 xfs_icsb_disable_counter(mp
, field
);
2040 /* update counters - first CPU gets residual*/
2042 case XFS_SBS_ICOUNT
:
2043 count
= mp
->m_sb
.sb_icount
;
2044 resid
= do_div(count
, weight
);
2045 if (count
< XFS_ICSB_INO_CNTR_REENABLE
)
2049 count
= mp
->m_sb
.sb_ifree
;
2050 resid
= do_div(count
, weight
);
2051 if (count
< XFS_ICSB_INO_CNTR_REENABLE
)
2054 case XFS_SBS_FDBLOCKS
:
2055 count
= mp
->m_sb
.sb_fdblocks
;
2056 resid
= do_div(count
, weight
);
2057 if (count
< XFS_ICSB_FDBLK_CNTR_REENABLE
)
2065 xfs_icsb_enable_counter(mp
, field
, count
, resid
);
2067 if (!(flags
& XFS_ICSB_SB_LOCKED
))
2068 XFS_SB_UNLOCK(mp
, s
);
2072 xfs_icsb_modify_counters_int(
2074 xfs_sb_field_t field
,
2079 xfs_icsb_cnts_t
*icsbp
;
2080 long long lcounter
; /* long counter for 64 bit fields */
2081 int cpu
, s
, locked
= 0;
2082 int ret
= 0, balance_done
= 0;
2086 icsbp
= (xfs_icsb_cnts_t
*)per_cpu_ptr(mp
->m_sb_cnts
, cpu
),
2087 xfs_icsb_lock_cntr(icsbp
);
2088 if (unlikely(xfs_icsb_counter_disabled(mp
, field
)))
2092 case XFS_SBS_ICOUNT
:
2093 lcounter
= icsbp
->icsb_icount
;
2095 if (unlikely(lcounter
< 0))
2097 icsbp
->icsb_icount
= lcounter
;
2101 lcounter
= icsbp
->icsb_ifree
;
2103 if (unlikely(lcounter
< 0))
2105 icsbp
->icsb_ifree
= lcounter
;
2108 case XFS_SBS_FDBLOCKS
:
2109 BUG_ON((mp
->m_resblks
- mp
->m_resblks_avail
) != 0);
2111 lcounter
= icsbp
->icsb_fdblocks
;
2113 if (unlikely(lcounter
< 0))
2115 icsbp
->icsb_fdblocks
= lcounter
;
2121 xfs_icsb_unlock_cntr(icsbp
);
2124 XFS_SB_UNLOCK(mp
, s
);
2128 * The slow path needs to be run with the SBLOCK
2129 * held so that we prevent other threads from
2130 * attempting to run this path at the same time.
2131 * this provides exclusion for the balancing code,
2132 * and exclusive fallback if the balance does not
2133 * provide enough resources to continue in an unlocked
2137 xfs_icsb_unlock_cntr(icsbp
);
2140 /* need to hold superblock incase we need
2141 * to disable a counter */
2142 if (!(flags
& XFS_ICSB_SB_LOCKED
)) {
2143 s
= XFS_SB_LOCK(mp
);
2145 flags
|= XFS_ICSB_SB_LOCKED
;
2147 if (!balance_done
) {
2148 xfs_icsb_balance_counter(mp
, field
, flags
);
2153 * we might not have enough on this local
2154 * cpu to allocate for a bulk request.
2155 * We need to drain this field from all CPUs
2156 * and disable the counter fastpath
2158 xfs_icsb_disable_counter(mp
, field
);
2161 ret
= xfs_mod_incore_sb_unlocked(mp
, field
, delta
, rsvd
);
2164 XFS_SB_UNLOCK(mp
, s
);
2169 xfs_icsb_modify_counters(
2171 xfs_sb_field_t field
,
2175 return xfs_icsb_modify_counters_int(mp
, field
, delta
, rsvd
, 0);
2179 * Called when superblock is already locked
2182 xfs_icsb_modify_counters_locked(
2184 xfs_sb_field_t field
,
2188 return xfs_icsb_modify_counters_int(mp
, field
, delta
,
2189 rsvd
, XFS_ICSB_SB_LOCKED
);