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
165 for (agno = 0; agno < mp->m_maxagi; agno++)
166 if (mp->m_perag[agno].pagb_list)
167 kmem_free(mp->m_perag[agno].pagb_list,
168 sizeof(xfs_perag_busy_t) *
170 kmem_free(mp->m_perag,
171 sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
174 AIL_LOCK_DESTROY(&mp->m_ail_lock);
175 spinlock_destroy(&mp->m_sb_lock);
176 mutex_destroy(&mp->m_ilock);
177 freesema(&mp->m_growlock);
181 if (mp->m_fsname != NULL)
182 kmem_free(mp->m_fsname, mp->m_fsname_len);
183 if (mp->m_rtname != NULL)
184 kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
185 if (mp->m_logname != NULL)
186 kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
188 xfs_icsb_destroy_counters(mp);
192 * Check size of device based on the (data/realtime) block count.
193 * Note: this check is used by the growfs code as well as mount.
196 xfs_sb_validate_fsb_count(
200 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
201 ASSERT(sbp->sb_blocklog >= BBSHIFT);
203 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
204 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
206 #else /* Limited by UINT_MAX of sectors */
207 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
214 * Check the validity of the SB found.
217 xfs_mount_validate_sb(
223 * If the log device and data device have the
224 * same device number, the log is internal.
225 * Consequently, the sb_logstart should be non-zero. If
226 * we have a zero sb_logstart in this case, we may be trying to mount
227 * a volume filesystem in a non-volume manner.
229 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
230 xfs_fs_mount_cmn_err(flags, "bad magic number");
231 return XFS_ERROR(EWRONGFS);
234 if (!XFS_SB_GOOD_VERSION(sbp)) {
235 xfs_fs_mount_cmn_err(flags, "bad version");
236 return XFS_ERROR(EWRONGFS);
240 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
241 xfs_fs_mount_cmn_err(flags,
242 "filesystem is marked as having an external log; "
243 "specify logdev on the\nmount command line.");
244 return XFS_ERROR(EINVAL);
248 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
249 xfs_fs_mount_cmn_err(flags,
250 "filesystem is marked as having an internal log; "
251 "do not specify logdev on\nthe mount command line.");
252 return XFS_ERROR(EINVAL);
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 /* zero sb_imax_pct is valid */))) {
277 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
278 return XFS_ERROR(EFSCORRUPTED);
282 * Sanity check AG count, size fields against data size field
285 sbp->sb_dblocks == 0 ||
287 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
288 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
289 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
290 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
291 return XFS_ERROR(EFSCORRUPTED);
294 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
295 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
296 xfs_fs_mount_cmn_err(flags,
297 "file system too large to be mounted on this system.");
298 return XFS_ERROR(E2BIG);
301 if (unlikely(sbp->sb_inprogress)) {
302 xfs_fs_mount_cmn_err(flags, "file system busy");
303 return XFS_ERROR(EFSCORRUPTED);
307 * Version 1 directory format has never worked on Linux.
309 if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
310 xfs_fs_mount_cmn_err(flags,
311 "file system using version 1 directory format");
312 return XFS_ERROR(ENOSYS);
316 * Until this is fixed only page-sized or smaller data blocks work.
318 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
319 xfs_fs_mount_cmn_err(flags,
320 "file system with blocksize %d bytes",
322 xfs_fs_mount_cmn_err(flags,
323 "only pagesize (%ld) or less will currently work.",
325 return XFS_ERROR(ENOSYS);
332 xfs_initialize_perag_icache(
335 if (!pag->pag_ici_init) {
336 rwlock_init(&pag->pag_ici_lock);
337 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
338 pag->pag_ici_init = 1;
343 xfs_initialize_perag(
346 xfs_agnumber_t agcount)
348 xfs_agnumber_t index, max_metadata;
352 xfs_sb_t *sbp = &mp->m_sb;
353 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
355 /* Check to see if the filesystem can overflow 32 bit inodes */
356 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
357 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
359 /* Clear the mount flag if no inode can overflow 32 bits
360 * on this filesystem, or if specifically requested..
362 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
363 mp->m_flags |= XFS_MOUNT_32BITINODES;
365 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
368 /* If we can overflow then setup the ag headers accordingly */
369 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
370 /* Calculate how much should be reserved for inodes to
371 * meet the max inode percentage.
373 if (mp->m_maxicount) {
376 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
378 icount += sbp->sb_agblocks - 1;
379 do_div(icount, sbp->sb_agblocks);
380 max_metadata = icount;
382 max_metadata = agcount;
384 for (index = 0; index < agcount; index++) {
385 ino = XFS_AGINO_TO_INO(mp, index, agino);
386 if (ino > max_inum) {
391 /* This ag is preferred for inodes */
392 pag = &mp->m_perag[index];
393 pag->pagi_inodeok = 1;
394 if (index < max_metadata)
395 pag->pagf_metadata = 1;
396 xfs_initialize_perag_icache(pag);
399 /* Setup default behavior for smaller filesystems */
400 for (index = 0; index < agcount; index++) {
401 pag = &mp->m_perag[index];
402 pag->pagi_inodeok = 1;
403 xfs_initialize_perag_icache(pag);
414 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
415 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
416 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
417 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
418 to->sb_rextents = be64_to_cpu(from->sb_rextents);
419 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
420 to->sb_logstart = be64_to_cpu(from->sb_logstart);
421 to->sb_rootino = be64_to_cpu(from->sb_rootino);
422 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
423 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
424 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
425 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
426 to->sb_agcount = be32_to_cpu(from->sb_agcount);
427 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
428 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
429 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
430 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
431 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
432 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
433 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
434 to->sb_blocklog = from->sb_blocklog;
435 to->sb_sectlog = from->sb_sectlog;
436 to->sb_inodelog = from->sb_inodelog;
437 to->sb_inopblog = from->sb_inopblog;
438 to->sb_agblklog = from->sb_agblklog;
439 to->sb_rextslog = from->sb_rextslog;
440 to->sb_inprogress = from->sb_inprogress;
441 to->sb_imax_pct = from->sb_imax_pct;
442 to->sb_icount = be64_to_cpu(from->sb_icount);
443 to->sb_ifree = be64_to_cpu(from->sb_ifree);
444 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
445 to->sb_frextents = be64_to_cpu(from->sb_frextents);
446 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
447 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
448 to->sb_qflags = be16_to_cpu(from->sb_qflags);
449 to->sb_flags = from->sb_flags;
450 to->sb_shared_vn = from->sb_shared_vn;
451 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
452 to->sb_unit = be32_to_cpu(from->sb_unit);
453 to->sb_width = be32_to_cpu(from->sb_width);
454 to->sb_dirblklog = from->sb_dirblklog;
455 to->sb_logsectlog = from->sb_logsectlog;
456 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
457 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
458 to->sb_features2 = be32_to_cpu(from->sb_features2);
462 * Copy in core superblock to ondisk one.
464 * The fields argument is mask of superblock fields to copy.
472 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
473 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
483 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
484 first = xfs_sb_info[f].offset;
485 size = xfs_sb_info[f + 1].offset - first;
487 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
489 if (size == 1 || xfs_sb_info[f].type == 1) {
490 memcpy(to_ptr + first, from_ptr + first, size);
494 *(__be16 *)(to_ptr + first) =
495 cpu_to_be16(*(__u16 *)(from_ptr + first));
498 *(__be32 *)(to_ptr + first) =
499 cpu_to_be32(*(__u32 *)(from_ptr + first));
502 *(__be64 *)(to_ptr + first) =
503 cpu_to_be64(*(__u64 *)(from_ptr + first));
510 fields &= ~(1LL << f);
517 * Does the initial read of the superblock.
520 xfs_readsb(xfs_mount_t *mp, int flags)
522 unsigned int sector_size;
523 unsigned int extra_flags;
527 ASSERT(mp->m_sb_bp == NULL);
528 ASSERT(mp->m_ddev_targp != NULL);
531 * Allocate a (locked) buffer to hold the superblock.
532 * This will be kept around at all times to optimize
533 * access to the superblock.
535 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
536 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
538 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
539 BTOBB(sector_size), extra_flags);
540 if (!bp || XFS_BUF_ISERROR(bp)) {
541 xfs_fs_mount_cmn_err(flags, "SB read failed");
542 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
545 ASSERT(XFS_BUF_ISBUSY(bp));
546 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
549 * Initialize the mount structure from the superblock.
550 * But first do some basic consistency checking.
552 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
554 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
556 xfs_fs_mount_cmn_err(flags, "SB validate failed");
561 * We must be able to do sector-sized and sector-aligned IO.
563 if (sector_size > mp->m_sb.sb_sectsize) {
564 xfs_fs_mount_cmn_err(flags,
565 "device supports only %u byte sectors (not %u)",
566 sector_size, mp->m_sb.sb_sectsize);
572 * If device sector size is smaller than the superblock size,
573 * re-read the superblock so the buffer is correctly sized.
575 if (sector_size < mp->m_sb.sb_sectsize) {
576 XFS_BUF_UNMANAGE(bp);
578 sector_size = mp->m_sb.sb_sectsize;
579 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
580 BTOBB(sector_size), extra_flags);
581 if (!bp || XFS_BUF_ISERROR(bp)) {
582 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
583 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
586 ASSERT(XFS_BUF_ISBUSY(bp));
587 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
590 /* Initialize per-cpu counters */
591 xfs_icsb_reinit_counters(mp);
595 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
600 XFS_BUF_UNMANAGE(bp);
610 * Mount initialization code establishing various mount
611 * fields from the superblock associated with the given
615 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
619 mp->m_agfrotor = mp->m_agirotor = 0;
620 spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
621 mp->m_maxagi = mp->m_sb.sb_agcount;
622 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
623 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
624 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
625 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
626 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
627 mp->m_litino = sbp->sb_inodesize -
628 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
629 mp->m_blockmask = sbp->sb_blocksize - 1;
630 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
631 mp->m_blockwmask = mp->m_blockwsize - 1;
632 INIT_LIST_HEAD(&mp->m_del_inodes);
635 * Setup for attributes, in case they get created.
636 * This value is for inodes getting attributes for the first time,
637 * the per-inode value is for old attribute values.
639 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
640 switch (sbp->sb_inodesize) {
642 mp->m_attroffset = XFS_LITINO(mp) -
643 XFS_BMDR_SPACE_CALC(MINABTPTRS);
648 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
653 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
655 for (i = 0; i < 2; i++) {
656 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
658 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
661 for (i = 0; i < 2; i++) {
662 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
664 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
667 for (i = 0; i < 2; i++) {
668 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
670 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
674 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
675 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
677 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
681 * xfs_initialize_perag_data
683 * Read in each per-ag structure so we can count up the number of
684 * allocated inodes, free inodes and used filesystem blocks as this
685 * information is no longer persistent in the superblock. Once we have
686 * this information, write it into the in-core superblock structure.
689 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
691 xfs_agnumber_t index;
693 xfs_sb_t *sbp = &mp->m_sb;
697 uint64_t bfreelst = 0;
702 for (index = 0; index < agcount; index++) {
704 * read the agf, then the agi. This gets us
705 * all the inforamtion we need and populates the
706 * per-ag structures for us.
708 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
712 error = xfs_ialloc_pagi_init(mp, NULL, index);
715 pag = &mp->m_perag[index];
716 ifree += pag->pagi_freecount;
717 ialloc += pag->pagi_count;
718 bfree += pag->pagf_freeblks;
719 bfreelst += pag->pagf_flcount;
720 btree += pag->pagf_btreeblks;
723 * Overwrite incore superblock counters with just-read data
726 sbp->sb_ifree = ifree;
727 sbp->sb_icount = ialloc;
728 sbp->sb_fdblocks = bfree + bfreelst + btree;
729 XFS_SB_UNLOCK(mp, s);
731 /* Fixup the per-cpu counters as well. */
732 xfs_icsb_reinit_counters(mp);
740 * This function does the following on an initial mount of a file system:
741 * - reads the superblock from disk and init the mount struct
742 * - if we're a 32-bit kernel, do a size check on the superblock
743 * so we don't mount terabyte filesystems
744 * - init mount struct realtime fields
745 * - allocate inode hash table for fs
746 * - init directory manager
747 * - perform recovery and init the log manager
756 xfs_sb_t *sbp = &(mp->m_sb);
758 bhv_vnode_t *rvp = NULL;
759 int readio_log, writeio_log;
762 __int64_t update_flags;
763 uint quotamount, quotaflags;
765 int uuid_mounted = 0;
768 if (mp->m_sb_bp == NULL) {
769 if ((error = xfs_readsb(mp, mfsi_flags))) {
773 xfs_mount_common(mp, sbp);
776 * Check if sb_agblocks is aligned at stripe boundary
777 * If sb_agblocks is NOT aligned turn off m_dalign since
778 * allocator alignment is within an ag, therefore ag has
779 * to be aligned at stripe boundary.
782 if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
784 * If stripe unit and stripe width are not multiples
785 * of the fs blocksize turn off alignment.
787 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
788 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
789 if (mp->m_flags & XFS_MOUNT_RETERR) {
791 "XFS: alignment check 1 failed");
792 error = XFS_ERROR(EINVAL);
795 mp->m_dalign = mp->m_swidth = 0;
798 * Convert the stripe unit and width to FSBs.
800 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
801 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
802 if (mp->m_flags & XFS_MOUNT_RETERR) {
803 error = XFS_ERROR(EINVAL);
806 xfs_fs_cmn_err(CE_WARN, mp,
807 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
808 mp->m_dalign, mp->m_swidth,
813 } else if (mp->m_dalign) {
814 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
816 if (mp->m_flags & XFS_MOUNT_RETERR) {
817 xfs_fs_cmn_err(CE_WARN, mp,
818 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
821 error = XFS_ERROR(EINVAL);
829 * Update superblock with new values
832 if (XFS_SB_VERSION_HASDALIGN(sbp)) {
833 if (sbp->sb_unit != mp->m_dalign) {
834 sbp->sb_unit = mp->m_dalign;
835 update_flags |= XFS_SB_UNIT;
837 if (sbp->sb_width != mp->m_swidth) {
838 sbp->sb_width = mp->m_swidth;
839 update_flags |= XFS_SB_WIDTH;
842 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
843 XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
844 mp->m_dalign = sbp->sb_unit;
845 mp->m_swidth = sbp->sb_width;
848 xfs_alloc_compute_maxlevels(mp);
849 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
850 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
851 xfs_ialloc_compute_maxlevels(mp);
853 if (sbp->sb_imax_pct) {
856 /* Make sure the maximum inode count is a multiple of the
857 * units we allocate inodes in.
860 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
862 do_div(icount, mp->m_ialloc_blks);
863 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
868 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
871 * XFS uses the uuid from the superblock as the unique
872 * identifier for fsid. We can not use the uuid from the volume
873 * since a single partition filesystem is identical to a single
874 * partition volume/filesystem.
876 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
877 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
878 if (xfs_uuid_mount(mp)) {
879 error = XFS_ERROR(EINVAL);
886 * Set the default minimum read and write sizes unless
887 * already specified in a mount option.
888 * We use smaller I/O sizes when the file system
889 * is being used for NFS service (wsync mount option).
891 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
892 if (mp->m_flags & XFS_MOUNT_WSYNC) {
893 readio_log = XFS_WSYNC_READIO_LOG;
894 writeio_log = XFS_WSYNC_WRITEIO_LOG;
896 readio_log = XFS_READIO_LOG_LARGE;
897 writeio_log = XFS_WRITEIO_LOG_LARGE;
900 readio_log = mp->m_readio_log;
901 writeio_log = mp->m_writeio_log;
904 if (sbp->sb_blocklog > readio_log) {
905 mp->m_readio_log = sbp->sb_blocklog;
907 mp->m_readio_log = readio_log;
909 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
910 if (sbp->sb_blocklog > writeio_log) {
911 mp->m_writeio_log = sbp->sb_blocklog;
913 mp->m_writeio_log = writeio_log;
915 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
918 * Set the inode cluster size.
919 * This may still be overridden by the file system
920 * block size if it is larger than the chosen cluster size.
922 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
925 * Set whether we're using inode alignment.
927 if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
928 mp->m_sb.sb_inoalignmt >=
929 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
930 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
932 mp->m_inoalign_mask = 0;
934 * If we are using stripe alignment, check whether
935 * the stripe unit is a multiple of the inode alignment
937 if (mp->m_dalign && mp->m_inoalign_mask &&
938 !(mp->m_dalign & mp->m_inoalign_mask))
939 mp->m_sinoalign = mp->m_dalign;
943 * Check that the data (and log if separate) are an ok size.
945 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
946 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
947 cmn_err(CE_WARN, "XFS: size check 1 failed");
948 error = XFS_ERROR(E2BIG);
951 error = xfs_read_buf(mp, mp->m_ddev_targp,
952 d - XFS_FSS_TO_BB(mp, 1),
953 XFS_FSS_TO_BB(mp, 1), 0, &bp);
957 cmn_err(CE_WARN, "XFS: size check 2 failed");
958 if (error == ENOSPC) {
959 error = XFS_ERROR(E2BIG);
964 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
965 mp->m_logdev_targp != mp->m_ddev_targp) {
966 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
967 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
968 cmn_err(CE_WARN, "XFS: size check 3 failed");
969 error = XFS_ERROR(E2BIG);
972 error = xfs_read_buf(mp, mp->m_logdev_targp,
973 d - XFS_FSB_TO_BB(mp, 1),
974 XFS_FSB_TO_BB(mp, 1), 0, &bp);
978 cmn_err(CE_WARN, "XFS: size check 3 failed");
979 if (error == ENOSPC) {
980 error = XFS_ERROR(E2BIG);
987 * Initialize realtime fields in the mount structure
989 if ((error = xfs_rtmount_init(mp))) {
990 cmn_err(CE_WARN, "XFS: RT mount failed");
995 * For client case we are done now
997 if (mfsi_flags & XFS_MFSI_CLIENT) {
1002 * Copies the low order bits of the timestamp and the randomly
1003 * set "sequence" number out of a UUID.
1005 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1007 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1012 * Initialize the attribute manager's entries.
1014 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1017 * Initialize the precomputed transaction reservations values.
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 && !(mp->m_flags & XFS_MOUNT_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 = mp->m_sb.sb_dblocks;
1158 do_div(resblks, 20);
1159 resblks = min_t(__uint64_t, resblks, 1024);
1160 xfs_reserve_blocks(mp, &resblks, NULL);
1166 * Free up the root inode.
1170 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);
1199 * We can potentially deadlock here if we have an inode cluster
1200 * that has been freed has it's buffer still pinned in memory because
1201 * the transaction is still sitting in a iclog. The stale inodes
1202 * on that buffer will have their flush locks held until the
1203 * transaction hits the disk and the callbacks run. the inode
1204 * flush takes the flush lock unconditionally and with nothing to
1205 * push out the iclog we will never get that unlocked. hence we
1206 * need to force the log first.
1208 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1211 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1214 * Flush out the log synchronously so that we know for sure
1215 * that nothing is pinned. This is important because bflush()
1216 * will skip pinned buffers.
1218 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1220 xfs_binval(mp->m_ddev_targp);
1221 if (mp->m_rtdev_targp) {
1222 xfs_binval(mp->m_rtdev_targp);
1226 * Unreserve any blocks we have so that when we unmount we don't account
1227 * the reserved free space as used. This is really only necessary for
1228 * lazy superblock counting because it trusts the incore superblock
1229 * counters to be aboslutely correct on clean unmount.
1231 * We don't bother correcting this elsewhere for lazy superblock
1232 * counting because on mount of an unclean filesystem we reconstruct the
1233 * correct counter value and this is irrelevant.
1235 * For non-lazy counter filesystems, this doesn't matter at all because
1236 * we only every apply deltas to the superblock and hence the incore
1237 * value does not matter....
1240 xfs_reserve_blocks(mp, &resblks, NULL);
1242 xfs_log_sbcount(mp, 1);
1243 xfs_unmountfs_writesb(mp);
1244 xfs_unmountfs_wait(mp); /* wait for async bufs */
1245 xfs_log_unmount(mp); /* Done! No more fs ops. */
1250 * All inodes from this mount point should be freed.
1252 ASSERT(mp->m_inodes == NULL);
1254 xfs_unmountfs_close(mp, cr);
1255 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1256 xfs_uuid_unmount(mp);
1258 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1259 xfs_errortag_clearall(mp, 0);
1267 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1269 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1270 xfs_free_buftarg(mp->m_logdev_targp, 1);
1271 if (mp->m_rtdev_targp)
1272 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1273 xfs_free_buftarg(mp->m_ddev_targp, 0);
1277 xfs_unmountfs_wait(xfs_mount_t *mp)
1279 if (mp->m_logdev_targp != mp->m_ddev_targp)
1280 xfs_wait_buftarg(mp->m_logdev_targp);
1281 if (mp->m_rtdev_targp)
1282 xfs_wait_buftarg(mp->m_rtdev_targp);
1283 xfs_wait_buftarg(mp->m_ddev_targp);
1287 xfs_fs_writable(xfs_mount_t *mp)
1289 bhv_vfs_t *vfsp = XFS_MTOVFS(mp);
1291 return !(vfs_test_for_freeze(vfsp) || XFS_FORCED_SHUTDOWN(mp) ||
1292 (mp->m_flags & XFS_MOUNT_RDONLY));
1298 * Called either periodically to keep the on disk superblock values
1299 * roughly up to date or from unmount to make sure the values are
1300 * correct on a clean unmount.
1302 * Note this code can be called during the process of freezing, so
1303 * we may need to use the transaction allocator which does not not
1304 * block when the transaction subsystem is in its frozen state.
1314 if (!xfs_fs_writable(mp))
1317 xfs_icsb_sync_counters(mp);
1320 * we don't need to do this if we are updating the superblock
1321 * counters on every modification.
1323 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1326 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1327 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1328 XFS_DEFAULT_LOG_COUNT);
1330 xfs_trans_cancel(tp, 0);
1334 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1336 xfs_trans_set_sync(tp);
1337 xfs_trans_commit(tp, 0);
1347 xfs_dsb_t *sb = XFS_BUF_TO_SBP(bp);
1350 if (!(sb->sb_flags & XFS_SBF_READONLY))
1351 sb->sb_flags |= XFS_SBF_READONLY;
1353 version = be16_to_cpu(sb->sb_versionnum);
1354 if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 ||
1355 !(version & XFS_SB_VERSION_SHAREDBIT))
1356 version |= XFS_SB_VERSION_SHAREDBIT;
1357 sb->sb_versionnum = cpu_to_be16(version);
1361 xfs_unmountfs_writesb(xfs_mount_t *mp)
1367 * skip superblock write if fs is read-only, or
1368 * if we are doing a forced umount.
1370 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1371 XFS_FORCED_SHUTDOWN(mp))) {
1373 sbp = xfs_getsb(mp, 0);
1376 * mark shared-readonly if desired
1378 if (mp->m_mk_sharedro)
1379 xfs_mark_shared_ro(mp, sbp);
1381 XFS_BUF_UNDONE(sbp);
1382 XFS_BUF_UNREAD(sbp);
1383 XFS_BUF_UNDELAYWRITE(sbp);
1385 XFS_BUF_UNASYNC(sbp);
1386 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1387 xfsbdstrat(mp, sbp);
1388 /* Nevermind errors we might get here. */
1389 error = xfs_iowait(sbp);
1391 xfs_ioerror_alert("xfs_unmountfs_writesb",
1392 mp, sbp, XFS_BUF_ADDR(sbp));
1393 if (error && mp->m_mk_sharedro)
1394 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1401 * xfs_mod_sb() can be used to copy arbitrary changes to the
1402 * in-core superblock into the superblock buffer to be logged.
1403 * It does not provide the higher level of locking that is
1404 * needed to protect the in-core superblock from concurrent
1408 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1420 bp = xfs_trans_getsb(tp, mp, 0);
1421 first = sizeof(xfs_sb_t);
1424 /* translate/copy */
1426 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1428 /* find modified range */
1430 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1431 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1432 first = xfs_sb_info[f].offset;
1434 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1435 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1436 last = xfs_sb_info[f + 1].offset - 1;
1438 xfs_trans_log_buf(tp, bp, first, last);
1443 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1444 * a delta to a specified field in the in-core superblock. Simply
1445 * switch on the field indicated and apply the delta to that field.
1446 * Fields are not allowed to dip below zero, so if the delta would
1447 * do this do not apply it and return EINVAL.
1449 * The SB_LOCK must be held when this routine is called.
1452 xfs_mod_incore_sb_unlocked(
1454 xfs_sb_field_t field,
1458 int scounter; /* short counter for 32 bit fields */
1459 long long lcounter; /* long counter for 64 bit fields */
1460 long long res_used, rem;
1463 * With the in-core superblock spin lock held, switch
1464 * on the indicated field. Apply the delta to the
1465 * proper field. If the fields value would dip below
1466 * 0, then do not apply the delta and return EINVAL.
1469 case XFS_SBS_ICOUNT:
1470 lcounter = (long long)mp->m_sb.sb_icount;
1474 return XFS_ERROR(EINVAL);
1476 mp->m_sb.sb_icount = lcounter;
1479 lcounter = (long long)mp->m_sb.sb_ifree;
1483 return XFS_ERROR(EINVAL);
1485 mp->m_sb.sb_ifree = lcounter;
1487 case XFS_SBS_FDBLOCKS:
1488 lcounter = (long long)
1489 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1490 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1492 if (delta > 0) { /* Putting blocks back */
1493 if (res_used > delta) {
1494 mp->m_resblks_avail += delta;
1496 rem = delta - res_used;
1497 mp->m_resblks_avail = mp->m_resblks;
1500 } else { /* Taking blocks away */
1505 * If were out of blocks, use any available reserved blocks if
1511 lcounter = (long long)mp->m_resblks_avail + delta;
1513 return XFS_ERROR(ENOSPC);
1515 mp->m_resblks_avail = lcounter;
1517 } else { /* not reserved */
1518 return XFS_ERROR(ENOSPC);
1523 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1525 case XFS_SBS_FREXTENTS:
1526 lcounter = (long long)mp->m_sb.sb_frextents;
1529 return XFS_ERROR(ENOSPC);
1531 mp->m_sb.sb_frextents = lcounter;
1533 case XFS_SBS_DBLOCKS:
1534 lcounter = (long long)mp->m_sb.sb_dblocks;
1538 return XFS_ERROR(EINVAL);
1540 mp->m_sb.sb_dblocks = lcounter;
1542 case XFS_SBS_AGCOUNT:
1543 scounter = mp->m_sb.sb_agcount;
1547 return XFS_ERROR(EINVAL);
1549 mp->m_sb.sb_agcount = scounter;
1551 case XFS_SBS_IMAX_PCT:
1552 scounter = mp->m_sb.sb_imax_pct;
1556 return XFS_ERROR(EINVAL);
1558 mp->m_sb.sb_imax_pct = scounter;
1560 case XFS_SBS_REXTSIZE:
1561 scounter = mp->m_sb.sb_rextsize;
1565 return XFS_ERROR(EINVAL);
1567 mp->m_sb.sb_rextsize = scounter;
1569 case XFS_SBS_RBMBLOCKS:
1570 scounter = mp->m_sb.sb_rbmblocks;
1574 return XFS_ERROR(EINVAL);
1576 mp->m_sb.sb_rbmblocks = scounter;
1578 case XFS_SBS_RBLOCKS:
1579 lcounter = (long long)mp->m_sb.sb_rblocks;
1583 return XFS_ERROR(EINVAL);
1585 mp->m_sb.sb_rblocks = lcounter;
1587 case XFS_SBS_REXTENTS:
1588 lcounter = (long long)mp->m_sb.sb_rextents;
1592 return XFS_ERROR(EINVAL);
1594 mp->m_sb.sb_rextents = lcounter;
1596 case XFS_SBS_REXTSLOG:
1597 scounter = mp->m_sb.sb_rextslog;
1601 return XFS_ERROR(EINVAL);
1603 mp->m_sb.sb_rextslog = scounter;
1607 return XFS_ERROR(EINVAL);
1612 * xfs_mod_incore_sb() is used to change a field in the in-core
1613 * superblock structure by the specified delta. This modification
1614 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1615 * routine to do the work.
1620 xfs_sb_field_t field,
1627 /* check for per-cpu counters */
1629 #ifdef HAVE_PERCPU_SB
1630 case XFS_SBS_ICOUNT:
1632 case XFS_SBS_FDBLOCKS:
1633 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1634 status = xfs_icsb_modify_counters(mp, field,
1641 s = XFS_SB_LOCK(mp);
1642 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1643 XFS_SB_UNLOCK(mp, s);
1651 * xfs_mod_incore_sb_batch() is used to change more than one field
1652 * in the in-core superblock structure at a time. This modification
1653 * is protected by a lock internal to this module. The fields and
1654 * changes to those fields are specified in the array of xfs_mod_sb
1655 * structures passed in.
1657 * Either all of the specified deltas will be applied or none of
1658 * them will. If any modified field dips below 0, then all modifications
1659 * will be backed out and EINVAL will be returned.
1662 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1669 * Loop through the array of mod structures and apply each
1670 * individually. If any fail, then back out all those
1671 * which have already been applied. Do all of this within
1672 * the scope of the SB_LOCK so that all of the changes will
1675 s = XFS_SB_LOCK(mp);
1677 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1679 * Apply the delta at index n. If it fails, break
1680 * from the loop so we'll fall into the undo loop
1683 switch (msbp->msb_field) {
1684 #ifdef HAVE_PERCPU_SB
1685 case XFS_SBS_ICOUNT:
1687 case XFS_SBS_FDBLOCKS:
1688 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1689 XFS_SB_UNLOCK(mp, s);
1690 status = xfs_icsb_modify_counters(mp,
1692 msbp->msb_delta, rsvd);
1693 s = XFS_SB_LOCK(mp);
1699 status = xfs_mod_incore_sb_unlocked(mp,
1701 msbp->msb_delta, rsvd);
1711 * If we didn't complete the loop above, then back out
1712 * any changes made to the superblock. If you add code
1713 * between the loop above and here, make sure that you
1714 * preserve the value of status. Loop back until
1715 * we step below the beginning of the array. Make sure
1716 * we don't touch anything back there.
1720 while (msbp >= msb) {
1721 switch (msbp->msb_field) {
1722 #ifdef HAVE_PERCPU_SB
1723 case XFS_SBS_ICOUNT:
1725 case XFS_SBS_FDBLOCKS:
1726 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1727 XFS_SB_UNLOCK(mp, s);
1728 status = xfs_icsb_modify_counters(mp,
1732 s = XFS_SB_LOCK(mp);
1738 status = xfs_mod_incore_sb_unlocked(mp,
1744 ASSERT(status == 0);
1748 XFS_SB_UNLOCK(mp, s);
1753 * xfs_getsb() is called to obtain the buffer for the superblock.
1754 * The buffer is returned locked and read in from disk.
1755 * The buffer should be released with a call to xfs_brelse().
1757 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1758 * the superblock buffer if it can be locked without sleeping.
1759 * If it can't then we'll return NULL.
1768 ASSERT(mp->m_sb_bp != NULL);
1770 if (flags & XFS_BUF_TRYLOCK) {
1771 if (!XFS_BUF_CPSEMA(bp)) {
1775 XFS_BUF_PSEMA(bp, PRIBIO);
1778 ASSERT(XFS_BUF_ISDONE(bp));
1783 * Used to free the superblock along various error paths.
1792 * Use xfs_getsb() so that the buffer will be locked
1793 * when we call xfs_buf_relse().
1795 bp = xfs_getsb(mp, 0);
1796 XFS_BUF_UNMANAGE(bp);
1802 * See if the UUID is unique among mounted XFS filesystems.
1803 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1809 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1811 "XFS: Filesystem %s has nil UUID - can't mount",
1815 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1817 "XFS: Filesystem %s has duplicate UUID - can't mount",
1825 * Remove filesystem from the UUID table.
1831 uuid_table_remove(&mp->m_sb.sb_uuid);
1835 * Used to log changes to the superblock unit and width fields which could
1836 * be altered by the mount options. Only the first superblock is updated.
1839 xfs_mount_log_sbunit(
1845 ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1847 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1848 if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1849 XFS_DEFAULT_LOG_COUNT)) {
1850 xfs_trans_cancel(tp, 0);
1853 xfs_mod_sb(tp, fields);
1854 xfs_trans_commit(tp, 0);
1858 #ifdef HAVE_PERCPU_SB
1860 * Per-cpu incore superblock counters
1862 * Simple concept, difficult implementation
1864 * Basically, replace the incore superblock counters with a distributed per cpu
1865 * counter for contended fields (e.g. free block count).
1867 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1868 * hence needs to be accurately read when we are running low on space. Hence
1869 * there is a method to enable and disable the per-cpu counters based on how
1870 * much "stuff" is available in them.
1872 * Basically, a counter is enabled if there is enough free resource to justify
1873 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1874 * ENOSPC), then we disable the counters to synchronise all callers and
1875 * re-distribute the available resources.
1877 * If, once we redistributed the available resources, we still get a failure,
1878 * we disable the per-cpu counter and go through the slow path.
1880 * The slow path is the current xfs_mod_incore_sb() function. This means that
1881 * when we disable a per-cpu counter, we need to drain it's resources back to
1882 * the global superblock. We do this after disabling the counter to prevent
1883 * more threads from queueing up on the counter.
1885 * Essentially, this means that we still need a lock in the fast path to enable
1886 * synchronisation between the global counters and the per-cpu counters. This
1887 * is not a problem because the lock will be local to a CPU almost all the time
1888 * and have little contention except when we get to ENOSPC conditions.
1890 * Basically, this lock becomes a barrier that enables us to lock out the fast
1891 * path while we do things like enabling and disabling counters and
1892 * synchronising the counters.
1896 * 1. XFS_SB_LOCK() before picking up per-cpu locks
1897 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1898 * 3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
1899 * 4. modifying per-cpu counters requires holding per-cpu lock
1900 * 5. modifying global counters requires holding XFS_SB_LOCK
1901 * 6. enabling or disabling a counter requires holding the XFS_SB_LOCK
1902 * and _none_ of the per-cpu locks.
1904 * Disabled counters are only ever re-enabled by a balance operation
1905 * that results in more free resources per CPU than a given threshold.
1906 * To ensure counters don't remain disabled, they are rebalanced when
1907 * the global resource goes above a higher threshold (i.e. some hysteresis
1908 * is present to prevent thrashing).
1911 #ifdef CONFIG_HOTPLUG_CPU
1913 * hot-plug CPU notifier support.
1915 * We need a notifier per filesystem as we need to be able to identify
1916 * the filesystem to balance the counters out. This is achieved by
1917 * having a notifier block embedded in the xfs_mount_t and doing pointer
1918 * magic to get the mount pointer from the notifier block address.
1921 xfs_icsb_cpu_notify(
1922 struct notifier_block *nfb,
1923 unsigned long action,
1926 xfs_icsb_cnts_t *cntp;
1930 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1931 cntp = (xfs_icsb_cnts_t *)
1932 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1934 case CPU_UP_PREPARE:
1935 case CPU_UP_PREPARE_FROZEN:
1936 /* Easy Case - initialize the area and locks, and
1937 * then rebalance when online does everything else for us. */
1938 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1941 case CPU_ONLINE_FROZEN:
1943 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
1944 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
1945 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
1946 xfs_icsb_unlock(mp);
1949 case CPU_DEAD_FROZEN:
1950 /* Disable all the counters, then fold the dead cpu's
1951 * count into the total on the global superblock and
1952 * re-enable the counters. */
1954 s = XFS_SB_LOCK(mp);
1955 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1956 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1957 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1959 mp->m_sb.sb_icount += cntp->icsb_icount;
1960 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1961 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1963 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1965 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT,
1966 XFS_ICSB_SB_LOCKED, 0);
1967 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE,
1968 XFS_ICSB_SB_LOCKED, 0);
1969 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS,
1970 XFS_ICSB_SB_LOCKED, 0);
1971 XFS_SB_UNLOCK(mp, s);
1972 xfs_icsb_unlock(mp);
1978 #endif /* CONFIG_HOTPLUG_CPU */
1981 xfs_icsb_init_counters(
1984 xfs_icsb_cnts_t *cntp;
1987 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1988 if (mp->m_sb_cnts == NULL)
1991 #ifdef CONFIG_HOTPLUG_CPU
1992 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1993 mp->m_icsb_notifier.priority = 0;
1994 register_hotcpu_notifier(&mp->m_icsb_notifier);
1995 #endif /* CONFIG_HOTPLUG_CPU */
1997 for_each_online_cpu(i) {
1998 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1999 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2002 mutex_init(&mp->m_icsb_mutex);
2005 * start with all counters disabled so that the
2006 * initial balance kicks us off correctly
2008 mp->m_icsb_counters = -1;
2013 xfs_icsb_reinit_counters(
2018 * start with all counters disabled so that the
2019 * initial balance kicks us off correctly
2021 mp->m_icsb_counters = -1;
2022 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
2023 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
2024 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
2025 xfs_icsb_unlock(mp);
2029 xfs_icsb_destroy_counters(
2032 if (mp->m_sb_cnts) {
2033 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2034 free_percpu(mp->m_sb_cnts);
2036 mutex_destroy(&mp->m_icsb_mutex);
2041 xfs_icsb_cnts_t *icsbp)
2043 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2049 xfs_icsb_unlock_cntr(
2050 xfs_icsb_cnts_t *icsbp)
2052 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2057 xfs_icsb_lock_all_counters(
2060 xfs_icsb_cnts_t *cntp;
2063 for_each_online_cpu(i) {
2064 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2065 xfs_icsb_lock_cntr(cntp);
2070 xfs_icsb_unlock_all_counters(
2073 xfs_icsb_cnts_t *cntp;
2076 for_each_online_cpu(i) {
2077 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2078 xfs_icsb_unlock_cntr(cntp);
2085 xfs_icsb_cnts_t *cnt,
2088 xfs_icsb_cnts_t *cntp;
2091 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2093 if (!(flags & XFS_ICSB_LAZY_COUNT))
2094 xfs_icsb_lock_all_counters(mp);
2096 for_each_online_cpu(i) {
2097 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2098 cnt->icsb_icount += cntp->icsb_icount;
2099 cnt->icsb_ifree += cntp->icsb_ifree;
2100 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2103 if (!(flags & XFS_ICSB_LAZY_COUNT))
2104 xfs_icsb_unlock_all_counters(mp);
2108 xfs_icsb_counter_disabled(
2110 xfs_sb_field_t field)
2112 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2113 return test_bit(field, &mp->m_icsb_counters);
2117 xfs_icsb_disable_counter(
2119 xfs_sb_field_t field)
2121 xfs_icsb_cnts_t cnt;
2123 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2126 * If we are already disabled, then there is nothing to do
2127 * here. We check before locking all the counters to avoid
2128 * the expensive lock operation when being called in the
2129 * slow path and the counter is already disabled. This is
2130 * safe because the only time we set or clear this state is under
2133 if (xfs_icsb_counter_disabled(mp, field))
2136 xfs_icsb_lock_all_counters(mp);
2137 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2138 /* drain back to superblock */
2140 xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
2142 case XFS_SBS_ICOUNT:
2143 mp->m_sb.sb_icount = cnt.icsb_icount;
2146 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2148 case XFS_SBS_FDBLOCKS:
2149 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2156 xfs_icsb_unlock_all_counters(mp);
2162 xfs_icsb_enable_counter(
2164 xfs_sb_field_t field,
2168 xfs_icsb_cnts_t *cntp;
2171 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2173 xfs_icsb_lock_all_counters(mp);
2174 for_each_online_cpu(i) {
2175 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2177 case XFS_SBS_ICOUNT:
2178 cntp->icsb_icount = count + resid;
2181 cntp->icsb_ifree = count + resid;
2183 case XFS_SBS_FDBLOCKS:
2184 cntp->icsb_fdblocks = count + resid;
2192 clear_bit(field, &mp->m_icsb_counters);
2193 xfs_icsb_unlock_all_counters(mp);
2197 xfs_icsb_sync_counters_flags(
2201 xfs_icsb_cnts_t cnt;
2204 /* Pass 1: lock all counters */
2205 if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2206 s = XFS_SB_LOCK(mp);
2208 xfs_icsb_count(mp, &cnt, flags);
2210 /* Step 3: update mp->m_sb fields */
2211 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2212 mp->m_sb.sb_icount = cnt.icsb_icount;
2213 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2214 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2215 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2216 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2218 if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2219 XFS_SB_UNLOCK(mp, s);
2223 * Accurate update of per-cpu counters to incore superblock
2226 xfs_icsb_sync_counters(
2229 xfs_icsb_sync_counters_flags(mp, 0);
2233 * Balance and enable/disable counters as necessary.
2235 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2236 * chosen to be the same number as single on disk allocation chunk per CPU, and
2237 * free blocks is something far enough zero that we aren't going thrash when we
2238 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2239 * prevent looping endlessly when xfs_alloc_space asks for more than will
2240 * be distributed to a single CPU but each CPU has enough blocks to be
2243 * Note that we can be called when counters are already disabled.
2244 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2245 * prevent locking every per-cpu counter needlessly.
2248 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2249 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2250 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2252 xfs_icsb_balance_counter(
2254 xfs_sb_field_t field,
2258 uint64_t count, resid;
2259 int weight = num_online_cpus();
2261 uint64_t min = (uint64_t)min_per_cpu;
2263 if (!(flags & XFS_ICSB_SB_LOCKED))
2264 s = XFS_SB_LOCK(mp);
2266 /* disable counter and sync counter */
2267 xfs_icsb_disable_counter(mp, field);
2269 /* update counters - first CPU gets residual*/
2271 case XFS_SBS_ICOUNT:
2272 count = mp->m_sb.sb_icount;
2273 resid = do_div(count, weight);
2274 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2278 count = mp->m_sb.sb_ifree;
2279 resid = do_div(count, weight);
2280 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2283 case XFS_SBS_FDBLOCKS:
2284 count = mp->m_sb.sb_fdblocks;
2285 resid = do_div(count, weight);
2286 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2291 count = resid = 0; /* quiet, gcc */
2295 xfs_icsb_enable_counter(mp, field, count, resid);
2297 if (!(flags & XFS_ICSB_SB_LOCKED))
2298 XFS_SB_UNLOCK(mp, s);
2302 xfs_icsb_modify_counters(
2304 xfs_sb_field_t field,
2308 xfs_icsb_cnts_t *icsbp;
2309 long long lcounter; /* long counter for 64 bit fields */
2310 int cpu, ret = 0, s;
2315 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2318 * if the counter is disabled, go to slow path
2320 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2322 xfs_icsb_lock_cntr(icsbp);
2323 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2324 xfs_icsb_unlock_cntr(icsbp);
2329 case XFS_SBS_ICOUNT:
2330 lcounter = icsbp->icsb_icount;
2332 if (unlikely(lcounter < 0))
2333 goto balance_counter;
2334 icsbp->icsb_icount = lcounter;
2338 lcounter = icsbp->icsb_ifree;
2340 if (unlikely(lcounter < 0))
2341 goto balance_counter;
2342 icsbp->icsb_ifree = lcounter;
2345 case XFS_SBS_FDBLOCKS:
2346 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2348 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2350 if (unlikely(lcounter < 0))
2351 goto balance_counter;
2352 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2358 xfs_icsb_unlock_cntr(icsbp);
2366 * serialise with a mutex so we don't burn lots of cpu on
2367 * the superblock lock. We still need to hold the superblock
2368 * lock, however, when we modify the global structures.
2373 * Now running atomically.
2375 * If the counter is enabled, someone has beaten us to rebalancing.
2376 * Drop the lock and try again in the fast path....
2378 if (!(xfs_icsb_counter_disabled(mp, field))) {
2379 xfs_icsb_unlock(mp);
2384 * The counter is currently disabled. Because we are
2385 * running atomically here, we know a rebalance cannot
2386 * be in progress. Hence we can go straight to operating
2387 * on the global superblock. We do not call xfs_mod_incore_sb()
2388 * here even though we need to get the SB_LOCK. Doing so
2389 * will cause us to re-enter this function and deadlock.
2390 * Hence we get the SB_LOCK ourselves and then call
2391 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2392 * directly on the global counters.
2394 s = XFS_SB_LOCK(mp);
2395 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2396 XFS_SB_UNLOCK(mp, s);
2399 * Now that we've modified the global superblock, we
2400 * may be able to re-enable the distributed counters
2401 * (e.g. lots of space just got freed). After that
2405 xfs_icsb_balance_counter(mp, field, 0, 0);
2406 xfs_icsb_unlock(mp);
2410 xfs_icsb_unlock_cntr(icsbp);
2414 * We may have multiple threads here if multiple per-cpu
2415 * counters run dry at the same time. This will mean we can
2416 * do more balances than strictly necessary but it is not
2417 * the common slowpath case.
2422 * running atomically.
2424 * This will leave the counter in the correct state for future
2425 * accesses. After the rebalance, we simply try again and our retry
2426 * will either succeed through the fast path or slow path without
2427 * another balance operation being required.
2429 xfs_icsb_balance_counter(mp, field, 0, delta);
2430 xfs_icsb_unlock(mp);