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"
25 #include "xfs_trans_priv.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_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_alloc.h"
38 #include "xfs_rtalloc.h"
40 #include "xfs_error.h"
41 #include "xfs_quota.h"
42 #include "xfs_fsops.h"
43 #include "xfs_utils.h"
44 #include "xfs_trace.h"
45 #include "xfs_icache.h"
46 #include "xfs_cksum.h"
47 #include "xfs_buf_item.h"
51 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
53 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
55 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
58 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
59 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
64 short type; /* 0 = integer
65 * 1 = binary / string (no translation)
68 { offsetof(xfs_sb_t, sb_magicnum), 0 },
69 { offsetof(xfs_sb_t, sb_blocksize), 0 },
70 { offsetof(xfs_sb_t, sb_dblocks), 0 },
71 { offsetof(xfs_sb_t, sb_rblocks), 0 },
72 { offsetof(xfs_sb_t, sb_rextents), 0 },
73 { offsetof(xfs_sb_t, sb_uuid), 1 },
74 { offsetof(xfs_sb_t, sb_logstart), 0 },
75 { offsetof(xfs_sb_t, sb_rootino), 0 },
76 { offsetof(xfs_sb_t, sb_rbmino), 0 },
77 { offsetof(xfs_sb_t, sb_rsumino), 0 },
78 { offsetof(xfs_sb_t, sb_rextsize), 0 },
79 { offsetof(xfs_sb_t, sb_agblocks), 0 },
80 { offsetof(xfs_sb_t, sb_agcount), 0 },
81 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
82 { offsetof(xfs_sb_t, sb_logblocks), 0 },
83 { offsetof(xfs_sb_t, sb_versionnum), 0 },
84 { offsetof(xfs_sb_t, sb_sectsize), 0 },
85 { offsetof(xfs_sb_t, sb_inodesize), 0 },
86 { offsetof(xfs_sb_t, sb_inopblock), 0 },
87 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
88 { offsetof(xfs_sb_t, sb_blocklog), 0 },
89 { offsetof(xfs_sb_t, sb_sectlog), 0 },
90 { offsetof(xfs_sb_t, sb_inodelog), 0 },
91 { offsetof(xfs_sb_t, sb_inopblog), 0 },
92 { offsetof(xfs_sb_t, sb_agblklog), 0 },
93 { offsetof(xfs_sb_t, sb_rextslog), 0 },
94 { offsetof(xfs_sb_t, sb_inprogress), 0 },
95 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
96 { offsetof(xfs_sb_t, sb_icount), 0 },
97 { offsetof(xfs_sb_t, sb_ifree), 0 },
98 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
99 { offsetof(xfs_sb_t, sb_frextents), 0 },
100 { offsetof(xfs_sb_t, sb_uquotino), 0 },
101 { offsetof(xfs_sb_t, sb_gquotino), 0 },
102 { offsetof(xfs_sb_t, sb_qflags), 0 },
103 { offsetof(xfs_sb_t, sb_flags), 0 },
104 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
105 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
106 { offsetof(xfs_sb_t, sb_unit), 0 },
107 { offsetof(xfs_sb_t, sb_width), 0 },
108 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
109 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
110 { offsetof(xfs_sb_t, sb_logsectsize),0 },
111 { offsetof(xfs_sb_t, sb_logsunit), 0 },
112 { offsetof(xfs_sb_t, sb_features2), 0 },
113 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
114 { offsetof(xfs_sb_t, sb_features_compat), 0 },
115 { offsetof(xfs_sb_t, sb_features_ro_compat), 0 },
116 { offsetof(xfs_sb_t, sb_features_incompat), 0 },
117 { offsetof(xfs_sb_t, sb_features_log_incompat), 0 },
118 { offsetof(xfs_sb_t, sb_crc), 0 },
119 { offsetof(xfs_sb_t, sb_pad), 0 },
120 { offsetof(xfs_sb_t, sb_pquotino), 0 },
121 { offsetof(xfs_sb_t, sb_lsn), 0 },
122 { sizeof(xfs_sb_t), 0 }
125 static DEFINE_MUTEX(xfs_uuid_table_mutex);
126 static int xfs_uuid_table_size;
127 static uuid_t *xfs_uuid_table;
130 * See if the UUID is unique among mounted XFS filesystems.
131 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
135 struct xfs_mount *mp)
137 uuid_t *uuid = &mp->m_sb.sb_uuid;
140 if (mp->m_flags & XFS_MOUNT_NOUUID)
143 if (uuid_is_nil(uuid)) {
144 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
145 return XFS_ERROR(EINVAL);
148 mutex_lock(&xfs_uuid_table_mutex);
149 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
150 if (uuid_is_nil(&xfs_uuid_table[i])) {
154 if (uuid_equal(uuid, &xfs_uuid_table[i]))
159 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
160 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
161 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
163 hole = xfs_uuid_table_size++;
165 xfs_uuid_table[hole] = *uuid;
166 mutex_unlock(&xfs_uuid_table_mutex);
171 mutex_unlock(&xfs_uuid_table_mutex);
172 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
173 return XFS_ERROR(EINVAL);
178 struct xfs_mount *mp)
180 uuid_t *uuid = &mp->m_sb.sb_uuid;
183 if (mp->m_flags & XFS_MOUNT_NOUUID)
186 mutex_lock(&xfs_uuid_table_mutex);
187 for (i = 0; i < xfs_uuid_table_size; i++) {
188 if (uuid_is_nil(&xfs_uuid_table[i]))
190 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
192 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
195 ASSERT(i < xfs_uuid_table_size);
196 mutex_unlock(&xfs_uuid_table_mutex);
201 * Reference counting access wrappers to the perag structures.
202 * Because we never free per-ag structures, the only thing we
203 * have to protect against changes is the tree structure itself.
206 xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
208 struct xfs_perag *pag;
212 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
214 ASSERT(atomic_read(&pag->pag_ref) >= 0);
215 ref = atomic_inc_return(&pag->pag_ref);
218 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
223 * search from @first to find the next perag with the given tag set.
227 struct xfs_mount *mp,
228 xfs_agnumber_t first,
231 struct xfs_perag *pag;
236 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
237 (void **)&pag, first, 1, tag);
242 ref = atomic_inc_return(&pag->pag_ref);
244 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
249 xfs_perag_put(struct xfs_perag *pag)
253 ASSERT(atomic_read(&pag->pag_ref) > 0);
254 ref = atomic_dec_return(&pag->pag_ref);
255 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
260 struct rcu_head *head)
262 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
264 ASSERT(atomic_read(&pag->pag_ref) == 0);
269 * Free up the per-ag resources associated with the mount structure.
276 struct xfs_perag *pag;
278 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
279 spin_lock(&mp->m_perag_lock);
280 pag = radix_tree_delete(&mp->m_perag_tree, agno);
281 spin_unlock(&mp->m_perag_lock);
283 ASSERT(atomic_read(&pag->pag_ref) == 0);
284 call_rcu(&pag->rcu_head, __xfs_free_perag);
289 * Check size of device based on the (data/realtime) block count.
290 * Note: this check is used by the growfs code as well as mount.
293 xfs_sb_validate_fsb_count(
297 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
298 ASSERT(sbp->sb_blocklog >= BBSHIFT);
300 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
301 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
303 #else /* Limited by UINT_MAX of sectors */
304 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
311 * Check the validity of the SB found.
314 xfs_mount_validate_sb(
317 bool check_inprogress)
321 * If the log device and data device have the
322 * same device number, the log is internal.
323 * Consequently, the sb_logstart should be non-zero. If
324 * we have a zero sb_logstart in this case, we may be trying to mount
325 * a volume filesystem in a non-volume manner.
327 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
328 xfs_warn(mp, "bad magic number");
329 return XFS_ERROR(EWRONGFS);
333 if (!xfs_sb_good_version(sbp)) {
334 xfs_warn(mp, "bad version");
335 return XFS_ERROR(EWRONGFS);
339 * Version 5 superblock feature mask validation. Reject combinations the
340 * kernel cannot support up front before checking anything else.
342 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) {
344 "Version 5 superblock detected. This kernel has EXPERIMENTAL support enabled!\n"
345 "Use of these features in this kernel is at your own risk!");
347 if (xfs_sb_has_compat_feature(sbp,
348 XFS_SB_FEAT_COMPAT_UNKNOWN)) {
350 "Superblock has unknown compatible features (0x%x) enabled.\n"
351 "Using a more recent kernel is recommended.",
352 (sbp->sb_features_compat &
353 XFS_SB_FEAT_COMPAT_UNKNOWN));
356 if (xfs_sb_has_ro_compat_feature(sbp,
357 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
359 "Superblock has unknown read-only compatible features (0x%x) enabled.",
360 (sbp->sb_features_ro_compat &
361 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
362 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
364 "Attempted to mount read-only compatible filesystem read-write.\n"
365 "Filesystem can only be safely mounted read only.");
366 return XFS_ERROR(EINVAL);
369 if (xfs_sb_has_incompat_feature(sbp,
370 XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
372 "Superblock has unknown incompatible features (0x%x) enabled.\n"
373 "Filesystem can not be safely mounted by this kernel.",
374 (sbp->sb_features_incompat &
375 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
376 return XFS_ERROR(EINVAL);
381 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
383 "filesystem is marked as having an external log; "
384 "specify logdev on the mount command line.");
385 return XFS_ERROR(EINVAL);
389 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
391 "filesystem is marked as having an internal log; "
392 "do not specify logdev on the mount command line.");
393 return XFS_ERROR(EINVAL);
397 * More sanity checking. Most of these were stolen directly from
401 sbp->sb_agcount <= 0 ||
402 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
403 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
404 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
405 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
406 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
407 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
408 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
409 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
410 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
411 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
412 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
413 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
414 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
415 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
416 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
417 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
418 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
419 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
420 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
421 sbp->sb_dblocks == 0 ||
422 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
423 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
424 XFS_CORRUPTION_ERROR("SB sanity check failed",
425 XFS_ERRLEVEL_LOW, mp, sbp);
426 return XFS_ERROR(EFSCORRUPTED);
430 * Until this is fixed only page-sized or smaller data blocks work.
432 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
434 "File system with blocksize %d bytes. "
435 "Only pagesize (%ld) or less will currently work.",
436 sbp->sb_blocksize, PAGE_SIZE);
437 return XFS_ERROR(ENOSYS);
441 * Currently only very few inode sizes are supported.
443 switch (sbp->sb_inodesize) {
450 xfs_warn(mp, "inode size of %d bytes not supported",
452 return XFS_ERROR(ENOSYS);
455 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
456 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
458 "file system too large to be mounted on this system.");
459 return XFS_ERROR(EFBIG);
462 if (check_inprogress && sbp->sb_inprogress) {
463 xfs_warn(mp, "Offline file system operation in progress!");
464 return XFS_ERROR(EFSCORRUPTED);
468 * Version 1 directory format has never worked on Linux.
470 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
471 xfs_warn(mp, "file system using version 1 directory format");
472 return XFS_ERROR(ENOSYS);
479 xfs_initialize_perag(
481 xfs_agnumber_t agcount,
482 xfs_agnumber_t *maxagi)
484 xfs_agnumber_t index;
485 xfs_agnumber_t first_initialised = 0;
489 xfs_sb_t *sbp = &mp->m_sb;
493 * Walk the current per-ag tree so we don't try to initialise AGs
494 * that already exist (growfs case). Allocate and insert all the
495 * AGs we don't find ready for initialisation.
497 for (index = 0; index < agcount; index++) {
498 pag = xfs_perag_get(mp, index);
503 if (!first_initialised)
504 first_initialised = index;
506 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
509 pag->pag_agno = index;
511 spin_lock_init(&pag->pag_ici_lock);
512 mutex_init(&pag->pag_ici_reclaim_lock);
513 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
514 spin_lock_init(&pag->pag_buf_lock);
515 pag->pag_buf_tree = RB_ROOT;
517 if (radix_tree_preload(GFP_NOFS))
520 spin_lock(&mp->m_perag_lock);
521 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
523 spin_unlock(&mp->m_perag_lock);
524 radix_tree_preload_end();
528 spin_unlock(&mp->m_perag_lock);
529 radix_tree_preload_end();
533 * If we mount with the inode64 option, or no inode overflows
534 * the legacy 32-bit address space clear the inode32 option.
536 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
537 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
539 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
540 mp->m_flags |= XFS_MOUNT_32BITINODES;
542 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
544 if (mp->m_flags & XFS_MOUNT_32BITINODES)
545 index = xfs_set_inode32(mp);
547 index = xfs_set_inode64(mp);
555 for (; index > first_initialised; index--) {
556 pag = radix_tree_delete(&mp->m_perag_tree, index);
567 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
568 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
569 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
570 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
571 to->sb_rextents = be64_to_cpu(from->sb_rextents);
572 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
573 to->sb_logstart = be64_to_cpu(from->sb_logstart);
574 to->sb_rootino = be64_to_cpu(from->sb_rootino);
575 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
576 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
577 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
578 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
579 to->sb_agcount = be32_to_cpu(from->sb_agcount);
580 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
581 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
582 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
583 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
584 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
585 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
586 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
587 to->sb_blocklog = from->sb_blocklog;
588 to->sb_sectlog = from->sb_sectlog;
589 to->sb_inodelog = from->sb_inodelog;
590 to->sb_inopblog = from->sb_inopblog;
591 to->sb_agblklog = from->sb_agblklog;
592 to->sb_rextslog = from->sb_rextslog;
593 to->sb_inprogress = from->sb_inprogress;
594 to->sb_imax_pct = from->sb_imax_pct;
595 to->sb_icount = be64_to_cpu(from->sb_icount);
596 to->sb_ifree = be64_to_cpu(from->sb_ifree);
597 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
598 to->sb_frextents = be64_to_cpu(from->sb_frextents);
599 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
600 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
601 to->sb_qflags = be16_to_cpu(from->sb_qflags);
602 to->sb_flags = from->sb_flags;
603 to->sb_shared_vn = from->sb_shared_vn;
604 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
605 to->sb_unit = be32_to_cpu(from->sb_unit);
606 to->sb_width = be32_to_cpu(from->sb_width);
607 to->sb_dirblklog = from->sb_dirblklog;
608 to->sb_logsectlog = from->sb_logsectlog;
609 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
610 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
611 to->sb_features2 = be32_to_cpu(from->sb_features2);
612 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
613 to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
614 to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
615 to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
616 to->sb_features_log_incompat =
617 be32_to_cpu(from->sb_features_log_incompat);
619 to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
620 to->sb_lsn = be64_to_cpu(from->sb_lsn);
624 * Copy in core superblock to ondisk one.
626 * The fields argument is mask of superblock fields to copy.
634 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
635 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
645 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
646 first = xfs_sb_info[f].offset;
647 size = xfs_sb_info[f + 1].offset - first;
649 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
651 if (size == 1 || xfs_sb_info[f].type == 1) {
652 memcpy(to_ptr + first, from_ptr + first, size);
656 *(__be16 *)(to_ptr + first) =
657 cpu_to_be16(*(__u16 *)(from_ptr + first));
660 *(__be32 *)(to_ptr + first) =
661 cpu_to_be32(*(__u32 *)(from_ptr + first));
664 *(__be64 *)(to_ptr + first) =
665 cpu_to_be64(*(__u64 *)(from_ptr + first));
672 fields &= ~(1LL << f);
680 struct xfs_mount *mp = bp->b_target->bt_mount;
683 xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp));
686 * Only check the in progress field for the primary superblock as
687 * mkfs.xfs doesn't clear it from secondary superblocks.
689 return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR);
693 * If the superblock has the CRC feature bit set or the CRC field is non-null,
694 * check that the CRC is valid. We check the CRC field is non-null because a
695 * single bit error could clear the feature bit and unused parts of the
696 * superblock are supposed to be zero. Hence a non-null crc field indicates that
697 * we've potentially lost a feature bit and we should check it anyway.
703 struct xfs_mount *mp = bp->b_target->bt_mount;
704 struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp);
708 * open code the version check to avoid needing to convert the entire
709 * superblock from disk order just to check the version number
711 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
712 (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
716 if (!xfs_verify_cksum(bp->b_addr, be16_to_cpu(dsb->sb_sectsize),
717 offsetof(struct xfs_sb, sb_crc))) {
718 error = EFSCORRUPTED;
722 error = xfs_sb_verify(bp);
726 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
727 xfs_buf_ioerror(bp, error);
732 * We may be probed for a filesystem match, so we may not want to emit
733 * messages when the superblock buffer is not actually an XFS superblock.
734 * If we find an XFS superblock, the run a normal, noisy mount because we are
735 * really going to mount it and want to know about errors.
738 xfs_sb_quiet_read_verify(
741 struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp);
744 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
745 /* XFS filesystem, verify noisily! */
746 xfs_sb_read_verify(bp);
750 xfs_buf_ioerror(bp, EWRONGFS);
757 struct xfs_mount *mp = bp->b_target->bt_mount;
758 struct xfs_buf_log_item *bip = bp->b_fspriv;
761 error = xfs_sb_verify(bp);
763 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
764 xfs_buf_ioerror(bp, error);
768 if (!xfs_sb_version_hascrc(&mp->m_sb))
772 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
774 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
775 offsetof(struct xfs_sb, sb_crc));
778 const struct xfs_buf_ops xfs_sb_buf_ops = {
779 .verify_read = xfs_sb_read_verify,
780 .verify_write = xfs_sb_write_verify,
783 static const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
784 .verify_read = xfs_sb_quiet_read_verify,
785 .verify_write = xfs_sb_write_verify,
791 * Does the initial read of the superblock.
794 xfs_readsb(xfs_mount_t *mp, int flags)
796 unsigned int sector_size;
798 struct xfs_sb *sbp = &mp->m_sb;
800 int loud = !(flags & XFS_MFSI_QUIET);
802 ASSERT(mp->m_sb_bp == NULL);
803 ASSERT(mp->m_ddev_targp != NULL);
806 * Allocate a (locked) buffer to hold the superblock.
807 * This will be kept around at all times to optimize
808 * access to the superblock.
810 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
813 bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
814 BTOBB(sector_size), 0,
815 loud ? &xfs_sb_buf_ops
816 : &xfs_sb_quiet_buf_ops);
819 xfs_warn(mp, "SB buffer read failed");
825 xfs_warn(mp, "SB validate failed with error %d.", error);
830 * Initialize the mount structure from the superblock.
832 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
835 * We must be able to do sector-sized and sector-aligned IO.
837 if (sector_size > sbp->sb_sectsize) {
839 xfs_warn(mp, "device supports %u byte sectors (not %u)",
840 sector_size, sbp->sb_sectsize);
846 * If device sector size is smaller than the superblock size,
847 * re-read the superblock so the buffer is correctly sized.
849 if (sector_size < sbp->sb_sectsize) {
851 sector_size = sbp->sb_sectsize;
855 /* Initialize per-cpu counters */
856 xfs_icsb_reinit_counters(mp);
858 /* no need to be quiet anymore, so reset the buf ops */
859 bp->b_ops = &xfs_sb_buf_ops;
874 * Mount initialization code establishing various mount
875 * fields from the superblock associated with the given
879 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
881 mp->m_agfrotor = mp->m_agirotor = 0;
882 spin_lock_init(&mp->m_agirotor_lock);
883 mp->m_maxagi = mp->m_sb.sb_agcount;
884 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
885 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
886 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
887 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
888 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
889 mp->m_blockmask = sbp->sb_blocksize - 1;
890 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
891 mp->m_blockwmask = mp->m_blockwsize - 1;
893 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
894 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
895 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
896 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
898 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
899 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
900 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
901 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
903 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
904 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
905 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
906 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
908 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
909 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
911 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
915 * xfs_initialize_perag_data
917 * Read in each per-ag structure so we can count up the number of
918 * allocated inodes, free inodes and used filesystem blocks as this
919 * information is no longer persistent in the superblock. Once we have
920 * this information, write it into the in-core superblock structure.
923 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
925 xfs_agnumber_t index;
927 xfs_sb_t *sbp = &mp->m_sb;
931 uint64_t bfreelst = 0;
935 for (index = 0; index < agcount; index++) {
937 * read the agf, then the agi. This gets us
938 * all the information we need and populates the
939 * per-ag structures for us.
941 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
945 error = xfs_ialloc_pagi_init(mp, NULL, index);
948 pag = xfs_perag_get(mp, index);
949 ifree += pag->pagi_freecount;
950 ialloc += pag->pagi_count;
951 bfree += pag->pagf_freeblks;
952 bfreelst += pag->pagf_flcount;
953 btree += pag->pagf_btreeblks;
957 * Overwrite incore superblock counters with just-read data
959 spin_lock(&mp->m_sb_lock);
960 sbp->sb_ifree = ifree;
961 sbp->sb_icount = ialloc;
962 sbp->sb_fdblocks = bfree + bfreelst + btree;
963 spin_unlock(&mp->m_sb_lock);
965 /* Fixup the per-cpu counters as well. */
966 xfs_icsb_reinit_counters(mp);
972 * Update alignment values based on mount options and sb values
975 xfs_update_alignment(xfs_mount_t *mp)
977 xfs_sb_t *sbp = &(mp->m_sb);
981 * If stripe unit and stripe width are not multiples
982 * of the fs blocksize turn off alignment.
984 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
985 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
986 if (mp->m_flags & XFS_MOUNT_RETERR) {
987 xfs_warn(mp, "alignment check failed: "
988 "(sunit/swidth vs. blocksize)");
989 return XFS_ERROR(EINVAL);
991 mp->m_dalign = mp->m_swidth = 0;
994 * Convert the stripe unit and width to FSBs.
996 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
997 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
998 if (mp->m_flags & XFS_MOUNT_RETERR) {
999 xfs_warn(mp, "alignment check failed: "
1000 "(sunit/swidth vs. ag size)");
1001 return XFS_ERROR(EINVAL);
1004 "stripe alignment turned off: sunit(%d)/swidth(%d) "
1005 "incompatible with agsize(%d)",
1006 mp->m_dalign, mp->m_swidth,
1011 } else if (mp->m_dalign) {
1012 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
1014 if (mp->m_flags & XFS_MOUNT_RETERR) {
1015 xfs_warn(mp, "alignment check failed: "
1016 "sunit(%d) less than bsize(%d)",
1018 mp->m_blockmask +1);
1019 return XFS_ERROR(EINVAL);
1026 * Update superblock with new values
1029 if (xfs_sb_version_hasdalign(sbp)) {
1030 if (sbp->sb_unit != mp->m_dalign) {
1031 sbp->sb_unit = mp->m_dalign;
1032 mp->m_update_flags |= XFS_SB_UNIT;
1034 if (sbp->sb_width != mp->m_swidth) {
1035 sbp->sb_width = mp->m_swidth;
1036 mp->m_update_flags |= XFS_SB_WIDTH;
1039 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
1040 xfs_sb_version_hasdalign(&mp->m_sb)) {
1041 mp->m_dalign = sbp->sb_unit;
1042 mp->m_swidth = sbp->sb_width;
1049 * Set the maximum inode count for this filesystem
1052 xfs_set_maxicount(xfs_mount_t *mp)
1054 xfs_sb_t *sbp = &(mp->m_sb);
1057 if (sbp->sb_imax_pct) {
1059 * Make sure the maximum inode count is a multiple
1060 * of the units we allocate inodes in.
1062 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
1063 do_div(icount, 100);
1064 do_div(icount, mp->m_ialloc_blks);
1065 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
1068 mp->m_maxicount = 0;
1073 * Set the default minimum read and write sizes unless
1074 * already specified in a mount option.
1075 * We use smaller I/O sizes when the file system
1076 * is being used for NFS service (wsync mount option).
1079 xfs_set_rw_sizes(xfs_mount_t *mp)
1081 xfs_sb_t *sbp = &(mp->m_sb);
1082 int readio_log, writeio_log;
1084 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
1085 if (mp->m_flags & XFS_MOUNT_WSYNC) {
1086 readio_log = XFS_WSYNC_READIO_LOG;
1087 writeio_log = XFS_WSYNC_WRITEIO_LOG;
1089 readio_log = XFS_READIO_LOG_LARGE;
1090 writeio_log = XFS_WRITEIO_LOG_LARGE;
1093 readio_log = mp->m_readio_log;
1094 writeio_log = mp->m_writeio_log;
1097 if (sbp->sb_blocklog > readio_log) {
1098 mp->m_readio_log = sbp->sb_blocklog;
1100 mp->m_readio_log = readio_log;
1102 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
1103 if (sbp->sb_blocklog > writeio_log) {
1104 mp->m_writeio_log = sbp->sb_blocklog;
1106 mp->m_writeio_log = writeio_log;
1108 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
1112 * precalculate the low space thresholds for dynamic speculative preallocation.
1115 xfs_set_low_space_thresholds(
1116 struct xfs_mount *mp)
1120 for (i = 0; i < XFS_LOWSP_MAX; i++) {
1121 __uint64_t space = mp->m_sb.sb_dblocks;
1124 mp->m_low_space[i] = space * (i + 1);
1130 * Set whether we're using inode alignment.
1133 xfs_set_inoalignment(xfs_mount_t *mp)
1135 if (xfs_sb_version_hasalign(&mp->m_sb) &&
1136 mp->m_sb.sb_inoalignmt >=
1137 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1138 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1140 mp->m_inoalign_mask = 0;
1142 * If we are using stripe alignment, check whether
1143 * the stripe unit is a multiple of the inode alignment
1145 if (mp->m_dalign && mp->m_inoalign_mask &&
1146 !(mp->m_dalign & mp->m_inoalign_mask))
1147 mp->m_sinoalign = mp->m_dalign;
1149 mp->m_sinoalign = 0;
1153 * Check that the data (and log if separate) are an ok size.
1156 xfs_check_sizes(xfs_mount_t *mp)
1161 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1162 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1163 xfs_warn(mp, "filesystem size mismatch detected");
1164 return XFS_ERROR(EFBIG);
1166 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
1167 d - XFS_FSS_TO_BB(mp, 1),
1168 XFS_FSS_TO_BB(mp, 1), 0, NULL);
1170 xfs_warn(mp, "last sector read failed");
1175 if (mp->m_logdev_targp != mp->m_ddev_targp) {
1176 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1177 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1178 xfs_warn(mp, "log size mismatch detected");
1179 return XFS_ERROR(EFBIG);
1181 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
1182 d - XFS_FSB_TO_BB(mp, 1),
1183 XFS_FSB_TO_BB(mp, 1), 0, NULL);
1185 xfs_warn(mp, "log device read failed");
1194 * Clear the quotaflags in memory and in the superblock.
1197 xfs_mount_reset_sbqflags(
1198 struct xfs_mount *mp)
1201 struct xfs_trans *tp;
1206 * It is OK to look at sb_qflags here in mount path,
1207 * without m_sb_lock.
1209 if (mp->m_sb.sb_qflags == 0)
1211 spin_lock(&mp->m_sb_lock);
1212 mp->m_sb.sb_qflags = 0;
1213 spin_unlock(&mp->m_sb_lock);
1216 * If the fs is readonly, let the incore superblock run
1217 * with quotas off but don't flush the update out to disk
1219 if (mp->m_flags & XFS_MOUNT_RDONLY)
1222 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1223 error = xfs_trans_reserve(tp, 0, XFS_QM_SBCHANGE_LOG_RES(mp),
1224 0, 0, XFS_DEFAULT_LOG_COUNT);
1226 xfs_trans_cancel(tp, 0);
1227 xfs_alert(mp, "%s: Superblock update failed!", __func__);
1231 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1232 return xfs_trans_commit(tp, 0);
1236 xfs_default_resblks(xfs_mount_t *mp)
1241 * We default to 5% or 8192 fsbs of space reserved, whichever is
1242 * smaller. This is intended to cover concurrent allocation
1243 * transactions when we initially hit enospc. These each require a 4
1244 * block reservation. Hence by default we cover roughly 2000 concurrent
1245 * allocation reservations.
1247 resblks = mp->m_sb.sb_dblocks;
1248 do_div(resblks, 20);
1249 resblks = min_t(__uint64_t, resblks, 8192);
1254 * This function does the following on an initial mount of a file system:
1255 * - reads the superblock from disk and init the mount struct
1256 * - if we're a 32-bit kernel, do a size check on the superblock
1257 * so we don't mount terabyte filesystems
1258 * - init mount struct realtime fields
1259 * - allocate inode hash table for fs
1260 * - init directory manager
1261 * - perform recovery and init the log manager
1267 xfs_sb_t *sbp = &(mp->m_sb);
1270 uint quotamount = 0;
1271 uint quotaflags = 0;
1274 xfs_mount_common(mp, sbp);
1277 * Check for a mismatched features2 values. Older kernels
1278 * read & wrote into the wrong sb offset for sb_features2
1279 * on some platforms due to xfs_sb_t not being 64bit size aligned
1280 * when sb_features2 was added, which made older superblock
1281 * reading/writing routines swap it as a 64-bit value.
1283 * For backwards compatibility, we make both slots equal.
1285 * If we detect a mismatched field, we OR the set bits into the
1286 * existing features2 field in case it has already been modified; we
1287 * don't want to lose any features. We then update the bad location
1288 * with the ORed value so that older kernels will see any features2
1289 * flags, and mark the two fields as needing updates once the
1290 * transaction subsystem is online.
1292 if (xfs_sb_has_mismatched_features2(sbp)) {
1293 xfs_warn(mp, "correcting sb_features alignment problem");
1294 sbp->sb_features2 |= sbp->sb_bad_features2;
1295 sbp->sb_bad_features2 = sbp->sb_features2;
1296 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1299 * Re-check for ATTR2 in case it was found in bad_features2
1302 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1303 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1304 mp->m_flags |= XFS_MOUNT_ATTR2;
1307 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1308 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1309 xfs_sb_version_removeattr2(&mp->m_sb);
1310 mp->m_update_flags |= XFS_SB_FEATURES2;
1312 /* update sb_versionnum for the clearing of the morebits */
1313 if (!sbp->sb_features2)
1314 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1318 * Check if sb_agblocks is aligned at stripe boundary
1319 * If sb_agblocks is NOT aligned turn off m_dalign since
1320 * allocator alignment is within an ag, therefore ag has
1321 * to be aligned at stripe boundary.
1323 error = xfs_update_alignment(mp);
1327 xfs_alloc_compute_maxlevels(mp);
1328 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1329 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1330 xfs_ialloc_compute_maxlevels(mp);
1332 xfs_set_maxicount(mp);
1334 error = xfs_uuid_mount(mp);
1339 * Set the minimum read and write sizes
1341 xfs_set_rw_sizes(mp);
1343 /* set the low space thresholds for dynamic preallocation */
1344 xfs_set_low_space_thresholds(mp);
1347 * Set the inode cluster size.
1348 * This may still be overridden by the file system
1349 * block size if it is larger than the chosen cluster size.
1351 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1354 * Set inode alignment fields
1356 xfs_set_inoalignment(mp);
1359 * Check that the data (and log if separate) are an ok size.
1361 error = xfs_check_sizes(mp);
1363 goto out_remove_uuid;
1366 * Initialize realtime fields in the mount structure
1368 error = xfs_rtmount_init(mp);
1370 xfs_warn(mp, "RT mount failed");
1371 goto out_remove_uuid;
1375 * Copies the low order bits of the timestamp and the randomly
1376 * set "sequence" number out of a UUID.
1378 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1380 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1385 * Initialize the attribute manager's entries.
1387 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1390 * Initialize the precomputed transaction reservations values.
1395 * Allocate and initialize the per-ag data.
1397 spin_lock_init(&mp->m_perag_lock);
1398 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1399 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1401 xfs_warn(mp, "Failed per-ag init: %d", error);
1402 goto out_remove_uuid;
1405 if (!sbp->sb_logblocks) {
1406 xfs_warn(mp, "no log defined");
1407 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1408 error = XFS_ERROR(EFSCORRUPTED);
1409 goto out_free_perag;
1413 * log's mount-time initialization. Perform 1st part recovery if needed
1415 error = xfs_log_mount(mp, mp->m_logdev_targp,
1416 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1417 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1419 xfs_warn(mp, "log mount failed");
1424 * Now the log is mounted, we know if it was an unclean shutdown or
1425 * not. If it was, with the first phase of recovery has completed, we
1426 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1427 * but they are recovered transactionally in the second recovery phase
1430 * Hence we can safely re-initialise incore superblock counters from
1431 * the per-ag data. These may not be correct if the filesystem was not
1432 * cleanly unmounted, so we need to wait for recovery to finish before
1435 * If the filesystem was cleanly unmounted, then we can trust the
1436 * values in the superblock to be correct and we don't need to do
1439 * If we are currently making the filesystem, the initialisation will
1440 * fail as the perag data is in an undefined state.
1442 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1443 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1444 !mp->m_sb.sb_inprogress) {
1445 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1451 * Get and sanity-check the root inode.
1452 * Save the pointer to it in the mount structure.
1454 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1456 xfs_warn(mp, "failed to read root inode");
1457 goto out_log_dealloc;
1460 ASSERT(rip != NULL);
1462 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
1463 xfs_warn(mp, "corrupted root inode %llu: not a directory",
1464 (unsigned long long)rip->i_ino);
1465 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1466 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1468 error = XFS_ERROR(EFSCORRUPTED);
1471 mp->m_rootip = rip; /* save it */
1473 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1476 * Initialize realtime inode pointers in the mount structure
1478 error = xfs_rtmount_inodes(mp);
1481 * Free up the root inode.
1483 xfs_warn(mp, "failed to read RT inodes");
1488 * If this is a read-only mount defer the superblock updates until
1489 * the next remount into writeable mode. Otherwise we would never
1490 * perform the update e.g. for the root filesystem.
1492 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1493 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1495 xfs_warn(mp, "failed to write sb changes");
1501 * Initialise the XFS quota management subsystem for this mount
1503 if (XFS_IS_QUOTA_RUNNING(mp)) {
1504 error = xfs_qm_newmount(mp, "amount, "aflags);
1508 ASSERT(!XFS_IS_QUOTA_ON(mp));
1511 * If a file system had quotas running earlier, but decided to
1512 * mount without -o uquota/pquota/gquota options, revoke the
1513 * quotachecked license.
1515 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1516 xfs_notice(mp, "resetting quota flags");
1517 error = xfs_mount_reset_sbqflags(mp);
1524 * Finish recovering the file system. This part needed to be
1525 * delayed until after the root and real-time bitmap inodes
1526 * were consistently read in.
1528 error = xfs_log_mount_finish(mp);
1530 xfs_warn(mp, "log mount finish failed");
1535 * Complete the quota initialisation, post-log-replay component.
1538 ASSERT(mp->m_qflags == 0);
1539 mp->m_qflags = quotaflags;
1541 xfs_qm_mount_quotas(mp);
1545 * Now we are mounted, reserve a small amount of unused space for
1546 * privileged transactions. This is needed so that transaction
1547 * space required for critical operations can dip into this pool
1548 * when at ENOSPC. This is needed for operations like create with
1549 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1550 * are not allowed to use this reserved space.
1552 * This may drive us straight to ENOSPC on mount, but that implies
1553 * we were already there on the last unmount. Warn if this occurs.
1555 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1556 resblks = xfs_default_resblks(mp);
1557 error = xfs_reserve_blocks(mp, &resblks, NULL);
1560 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1566 xfs_rtunmount_inodes(mp);
1570 xfs_log_unmount(mp);
1572 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1573 xfs_wait_buftarg(mp->m_logdev_targp);
1574 xfs_wait_buftarg(mp->m_ddev_targp);
1578 xfs_uuid_unmount(mp);
1584 * This flushes out the inodes,dquots and the superblock, unmounts the
1585 * log and makes sure that incore structures are freed.
1589 struct xfs_mount *mp)
1594 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1596 xfs_qm_unmount_quotas(mp);
1597 xfs_rtunmount_inodes(mp);
1598 IRELE(mp->m_rootip);
1601 * We can potentially deadlock here if we have an inode cluster
1602 * that has been freed has its buffer still pinned in memory because
1603 * the transaction is still sitting in a iclog. The stale inodes
1604 * on that buffer will have their flush locks held until the
1605 * transaction hits the disk and the callbacks run. the inode
1606 * flush takes the flush lock unconditionally and with nothing to
1607 * push out the iclog we will never get that unlocked. hence we
1608 * need to force the log first.
1610 xfs_log_force(mp, XFS_LOG_SYNC);
1613 * Flush all pending changes from the AIL.
1615 xfs_ail_push_all_sync(mp->m_ail);
1618 * And reclaim all inodes. At this point there should be no dirty
1619 * inodes and none should be pinned or locked, but use synchronous
1620 * reclaim just to be sure. We can stop background inode reclaim
1621 * here as well if it is still running.
1623 cancel_delayed_work_sync(&mp->m_reclaim_work);
1624 xfs_reclaim_inodes(mp, SYNC_WAIT);
1629 * Unreserve any blocks we have so that when we unmount we don't account
1630 * the reserved free space as used. This is really only necessary for
1631 * lazy superblock counting because it trusts the incore superblock
1632 * counters to be absolutely correct on clean unmount.
1634 * We don't bother correcting this elsewhere for lazy superblock
1635 * counting because on mount of an unclean filesystem we reconstruct the
1636 * correct counter value and this is irrelevant.
1638 * For non-lazy counter filesystems, this doesn't matter at all because
1639 * we only every apply deltas to the superblock and hence the incore
1640 * value does not matter....
1643 error = xfs_reserve_blocks(mp, &resblks, NULL);
1645 xfs_warn(mp, "Unable to free reserved block pool. "
1646 "Freespace may not be correct on next mount.");
1648 error = xfs_log_sbcount(mp);
1650 xfs_warn(mp, "Unable to update superblock counters. "
1651 "Freespace may not be correct on next mount.");
1653 xfs_log_unmount(mp);
1654 xfs_uuid_unmount(mp);
1657 xfs_errortag_clearall(mp, 0);
1663 xfs_fs_writable(xfs_mount_t *mp)
1665 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1666 (mp->m_flags & XFS_MOUNT_RDONLY));
1672 * Sync the superblock counters to disk.
1674 * Note this code can be called during the process of freezing, so
1675 * we may need to use the transaction allocator which does not
1676 * block when the transaction subsystem is in its frozen state.
1679 xfs_log_sbcount(xfs_mount_t *mp)
1684 if (!xfs_fs_writable(mp))
1687 xfs_icsb_sync_counters(mp, 0);
1690 * we don't need to do this if we are updating the superblock
1691 * counters on every modification.
1693 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1696 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1697 error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0,
1698 XFS_DEFAULT_LOG_COUNT);
1700 xfs_trans_cancel(tp, 0);
1704 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1705 xfs_trans_set_sync(tp);
1706 error = xfs_trans_commit(tp, 0);
1711 * xfs_mod_sb() can be used to copy arbitrary changes to the
1712 * in-core superblock into the superblock buffer to be logged.
1713 * It does not provide the higher level of locking that is
1714 * needed to protect the in-core superblock from concurrent
1718 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1730 bp = xfs_trans_getsb(tp, mp, 0);
1731 first = sizeof(xfs_sb_t);
1734 /* translate/copy */
1736 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1738 /* find modified range */
1739 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1740 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1741 last = xfs_sb_info[f + 1].offset - 1;
1743 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1744 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1745 first = xfs_sb_info[f].offset;
1747 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1748 xfs_trans_log_buf(tp, bp, first, last);
1753 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1754 * a delta to a specified field in the in-core superblock. Simply
1755 * switch on the field indicated and apply the delta to that field.
1756 * Fields are not allowed to dip below zero, so if the delta would
1757 * do this do not apply it and return EINVAL.
1759 * The m_sb_lock must be held when this routine is called.
1762 xfs_mod_incore_sb_unlocked(
1764 xfs_sb_field_t field,
1768 int scounter; /* short counter for 32 bit fields */
1769 long long lcounter; /* long counter for 64 bit fields */
1770 long long res_used, rem;
1773 * With the in-core superblock spin lock held, switch
1774 * on the indicated field. Apply the delta to the
1775 * proper field. If the fields value would dip below
1776 * 0, then do not apply the delta and return EINVAL.
1779 case XFS_SBS_ICOUNT:
1780 lcounter = (long long)mp->m_sb.sb_icount;
1784 return XFS_ERROR(EINVAL);
1786 mp->m_sb.sb_icount = lcounter;
1789 lcounter = (long long)mp->m_sb.sb_ifree;
1793 return XFS_ERROR(EINVAL);
1795 mp->m_sb.sb_ifree = lcounter;
1797 case XFS_SBS_FDBLOCKS:
1798 lcounter = (long long)
1799 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1800 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1802 if (delta > 0) { /* Putting blocks back */
1803 if (res_used > delta) {
1804 mp->m_resblks_avail += delta;
1806 rem = delta - res_used;
1807 mp->m_resblks_avail = mp->m_resblks;
1810 } else { /* Taking blocks away */
1812 if (lcounter >= 0) {
1813 mp->m_sb.sb_fdblocks = lcounter +
1814 XFS_ALLOC_SET_ASIDE(mp);
1819 * We are out of blocks, use any available reserved
1820 * blocks if were allowed to.
1823 return XFS_ERROR(ENOSPC);
1825 lcounter = (long long)mp->m_resblks_avail + delta;
1826 if (lcounter >= 0) {
1827 mp->m_resblks_avail = lcounter;
1830 printk_once(KERN_WARNING
1831 "Filesystem \"%s\": reserve blocks depleted! "
1832 "Consider increasing reserve pool size.",
1834 return XFS_ERROR(ENOSPC);
1837 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1839 case XFS_SBS_FREXTENTS:
1840 lcounter = (long long)mp->m_sb.sb_frextents;
1843 return XFS_ERROR(ENOSPC);
1845 mp->m_sb.sb_frextents = lcounter;
1847 case XFS_SBS_DBLOCKS:
1848 lcounter = (long long)mp->m_sb.sb_dblocks;
1852 return XFS_ERROR(EINVAL);
1854 mp->m_sb.sb_dblocks = lcounter;
1856 case XFS_SBS_AGCOUNT:
1857 scounter = mp->m_sb.sb_agcount;
1861 return XFS_ERROR(EINVAL);
1863 mp->m_sb.sb_agcount = scounter;
1865 case XFS_SBS_IMAX_PCT:
1866 scounter = mp->m_sb.sb_imax_pct;
1870 return XFS_ERROR(EINVAL);
1872 mp->m_sb.sb_imax_pct = scounter;
1874 case XFS_SBS_REXTSIZE:
1875 scounter = mp->m_sb.sb_rextsize;
1879 return XFS_ERROR(EINVAL);
1881 mp->m_sb.sb_rextsize = scounter;
1883 case XFS_SBS_RBMBLOCKS:
1884 scounter = mp->m_sb.sb_rbmblocks;
1888 return XFS_ERROR(EINVAL);
1890 mp->m_sb.sb_rbmblocks = scounter;
1892 case XFS_SBS_RBLOCKS:
1893 lcounter = (long long)mp->m_sb.sb_rblocks;
1897 return XFS_ERROR(EINVAL);
1899 mp->m_sb.sb_rblocks = lcounter;
1901 case XFS_SBS_REXTENTS:
1902 lcounter = (long long)mp->m_sb.sb_rextents;
1906 return XFS_ERROR(EINVAL);
1908 mp->m_sb.sb_rextents = lcounter;
1910 case XFS_SBS_REXTSLOG:
1911 scounter = mp->m_sb.sb_rextslog;
1915 return XFS_ERROR(EINVAL);
1917 mp->m_sb.sb_rextslog = scounter;
1921 return XFS_ERROR(EINVAL);
1926 * xfs_mod_incore_sb() is used to change a field in the in-core
1927 * superblock structure by the specified delta. This modification
1928 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1929 * routine to do the work.
1933 struct xfs_mount *mp,
1934 xfs_sb_field_t field,
1940 #ifdef HAVE_PERCPU_SB
1941 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1943 spin_lock(&mp->m_sb_lock);
1944 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1945 spin_unlock(&mp->m_sb_lock);
1951 * Change more than one field in the in-core superblock structure at a time.
1953 * The fields and changes to those fields are specified in the array of
1954 * xfs_mod_sb structures passed in. Either all of the specified deltas
1955 * will be applied or none of them will. If any modified field dips below 0,
1956 * then all modifications will be backed out and EINVAL will be returned.
1958 * Note that this function may not be used for the superblock values that
1959 * are tracked with the in-memory per-cpu counters - a direct call to
1960 * xfs_icsb_modify_counters is required for these.
1963 xfs_mod_incore_sb_batch(
1964 struct xfs_mount *mp,
1973 * Loop through the array of mod structures and apply each individually.
1974 * If any fail, then back out all those which have already been applied.
1975 * Do all of this within the scope of the m_sb_lock so that all of the
1976 * changes will be atomic.
1978 spin_lock(&mp->m_sb_lock);
1979 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1980 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1981 msbp->msb_field > XFS_SBS_FDBLOCKS);
1983 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1984 msbp->msb_delta, rsvd);
1988 spin_unlock(&mp->m_sb_lock);
1992 while (--msbp >= msb) {
1993 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1994 -msbp->msb_delta, rsvd);
1997 spin_unlock(&mp->m_sb_lock);
2002 * xfs_getsb() is called to obtain the buffer for the superblock.
2003 * The buffer is returned locked and read in from disk.
2004 * The buffer should be released with a call to xfs_brelse().
2006 * If the flags parameter is BUF_TRYLOCK, then we'll only return
2007 * the superblock buffer if it can be locked without sleeping.
2008 * If it can't then we'll return NULL.
2012 struct xfs_mount *mp,
2015 struct xfs_buf *bp = mp->m_sb_bp;
2017 if (!xfs_buf_trylock(bp)) {
2018 if (flags & XBF_TRYLOCK)
2024 ASSERT(XFS_BUF_ISDONE(bp));
2029 * Used to free the superblock along various error paths.
2033 struct xfs_mount *mp)
2035 struct xfs_buf *bp = mp->m_sb_bp;
2043 * Used to log changes to the superblock unit and width fields which could
2044 * be altered by the mount options, as well as any potential sb_features2
2045 * fixup. Only the first superblock is updated.
2055 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
2056 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
2057 XFS_SB_VERSIONNUM));
2059 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
2060 error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0,
2061 XFS_DEFAULT_LOG_COUNT);
2063 xfs_trans_cancel(tp, 0);
2066 xfs_mod_sb(tp, fields);
2067 error = xfs_trans_commit(tp, 0);
2072 * If the underlying (data/log/rt) device is readonly, there are some
2073 * operations that cannot proceed.
2076 xfs_dev_is_read_only(
2077 struct xfs_mount *mp,
2080 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2081 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2082 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2083 xfs_notice(mp, "%s required on read-only device.", message);
2084 xfs_notice(mp, "write access unavailable, cannot proceed.");
2090 #ifdef HAVE_PERCPU_SB
2092 * Per-cpu incore superblock counters
2094 * Simple concept, difficult implementation
2096 * Basically, replace the incore superblock counters with a distributed per cpu
2097 * counter for contended fields (e.g. free block count).
2099 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2100 * hence needs to be accurately read when we are running low on space. Hence
2101 * there is a method to enable and disable the per-cpu counters based on how
2102 * much "stuff" is available in them.
2104 * Basically, a counter is enabled if there is enough free resource to justify
2105 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2106 * ENOSPC), then we disable the counters to synchronise all callers and
2107 * re-distribute the available resources.
2109 * If, once we redistributed the available resources, we still get a failure,
2110 * we disable the per-cpu counter and go through the slow path.
2112 * The slow path is the current xfs_mod_incore_sb() function. This means that
2113 * when we disable a per-cpu counter, we need to drain its resources back to
2114 * the global superblock. We do this after disabling the counter to prevent
2115 * more threads from queueing up on the counter.
2117 * Essentially, this means that we still need a lock in the fast path to enable
2118 * synchronisation between the global counters and the per-cpu counters. This
2119 * is not a problem because the lock will be local to a CPU almost all the time
2120 * and have little contention except when we get to ENOSPC conditions.
2122 * Basically, this lock becomes a barrier that enables us to lock out the fast
2123 * path while we do things like enabling and disabling counters and
2124 * synchronising the counters.
2128 * 1. m_sb_lock before picking up per-cpu locks
2129 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2130 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2131 * 4. modifying per-cpu counters requires holding per-cpu lock
2132 * 5. modifying global counters requires holding m_sb_lock
2133 * 6. enabling or disabling a counter requires holding the m_sb_lock
2134 * and _none_ of the per-cpu locks.
2136 * Disabled counters are only ever re-enabled by a balance operation
2137 * that results in more free resources per CPU than a given threshold.
2138 * To ensure counters don't remain disabled, they are rebalanced when
2139 * the global resource goes above a higher threshold (i.e. some hysteresis
2140 * is present to prevent thrashing).
2143 #ifdef CONFIG_HOTPLUG_CPU
2145 * hot-plug CPU notifier support.
2147 * We need a notifier per filesystem as we need to be able to identify
2148 * the filesystem to balance the counters out. This is achieved by
2149 * having a notifier block embedded in the xfs_mount_t and doing pointer
2150 * magic to get the mount pointer from the notifier block address.
2153 xfs_icsb_cpu_notify(
2154 struct notifier_block *nfb,
2155 unsigned long action,
2158 xfs_icsb_cnts_t *cntp;
2161 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2162 cntp = (xfs_icsb_cnts_t *)
2163 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2165 case CPU_UP_PREPARE:
2166 case CPU_UP_PREPARE_FROZEN:
2167 /* Easy Case - initialize the area and locks, and
2168 * then rebalance when online does everything else for us. */
2169 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2172 case CPU_ONLINE_FROZEN:
2174 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2175 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2176 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2177 xfs_icsb_unlock(mp);
2180 case CPU_DEAD_FROZEN:
2181 /* Disable all the counters, then fold the dead cpu's
2182 * count into the total on the global superblock and
2183 * re-enable the counters. */
2185 spin_lock(&mp->m_sb_lock);
2186 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2187 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2188 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2190 mp->m_sb.sb_icount += cntp->icsb_icount;
2191 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2192 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2194 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2196 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2197 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2198 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2199 spin_unlock(&mp->m_sb_lock);
2200 xfs_icsb_unlock(mp);
2206 #endif /* CONFIG_HOTPLUG_CPU */
2209 xfs_icsb_init_counters(
2212 xfs_icsb_cnts_t *cntp;
2215 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2216 if (mp->m_sb_cnts == NULL)
2219 #ifdef CONFIG_HOTPLUG_CPU
2220 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2221 mp->m_icsb_notifier.priority = 0;
2222 register_hotcpu_notifier(&mp->m_icsb_notifier);
2223 #endif /* CONFIG_HOTPLUG_CPU */
2225 for_each_online_cpu(i) {
2226 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2227 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2230 mutex_init(&mp->m_icsb_mutex);
2233 * start with all counters disabled so that the
2234 * initial balance kicks us off correctly
2236 mp->m_icsb_counters = -1;
2241 xfs_icsb_reinit_counters(
2246 * start with all counters disabled so that the
2247 * initial balance kicks us off correctly
2249 mp->m_icsb_counters = -1;
2250 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2251 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2252 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2253 xfs_icsb_unlock(mp);
2257 xfs_icsb_destroy_counters(
2260 if (mp->m_sb_cnts) {
2261 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2262 free_percpu(mp->m_sb_cnts);
2264 mutex_destroy(&mp->m_icsb_mutex);
2269 xfs_icsb_cnts_t *icsbp)
2271 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2277 xfs_icsb_unlock_cntr(
2278 xfs_icsb_cnts_t *icsbp)
2280 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2285 xfs_icsb_lock_all_counters(
2288 xfs_icsb_cnts_t *cntp;
2291 for_each_online_cpu(i) {
2292 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2293 xfs_icsb_lock_cntr(cntp);
2298 xfs_icsb_unlock_all_counters(
2301 xfs_icsb_cnts_t *cntp;
2304 for_each_online_cpu(i) {
2305 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2306 xfs_icsb_unlock_cntr(cntp);
2313 xfs_icsb_cnts_t *cnt,
2316 xfs_icsb_cnts_t *cntp;
2319 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2321 if (!(flags & XFS_ICSB_LAZY_COUNT))
2322 xfs_icsb_lock_all_counters(mp);
2324 for_each_online_cpu(i) {
2325 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2326 cnt->icsb_icount += cntp->icsb_icount;
2327 cnt->icsb_ifree += cntp->icsb_ifree;
2328 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2331 if (!(flags & XFS_ICSB_LAZY_COUNT))
2332 xfs_icsb_unlock_all_counters(mp);
2336 xfs_icsb_counter_disabled(
2338 xfs_sb_field_t field)
2340 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2341 return test_bit(field, &mp->m_icsb_counters);
2345 xfs_icsb_disable_counter(
2347 xfs_sb_field_t field)
2349 xfs_icsb_cnts_t cnt;
2351 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2354 * If we are already disabled, then there is nothing to do
2355 * here. We check before locking all the counters to avoid
2356 * the expensive lock operation when being called in the
2357 * slow path and the counter is already disabled. This is
2358 * safe because the only time we set or clear this state is under
2361 if (xfs_icsb_counter_disabled(mp, field))
2364 xfs_icsb_lock_all_counters(mp);
2365 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2366 /* drain back to superblock */
2368 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2370 case XFS_SBS_ICOUNT:
2371 mp->m_sb.sb_icount = cnt.icsb_icount;
2374 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2376 case XFS_SBS_FDBLOCKS:
2377 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2384 xfs_icsb_unlock_all_counters(mp);
2388 xfs_icsb_enable_counter(
2390 xfs_sb_field_t field,
2394 xfs_icsb_cnts_t *cntp;
2397 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2399 xfs_icsb_lock_all_counters(mp);
2400 for_each_online_cpu(i) {
2401 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2403 case XFS_SBS_ICOUNT:
2404 cntp->icsb_icount = count + resid;
2407 cntp->icsb_ifree = count + resid;
2409 case XFS_SBS_FDBLOCKS:
2410 cntp->icsb_fdblocks = count + resid;
2418 clear_bit(field, &mp->m_icsb_counters);
2419 xfs_icsb_unlock_all_counters(mp);
2423 xfs_icsb_sync_counters_locked(
2427 xfs_icsb_cnts_t cnt;
2429 xfs_icsb_count(mp, &cnt, flags);
2431 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2432 mp->m_sb.sb_icount = cnt.icsb_icount;
2433 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2434 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2435 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2436 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2440 * Accurate update of per-cpu counters to incore superblock
2443 xfs_icsb_sync_counters(
2447 spin_lock(&mp->m_sb_lock);
2448 xfs_icsb_sync_counters_locked(mp, flags);
2449 spin_unlock(&mp->m_sb_lock);
2453 * Balance and enable/disable counters as necessary.
2455 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2456 * chosen to be the same number as single on disk allocation chunk per CPU, and
2457 * free blocks is something far enough zero that we aren't going thrash when we
2458 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2459 * prevent looping endlessly when xfs_alloc_space asks for more than will
2460 * be distributed to a single CPU but each CPU has enough blocks to be
2463 * Note that we can be called when counters are already disabled.
2464 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2465 * prevent locking every per-cpu counter needlessly.
2468 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2469 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2470 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2472 xfs_icsb_balance_counter_locked(
2474 xfs_sb_field_t field,
2477 uint64_t count, resid;
2478 int weight = num_online_cpus();
2479 uint64_t min = (uint64_t)min_per_cpu;
2481 /* disable counter and sync counter */
2482 xfs_icsb_disable_counter(mp, field);
2484 /* update counters - first CPU gets residual*/
2486 case XFS_SBS_ICOUNT:
2487 count = mp->m_sb.sb_icount;
2488 resid = do_div(count, weight);
2489 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2493 count = mp->m_sb.sb_ifree;
2494 resid = do_div(count, weight);
2495 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2498 case XFS_SBS_FDBLOCKS:
2499 count = mp->m_sb.sb_fdblocks;
2500 resid = do_div(count, weight);
2501 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2506 count = resid = 0; /* quiet, gcc */
2510 xfs_icsb_enable_counter(mp, field, count, resid);
2514 xfs_icsb_balance_counter(
2516 xfs_sb_field_t fields,
2519 spin_lock(&mp->m_sb_lock);
2520 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2521 spin_unlock(&mp->m_sb_lock);
2525 xfs_icsb_modify_counters(
2527 xfs_sb_field_t field,
2531 xfs_icsb_cnts_t *icsbp;
2532 long long lcounter; /* long counter for 64 bit fields */
2538 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2541 * if the counter is disabled, go to slow path
2543 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2545 xfs_icsb_lock_cntr(icsbp);
2546 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2547 xfs_icsb_unlock_cntr(icsbp);
2552 case XFS_SBS_ICOUNT:
2553 lcounter = icsbp->icsb_icount;
2555 if (unlikely(lcounter < 0))
2556 goto balance_counter;
2557 icsbp->icsb_icount = lcounter;
2561 lcounter = icsbp->icsb_ifree;
2563 if (unlikely(lcounter < 0))
2564 goto balance_counter;
2565 icsbp->icsb_ifree = lcounter;
2568 case XFS_SBS_FDBLOCKS:
2569 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2571 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2573 if (unlikely(lcounter < 0))
2574 goto balance_counter;
2575 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2581 xfs_icsb_unlock_cntr(icsbp);
2589 * serialise with a mutex so we don't burn lots of cpu on
2590 * the superblock lock. We still need to hold the superblock
2591 * lock, however, when we modify the global structures.
2596 * Now running atomically.
2598 * If the counter is enabled, someone has beaten us to rebalancing.
2599 * Drop the lock and try again in the fast path....
2601 if (!(xfs_icsb_counter_disabled(mp, field))) {
2602 xfs_icsb_unlock(mp);
2607 * The counter is currently disabled. Because we are
2608 * running atomically here, we know a rebalance cannot
2609 * be in progress. Hence we can go straight to operating
2610 * on the global superblock. We do not call xfs_mod_incore_sb()
2611 * here even though we need to get the m_sb_lock. Doing so
2612 * will cause us to re-enter this function and deadlock.
2613 * Hence we get the m_sb_lock ourselves and then call
2614 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2615 * directly on the global counters.
2617 spin_lock(&mp->m_sb_lock);
2618 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2619 spin_unlock(&mp->m_sb_lock);
2622 * Now that we've modified the global superblock, we
2623 * may be able to re-enable the distributed counters
2624 * (e.g. lots of space just got freed). After that
2628 xfs_icsb_balance_counter(mp, field, 0);
2629 xfs_icsb_unlock(mp);
2633 xfs_icsb_unlock_cntr(icsbp);
2637 * We may have multiple threads here if multiple per-cpu
2638 * counters run dry at the same time. This will mean we can
2639 * do more balances than strictly necessary but it is not
2640 * the common slowpath case.
2645 * running atomically.
2647 * This will leave the counter in the correct state for future
2648 * accesses. After the rebalance, we simply try again and our retry
2649 * will either succeed through the fast path or slow path without
2650 * another balance operation being required.
2652 xfs_icsb_balance_counter(mp, field, delta);
2653 xfs_icsb_unlock(mp);