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"
42 #include "xfs_quota.h"
43 #include "xfs_fsops.h"
44 #include "xfs_utils.h"
45 #include "xfs_trace.h"
49 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
51 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
53 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
56 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
57 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
62 short type; /* 0 = integer
63 * 1 = binary / string (no translation)
66 { offsetof(xfs_sb_t, sb_magicnum), 0 },
67 { offsetof(xfs_sb_t, sb_blocksize), 0 },
68 { offsetof(xfs_sb_t, sb_dblocks), 0 },
69 { offsetof(xfs_sb_t, sb_rblocks), 0 },
70 { offsetof(xfs_sb_t, sb_rextents), 0 },
71 { offsetof(xfs_sb_t, sb_uuid), 1 },
72 { offsetof(xfs_sb_t, sb_logstart), 0 },
73 { offsetof(xfs_sb_t, sb_rootino), 0 },
74 { offsetof(xfs_sb_t, sb_rbmino), 0 },
75 { offsetof(xfs_sb_t, sb_rsumino), 0 },
76 { offsetof(xfs_sb_t, sb_rextsize), 0 },
77 { offsetof(xfs_sb_t, sb_agblocks), 0 },
78 { offsetof(xfs_sb_t, sb_agcount), 0 },
79 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
80 { offsetof(xfs_sb_t, sb_logblocks), 0 },
81 { offsetof(xfs_sb_t, sb_versionnum), 0 },
82 { offsetof(xfs_sb_t, sb_sectsize), 0 },
83 { offsetof(xfs_sb_t, sb_inodesize), 0 },
84 { offsetof(xfs_sb_t, sb_inopblock), 0 },
85 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
86 { offsetof(xfs_sb_t, sb_blocklog), 0 },
87 { offsetof(xfs_sb_t, sb_sectlog), 0 },
88 { offsetof(xfs_sb_t, sb_inodelog), 0 },
89 { offsetof(xfs_sb_t, sb_inopblog), 0 },
90 { offsetof(xfs_sb_t, sb_agblklog), 0 },
91 { offsetof(xfs_sb_t, sb_rextslog), 0 },
92 { offsetof(xfs_sb_t, sb_inprogress), 0 },
93 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
94 { offsetof(xfs_sb_t, sb_icount), 0 },
95 { offsetof(xfs_sb_t, sb_ifree), 0 },
96 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
97 { offsetof(xfs_sb_t, sb_frextents), 0 },
98 { offsetof(xfs_sb_t, sb_uquotino), 0 },
99 { offsetof(xfs_sb_t, sb_gquotino), 0 },
100 { offsetof(xfs_sb_t, sb_qflags), 0 },
101 { offsetof(xfs_sb_t, sb_flags), 0 },
102 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
103 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
104 { offsetof(xfs_sb_t, sb_unit), 0 },
105 { offsetof(xfs_sb_t, sb_width), 0 },
106 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
107 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
108 { offsetof(xfs_sb_t, sb_logsectsize),0 },
109 { offsetof(xfs_sb_t, sb_logsunit), 0 },
110 { offsetof(xfs_sb_t, sb_features2), 0 },
111 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
112 { sizeof(xfs_sb_t), 0 }
115 static DEFINE_MUTEX(xfs_uuid_table_mutex);
116 static int xfs_uuid_table_size;
117 static uuid_t *xfs_uuid_table;
120 * See if the UUID is unique among mounted XFS filesystems.
121 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
125 struct xfs_mount *mp)
127 uuid_t *uuid = &mp->m_sb.sb_uuid;
130 if (mp->m_flags & XFS_MOUNT_NOUUID)
133 if (uuid_is_nil(uuid)) {
134 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
135 return XFS_ERROR(EINVAL);
138 mutex_lock(&xfs_uuid_table_mutex);
139 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
140 if (uuid_is_nil(&xfs_uuid_table[i])) {
144 if (uuid_equal(uuid, &xfs_uuid_table[i]))
149 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
150 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
151 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
153 hole = xfs_uuid_table_size++;
155 xfs_uuid_table[hole] = *uuid;
156 mutex_unlock(&xfs_uuid_table_mutex);
161 mutex_unlock(&xfs_uuid_table_mutex);
162 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
163 return XFS_ERROR(EINVAL);
168 struct xfs_mount *mp)
170 uuid_t *uuid = &mp->m_sb.sb_uuid;
173 if (mp->m_flags & XFS_MOUNT_NOUUID)
176 mutex_lock(&xfs_uuid_table_mutex);
177 for (i = 0; i < xfs_uuid_table_size; i++) {
178 if (uuid_is_nil(&xfs_uuid_table[i]))
180 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
182 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
185 ASSERT(i < xfs_uuid_table_size);
186 mutex_unlock(&xfs_uuid_table_mutex);
191 * Reference counting access wrappers to the perag structures.
192 * Because we never free per-ag structures, the only thing we
193 * have to protect against changes is the tree structure itself.
196 xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
198 struct xfs_perag *pag;
202 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
204 ASSERT(atomic_read(&pag->pag_ref) >= 0);
205 ref = atomic_inc_return(&pag->pag_ref);
208 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
213 * search from @first to find the next perag with the given tag set.
217 struct xfs_mount *mp,
218 xfs_agnumber_t first,
221 struct xfs_perag *pag;
226 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
227 (void **)&pag, first, 1, tag);
232 ref = atomic_inc_return(&pag->pag_ref);
234 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
239 xfs_perag_put(struct xfs_perag *pag)
243 ASSERT(atomic_read(&pag->pag_ref) > 0);
244 ref = atomic_dec_return(&pag->pag_ref);
245 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
250 struct rcu_head *head)
252 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
254 ASSERT(atomic_read(&pag->pag_ref) == 0);
259 * Free up the per-ag resources associated with the mount structure.
266 struct xfs_perag *pag;
268 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
269 spin_lock(&mp->m_perag_lock);
270 pag = radix_tree_delete(&mp->m_perag_tree, agno);
271 spin_unlock(&mp->m_perag_lock);
273 ASSERT(atomic_read(&pag->pag_ref) == 0);
274 call_rcu(&pag->rcu_head, __xfs_free_perag);
279 * Check size of device based on the (data/realtime) block count.
280 * Note: this check is used by the growfs code as well as mount.
283 xfs_sb_validate_fsb_count(
287 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
288 ASSERT(sbp->sb_blocklog >= BBSHIFT);
290 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
291 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
293 #else /* Limited by UINT_MAX of sectors */
294 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
301 * Check the validity of the SB found.
304 xfs_mount_validate_sb(
309 int loud = !(flags & XFS_MFSI_QUIET);
312 * If the log device and data device have the
313 * same device number, the log is internal.
314 * Consequently, the sb_logstart should be non-zero. If
315 * we have a zero sb_logstart in this case, we may be trying to mount
316 * a volume filesystem in a non-volume manner.
318 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
320 xfs_warn(mp, "bad magic number");
321 return XFS_ERROR(EWRONGFS);
324 if (!xfs_sb_good_version(sbp)) {
326 xfs_warn(mp, "bad version");
327 return XFS_ERROR(EWRONGFS);
331 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
334 "filesystem is marked as having an external log; "
335 "specify logdev on the mount command line.");
336 return XFS_ERROR(EINVAL);
340 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
343 "filesystem is marked as having an internal log; "
344 "do not specify logdev on the mount command line.");
345 return XFS_ERROR(EINVAL);
349 * More sanity checking. Most of these were stolen directly from
353 sbp->sb_agcount <= 0 ||
354 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
355 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
356 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
357 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
358 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
359 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
360 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
361 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
362 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
363 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
364 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
365 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
366 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
367 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
368 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
369 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
370 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
371 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
372 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
373 sbp->sb_dblocks == 0 ||
374 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
375 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
377 XFS_CORRUPTION_ERROR("SB sanity check failed",
378 XFS_ERRLEVEL_LOW, mp, sbp);
379 return XFS_ERROR(EFSCORRUPTED);
383 * Until this is fixed only page-sized or smaller data blocks work.
385 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
388 "File system with blocksize %d bytes. "
389 "Only pagesize (%ld) or less will currently work.",
390 sbp->sb_blocksize, PAGE_SIZE);
392 return XFS_ERROR(ENOSYS);
396 * Currently only very few inode sizes are supported.
398 switch (sbp->sb_inodesize) {
406 xfs_warn(mp, "inode size of %d bytes not supported",
408 return XFS_ERROR(ENOSYS);
411 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
412 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
415 "file system too large to be mounted on this system.");
416 return XFS_ERROR(EFBIG);
419 if (unlikely(sbp->sb_inprogress)) {
421 xfs_warn(mp, "file system busy");
422 return XFS_ERROR(EFSCORRUPTED);
426 * Version 1 directory format has never worked on Linux.
428 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
431 "file system using version 1 directory format");
432 return XFS_ERROR(ENOSYS);
439 xfs_initialize_perag(
441 xfs_agnumber_t agcount,
442 xfs_agnumber_t *maxagi)
444 xfs_agnumber_t index, max_metadata;
445 xfs_agnumber_t first_initialised = 0;
449 xfs_sb_t *sbp = &mp->m_sb;
453 * Walk the current per-ag tree so we don't try to initialise AGs
454 * that already exist (growfs case). Allocate and insert all the
455 * AGs we don't find ready for initialisation.
457 for (index = 0; index < agcount; index++) {
458 pag = xfs_perag_get(mp, index);
463 if (!first_initialised)
464 first_initialised = index;
466 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
469 pag->pag_agno = index;
471 spin_lock_init(&pag->pag_ici_lock);
472 mutex_init(&pag->pag_ici_reclaim_lock);
473 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
474 spin_lock_init(&pag->pag_buf_lock);
475 pag->pag_buf_tree = RB_ROOT;
477 if (radix_tree_preload(GFP_NOFS))
480 spin_lock(&mp->m_perag_lock);
481 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
483 spin_unlock(&mp->m_perag_lock);
484 radix_tree_preload_end();
488 spin_unlock(&mp->m_perag_lock);
489 radix_tree_preload_end();
493 * If we mount with the inode64 option, or no inode overflows
494 * the legacy 32-bit address space clear the inode32 option.
496 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
497 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
499 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
500 mp->m_flags |= XFS_MOUNT_32BITINODES;
502 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
504 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
506 * Calculate how much should be reserved for inodes to meet
507 * the max inode percentage.
509 if (mp->m_maxicount) {
512 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
514 icount += sbp->sb_agblocks - 1;
515 do_div(icount, sbp->sb_agblocks);
516 max_metadata = icount;
518 max_metadata = agcount;
521 for (index = 0; index < agcount; index++) {
522 ino = XFS_AGINO_TO_INO(mp, index, agino);
523 if (ino > XFS_MAXINUMBER_32) {
528 pag = xfs_perag_get(mp, index);
529 pag->pagi_inodeok = 1;
530 if (index < max_metadata)
531 pag->pagf_metadata = 1;
535 for (index = 0; index < agcount; index++) {
536 pag = xfs_perag_get(mp, index);
537 pag->pagi_inodeok = 1;
548 for (; index > first_initialised; index--) {
549 pag = radix_tree_delete(&mp->m_perag_tree, index);
557 struct xfs_mount *mp,
560 struct xfs_sb *to = &mp->m_sb;
562 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
563 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
564 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
565 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
566 to->sb_rextents = be64_to_cpu(from->sb_rextents);
567 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
568 to->sb_logstart = be64_to_cpu(from->sb_logstart);
569 to->sb_rootino = be64_to_cpu(from->sb_rootino);
570 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
571 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
572 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
573 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
574 to->sb_agcount = be32_to_cpu(from->sb_agcount);
575 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
576 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
577 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
578 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
579 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
580 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
581 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
582 to->sb_blocklog = from->sb_blocklog;
583 to->sb_sectlog = from->sb_sectlog;
584 to->sb_inodelog = from->sb_inodelog;
585 to->sb_inopblog = from->sb_inopblog;
586 to->sb_agblklog = from->sb_agblklog;
587 to->sb_rextslog = from->sb_rextslog;
588 to->sb_inprogress = from->sb_inprogress;
589 to->sb_imax_pct = from->sb_imax_pct;
590 to->sb_icount = be64_to_cpu(from->sb_icount);
591 to->sb_ifree = be64_to_cpu(from->sb_ifree);
592 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
593 to->sb_frextents = be64_to_cpu(from->sb_frextents);
594 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
595 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
596 to->sb_qflags = be16_to_cpu(from->sb_qflags);
597 to->sb_flags = from->sb_flags;
598 to->sb_shared_vn = from->sb_shared_vn;
599 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
600 to->sb_unit = be32_to_cpu(from->sb_unit);
601 to->sb_width = be32_to_cpu(from->sb_width);
602 to->sb_dirblklog = from->sb_dirblklog;
603 to->sb_logsectlog = from->sb_logsectlog;
604 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
605 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
606 to->sb_features2 = be32_to_cpu(from->sb_features2);
607 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
611 * Copy in core superblock to ondisk one.
613 * The fields argument is mask of superblock fields to copy.
621 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
622 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
632 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
633 first = xfs_sb_info[f].offset;
634 size = xfs_sb_info[f + 1].offset - first;
636 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
638 if (size == 1 || xfs_sb_info[f].type == 1) {
639 memcpy(to_ptr + first, from_ptr + first, size);
643 *(__be16 *)(to_ptr + first) =
644 cpu_to_be16(*(__u16 *)(from_ptr + first));
647 *(__be32 *)(to_ptr + first) =
648 cpu_to_be32(*(__u32 *)(from_ptr + first));
651 *(__be64 *)(to_ptr + first) =
652 cpu_to_be64(*(__u64 *)(from_ptr + first));
659 fields &= ~(1LL << f);
666 * Does the initial read of the superblock.
669 xfs_readsb(xfs_mount_t *mp, int flags)
671 unsigned int sector_size;
674 int loud = !(flags & XFS_MFSI_QUIET);
676 ASSERT(mp->m_sb_bp == NULL);
677 ASSERT(mp->m_ddev_targp != NULL);
680 * Allocate a (locked) buffer to hold the superblock.
681 * This will be kept around at all times to optimize
682 * access to the superblock.
684 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
687 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
688 XFS_SB_DADDR, sector_size, 0);
691 xfs_warn(mp, "SB buffer read failed");
696 * Initialize the mount structure from the superblock.
697 * But first do some basic consistency checking.
699 xfs_sb_from_disk(mp, XFS_BUF_TO_SBP(bp));
700 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
703 xfs_warn(mp, "SB validate failed");
708 * We must be able to do sector-sized and sector-aligned IO.
710 if (sector_size > mp->m_sb.sb_sectsize) {
712 xfs_warn(mp, "device supports %u byte sectors (not %u)",
713 sector_size, mp->m_sb.sb_sectsize);
719 * If device sector size is smaller than the superblock size,
720 * re-read the superblock so the buffer is correctly sized.
722 if (sector_size < mp->m_sb.sb_sectsize) {
724 sector_size = mp->m_sb.sb_sectsize;
728 /* Initialize per-cpu counters */
729 xfs_icsb_reinit_counters(mp);
744 * Mount initialization code establishing various mount
745 * fields from the superblock associated with the given
749 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
751 mp->m_agfrotor = mp->m_agirotor = 0;
752 spin_lock_init(&mp->m_agirotor_lock);
753 mp->m_maxagi = mp->m_sb.sb_agcount;
754 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
755 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
756 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
757 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
758 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
759 mp->m_blockmask = sbp->sb_blocksize - 1;
760 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
761 mp->m_blockwmask = mp->m_blockwsize - 1;
763 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
764 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
765 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
766 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
768 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
769 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
770 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
771 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
773 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
774 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
775 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
776 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
778 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
779 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
781 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
785 * xfs_initialize_perag_data
787 * Read in each per-ag structure so we can count up the number of
788 * allocated inodes, free inodes and used filesystem blocks as this
789 * information is no longer persistent in the superblock. Once we have
790 * this information, write it into the in-core superblock structure.
793 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
795 xfs_agnumber_t index;
797 xfs_sb_t *sbp = &mp->m_sb;
801 uint64_t bfreelst = 0;
805 for (index = 0; index < agcount; index++) {
807 * read the agf, then the agi. This gets us
808 * all the information we need and populates the
809 * per-ag structures for us.
811 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
815 error = xfs_ialloc_pagi_init(mp, NULL, index);
818 pag = xfs_perag_get(mp, index);
819 ifree += pag->pagi_freecount;
820 ialloc += pag->pagi_count;
821 bfree += pag->pagf_freeblks;
822 bfreelst += pag->pagf_flcount;
823 btree += pag->pagf_btreeblks;
827 * Overwrite incore superblock counters with just-read data
829 spin_lock(&mp->m_sb_lock);
830 sbp->sb_ifree = ifree;
831 sbp->sb_icount = ialloc;
832 sbp->sb_fdblocks = bfree + bfreelst + btree;
833 spin_unlock(&mp->m_sb_lock);
835 /* Fixup the per-cpu counters as well. */
836 xfs_icsb_reinit_counters(mp);
842 * Update alignment values based on mount options and sb values
845 xfs_update_alignment(xfs_mount_t *mp)
847 xfs_sb_t *sbp = &(mp->m_sb);
851 * If stripe unit and stripe width are not multiples
852 * of the fs blocksize turn off alignment.
854 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
855 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
856 if (mp->m_flags & XFS_MOUNT_RETERR) {
857 xfs_warn(mp, "alignment check failed: "
858 "(sunit/swidth vs. blocksize)");
859 return XFS_ERROR(EINVAL);
861 mp->m_dalign = mp->m_swidth = 0;
864 * Convert the stripe unit and width to FSBs.
866 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
867 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
868 if (mp->m_flags & XFS_MOUNT_RETERR) {
869 xfs_warn(mp, "alignment check failed: "
870 "(sunit/swidth vs. ag size)");
871 return XFS_ERROR(EINVAL);
874 "stripe alignment turned off: sunit(%d)/swidth(%d) "
875 "incompatible with agsize(%d)",
876 mp->m_dalign, mp->m_swidth,
881 } else if (mp->m_dalign) {
882 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
884 if (mp->m_flags & XFS_MOUNT_RETERR) {
885 xfs_warn(mp, "alignment check failed: "
886 "sunit(%d) less than bsize(%d)",
889 return XFS_ERROR(EINVAL);
896 * Update superblock with new values
899 if (xfs_sb_version_hasdalign(sbp)) {
900 if (sbp->sb_unit != mp->m_dalign) {
901 sbp->sb_unit = mp->m_dalign;
902 mp->m_update_flags |= XFS_SB_UNIT;
904 if (sbp->sb_width != mp->m_swidth) {
905 sbp->sb_width = mp->m_swidth;
906 mp->m_update_flags |= XFS_SB_WIDTH;
909 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
910 xfs_sb_version_hasdalign(&mp->m_sb)) {
911 mp->m_dalign = sbp->sb_unit;
912 mp->m_swidth = sbp->sb_width;
919 * Set the maximum inode count for this filesystem
922 xfs_set_maxicount(xfs_mount_t *mp)
924 xfs_sb_t *sbp = &(mp->m_sb);
927 if (sbp->sb_imax_pct) {
929 * Make sure the maximum inode count is a multiple
930 * of the units we allocate inodes in.
932 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
934 do_div(icount, mp->m_ialloc_blks);
935 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
943 * Set the default minimum read and write sizes unless
944 * already specified in a mount option.
945 * We use smaller I/O sizes when the file system
946 * is being used for NFS service (wsync mount option).
949 xfs_set_rw_sizes(xfs_mount_t *mp)
951 xfs_sb_t *sbp = &(mp->m_sb);
952 int readio_log, writeio_log;
954 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
955 if (mp->m_flags & XFS_MOUNT_WSYNC) {
956 readio_log = XFS_WSYNC_READIO_LOG;
957 writeio_log = XFS_WSYNC_WRITEIO_LOG;
959 readio_log = XFS_READIO_LOG_LARGE;
960 writeio_log = XFS_WRITEIO_LOG_LARGE;
963 readio_log = mp->m_readio_log;
964 writeio_log = mp->m_writeio_log;
967 if (sbp->sb_blocklog > readio_log) {
968 mp->m_readio_log = sbp->sb_blocklog;
970 mp->m_readio_log = readio_log;
972 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
973 if (sbp->sb_blocklog > writeio_log) {
974 mp->m_writeio_log = sbp->sb_blocklog;
976 mp->m_writeio_log = writeio_log;
978 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
982 * precalculate the low space thresholds for dynamic speculative preallocation.
985 xfs_set_low_space_thresholds(
986 struct xfs_mount *mp)
990 for (i = 0; i < XFS_LOWSP_MAX; i++) {
991 __uint64_t space = mp->m_sb.sb_dblocks;
994 mp->m_low_space[i] = space * (i + 1);
1000 * Set whether we're using inode alignment.
1003 xfs_set_inoalignment(xfs_mount_t *mp)
1005 if (xfs_sb_version_hasalign(&mp->m_sb) &&
1006 mp->m_sb.sb_inoalignmt >=
1007 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1008 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1010 mp->m_inoalign_mask = 0;
1012 * If we are using stripe alignment, check whether
1013 * the stripe unit is a multiple of the inode alignment
1015 if (mp->m_dalign && mp->m_inoalign_mask &&
1016 !(mp->m_dalign & mp->m_inoalign_mask))
1017 mp->m_sinoalign = mp->m_dalign;
1019 mp->m_sinoalign = 0;
1023 * Check that the data (and log if separate) are an ok size.
1026 xfs_check_sizes(xfs_mount_t *mp)
1031 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1032 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1033 xfs_warn(mp, "filesystem size mismatch detected");
1034 return XFS_ERROR(EFBIG);
1036 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
1037 d - XFS_FSS_TO_BB(mp, 1),
1038 BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
1040 xfs_warn(mp, "last sector read failed");
1045 if (mp->m_logdev_targp != mp->m_ddev_targp) {
1046 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1047 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1048 xfs_warn(mp, "log size mismatch detected");
1049 return XFS_ERROR(EFBIG);
1051 bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp,
1052 d - XFS_FSB_TO_BB(mp, 1),
1053 XFS_FSB_TO_B(mp, 1), 0);
1055 xfs_warn(mp, "log device read failed");
1064 * Clear the quotaflags in memory and in the superblock.
1067 xfs_mount_reset_sbqflags(
1068 struct xfs_mount *mp)
1071 struct xfs_trans *tp;
1076 * It is OK to look at sb_qflags here in mount path,
1077 * without m_sb_lock.
1079 if (mp->m_sb.sb_qflags == 0)
1081 spin_lock(&mp->m_sb_lock);
1082 mp->m_sb.sb_qflags = 0;
1083 spin_unlock(&mp->m_sb_lock);
1086 * If the fs is readonly, let the incore superblock run
1087 * with quotas off but don't flush the update out to disk
1089 if (mp->m_flags & XFS_MOUNT_RDONLY)
1092 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1093 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1094 XFS_DEFAULT_LOG_COUNT);
1096 xfs_trans_cancel(tp, 0);
1097 xfs_alert(mp, "%s: Superblock update failed!", __func__);
1101 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1102 return xfs_trans_commit(tp, 0);
1106 xfs_default_resblks(xfs_mount_t *mp)
1111 * We default to 5% or 8192 fsbs of space reserved, whichever is
1112 * smaller. This is intended to cover concurrent allocation
1113 * transactions when we initially hit enospc. These each require a 4
1114 * block reservation. Hence by default we cover roughly 2000 concurrent
1115 * allocation reservations.
1117 resblks = mp->m_sb.sb_dblocks;
1118 do_div(resblks, 20);
1119 resblks = min_t(__uint64_t, resblks, 8192);
1124 * This function does the following on an initial mount of a file system:
1125 * - reads the superblock from disk and init the mount struct
1126 * - if we're a 32-bit kernel, do a size check on the superblock
1127 * so we don't mount terabyte filesystems
1128 * - init mount struct realtime fields
1129 * - allocate inode hash table for fs
1130 * - init directory manager
1131 * - perform recovery and init the log manager
1137 xfs_sb_t *sbp = &(mp->m_sb);
1140 uint quotamount = 0;
1141 uint quotaflags = 0;
1144 xfs_mount_common(mp, sbp);
1147 * Check for a mismatched features2 values. Older kernels
1148 * read & wrote into the wrong sb offset for sb_features2
1149 * on some platforms due to xfs_sb_t not being 64bit size aligned
1150 * when sb_features2 was added, which made older superblock
1151 * reading/writing routines swap it as a 64-bit value.
1153 * For backwards compatibility, we make both slots equal.
1155 * If we detect a mismatched field, we OR the set bits into the
1156 * existing features2 field in case it has already been modified; we
1157 * don't want to lose any features. We then update the bad location
1158 * with the ORed value so that older kernels will see any features2
1159 * flags, and mark the two fields as needing updates once the
1160 * transaction subsystem is online.
1162 if (xfs_sb_has_mismatched_features2(sbp)) {
1163 xfs_warn(mp, "correcting sb_features alignment problem");
1164 sbp->sb_features2 |= sbp->sb_bad_features2;
1165 sbp->sb_bad_features2 = sbp->sb_features2;
1166 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1169 * Re-check for ATTR2 in case it was found in bad_features2
1172 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1173 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1174 mp->m_flags |= XFS_MOUNT_ATTR2;
1177 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1178 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1179 xfs_sb_version_removeattr2(&mp->m_sb);
1180 mp->m_update_flags |= XFS_SB_FEATURES2;
1182 /* update sb_versionnum for the clearing of the morebits */
1183 if (!sbp->sb_features2)
1184 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1188 * Check if sb_agblocks is aligned at stripe boundary
1189 * If sb_agblocks is NOT aligned turn off m_dalign since
1190 * allocator alignment is within an ag, therefore ag has
1191 * to be aligned at stripe boundary.
1193 error = xfs_update_alignment(mp);
1197 xfs_alloc_compute_maxlevels(mp);
1198 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1199 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1200 xfs_ialloc_compute_maxlevels(mp);
1202 xfs_set_maxicount(mp);
1204 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1206 error = xfs_uuid_mount(mp);
1211 * Set the minimum read and write sizes
1213 xfs_set_rw_sizes(mp);
1215 /* set the low space thresholds for dynamic preallocation */
1216 xfs_set_low_space_thresholds(mp);
1219 * Set the inode cluster size.
1220 * This may still be overridden by the file system
1221 * block size if it is larger than the chosen cluster size.
1223 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1226 * Set inode alignment fields
1228 xfs_set_inoalignment(mp);
1231 * Check that the data (and log if separate) are an ok size.
1233 error = xfs_check_sizes(mp);
1235 goto out_remove_uuid;
1238 * Initialize realtime fields in the mount structure
1240 error = xfs_rtmount_init(mp);
1242 xfs_warn(mp, "RT mount failed");
1243 goto out_remove_uuid;
1247 * Copies the low order bits of the timestamp and the randomly
1248 * set "sequence" number out of a UUID.
1250 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1252 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1257 * Initialize the attribute manager's entries.
1259 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1262 * Initialize the precomputed transaction reservations values.
1267 * Allocate and initialize the per-ag data.
1269 spin_lock_init(&mp->m_perag_lock);
1270 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1271 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1273 xfs_warn(mp, "Failed per-ag init: %d", error);
1274 goto out_remove_uuid;
1277 if (!sbp->sb_logblocks) {
1278 xfs_warn(mp, "no log defined");
1279 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1280 error = XFS_ERROR(EFSCORRUPTED);
1281 goto out_free_perag;
1285 * log's mount-time initialization. Perform 1st part recovery if needed
1287 error = xfs_log_mount(mp, mp->m_logdev_targp,
1288 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1289 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1291 xfs_warn(mp, "log mount failed");
1292 goto out_free_perag;
1296 * Now the log is mounted, we know if it was an unclean shutdown or
1297 * not. If it was, with the first phase of recovery has completed, we
1298 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1299 * but they are recovered transactionally in the second recovery phase
1302 * Hence we can safely re-initialise incore superblock counters from
1303 * the per-ag data. These may not be correct if the filesystem was not
1304 * cleanly unmounted, so we need to wait for recovery to finish before
1307 * If the filesystem was cleanly unmounted, then we can trust the
1308 * values in the superblock to be correct and we don't need to do
1311 * If we are currently making the filesystem, the initialisation will
1312 * fail as the perag data is in an undefined state.
1314 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1315 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1316 !mp->m_sb.sb_inprogress) {
1317 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1319 goto out_free_perag;
1323 * Get and sanity-check the root inode.
1324 * Save the pointer to it in the mount structure.
1326 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1328 xfs_warn(mp, "failed to read root inode");
1329 goto out_log_dealloc;
1332 ASSERT(rip != NULL);
1334 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
1335 xfs_warn(mp, "corrupted root inode %llu: not a directory",
1336 (unsigned long long)rip->i_ino);
1337 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1338 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1340 error = XFS_ERROR(EFSCORRUPTED);
1343 mp->m_rootip = rip; /* save it */
1345 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1348 * Initialize realtime inode pointers in the mount structure
1350 error = xfs_rtmount_inodes(mp);
1353 * Free up the root inode.
1355 xfs_warn(mp, "failed to read RT inodes");
1360 * If this is a read-only mount defer the superblock updates until
1361 * the next remount into writeable mode. Otherwise we would never
1362 * perform the update e.g. for the root filesystem.
1364 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1365 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1367 xfs_warn(mp, "failed to write sb changes");
1373 * Initialise the XFS quota management subsystem for this mount
1375 if (XFS_IS_QUOTA_RUNNING(mp)) {
1376 error = xfs_qm_newmount(mp, "amount, "aflags);
1380 ASSERT(!XFS_IS_QUOTA_ON(mp));
1383 * If a file system had quotas running earlier, but decided to
1384 * mount without -o uquota/pquota/gquota options, revoke the
1385 * quotachecked license.
1387 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1388 xfs_notice(mp, "resetting quota flags");
1389 error = xfs_mount_reset_sbqflags(mp);
1396 * Finish recovering the file system. This part needed to be
1397 * delayed until after the root and real-time bitmap inodes
1398 * were consistently read in.
1400 error = xfs_log_mount_finish(mp);
1402 xfs_warn(mp, "log mount finish failed");
1407 * Complete the quota initialisation, post-log-replay component.
1410 ASSERT(mp->m_qflags == 0);
1411 mp->m_qflags = quotaflags;
1413 xfs_qm_mount_quotas(mp);
1417 * Now we are mounted, reserve a small amount of unused space for
1418 * privileged transactions. This is needed so that transaction
1419 * space required for critical operations can dip into this pool
1420 * when at ENOSPC. This is needed for operations like create with
1421 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1422 * are not allowed to use this reserved space.
1424 * This may drive us straight to ENOSPC on mount, but that implies
1425 * we were already there on the last unmount. Warn if this occurs.
1427 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1428 resblks = xfs_default_resblks(mp);
1429 error = xfs_reserve_blocks(mp, &resblks, NULL);
1432 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1438 xfs_rtunmount_inodes(mp);
1442 xfs_log_unmount(mp);
1446 xfs_uuid_unmount(mp);
1452 * This flushes out the inodes,dquots and the superblock, unmounts the
1453 * log and makes sure that incore structures are freed.
1457 struct xfs_mount *mp)
1462 xfs_qm_unmount_quotas(mp);
1463 xfs_rtunmount_inodes(mp);
1464 IRELE(mp->m_rootip);
1467 * We can potentially deadlock here if we have an inode cluster
1468 * that has been freed has its buffer still pinned in memory because
1469 * the transaction is still sitting in a iclog. The stale inodes
1470 * on that buffer will have their flush locks held until the
1471 * transaction hits the disk and the callbacks run. the inode
1472 * flush takes the flush lock unconditionally and with nothing to
1473 * push out the iclog we will never get that unlocked. hence we
1474 * need to force the log first.
1476 xfs_log_force(mp, XFS_LOG_SYNC);
1479 * Flush all pending changes from the AIL.
1481 xfs_ail_push_all_sync(mp->m_ail);
1484 * And reclaim all inodes. At this point there should be no dirty
1485 * inode, and none should be pinned or locked, but use synchronous
1486 * reclaim just to be sure.
1488 xfs_reclaim_inodes(mp, SYNC_WAIT);
1493 * Flush out the log synchronously so that we know for sure
1494 * that nothing is pinned. This is important because bflush()
1495 * will skip pinned buffers.
1497 xfs_log_force(mp, XFS_LOG_SYNC);
1500 * Unreserve any blocks we have so that when we unmount we don't account
1501 * the reserved free space as used. This is really only necessary for
1502 * lazy superblock counting because it trusts the incore superblock
1503 * counters to be absolutely correct on clean unmount.
1505 * We don't bother correcting this elsewhere for lazy superblock
1506 * counting because on mount of an unclean filesystem we reconstruct the
1507 * correct counter value and this is irrelevant.
1509 * For non-lazy counter filesystems, this doesn't matter at all because
1510 * we only every apply deltas to the superblock and hence the incore
1511 * value does not matter....
1514 error = xfs_reserve_blocks(mp, &resblks, NULL);
1516 xfs_warn(mp, "Unable to free reserved block pool. "
1517 "Freespace may not be correct on next mount.");
1519 error = xfs_log_sbcount(mp);
1521 xfs_warn(mp, "Unable to update superblock counters. "
1522 "Freespace may not be correct on next mount.");
1525 * At this point we might have modified the superblock again and thus
1526 * added an item to the AIL, thus flush it again.
1528 xfs_ail_push_all_sync(mp->m_ail);
1529 xfs_wait_buftarg(mp->m_ddev_targp);
1531 xfs_log_unmount_write(mp);
1532 xfs_log_unmount(mp);
1533 xfs_uuid_unmount(mp);
1536 xfs_errortag_clearall(mp, 0);
1542 xfs_fs_writable(xfs_mount_t *mp)
1544 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1545 (mp->m_flags & XFS_MOUNT_RDONLY));
1551 * Sync the superblock counters to disk.
1553 * Note this code can be called during the process of freezing, so
1554 * we may need to use the transaction allocator which does not
1555 * block when the transaction subsystem is in its frozen state.
1558 xfs_log_sbcount(xfs_mount_t *mp)
1563 if (!xfs_fs_writable(mp))
1566 xfs_icsb_sync_counters(mp, 0);
1569 * we don't need to do this if we are updating the superblock
1570 * counters on every modification.
1572 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1575 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1576 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1577 XFS_DEFAULT_LOG_COUNT);
1579 xfs_trans_cancel(tp, 0);
1583 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1584 xfs_trans_set_sync(tp);
1585 error = xfs_trans_commit(tp, 0);
1590 * xfs_mod_sb() can be used to copy arbitrary changes to the
1591 * in-core superblock into the superblock buffer to be logged.
1592 * It does not provide the higher level of locking that is
1593 * needed to protect the in-core superblock from concurrent
1597 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1609 bp = xfs_trans_getsb(tp, mp, 0);
1610 first = sizeof(xfs_sb_t);
1613 /* translate/copy */
1615 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1617 /* find modified range */
1618 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1619 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1620 last = xfs_sb_info[f + 1].offset - 1;
1622 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1623 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1624 first = xfs_sb_info[f].offset;
1626 xfs_trans_log_buf(tp, bp, first, last);
1631 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1632 * a delta to a specified field in the in-core superblock. Simply
1633 * switch on the field indicated and apply the delta to that field.
1634 * Fields are not allowed to dip below zero, so if the delta would
1635 * do this do not apply it and return EINVAL.
1637 * The m_sb_lock must be held when this routine is called.
1640 xfs_mod_incore_sb_unlocked(
1642 xfs_sb_field_t field,
1646 int scounter; /* short counter for 32 bit fields */
1647 long long lcounter; /* long counter for 64 bit fields */
1648 long long res_used, rem;
1651 * With the in-core superblock spin lock held, switch
1652 * on the indicated field. Apply the delta to the
1653 * proper field. If the fields value would dip below
1654 * 0, then do not apply the delta and return EINVAL.
1657 case XFS_SBS_ICOUNT:
1658 lcounter = (long long)mp->m_sb.sb_icount;
1662 return XFS_ERROR(EINVAL);
1664 mp->m_sb.sb_icount = lcounter;
1667 lcounter = (long long)mp->m_sb.sb_ifree;
1671 return XFS_ERROR(EINVAL);
1673 mp->m_sb.sb_ifree = lcounter;
1675 case XFS_SBS_FDBLOCKS:
1676 lcounter = (long long)
1677 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1678 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1680 if (delta > 0) { /* Putting blocks back */
1681 if (res_used > delta) {
1682 mp->m_resblks_avail += delta;
1684 rem = delta - res_used;
1685 mp->m_resblks_avail = mp->m_resblks;
1688 } else { /* Taking blocks away */
1690 if (lcounter >= 0) {
1691 mp->m_sb.sb_fdblocks = lcounter +
1692 XFS_ALLOC_SET_ASIDE(mp);
1697 * We are out of blocks, use any available reserved
1698 * blocks if were allowed to.
1701 return XFS_ERROR(ENOSPC);
1703 lcounter = (long long)mp->m_resblks_avail + delta;
1704 if (lcounter >= 0) {
1705 mp->m_resblks_avail = lcounter;
1708 printk_once(KERN_WARNING
1709 "Filesystem \"%s\": reserve blocks depleted! "
1710 "Consider increasing reserve pool size.",
1712 return XFS_ERROR(ENOSPC);
1715 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1717 case XFS_SBS_FREXTENTS:
1718 lcounter = (long long)mp->m_sb.sb_frextents;
1721 return XFS_ERROR(ENOSPC);
1723 mp->m_sb.sb_frextents = lcounter;
1725 case XFS_SBS_DBLOCKS:
1726 lcounter = (long long)mp->m_sb.sb_dblocks;
1730 return XFS_ERROR(EINVAL);
1732 mp->m_sb.sb_dblocks = lcounter;
1734 case XFS_SBS_AGCOUNT:
1735 scounter = mp->m_sb.sb_agcount;
1739 return XFS_ERROR(EINVAL);
1741 mp->m_sb.sb_agcount = scounter;
1743 case XFS_SBS_IMAX_PCT:
1744 scounter = mp->m_sb.sb_imax_pct;
1748 return XFS_ERROR(EINVAL);
1750 mp->m_sb.sb_imax_pct = scounter;
1752 case XFS_SBS_REXTSIZE:
1753 scounter = mp->m_sb.sb_rextsize;
1757 return XFS_ERROR(EINVAL);
1759 mp->m_sb.sb_rextsize = scounter;
1761 case XFS_SBS_RBMBLOCKS:
1762 scounter = mp->m_sb.sb_rbmblocks;
1766 return XFS_ERROR(EINVAL);
1768 mp->m_sb.sb_rbmblocks = scounter;
1770 case XFS_SBS_RBLOCKS:
1771 lcounter = (long long)mp->m_sb.sb_rblocks;
1775 return XFS_ERROR(EINVAL);
1777 mp->m_sb.sb_rblocks = lcounter;
1779 case XFS_SBS_REXTENTS:
1780 lcounter = (long long)mp->m_sb.sb_rextents;
1784 return XFS_ERROR(EINVAL);
1786 mp->m_sb.sb_rextents = lcounter;
1788 case XFS_SBS_REXTSLOG:
1789 scounter = mp->m_sb.sb_rextslog;
1793 return XFS_ERROR(EINVAL);
1795 mp->m_sb.sb_rextslog = scounter;
1799 return XFS_ERROR(EINVAL);
1804 * xfs_mod_incore_sb() is used to change a field in the in-core
1805 * superblock structure by the specified delta. This modification
1806 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1807 * routine to do the work.
1811 struct xfs_mount *mp,
1812 xfs_sb_field_t field,
1818 #ifdef HAVE_PERCPU_SB
1819 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1821 spin_lock(&mp->m_sb_lock);
1822 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1823 spin_unlock(&mp->m_sb_lock);
1829 * Change more than one field in the in-core superblock structure at a time.
1831 * The fields and changes to those fields are specified in the array of
1832 * xfs_mod_sb structures passed in. Either all of the specified deltas
1833 * will be applied or none of them will. If any modified field dips below 0,
1834 * then all modifications will be backed out and EINVAL will be returned.
1836 * Note that this function may not be used for the superblock values that
1837 * are tracked with the in-memory per-cpu counters - a direct call to
1838 * xfs_icsb_modify_counters is required for these.
1841 xfs_mod_incore_sb_batch(
1842 struct xfs_mount *mp,
1851 * Loop through the array of mod structures and apply each individually.
1852 * If any fail, then back out all those which have already been applied.
1853 * Do all of this within the scope of the m_sb_lock so that all of the
1854 * changes will be atomic.
1856 spin_lock(&mp->m_sb_lock);
1857 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1858 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1859 msbp->msb_field > XFS_SBS_FDBLOCKS);
1861 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1862 msbp->msb_delta, rsvd);
1866 spin_unlock(&mp->m_sb_lock);
1870 while (--msbp >= msb) {
1871 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1872 -msbp->msb_delta, rsvd);
1875 spin_unlock(&mp->m_sb_lock);
1880 * xfs_getsb() is called to obtain the buffer for the superblock.
1881 * The buffer is returned locked and read in from disk.
1882 * The buffer should be released with a call to xfs_brelse().
1884 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1885 * the superblock buffer if it can be locked without sleeping.
1886 * If it can't then we'll return NULL.
1890 struct xfs_mount *mp,
1893 struct xfs_buf *bp = mp->m_sb_bp;
1895 if (!xfs_buf_trylock(bp)) {
1896 if (flags & XBF_TRYLOCK)
1902 ASSERT(XFS_BUF_ISDONE(bp));
1907 * Used to free the superblock along various error paths.
1911 struct xfs_mount *mp)
1913 struct xfs_buf *bp = mp->m_sb_bp;
1921 * Used to log changes to the superblock unit and width fields which could
1922 * be altered by the mount options, as well as any potential sb_features2
1923 * fixup. Only the first superblock is updated.
1933 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1934 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1935 XFS_SB_VERSIONNUM));
1937 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1938 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1939 XFS_DEFAULT_LOG_COUNT);
1941 xfs_trans_cancel(tp, 0);
1944 xfs_mod_sb(tp, fields);
1945 error = xfs_trans_commit(tp, 0);
1950 * If the underlying (data/log/rt) device is readonly, there are some
1951 * operations that cannot proceed.
1954 xfs_dev_is_read_only(
1955 struct xfs_mount *mp,
1958 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1959 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1960 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1961 xfs_notice(mp, "%s required on read-only device.", message);
1962 xfs_notice(mp, "write access unavailable, cannot proceed.");
1968 #ifdef HAVE_PERCPU_SB
1970 * Per-cpu incore superblock counters
1972 * Simple concept, difficult implementation
1974 * Basically, replace the incore superblock counters with a distributed per cpu
1975 * counter for contended fields (e.g. free block count).
1977 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1978 * hence needs to be accurately read when we are running low on space. Hence
1979 * there is a method to enable and disable the per-cpu counters based on how
1980 * much "stuff" is available in them.
1982 * Basically, a counter is enabled if there is enough free resource to justify
1983 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1984 * ENOSPC), then we disable the counters to synchronise all callers and
1985 * re-distribute the available resources.
1987 * If, once we redistributed the available resources, we still get a failure,
1988 * we disable the per-cpu counter and go through the slow path.
1990 * The slow path is the current xfs_mod_incore_sb() function. This means that
1991 * when we disable a per-cpu counter, we need to drain its resources back to
1992 * the global superblock. We do this after disabling the counter to prevent
1993 * more threads from queueing up on the counter.
1995 * Essentially, this means that we still need a lock in the fast path to enable
1996 * synchronisation between the global counters and the per-cpu counters. This
1997 * is not a problem because the lock will be local to a CPU almost all the time
1998 * and have little contention except when we get to ENOSPC conditions.
2000 * Basically, this lock becomes a barrier that enables us to lock out the fast
2001 * path while we do things like enabling and disabling counters and
2002 * synchronising the counters.
2006 * 1. m_sb_lock before picking up per-cpu locks
2007 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2008 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2009 * 4. modifying per-cpu counters requires holding per-cpu lock
2010 * 5. modifying global counters requires holding m_sb_lock
2011 * 6. enabling or disabling a counter requires holding the m_sb_lock
2012 * and _none_ of the per-cpu locks.
2014 * Disabled counters are only ever re-enabled by a balance operation
2015 * that results in more free resources per CPU than a given threshold.
2016 * To ensure counters don't remain disabled, they are rebalanced when
2017 * the global resource goes above a higher threshold (i.e. some hysteresis
2018 * is present to prevent thrashing).
2021 #ifdef CONFIG_HOTPLUG_CPU
2023 * hot-plug CPU notifier support.
2025 * We need a notifier per filesystem as we need to be able to identify
2026 * the filesystem to balance the counters out. This is achieved by
2027 * having a notifier block embedded in the xfs_mount_t and doing pointer
2028 * magic to get the mount pointer from the notifier block address.
2031 xfs_icsb_cpu_notify(
2032 struct notifier_block *nfb,
2033 unsigned long action,
2036 xfs_icsb_cnts_t *cntp;
2039 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2040 cntp = (xfs_icsb_cnts_t *)
2041 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2043 case CPU_UP_PREPARE:
2044 case CPU_UP_PREPARE_FROZEN:
2045 /* Easy Case - initialize the area and locks, and
2046 * then rebalance when online does everything else for us. */
2047 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2050 case CPU_ONLINE_FROZEN:
2052 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2053 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2054 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2055 xfs_icsb_unlock(mp);
2058 case CPU_DEAD_FROZEN:
2059 /* Disable all the counters, then fold the dead cpu's
2060 * count into the total on the global superblock and
2061 * re-enable the counters. */
2063 spin_lock(&mp->m_sb_lock);
2064 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2065 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2066 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2068 mp->m_sb.sb_icount += cntp->icsb_icount;
2069 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2070 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2072 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2074 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2075 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2076 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2077 spin_unlock(&mp->m_sb_lock);
2078 xfs_icsb_unlock(mp);
2084 #endif /* CONFIG_HOTPLUG_CPU */
2087 xfs_icsb_init_counters(
2090 xfs_icsb_cnts_t *cntp;
2093 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2094 if (mp->m_sb_cnts == NULL)
2097 #ifdef CONFIG_HOTPLUG_CPU
2098 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2099 mp->m_icsb_notifier.priority = 0;
2100 register_hotcpu_notifier(&mp->m_icsb_notifier);
2101 #endif /* CONFIG_HOTPLUG_CPU */
2103 for_each_online_cpu(i) {
2104 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2105 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2108 mutex_init(&mp->m_icsb_mutex);
2111 * start with all counters disabled so that the
2112 * initial balance kicks us off correctly
2114 mp->m_icsb_counters = -1;
2119 xfs_icsb_reinit_counters(
2124 * start with all counters disabled so that the
2125 * initial balance kicks us off correctly
2127 mp->m_icsb_counters = -1;
2128 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2129 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2130 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2131 xfs_icsb_unlock(mp);
2135 xfs_icsb_destroy_counters(
2138 if (mp->m_sb_cnts) {
2139 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2140 free_percpu(mp->m_sb_cnts);
2142 mutex_destroy(&mp->m_icsb_mutex);
2147 xfs_icsb_cnts_t *icsbp)
2149 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2155 xfs_icsb_unlock_cntr(
2156 xfs_icsb_cnts_t *icsbp)
2158 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2163 xfs_icsb_lock_all_counters(
2166 xfs_icsb_cnts_t *cntp;
2169 for_each_online_cpu(i) {
2170 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2171 xfs_icsb_lock_cntr(cntp);
2176 xfs_icsb_unlock_all_counters(
2179 xfs_icsb_cnts_t *cntp;
2182 for_each_online_cpu(i) {
2183 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2184 xfs_icsb_unlock_cntr(cntp);
2191 xfs_icsb_cnts_t *cnt,
2194 xfs_icsb_cnts_t *cntp;
2197 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2199 if (!(flags & XFS_ICSB_LAZY_COUNT))
2200 xfs_icsb_lock_all_counters(mp);
2202 for_each_online_cpu(i) {
2203 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2204 cnt->icsb_icount += cntp->icsb_icount;
2205 cnt->icsb_ifree += cntp->icsb_ifree;
2206 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2209 if (!(flags & XFS_ICSB_LAZY_COUNT))
2210 xfs_icsb_unlock_all_counters(mp);
2214 xfs_icsb_counter_disabled(
2216 xfs_sb_field_t field)
2218 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2219 return test_bit(field, &mp->m_icsb_counters);
2223 xfs_icsb_disable_counter(
2225 xfs_sb_field_t field)
2227 xfs_icsb_cnts_t cnt;
2229 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2232 * If we are already disabled, then there is nothing to do
2233 * here. We check before locking all the counters to avoid
2234 * the expensive lock operation when being called in the
2235 * slow path and the counter is already disabled. This is
2236 * safe because the only time we set or clear this state is under
2239 if (xfs_icsb_counter_disabled(mp, field))
2242 xfs_icsb_lock_all_counters(mp);
2243 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2244 /* drain back to superblock */
2246 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2248 case XFS_SBS_ICOUNT:
2249 mp->m_sb.sb_icount = cnt.icsb_icount;
2252 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2254 case XFS_SBS_FDBLOCKS:
2255 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2262 xfs_icsb_unlock_all_counters(mp);
2266 xfs_icsb_enable_counter(
2268 xfs_sb_field_t field,
2272 xfs_icsb_cnts_t *cntp;
2275 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2277 xfs_icsb_lock_all_counters(mp);
2278 for_each_online_cpu(i) {
2279 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2281 case XFS_SBS_ICOUNT:
2282 cntp->icsb_icount = count + resid;
2285 cntp->icsb_ifree = count + resid;
2287 case XFS_SBS_FDBLOCKS:
2288 cntp->icsb_fdblocks = count + resid;
2296 clear_bit(field, &mp->m_icsb_counters);
2297 xfs_icsb_unlock_all_counters(mp);
2301 xfs_icsb_sync_counters_locked(
2305 xfs_icsb_cnts_t cnt;
2307 xfs_icsb_count(mp, &cnt, flags);
2309 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2310 mp->m_sb.sb_icount = cnt.icsb_icount;
2311 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2312 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2313 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2314 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2318 * Accurate update of per-cpu counters to incore superblock
2321 xfs_icsb_sync_counters(
2325 spin_lock(&mp->m_sb_lock);
2326 xfs_icsb_sync_counters_locked(mp, flags);
2327 spin_unlock(&mp->m_sb_lock);
2331 * Balance and enable/disable counters as necessary.
2333 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2334 * chosen to be the same number as single on disk allocation chunk per CPU, and
2335 * free blocks is something far enough zero that we aren't going thrash when we
2336 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2337 * prevent looping endlessly when xfs_alloc_space asks for more than will
2338 * be distributed to a single CPU but each CPU has enough blocks to be
2341 * Note that we can be called when counters are already disabled.
2342 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2343 * prevent locking every per-cpu counter needlessly.
2346 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2347 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2348 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2350 xfs_icsb_balance_counter_locked(
2352 xfs_sb_field_t field,
2355 uint64_t count, resid;
2356 int weight = num_online_cpus();
2357 uint64_t min = (uint64_t)min_per_cpu;
2359 /* disable counter and sync counter */
2360 xfs_icsb_disable_counter(mp, field);
2362 /* update counters - first CPU gets residual*/
2364 case XFS_SBS_ICOUNT:
2365 count = mp->m_sb.sb_icount;
2366 resid = do_div(count, weight);
2367 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2371 count = mp->m_sb.sb_ifree;
2372 resid = do_div(count, weight);
2373 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2376 case XFS_SBS_FDBLOCKS:
2377 count = mp->m_sb.sb_fdblocks;
2378 resid = do_div(count, weight);
2379 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2384 count = resid = 0; /* quiet, gcc */
2388 xfs_icsb_enable_counter(mp, field, count, resid);
2392 xfs_icsb_balance_counter(
2394 xfs_sb_field_t fields,
2397 spin_lock(&mp->m_sb_lock);
2398 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2399 spin_unlock(&mp->m_sb_lock);
2403 xfs_icsb_modify_counters(
2405 xfs_sb_field_t field,
2409 xfs_icsb_cnts_t *icsbp;
2410 long long lcounter; /* long counter for 64 bit fields */
2416 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2419 * if the counter is disabled, go to slow path
2421 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2423 xfs_icsb_lock_cntr(icsbp);
2424 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2425 xfs_icsb_unlock_cntr(icsbp);
2430 case XFS_SBS_ICOUNT:
2431 lcounter = icsbp->icsb_icount;
2433 if (unlikely(lcounter < 0))
2434 goto balance_counter;
2435 icsbp->icsb_icount = lcounter;
2439 lcounter = icsbp->icsb_ifree;
2441 if (unlikely(lcounter < 0))
2442 goto balance_counter;
2443 icsbp->icsb_ifree = lcounter;
2446 case XFS_SBS_FDBLOCKS:
2447 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2449 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2451 if (unlikely(lcounter < 0))
2452 goto balance_counter;
2453 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2459 xfs_icsb_unlock_cntr(icsbp);
2467 * serialise with a mutex so we don't burn lots of cpu on
2468 * the superblock lock. We still need to hold the superblock
2469 * lock, however, when we modify the global structures.
2474 * Now running atomically.
2476 * If the counter is enabled, someone has beaten us to rebalancing.
2477 * Drop the lock and try again in the fast path....
2479 if (!(xfs_icsb_counter_disabled(mp, field))) {
2480 xfs_icsb_unlock(mp);
2485 * The counter is currently disabled. Because we are
2486 * running atomically here, we know a rebalance cannot
2487 * be in progress. Hence we can go straight to operating
2488 * on the global superblock. We do not call xfs_mod_incore_sb()
2489 * here even though we need to get the m_sb_lock. Doing so
2490 * will cause us to re-enter this function and deadlock.
2491 * Hence we get the m_sb_lock ourselves and then call
2492 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2493 * directly on the global counters.
2495 spin_lock(&mp->m_sb_lock);
2496 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2497 spin_unlock(&mp->m_sb_lock);
2500 * Now that we've modified the global superblock, we
2501 * may be able to re-enable the distributed counters
2502 * (e.g. lots of space just got freed). After that
2506 xfs_icsb_balance_counter(mp, field, 0);
2507 xfs_icsb_unlock(mp);
2511 xfs_icsb_unlock_cntr(icsbp);
2515 * We may have multiple threads here if multiple per-cpu
2516 * counters run dry at the same time. This will mean we can
2517 * do more balances than strictly necessary but it is not
2518 * the common slowpath case.
2523 * running atomically.
2525 * This will leave the counter in the correct state for future
2526 * accesses. After the rebalance, we simply try again and our retry
2527 * will either succeed through the fast path or slow path without
2528 * another balance operation being required.
2530 xfs_icsb_balance_counter(mp, field, delta);
2531 xfs_icsb_unlock(mp);