2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_btree.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_alloc.h"
39 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_fsops.h"
45 #include "xfs_utils.h"
46 #include "xfs_trace.h"
49 STATIC void xfs_unmountfs_wait(xfs_mount_t *);
53 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
55 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
57 STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
59 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
63 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
64 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
65 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
71 short type; /* 0 = integer
72 * 1 = binary / string (no translation)
75 { offsetof(xfs_sb_t, sb_magicnum), 0 },
76 { offsetof(xfs_sb_t, sb_blocksize), 0 },
77 { offsetof(xfs_sb_t, sb_dblocks), 0 },
78 { offsetof(xfs_sb_t, sb_rblocks), 0 },
79 { offsetof(xfs_sb_t, sb_rextents), 0 },
80 { offsetof(xfs_sb_t, sb_uuid), 1 },
81 { offsetof(xfs_sb_t, sb_logstart), 0 },
82 { offsetof(xfs_sb_t, sb_rootino), 0 },
83 { offsetof(xfs_sb_t, sb_rbmino), 0 },
84 { offsetof(xfs_sb_t, sb_rsumino), 0 },
85 { offsetof(xfs_sb_t, sb_rextsize), 0 },
86 { offsetof(xfs_sb_t, sb_agblocks), 0 },
87 { offsetof(xfs_sb_t, sb_agcount), 0 },
88 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
89 { offsetof(xfs_sb_t, sb_logblocks), 0 },
90 { offsetof(xfs_sb_t, sb_versionnum), 0 },
91 { offsetof(xfs_sb_t, sb_sectsize), 0 },
92 { offsetof(xfs_sb_t, sb_inodesize), 0 },
93 { offsetof(xfs_sb_t, sb_inopblock), 0 },
94 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
95 { offsetof(xfs_sb_t, sb_blocklog), 0 },
96 { offsetof(xfs_sb_t, sb_sectlog), 0 },
97 { offsetof(xfs_sb_t, sb_inodelog), 0 },
98 { offsetof(xfs_sb_t, sb_inopblog), 0 },
99 { offsetof(xfs_sb_t, sb_agblklog), 0 },
100 { offsetof(xfs_sb_t, sb_rextslog), 0 },
101 { offsetof(xfs_sb_t, sb_inprogress), 0 },
102 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
103 { offsetof(xfs_sb_t, sb_icount), 0 },
104 { offsetof(xfs_sb_t, sb_ifree), 0 },
105 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
106 { offsetof(xfs_sb_t, sb_frextents), 0 },
107 { offsetof(xfs_sb_t, sb_uquotino), 0 },
108 { offsetof(xfs_sb_t, sb_gquotino), 0 },
109 { offsetof(xfs_sb_t, sb_qflags), 0 },
110 { offsetof(xfs_sb_t, sb_flags), 0 },
111 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
112 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
113 { offsetof(xfs_sb_t, sb_unit), 0 },
114 { offsetof(xfs_sb_t, sb_width), 0 },
115 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
116 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
117 { offsetof(xfs_sb_t, sb_logsectsize),0 },
118 { offsetof(xfs_sb_t, sb_logsunit), 0 },
119 { offsetof(xfs_sb_t, sb_features2), 0 },
120 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
121 { sizeof(xfs_sb_t), 0 }
124 static DEFINE_MUTEX(xfs_uuid_table_mutex);
125 static int xfs_uuid_table_size;
126 static uuid_t *xfs_uuid_table;
129 * See if the UUID is unique among mounted XFS filesystems.
130 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
134 struct xfs_mount *mp)
136 uuid_t *uuid = &mp->m_sb.sb_uuid;
139 if (mp->m_flags & XFS_MOUNT_NOUUID)
142 if (uuid_is_nil(uuid)) {
144 "XFS: Filesystem %s has nil UUID - can't mount",
146 return XFS_ERROR(EINVAL);
149 mutex_lock(&xfs_uuid_table_mutex);
150 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
151 if (uuid_is_nil(&xfs_uuid_table[i])) {
155 if (uuid_equal(uuid, &xfs_uuid_table[i]))
160 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
161 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
162 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
164 hole = xfs_uuid_table_size++;
166 xfs_uuid_table[hole] = *uuid;
167 mutex_unlock(&xfs_uuid_table_mutex);
172 mutex_unlock(&xfs_uuid_table_mutex);
173 cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
175 return XFS_ERROR(EINVAL);
180 struct xfs_mount *mp)
182 uuid_t *uuid = &mp->m_sb.sb_uuid;
185 if (mp->m_flags & XFS_MOUNT_NOUUID)
188 mutex_lock(&xfs_uuid_table_mutex);
189 for (i = 0; i < xfs_uuid_table_size; i++) {
190 if (uuid_is_nil(&xfs_uuid_table[i]))
192 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
194 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
197 ASSERT(i < xfs_uuid_table_size);
198 mutex_unlock(&xfs_uuid_table_mutex);
203 * Reference counting access wrappers to the perag structures.
206 xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
208 struct xfs_perag *pag;
211 spin_lock(&mp->m_perag_lock);
212 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
214 ASSERT(atomic_read(&pag->pag_ref) >= 0);
215 /* catch leaks in the positive direction during testing */
216 ASSERT(atomic_read(&pag->pag_ref) < 1000);
217 ref = atomic_inc_return(&pag->pag_ref);
219 spin_unlock(&mp->m_perag_lock);
220 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
225 xfs_perag_put(struct xfs_perag *pag)
229 ASSERT(atomic_read(&pag->pag_ref) > 0);
230 ref = atomic_dec_return(&pag->pag_ref);
231 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
235 * Free up the resources associated with a mount structure. Assume that
236 * the structure was initially zeroed, so we can tell which fields got
244 struct xfs_perag *pag;
246 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
247 spin_lock(&mp->m_perag_lock);
248 pag = radix_tree_delete(&mp->m_perag_tree, agno);
250 ASSERT(atomic_read(&pag->pag_ref) == 0);
251 spin_unlock(&mp->m_perag_lock);
257 * Check size of device based on the (data/realtime) block count.
258 * Note: this check is used by the growfs code as well as mount.
261 xfs_sb_validate_fsb_count(
265 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
266 ASSERT(sbp->sb_blocklog >= BBSHIFT);
268 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
269 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
271 #else /* Limited by UINT_MAX of sectors */
272 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
279 * Check the validity of the SB found.
282 xfs_mount_validate_sb(
288 * If the log device and data device have the
289 * same device number, the log is internal.
290 * Consequently, the sb_logstart should be non-zero. If
291 * we have a zero sb_logstart in this case, we may be trying to mount
292 * a volume filesystem in a non-volume manner.
294 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
295 xfs_fs_mount_cmn_err(flags, "bad magic number");
296 return XFS_ERROR(EWRONGFS);
299 if (!xfs_sb_good_version(sbp)) {
300 xfs_fs_mount_cmn_err(flags, "bad version");
301 return XFS_ERROR(EWRONGFS);
305 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
306 xfs_fs_mount_cmn_err(flags,
307 "filesystem is marked as having an external log; "
308 "specify logdev on the\nmount command line.");
309 return XFS_ERROR(EINVAL);
313 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
314 xfs_fs_mount_cmn_err(flags,
315 "filesystem is marked as having an internal log; "
316 "do not specify logdev on\nthe mount command line.");
317 return XFS_ERROR(EINVAL);
321 * More sanity checking. These were stolen directly from
325 sbp->sb_agcount <= 0 ||
326 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
327 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
328 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
329 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
330 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
331 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
332 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
333 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
334 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
335 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
336 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
337 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
338 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
339 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
340 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
341 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
342 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
343 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
344 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
345 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
346 return XFS_ERROR(EFSCORRUPTED);
350 * Sanity check AG count, size fields against data size field
353 sbp->sb_dblocks == 0 ||
355 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
356 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
357 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
358 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
359 return XFS_ERROR(EFSCORRUPTED);
363 * Until this is fixed only page-sized or smaller data blocks work.
365 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
366 xfs_fs_mount_cmn_err(flags,
367 "file system with blocksize %d bytes",
369 xfs_fs_mount_cmn_err(flags,
370 "only pagesize (%ld) or less will currently work.",
372 return XFS_ERROR(ENOSYS);
376 * Currently only very few inode sizes are supported.
378 switch (sbp->sb_inodesize) {
385 xfs_fs_mount_cmn_err(flags,
386 "inode size of %d bytes not supported",
388 return XFS_ERROR(ENOSYS);
391 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
392 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
393 xfs_fs_mount_cmn_err(flags,
394 "file system too large to be mounted on this system.");
395 return XFS_ERROR(EFBIG);
398 if (unlikely(sbp->sb_inprogress)) {
399 xfs_fs_mount_cmn_err(flags, "file system busy");
400 return XFS_ERROR(EFSCORRUPTED);
404 * Version 1 directory format has never worked on Linux.
406 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
407 xfs_fs_mount_cmn_err(flags,
408 "file system using version 1 directory format");
409 return XFS_ERROR(ENOSYS);
416 xfs_initialize_perag(
418 xfs_agnumber_t agcount,
419 xfs_agnumber_t *maxagi)
421 xfs_agnumber_t index, max_metadata;
422 xfs_agnumber_t first_initialised = 0;
426 xfs_sb_t *sbp = &mp->m_sb;
430 * Walk the current per-ag tree so we don't try to initialise AGs
431 * that already exist (growfs case). Allocate and insert all the
432 * AGs we don't find ready for initialisation.
434 for (index = 0; index < agcount; index++) {
435 pag = xfs_perag_get(mp, index);
440 if (!first_initialised)
441 first_initialised = index;
443 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
446 pag->pag_agno = index;
448 rwlock_init(&pag->pag_ici_lock);
449 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
451 if (radix_tree_preload(GFP_NOFS))
454 spin_lock(&mp->m_perag_lock);
455 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
457 spin_unlock(&mp->m_perag_lock);
458 radix_tree_preload_end();
462 spin_unlock(&mp->m_perag_lock);
463 radix_tree_preload_end();
467 * If we mount with the inode64 option, or no inode overflows
468 * the legacy 32-bit address space clear the inode32 option.
470 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
471 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
473 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
474 mp->m_flags |= XFS_MOUNT_32BITINODES;
476 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
478 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
480 * Calculate how much should be reserved for inodes to meet
481 * the max inode percentage.
483 if (mp->m_maxicount) {
486 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
488 icount += sbp->sb_agblocks - 1;
489 do_div(icount, sbp->sb_agblocks);
490 max_metadata = icount;
492 max_metadata = agcount;
495 for (index = 0; index < agcount; index++) {
496 ino = XFS_AGINO_TO_INO(mp, index, agino);
497 if (ino > XFS_MAXINUMBER_32) {
502 pag = xfs_perag_get(mp, index);
503 pag->pagi_inodeok = 1;
504 if (index < max_metadata)
505 pag->pagf_metadata = 1;
509 for (index = 0; index < agcount; index++) {
510 pag = xfs_perag_get(mp, index);
511 pag->pagi_inodeok = 1;
522 for (; index > first_initialised; index--) {
523 pag = radix_tree_delete(&mp->m_perag_tree, index);
534 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
535 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
536 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
537 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
538 to->sb_rextents = be64_to_cpu(from->sb_rextents);
539 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
540 to->sb_logstart = be64_to_cpu(from->sb_logstart);
541 to->sb_rootino = be64_to_cpu(from->sb_rootino);
542 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
543 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
544 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
545 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
546 to->sb_agcount = be32_to_cpu(from->sb_agcount);
547 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
548 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
549 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
550 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
551 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
552 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
553 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
554 to->sb_blocklog = from->sb_blocklog;
555 to->sb_sectlog = from->sb_sectlog;
556 to->sb_inodelog = from->sb_inodelog;
557 to->sb_inopblog = from->sb_inopblog;
558 to->sb_agblklog = from->sb_agblklog;
559 to->sb_rextslog = from->sb_rextslog;
560 to->sb_inprogress = from->sb_inprogress;
561 to->sb_imax_pct = from->sb_imax_pct;
562 to->sb_icount = be64_to_cpu(from->sb_icount);
563 to->sb_ifree = be64_to_cpu(from->sb_ifree);
564 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
565 to->sb_frextents = be64_to_cpu(from->sb_frextents);
566 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
567 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
568 to->sb_qflags = be16_to_cpu(from->sb_qflags);
569 to->sb_flags = from->sb_flags;
570 to->sb_shared_vn = from->sb_shared_vn;
571 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
572 to->sb_unit = be32_to_cpu(from->sb_unit);
573 to->sb_width = be32_to_cpu(from->sb_width);
574 to->sb_dirblklog = from->sb_dirblklog;
575 to->sb_logsectlog = from->sb_logsectlog;
576 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
577 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
578 to->sb_features2 = be32_to_cpu(from->sb_features2);
579 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
583 * Copy in core superblock to ondisk one.
585 * The fields argument is mask of superblock fields to copy.
593 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
594 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
604 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
605 first = xfs_sb_info[f].offset;
606 size = xfs_sb_info[f + 1].offset - first;
608 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
610 if (size == 1 || xfs_sb_info[f].type == 1) {
611 memcpy(to_ptr + first, from_ptr + first, size);
615 *(__be16 *)(to_ptr + first) =
616 cpu_to_be16(*(__u16 *)(from_ptr + first));
619 *(__be32 *)(to_ptr + first) =
620 cpu_to_be32(*(__u32 *)(from_ptr + first));
623 *(__be64 *)(to_ptr + first) =
624 cpu_to_be64(*(__u64 *)(from_ptr + first));
631 fields &= ~(1LL << f);
638 * Does the initial read of the superblock.
641 xfs_readsb(xfs_mount_t *mp, int flags)
643 unsigned int sector_size;
644 unsigned int extra_flags;
648 ASSERT(mp->m_sb_bp == NULL);
649 ASSERT(mp->m_ddev_targp != NULL);
652 * Allocate a (locked) buffer to hold the superblock.
653 * This will be kept around at all times to optimize
654 * access to the superblock.
656 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
657 extra_flags = XBF_LOCK | XBF_FS_MANAGED | XBF_MAPPED;
659 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size),
661 if (!bp || XFS_BUF_ISERROR(bp)) {
662 xfs_fs_mount_cmn_err(flags, "SB read failed");
663 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
666 ASSERT(XFS_BUF_ISBUSY(bp));
667 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
670 * Initialize the mount structure from the superblock.
671 * But first do some basic consistency checking.
673 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
675 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
677 xfs_fs_mount_cmn_err(flags, "SB validate failed");
682 * We must be able to do sector-sized and sector-aligned IO.
684 if (sector_size > mp->m_sb.sb_sectsize) {
685 xfs_fs_mount_cmn_err(flags,
686 "device supports only %u byte sectors (not %u)",
687 sector_size, mp->m_sb.sb_sectsize);
693 * If device sector size is smaller than the superblock size,
694 * re-read the superblock so the buffer is correctly sized.
696 if (sector_size < mp->m_sb.sb_sectsize) {
697 XFS_BUF_UNMANAGE(bp);
699 sector_size = mp->m_sb.sb_sectsize;
700 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR,
701 BTOBB(sector_size), extra_flags);
702 if (!bp || XFS_BUF_ISERROR(bp)) {
703 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
704 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
707 ASSERT(XFS_BUF_ISBUSY(bp));
708 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
711 /* Initialize per-cpu counters */
712 xfs_icsb_reinit_counters(mp);
716 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
721 XFS_BUF_UNMANAGE(bp);
731 * Mount initialization code establishing various mount
732 * fields from the superblock associated with the given
736 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
738 mp->m_agfrotor = mp->m_agirotor = 0;
739 spin_lock_init(&mp->m_agirotor_lock);
740 mp->m_maxagi = mp->m_sb.sb_agcount;
741 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
742 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
743 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
744 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
745 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
746 mp->m_blockmask = sbp->sb_blocksize - 1;
747 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
748 mp->m_blockwmask = mp->m_blockwsize - 1;
750 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
751 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
752 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
753 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
755 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
756 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
757 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
758 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
760 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
761 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
762 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
763 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
765 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
766 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
768 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
772 * xfs_initialize_perag_data
774 * Read in each per-ag structure so we can count up the number of
775 * allocated inodes, free inodes and used filesystem blocks as this
776 * information is no longer persistent in the superblock. Once we have
777 * this information, write it into the in-core superblock structure.
780 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
782 xfs_agnumber_t index;
784 xfs_sb_t *sbp = &mp->m_sb;
788 uint64_t bfreelst = 0;
792 for (index = 0; index < agcount; index++) {
794 * read the agf, then the agi. This gets us
795 * all the information we need and populates the
796 * per-ag structures for us.
798 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
802 error = xfs_ialloc_pagi_init(mp, NULL, index);
805 pag = xfs_perag_get(mp, index);
806 ifree += pag->pagi_freecount;
807 ialloc += pag->pagi_count;
808 bfree += pag->pagf_freeblks;
809 bfreelst += pag->pagf_flcount;
810 btree += pag->pagf_btreeblks;
814 * Overwrite incore superblock counters with just-read data
816 spin_lock(&mp->m_sb_lock);
817 sbp->sb_ifree = ifree;
818 sbp->sb_icount = ialloc;
819 sbp->sb_fdblocks = bfree + bfreelst + btree;
820 spin_unlock(&mp->m_sb_lock);
822 /* Fixup the per-cpu counters as well. */
823 xfs_icsb_reinit_counters(mp);
829 * Update alignment values based on mount options and sb values
832 xfs_update_alignment(xfs_mount_t *mp)
834 xfs_sb_t *sbp = &(mp->m_sb);
838 * If stripe unit and stripe width are not multiples
839 * of the fs blocksize turn off alignment.
841 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
842 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
843 if (mp->m_flags & XFS_MOUNT_RETERR) {
845 "XFS: alignment check 1 failed");
846 return XFS_ERROR(EINVAL);
848 mp->m_dalign = mp->m_swidth = 0;
851 * Convert the stripe unit and width to FSBs.
853 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
854 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
855 if (mp->m_flags & XFS_MOUNT_RETERR) {
856 return XFS_ERROR(EINVAL);
858 xfs_fs_cmn_err(CE_WARN, mp,
859 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
860 mp->m_dalign, mp->m_swidth,
865 } else if (mp->m_dalign) {
866 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
868 if (mp->m_flags & XFS_MOUNT_RETERR) {
869 xfs_fs_cmn_err(CE_WARN, mp,
870 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
873 return XFS_ERROR(EINVAL);
880 * Update superblock with new values
883 if (xfs_sb_version_hasdalign(sbp)) {
884 if (sbp->sb_unit != mp->m_dalign) {
885 sbp->sb_unit = mp->m_dalign;
886 mp->m_update_flags |= XFS_SB_UNIT;
888 if (sbp->sb_width != mp->m_swidth) {
889 sbp->sb_width = mp->m_swidth;
890 mp->m_update_flags |= XFS_SB_WIDTH;
893 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
894 xfs_sb_version_hasdalign(&mp->m_sb)) {
895 mp->m_dalign = sbp->sb_unit;
896 mp->m_swidth = sbp->sb_width;
903 * Set the maximum inode count for this filesystem
906 xfs_set_maxicount(xfs_mount_t *mp)
908 xfs_sb_t *sbp = &(mp->m_sb);
911 if (sbp->sb_imax_pct) {
913 * Make sure the maximum inode count is a multiple
914 * of the units we allocate inodes in.
916 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
918 do_div(icount, mp->m_ialloc_blks);
919 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
927 * Set the default minimum read and write sizes unless
928 * already specified in a mount option.
929 * We use smaller I/O sizes when the file system
930 * is being used for NFS service (wsync mount option).
933 xfs_set_rw_sizes(xfs_mount_t *mp)
935 xfs_sb_t *sbp = &(mp->m_sb);
936 int readio_log, writeio_log;
938 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
939 if (mp->m_flags & XFS_MOUNT_WSYNC) {
940 readio_log = XFS_WSYNC_READIO_LOG;
941 writeio_log = XFS_WSYNC_WRITEIO_LOG;
943 readio_log = XFS_READIO_LOG_LARGE;
944 writeio_log = XFS_WRITEIO_LOG_LARGE;
947 readio_log = mp->m_readio_log;
948 writeio_log = mp->m_writeio_log;
951 if (sbp->sb_blocklog > readio_log) {
952 mp->m_readio_log = sbp->sb_blocklog;
954 mp->m_readio_log = readio_log;
956 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
957 if (sbp->sb_blocklog > writeio_log) {
958 mp->m_writeio_log = sbp->sb_blocklog;
960 mp->m_writeio_log = writeio_log;
962 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
966 * Set whether we're using inode alignment.
969 xfs_set_inoalignment(xfs_mount_t *mp)
971 if (xfs_sb_version_hasalign(&mp->m_sb) &&
972 mp->m_sb.sb_inoalignmt >=
973 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
974 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
976 mp->m_inoalign_mask = 0;
978 * If we are using stripe alignment, check whether
979 * the stripe unit is a multiple of the inode alignment
981 if (mp->m_dalign && mp->m_inoalign_mask &&
982 !(mp->m_dalign & mp->m_inoalign_mask))
983 mp->m_sinoalign = mp->m_dalign;
989 * Check that the data (and log if separate) are an ok size.
992 xfs_check_sizes(xfs_mount_t *mp)
998 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
999 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1000 cmn_err(CE_WARN, "XFS: size check 1 failed");
1001 return XFS_ERROR(EFBIG);
1003 error = xfs_read_buf(mp, mp->m_ddev_targp,
1004 d - XFS_FSS_TO_BB(mp, 1),
1005 XFS_FSS_TO_BB(mp, 1), 0, &bp);
1009 cmn_err(CE_WARN, "XFS: size check 2 failed");
1010 if (error == ENOSPC)
1011 error = XFS_ERROR(EFBIG);
1015 if (mp->m_logdev_targp != mp->m_ddev_targp) {
1016 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1017 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1018 cmn_err(CE_WARN, "XFS: size check 3 failed");
1019 return XFS_ERROR(EFBIG);
1021 error = xfs_read_buf(mp, mp->m_logdev_targp,
1022 d - XFS_FSB_TO_BB(mp, 1),
1023 XFS_FSB_TO_BB(mp, 1), 0, &bp);
1027 cmn_err(CE_WARN, "XFS: size check 3 failed");
1028 if (error == ENOSPC)
1029 error = XFS_ERROR(EFBIG);
1037 * Clear the quotaflags in memory and in the superblock.
1040 xfs_mount_reset_sbqflags(
1041 struct xfs_mount *mp)
1044 struct xfs_trans *tp;
1049 * It is OK to look at sb_qflags here in mount path,
1050 * without m_sb_lock.
1052 if (mp->m_sb.sb_qflags == 0)
1054 spin_lock(&mp->m_sb_lock);
1055 mp->m_sb.sb_qflags = 0;
1056 spin_unlock(&mp->m_sb_lock);
1059 * If the fs is readonly, let the incore superblock run
1060 * with quotas off but don't flush the update out to disk
1062 if (mp->m_flags & XFS_MOUNT_RDONLY)
1066 xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
1069 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1070 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1071 XFS_DEFAULT_LOG_COUNT);
1073 xfs_trans_cancel(tp, 0);
1074 xfs_fs_cmn_err(CE_ALERT, mp,
1075 "xfs_mount_reset_sbqflags: Superblock update failed!");
1079 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1080 return xfs_trans_commit(tp, 0);
1084 xfs_default_resblks(xfs_mount_t *mp)
1089 * We default to 5% or 8192 fsbs of space reserved, whichever is
1090 * smaller. This is intended to cover concurrent allocation
1091 * transactions when we initially hit enospc. These each require a 4
1092 * block reservation. Hence by default we cover roughly 2000 concurrent
1093 * allocation reservations.
1095 resblks = mp->m_sb.sb_dblocks;
1096 do_div(resblks, 20);
1097 resblks = min_t(__uint64_t, resblks, 8192);
1102 * This function does the following on an initial mount of a file system:
1103 * - reads the superblock from disk and init the mount struct
1104 * - if we're a 32-bit kernel, do a size check on the superblock
1105 * so we don't mount terabyte filesystems
1106 * - init mount struct realtime fields
1107 * - allocate inode hash table for fs
1108 * - init directory manager
1109 * - perform recovery and init the log manager
1115 xfs_sb_t *sbp = &(mp->m_sb);
1118 uint quotamount = 0;
1119 uint quotaflags = 0;
1122 xfs_mount_common(mp, sbp);
1125 * Check for a mismatched features2 values. Older kernels
1126 * read & wrote into the wrong sb offset for sb_features2
1127 * on some platforms due to xfs_sb_t not being 64bit size aligned
1128 * when sb_features2 was added, which made older superblock
1129 * reading/writing routines swap it as a 64-bit value.
1131 * For backwards compatibility, we make both slots equal.
1133 * If we detect a mismatched field, we OR the set bits into the
1134 * existing features2 field in case it has already been modified; we
1135 * don't want to lose any features. We then update the bad location
1136 * with the ORed value so that older kernels will see any features2
1137 * flags, and mark the two fields as needing updates once the
1138 * transaction subsystem is online.
1140 if (xfs_sb_has_mismatched_features2(sbp)) {
1142 "XFS: correcting sb_features alignment problem");
1143 sbp->sb_features2 |= sbp->sb_bad_features2;
1144 sbp->sb_bad_features2 = sbp->sb_features2;
1145 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1148 * Re-check for ATTR2 in case it was found in bad_features2
1151 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1152 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1153 mp->m_flags |= XFS_MOUNT_ATTR2;
1156 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1157 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1158 xfs_sb_version_removeattr2(&mp->m_sb);
1159 mp->m_update_flags |= XFS_SB_FEATURES2;
1161 /* update sb_versionnum for the clearing of the morebits */
1162 if (!sbp->sb_features2)
1163 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1167 * Check if sb_agblocks is aligned at stripe boundary
1168 * If sb_agblocks is NOT aligned turn off m_dalign since
1169 * allocator alignment is within an ag, therefore ag has
1170 * to be aligned at stripe boundary.
1172 error = xfs_update_alignment(mp);
1176 xfs_alloc_compute_maxlevels(mp);
1177 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1178 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1179 xfs_ialloc_compute_maxlevels(mp);
1181 xfs_set_maxicount(mp);
1183 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1185 error = xfs_uuid_mount(mp);
1190 * Set the minimum read and write sizes
1192 xfs_set_rw_sizes(mp);
1195 * Set the inode cluster size.
1196 * This may still be overridden by the file system
1197 * block size if it is larger than the chosen cluster size.
1199 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1202 * Set inode alignment fields
1204 xfs_set_inoalignment(mp);
1207 * Check that the data (and log if separate) are an ok size.
1209 error = xfs_check_sizes(mp);
1211 goto out_remove_uuid;
1214 * Initialize realtime fields in the mount structure
1216 error = xfs_rtmount_init(mp);
1218 cmn_err(CE_WARN, "XFS: RT mount failed");
1219 goto out_remove_uuid;
1223 * Copies the low order bits of the timestamp and the randomly
1224 * set "sequence" number out of a UUID.
1226 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1228 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1233 * Initialize the attribute manager's entries.
1235 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1238 * Initialize the precomputed transaction reservations values.
1243 * Allocate and initialize the per-ag data.
1245 spin_lock_init(&mp->m_perag_lock);
1246 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1247 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1249 cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error);
1250 goto out_remove_uuid;
1253 if (!sbp->sb_logblocks) {
1254 cmn_err(CE_WARN, "XFS: no log defined");
1255 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1256 error = XFS_ERROR(EFSCORRUPTED);
1257 goto out_free_perag;
1261 * log's mount-time initialization. Perform 1st part recovery if needed
1263 error = xfs_log_mount(mp, mp->m_logdev_targp,
1264 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1265 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1267 cmn_err(CE_WARN, "XFS: log mount failed");
1268 goto out_free_perag;
1272 * Now the log is mounted, we know if it was an unclean shutdown or
1273 * not. If it was, with the first phase of recovery has completed, we
1274 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1275 * but they are recovered transactionally in the second recovery phase
1278 * Hence we can safely re-initialise incore superblock counters from
1279 * the per-ag data. These may not be correct if the filesystem was not
1280 * cleanly unmounted, so we need to wait for recovery to finish before
1283 * If the filesystem was cleanly unmounted, then we can trust the
1284 * values in the superblock to be correct and we don't need to do
1287 * If we are currently making the filesystem, the initialisation will
1288 * fail as the perag data is in an undefined state.
1290 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1291 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1292 !mp->m_sb.sb_inprogress) {
1293 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1295 goto out_free_perag;
1299 * Get and sanity-check the root inode.
1300 * Save the pointer to it in the mount structure.
1302 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1304 cmn_err(CE_WARN, "XFS: failed to read root inode");
1305 goto out_log_dealloc;
1308 ASSERT(rip != NULL);
1310 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1311 cmn_err(CE_WARN, "XFS: corrupted root inode");
1312 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1313 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1314 (unsigned long long)rip->i_ino);
1315 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1316 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1318 error = XFS_ERROR(EFSCORRUPTED);
1321 mp->m_rootip = rip; /* save it */
1323 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1326 * Initialize realtime inode pointers in the mount structure
1328 error = xfs_rtmount_inodes(mp);
1331 * Free up the root inode.
1333 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1338 * If this is a read-only mount defer the superblock updates until
1339 * the next remount into writeable mode. Otherwise we would never
1340 * perform the update e.g. for the root filesystem.
1342 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1343 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1345 cmn_err(CE_WARN, "XFS: failed to write sb changes");
1351 * Initialise the XFS quota management subsystem for this mount
1353 if (XFS_IS_QUOTA_RUNNING(mp)) {
1354 error = xfs_qm_newmount(mp, "amount, "aflags);
1358 ASSERT(!XFS_IS_QUOTA_ON(mp));
1361 * If a file system had quotas running earlier, but decided to
1362 * mount without -o uquota/pquota/gquota options, revoke the
1363 * quotachecked license.
1365 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1367 "XFS: resetting qflags for filesystem %s",
1370 error = xfs_mount_reset_sbqflags(mp);
1377 * Finish recovering the file system. This part needed to be
1378 * delayed until after the root and real-time bitmap inodes
1379 * were consistently read in.
1381 error = xfs_log_mount_finish(mp);
1383 cmn_err(CE_WARN, "XFS: log mount finish failed");
1388 * Complete the quota initialisation, post-log-replay component.
1391 ASSERT(mp->m_qflags == 0);
1392 mp->m_qflags = quotaflags;
1394 xfs_qm_mount_quotas(mp);
1398 * Now we are mounted, reserve a small amount of unused space for
1399 * privileged transactions. This is needed so that transaction
1400 * space required for critical operations can dip into this pool
1401 * when at ENOSPC. This is needed for operations like create with
1402 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1403 * are not allowed to use this reserved space.
1405 * This may drive us straight to ENOSPC on mount, but that implies
1406 * we were already there on the last unmount. Warn if this occurs.
1408 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1409 resblks = xfs_default_resblks(mp);
1410 error = xfs_reserve_blocks(mp, &resblks, NULL);
1412 cmn_err(CE_WARN, "XFS: Unable to allocate reserve "
1413 "blocks. Continuing without a reserve pool.");
1419 xfs_rtunmount_inodes(mp);
1423 xfs_log_unmount(mp);
1427 xfs_uuid_unmount(mp);
1433 * This flushes out the inodes,dquots and the superblock, unmounts the
1434 * log and makes sure that incore structures are freed.
1438 struct xfs_mount *mp)
1443 xfs_qm_unmount_quotas(mp);
1444 xfs_rtunmount_inodes(mp);
1445 IRELE(mp->m_rootip);
1448 * We can potentially deadlock here if we have an inode cluster
1449 * that has been freed has its buffer still pinned in memory because
1450 * the transaction is still sitting in a iclog. The stale inodes
1451 * on that buffer will have their flush locks held until the
1452 * transaction hits the disk and the callbacks run. the inode
1453 * flush takes the flush lock unconditionally and with nothing to
1454 * push out the iclog we will never get that unlocked. hence we
1455 * need to force the log first.
1457 xfs_log_force(mp, XFS_LOG_SYNC);
1460 * Do a delwri reclaim pass first so that as many dirty inodes are
1461 * queued up for IO as possible. Then flush the buffers before making
1462 * a synchronous path to catch all the remaining inodes are reclaimed.
1463 * This makes the reclaim process as quick as possible by avoiding
1464 * synchronous writeout and blocking on inodes already in the delwri
1465 * state as much as possible.
1467 xfs_reclaim_inodes(mp, 0);
1468 XFS_bflush(mp->m_ddev_targp);
1469 xfs_reclaim_inodes(mp, SYNC_WAIT);
1474 * Flush out the log synchronously so that we know for sure
1475 * that nothing is pinned. This is important because bflush()
1476 * will skip pinned buffers.
1478 xfs_log_force(mp, XFS_LOG_SYNC);
1480 xfs_binval(mp->m_ddev_targp);
1481 if (mp->m_rtdev_targp) {
1482 xfs_binval(mp->m_rtdev_targp);
1486 * Unreserve any blocks we have so that when we unmount we don't account
1487 * the reserved free space as used. This is really only necessary for
1488 * lazy superblock counting because it trusts the incore superblock
1489 * counters to be absolutely correct on clean unmount.
1491 * We don't bother correcting this elsewhere for lazy superblock
1492 * counting because on mount of an unclean filesystem we reconstruct the
1493 * correct counter value and this is irrelevant.
1495 * For non-lazy counter filesystems, this doesn't matter at all because
1496 * we only every apply deltas to the superblock and hence the incore
1497 * value does not matter....
1500 error = xfs_reserve_blocks(mp, &resblks, NULL);
1502 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1503 "Freespace may not be correct on next mount.");
1505 error = xfs_log_sbcount(mp, 1);
1507 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1508 "Freespace may not be correct on next mount.");
1509 xfs_unmountfs_writesb(mp);
1510 xfs_unmountfs_wait(mp); /* wait for async bufs */
1511 xfs_log_unmount_write(mp);
1512 xfs_log_unmount(mp);
1513 xfs_uuid_unmount(mp);
1516 xfs_errortag_clearall(mp, 0);
1522 xfs_unmountfs_wait(xfs_mount_t *mp)
1524 if (mp->m_logdev_targp != mp->m_ddev_targp)
1525 xfs_wait_buftarg(mp->m_logdev_targp);
1526 if (mp->m_rtdev_targp)
1527 xfs_wait_buftarg(mp->m_rtdev_targp);
1528 xfs_wait_buftarg(mp->m_ddev_targp);
1532 xfs_fs_writable(xfs_mount_t *mp)
1534 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1535 (mp->m_flags & XFS_MOUNT_RDONLY));
1541 * Called either periodically to keep the on disk superblock values
1542 * roughly up to date or from unmount to make sure the values are
1543 * correct on a clean unmount.
1545 * Note this code can be called during the process of freezing, so
1546 * we may need to use the transaction allocator which does not not
1547 * block when the transaction subsystem is in its frozen state.
1557 if (!xfs_fs_writable(mp))
1560 xfs_icsb_sync_counters(mp, 0);
1563 * we don't need to do this if we are updating the superblock
1564 * counters on every modification.
1566 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1569 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1570 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1571 XFS_DEFAULT_LOG_COUNT);
1573 xfs_trans_cancel(tp, 0);
1577 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1579 xfs_trans_set_sync(tp);
1580 error = xfs_trans_commit(tp, 0);
1585 xfs_unmountfs_writesb(xfs_mount_t *mp)
1591 * skip superblock write if fs is read-only, or
1592 * if we are doing a forced umount.
1594 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1595 XFS_FORCED_SHUTDOWN(mp))) {
1597 sbp = xfs_getsb(mp, 0);
1599 XFS_BUF_UNDONE(sbp);
1600 XFS_BUF_UNREAD(sbp);
1601 XFS_BUF_UNDELAYWRITE(sbp);
1603 XFS_BUF_UNASYNC(sbp);
1604 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1605 xfsbdstrat(mp, sbp);
1606 error = xfs_iowait(sbp);
1608 xfs_ioerror_alert("xfs_unmountfs_writesb",
1609 mp, sbp, XFS_BUF_ADDR(sbp));
1616 * xfs_mod_sb() can be used to copy arbitrary changes to the
1617 * in-core superblock into the superblock buffer to be logged.
1618 * It does not provide the higher level of locking that is
1619 * needed to protect the in-core superblock from concurrent
1623 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1635 bp = xfs_trans_getsb(tp, mp, 0);
1636 first = sizeof(xfs_sb_t);
1639 /* translate/copy */
1641 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1643 /* find modified range */
1644 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1645 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1646 last = xfs_sb_info[f + 1].offset - 1;
1648 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1649 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1650 first = xfs_sb_info[f].offset;
1652 xfs_trans_log_buf(tp, bp, first, last);
1657 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1658 * a delta to a specified field in the in-core superblock. Simply
1659 * switch on the field indicated and apply the delta to that field.
1660 * Fields are not allowed to dip below zero, so if the delta would
1661 * do this do not apply it and return EINVAL.
1663 * The m_sb_lock must be held when this routine is called.
1666 xfs_mod_incore_sb_unlocked(
1668 xfs_sb_field_t field,
1672 int scounter; /* short counter for 32 bit fields */
1673 long long lcounter; /* long counter for 64 bit fields */
1674 long long res_used, rem;
1677 * With the in-core superblock spin lock held, switch
1678 * on the indicated field. Apply the delta to the
1679 * proper field. If the fields value would dip below
1680 * 0, then do not apply the delta and return EINVAL.
1683 case XFS_SBS_ICOUNT:
1684 lcounter = (long long)mp->m_sb.sb_icount;
1688 return XFS_ERROR(EINVAL);
1690 mp->m_sb.sb_icount = lcounter;
1693 lcounter = (long long)mp->m_sb.sb_ifree;
1697 return XFS_ERROR(EINVAL);
1699 mp->m_sb.sb_ifree = lcounter;
1701 case XFS_SBS_FDBLOCKS:
1702 lcounter = (long long)
1703 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1704 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1706 if (delta > 0) { /* Putting blocks back */
1707 if (res_used > delta) {
1708 mp->m_resblks_avail += delta;
1710 rem = delta - res_used;
1711 mp->m_resblks_avail = mp->m_resblks;
1714 } else { /* Taking blocks away */
1716 if (lcounter >= 0) {
1717 mp->m_sb.sb_fdblocks = lcounter +
1718 XFS_ALLOC_SET_ASIDE(mp);
1723 * We are out of blocks, use any available reserved
1724 * blocks if were allowed to.
1727 return XFS_ERROR(ENOSPC);
1729 lcounter = (long long)mp->m_resblks_avail + delta;
1730 if (lcounter >= 0) {
1731 mp->m_resblks_avail = lcounter;
1734 printk_once(KERN_WARNING
1735 "Filesystem \"%s\": reserve blocks depleted! "
1736 "Consider increasing reserve pool size.",
1738 return XFS_ERROR(ENOSPC);
1741 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1743 case XFS_SBS_FREXTENTS:
1744 lcounter = (long long)mp->m_sb.sb_frextents;
1747 return XFS_ERROR(ENOSPC);
1749 mp->m_sb.sb_frextents = lcounter;
1751 case XFS_SBS_DBLOCKS:
1752 lcounter = (long long)mp->m_sb.sb_dblocks;
1756 return XFS_ERROR(EINVAL);
1758 mp->m_sb.sb_dblocks = lcounter;
1760 case XFS_SBS_AGCOUNT:
1761 scounter = mp->m_sb.sb_agcount;
1765 return XFS_ERROR(EINVAL);
1767 mp->m_sb.sb_agcount = scounter;
1769 case XFS_SBS_IMAX_PCT:
1770 scounter = mp->m_sb.sb_imax_pct;
1774 return XFS_ERROR(EINVAL);
1776 mp->m_sb.sb_imax_pct = scounter;
1778 case XFS_SBS_REXTSIZE:
1779 scounter = mp->m_sb.sb_rextsize;
1783 return XFS_ERROR(EINVAL);
1785 mp->m_sb.sb_rextsize = scounter;
1787 case XFS_SBS_RBMBLOCKS:
1788 scounter = mp->m_sb.sb_rbmblocks;
1792 return XFS_ERROR(EINVAL);
1794 mp->m_sb.sb_rbmblocks = scounter;
1796 case XFS_SBS_RBLOCKS:
1797 lcounter = (long long)mp->m_sb.sb_rblocks;
1801 return XFS_ERROR(EINVAL);
1803 mp->m_sb.sb_rblocks = lcounter;
1805 case XFS_SBS_REXTENTS:
1806 lcounter = (long long)mp->m_sb.sb_rextents;
1810 return XFS_ERROR(EINVAL);
1812 mp->m_sb.sb_rextents = lcounter;
1814 case XFS_SBS_REXTSLOG:
1815 scounter = mp->m_sb.sb_rextslog;
1819 return XFS_ERROR(EINVAL);
1821 mp->m_sb.sb_rextslog = scounter;
1825 return XFS_ERROR(EINVAL);
1830 * xfs_mod_incore_sb() is used to change a field in the in-core
1831 * superblock structure by the specified delta. This modification
1832 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1833 * routine to do the work.
1838 xfs_sb_field_t field,
1844 /* check for per-cpu counters */
1846 #ifdef HAVE_PERCPU_SB
1847 case XFS_SBS_ICOUNT:
1849 case XFS_SBS_FDBLOCKS:
1850 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1851 status = xfs_icsb_modify_counters(mp, field,
1858 spin_lock(&mp->m_sb_lock);
1859 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1860 spin_unlock(&mp->m_sb_lock);
1868 * xfs_mod_incore_sb_batch() is used to change more than one field
1869 * in the in-core superblock structure at a time. This modification
1870 * is protected by a lock internal to this module. The fields and
1871 * changes to those fields are specified in the array of xfs_mod_sb
1872 * structures passed in.
1874 * Either all of the specified deltas will be applied or none of
1875 * them will. If any modified field dips below 0, then all modifications
1876 * will be backed out and EINVAL will be returned.
1879 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1885 * Loop through the array of mod structures and apply each
1886 * individually. If any fail, then back out all those
1887 * which have already been applied. Do all of this within
1888 * the scope of the m_sb_lock so that all of the changes will
1891 spin_lock(&mp->m_sb_lock);
1893 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1895 * Apply the delta at index n. If it fails, break
1896 * from the loop so we'll fall into the undo loop
1899 switch (msbp->msb_field) {
1900 #ifdef HAVE_PERCPU_SB
1901 case XFS_SBS_ICOUNT:
1903 case XFS_SBS_FDBLOCKS:
1904 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1905 spin_unlock(&mp->m_sb_lock);
1906 status = xfs_icsb_modify_counters(mp,
1908 msbp->msb_delta, rsvd);
1909 spin_lock(&mp->m_sb_lock);
1915 status = xfs_mod_incore_sb_unlocked(mp,
1917 msbp->msb_delta, rsvd);
1927 * If we didn't complete the loop above, then back out
1928 * any changes made to the superblock. If you add code
1929 * between the loop above and here, make sure that you
1930 * preserve the value of status. Loop back until
1931 * we step below the beginning of the array. Make sure
1932 * we don't touch anything back there.
1936 while (msbp >= msb) {
1937 switch (msbp->msb_field) {
1938 #ifdef HAVE_PERCPU_SB
1939 case XFS_SBS_ICOUNT:
1941 case XFS_SBS_FDBLOCKS:
1942 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1943 spin_unlock(&mp->m_sb_lock);
1944 status = xfs_icsb_modify_counters(mp,
1948 spin_lock(&mp->m_sb_lock);
1954 status = xfs_mod_incore_sb_unlocked(mp,
1960 ASSERT(status == 0);
1964 spin_unlock(&mp->m_sb_lock);
1969 * xfs_getsb() is called to obtain the buffer for the superblock.
1970 * The buffer is returned locked and read in from disk.
1971 * The buffer should be released with a call to xfs_brelse().
1973 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1974 * the superblock buffer if it can be locked without sleeping.
1975 * If it can't then we'll return NULL.
1984 ASSERT(mp->m_sb_bp != NULL);
1986 if (flags & XBF_TRYLOCK) {
1987 if (!XFS_BUF_CPSEMA(bp)) {
1991 XFS_BUF_PSEMA(bp, PRIBIO);
1994 ASSERT(XFS_BUF_ISDONE(bp));
1999 * Used to free the superblock along various error paths.
2008 * Use xfs_getsb() so that the buffer will be locked
2009 * when we call xfs_buf_relse().
2011 bp = xfs_getsb(mp, 0);
2012 XFS_BUF_UNMANAGE(bp);
2018 * Used to log changes to the superblock unit and width fields which could
2019 * be altered by the mount options, as well as any potential sb_features2
2020 * fixup. Only the first superblock is updated.
2030 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
2031 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
2032 XFS_SB_VERSIONNUM));
2034 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
2035 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
2036 XFS_DEFAULT_LOG_COUNT);
2038 xfs_trans_cancel(tp, 0);
2041 xfs_mod_sb(tp, fields);
2042 error = xfs_trans_commit(tp, 0);
2047 * If the underlying (data/log/rt) device is readonly, there are some
2048 * operations that cannot proceed.
2051 xfs_dev_is_read_only(
2052 struct xfs_mount *mp,
2055 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2056 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2057 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2059 "XFS: %s required on read-only device.", message);
2061 "XFS: write access unavailable, cannot proceed.");
2067 #ifdef HAVE_PERCPU_SB
2069 * Per-cpu incore superblock counters
2071 * Simple concept, difficult implementation
2073 * Basically, replace the incore superblock counters with a distributed per cpu
2074 * counter for contended fields (e.g. free block count).
2076 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2077 * hence needs to be accurately read when we are running low on space. Hence
2078 * there is a method to enable and disable the per-cpu counters based on how
2079 * much "stuff" is available in them.
2081 * Basically, a counter is enabled if there is enough free resource to justify
2082 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2083 * ENOSPC), then we disable the counters to synchronise all callers and
2084 * re-distribute the available resources.
2086 * If, once we redistributed the available resources, we still get a failure,
2087 * we disable the per-cpu counter and go through the slow path.
2089 * The slow path is the current xfs_mod_incore_sb() function. This means that
2090 * when we disable a per-cpu counter, we need to drain its resources back to
2091 * the global superblock. We do this after disabling the counter to prevent
2092 * more threads from queueing up on the counter.
2094 * Essentially, this means that we still need a lock in the fast path to enable
2095 * synchronisation between the global counters and the per-cpu counters. This
2096 * is not a problem because the lock will be local to a CPU almost all the time
2097 * and have little contention except when we get to ENOSPC conditions.
2099 * Basically, this lock becomes a barrier that enables us to lock out the fast
2100 * path while we do things like enabling and disabling counters and
2101 * synchronising the counters.
2105 * 1. m_sb_lock before picking up per-cpu locks
2106 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2107 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2108 * 4. modifying per-cpu counters requires holding per-cpu lock
2109 * 5. modifying global counters requires holding m_sb_lock
2110 * 6. enabling or disabling a counter requires holding the m_sb_lock
2111 * and _none_ of the per-cpu locks.
2113 * Disabled counters are only ever re-enabled by a balance operation
2114 * that results in more free resources per CPU than a given threshold.
2115 * To ensure counters don't remain disabled, they are rebalanced when
2116 * the global resource goes above a higher threshold (i.e. some hysteresis
2117 * is present to prevent thrashing).
2120 #ifdef CONFIG_HOTPLUG_CPU
2122 * hot-plug CPU notifier support.
2124 * We need a notifier per filesystem as we need to be able to identify
2125 * the filesystem to balance the counters out. This is achieved by
2126 * having a notifier block embedded in the xfs_mount_t and doing pointer
2127 * magic to get the mount pointer from the notifier block address.
2130 xfs_icsb_cpu_notify(
2131 struct notifier_block *nfb,
2132 unsigned long action,
2135 xfs_icsb_cnts_t *cntp;
2138 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2139 cntp = (xfs_icsb_cnts_t *)
2140 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2142 case CPU_UP_PREPARE:
2143 case CPU_UP_PREPARE_FROZEN:
2144 /* Easy Case - initialize the area and locks, and
2145 * then rebalance when online does everything else for us. */
2146 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2149 case CPU_ONLINE_FROZEN:
2151 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2152 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2153 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2154 xfs_icsb_unlock(mp);
2157 case CPU_DEAD_FROZEN:
2158 /* Disable all the counters, then fold the dead cpu's
2159 * count into the total on the global superblock and
2160 * re-enable the counters. */
2162 spin_lock(&mp->m_sb_lock);
2163 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2164 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2165 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2167 mp->m_sb.sb_icount += cntp->icsb_icount;
2168 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2169 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2171 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2173 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2174 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2175 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2176 spin_unlock(&mp->m_sb_lock);
2177 xfs_icsb_unlock(mp);
2183 #endif /* CONFIG_HOTPLUG_CPU */
2186 xfs_icsb_init_counters(
2189 xfs_icsb_cnts_t *cntp;
2192 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2193 if (mp->m_sb_cnts == NULL)
2196 #ifdef CONFIG_HOTPLUG_CPU
2197 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2198 mp->m_icsb_notifier.priority = 0;
2199 register_hotcpu_notifier(&mp->m_icsb_notifier);
2200 #endif /* CONFIG_HOTPLUG_CPU */
2202 for_each_online_cpu(i) {
2203 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2204 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2207 mutex_init(&mp->m_icsb_mutex);
2210 * start with all counters disabled so that the
2211 * initial balance kicks us off correctly
2213 mp->m_icsb_counters = -1;
2218 xfs_icsb_reinit_counters(
2223 * start with all counters disabled so that the
2224 * initial balance kicks us off correctly
2226 mp->m_icsb_counters = -1;
2227 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2228 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2229 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2230 xfs_icsb_unlock(mp);
2234 xfs_icsb_destroy_counters(
2237 if (mp->m_sb_cnts) {
2238 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2239 free_percpu(mp->m_sb_cnts);
2241 mutex_destroy(&mp->m_icsb_mutex);
2246 xfs_icsb_cnts_t *icsbp)
2248 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2254 xfs_icsb_unlock_cntr(
2255 xfs_icsb_cnts_t *icsbp)
2257 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2262 xfs_icsb_lock_all_counters(
2265 xfs_icsb_cnts_t *cntp;
2268 for_each_online_cpu(i) {
2269 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2270 xfs_icsb_lock_cntr(cntp);
2275 xfs_icsb_unlock_all_counters(
2278 xfs_icsb_cnts_t *cntp;
2281 for_each_online_cpu(i) {
2282 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2283 xfs_icsb_unlock_cntr(cntp);
2290 xfs_icsb_cnts_t *cnt,
2293 xfs_icsb_cnts_t *cntp;
2296 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2298 if (!(flags & XFS_ICSB_LAZY_COUNT))
2299 xfs_icsb_lock_all_counters(mp);
2301 for_each_online_cpu(i) {
2302 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2303 cnt->icsb_icount += cntp->icsb_icount;
2304 cnt->icsb_ifree += cntp->icsb_ifree;
2305 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2308 if (!(flags & XFS_ICSB_LAZY_COUNT))
2309 xfs_icsb_unlock_all_counters(mp);
2313 xfs_icsb_counter_disabled(
2315 xfs_sb_field_t field)
2317 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2318 return test_bit(field, &mp->m_icsb_counters);
2322 xfs_icsb_disable_counter(
2324 xfs_sb_field_t field)
2326 xfs_icsb_cnts_t cnt;
2328 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2331 * If we are already disabled, then there is nothing to do
2332 * here. We check before locking all the counters to avoid
2333 * the expensive lock operation when being called in the
2334 * slow path and the counter is already disabled. This is
2335 * safe because the only time we set or clear this state is under
2338 if (xfs_icsb_counter_disabled(mp, field))
2341 xfs_icsb_lock_all_counters(mp);
2342 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2343 /* drain back to superblock */
2345 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2347 case XFS_SBS_ICOUNT:
2348 mp->m_sb.sb_icount = cnt.icsb_icount;
2351 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2353 case XFS_SBS_FDBLOCKS:
2354 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2361 xfs_icsb_unlock_all_counters(mp);
2365 xfs_icsb_enable_counter(
2367 xfs_sb_field_t field,
2371 xfs_icsb_cnts_t *cntp;
2374 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2376 xfs_icsb_lock_all_counters(mp);
2377 for_each_online_cpu(i) {
2378 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2380 case XFS_SBS_ICOUNT:
2381 cntp->icsb_icount = count + resid;
2384 cntp->icsb_ifree = count + resid;
2386 case XFS_SBS_FDBLOCKS:
2387 cntp->icsb_fdblocks = count + resid;
2395 clear_bit(field, &mp->m_icsb_counters);
2396 xfs_icsb_unlock_all_counters(mp);
2400 xfs_icsb_sync_counters_locked(
2404 xfs_icsb_cnts_t cnt;
2406 xfs_icsb_count(mp, &cnt, flags);
2408 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2409 mp->m_sb.sb_icount = cnt.icsb_icount;
2410 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2411 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2412 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2413 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2417 * Accurate update of per-cpu counters to incore superblock
2420 xfs_icsb_sync_counters(
2424 spin_lock(&mp->m_sb_lock);
2425 xfs_icsb_sync_counters_locked(mp, flags);
2426 spin_unlock(&mp->m_sb_lock);
2430 * Balance and enable/disable counters as necessary.
2432 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2433 * chosen to be the same number as single on disk allocation chunk per CPU, and
2434 * free blocks is something far enough zero that we aren't going thrash when we
2435 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2436 * prevent looping endlessly when xfs_alloc_space asks for more than will
2437 * be distributed to a single CPU but each CPU has enough blocks to be
2440 * Note that we can be called when counters are already disabled.
2441 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2442 * prevent locking every per-cpu counter needlessly.
2445 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2446 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2447 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2449 xfs_icsb_balance_counter_locked(
2451 xfs_sb_field_t field,
2454 uint64_t count, resid;
2455 int weight = num_online_cpus();
2456 uint64_t min = (uint64_t)min_per_cpu;
2458 /* disable counter and sync counter */
2459 xfs_icsb_disable_counter(mp, field);
2461 /* update counters - first CPU gets residual*/
2463 case XFS_SBS_ICOUNT:
2464 count = mp->m_sb.sb_icount;
2465 resid = do_div(count, weight);
2466 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2470 count = mp->m_sb.sb_ifree;
2471 resid = do_div(count, weight);
2472 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2475 case XFS_SBS_FDBLOCKS:
2476 count = mp->m_sb.sb_fdblocks;
2477 resid = do_div(count, weight);
2478 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2483 count = resid = 0; /* quiet, gcc */
2487 xfs_icsb_enable_counter(mp, field, count, resid);
2491 xfs_icsb_balance_counter(
2493 xfs_sb_field_t fields,
2496 spin_lock(&mp->m_sb_lock);
2497 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2498 spin_unlock(&mp->m_sb_lock);
2502 xfs_icsb_modify_counters(
2504 xfs_sb_field_t field,
2508 xfs_icsb_cnts_t *icsbp;
2509 long long lcounter; /* long counter for 64 bit fields */
2515 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2518 * if the counter is disabled, go to slow path
2520 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2522 xfs_icsb_lock_cntr(icsbp);
2523 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2524 xfs_icsb_unlock_cntr(icsbp);
2529 case XFS_SBS_ICOUNT:
2530 lcounter = icsbp->icsb_icount;
2532 if (unlikely(lcounter < 0))
2533 goto balance_counter;
2534 icsbp->icsb_icount = lcounter;
2538 lcounter = icsbp->icsb_ifree;
2540 if (unlikely(lcounter < 0))
2541 goto balance_counter;
2542 icsbp->icsb_ifree = lcounter;
2545 case XFS_SBS_FDBLOCKS:
2546 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2548 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2550 if (unlikely(lcounter < 0))
2551 goto balance_counter;
2552 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2558 xfs_icsb_unlock_cntr(icsbp);
2566 * serialise with a mutex so we don't burn lots of cpu on
2567 * the superblock lock. We still need to hold the superblock
2568 * lock, however, when we modify the global structures.
2573 * Now running atomically.
2575 * If the counter is enabled, someone has beaten us to rebalancing.
2576 * Drop the lock and try again in the fast path....
2578 if (!(xfs_icsb_counter_disabled(mp, field))) {
2579 xfs_icsb_unlock(mp);
2584 * The counter is currently disabled. Because we are
2585 * running atomically here, we know a rebalance cannot
2586 * be in progress. Hence we can go straight to operating
2587 * on the global superblock. We do not call xfs_mod_incore_sb()
2588 * here even though we need to get the m_sb_lock. Doing so
2589 * will cause us to re-enter this function and deadlock.
2590 * Hence we get the m_sb_lock ourselves and then call
2591 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2592 * directly on the global counters.
2594 spin_lock(&mp->m_sb_lock);
2595 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2596 spin_unlock(&mp->m_sb_lock);
2599 * Now that we've modified the global superblock, we
2600 * may be able to re-enable the distributed counters
2601 * (e.g. lots of space just got freed). After that
2605 xfs_icsb_balance_counter(mp, field, 0);
2606 xfs_icsb_unlock(mp);
2610 xfs_icsb_unlock_cntr(icsbp);
2614 * We may have multiple threads here if multiple per-cpu
2615 * counters run dry at the same time. This will mean we can
2616 * do more balances than strictly necessary but it is not
2617 * the common slowpath case.
2622 * running atomically.
2624 * This will leave the counter in the correct state for future
2625 * accesses. After the rebalance, we simply try again and our retry
2626 * will either succeed through the fast path or slow path without
2627 * another balance operation being required.
2629 xfs_icsb_balance_counter(mp, field, delta);
2630 xfs_icsb_unlock(mp);