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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_inode.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
36 #include "xfs_trans.h"
37 #include "xfs_trans_priv.h"
39 #include "xfs_error.h"
40 #include "xfs_quota.h"
41 #include "xfs_fsops.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_dinode.h"
45 #include "xfs_sysfs.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)
60 static DEFINE_MUTEX(xfs_uuid_table_mutex);
61 static int xfs_uuid_table_size;
62 static uuid_t *xfs_uuid_table;
65 * See if the UUID is unique among mounted XFS filesystems.
66 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
72 uuid_t *uuid = &mp->m_sb.sb_uuid;
75 if (mp->m_flags & XFS_MOUNT_NOUUID)
78 if (uuid_is_nil(uuid)) {
79 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
83 mutex_lock(&xfs_uuid_table_mutex);
84 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
85 if (uuid_is_nil(&xfs_uuid_table[i])) {
89 if (uuid_equal(uuid, &xfs_uuid_table[i]))
94 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
95 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
96 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
98 hole = xfs_uuid_table_size++;
100 xfs_uuid_table[hole] = *uuid;
101 mutex_unlock(&xfs_uuid_table_mutex);
106 mutex_unlock(&xfs_uuid_table_mutex);
107 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
113 struct xfs_mount *mp)
115 uuid_t *uuid = &mp->m_sb.sb_uuid;
118 if (mp->m_flags & XFS_MOUNT_NOUUID)
121 mutex_lock(&xfs_uuid_table_mutex);
122 for (i = 0; i < xfs_uuid_table_size; i++) {
123 if (uuid_is_nil(&xfs_uuid_table[i]))
125 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
127 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
130 ASSERT(i < xfs_uuid_table_size);
131 mutex_unlock(&xfs_uuid_table_mutex);
137 struct rcu_head *head)
139 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
141 ASSERT(atomic_read(&pag->pag_ref) == 0);
146 * Free up the per-ag resources associated with the mount structure.
153 struct xfs_perag *pag;
155 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
156 spin_lock(&mp->m_perag_lock);
157 pag = radix_tree_delete(&mp->m_perag_tree, agno);
158 spin_unlock(&mp->m_perag_lock);
160 ASSERT(atomic_read(&pag->pag_ref) == 0);
161 call_rcu(&pag->rcu_head, __xfs_free_perag);
166 * Check size of device based on the (data/realtime) block count.
167 * Note: this check is used by the growfs code as well as mount.
170 xfs_sb_validate_fsb_count(
174 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
175 ASSERT(sbp->sb_blocklog >= BBSHIFT);
177 /* Limited by ULONG_MAX of page cache index */
178 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
184 xfs_initialize_perag(
186 xfs_agnumber_t agcount,
187 xfs_agnumber_t *maxagi)
189 xfs_agnumber_t index;
190 xfs_agnumber_t first_initialised = 0;
194 xfs_sb_t *sbp = &mp->m_sb;
198 * Walk the current per-ag tree so we don't try to initialise AGs
199 * that already exist (growfs case). Allocate and insert all the
200 * AGs we don't find ready for initialisation.
202 for (index = 0; index < agcount; index++) {
203 pag = xfs_perag_get(mp, index);
208 if (!first_initialised)
209 first_initialised = index;
211 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
214 pag->pag_agno = index;
216 spin_lock_init(&pag->pag_ici_lock);
217 mutex_init(&pag->pag_ici_reclaim_lock);
218 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
219 spin_lock_init(&pag->pag_buf_lock);
220 pag->pag_buf_tree = RB_ROOT;
222 if (radix_tree_preload(GFP_NOFS))
225 spin_lock(&mp->m_perag_lock);
226 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
228 spin_unlock(&mp->m_perag_lock);
229 radix_tree_preload_end();
233 spin_unlock(&mp->m_perag_lock);
234 radix_tree_preload_end();
238 * If we mount with the inode64 option, or no inode overflows
239 * the legacy 32-bit address space clear the inode32 option.
241 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
242 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
244 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
245 mp->m_flags |= XFS_MOUNT_32BITINODES;
247 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
249 if (mp->m_flags & XFS_MOUNT_32BITINODES)
250 index = xfs_set_inode32(mp, agcount);
252 index = xfs_set_inode64(mp, agcount);
260 for (; index > first_initialised; index--) {
261 pag = radix_tree_delete(&mp->m_perag_tree, index);
270 * Does the initial read of the superblock.
274 struct xfs_mount *mp,
277 unsigned int sector_size;
279 struct xfs_sb *sbp = &mp->m_sb;
281 int loud = !(flags & XFS_MFSI_QUIET);
282 const struct xfs_buf_ops *buf_ops;
284 ASSERT(mp->m_sb_bp == NULL);
285 ASSERT(mp->m_ddev_targp != NULL);
288 * For the initial read, we must guess at the sector
289 * size based on the block device. It's enough to
290 * get the sb_sectsize out of the superblock and
291 * then reread with the proper length.
292 * We don't verify it yet, because it may not be complete.
294 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
298 * Allocate a (locked) buffer to hold the superblock.
299 * This will be kept around at all times to optimize
300 * access to the superblock.
303 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
304 BTOBB(sector_size), 0, &bp, buf_ops);
307 xfs_warn(mp, "SB validate failed with error %d.", error);
308 /* bad CRC means corrupted metadata */
309 if (error == -EFSBADCRC)
310 error = -EFSCORRUPTED;
315 * Initialize the mount structure from the superblock.
317 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
320 * If we haven't validated the superblock, do so now before we try
321 * to check the sector size and reread the superblock appropriately.
323 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
325 xfs_warn(mp, "Invalid superblock magic number");
331 * We must be able to do sector-sized and sector-aligned IO.
333 if (sector_size > sbp->sb_sectsize) {
335 xfs_warn(mp, "device supports %u byte sectors (not %u)",
336 sector_size, sbp->sb_sectsize);
341 if (buf_ops == NULL) {
343 * Re-read the superblock so the buffer is correctly sized,
344 * and properly verified.
347 sector_size = sbp->sb_sectsize;
348 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
352 /* Initialize per-cpu counters */
353 xfs_icsb_reinit_counters(mp);
355 /* no need to be quiet anymore, so reset the buf ops */
356 bp->b_ops = &xfs_sb_buf_ops;
368 * Update alignment values based on mount options and sb values
371 xfs_update_alignment(xfs_mount_t *mp)
373 xfs_sb_t *sbp = &(mp->m_sb);
377 * If stripe unit and stripe width are not multiples
378 * of the fs blocksize turn off alignment.
380 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
381 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
383 "alignment check failed: sunit/swidth vs. blocksize(%d)",
388 * Convert the stripe unit and width to FSBs.
390 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
391 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
393 "alignment check failed: sunit/swidth vs. agsize(%d)",
396 } else if (mp->m_dalign) {
397 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
400 "alignment check failed: sunit(%d) less than bsize(%d)",
401 mp->m_dalign, sbp->sb_blocksize);
407 * Update superblock with new values
410 if (xfs_sb_version_hasdalign(sbp)) {
411 if (sbp->sb_unit != mp->m_dalign) {
412 sbp->sb_unit = mp->m_dalign;
413 mp->m_update_flags |= XFS_SB_UNIT;
415 if (sbp->sb_width != mp->m_swidth) {
416 sbp->sb_width = mp->m_swidth;
417 mp->m_update_flags |= XFS_SB_WIDTH;
421 "cannot change alignment: superblock does not support data alignment");
424 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
425 xfs_sb_version_hasdalign(&mp->m_sb)) {
426 mp->m_dalign = sbp->sb_unit;
427 mp->m_swidth = sbp->sb_width;
434 * Set the maximum inode count for this filesystem
437 xfs_set_maxicount(xfs_mount_t *mp)
439 xfs_sb_t *sbp = &(mp->m_sb);
442 if (sbp->sb_imax_pct) {
444 * Make sure the maximum inode count is a multiple
445 * of the units we allocate inodes in.
447 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
449 do_div(icount, mp->m_ialloc_blks);
450 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
458 * Set the default minimum read and write sizes unless
459 * already specified in a mount option.
460 * We use smaller I/O sizes when the file system
461 * is being used for NFS service (wsync mount option).
464 xfs_set_rw_sizes(xfs_mount_t *mp)
466 xfs_sb_t *sbp = &(mp->m_sb);
467 int readio_log, writeio_log;
469 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
470 if (mp->m_flags & XFS_MOUNT_WSYNC) {
471 readio_log = XFS_WSYNC_READIO_LOG;
472 writeio_log = XFS_WSYNC_WRITEIO_LOG;
474 readio_log = XFS_READIO_LOG_LARGE;
475 writeio_log = XFS_WRITEIO_LOG_LARGE;
478 readio_log = mp->m_readio_log;
479 writeio_log = mp->m_writeio_log;
482 if (sbp->sb_blocklog > readio_log) {
483 mp->m_readio_log = sbp->sb_blocklog;
485 mp->m_readio_log = readio_log;
487 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
488 if (sbp->sb_blocklog > writeio_log) {
489 mp->m_writeio_log = sbp->sb_blocklog;
491 mp->m_writeio_log = writeio_log;
493 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
497 * precalculate the low space thresholds for dynamic speculative preallocation.
500 xfs_set_low_space_thresholds(
501 struct xfs_mount *mp)
505 for (i = 0; i < XFS_LOWSP_MAX; i++) {
506 __uint64_t space = mp->m_sb.sb_dblocks;
509 mp->m_low_space[i] = space * (i + 1);
515 * Set whether we're using inode alignment.
518 xfs_set_inoalignment(xfs_mount_t *mp)
520 if (xfs_sb_version_hasalign(&mp->m_sb) &&
521 mp->m_sb.sb_inoalignmt >=
522 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
523 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
525 mp->m_inoalign_mask = 0;
527 * If we are using stripe alignment, check whether
528 * the stripe unit is a multiple of the inode alignment
530 if (mp->m_dalign && mp->m_inoalign_mask &&
531 !(mp->m_dalign & mp->m_inoalign_mask))
532 mp->m_sinoalign = mp->m_dalign;
538 * Check that the data (and log if separate) is an ok size.
542 struct xfs_mount *mp)
548 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
549 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
550 xfs_warn(mp, "filesystem size mismatch detected");
553 error = xfs_buf_read_uncached(mp->m_ddev_targp,
554 d - XFS_FSS_TO_BB(mp, 1),
555 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
557 xfs_warn(mp, "last sector read failed");
562 if (mp->m_logdev_targp == mp->m_ddev_targp)
565 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
566 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
567 xfs_warn(mp, "log size mismatch detected");
570 error = xfs_buf_read_uncached(mp->m_logdev_targp,
571 d - XFS_FSB_TO_BB(mp, 1),
572 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
574 xfs_warn(mp, "log device read failed");
582 * Clear the quotaflags in memory and in the superblock.
585 xfs_mount_reset_sbqflags(
586 struct xfs_mount *mp)
589 struct xfs_trans *tp;
594 * It is OK to look at sb_qflags here in mount path,
597 if (mp->m_sb.sb_qflags == 0)
599 spin_lock(&mp->m_sb_lock);
600 mp->m_sb.sb_qflags = 0;
601 spin_unlock(&mp->m_sb_lock);
604 * If the fs is readonly, let the incore superblock run
605 * with quotas off but don't flush the update out to disk
607 if (mp->m_flags & XFS_MOUNT_RDONLY)
610 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
611 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
613 xfs_trans_cancel(tp, 0);
614 xfs_alert(mp, "%s: Superblock update failed!", __func__);
618 xfs_mod_sb(tp, XFS_SB_QFLAGS);
619 return xfs_trans_commit(tp, 0);
623 xfs_default_resblks(xfs_mount_t *mp)
628 * We default to 5% or 8192 fsbs of space reserved, whichever is
629 * smaller. This is intended to cover concurrent allocation
630 * transactions when we initially hit enospc. These each require a 4
631 * block reservation. Hence by default we cover roughly 2000 concurrent
632 * allocation reservations.
634 resblks = mp->m_sb.sb_dblocks;
636 resblks = min_t(__uint64_t, resblks, 8192);
641 * This function does the following on an initial mount of a file system:
642 * - reads the superblock from disk and init the mount struct
643 * - if we're a 32-bit kernel, do a size check on the superblock
644 * so we don't mount terabyte filesystems
645 * - init mount struct realtime fields
646 * - allocate inode hash table for fs
647 * - init directory manager
648 * - perform recovery and init the log manager
654 xfs_sb_t *sbp = &(mp->m_sb);
661 xfs_sb_mount_common(mp, sbp);
664 * Check for a mismatched features2 values. Older kernels
665 * read & wrote into the wrong sb offset for sb_features2
666 * on some platforms due to xfs_sb_t not being 64bit size aligned
667 * when sb_features2 was added, which made older superblock
668 * reading/writing routines swap it as a 64-bit value.
670 * For backwards compatibility, we make both slots equal.
672 * If we detect a mismatched field, we OR the set bits into the
673 * existing features2 field in case it has already been modified; we
674 * don't want to lose any features. We then update the bad location
675 * with the ORed value so that older kernels will see any features2
676 * flags, and mark the two fields as needing updates once the
677 * transaction subsystem is online.
679 if (xfs_sb_has_mismatched_features2(sbp)) {
680 xfs_warn(mp, "correcting sb_features alignment problem");
681 sbp->sb_features2 |= sbp->sb_bad_features2;
682 sbp->sb_bad_features2 = sbp->sb_features2;
683 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
686 * Re-check for ATTR2 in case it was found in bad_features2
689 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
690 !(mp->m_flags & XFS_MOUNT_NOATTR2))
691 mp->m_flags |= XFS_MOUNT_ATTR2;
694 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
695 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
696 xfs_sb_version_removeattr2(&mp->m_sb);
697 mp->m_update_flags |= XFS_SB_FEATURES2;
699 /* update sb_versionnum for the clearing of the morebits */
700 if (!sbp->sb_features2)
701 mp->m_update_flags |= XFS_SB_VERSIONNUM;
704 /* always use v2 inodes by default now */
705 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
706 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
707 mp->m_update_flags |= XFS_SB_VERSIONNUM;
711 * Check if sb_agblocks is aligned at stripe boundary
712 * If sb_agblocks is NOT aligned turn off m_dalign since
713 * allocator alignment is within an ag, therefore ag has
714 * to be aligned at stripe boundary.
716 error = xfs_update_alignment(mp);
720 xfs_alloc_compute_maxlevels(mp);
721 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
722 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
723 xfs_ialloc_compute_maxlevels(mp);
725 xfs_set_maxicount(mp);
727 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
731 error = xfs_uuid_mount(mp);
733 goto out_remove_sysfs;
736 * Set the minimum read and write sizes
738 xfs_set_rw_sizes(mp);
740 /* set the low space thresholds for dynamic preallocation */
741 xfs_set_low_space_thresholds(mp);
744 * Set the inode cluster size.
745 * This may still be overridden by the file system
746 * block size if it is larger than the chosen cluster size.
748 * For v5 filesystems, scale the cluster size with the inode size to
749 * keep a constant ratio of inode per cluster buffer, but only if mkfs
750 * has set the inode alignment value appropriately for larger cluster
753 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
754 if (xfs_sb_version_hascrc(&mp->m_sb)) {
755 int new_size = mp->m_inode_cluster_size;
757 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
758 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
759 mp->m_inode_cluster_size = new_size;
763 * Set inode alignment fields
765 xfs_set_inoalignment(mp);
768 * Check that the data (and log if separate) is an ok size.
770 error = xfs_check_sizes(mp);
772 goto out_remove_uuid;
775 * Initialize realtime fields in the mount structure
777 error = xfs_rtmount_init(mp);
779 xfs_warn(mp, "RT mount failed");
780 goto out_remove_uuid;
784 * Copies the low order bits of the timestamp and the randomly
785 * set "sequence" number out of a UUID.
787 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
789 mp->m_dmevmask = 0; /* not persistent; set after each mount */
791 error = xfs_da_mount(mp);
793 xfs_warn(mp, "Failed dir/attr init: %d", error);
794 goto out_remove_uuid;
798 * Initialize the precomputed transaction reservations values.
803 * Allocate and initialize the per-ag data.
805 spin_lock_init(&mp->m_perag_lock);
806 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
807 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
809 xfs_warn(mp, "Failed per-ag init: %d", error);
813 if (!sbp->sb_logblocks) {
814 xfs_warn(mp, "no log defined");
815 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
816 error = -EFSCORRUPTED;
821 * log's mount-time initialization. Perform 1st part recovery if needed
823 error = xfs_log_mount(mp, mp->m_logdev_targp,
824 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
825 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
827 xfs_warn(mp, "log mount failed");
832 * Now the log is mounted, we know if it was an unclean shutdown or
833 * not. If it was, with the first phase of recovery has completed, we
834 * have consistent AG blocks on disk. We have not recovered EFIs yet,
835 * but they are recovered transactionally in the second recovery phase
838 * Hence we can safely re-initialise incore superblock counters from
839 * the per-ag data. These may not be correct if the filesystem was not
840 * cleanly unmounted, so we need to wait for recovery to finish before
843 * If the filesystem was cleanly unmounted, then we can trust the
844 * values in the superblock to be correct and we don't need to do
847 * If we are currently making the filesystem, the initialisation will
848 * fail as the perag data is in an undefined state.
850 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
851 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
852 !mp->m_sb.sb_inprogress) {
853 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
855 goto out_log_dealloc;
859 * Get and sanity-check the root inode.
860 * Save the pointer to it in the mount structure.
862 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
864 xfs_warn(mp, "failed to read root inode");
865 goto out_log_dealloc;
870 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
871 xfs_warn(mp, "corrupted root inode %llu: not a directory",
872 (unsigned long long)rip->i_ino);
873 xfs_iunlock(rip, XFS_ILOCK_EXCL);
874 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
876 error = -EFSCORRUPTED;
879 mp->m_rootip = rip; /* save it */
881 xfs_iunlock(rip, XFS_ILOCK_EXCL);
884 * Initialize realtime inode pointers in the mount structure
886 error = xfs_rtmount_inodes(mp);
889 * Free up the root inode.
891 xfs_warn(mp, "failed to read RT inodes");
896 * If this is a read-only mount defer the superblock updates until
897 * the next remount into writeable mode. Otherwise we would never
898 * perform the update e.g. for the root filesystem.
900 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
901 error = xfs_mount_log_sb(mp, mp->m_update_flags);
903 xfs_warn(mp, "failed to write sb changes");
909 * Initialise the XFS quota management subsystem for this mount
911 if (XFS_IS_QUOTA_RUNNING(mp)) {
912 error = xfs_qm_newmount(mp, "amount, "aflags);
916 ASSERT(!XFS_IS_QUOTA_ON(mp));
919 * If a file system had quotas running earlier, but decided to
920 * mount without -o uquota/pquota/gquota options, revoke the
921 * quotachecked license.
923 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
924 xfs_notice(mp, "resetting quota flags");
925 error = xfs_mount_reset_sbqflags(mp);
932 * Finish recovering the file system. This part needed to be
933 * delayed until after the root and real-time bitmap inodes
934 * were consistently read in.
936 error = xfs_log_mount_finish(mp);
938 xfs_warn(mp, "log mount finish failed");
943 * Complete the quota initialisation, post-log-replay component.
946 ASSERT(mp->m_qflags == 0);
947 mp->m_qflags = quotaflags;
949 xfs_qm_mount_quotas(mp);
953 * Now we are mounted, reserve a small amount of unused space for
954 * privileged transactions. This is needed so that transaction
955 * space required for critical operations can dip into this pool
956 * when at ENOSPC. This is needed for operations like create with
957 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
958 * are not allowed to use this reserved space.
960 * This may drive us straight to ENOSPC on mount, but that implies
961 * we were already there on the last unmount. Warn if this occurs.
963 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
964 resblks = xfs_default_resblks(mp);
965 error = xfs_reserve_blocks(mp, &resblks, NULL);
968 "Unable to allocate reserve blocks. Continuing without reserve pool.");
974 xfs_rtunmount_inodes(mp);
980 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
981 xfs_wait_buftarg(mp->m_logdev_targp);
982 xfs_wait_buftarg(mp->m_ddev_targp);
988 xfs_uuid_unmount(mp);
990 xfs_sysfs_del(&mp->m_kobj);
996 * This flushes out the inodes,dquots and the superblock, unmounts the
997 * log and makes sure that incore structures are freed.
1001 struct xfs_mount *mp)
1006 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1008 xfs_qm_unmount_quotas(mp);
1009 xfs_rtunmount_inodes(mp);
1010 IRELE(mp->m_rootip);
1013 * We can potentially deadlock here if we have an inode cluster
1014 * that has been freed has its buffer still pinned in memory because
1015 * the transaction is still sitting in a iclog. The stale inodes
1016 * on that buffer will have their flush locks held until the
1017 * transaction hits the disk and the callbacks run. the inode
1018 * flush takes the flush lock unconditionally and with nothing to
1019 * push out the iclog we will never get that unlocked. hence we
1020 * need to force the log first.
1022 xfs_log_force(mp, XFS_LOG_SYNC);
1025 * Flush all pending changes from the AIL.
1027 xfs_ail_push_all_sync(mp->m_ail);
1030 * And reclaim all inodes. At this point there should be no dirty
1031 * inodes and none should be pinned or locked, but use synchronous
1032 * reclaim just to be sure. We can stop background inode reclaim
1033 * here as well if it is still running.
1035 cancel_delayed_work_sync(&mp->m_reclaim_work);
1036 xfs_reclaim_inodes(mp, SYNC_WAIT);
1041 * Unreserve any blocks we have so that when we unmount we don't account
1042 * the reserved free space as used. This is really only necessary for
1043 * lazy superblock counting because it trusts the incore superblock
1044 * counters to be absolutely correct on clean unmount.
1046 * We don't bother correcting this elsewhere for lazy superblock
1047 * counting because on mount of an unclean filesystem we reconstruct the
1048 * correct counter value and this is irrelevant.
1050 * For non-lazy counter filesystems, this doesn't matter at all because
1051 * we only every apply deltas to the superblock and hence the incore
1052 * value does not matter....
1055 error = xfs_reserve_blocks(mp, &resblks, NULL);
1057 xfs_warn(mp, "Unable to free reserved block pool. "
1058 "Freespace may not be correct on next mount.");
1060 error = xfs_log_sbcount(mp);
1062 xfs_warn(mp, "Unable to update superblock counters. "
1063 "Freespace may not be correct on next mount.");
1065 xfs_log_unmount(mp);
1067 xfs_uuid_unmount(mp);
1070 xfs_errortag_clearall(mp, 0);
1074 xfs_sysfs_del(&mp->m_kobj);
1078 xfs_fs_writable(xfs_mount_t *mp)
1080 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1081 (mp->m_flags & XFS_MOUNT_RDONLY));
1087 * Sync the superblock counters to disk.
1089 * Note this code can be called during the process of freezing, so
1090 * we may need to use the transaction allocator which does not
1091 * block when the transaction subsystem is in its frozen state.
1094 xfs_log_sbcount(xfs_mount_t *mp)
1099 if (!xfs_fs_writable(mp))
1102 xfs_icsb_sync_counters(mp, 0);
1105 * we don't need to do this if we are updating the superblock
1106 * counters on every modification.
1108 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1111 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1112 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1114 xfs_trans_cancel(tp, 0);
1118 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1119 xfs_trans_set_sync(tp);
1120 error = xfs_trans_commit(tp, 0);
1125 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1126 * a delta to a specified field in the in-core superblock. Simply
1127 * switch on the field indicated and apply the delta to that field.
1128 * Fields are not allowed to dip below zero, so if the delta would
1129 * do this do not apply it and return EINVAL.
1131 * The m_sb_lock must be held when this routine is called.
1134 xfs_mod_incore_sb_unlocked(
1136 xfs_sb_field_t field,
1140 int scounter; /* short counter for 32 bit fields */
1141 long long lcounter; /* long counter for 64 bit fields */
1142 long long res_used, rem;
1145 * With the in-core superblock spin lock held, switch
1146 * on the indicated field. Apply the delta to the
1147 * proper field. If the fields value would dip below
1148 * 0, then do not apply the delta and return EINVAL.
1151 case XFS_SBS_ICOUNT:
1152 lcounter = (long long)mp->m_sb.sb_icount;
1158 mp->m_sb.sb_icount = lcounter;
1161 lcounter = (long long)mp->m_sb.sb_ifree;
1167 mp->m_sb.sb_ifree = lcounter;
1169 case XFS_SBS_FDBLOCKS:
1170 lcounter = (long long)
1171 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1172 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1174 if (delta > 0) { /* Putting blocks back */
1175 if (res_used > delta) {
1176 mp->m_resblks_avail += delta;
1178 rem = delta - res_used;
1179 mp->m_resblks_avail = mp->m_resblks;
1182 } else { /* Taking blocks away */
1184 if (lcounter >= 0) {
1185 mp->m_sb.sb_fdblocks = lcounter +
1186 XFS_ALLOC_SET_ASIDE(mp);
1191 * We are out of blocks, use any available reserved
1192 * blocks if were allowed to.
1197 lcounter = (long long)mp->m_resblks_avail + delta;
1198 if (lcounter >= 0) {
1199 mp->m_resblks_avail = lcounter;
1202 printk_once(KERN_WARNING
1203 "Filesystem \"%s\": reserve blocks depleted! "
1204 "Consider increasing reserve pool size.",
1209 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1211 case XFS_SBS_FREXTENTS:
1212 lcounter = (long long)mp->m_sb.sb_frextents;
1217 mp->m_sb.sb_frextents = lcounter;
1219 case XFS_SBS_DBLOCKS:
1220 lcounter = (long long)mp->m_sb.sb_dblocks;
1226 mp->m_sb.sb_dblocks = lcounter;
1228 case XFS_SBS_AGCOUNT:
1229 scounter = mp->m_sb.sb_agcount;
1235 mp->m_sb.sb_agcount = scounter;
1237 case XFS_SBS_IMAX_PCT:
1238 scounter = mp->m_sb.sb_imax_pct;
1244 mp->m_sb.sb_imax_pct = scounter;
1246 case XFS_SBS_REXTSIZE:
1247 scounter = mp->m_sb.sb_rextsize;
1253 mp->m_sb.sb_rextsize = scounter;
1255 case XFS_SBS_RBMBLOCKS:
1256 scounter = mp->m_sb.sb_rbmblocks;
1262 mp->m_sb.sb_rbmblocks = scounter;
1264 case XFS_SBS_RBLOCKS:
1265 lcounter = (long long)mp->m_sb.sb_rblocks;
1271 mp->m_sb.sb_rblocks = lcounter;
1273 case XFS_SBS_REXTENTS:
1274 lcounter = (long long)mp->m_sb.sb_rextents;
1280 mp->m_sb.sb_rextents = lcounter;
1282 case XFS_SBS_REXTSLOG:
1283 scounter = mp->m_sb.sb_rextslog;
1289 mp->m_sb.sb_rextslog = scounter;
1298 * xfs_mod_incore_sb() is used to change a field in the in-core
1299 * superblock structure by the specified delta. This modification
1300 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1301 * routine to do the work.
1305 struct xfs_mount *mp,
1306 xfs_sb_field_t field,
1312 #ifdef HAVE_PERCPU_SB
1313 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1315 spin_lock(&mp->m_sb_lock);
1316 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1317 spin_unlock(&mp->m_sb_lock);
1323 * Change more than one field in the in-core superblock structure at a time.
1325 * The fields and changes to those fields are specified in the array of
1326 * xfs_mod_sb structures passed in. Either all of the specified deltas
1327 * will be applied or none of them will. If any modified field dips below 0,
1328 * then all modifications will be backed out and EINVAL will be returned.
1330 * Note that this function may not be used for the superblock values that
1331 * are tracked with the in-memory per-cpu counters - a direct call to
1332 * xfs_icsb_modify_counters is required for these.
1335 xfs_mod_incore_sb_batch(
1336 struct xfs_mount *mp,
1345 * Loop through the array of mod structures and apply each individually.
1346 * If any fail, then back out all those which have already been applied.
1347 * Do all of this within the scope of the m_sb_lock so that all of the
1348 * changes will be atomic.
1350 spin_lock(&mp->m_sb_lock);
1351 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1352 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1353 msbp->msb_field > XFS_SBS_FDBLOCKS);
1355 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1356 msbp->msb_delta, rsvd);
1360 spin_unlock(&mp->m_sb_lock);
1364 while (--msbp >= msb) {
1365 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1366 -msbp->msb_delta, rsvd);
1369 spin_unlock(&mp->m_sb_lock);
1374 * xfs_getsb() is called to obtain the buffer for the superblock.
1375 * The buffer is returned locked and read in from disk.
1376 * The buffer should be released with a call to xfs_brelse().
1378 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1379 * the superblock buffer if it can be locked without sleeping.
1380 * If it can't then we'll return NULL.
1384 struct xfs_mount *mp,
1387 struct xfs_buf *bp = mp->m_sb_bp;
1389 if (!xfs_buf_trylock(bp)) {
1390 if (flags & XBF_TRYLOCK)
1396 ASSERT(XFS_BUF_ISDONE(bp));
1401 * Used to free the superblock along various error paths.
1405 struct xfs_mount *mp)
1407 struct xfs_buf *bp = mp->m_sb_bp;
1415 * Used to log changes to the superblock unit and width fields which could
1416 * be altered by the mount options, as well as any potential sb_features2
1417 * fixup. Only the first superblock is updated.
1427 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1428 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1429 XFS_SB_VERSIONNUM));
1431 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1432 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1434 xfs_trans_cancel(tp, 0);
1437 xfs_mod_sb(tp, fields);
1438 error = xfs_trans_commit(tp, 0);
1443 * If the underlying (data/log/rt) device is readonly, there are some
1444 * operations that cannot proceed.
1447 xfs_dev_is_read_only(
1448 struct xfs_mount *mp,
1451 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1452 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1453 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1454 xfs_notice(mp, "%s required on read-only device.", message);
1455 xfs_notice(mp, "write access unavailable, cannot proceed.");
1461 #ifdef HAVE_PERCPU_SB
1463 * Per-cpu incore superblock counters
1465 * Simple concept, difficult implementation
1467 * Basically, replace the incore superblock counters with a distributed per cpu
1468 * counter for contended fields (e.g. free block count).
1470 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1471 * hence needs to be accurately read when we are running low on space. Hence
1472 * there is a method to enable and disable the per-cpu counters based on how
1473 * much "stuff" is available in them.
1475 * Basically, a counter is enabled if there is enough free resource to justify
1476 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1477 * ENOSPC), then we disable the counters to synchronise all callers and
1478 * re-distribute the available resources.
1480 * If, once we redistributed the available resources, we still get a failure,
1481 * we disable the per-cpu counter and go through the slow path.
1483 * The slow path is the current xfs_mod_incore_sb() function. This means that
1484 * when we disable a per-cpu counter, we need to drain its resources back to
1485 * the global superblock. We do this after disabling the counter to prevent
1486 * more threads from queueing up on the counter.
1488 * Essentially, this means that we still need a lock in the fast path to enable
1489 * synchronisation between the global counters and the per-cpu counters. This
1490 * is not a problem because the lock will be local to a CPU almost all the time
1491 * and have little contention except when we get to ENOSPC conditions.
1493 * Basically, this lock becomes a barrier that enables us to lock out the fast
1494 * path while we do things like enabling and disabling counters and
1495 * synchronising the counters.
1499 * 1. m_sb_lock before picking up per-cpu locks
1500 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1501 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1502 * 4. modifying per-cpu counters requires holding per-cpu lock
1503 * 5. modifying global counters requires holding m_sb_lock
1504 * 6. enabling or disabling a counter requires holding the m_sb_lock
1505 * and _none_ of the per-cpu locks.
1507 * Disabled counters are only ever re-enabled by a balance operation
1508 * that results in more free resources per CPU than a given threshold.
1509 * To ensure counters don't remain disabled, they are rebalanced when
1510 * the global resource goes above a higher threshold (i.e. some hysteresis
1511 * is present to prevent thrashing).
1514 #ifdef CONFIG_HOTPLUG_CPU
1516 * hot-plug CPU notifier support.
1518 * We need a notifier per filesystem as we need to be able to identify
1519 * the filesystem to balance the counters out. This is achieved by
1520 * having a notifier block embedded in the xfs_mount_t and doing pointer
1521 * magic to get the mount pointer from the notifier block address.
1524 xfs_icsb_cpu_notify(
1525 struct notifier_block *nfb,
1526 unsigned long action,
1529 xfs_icsb_cnts_t *cntp;
1532 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1533 cntp = (xfs_icsb_cnts_t *)
1534 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1536 case CPU_UP_PREPARE:
1537 case CPU_UP_PREPARE_FROZEN:
1538 /* Easy Case - initialize the area and locks, and
1539 * then rebalance when online does everything else for us. */
1540 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1543 case CPU_ONLINE_FROZEN:
1545 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1546 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1547 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1548 xfs_icsb_unlock(mp);
1551 case CPU_DEAD_FROZEN:
1552 /* Disable all the counters, then fold the dead cpu's
1553 * count into the total on the global superblock and
1554 * re-enable the counters. */
1556 spin_lock(&mp->m_sb_lock);
1557 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1558 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1559 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1561 mp->m_sb.sb_icount += cntp->icsb_icount;
1562 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1563 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1565 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1567 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1568 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1569 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1570 spin_unlock(&mp->m_sb_lock);
1571 xfs_icsb_unlock(mp);
1577 #endif /* CONFIG_HOTPLUG_CPU */
1580 xfs_icsb_init_counters(
1583 xfs_icsb_cnts_t *cntp;
1586 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1587 if (mp->m_sb_cnts == NULL)
1590 for_each_online_cpu(i) {
1591 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1592 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1595 mutex_init(&mp->m_icsb_mutex);
1598 * start with all counters disabled so that the
1599 * initial balance kicks us off correctly
1601 mp->m_icsb_counters = -1;
1603 #ifdef CONFIG_HOTPLUG_CPU
1604 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1605 mp->m_icsb_notifier.priority = 0;
1606 register_hotcpu_notifier(&mp->m_icsb_notifier);
1607 #endif /* CONFIG_HOTPLUG_CPU */
1613 xfs_icsb_reinit_counters(
1618 * start with all counters disabled so that the
1619 * initial balance kicks us off correctly
1621 mp->m_icsb_counters = -1;
1622 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1623 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1624 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1625 xfs_icsb_unlock(mp);
1629 xfs_icsb_destroy_counters(
1632 if (mp->m_sb_cnts) {
1633 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1634 free_percpu(mp->m_sb_cnts);
1636 mutex_destroy(&mp->m_icsb_mutex);
1641 xfs_icsb_cnts_t *icsbp)
1643 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1649 xfs_icsb_unlock_cntr(
1650 xfs_icsb_cnts_t *icsbp)
1652 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1657 xfs_icsb_lock_all_counters(
1660 xfs_icsb_cnts_t *cntp;
1663 for_each_online_cpu(i) {
1664 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1665 xfs_icsb_lock_cntr(cntp);
1670 xfs_icsb_unlock_all_counters(
1673 xfs_icsb_cnts_t *cntp;
1676 for_each_online_cpu(i) {
1677 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1678 xfs_icsb_unlock_cntr(cntp);
1685 xfs_icsb_cnts_t *cnt,
1688 xfs_icsb_cnts_t *cntp;
1691 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1693 if (!(flags & XFS_ICSB_LAZY_COUNT))
1694 xfs_icsb_lock_all_counters(mp);
1696 for_each_online_cpu(i) {
1697 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1698 cnt->icsb_icount += cntp->icsb_icount;
1699 cnt->icsb_ifree += cntp->icsb_ifree;
1700 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1703 if (!(flags & XFS_ICSB_LAZY_COUNT))
1704 xfs_icsb_unlock_all_counters(mp);
1708 xfs_icsb_counter_disabled(
1710 xfs_sb_field_t field)
1712 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1713 return test_bit(field, &mp->m_icsb_counters);
1717 xfs_icsb_disable_counter(
1719 xfs_sb_field_t field)
1721 xfs_icsb_cnts_t cnt;
1723 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1726 * If we are already disabled, then there is nothing to do
1727 * here. We check before locking all the counters to avoid
1728 * the expensive lock operation when being called in the
1729 * slow path and the counter is already disabled. This is
1730 * safe because the only time we set or clear this state is under
1733 if (xfs_icsb_counter_disabled(mp, field))
1736 xfs_icsb_lock_all_counters(mp);
1737 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1738 /* drain back to superblock */
1740 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1742 case XFS_SBS_ICOUNT:
1743 mp->m_sb.sb_icount = cnt.icsb_icount;
1746 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1748 case XFS_SBS_FDBLOCKS:
1749 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1756 xfs_icsb_unlock_all_counters(mp);
1760 xfs_icsb_enable_counter(
1762 xfs_sb_field_t field,
1766 xfs_icsb_cnts_t *cntp;
1769 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1771 xfs_icsb_lock_all_counters(mp);
1772 for_each_online_cpu(i) {
1773 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1775 case XFS_SBS_ICOUNT:
1776 cntp->icsb_icount = count + resid;
1779 cntp->icsb_ifree = count + resid;
1781 case XFS_SBS_FDBLOCKS:
1782 cntp->icsb_fdblocks = count + resid;
1790 clear_bit(field, &mp->m_icsb_counters);
1791 xfs_icsb_unlock_all_counters(mp);
1795 xfs_icsb_sync_counters_locked(
1799 xfs_icsb_cnts_t cnt;
1801 xfs_icsb_count(mp, &cnt, flags);
1803 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1804 mp->m_sb.sb_icount = cnt.icsb_icount;
1805 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1806 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1807 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1808 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1812 * Accurate update of per-cpu counters to incore superblock
1815 xfs_icsb_sync_counters(
1819 spin_lock(&mp->m_sb_lock);
1820 xfs_icsb_sync_counters_locked(mp, flags);
1821 spin_unlock(&mp->m_sb_lock);
1825 * Balance and enable/disable counters as necessary.
1827 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1828 * chosen to be the same number as single on disk allocation chunk per CPU, and
1829 * free blocks is something far enough zero that we aren't going thrash when we
1830 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1831 * prevent looping endlessly when xfs_alloc_space asks for more than will
1832 * be distributed to a single CPU but each CPU has enough blocks to be
1835 * Note that we can be called when counters are already disabled.
1836 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1837 * prevent locking every per-cpu counter needlessly.
1840 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1841 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1842 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1844 xfs_icsb_balance_counter_locked(
1846 xfs_sb_field_t field,
1849 uint64_t count, resid;
1850 int weight = num_online_cpus();
1851 uint64_t min = (uint64_t)min_per_cpu;
1853 /* disable counter and sync counter */
1854 xfs_icsb_disable_counter(mp, field);
1856 /* update counters - first CPU gets residual*/
1858 case XFS_SBS_ICOUNT:
1859 count = mp->m_sb.sb_icount;
1860 resid = do_div(count, weight);
1861 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1865 count = mp->m_sb.sb_ifree;
1866 resid = do_div(count, weight);
1867 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1870 case XFS_SBS_FDBLOCKS:
1871 count = mp->m_sb.sb_fdblocks;
1872 resid = do_div(count, weight);
1873 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
1878 count = resid = 0; /* quiet, gcc */
1882 xfs_icsb_enable_counter(mp, field, count, resid);
1886 xfs_icsb_balance_counter(
1888 xfs_sb_field_t fields,
1891 spin_lock(&mp->m_sb_lock);
1892 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1893 spin_unlock(&mp->m_sb_lock);
1897 xfs_icsb_modify_counters(
1899 xfs_sb_field_t field,
1903 xfs_icsb_cnts_t *icsbp;
1904 long long lcounter; /* long counter for 64 bit fields */
1910 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1913 * if the counter is disabled, go to slow path
1915 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1917 xfs_icsb_lock_cntr(icsbp);
1918 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1919 xfs_icsb_unlock_cntr(icsbp);
1924 case XFS_SBS_ICOUNT:
1925 lcounter = icsbp->icsb_icount;
1927 if (unlikely(lcounter < 0))
1928 goto balance_counter;
1929 icsbp->icsb_icount = lcounter;
1933 lcounter = icsbp->icsb_ifree;
1935 if (unlikely(lcounter < 0))
1936 goto balance_counter;
1937 icsbp->icsb_ifree = lcounter;
1940 case XFS_SBS_FDBLOCKS:
1941 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1943 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1945 if (unlikely(lcounter < 0))
1946 goto balance_counter;
1947 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1953 xfs_icsb_unlock_cntr(icsbp);
1961 * serialise with a mutex so we don't burn lots of cpu on
1962 * the superblock lock. We still need to hold the superblock
1963 * lock, however, when we modify the global structures.
1968 * Now running atomically.
1970 * If the counter is enabled, someone has beaten us to rebalancing.
1971 * Drop the lock and try again in the fast path....
1973 if (!(xfs_icsb_counter_disabled(mp, field))) {
1974 xfs_icsb_unlock(mp);
1979 * The counter is currently disabled. Because we are
1980 * running atomically here, we know a rebalance cannot
1981 * be in progress. Hence we can go straight to operating
1982 * on the global superblock. We do not call xfs_mod_incore_sb()
1983 * here even though we need to get the m_sb_lock. Doing so
1984 * will cause us to re-enter this function and deadlock.
1985 * Hence we get the m_sb_lock ourselves and then call
1986 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1987 * directly on the global counters.
1989 spin_lock(&mp->m_sb_lock);
1990 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1991 spin_unlock(&mp->m_sb_lock);
1994 * Now that we've modified the global superblock, we
1995 * may be able to re-enable the distributed counters
1996 * (e.g. lots of space just got freed). After that
2000 xfs_icsb_balance_counter(mp, field, 0);
2001 xfs_icsb_unlock(mp);
2005 xfs_icsb_unlock_cntr(icsbp);
2009 * We may have multiple threads here if multiple per-cpu
2010 * counters run dry at the same time. This will mean we can
2011 * do more balances than strictly necessary but it is not
2012 * the common slowpath case.
2017 * running atomically.
2019 * This will leave the counter in the correct state for future
2020 * accesses. After the rebalance, we simply try again and our retry
2021 * will either succeed through the fast path or slow path without
2022 * another balance operation being required.
2024 xfs_icsb_balance_counter(mp, field, delta);
2025 xfs_icsb_unlock(mp);