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"
48 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
50 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
52 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
55 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
56 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
59 static DEFINE_MUTEX(xfs_uuid_table_mutex);
60 static int xfs_uuid_table_size;
61 static uuid_t *xfs_uuid_table;
64 * See if the UUID is unique among mounted XFS filesystems.
65 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
71 uuid_t *uuid = &mp->m_sb.sb_uuid;
74 if (mp->m_flags & XFS_MOUNT_NOUUID)
77 if (uuid_is_nil(uuid)) {
78 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
82 mutex_lock(&xfs_uuid_table_mutex);
83 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
84 if (uuid_is_nil(&xfs_uuid_table[i])) {
88 if (uuid_equal(uuid, &xfs_uuid_table[i]))
93 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
94 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
95 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
97 hole = xfs_uuid_table_size++;
99 xfs_uuid_table[hole] = *uuid;
100 mutex_unlock(&xfs_uuid_table_mutex);
105 mutex_unlock(&xfs_uuid_table_mutex);
106 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
112 struct xfs_mount *mp)
114 uuid_t *uuid = &mp->m_sb.sb_uuid;
117 if (mp->m_flags & XFS_MOUNT_NOUUID)
120 mutex_lock(&xfs_uuid_table_mutex);
121 for (i = 0; i < xfs_uuid_table_size; i++) {
122 if (uuid_is_nil(&xfs_uuid_table[i]))
124 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
126 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
129 ASSERT(i < xfs_uuid_table_size);
130 mutex_unlock(&xfs_uuid_table_mutex);
136 struct rcu_head *head)
138 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
140 ASSERT(atomic_read(&pag->pag_ref) == 0);
145 * Free up the per-ag resources associated with the mount structure.
152 struct xfs_perag *pag;
154 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
155 spin_lock(&mp->m_perag_lock);
156 pag = radix_tree_delete(&mp->m_perag_tree, agno);
157 spin_unlock(&mp->m_perag_lock);
159 ASSERT(atomic_read(&pag->pag_ref) == 0);
160 call_rcu(&pag->rcu_head, __xfs_free_perag);
165 * Check size of device based on the (data/realtime) block count.
166 * Note: this check is used by the growfs code as well as mount.
169 xfs_sb_validate_fsb_count(
173 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
174 ASSERT(sbp->sb_blocklog >= BBSHIFT);
176 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
177 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
179 #else /* Limited by UINT_MAX of sectors */
180 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
187 xfs_initialize_perag(
189 xfs_agnumber_t agcount,
190 xfs_agnumber_t *maxagi)
192 xfs_agnumber_t index;
193 xfs_agnumber_t first_initialised = 0;
197 xfs_sb_t *sbp = &mp->m_sb;
201 * Walk the current per-ag tree so we don't try to initialise AGs
202 * that already exist (growfs case). Allocate and insert all the
203 * AGs we don't find ready for initialisation.
205 for (index = 0; index < agcount; index++) {
206 pag = xfs_perag_get(mp, index);
211 if (!first_initialised)
212 first_initialised = index;
214 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
217 pag->pag_agno = index;
219 spin_lock_init(&pag->pag_ici_lock);
220 mutex_init(&pag->pag_ici_reclaim_lock);
221 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
222 spin_lock_init(&pag->pag_buf_lock);
223 pag->pag_buf_tree = RB_ROOT;
225 if (radix_tree_preload(GFP_NOFS))
228 spin_lock(&mp->m_perag_lock);
229 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
231 spin_unlock(&mp->m_perag_lock);
232 radix_tree_preload_end();
236 spin_unlock(&mp->m_perag_lock);
237 radix_tree_preload_end();
241 * If we mount with the inode64 option, or no inode overflows
242 * the legacy 32-bit address space clear the inode32 option.
244 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
245 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
247 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
248 mp->m_flags |= XFS_MOUNT_32BITINODES;
250 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
252 if (mp->m_flags & XFS_MOUNT_32BITINODES)
253 index = xfs_set_inode32(mp);
255 index = xfs_set_inode64(mp);
263 for (; index > first_initialised; index--) {
264 pag = radix_tree_delete(&mp->m_perag_tree, index);
273 * Does the initial read of the superblock.
277 struct xfs_mount *mp,
280 unsigned int sector_size;
282 struct xfs_sb *sbp = &mp->m_sb;
284 int loud = !(flags & XFS_MFSI_QUIET);
285 const struct xfs_buf_ops *buf_ops;
287 ASSERT(mp->m_sb_bp == NULL);
288 ASSERT(mp->m_ddev_targp != NULL);
291 * For the initial read, we must guess at the sector
292 * size based on the block device. It's enough to
293 * get the sb_sectsize out of the superblock and
294 * then reread with the proper length.
295 * We don't verify it yet, because it may not be complete.
297 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
301 * Allocate a (locked) buffer to hold the superblock.
302 * This will be kept around at all times to optimize
303 * access to the superblock.
306 bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
307 BTOBB(sector_size), 0, buf_ops);
310 xfs_warn(mp, "SB buffer read failed");
316 xfs_warn(mp, "SB validate failed with error %d.", error);
317 /* bad CRC means corrupted metadata */
318 if (error == -EFSBADCRC)
319 error = -EFSCORRUPTED;
324 * Initialize the mount structure from the superblock.
326 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
329 * If we haven't validated the superblock, do so now before we try
330 * to check the sector size and reread the superblock appropriately.
332 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
334 xfs_warn(mp, "Invalid superblock magic number");
340 * We must be able to do sector-sized and sector-aligned IO.
342 if (sector_size > sbp->sb_sectsize) {
344 xfs_warn(mp, "device supports %u byte sectors (not %u)",
345 sector_size, sbp->sb_sectsize);
350 if (buf_ops == NULL) {
352 * Re-read the superblock so the buffer is correctly sized,
353 * and properly verified.
356 sector_size = sbp->sb_sectsize;
357 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
361 /* Initialize per-cpu counters */
362 xfs_icsb_reinit_counters(mp);
364 /* no need to be quiet anymore, so reset the buf ops */
365 bp->b_ops = &xfs_sb_buf_ops;
377 * Update alignment values based on mount options and sb values
380 xfs_update_alignment(xfs_mount_t *mp)
382 xfs_sb_t *sbp = &(mp->m_sb);
386 * If stripe unit and stripe width are not multiples
387 * of the fs blocksize turn off alignment.
389 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
390 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
392 "alignment check failed: sunit/swidth vs. blocksize(%d)",
397 * Convert the stripe unit and width to FSBs.
399 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
400 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
402 "alignment check failed: sunit/swidth vs. agsize(%d)",
405 } else if (mp->m_dalign) {
406 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
409 "alignment check failed: sunit(%d) less than bsize(%d)",
410 mp->m_dalign, sbp->sb_blocksize);
416 * Update superblock with new values
419 if (xfs_sb_version_hasdalign(sbp)) {
420 if (sbp->sb_unit != mp->m_dalign) {
421 sbp->sb_unit = mp->m_dalign;
422 mp->m_update_flags |= XFS_SB_UNIT;
424 if (sbp->sb_width != mp->m_swidth) {
425 sbp->sb_width = mp->m_swidth;
426 mp->m_update_flags |= XFS_SB_WIDTH;
430 "cannot change alignment: superblock does not support data alignment");
433 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
434 xfs_sb_version_hasdalign(&mp->m_sb)) {
435 mp->m_dalign = sbp->sb_unit;
436 mp->m_swidth = sbp->sb_width;
443 * Set the maximum inode count for this filesystem
446 xfs_set_maxicount(xfs_mount_t *mp)
448 xfs_sb_t *sbp = &(mp->m_sb);
451 if (sbp->sb_imax_pct) {
453 * Make sure the maximum inode count is a multiple
454 * of the units we allocate inodes in.
456 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
458 do_div(icount, mp->m_ialloc_blks);
459 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
467 * Set the default minimum read and write sizes unless
468 * already specified in a mount option.
469 * We use smaller I/O sizes when the file system
470 * is being used for NFS service (wsync mount option).
473 xfs_set_rw_sizes(xfs_mount_t *mp)
475 xfs_sb_t *sbp = &(mp->m_sb);
476 int readio_log, writeio_log;
478 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
479 if (mp->m_flags & XFS_MOUNT_WSYNC) {
480 readio_log = XFS_WSYNC_READIO_LOG;
481 writeio_log = XFS_WSYNC_WRITEIO_LOG;
483 readio_log = XFS_READIO_LOG_LARGE;
484 writeio_log = XFS_WRITEIO_LOG_LARGE;
487 readio_log = mp->m_readio_log;
488 writeio_log = mp->m_writeio_log;
491 if (sbp->sb_blocklog > readio_log) {
492 mp->m_readio_log = sbp->sb_blocklog;
494 mp->m_readio_log = readio_log;
496 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
497 if (sbp->sb_blocklog > writeio_log) {
498 mp->m_writeio_log = sbp->sb_blocklog;
500 mp->m_writeio_log = writeio_log;
502 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
506 * precalculate the low space thresholds for dynamic speculative preallocation.
509 xfs_set_low_space_thresholds(
510 struct xfs_mount *mp)
514 for (i = 0; i < XFS_LOWSP_MAX; i++) {
515 __uint64_t space = mp->m_sb.sb_dblocks;
518 mp->m_low_space[i] = space * (i + 1);
524 * Set whether we're using inode alignment.
527 xfs_set_inoalignment(xfs_mount_t *mp)
529 if (xfs_sb_version_hasalign(&mp->m_sb) &&
530 mp->m_sb.sb_inoalignmt >=
531 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
532 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
534 mp->m_inoalign_mask = 0;
536 * If we are using stripe alignment, check whether
537 * the stripe unit is a multiple of the inode alignment
539 if (mp->m_dalign && mp->m_inoalign_mask &&
540 !(mp->m_dalign & mp->m_inoalign_mask))
541 mp->m_sinoalign = mp->m_dalign;
547 * Check that the data (and log if separate) is an ok size.
550 xfs_check_sizes(xfs_mount_t *mp)
555 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
556 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
557 xfs_warn(mp, "filesystem size mismatch detected");
560 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
561 d - XFS_FSS_TO_BB(mp, 1),
562 XFS_FSS_TO_BB(mp, 1), 0, NULL);
564 xfs_warn(mp, "last sector read failed");
569 if (mp->m_logdev_targp != mp->m_ddev_targp) {
570 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
571 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
572 xfs_warn(mp, "log size mismatch detected");
575 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
576 d - XFS_FSB_TO_BB(mp, 1),
577 XFS_FSB_TO_BB(mp, 1), 0, NULL);
579 xfs_warn(mp, "log device read failed");
588 * Clear the quotaflags in memory and in the superblock.
591 xfs_mount_reset_sbqflags(
592 struct xfs_mount *mp)
595 struct xfs_trans *tp;
600 * It is OK to look at sb_qflags here in mount path,
603 if (mp->m_sb.sb_qflags == 0)
605 spin_lock(&mp->m_sb_lock);
606 mp->m_sb.sb_qflags = 0;
607 spin_unlock(&mp->m_sb_lock);
610 * If the fs is readonly, let the incore superblock run
611 * with quotas off but don't flush the update out to disk
613 if (mp->m_flags & XFS_MOUNT_RDONLY)
616 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
617 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
619 xfs_trans_cancel(tp, 0);
620 xfs_alert(mp, "%s: Superblock update failed!", __func__);
624 xfs_mod_sb(tp, XFS_SB_QFLAGS);
625 return xfs_trans_commit(tp, 0);
629 xfs_default_resblks(xfs_mount_t *mp)
634 * We default to 5% or 8192 fsbs of space reserved, whichever is
635 * smaller. This is intended to cover concurrent allocation
636 * transactions when we initially hit enospc. These each require a 4
637 * block reservation. Hence by default we cover roughly 2000 concurrent
638 * allocation reservations.
640 resblks = mp->m_sb.sb_dblocks;
642 resblks = min_t(__uint64_t, resblks, 8192);
647 * This function does the following on an initial mount of a file system:
648 * - reads the superblock from disk and init the mount struct
649 * - if we're a 32-bit kernel, do a size check on the superblock
650 * so we don't mount terabyte filesystems
651 * - init mount struct realtime fields
652 * - allocate inode hash table for fs
653 * - init directory manager
654 * - perform recovery and init the log manager
660 xfs_sb_t *sbp = &(mp->m_sb);
667 xfs_sb_mount_common(mp, sbp);
670 * Check for a mismatched features2 values. Older kernels
671 * read & wrote into the wrong sb offset for sb_features2
672 * on some platforms due to xfs_sb_t not being 64bit size aligned
673 * when sb_features2 was added, which made older superblock
674 * reading/writing routines swap it as a 64-bit value.
676 * For backwards compatibility, we make both slots equal.
678 * If we detect a mismatched field, we OR the set bits into the
679 * existing features2 field in case it has already been modified; we
680 * don't want to lose any features. We then update the bad location
681 * with the ORed value so that older kernels will see any features2
682 * flags, and mark the two fields as needing updates once the
683 * transaction subsystem is online.
685 if (xfs_sb_has_mismatched_features2(sbp)) {
686 xfs_warn(mp, "correcting sb_features alignment problem");
687 sbp->sb_features2 |= sbp->sb_bad_features2;
688 sbp->sb_bad_features2 = sbp->sb_features2;
689 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
692 * Re-check for ATTR2 in case it was found in bad_features2
695 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
696 !(mp->m_flags & XFS_MOUNT_NOATTR2))
697 mp->m_flags |= XFS_MOUNT_ATTR2;
700 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
701 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
702 xfs_sb_version_removeattr2(&mp->m_sb);
703 mp->m_update_flags |= XFS_SB_FEATURES2;
705 /* update sb_versionnum for the clearing of the morebits */
706 if (!sbp->sb_features2)
707 mp->m_update_flags |= XFS_SB_VERSIONNUM;
710 /* always use v2 inodes by default now */
711 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
712 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
713 mp->m_update_flags |= XFS_SB_VERSIONNUM;
717 * Check if sb_agblocks is aligned at stripe boundary
718 * If sb_agblocks is NOT aligned turn off m_dalign since
719 * allocator alignment is within an ag, therefore ag has
720 * to be aligned at stripe boundary.
722 error = xfs_update_alignment(mp);
726 xfs_alloc_compute_maxlevels(mp);
727 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
728 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
729 xfs_ialloc_compute_maxlevels(mp);
731 xfs_set_maxicount(mp);
733 error = xfs_uuid_mount(mp);
738 * Set the minimum read and write sizes
740 xfs_set_rw_sizes(mp);
742 /* set the low space thresholds for dynamic preallocation */
743 xfs_set_low_space_thresholds(mp);
746 * Set the inode cluster size.
747 * This may still be overridden by the file system
748 * block size if it is larger than the chosen cluster size.
750 * For v5 filesystems, scale the cluster size with the inode size to
751 * keep a constant ratio of inode per cluster buffer, but only if mkfs
752 * has set the inode alignment value appropriately for larger cluster
755 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
756 if (xfs_sb_version_hascrc(&mp->m_sb)) {
757 int new_size = mp->m_inode_cluster_size;
759 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
760 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
761 mp->m_inode_cluster_size = new_size;
765 * Set inode alignment fields
767 xfs_set_inoalignment(mp);
770 * Check that the data (and log if separate) is an ok size.
772 error = xfs_check_sizes(mp);
774 goto out_remove_uuid;
777 * Initialize realtime fields in the mount structure
779 error = xfs_rtmount_init(mp);
781 xfs_warn(mp, "RT mount failed");
782 goto out_remove_uuid;
786 * Copies the low order bits of the timestamp and the randomly
787 * set "sequence" number out of a UUID.
789 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
791 mp->m_dmevmask = 0; /* not persistent; set after each mount */
793 error = xfs_da_mount(mp);
795 xfs_warn(mp, "Failed dir/attr init: %d", error);
796 goto out_remove_uuid;
800 * Initialize the precomputed transaction reservations values.
805 * Allocate and initialize the per-ag data.
807 spin_lock_init(&mp->m_perag_lock);
808 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
809 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
811 xfs_warn(mp, "Failed per-ag init: %d", error);
815 if (!sbp->sb_logblocks) {
816 xfs_warn(mp, "no log defined");
817 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
818 error = -EFSCORRUPTED;
823 * log's mount-time initialization. Perform 1st part recovery if needed
825 error = xfs_log_mount(mp, mp->m_logdev_targp,
826 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
827 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
829 xfs_warn(mp, "log mount failed");
834 * Now the log is mounted, we know if it was an unclean shutdown or
835 * not. If it was, with the first phase of recovery has completed, we
836 * have consistent AG blocks on disk. We have not recovered EFIs yet,
837 * but they are recovered transactionally in the second recovery phase
840 * Hence we can safely re-initialise incore superblock counters from
841 * the per-ag data. These may not be correct if the filesystem was not
842 * cleanly unmounted, so we need to wait for recovery to finish before
845 * If the filesystem was cleanly unmounted, then we can trust the
846 * values in the superblock to be correct and we don't need to do
849 * If we are currently making the filesystem, the initialisation will
850 * fail as the perag data is in an undefined state.
852 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
853 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
854 !mp->m_sb.sb_inprogress) {
855 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
861 * Get and sanity-check the root inode.
862 * Save the pointer to it in the mount structure.
864 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
866 xfs_warn(mp, "failed to read root inode");
867 goto out_log_dealloc;
872 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
873 xfs_warn(mp, "corrupted root inode %llu: not a directory",
874 (unsigned long long)rip->i_ino);
875 xfs_iunlock(rip, XFS_ILOCK_EXCL);
876 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
878 error = -EFSCORRUPTED;
881 mp->m_rootip = rip; /* save it */
883 xfs_iunlock(rip, XFS_ILOCK_EXCL);
886 * Initialize realtime inode pointers in the mount structure
888 error = xfs_rtmount_inodes(mp);
891 * Free up the root inode.
893 xfs_warn(mp, "failed to read RT inodes");
898 * If this is a read-only mount defer the superblock updates until
899 * the next remount into writeable mode. Otherwise we would never
900 * perform the update e.g. for the root filesystem.
902 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
903 error = xfs_mount_log_sb(mp, mp->m_update_flags);
905 xfs_warn(mp, "failed to write sb changes");
911 * Initialise the XFS quota management subsystem for this mount
913 if (XFS_IS_QUOTA_RUNNING(mp)) {
914 error = xfs_qm_newmount(mp, "amount, "aflags);
918 ASSERT(!XFS_IS_QUOTA_ON(mp));
921 * If a file system had quotas running earlier, but decided to
922 * mount without -o uquota/pquota/gquota options, revoke the
923 * quotachecked license.
925 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
926 xfs_notice(mp, "resetting quota flags");
927 error = xfs_mount_reset_sbqflags(mp);
934 * Finish recovering the file system. This part needed to be
935 * delayed until after the root and real-time bitmap inodes
936 * were consistently read in.
938 error = xfs_log_mount_finish(mp);
940 xfs_warn(mp, "log mount finish failed");
945 * Complete the quota initialisation, post-log-replay component.
948 ASSERT(mp->m_qflags == 0);
949 mp->m_qflags = quotaflags;
951 xfs_qm_mount_quotas(mp);
955 * Now we are mounted, reserve a small amount of unused space for
956 * privileged transactions. This is needed so that transaction
957 * space required for critical operations can dip into this pool
958 * when at ENOSPC. This is needed for operations like create with
959 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
960 * are not allowed to use this reserved space.
962 * This may drive us straight to ENOSPC on mount, but that implies
963 * we were already there on the last unmount. Warn if this occurs.
965 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
966 resblks = xfs_default_resblks(mp);
967 error = xfs_reserve_blocks(mp, &resblks, NULL);
970 "Unable to allocate reserve blocks. Continuing without reserve pool.");
976 xfs_rtunmount_inodes(mp);
982 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
983 xfs_wait_buftarg(mp->m_logdev_targp);
984 xfs_wait_buftarg(mp->m_ddev_targp);
990 xfs_uuid_unmount(mp);
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);
1076 xfs_fs_writable(xfs_mount_t *mp)
1078 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1079 (mp->m_flags & XFS_MOUNT_RDONLY));
1085 * Sync the superblock counters to disk.
1087 * Note this code can be called during the process of freezing, so
1088 * we may need to use the transaction allocator which does not
1089 * block when the transaction subsystem is in its frozen state.
1092 xfs_log_sbcount(xfs_mount_t *mp)
1097 if (!xfs_fs_writable(mp))
1100 xfs_icsb_sync_counters(mp, 0);
1103 * we don't need to do this if we are updating the superblock
1104 * counters on every modification.
1106 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1109 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1110 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1112 xfs_trans_cancel(tp, 0);
1116 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1117 xfs_trans_set_sync(tp);
1118 error = xfs_trans_commit(tp, 0);
1123 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1124 * a delta to a specified field in the in-core superblock. Simply
1125 * switch on the field indicated and apply the delta to that field.
1126 * Fields are not allowed to dip below zero, so if the delta would
1127 * do this do not apply it and return EINVAL.
1129 * The m_sb_lock must be held when this routine is called.
1132 xfs_mod_incore_sb_unlocked(
1134 xfs_sb_field_t field,
1138 int scounter; /* short counter for 32 bit fields */
1139 long long lcounter; /* long counter for 64 bit fields */
1140 long long res_used, rem;
1143 * With the in-core superblock spin lock held, switch
1144 * on the indicated field. Apply the delta to the
1145 * proper field. If the fields value would dip below
1146 * 0, then do not apply the delta and return EINVAL.
1149 case XFS_SBS_ICOUNT:
1150 lcounter = (long long)mp->m_sb.sb_icount;
1156 mp->m_sb.sb_icount = lcounter;
1159 lcounter = (long long)mp->m_sb.sb_ifree;
1165 mp->m_sb.sb_ifree = lcounter;
1167 case XFS_SBS_FDBLOCKS:
1168 lcounter = (long long)
1169 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1170 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1172 if (delta > 0) { /* Putting blocks back */
1173 if (res_used > delta) {
1174 mp->m_resblks_avail += delta;
1176 rem = delta - res_used;
1177 mp->m_resblks_avail = mp->m_resblks;
1180 } else { /* Taking blocks away */
1182 if (lcounter >= 0) {
1183 mp->m_sb.sb_fdblocks = lcounter +
1184 XFS_ALLOC_SET_ASIDE(mp);
1189 * We are out of blocks, use any available reserved
1190 * blocks if were allowed to.
1195 lcounter = (long long)mp->m_resblks_avail + delta;
1196 if (lcounter >= 0) {
1197 mp->m_resblks_avail = lcounter;
1200 printk_once(KERN_WARNING
1201 "Filesystem \"%s\": reserve blocks depleted! "
1202 "Consider increasing reserve pool size.",
1207 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1209 case XFS_SBS_FREXTENTS:
1210 lcounter = (long long)mp->m_sb.sb_frextents;
1215 mp->m_sb.sb_frextents = lcounter;
1217 case XFS_SBS_DBLOCKS:
1218 lcounter = (long long)mp->m_sb.sb_dblocks;
1224 mp->m_sb.sb_dblocks = lcounter;
1226 case XFS_SBS_AGCOUNT:
1227 scounter = mp->m_sb.sb_agcount;
1233 mp->m_sb.sb_agcount = scounter;
1235 case XFS_SBS_IMAX_PCT:
1236 scounter = mp->m_sb.sb_imax_pct;
1242 mp->m_sb.sb_imax_pct = scounter;
1244 case XFS_SBS_REXTSIZE:
1245 scounter = mp->m_sb.sb_rextsize;
1251 mp->m_sb.sb_rextsize = scounter;
1253 case XFS_SBS_RBMBLOCKS:
1254 scounter = mp->m_sb.sb_rbmblocks;
1260 mp->m_sb.sb_rbmblocks = scounter;
1262 case XFS_SBS_RBLOCKS:
1263 lcounter = (long long)mp->m_sb.sb_rblocks;
1269 mp->m_sb.sb_rblocks = lcounter;
1271 case XFS_SBS_REXTENTS:
1272 lcounter = (long long)mp->m_sb.sb_rextents;
1278 mp->m_sb.sb_rextents = lcounter;
1280 case XFS_SBS_REXTSLOG:
1281 scounter = mp->m_sb.sb_rextslog;
1287 mp->m_sb.sb_rextslog = scounter;
1296 * xfs_mod_incore_sb() is used to change a field in the in-core
1297 * superblock structure by the specified delta. This modification
1298 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1299 * routine to do the work.
1303 struct xfs_mount *mp,
1304 xfs_sb_field_t field,
1310 #ifdef HAVE_PERCPU_SB
1311 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1313 spin_lock(&mp->m_sb_lock);
1314 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1315 spin_unlock(&mp->m_sb_lock);
1321 * Change more than one field in the in-core superblock structure at a time.
1323 * The fields and changes to those fields are specified in the array of
1324 * xfs_mod_sb structures passed in. Either all of the specified deltas
1325 * will be applied or none of them will. If any modified field dips below 0,
1326 * then all modifications will be backed out and EINVAL will be returned.
1328 * Note that this function may not be used for the superblock values that
1329 * are tracked with the in-memory per-cpu counters - a direct call to
1330 * xfs_icsb_modify_counters is required for these.
1333 xfs_mod_incore_sb_batch(
1334 struct xfs_mount *mp,
1343 * Loop through the array of mod structures and apply each individually.
1344 * If any fail, then back out all those which have already been applied.
1345 * Do all of this within the scope of the m_sb_lock so that all of the
1346 * changes will be atomic.
1348 spin_lock(&mp->m_sb_lock);
1349 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1350 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1351 msbp->msb_field > XFS_SBS_FDBLOCKS);
1353 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1354 msbp->msb_delta, rsvd);
1358 spin_unlock(&mp->m_sb_lock);
1362 while (--msbp >= msb) {
1363 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1364 -msbp->msb_delta, rsvd);
1367 spin_unlock(&mp->m_sb_lock);
1372 * xfs_getsb() is called to obtain the buffer for the superblock.
1373 * The buffer is returned locked and read in from disk.
1374 * The buffer should be released with a call to xfs_brelse().
1376 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1377 * the superblock buffer if it can be locked without sleeping.
1378 * If it can't then we'll return NULL.
1382 struct xfs_mount *mp,
1385 struct xfs_buf *bp = mp->m_sb_bp;
1387 if (!xfs_buf_trylock(bp)) {
1388 if (flags & XBF_TRYLOCK)
1394 ASSERT(XFS_BUF_ISDONE(bp));
1399 * Used to free the superblock along various error paths.
1403 struct xfs_mount *mp)
1405 struct xfs_buf *bp = mp->m_sb_bp;
1413 * Used to log changes to the superblock unit and width fields which could
1414 * be altered by the mount options, as well as any potential sb_features2
1415 * fixup. Only the first superblock is updated.
1425 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1426 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1427 XFS_SB_VERSIONNUM));
1429 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1430 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1432 xfs_trans_cancel(tp, 0);
1435 xfs_mod_sb(tp, fields);
1436 error = xfs_trans_commit(tp, 0);
1441 * If the underlying (data/log/rt) device is readonly, there are some
1442 * operations that cannot proceed.
1445 xfs_dev_is_read_only(
1446 struct xfs_mount *mp,
1449 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1450 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1451 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1452 xfs_notice(mp, "%s required on read-only device.", message);
1453 xfs_notice(mp, "write access unavailable, cannot proceed.");
1459 #ifdef HAVE_PERCPU_SB
1461 * Per-cpu incore superblock counters
1463 * Simple concept, difficult implementation
1465 * Basically, replace the incore superblock counters with a distributed per cpu
1466 * counter for contended fields (e.g. free block count).
1468 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1469 * hence needs to be accurately read when we are running low on space. Hence
1470 * there is a method to enable and disable the per-cpu counters based on how
1471 * much "stuff" is available in them.
1473 * Basically, a counter is enabled if there is enough free resource to justify
1474 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1475 * ENOSPC), then we disable the counters to synchronise all callers and
1476 * re-distribute the available resources.
1478 * If, once we redistributed the available resources, we still get a failure,
1479 * we disable the per-cpu counter and go through the slow path.
1481 * The slow path is the current xfs_mod_incore_sb() function. This means that
1482 * when we disable a per-cpu counter, we need to drain its resources back to
1483 * the global superblock. We do this after disabling the counter to prevent
1484 * more threads from queueing up on the counter.
1486 * Essentially, this means that we still need a lock in the fast path to enable
1487 * synchronisation between the global counters and the per-cpu counters. This
1488 * is not a problem because the lock will be local to a CPU almost all the time
1489 * and have little contention except when we get to ENOSPC conditions.
1491 * Basically, this lock becomes a barrier that enables us to lock out the fast
1492 * path while we do things like enabling and disabling counters and
1493 * synchronising the counters.
1497 * 1. m_sb_lock before picking up per-cpu locks
1498 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1499 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1500 * 4. modifying per-cpu counters requires holding per-cpu lock
1501 * 5. modifying global counters requires holding m_sb_lock
1502 * 6. enabling or disabling a counter requires holding the m_sb_lock
1503 * and _none_ of the per-cpu locks.
1505 * Disabled counters are only ever re-enabled by a balance operation
1506 * that results in more free resources per CPU than a given threshold.
1507 * To ensure counters don't remain disabled, they are rebalanced when
1508 * the global resource goes above a higher threshold (i.e. some hysteresis
1509 * is present to prevent thrashing).
1512 #ifdef CONFIG_HOTPLUG_CPU
1514 * hot-plug CPU notifier support.
1516 * We need a notifier per filesystem as we need to be able to identify
1517 * the filesystem to balance the counters out. This is achieved by
1518 * having a notifier block embedded in the xfs_mount_t and doing pointer
1519 * magic to get the mount pointer from the notifier block address.
1522 xfs_icsb_cpu_notify(
1523 struct notifier_block *nfb,
1524 unsigned long action,
1527 xfs_icsb_cnts_t *cntp;
1530 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1531 cntp = (xfs_icsb_cnts_t *)
1532 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1534 case CPU_UP_PREPARE:
1535 case CPU_UP_PREPARE_FROZEN:
1536 /* Easy Case - initialize the area and locks, and
1537 * then rebalance when online does everything else for us. */
1538 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1541 case CPU_ONLINE_FROZEN:
1543 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1544 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1545 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1546 xfs_icsb_unlock(mp);
1549 case CPU_DEAD_FROZEN:
1550 /* Disable all the counters, then fold the dead cpu's
1551 * count into the total on the global superblock and
1552 * re-enable the counters. */
1554 spin_lock(&mp->m_sb_lock);
1555 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1556 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1557 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1559 mp->m_sb.sb_icount += cntp->icsb_icount;
1560 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1561 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1563 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1565 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1566 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1567 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1568 spin_unlock(&mp->m_sb_lock);
1569 xfs_icsb_unlock(mp);
1575 #endif /* CONFIG_HOTPLUG_CPU */
1578 xfs_icsb_init_counters(
1581 xfs_icsb_cnts_t *cntp;
1584 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1585 if (mp->m_sb_cnts == NULL)
1588 for_each_online_cpu(i) {
1589 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1590 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1593 mutex_init(&mp->m_icsb_mutex);
1596 * start with all counters disabled so that the
1597 * initial balance kicks us off correctly
1599 mp->m_icsb_counters = -1;
1601 #ifdef CONFIG_HOTPLUG_CPU
1602 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1603 mp->m_icsb_notifier.priority = 0;
1604 register_hotcpu_notifier(&mp->m_icsb_notifier);
1605 #endif /* CONFIG_HOTPLUG_CPU */
1611 xfs_icsb_reinit_counters(
1616 * start with all counters disabled so that the
1617 * initial balance kicks us off correctly
1619 mp->m_icsb_counters = -1;
1620 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1621 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1622 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1623 xfs_icsb_unlock(mp);
1627 xfs_icsb_destroy_counters(
1630 if (mp->m_sb_cnts) {
1631 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1632 free_percpu(mp->m_sb_cnts);
1634 mutex_destroy(&mp->m_icsb_mutex);
1639 xfs_icsb_cnts_t *icsbp)
1641 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1647 xfs_icsb_unlock_cntr(
1648 xfs_icsb_cnts_t *icsbp)
1650 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1655 xfs_icsb_lock_all_counters(
1658 xfs_icsb_cnts_t *cntp;
1661 for_each_online_cpu(i) {
1662 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1663 xfs_icsb_lock_cntr(cntp);
1668 xfs_icsb_unlock_all_counters(
1671 xfs_icsb_cnts_t *cntp;
1674 for_each_online_cpu(i) {
1675 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1676 xfs_icsb_unlock_cntr(cntp);
1683 xfs_icsb_cnts_t *cnt,
1686 xfs_icsb_cnts_t *cntp;
1689 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1691 if (!(flags & XFS_ICSB_LAZY_COUNT))
1692 xfs_icsb_lock_all_counters(mp);
1694 for_each_online_cpu(i) {
1695 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1696 cnt->icsb_icount += cntp->icsb_icount;
1697 cnt->icsb_ifree += cntp->icsb_ifree;
1698 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1701 if (!(flags & XFS_ICSB_LAZY_COUNT))
1702 xfs_icsb_unlock_all_counters(mp);
1706 xfs_icsb_counter_disabled(
1708 xfs_sb_field_t field)
1710 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1711 return test_bit(field, &mp->m_icsb_counters);
1715 xfs_icsb_disable_counter(
1717 xfs_sb_field_t field)
1719 xfs_icsb_cnts_t cnt;
1721 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1724 * If we are already disabled, then there is nothing to do
1725 * here. We check before locking all the counters to avoid
1726 * the expensive lock operation when being called in the
1727 * slow path and the counter is already disabled. This is
1728 * safe because the only time we set or clear this state is under
1731 if (xfs_icsb_counter_disabled(mp, field))
1734 xfs_icsb_lock_all_counters(mp);
1735 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1736 /* drain back to superblock */
1738 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1740 case XFS_SBS_ICOUNT:
1741 mp->m_sb.sb_icount = cnt.icsb_icount;
1744 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1746 case XFS_SBS_FDBLOCKS:
1747 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1754 xfs_icsb_unlock_all_counters(mp);
1758 xfs_icsb_enable_counter(
1760 xfs_sb_field_t field,
1764 xfs_icsb_cnts_t *cntp;
1767 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1769 xfs_icsb_lock_all_counters(mp);
1770 for_each_online_cpu(i) {
1771 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1773 case XFS_SBS_ICOUNT:
1774 cntp->icsb_icount = count + resid;
1777 cntp->icsb_ifree = count + resid;
1779 case XFS_SBS_FDBLOCKS:
1780 cntp->icsb_fdblocks = count + resid;
1788 clear_bit(field, &mp->m_icsb_counters);
1789 xfs_icsb_unlock_all_counters(mp);
1793 xfs_icsb_sync_counters_locked(
1797 xfs_icsb_cnts_t cnt;
1799 xfs_icsb_count(mp, &cnt, flags);
1801 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1802 mp->m_sb.sb_icount = cnt.icsb_icount;
1803 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1804 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1805 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1806 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1810 * Accurate update of per-cpu counters to incore superblock
1813 xfs_icsb_sync_counters(
1817 spin_lock(&mp->m_sb_lock);
1818 xfs_icsb_sync_counters_locked(mp, flags);
1819 spin_unlock(&mp->m_sb_lock);
1823 * Balance and enable/disable counters as necessary.
1825 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1826 * chosen to be the same number as single on disk allocation chunk per CPU, and
1827 * free blocks is something far enough zero that we aren't going thrash when we
1828 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1829 * prevent looping endlessly when xfs_alloc_space asks for more than will
1830 * be distributed to a single CPU but each CPU has enough blocks to be
1833 * Note that we can be called when counters are already disabled.
1834 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1835 * prevent locking every per-cpu counter needlessly.
1838 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1839 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1840 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1842 xfs_icsb_balance_counter_locked(
1844 xfs_sb_field_t field,
1847 uint64_t count, resid;
1848 int weight = num_online_cpus();
1849 uint64_t min = (uint64_t)min_per_cpu;
1851 /* disable counter and sync counter */
1852 xfs_icsb_disable_counter(mp, field);
1854 /* update counters - first CPU gets residual*/
1856 case XFS_SBS_ICOUNT:
1857 count = mp->m_sb.sb_icount;
1858 resid = do_div(count, weight);
1859 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1863 count = mp->m_sb.sb_ifree;
1864 resid = do_div(count, weight);
1865 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1868 case XFS_SBS_FDBLOCKS:
1869 count = mp->m_sb.sb_fdblocks;
1870 resid = do_div(count, weight);
1871 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
1876 count = resid = 0; /* quiet, gcc */
1880 xfs_icsb_enable_counter(mp, field, count, resid);
1884 xfs_icsb_balance_counter(
1886 xfs_sb_field_t fields,
1889 spin_lock(&mp->m_sb_lock);
1890 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1891 spin_unlock(&mp->m_sb_lock);
1895 xfs_icsb_modify_counters(
1897 xfs_sb_field_t field,
1901 xfs_icsb_cnts_t *icsbp;
1902 long long lcounter; /* long counter for 64 bit fields */
1908 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1911 * if the counter is disabled, go to slow path
1913 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1915 xfs_icsb_lock_cntr(icsbp);
1916 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1917 xfs_icsb_unlock_cntr(icsbp);
1922 case XFS_SBS_ICOUNT:
1923 lcounter = icsbp->icsb_icount;
1925 if (unlikely(lcounter < 0))
1926 goto balance_counter;
1927 icsbp->icsb_icount = lcounter;
1931 lcounter = icsbp->icsb_ifree;
1933 if (unlikely(lcounter < 0))
1934 goto balance_counter;
1935 icsbp->icsb_ifree = lcounter;
1938 case XFS_SBS_FDBLOCKS:
1939 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1941 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1943 if (unlikely(lcounter < 0))
1944 goto balance_counter;
1945 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1951 xfs_icsb_unlock_cntr(icsbp);
1959 * serialise with a mutex so we don't burn lots of cpu on
1960 * the superblock lock. We still need to hold the superblock
1961 * lock, however, when we modify the global structures.
1966 * Now running atomically.
1968 * If the counter is enabled, someone has beaten us to rebalancing.
1969 * Drop the lock and try again in the fast path....
1971 if (!(xfs_icsb_counter_disabled(mp, field))) {
1972 xfs_icsb_unlock(mp);
1977 * The counter is currently disabled. Because we are
1978 * running atomically here, we know a rebalance cannot
1979 * be in progress. Hence we can go straight to operating
1980 * on the global superblock. We do not call xfs_mod_incore_sb()
1981 * here even though we need to get the m_sb_lock. Doing so
1982 * will cause us to re-enter this function and deadlock.
1983 * Hence we get the m_sb_lock ourselves and then call
1984 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1985 * directly on the global counters.
1987 spin_lock(&mp->m_sb_lock);
1988 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1989 spin_unlock(&mp->m_sb_lock);
1992 * Now that we've modified the global superblock, we
1993 * may be able to re-enable the distributed counters
1994 * (e.g. lots of space just got freed). After that
1998 xfs_icsb_balance_counter(mp, field, 0);
1999 xfs_icsb_unlock(mp);
2003 xfs_icsb_unlock_cntr(icsbp);
2007 * We may have multiple threads here if multiple per-cpu
2008 * counters run dry at the same time. This will mean we can
2009 * do more balances than strictly necessary but it is not
2010 * the common slowpath case.
2015 * running atomically.
2017 * This will leave the counter in the correct state for future
2018 * accesses. After the rebalance, we simply try again and our retry
2019 * will either succeed through the fast path or slow path without
2020 * another balance operation being required.
2022 xfs_icsb_balance_counter(mp, field, delta);
2023 xfs_icsb_unlock(mp);