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;
64 extern struct kset *xfs_kset;
67 * See if the UUID is unique among mounted XFS filesystems.
68 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
74 uuid_t *uuid = &mp->m_sb.sb_uuid;
77 if (mp->m_flags & XFS_MOUNT_NOUUID)
80 if (uuid_is_nil(uuid)) {
81 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
85 mutex_lock(&xfs_uuid_table_mutex);
86 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
87 if (uuid_is_nil(&xfs_uuid_table[i])) {
91 if (uuid_equal(uuid, &xfs_uuid_table[i]))
96 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
97 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
98 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
100 hole = xfs_uuid_table_size++;
102 xfs_uuid_table[hole] = *uuid;
103 mutex_unlock(&xfs_uuid_table_mutex);
108 mutex_unlock(&xfs_uuid_table_mutex);
109 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
115 struct xfs_mount *mp)
117 uuid_t *uuid = &mp->m_sb.sb_uuid;
120 if (mp->m_flags & XFS_MOUNT_NOUUID)
123 mutex_lock(&xfs_uuid_table_mutex);
124 for (i = 0; i < xfs_uuid_table_size; i++) {
125 if (uuid_is_nil(&xfs_uuid_table[i]))
127 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
129 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
132 ASSERT(i < xfs_uuid_table_size);
133 mutex_unlock(&xfs_uuid_table_mutex);
139 struct rcu_head *head)
141 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
143 ASSERT(atomic_read(&pag->pag_ref) == 0);
148 * Free up the per-ag resources associated with the mount structure.
155 struct xfs_perag *pag;
157 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
158 spin_lock(&mp->m_perag_lock);
159 pag = radix_tree_delete(&mp->m_perag_tree, agno);
160 spin_unlock(&mp->m_perag_lock);
162 ASSERT(atomic_read(&pag->pag_ref) == 0);
163 call_rcu(&pag->rcu_head, __xfs_free_perag);
168 * Check size of device based on the (data/realtime) block count.
169 * Note: this check is used by the growfs code as well as mount.
172 xfs_sb_validate_fsb_count(
176 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
177 ASSERT(sbp->sb_blocklog >= BBSHIFT);
179 /* Limited by ULONG_MAX of page cache index */
180 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
186 xfs_initialize_perag(
188 xfs_agnumber_t agcount,
189 xfs_agnumber_t *maxagi)
191 xfs_agnumber_t index;
192 xfs_agnumber_t first_initialised = 0;
196 xfs_sb_t *sbp = &mp->m_sb;
200 * Walk the current per-ag tree so we don't try to initialise AGs
201 * that already exist (growfs case). Allocate and insert all the
202 * AGs we don't find ready for initialisation.
204 for (index = 0; index < agcount; index++) {
205 pag = xfs_perag_get(mp, index);
210 if (!first_initialised)
211 first_initialised = index;
213 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
216 pag->pag_agno = index;
218 spin_lock_init(&pag->pag_ici_lock);
219 mutex_init(&pag->pag_ici_reclaim_lock);
220 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
221 spin_lock_init(&pag->pag_buf_lock);
222 pag->pag_buf_tree = RB_ROOT;
224 if (radix_tree_preload(GFP_NOFS))
227 spin_lock(&mp->m_perag_lock);
228 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
230 spin_unlock(&mp->m_perag_lock);
231 radix_tree_preload_end();
235 spin_unlock(&mp->m_perag_lock);
236 radix_tree_preload_end();
240 * If we mount with the inode64 option, or no inode overflows
241 * the legacy 32-bit address space clear the inode32 option.
243 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
244 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
246 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
247 mp->m_flags |= XFS_MOUNT_32BITINODES;
249 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
251 if (mp->m_flags & XFS_MOUNT_32BITINODES)
252 index = xfs_set_inode32(mp, agcount);
254 index = xfs_set_inode64(mp, agcount);
262 for (; index > first_initialised; index--) {
263 pag = radix_tree_delete(&mp->m_perag_tree, index);
272 * Does the initial read of the superblock.
276 struct xfs_mount *mp,
279 unsigned int sector_size;
281 struct xfs_sb *sbp = &mp->m_sb;
283 int loud = !(flags & XFS_MFSI_QUIET);
284 const struct xfs_buf_ops *buf_ops;
286 ASSERT(mp->m_sb_bp == NULL);
287 ASSERT(mp->m_ddev_targp != NULL);
290 * For the initial read, we must guess at the sector
291 * size based on the block device. It's enough to
292 * get the sb_sectsize out of the superblock and
293 * then reread with the proper length.
294 * We don't verify it yet, because it may not be complete.
296 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
300 * Allocate a (locked) buffer to hold the superblock.
301 * This will be kept around at all times to optimize
302 * access to the superblock.
305 bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
306 BTOBB(sector_size), 0, buf_ops);
309 xfs_warn(mp, "SB buffer read failed");
315 xfs_warn(mp, "SB validate failed with error %d.", error);
316 /* bad CRC means corrupted metadata */
317 if (error == -EFSBADCRC)
318 error = -EFSCORRUPTED;
323 * Initialize the mount structure from the superblock.
325 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
328 * If we haven't validated the superblock, do so now before we try
329 * to check the sector size and reread the superblock appropriately.
331 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
333 xfs_warn(mp, "Invalid superblock magic number");
339 * We must be able to do sector-sized and sector-aligned IO.
341 if (sector_size > sbp->sb_sectsize) {
343 xfs_warn(mp, "device supports %u byte sectors (not %u)",
344 sector_size, sbp->sb_sectsize);
349 if (buf_ops == NULL) {
351 * Re-read the superblock so the buffer is correctly sized,
352 * and properly verified.
355 sector_size = sbp->sb_sectsize;
356 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
360 /* Initialize per-cpu counters */
361 xfs_icsb_reinit_counters(mp);
363 /* no need to be quiet anymore, so reset the buf ops */
364 bp->b_ops = &xfs_sb_buf_ops;
376 * Update alignment values based on mount options and sb values
379 xfs_update_alignment(xfs_mount_t *mp)
381 xfs_sb_t *sbp = &(mp->m_sb);
385 * If stripe unit and stripe width are not multiples
386 * of the fs blocksize turn off alignment.
388 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
389 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
391 "alignment check failed: sunit/swidth vs. blocksize(%d)",
396 * Convert the stripe unit and width to FSBs.
398 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
399 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
401 "alignment check failed: sunit/swidth vs. agsize(%d)",
404 } else if (mp->m_dalign) {
405 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
408 "alignment check failed: sunit(%d) less than bsize(%d)",
409 mp->m_dalign, sbp->sb_blocksize);
415 * Update superblock with new values
418 if (xfs_sb_version_hasdalign(sbp)) {
419 if (sbp->sb_unit != mp->m_dalign) {
420 sbp->sb_unit = mp->m_dalign;
421 mp->m_update_flags |= XFS_SB_UNIT;
423 if (sbp->sb_width != mp->m_swidth) {
424 sbp->sb_width = mp->m_swidth;
425 mp->m_update_flags |= XFS_SB_WIDTH;
429 "cannot change alignment: superblock does not support data alignment");
432 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
433 xfs_sb_version_hasdalign(&mp->m_sb)) {
434 mp->m_dalign = sbp->sb_unit;
435 mp->m_swidth = sbp->sb_width;
442 * Set the maximum inode count for this filesystem
445 xfs_set_maxicount(xfs_mount_t *mp)
447 xfs_sb_t *sbp = &(mp->m_sb);
450 if (sbp->sb_imax_pct) {
452 * Make sure the maximum inode count is a multiple
453 * of the units we allocate inodes in.
455 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
457 do_div(icount, mp->m_ialloc_blks);
458 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
466 * Set the default minimum read and write sizes unless
467 * already specified in a mount option.
468 * We use smaller I/O sizes when the file system
469 * is being used for NFS service (wsync mount option).
472 xfs_set_rw_sizes(xfs_mount_t *mp)
474 xfs_sb_t *sbp = &(mp->m_sb);
475 int readio_log, writeio_log;
477 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
478 if (mp->m_flags & XFS_MOUNT_WSYNC) {
479 readio_log = XFS_WSYNC_READIO_LOG;
480 writeio_log = XFS_WSYNC_WRITEIO_LOG;
482 readio_log = XFS_READIO_LOG_LARGE;
483 writeio_log = XFS_WRITEIO_LOG_LARGE;
486 readio_log = mp->m_readio_log;
487 writeio_log = mp->m_writeio_log;
490 if (sbp->sb_blocklog > readio_log) {
491 mp->m_readio_log = sbp->sb_blocklog;
493 mp->m_readio_log = readio_log;
495 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
496 if (sbp->sb_blocklog > writeio_log) {
497 mp->m_writeio_log = sbp->sb_blocklog;
499 mp->m_writeio_log = writeio_log;
501 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
505 * precalculate the low space thresholds for dynamic speculative preallocation.
508 xfs_set_low_space_thresholds(
509 struct xfs_mount *mp)
513 for (i = 0; i < XFS_LOWSP_MAX; i++) {
514 __uint64_t space = mp->m_sb.sb_dblocks;
517 mp->m_low_space[i] = space * (i + 1);
523 * Set whether we're using inode alignment.
526 xfs_set_inoalignment(xfs_mount_t *mp)
528 if (xfs_sb_version_hasalign(&mp->m_sb) &&
529 mp->m_sb.sb_inoalignmt >=
530 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
531 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
533 mp->m_inoalign_mask = 0;
535 * If we are using stripe alignment, check whether
536 * the stripe unit is a multiple of the inode alignment
538 if (mp->m_dalign && mp->m_inoalign_mask &&
539 !(mp->m_dalign & mp->m_inoalign_mask))
540 mp->m_sinoalign = mp->m_dalign;
546 * Check that the data (and log if separate) is an ok size.
549 xfs_check_sizes(xfs_mount_t *mp)
554 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
555 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
556 xfs_warn(mp, "filesystem size mismatch detected");
559 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
560 d - XFS_FSS_TO_BB(mp, 1),
561 XFS_FSS_TO_BB(mp, 1), 0, NULL);
563 xfs_warn(mp, "last sector read failed");
568 if (mp->m_logdev_targp != mp->m_ddev_targp) {
569 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
570 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
571 xfs_warn(mp, "log size mismatch detected");
574 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
575 d - XFS_FSB_TO_BB(mp, 1),
576 XFS_FSB_TO_BB(mp, 1), 0, NULL);
578 xfs_warn(mp, "log device read failed");
587 * Clear the quotaflags in memory and in the superblock.
590 xfs_mount_reset_sbqflags(
591 struct xfs_mount *mp)
594 struct xfs_trans *tp;
599 * It is OK to look at sb_qflags here in mount path,
602 if (mp->m_sb.sb_qflags == 0)
604 spin_lock(&mp->m_sb_lock);
605 mp->m_sb.sb_qflags = 0;
606 spin_unlock(&mp->m_sb_lock);
609 * If the fs is readonly, let the incore superblock run
610 * with quotas off but don't flush the update out to disk
612 if (mp->m_flags & XFS_MOUNT_RDONLY)
615 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
616 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
618 xfs_trans_cancel(tp, 0);
619 xfs_alert(mp, "%s: Superblock update failed!", __func__);
623 xfs_mod_sb(tp, XFS_SB_QFLAGS);
624 return xfs_trans_commit(tp, 0);
628 xfs_default_resblks(xfs_mount_t *mp)
633 * We default to 5% or 8192 fsbs of space reserved, whichever is
634 * smaller. This is intended to cover concurrent allocation
635 * transactions when we initially hit enospc. These each require a 4
636 * block reservation. Hence by default we cover roughly 2000 concurrent
637 * allocation reservations.
639 resblks = mp->m_sb.sb_dblocks;
641 resblks = min_t(__uint64_t, resblks, 8192);
646 * This function does the following on an initial mount of a file system:
647 * - reads the superblock from disk and init the mount struct
648 * - if we're a 32-bit kernel, do a size check on the superblock
649 * so we don't mount terabyte filesystems
650 * - init mount struct realtime fields
651 * - allocate inode hash table for fs
652 * - init directory manager
653 * - perform recovery and init the log manager
659 xfs_sb_t *sbp = &(mp->m_sb);
666 xfs_sb_mount_common(mp, sbp);
669 * Check for a mismatched features2 values. Older kernels
670 * read & wrote into the wrong sb offset for sb_features2
671 * on some platforms due to xfs_sb_t not being 64bit size aligned
672 * when sb_features2 was added, which made older superblock
673 * reading/writing routines swap it as a 64-bit value.
675 * For backwards compatibility, we make both slots equal.
677 * If we detect a mismatched field, we OR the set bits into the
678 * existing features2 field in case it has already been modified; we
679 * don't want to lose any features. We then update the bad location
680 * with the ORed value so that older kernels will see any features2
681 * flags, and mark the two fields as needing updates once the
682 * transaction subsystem is online.
684 if (xfs_sb_has_mismatched_features2(sbp)) {
685 xfs_warn(mp, "correcting sb_features alignment problem");
686 sbp->sb_features2 |= sbp->sb_bad_features2;
687 sbp->sb_bad_features2 = sbp->sb_features2;
688 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
691 * Re-check for ATTR2 in case it was found in bad_features2
694 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
695 !(mp->m_flags & XFS_MOUNT_NOATTR2))
696 mp->m_flags |= XFS_MOUNT_ATTR2;
699 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
700 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
701 xfs_sb_version_removeattr2(&mp->m_sb);
702 mp->m_update_flags |= XFS_SB_FEATURES2;
704 /* update sb_versionnum for the clearing of the morebits */
705 if (!sbp->sb_features2)
706 mp->m_update_flags |= XFS_SB_VERSIONNUM;
709 /* always use v2 inodes by default now */
710 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
711 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
712 mp->m_update_flags |= XFS_SB_VERSIONNUM;
716 * Check if sb_agblocks is aligned at stripe boundary
717 * If sb_agblocks is NOT aligned turn off m_dalign since
718 * allocator alignment is within an ag, therefore ag has
719 * to be aligned at stripe boundary.
721 error = xfs_update_alignment(mp);
725 xfs_alloc_compute_maxlevels(mp);
726 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
727 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
728 xfs_ialloc_compute_maxlevels(mp);
730 xfs_set_maxicount(mp);
732 mp->m_kobj.kobject.kset = xfs_kset;
733 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
737 error = xfs_uuid_mount(mp);
739 goto out_remove_sysfs;
742 * Set the minimum read and write sizes
744 xfs_set_rw_sizes(mp);
746 /* set the low space thresholds for dynamic preallocation */
747 xfs_set_low_space_thresholds(mp);
750 * Set the inode cluster size.
751 * This may still be overridden by the file system
752 * block size if it is larger than the chosen cluster size.
754 * For v5 filesystems, scale the cluster size with the inode size to
755 * keep a constant ratio of inode per cluster buffer, but only if mkfs
756 * has set the inode alignment value appropriately for larger cluster
759 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
760 if (xfs_sb_version_hascrc(&mp->m_sb)) {
761 int new_size = mp->m_inode_cluster_size;
763 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
764 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
765 mp->m_inode_cluster_size = new_size;
769 * Set inode alignment fields
771 xfs_set_inoalignment(mp);
774 * Check that the data (and log if separate) is an ok size.
776 error = xfs_check_sizes(mp);
778 goto out_remove_uuid;
781 * Initialize realtime fields in the mount structure
783 error = xfs_rtmount_init(mp);
785 xfs_warn(mp, "RT mount failed");
786 goto out_remove_uuid;
790 * Copies the low order bits of the timestamp and the randomly
791 * set "sequence" number out of a UUID.
793 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
795 mp->m_dmevmask = 0; /* not persistent; set after each mount */
797 error = xfs_da_mount(mp);
799 xfs_warn(mp, "Failed dir/attr init: %d", error);
800 goto out_remove_uuid;
804 * Initialize the precomputed transaction reservations values.
809 * Allocate and initialize the per-ag data.
811 spin_lock_init(&mp->m_perag_lock);
812 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
813 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
815 xfs_warn(mp, "Failed per-ag init: %d", error);
819 if (!sbp->sb_logblocks) {
820 xfs_warn(mp, "no log defined");
821 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
822 error = -EFSCORRUPTED;
827 * log's mount-time initialization. Perform 1st part recovery if needed
829 error = xfs_log_mount(mp, mp->m_logdev_targp,
830 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
831 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
833 xfs_warn(mp, "log mount failed");
838 * Now the log is mounted, we know if it was an unclean shutdown or
839 * not. If it was, with the first phase of recovery has completed, we
840 * have consistent AG blocks on disk. We have not recovered EFIs yet,
841 * but they are recovered transactionally in the second recovery phase
844 * Hence we can safely re-initialise incore superblock counters from
845 * the per-ag data. These may not be correct if the filesystem was not
846 * cleanly unmounted, so we need to wait for recovery to finish before
849 * If the filesystem was cleanly unmounted, then we can trust the
850 * values in the superblock to be correct and we don't need to do
853 * If we are currently making the filesystem, the initialisation will
854 * fail as the perag data is in an undefined state.
856 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
857 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
858 !mp->m_sb.sb_inprogress) {
859 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
861 goto out_log_dealloc;
865 * Get and sanity-check the root inode.
866 * Save the pointer to it in the mount structure.
868 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
870 xfs_warn(mp, "failed to read root inode");
871 goto out_log_dealloc;
876 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
877 xfs_warn(mp, "corrupted root inode %llu: not a directory",
878 (unsigned long long)rip->i_ino);
879 xfs_iunlock(rip, XFS_ILOCK_EXCL);
880 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
882 error = -EFSCORRUPTED;
885 mp->m_rootip = rip; /* save it */
887 xfs_iunlock(rip, XFS_ILOCK_EXCL);
890 * Initialize realtime inode pointers in the mount structure
892 error = xfs_rtmount_inodes(mp);
895 * Free up the root inode.
897 xfs_warn(mp, "failed to read RT inodes");
902 * If this is a read-only mount defer the superblock updates until
903 * the next remount into writeable mode. Otherwise we would never
904 * perform the update e.g. for the root filesystem.
906 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
907 error = xfs_mount_log_sb(mp, mp->m_update_flags);
909 xfs_warn(mp, "failed to write sb changes");
915 * Initialise the XFS quota management subsystem for this mount
917 if (XFS_IS_QUOTA_RUNNING(mp)) {
918 error = xfs_qm_newmount(mp, "amount, "aflags);
922 ASSERT(!XFS_IS_QUOTA_ON(mp));
925 * If a file system had quotas running earlier, but decided to
926 * mount without -o uquota/pquota/gquota options, revoke the
927 * quotachecked license.
929 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
930 xfs_notice(mp, "resetting quota flags");
931 error = xfs_mount_reset_sbqflags(mp);
938 * Finish recovering the file system. This part needed to be
939 * delayed until after the root and real-time bitmap inodes
940 * were consistently read in.
942 error = xfs_log_mount_finish(mp);
944 xfs_warn(mp, "log mount finish failed");
949 * Complete the quota initialisation, post-log-replay component.
952 ASSERT(mp->m_qflags == 0);
953 mp->m_qflags = quotaflags;
955 xfs_qm_mount_quotas(mp);
959 * Now we are mounted, reserve a small amount of unused space for
960 * privileged transactions. This is needed so that transaction
961 * space required for critical operations can dip into this pool
962 * when at ENOSPC. This is needed for operations like create with
963 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
964 * are not allowed to use this reserved space.
966 * This may drive us straight to ENOSPC on mount, but that implies
967 * we were already there on the last unmount. Warn if this occurs.
969 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
970 resblks = xfs_default_resblks(mp);
971 error = xfs_reserve_blocks(mp, &resblks, NULL);
974 "Unable to allocate reserve blocks. Continuing without reserve pool.");
980 xfs_rtunmount_inodes(mp);
986 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
987 xfs_wait_buftarg(mp->m_logdev_targp);
988 xfs_wait_buftarg(mp->m_ddev_targp);
994 xfs_uuid_unmount(mp);
996 xfs_sysfs_del(&mp->m_kobj);
1002 * This flushes out the inodes,dquots and the superblock, unmounts the
1003 * log and makes sure that incore structures are freed.
1007 struct xfs_mount *mp)
1012 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1014 xfs_qm_unmount_quotas(mp);
1015 xfs_rtunmount_inodes(mp);
1016 IRELE(mp->m_rootip);
1019 * We can potentially deadlock here if we have an inode cluster
1020 * that has been freed has its buffer still pinned in memory because
1021 * the transaction is still sitting in a iclog. The stale inodes
1022 * on that buffer will have their flush locks held until the
1023 * transaction hits the disk and the callbacks run. the inode
1024 * flush takes the flush lock unconditionally and with nothing to
1025 * push out the iclog we will never get that unlocked. hence we
1026 * need to force the log first.
1028 xfs_log_force(mp, XFS_LOG_SYNC);
1031 * Flush all pending changes from the AIL.
1033 xfs_ail_push_all_sync(mp->m_ail);
1036 * And reclaim all inodes. At this point there should be no dirty
1037 * inodes and none should be pinned or locked, but use synchronous
1038 * reclaim just to be sure. We can stop background inode reclaim
1039 * here as well if it is still running.
1041 cancel_delayed_work_sync(&mp->m_reclaim_work);
1042 xfs_reclaim_inodes(mp, SYNC_WAIT);
1047 * Unreserve any blocks we have so that when we unmount we don't account
1048 * the reserved free space as used. This is really only necessary for
1049 * lazy superblock counting because it trusts the incore superblock
1050 * counters to be absolutely correct on clean unmount.
1052 * We don't bother correcting this elsewhere for lazy superblock
1053 * counting because on mount of an unclean filesystem we reconstruct the
1054 * correct counter value and this is irrelevant.
1056 * For non-lazy counter filesystems, this doesn't matter at all because
1057 * we only every apply deltas to the superblock and hence the incore
1058 * value does not matter....
1061 error = xfs_reserve_blocks(mp, &resblks, NULL);
1063 xfs_warn(mp, "Unable to free reserved block pool. "
1064 "Freespace may not be correct on next mount.");
1066 error = xfs_log_sbcount(mp);
1068 xfs_warn(mp, "Unable to update superblock counters. "
1069 "Freespace may not be correct on next mount.");
1071 xfs_log_unmount(mp);
1073 xfs_uuid_unmount(mp);
1076 xfs_errortag_clearall(mp, 0);
1080 xfs_sysfs_del(&mp->m_kobj);
1084 xfs_fs_writable(xfs_mount_t *mp)
1086 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1087 (mp->m_flags & XFS_MOUNT_RDONLY));
1093 * Sync the superblock counters to disk.
1095 * Note this code can be called during the process of freezing, so
1096 * we may need to use the transaction allocator which does not
1097 * block when the transaction subsystem is in its frozen state.
1100 xfs_log_sbcount(xfs_mount_t *mp)
1105 if (!xfs_fs_writable(mp))
1108 xfs_icsb_sync_counters(mp, 0);
1111 * we don't need to do this if we are updating the superblock
1112 * counters on every modification.
1114 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1117 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1118 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1120 xfs_trans_cancel(tp, 0);
1124 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1125 xfs_trans_set_sync(tp);
1126 error = xfs_trans_commit(tp, 0);
1131 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1132 * a delta to a specified field in the in-core superblock. Simply
1133 * switch on the field indicated and apply the delta to that field.
1134 * Fields are not allowed to dip below zero, so if the delta would
1135 * do this do not apply it and return EINVAL.
1137 * The m_sb_lock must be held when this routine is called.
1140 xfs_mod_incore_sb_unlocked(
1142 xfs_sb_field_t field,
1146 int scounter; /* short counter for 32 bit fields */
1147 long long lcounter; /* long counter for 64 bit fields */
1148 long long res_used, rem;
1151 * With the in-core superblock spin lock held, switch
1152 * on the indicated field. Apply the delta to the
1153 * proper field. If the fields value would dip below
1154 * 0, then do not apply the delta and return EINVAL.
1157 case XFS_SBS_ICOUNT:
1158 lcounter = (long long)mp->m_sb.sb_icount;
1164 mp->m_sb.sb_icount = lcounter;
1167 lcounter = (long long)mp->m_sb.sb_ifree;
1173 mp->m_sb.sb_ifree = lcounter;
1175 case XFS_SBS_FDBLOCKS:
1176 lcounter = (long long)
1177 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1178 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1180 if (delta > 0) { /* Putting blocks back */
1181 if (res_used > delta) {
1182 mp->m_resblks_avail += delta;
1184 rem = delta - res_used;
1185 mp->m_resblks_avail = mp->m_resblks;
1188 } else { /* Taking blocks away */
1190 if (lcounter >= 0) {
1191 mp->m_sb.sb_fdblocks = lcounter +
1192 XFS_ALLOC_SET_ASIDE(mp);
1197 * We are out of blocks, use any available reserved
1198 * blocks if were allowed to.
1203 lcounter = (long long)mp->m_resblks_avail + delta;
1204 if (lcounter >= 0) {
1205 mp->m_resblks_avail = lcounter;
1208 printk_once(KERN_WARNING
1209 "Filesystem \"%s\": reserve blocks depleted! "
1210 "Consider increasing reserve pool size.",
1215 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1217 case XFS_SBS_FREXTENTS:
1218 lcounter = (long long)mp->m_sb.sb_frextents;
1223 mp->m_sb.sb_frextents = lcounter;
1225 case XFS_SBS_DBLOCKS:
1226 lcounter = (long long)mp->m_sb.sb_dblocks;
1232 mp->m_sb.sb_dblocks = lcounter;
1234 case XFS_SBS_AGCOUNT:
1235 scounter = mp->m_sb.sb_agcount;
1241 mp->m_sb.sb_agcount = scounter;
1243 case XFS_SBS_IMAX_PCT:
1244 scounter = mp->m_sb.sb_imax_pct;
1250 mp->m_sb.sb_imax_pct = scounter;
1252 case XFS_SBS_REXTSIZE:
1253 scounter = mp->m_sb.sb_rextsize;
1259 mp->m_sb.sb_rextsize = scounter;
1261 case XFS_SBS_RBMBLOCKS:
1262 scounter = mp->m_sb.sb_rbmblocks;
1268 mp->m_sb.sb_rbmblocks = scounter;
1270 case XFS_SBS_RBLOCKS:
1271 lcounter = (long long)mp->m_sb.sb_rblocks;
1277 mp->m_sb.sb_rblocks = lcounter;
1279 case XFS_SBS_REXTENTS:
1280 lcounter = (long long)mp->m_sb.sb_rextents;
1286 mp->m_sb.sb_rextents = lcounter;
1288 case XFS_SBS_REXTSLOG:
1289 scounter = mp->m_sb.sb_rextslog;
1295 mp->m_sb.sb_rextslog = scounter;
1304 * xfs_mod_incore_sb() is used to change a field in the in-core
1305 * superblock structure by the specified delta. This modification
1306 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1307 * routine to do the work.
1311 struct xfs_mount *mp,
1312 xfs_sb_field_t field,
1318 #ifdef HAVE_PERCPU_SB
1319 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1321 spin_lock(&mp->m_sb_lock);
1322 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1323 spin_unlock(&mp->m_sb_lock);
1329 * Change more than one field in the in-core superblock structure at a time.
1331 * The fields and changes to those fields are specified in the array of
1332 * xfs_mod_sb structures passed in. Either all of the specified deltas
1333 * will be applied or none of them will. If any modified field dips below 0,
1334 * then all modifications will be backed out and EINVAL will be returned.
1336 * Note that this function may not be used for the superblock values that
1337 * are tracked with the in-memory per-cpu counters - a direct call to
1338 * xfs_icsb_modify_counters is required for these.
1341 xfs_mod_incore_sb_batch(
1342 struct xfs_mount *mp,
1351 * Loop through the array of mod structures and apply each individually.
1352 * If any fail, then back out all those which have already been applied.
1353 * Do all of this within the scope of the m_sb_lock so that all of the
1354 * changes will be atomic.
1356 spin_lock(&mp->m_sb_lock);
1357 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1358 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1359 msbp->msb_field > XFS_SBS_FDBLOCKS);
1361 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1362 msbp->msb_delta, rsvd);
1366 spin_unlock(&mp->m_sb_lock);
1370 while (--msbp >= msb) {
1371 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1372 -msbp->msb_delta, rsvd);
1375 spin_unlock(&mp->m_sb_lock);
1380 * xfs_getsb() is called to obtain the buffer for the superblock.
1381 * The buffer is returned locked and read in from disk.
1382 * The buffer should be released with a call to xfs_brelse().
1384 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1385 * the superblock buffer if it can be locked without sleeping.
1386 * If it can't then we'll return NULL.
1390 struct xfs_mount *mp,
1393 struct xfs_buf *bp = mp->m_sb_bp;
1395 if (!xfs_buf_trylock(bp)) {
1396 if (flags & XBF_TRYLOCK)
1402 ASSERT(XFS_BUF_ISDONE(bp));
1407 * Used to free the superblock along various error paths.
1411 struct xfs_mount *mp)
1413 struct xfs_buf *bp = mp->m_sb_bp;
1421 * Used to log changes to the superblock unit and width fields which could
1422 * be altered by the mount options, as well as any potential sb_features2
1423 * fixup. Only the first superblock is updated.
1433 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1434 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1435 XFS_SB_VERSIONNUM));
1437 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1438 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1440 xfs_trans_cancel(tp, 0);
1443 xfs_mod_sb(tp, fields);
1444 error = xfs_trans_commit(tp, 0);
1449 * If the underlying (data/log/rt) device is readonly, there are some
1450 * operations that cannot proceed.
1453 xfs_dev_is_read_only(
1454 struct xfs_mount *mp,
1457 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1458 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1459 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1460 xfs_notice(mp, "%s required on read-only device.", message);
1461 xfs_notice(mp, "write access unavailable, cannot proceed.");
1467 #ifdef HAVE_PERCPU_SB
1469 * Per-cpu incore superblock counters
1471 * Simple concept, difficult implementation
1473 * Basically, replace the incore superblock counters with a distributed per cpu
1474 * counter for contended fields (e.g. free block count).
1476 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1477 * hence needs to be accurately read when we are running low on space. Hence
1478 * there is a method to enable and disable the per-cpu counters based on how
1479 * much "stuff" is available in them.
1481 * Basically, a counter is enabled if there is enough free resource to justify
1482 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1483 * ENOSPC), then we disable the counters to synchronise all callers and
1484 * re-distribute the available resources.
1486 * If, once we redistributed the available resources, we still get a failure,
1487 * we disable the per-cpu counter and go through the slow path.
1489 * The slow path is the current xfs_mod_incore_sb() function. This means that
1490 * when we disable a per-cpu counter, we need to drain its resources back to
1491 * the global superblock. We do this after disabling the counter to prevent
1492 * more threads from queueing up on the counter.
1494 * Essentially, this means that we still need a lock in the fast path to enable
1495 * synchronisation between the global counters and the per-cpu counters. This
1496 * is not a problem because the lock will be local to a CPU almost all the time
1497 * and have little contention except when we get to ENOSPC conditions.
1499 * Basically, this lock becomes a barrier that enables us to lock out the fast
1500 * path while we do things like enabling and disabling counters and
1501 * synchronising the counters.
1505 * 1. m_sb_lock before picking up per-cpu locks
1506 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1507 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1508 * 4. modifying per-cpu counters requires holding per-cpu lock
1509 * 5. modifying global counters requires holding m_sb_lock
1510 * 6. enabling or disabling a counter requires holding the m_sb_lock
1511 * and _none_ of the per-cpu locks.
1513 * Disabled counters are only ever re-enabled by a balance operation
1514 * that results in more free resources per CPU than a given threshold.
1515 * To ensure counters don't remain disabled, they are rebalanced when
1516 * the global resource goes above a higher threshold (i.e. some hysteresis
1517 * is present to prevent thrashing).
1520 #ifdef CONFIG_HOTPLUG_CPU
1522 * hot-plug CPU notifier support.
1524 * We need a notifier per filesystem as we need to be able to identify
1525 * the filesystem to balance the counters out. This is achieved by
1526 * having a notifier block embedded in the xfs_mount_t and doing pointer
1527 * magic to get the mount pointer from the notifier block address.
1530 xfs_icsb_cpu_notify(
1531 struct notifier_block *nfb,
1532 unsigned long action,
1535 xfs_icsb_cnts_t *cntp;
1538 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1539 cntp = (xfs_icsb_cnts_t *)
1540 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1542 case CPU_UP_PREPARE:
1543 case CPU_UP_PREPARE_FROZEN:
1544 /* Easy Case - initialize the area and locks, and
1545 * then rebalance when online does everything else for us. */
1546 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1549 case CPU_ONLINE_FROZEN:
1551 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1552 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1553 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1554 xfs_icsb_unlock(mp);
1557 case CPU_DEAD_FROZEN:
1558 /* Disable all the counters, then fold the dead cpu's
1559 * count into the total on the global superblock and
1560 * re-enable the counters. */
1562 spin_lock(&mp->m_sb_lock);
1563 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1564 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1565 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1567 mp->m_sb.sb_icount += cntp->icsb_icount;
1568 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1569 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1571 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1573 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1574 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1575 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1576 spin_unlock(&mp->m_sb_lock);
1577 xfs_icsb_unlock(mp);
1583 #endif /* CONFIG_HOTPLUG_CPU */
1586 xfs_icsb_init_counters(
1589 xfs_icsb_cnts_t *cntp;
1592 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1593 if (mp->m_sb_cnts == NULL)
1596 for_each_online_cpu(i) {
1597 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1598 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1601 mutex_init(&mp->m_icsb_mutex);
1604 * start with all counters disabled so that the
1605 * initial balance kicks us off correctly
1607 mp->m_icsb_counters = -1;
1609 #ifdef CONFIG_HOTPLUG_CPU
1610 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1611 mp->m_icsb_notifier.priority = 0;
1612 register_hotcpu_notifier(&mp->m_icsb_notifier);
1613 #endif /* CONFIG_HOTPLUG_CPU */
1619 xfs_icsb_reinit_counters(
1624 * start with all counters disabled so that the
1625 * initial balance kicks us off correctly
1627 mp->m_icsb_counters = -1;
1628 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1629 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1630 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1631 xfs_icsb_unlock(mp);
1635 xfs_icsb_destroy_counters(
1638 if (mp->m_sb_cnts) {
1639 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1640 free_percpu(mp->m_sb_cnts);
1642 mutex_destroy(&mp->m_icsb_mutex);
1647 xfs_icsb_cnts_t *icsbp)
1649 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1655 xfs_icsb_unlock_cntr(
1656 xfs_icsb_cnts_t *icsbp)
1658 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1663 xfs_icsb_lock_all_counters(
1666 xfs_icsb_cnts_t *cntp;
1669 for_each_online_cpu(i) {
1670 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1671 xfs_icsb_lock_cntr(cntp);
1676 xfs_icsb_unlock_all_counters(
1679 xfs_icsb_cnts_t *cntp;
1682 for_each_online_cpu(i) {
1683 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1684 xfs_icsb_unlock_cntr(cntp);
1691 xfs_icsb_cnts_t *cnt,
1694 xfs_icsb_cnts_t *cntp;
1697 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1699 if (!(flags & XFS_ICSB_LAZY_COUNT))
1700 xfs_icsb_lock_all_counters(mp);
1702 for_each_online_cpu(i) {
1703 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1704 cnt->icsb_icount += cntp->icsb_icount;
1705 cnt->icsb_ifree += cntp->icsb_ifree;
1706 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1709 if (!(flags & XFS_ICSB_LAZY_COUNT))
1710 xfs_icsb_unlock_all_counters(mp);
1714 xfs_icsb_counter_disabled(
1716 xfs_sb_field_t field)
1718 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1719 return test_bit(field, &mp->m_icsb_counters);
1723 xfs_icsb_disable_counter(
1725 xfs_sb_field_t field)
1727 xfs_icsb_cnts_t cnt;
1729 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1732 * If we are already disabled, then there is nothing to do
1733 * here. We check before locking all the counters to avoid
1734 * the expensive lock operation when being called in the
1735 * slow path and the counter is already disabled. This is
1736 * safe because the only time we set or clear this state is under
1739 if (xfs_icsb_counter_disabled(mp, field))
1742 xfs_icsb_lock_all_counters(mp);
1743 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1744 /* drain back to superblock */
1746 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1748 case XFS_SBS_ICOUNT:
1749 mp->m_sb.sb_icount = cnt.icsb_icount;
1752 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1754 case XFS_SBS_FDBLOCKS:
1755 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1762 xfs_icsb_unlock_all_counters(mp);
1766 xfs_icsb_enable_counter(
1768 xfs_sb_field_t field,
1772 xfs_icsb_cnts_t *cntp;
1775 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1777 xfs_icsb_lock_all_counters(mp);
1778 for_each_online_cpu(i) {
1779 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1781 case XFS_SBS_ICOUNT:
1782 cntp->icsb_icount = count + resid;
1785 cntp->icsb_ifree = count + resid;
1787 case XFS_SBS_FDBLOCKS:
1788 cntp->icsb_fdblocks = count + resid;
1796 clear_bit(field, &mp->m_icsb_counters);
1797 xfs_icsb_unlock_all_counters(mp);
1801 xfs_icsb_sync_counters_locked(
1805 xfs_icsb_cnts_t cnt;
1807 xfs_icsb_count(mp, &cnt, flags);
1809 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1810 mp->m_sb.sb_icount = cnt.icsb_icount;
1811 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1812 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1813 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1814 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1818 * Accurate update of per-cpu counters to incore superblock
1821 xfs_icsb_sync_counters(
1825 spin_lock(&mp->m_sb_lock);
1826 xfs_icsb_sync_counters_locked(mp, flags);
1827 spin_unlock(&mp->m_sb_lock);
1831 * Balance and enable/disable counters as necessary.
1833 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1834 * chosen to be the same number as single on disk allocation chunk per CPU, and
1835 * free blocks is something far enough zero that we aren't going thrash when we
1836 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1837 * prevent looping endlessly when xfs_alloc_space asks for more than will
1838 * be distributed to a single CPU but each CPU has enough blocks to be
1841 * Note that we can be called when counters are already disabled.
1842 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1843 * prevent locking every per-cpu counter needlessly.
1846 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1847 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1848 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1850 xfs_icsb_balance_counter_locked(
1852 xfs_sb_field_t field,
1855 uint64_t count, resid;
1856 int weight = num_online_cpus();
1857 uint64_t min = (uint64_t)min_per_cpu;
1859 /* disable counter and sync counter */
1860 xfs_icsb_disable_counter(mp, field);
1862 /* update counters - first CPU gets residual*/
1864 case XFS_SBS_ICOUNT:
1865 count = mp->m_sb.sb_icount;
1866 resid = do_div(count, weight);
1867 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1871 count = mp->m_sb.sb_ifree;
1872 resid = do_div(count, weight);
1873 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1876 case XFS_SBS_FDBLOCKS:
1877 count = mp->m_sb.sb_fdblocks;
1878 resid = do_div(count, weight);
1879 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
1884 count = resid = 0; /* quiet, gcc */
1888 xfs_icsb_enable_counter(mp, field, count, resid);
1892 xfs_icsb_balance_counter(
1894 xfs_sb_field_t fields,
1897 spin_lock(&mp->m_sb_lock);
1898 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1899 spin_unlock(&mp->m_sb_lock);
1903 xfs_icsb_modify_counters(
1905 xfs_sb_field_t field,
1909 xfs_icsb_cnts_t *icsbp;
1910 long long lcounter; /* long counter for 64 bit fields */
1916 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1919 * if the counter is disabled, go to slow path
1921 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1923 xfs_icsb_lock_cntr(icsbp);
1924 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1925 xfs_icsb_unlock_cntr(icsbp);
1930 case XFS_SBS_ICOUNT:
1931 lcounter = icsbp->icsb_icount;
1933 if (unlikely(lcounter < 0))
1934 goto balance_counter;
1935 icsbp->icsb_icount = lcounter;
1939 lcounter = icsbp->icsb_ifree;
1941 if (unlikely(lcounter < 0))
1942 goto balance_counter;
1943 icsbp->icsb_ifree = lcounter;
1946 case XFS_SBS_FDBLOCKS:
1947 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1949 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1951 if (unlikely(lcounter < 0))
1952 goto balance_counter;
1953 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1959 xfs_icsb_unlock_cntr(icsbp);
1967 * serialise with a mutex so we don't burn lots of cpu on
1968 * the superblock lock. We still need to hold the superblock
1969 * lock, however, when we modify the global structures.
1974 * Now running atomically.
1976 * If the counter is enabled, someone has beaten us to rebalancing.
1977 * Drop the lock and try again in the fast path....
1979 if (!(xfs_icsb_counter_disabled(mp, field))) {
1980 xfs_icsb_unlock(mp);
1985 * The counter is currently disabled. Because we are
1986 * running atomically here, we know a rebalance cannot
1987 * be in progress. Hence we can go straight to operating
1988 * on the global superblock. We do not call xfs_mod_incore_sb()
1989 * here even though we need to get the m_sb_lock. Doing so
1990 * will cause us to re-enter this function and deadlock.
1991 * Hence we get the m_sb_lock ourselves and then call
1992 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1993 * directly on the global counters.
1995 spin_lock(&mp->m_sb_lock);
1996 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1997 spin_unlock(&mp->m_sb_lock);
2000 * Now that we've modified the global superblock, we
2001 * may be able to re-enable the distributed counters
2002 * (e.g. lots of space just got freed). After that
2006 xfs_icsb_balance_counter(mp, field, 0);
2007 xfs_icsb_unlock(mp);
2011 xfs_icsb_unlock_cntr(icsbp);
2015 * We may have multiple threads here if multiple per-cpu
2016 * counters run dry at the same time. This will mean we can
2017 * do more balances than strictly necessary but it is not
2018 * the common slowpath case.
2023 * running atomically.
2025 * This will leave the counter in the correct state for future
2026 * accesses. After the rebalance, we simply try again and our retry
2027 * will either succeed through the fast path or slow path without
2028 * another balance operation being required.
2030 xfs_icsb_balance_counter(mp, field, delta);
2031 xfs_icsb_unlock(mp);