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");
79 return XFS_ERROR(EINVAL);
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);
107 return XFS_ERROR(EINVAL);
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);
321 * Initialize the mount structure from the superblock.
323 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
324 xfs_sb_quota_from_disk(&mp->m_sb);
327 * We must be able to do sector-sized and sector-aligned IO.
329 if (sector_size > sbp->sb_sectsize) {
331 xfs_warn(mp, "device supports %u byte sectors (not %u)",
332 sector_size, sbp->sb_sectsize);
338 * Re-read the superblock so the buffer is correctly sized,
339 * and properly verified.
341 if (buf_ops == NULL) {
343 sector_size = sbp->sb_sectsize;
344 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
348 /* Initialize per-cpu counters */
349 xfs_icsb_reinit_counters(mp);
351 /* no need to be quiet anymore, so reset the buf ops */
352 bp->b_ops = &xfs_sb_buf_ops;
364 * Update alignment values based on mount options and sb values
367 xfs_update_alignment(xfs_mount_t *mp)
369 xfs_sb_t *sbp = &(mp->m_sb);
373 * If stripe unit and stripe width are not multiples
374 * of the fs blocksize turn off alignment.
376 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
377 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
379 "alignment check failed: sunit/swidth vs. blocksize(%d)",
381 return XFS_ERROR(EINVAL);
384 * Convert the stripe unit and width to FSBs.
386 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
387 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
389 "alignment check failed: sunit/swidth vs. agsize(%d)",
391 return XFS_ERROR(EINVAL);
392 } else if (mp->m_dalign) {
393 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
396 "alignment check failed: sunit(%d) less than bsize(%d)",
397 mp->m_dalign, sbp->sb_blocksize);
398 return XFS_ERROR(EINVAL);
403 * Update superblock with new values
406 if (xfs_sb_version_hasdalign(sbp)) {
407 if (sbp->sb_unit != mp->m_dalign) {
408 sbp->sb_unit = mp->m_dalign;
409 mp->m_update_flags |= XFS_SB_UNIT;
411 if (sbp->sb_width != mp->m_swidth) {
412 sbp->sb_width = mp->m_swidth;
413 mp->m_update_flags |= XFS_SB_WIDTH;
417 "cannot change alignment: superblock does not support data alignment");
418 return XFS_ERROR(EINVAL);
420 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
421 xfs_sb_version_hasdalign(&mp->m_sb)) {
422 mp->m_dalign = sbp->sb_unit;
423 mp->m_swidth = sbp->sb_width;
430 * Set the maximum inode count for this filesystem
433 xfs_set_maxicount(xfs_mount_t *mp)
435 xfs_sb_t *sbp = &(mp->m_sb);
438 if (sbp->sb_imax_pct) {
440 * Make sure the maximum inode count is a multiple
441 * of the units we allocate inodes in.
443 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
445 do_div(icount, mp->m_ialloc_blks);
446 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
454 * Set the default minimum read and write sizes unless
455 * already specified in a mount option.
456 * We use smaller I/O sizes when the file system
457 * is being used for NFS service (wsync mount option).
460 xfs_set_rw_sizes(xfs_mount_t *mp)
462 xfs_sb_t *sbp = &(mp->m_sb);
463 int readio_log, writeio_log;
465 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
466 if (mp->m_flags & XFS_MOUNT_WSYNC) {
467 readio_log = XFS_WSYNC_READIO_LOG;
468 writeio_log = XFS_WSYNC_WRITEIO_LOG;
470 readio_log = XFS_READIO_LOG_LARGE;
471 writeio_log = XFS_WRITEIO_LOG_LARGE;
474 readio_log = mp->m_readio_log;
475 writeio_log = mp->m_writeio_log;
478 if (sbp->sb_blocklog > readio_log) {
479 mp->m_readio_log = sbp->sb_blocklog;
481 mp->m_readio_log = readio_log;
483 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
484 if (sbp->sb_blocklog > writeio_log) {
485 mp->m_writeio_log = sbp->sb_blocklog;
487 mp->m_writeio_log = writeio_log;
489 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
493 * precalculate the low space thresholds for dynamic speculative preallocation.
496 xfs_set_low_space_thresholds(
497 struct xfs_mount *mp)
501 for (i = 0; i < XFS_LOWSP_MAX; i++) {
502 __uint64_t space = mp->m_sb.sb_dblocks;
505 mp->m_low_space[i] = space * (i + 1);
511 * Set whether we're using inode alignment.
514 xfs_set_inoalignment(xfs_mount_t *mp)
516 if (xfs_sb_version_hasalign(&mp->m_sb) &&
517 mp->m_sb.sb_inoalignmt >=
518 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
519 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
521 mp->m_inoalign_mask = 0;
523 * If we are using stripe alignment, check whether
524 * the stripe unit is a multiple of the inode alignment
526 if (mp->m_dalign && mp->m_inoalign_mask &&
527 !(mp->m_dalign & mp->m_inoalign_mask))
528 mp->m_sinoalign = mp->m_dalign;
534 * Check that the data (and log if separate) is an ok size.
537 xfs_check_sizes(xfs_mount_t *mp)
542 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
543 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
544 xfs_warn(mp, "filesystem size mismatch detected");
545 return XFS_ERROR(EFBIG);
547 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
548 d - XFS_FSS_TO_BB(mp, 1),
549 XFS_FSS_TO_BB(mp, 1), 0, NULL);
551 xfs_warn(mp, "last sector read failed");
556 if (mp->m_logdev_targp != mp->m_ddev_targp) {
557 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
558 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
559 xfs_warn(mp, "log size mismatch detected");
560 return XFS_ERROR(EFBIG);
562 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
563 d - XFS_FSB_TO_BB(mp, 1),
564 XFS_FSB_TO_BB(mp, 1), 0, NULL);
566 xfs_warn(mp, "log device read failed");
575 * Clear the quotaflags in memory and in the superblock.
578 xfs_mount_reset_sbqflags(
579 struct xfs_mount *mp)
582 struct xfs_trans *tp;
587 * It is OK to look at sb_qflags here in mount path,
590 if (mp->m_sb.sb_qflags == 0)
592 spin_lock(&mp->m_sb_lock);
593 mp->m_sb.sb_qflags = 0;
594 spin_unlock(&mp->m_sb_lock);
597 * If the fs is readonly, let the incore superblock run
598 * with quotas off but don't flush the update out to disk
600 if (mp->m_flags & XFS_MOUNT_RDONLY)
603 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
604 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
606 xfs_trans_cancel(tp, 0);
607 xfs_alert(mp, "%s: Superblock update failed!", __func__);
611 xfs_mod_sb(tp, XFS_SB_QFLAGS);
612 return xfs_trans_commit(tp, 0);
616 xfs_default_resblks(xfs_mount_t *mp)
621 * We default to 5% or 8192 fsbs of space reserved, whichever is
622 * smaller. This is intended to cover concurrent allocation
623 * transactions when we initially hit enospc. These each require a 4
624 * block reservation. Hence by default we cover roughly 2000 concurrent
625 * allocation reservations.
627 resblks = mp->m_sb.sb_dblocks;
629 resblks = min_t(__uint64_t, resblks, 8192);
634 * This function does the following on an initial mount of a file system:
635 * - reads the superblock from disk and init the mount struct
636 * - if we're a 32-bit kernel, do a size check on the superblock
637 * so we don't mount terabyte filesystems
638 * - init mount struct realtime fields
639 * - allocate inode hash table for fs
640 * - init directory manager
641 * - perform recovery and init the log manager
647 xfs_sb_t *sbp = &(mp->m_sb);
654 xfs_sb_mount_common(mp, sbp);
657 * Check for a mismatched features2 values. Older kernels
658 * read & wrote into the wrong sb offset for sb_features2
659 * on some platforms due to xfs_sb_t not being 64bit size aligned
660 * when sb_features2 was added, which made older superblock
661 * reading/writing routines swap it as a 64-bit value.
663 * For backwards compatibility, we make both slots equal.
665 * If we detect a mismatched field, we OR the set bits into the
666 * existing features2 field in case it has already been modified; we
667 * don't want to lose any features. We then update the bad location
668 * with the ORed value so that older kernels will see any features2
669 * flags, and mark the two fields as needing updates once the
670 * transaction subsystem is online.
672 if (xfs_sb_has_mismatched_features2(sbp)) {
673 xfs_warn(mp, "correcting sb_features alignment problem");
674 sbp->sb_features2 |= sbp->sb_bad_features2;
675 sbp->sb_bad_features2 = sbp->sb_features2;
676 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
679 * Re-check for ATTR2 in case it was found in bad_features2
682 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
683 !(mp->m_flags & XFS_MOUNT_NOATTR2))
684 mp->m_flags |= XFS_MOUNT_ATTR2;
687 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
688 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
689 xfs_sb_version_removeattr2(&mp->m_sb);
690 mp->m_update_flags |= XFS_SB_FEATURES2;
692 /* update sb_versionnum for the clearing of the morebits */
693 if (!sbp->sb_features2)
694 mp->m_update_flags |= XFS_SB_VERSIONNUM;
698 * Check if sb_agblocks is aligned at stripe boundary
699 * If sb_agblocks is NOT aligned turn off m_dalign since
700 * allocator alignment is within an ag, therefore ag has
701 * to be aligned at stripe boundary.
703 error = xfs_update_alignment(mp);
707 xfs_alloc_compute_maxlevels(mp);
708 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
709 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
710 xfs_ialloc_compute_maxlevels(mp);
712 xfs_set_maxicount(mp);
714 error = xfs_uuid_mount(mp);
719 * Set the minimum read and write sizes
721 xfs_set_rw_sizes(mp);
723 /* set the low space thresholds for dynamic preallocation */
724 xfs_set_low_space_thresholds(mp);
727 * Set the inode cluster size.
728 * This may still be overridden by the file system
729 * block size if it is larger than the chosen cluster size.
731 * For v5 filesystems, scale the cluster size with the inode size to
732 * keep a constant ratio of inode per cluster buffer, but only if mkfs
733 * has set the inode alignment value appropriately for larger cluster
736 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
737 if (xfs_sb_version_hascrc(&mp->m_sb)) {
738 int new_size = mp->m_inode_cluster_size;
740 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
741 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
742 mp->m_inode_cluster_size = new_size;
743 xfs_info(mp, "Using inode cluster size of %d bytes",
744 mp->m_inode_cluster_size);
748 * Set inode alignment fields
750 xfs_set_inoalignment(mp);
753 * Check that the data (and log if separate) is an ok size.
755 error = xfs_check_sizes(mp);
757 goto out_remove_uuid;
760 * Initialize realtime fields in the mount structure
762 error = xfs_rtmount_init(mp);
764 xfs_warn(mp, "RT mount failed");
765 goto out_remove_uuid;
769 * Copies the low order bits of the timestamp and the randomly
770 * set "sequence" number out of a UUID.
772 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
774 mp->m_dmevmask = 0; /* not persistent; set after each mount */
779 * Initialize the attribute manager's entries.
781 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
784 * Initialize the precomputed transaction reservations values.
789 * Allocate and initialize the per-ag data.
791 spin_lock_init(&mp->m_perag_lock);
792 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
793 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
795 xfs_warn(mp, "Failed per-ag init: %d", error);
796 goto out_remove_uuid;
799 if (!sbp->sb_logblocks) {
800 xfs_warn(mp, "no log defined");
801 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
802 error = XFS_ERROR(EFSCORRUPTED);
807 * log's mount-time initialization. Perform 1st part recovery if needed
809 error = xfs_log_mount(mp, mp->m_logdev_targp,
810 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
811 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
813 xfs_warn(mp, "log mount failed");
818 * Now the log is mounted, we know if it was an unclean shutdown or
819 * not. If it was, with the first phase of recovery has completed, we
820 * have consistent AG blocks on disk. We have not recovered EFIs yet,
821 * but they are recovered transactionally in the second recovery phase
824 * Hence we can safely re-initialise incore superblock counters from
825 * the per-ag data. These may not be correct if the filesystem was not
826 * cleanly unmounted, so we need to wait for recovery to finish before
829 * If the filesystem was cleanly unmounted, then we can trust the
830 * values in the superblock to be correct and we don't need to do
833 * If we are currently making the filesystem, the initialisation will
834 * fail as the perag data is in an undefined state.
836 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
837 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
838 !mp->m_sb.sb_inprogress) {
839 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
845 * Get and sanity-check the root inode.
846 * Save the pointer to it in the mount structure.
848 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
850 xfs_warn(mp, "failed to read root inode");
851 goto out_log_dealloc;
856 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
857 xfs_warn(mp, "corrupted root inode %llu: not a directory",
858 (unsigned long long)rip->i_ino);
859 xfs_iunlock(rip, XFS_ILOCK_EXCL);
860 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
862 error = XFS_ERROR(EFSCORRUPTED);
865 mp->m_rootip = rip; /* save it */
867 xfs_iunlock(rip, XFS_ILOCK_EXCL);
870 * Initialize realtime inode pointers in the mount structure
872 error = xfs_rtmount_inodes(mp);
875 * Free up the root inode.
877 xfs_warn(mp, "failed to read RT inodes");
882 * If this is a read-only mount defer the superblock updates until
883 * the next remount into writeable mode. Otherwise we would never
884 * perform the update e.g. for the root filesystem.
886 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
887 error = xfs_mount_log_sb(mp, mp->m_update_flags);
889 xfs_warn(mp, "failed to write sb changes");
895 * Initialise the XFS quota management subsystem for this mount
897 if (XFS_IS_QUOTA_RUNNING(mp)) {
898 error = xfs_qm_newmount(mp, "amount, "aflags);
902 ASSERT(!XFS_IS_QUOTA_ON(mp));
905 * If a file system had quotas running earlier, but decided to
906 * mount without -o uquota/pquota/gquota options, revoke the
907 * quotachecked license.
909 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
910 xfs_notice(mp, "resetting quota flags");
911 error = xfs_mount_reset_sbqflags(mp);
918 * Finish recovering the file system. This part needed to be
919 * delayed until after the root and real-time bitmap inodes
920 * were consistently read in.
922 error = xfs_log_mount_finish(mp);
924 xfs_warn(mp, "log mount finish failed");
929 * Complete the quota initialisation, post-log-replay component.
932 ASSERT(mp->m_qflags == 0);
933 mp->m_qflags = quotaflags;
935 xfs_qm_mount_quotas(mp);
939 * Now we are mounted, reserve a small amount of unused space for
940 * privileged transactions. This is needed so that transaction
941 * space required for critical operations can dip into this pool
942 * when at ENOSPC. This is needed for operations like create with
943 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
944 * are not allowed to use this reserved space.
946 * This may drive us straight to ENOSPC on mount, but that implies
947 * we were already there on the last unmount. Warn if this occurs.
949 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
950 resblks = xfs_default_resblks(mp);
951 error = xfs_reserve_blocks(mp, &resblks, NULL);
954 "Unable to allocate reserve blocks. Continuing without reserve pool.");
960 xfs_rtunmount_inodes(mp);
966 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
967 xfs_wait_buftarg(mp->m_logdev_targp);
968 xfs_wait_buftarg(mp->m_ddev_targp);
972 xfs_uuid_unmount(mp);
978 * This flushes out the inodes,dquots and the superblock, unmounts the
979 * log and makes sure that incore structures are freed.
983 struct xfs_mount *mp)
988 cancel_delayed_work_sync(&mp->m_eofblocks_work);
990 xfs_qm_unmount_quotas(mp);
991 xfs_rtunmount_inodes(mp);
995 * We can potentially deadlock here if we have an inode cluster
996 * that has been freed has its buffer still pinned in memory because
997 * the transaction is still sitting in a iclog. The stale inodes
998 * on that buffer will have their flush locks held until the
999 * transaction hits the disk and the callbacks run. the inode
1000 * flush takes the flush lock unconditionally and with nothing to
1001 * push out the iclog we will never get that unlocked. hence we
1002 * need to force the log first.
1004 xfs_log_force(mp, XFS_LOG_SYNC);
1007 * Flush all pending changes from the AIL.
1009 xfs_ail_push_all_sync(mp->m_ail);
1012 * And reclaim all inodes. At this point there should be no dirty
1013 * inodes and none should be pinned or locked, but use synchronous
1014 * reclaim just to be sure. We can stop background inode reclaim
1015 * here as well if it is still running.
1017 cancel_delayed_work_sync(&mp->m_reclaim_work);
1018 xfs_reclaim_inodes(mp, SYNC_WAIT);
1023 * Unreserve any blocks we have so that when we unmount we don't account
1024 * the reserved free space as used. This is really only necessary for
1025 * lazy superblock counting because it trusts the incore superblock
1026 * counters to be absolutely correct on clean unmount.
1028 * We don't bother correcting this elsewhere for lazy superblock
1029 * counting because on mount of an unclean filesystem we reconstruct the
1030 * correct counter value and this is irrelevant.
1032 * For non-lazy counter filesystems, this doesn't matter at all because
1033 * we only every apply deltas to the superblock and hence the incore
1034 * value does not matter....
1037 error = xfs_reserve_blocks(mp, &resblks, NULL);
1039 xfs_warn(mp, "Unable to free reserved block pool. "
1040 "Freespace may not be correct on next mount.");
1042 error = xfs_log_sbcount(mp);
1044 xfs_warn(mp, "Unable to update superblock counters. "
1045 "Freespace may not be correct on next mount.");
1047 xfs_log_unmount(mp);
1048 xfs_uuid_unmount(mp);
1051 xfs_errortag_clearall(mp, 0);
1057 xfs_fs_writable(xfs_mount_t *mp)
1059 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1060 (mp->m_flags & XFS_MOUNT_RDONLY));
1066 * Sync the superblock counters to disk.
1068 * Note this code can be called during the process of freezing, so
1069 * we may need to use the transaction allocator which does not
1070 * block when the transaction subsystem is in its frozen state.
1073 xfs_log_sbcount(xfs_mount_t *mp)
1078 if (!xfs_fs_writable(mp))
1081 xfs_icsb_sync_counters(mp, 0);
1084 * we don't need to do this if we are updating the superblock
1085 * counters on every modification.
1087 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1090 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1091 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1093 xfs_trans_cancel(tp, 0);
1097 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1098 xfs_trans_set_sync(tp);
1099 error = xfs_trans_commit(tp, 0);
1104 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1105 * a delta to a specified field in the in-core superblock. Simply
1106 * switch on the field indicated and apply the delta to that field.
1107 * Fields are not allowed to dip below zero, so if the delta would
1108 * do this do not apply it and return EINVAL.
1110 * The m_sb_lock must be held when this routine is called.
1113 xfs_mod_incore_sb_unlocked(
1115 xfs_sb_field_t field,
1119 int scounter; /* short counter for 32 bit fields */
1120 long long lcounter; /* long counter for 64 bit fields */
1121 long long res_used, rem;
1124 * With the in-core superblock spin lock held, switch
1125 * on the indicated field. Apply the delta to the
1126 * proper field. If the fields value would dip below
1127 * 0, then do not apply the delta and return EINVAL.
1130 case XFS_SBS_ICOUNT:
1131 lcounter = (long long)mp->m_sb.sb_icount;
1135 return XFS_ERROR(EINVAL);
1137 mp->m_sb.sb_icount = lcounter;
1140 lcounter = (long long)mp->m_sb.sb_ifree;
1144 return XFS_ERROR(EINVAL);
1146 mp->m_sb.sb_ifree = lcounter;
1148 case XFS_SBS_FDBLOCKS:
1149 lcounter = (long long)
1150 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1151 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1153 if (delta > 0) { /* Putting blocks back */
1154 if (res_used > delta) {
1155 mp->m_resblks_avail += delta;
1157 rem = delta - res_used;
1158 mp->m_resblks_avail = mp->m_resblks;
1161 } else { /* Taking blocks away */
1163 if (lcounter >= 0) {
1164 mp->m_sb.sb_fdblocks = lcounter +
1165 XFS_ALLOC_SET_ASIDE(mp);
1170 * We are out of blocks, use any available reserved
1171 * blocks if were allowed to.
1174 return XFS_ERROR(ENOSPC);
1176 lcounter = (long long)mp->m_resblks_avail + delta;
1177 if (lcounter >= 0) {
1178 mp->m_resblks_avail = lcounter;
1181 printk_once(KERN_WARNING
1182 "Filesystem \"%s\": reserve blocks depleted! "
1183 "Consider increasing reserve pool size.",
1185 return XFS_ERROR(ENOSPC);
1188 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1190 case XFS_SBS_FREXTENTS:
1191 lcounter = (long long)mp->m_sb.sb_frextents;
1194 return XFS_ERROR(ENOSPC);
1196 mp->m_sb.sb_frextents = lcounter;
1198 case XFS_SBS_DBLOCKS:
1199 lcounter = (long long)mp->m_sb.sb_dblocks;
1203 return XFS_ERROR(EINVAL);
1205 mp->m_sb.sb_dblocks = lcounter;
1207 case XFS_SBS_AGCOUNT:
1208 scounter = mp->m_sb.sb_agcount;
1212 return XFS_ERROR(EINVAL);
1214 mp->m_sb.sb_agcount = scounter;
1216 case XFS_SBS_IMAX_PCT:
1217 scounter = mp->m_sb.sb_imax_pct;
1221 return XFS_ERROR(EINVAL);
1223 mp->m_sb.sb_imax_pct = scounter;
1225 case XFS_SBS_REXTSIZE:
1226 scounter = mp->m_sb.sb_rextsize;
1230 return XFS_ERROR(EINVAL);
1232 mp->m_sb.sb_rextsize = scounter;
1234 case XFS_SBS_RBMBLOCKS:
1235 scounter = mp->m_sb.sb_rbmblocks;
1239 return XFS_ERROR(EINVAL);
1241 mp->m_sb.sb_rbmblocks = scounter;
1243 case XFS_SBS_RBLOCKS:
1244 lcounter = (long long)mp->m_sb.sb_rblocks;
1248 return XFS_ERROR(EINVAL);
1250 mp->m_sb.sb_rblocks = lcounter;
1252 case XFS_SBS_REXTENTS:
1253 lcounter = (long long)mp->m_sb.sb_rextents;
1257 return XFS_ERROR(EINVAL);
1259 mp->m_sb.sb_rextents = lcounter;
1261 case XFS_SBS_REXTSLOG:
1262 scounter = mp->m_sb.sb_rextslog;
1266 return XFS_ERROR(EINVAL);
1268 mp->m_sb.sb_rextslog = scounter;
1272 return XFS_ERROR(EINVAL);
1277 * xfs_mod_incore_sb() is used to change a field in the in-core
1278 * superblock structure by the specified delta. This modification
1279 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1280 * routine to do the work.
1284 struct xfs_mount *mp,
1285 xfs_sb_field_t field,
1291 #ifdef HAVE_PERCPU_SB
1292 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1294 spin_lock(&mp->m_sb_lock);
1295 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1296 spin_unlock(&mp->m_sb_lock);
1302 * Change more than one field in the in-core superblock structure at a time.
1304 * The fields and changes to those fields are specified in the array of
1305 * xfs_mod_sb structures passed in. Either all of the specified deltas
1306 * will be applied or none of them will. If any modified field dips below 0,
1307 * then all modifications will be backed out and EINVAL will be returned.
1309 * Note that this function may not be used for the superblock values that
1310 * are tracked with the in-memory per-cpu counters - a direct call to
1311 * xfs_icsb_modify_counters is required for these.
1314 xfs_mod_incore_sb_batch(
1315 struct xfs_mount *mp,
1324 * Loop through the array of mod structures and apply each individually.
1325 * If any fail, then back out all those which have already been applied.
1326 * Do all of this within the scope of the m_sb_lock so that all of the
1327 * changes will be atomic.
1329 spin_lock(&mp->m_sb_lock);
1330 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1331 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1332 msbp->msb_field > XFS_SBS_FDBLOCKS);
1334 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1335 msbp->msb_delta, rsvd);
1339 spin_unlock(&mp->m_sb_lock);
1343 while (--msbp >= msb) {
1344 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1345 -msbp->msb_delta, rsvd);
1348 spin_unlock(&mp->m_sb_lock);
1353 * xfs_getsb() is called to obtain the buffer for the superblock.
1354 * The buffer is returned locked and read in from disk.
1355 * The buffer should be released with a call to xfs_brelse().
1357 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1358 * the superblock buffer if it can be locked without sleeping.
1359 * If it can't then we'll return NULL.
1363 struct xfs_mount *mp,
1366 struct xfs_buf *bp = mp->m_sb_bp;
1368 if (!xfs_buf_trylock(bp)) {
1369 if (flags & XBF_TRYLOCK)
1375 ASSERT(XFS_BUF_ISDONE(bp));
1380 * Used to free the superblock along various error paths.
1384 struct xfs_mount *mp)
1386 struct xfs_buf *bp = mp->m_sb_bp;
1394 * Used to log changes to the superblock unit and width fields which could
1395 * be altered by the mount options, as well as any potential sb_features2
1396 * fixup. Only the first superblock is updated.
1406 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1407 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1408 XFS_SB_VERSIONNUM));
1410 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1411 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1413 xfs_trans_cancel(tp, 0);
1416 xfs_mod_sb(tp, fields);
1417 error = xfs_trans_commit(tp, 0);
1422 * If the underlying (data/log/rt) device is readonly, there are some
1423 * operations that cannot proceed.
1426 xfs_dev_is_read_only(
1427 struct xfs_mount *mp,
1430 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1431 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1432 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1433 xfs_notice(mp, "%s required on read-only device.", message);
1434 xfs_notice(mp, "write access unavailable, cannot proceed.");
1440 #ifdef HAVE_PERCPU_SB
1442 * Per-cpu incore superblock counters
1444 * Simple concept, difficult implementation
1446 * Basically, replace the incore superblock counters with a distributed per cpu
1447 * counter for contended fields (e.g. free block count).
1449 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1450 * hence needs to be accurately read when we are running low on space. Hence
1451 * there is a method to enable and disable the per-cpu counters based on how
1452 * much "stuff" is available in them.
1454 * Basically, a counter is enabled if there is enough free resource to justify
1455 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1456 * ENOSPC), then we disable the counters to synchronise all callers and
1457 * re-distribute the available resources.
1459 * If, once we redistributed the available resources, we still get a failure,
1460 * we disable the per-cpu counter and go through the slow path.
1462 * The slow path is the current xfs_mod_incore_sb() function. This means that
1463 * when we disable a per-cpu counter, we need to drain its resources back to
1464 * the global superblock. We do this after disabling the counter to prevent
1465 * more threads from queueing up on the counter.
1467 * Essentially, this means that we still need a lock in the fast path to enable
1468 * synchronisation between the global counters and the per-cpu counters. This
1469 * is not a problem because the lock will be local to a CPU almost all the time
1470 * and have little contention except when we get to ENOSPC conditions.
1472 * Basically, this lock becomes a barrier that enables us to lock out the fast
1473 * path while we do things like enabling and disabling counters and
1474 * synchronising the counters.
1478 * 1. m_sb_lock before picking up per-cpu locks
1479 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1480 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1481 * 4. modifying per-cpu counters requires holding per-cpu lock
1482 * 5. modifying global counters requires holding m_sb_lock
1483 * 6. enabling or disabling a counter requires holding the m_sb_lock
1484 * and _none_ of the per-cpu locks.
1486 * Disabled counters are only ever re-enabled by a balance operation
1487 * that results in more free resources per CPU than a given threshold.
1488 * To ensure counters don't remain disabled, they are rebalanced when
1489 * the global resource goes above a higher threshold (i.e. some hysteresis
1490 * is present to prevent thrashing).
1493 #ifdef CONFIG_HOTPLUG_CPU
1495 * hot-plug CPU notifier support.
1497 * We need a notifier per filesystem as we need to be able to identify
1498 * the filesystem to balance the counters out. This is achieved by
1499 * having a notifier block embedded in the xfs_mount_t and doing pointer
1500 * magic to get the mount pointer from the notifier block address.
1503 xfs_icsb_cpu_notify(
1504 struct notifier_block *nfb,
1505 unsigned long action,
1508 xfs_icsb_cnts_t *cntp;
1511 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1512 cntp = (xfs_icsb_cnts_t *)
1513 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1515 case CPU_UP_PREPARE:
1516 case CPU_UP_PREPARE_FROZEN:
1517 /* Easy Case - initialize the area and locks, and
1518 * then rebalance when online does everything else for us. */
1519 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1522 case CPU_ONLINE_FROZEN:
1524 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1525 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1526 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1527 xfs_icsb_unlock(mp);
1530 case CPU_DEAD_FROZEN:
1531 /* Disable all the counters, then fold the dead cpu's
1532 * count into the total on the global superblock and
1533 * re-enable the counters. */
1535 spin_lock(&mp->m_sb_lock);
1536 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1537 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1538 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1540 mp->m_sb.sb_icount += cntp->icsb_icount;
1541 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1542 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1544 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1546 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1547 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1548 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1549 spin_unlock(&mp->m_sb_lock);
1550 xfs_icsb_unlock(mp);
1556 #endif /* CONFIG_HOTPLUG_CPU */
1559 xfs_icsb_init_counters(
1562 xfs_icsb_cnts_t *cntp;
1565 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1566 if (mp->m_sb_cnts == NULL)
1569 for_each_online_cpu(i) {
1570 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1571 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1574 mutex_init(&mp->m_icsb_mutex);
1577 * start with all counters disabled so that the
1578 * initial balance kicks us off correctly
1580 mp->m_icsb_counters = -1;
1582 #ifdef CONFIG_HOTPLUG_CPU
1583 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1584 mp->m_icsb_notifier.priority = 0;
1585 register_hotcpu_notifier(&mp->m_icsb_notifier);
1586 #endif /* CONFIG_HOTPLUG_CPU */
1592 xfs_icsb_reinit_counters(
1597 * start with all counters disabled so that the
1598 * initial balance kicks us off correctly
1600 mp->m_icsb_counters = -1;
1601 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1602 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1603 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1604 xfs_icsb_unlock(mp);
1608 xfs_icsb_destroy_counters(
1611 if (mp->m_sb_cnts) {
1612 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1613 free_percpu(mp->m_sb_cnts);
1615 mutex_destroy(&mp->m_icsb_mutex);
1620 xfs_icsb_cnts_t *icsbp)
1622 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1628 xfs_icsb_unlock_cntr(
1629 xfs_icsb_cnts_t *icsbp)
1631 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1636 xfs_icsb_lock_all_counters(
1639 xfs_icsb_cnts_t *cntp;
1642 for_each_online_cpu(i) {
1643 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1644 xfs_icsb_lock_cntr(cntp);
1649 xfs_icsb_unlock_all_counters(
1652 xfs_icsb_cnts_t *cntp;
1655 for_each_online_cpu(i) {
1656 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1657 xfs_icsb_unlock_cntr(cntp);
1664 xfs_icsb_cnts_t *cnt,
1667 xfs_icsb_cnts_t *cntp;
1670 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1672 if (!(flags & XFS_ICSB_LAZY_COUNT))
1673 xfs_icsb_lock_all_counters(mp);
1675 for_each_online_cpu(i) {
1676 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1677 cnt->icsb_icount += cntp->icsb_icount;
1678 cnt->icsb_ifree += cntp->icsb_ifree;
1679 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1682 if (!(flags & XFS_ICSB_LAZY_COUNT))
1683 xfs_icsb_unlock_all_counters(mp);
1687 xfs_icsb_counter_disabled(
1689 xfs_sb_field_t field)
1691 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1692 return test_bit(field, &mp->m_icsb_counters);
1696 xfs_icsb_disable_counter(
1698 xfs_sb_field_t field)
1700 xfs_icsb_cnts_t cnt;
1702 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1705 * If we are already disabled, then there is nothing to do
1706 * here. We check before locking all the counters to avoid
1707 * the expensive lock operation when being called in the
1708 * slow path and the counter is already disabled. This is
1709 * safe because the only time we set or clear this state is under
1712 if (xfs_icsb_counter_disabled(mp, field))
1715 xfs_icsb_lock_all_counters(mp);
1716 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1717 /* drain back to superblock */
1719 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1721 case XFS_SBS_ICOUNT:
1722 mp->m_sb.sb_icount = cnt.icsb_icount;
1725 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1727 case XFS_SBS_FDBLOCKS:
1728 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1735 xfs_icsb_unlock_all_counters(mp);
1739 xfs_icsb_enable_counter(
1741 xfs_sb_field_t field,
1745 xfs_icsb_cnts_t *cntp;
1748 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1750 xfs_icsb_lock_all_counters(mp);
1751 for_each_online_cpu(i) {
1752 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1754 case XFS_SBS_ICOUNT:
1755 cntp->icsb_icount = count + resid;
1758 cntp->icsb_ifree = count + resid;
1760 case XFS_SBS_FDBLOCKS:
1761 cntp->icsb_fdblocks = count + resid;
1769 clear_bit(field, &mp->m_icsb_counters);
1770 xfs_icsb_unlock_all_counters(mp);
1774 xfs_icsb_sync_counters_locked(
1778 xfs_icsb_cnts_t cnt;
1780 xfs_icsb_count(mp, &cnt, flags);
1782 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1783 mp->m_sb.sb_icount = cnt.icsb_icount;
1784 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1785 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1786 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1787 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1791 * Accurate update of per-cpu counters to incore superblock
1794 xfs_icsb_sync_counters(
1798 spin_lock(&mp->m_sb_lock);
1799 xfs_icsb_sync_counters_locked(mp, flags);
1800 spin_unlock(&mp->m_sb_lock);
1804 * Balance and enable/disable counters as necessary.
1806 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1807 * chosen to be the same number as single on disk allocation chunk per CPU, and
1808 * free blocks is something far enough zero that we aren't going thrash when we
1809 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1810 * prevent looping endlessly when xfs_alloc_space asks for more than will
1811 * be distributed to a single CPU but each CPU has enough blocks to be
1814 * Note that we can be called when counters are already disabled.
1815 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1816 * prevent locking every per-cpu counter needlessly.
1819 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1820 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1821 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1823 xfs_icsb_balance_counter_locked(
1825 xfs_sb_field_t field,
1828 uint64_t count, resid;
1829 int weight = num_online_cpus();
1830 uint64_t min = (uint64_t)min_per_cpu;
1832 /* disable counter and sync counter */
1833 xfs_icsb_disable_counter(mp, field);
1835 /* update counters - first CPU gets residual*/
1837 case XFS_SBS_ICOUNT:
1838 count = mp->m_sb.sb_icount;
1839 resid = do_div(count, weight);
1840 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1844 count = mp->m_sb.sb_ifree;
1845 resid = do_div(count, weight);
1846 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1849 case XFS_SBS_FDBLOCKS:
1850 count = mp->m_sb.sb_fdblocks;
1851 resid = do_div(count, weight);
1852 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
1857 count = resid = 0; /* quiet, gcc */
1861 xfs_icsb_enable_counter(mp, field, count, resid);
1865 xfs_icsb_balance_counter(
1867 xfs_sb_field_t fields,
1870 spin_lock(&mp->m_sb_lock);
1871 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1872 spin_unlock(&mp->m_sb_lock);
1876 xfs_icsb_modify_counters(
1878 xfs_sb_field_t field,
1882 xfs_icsb_cnts_t *icsbp;
1883 long long lcounter; /* long counter for 64 bit fields */
1889 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1892 * if the counter is disabled, go to slow path
1894 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1896 xfs_icsb_lock_cntr(icsbp);
1897 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1898 xfs_icsb_unlock_cntr(icsbp);
1903 case XFS_SBS_ICOUNT:
1904 lcounter = icsbp->icsb_icount;
1906 if (unlikely(lcounter < 0))
1907 goto balance_counter;
1908 icsbp->icsb_icount = lcounter;
1912 lcounter = icsbp->icsb_ifree;
1914 if (unlikely(lcounter < 0))
1915 goto balance_counter;
1916 icsbp->icsb_ifree = lcounter;
1919 case XFS_SBS_FDBLOCKS:
1920 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1922 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1924 if (unlikely(lcounter < 0))
1925 goto balance_counter;
1926 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1932 xfs_icsb_unlock_cntr(icsbp);
1940 * serialise with a mutex so we don't burn lots of cpu on
1941 * the superblock lock. We still need to hold the superblock
1942 * lock, however, when we modify the global structures.
1947 * Now running atomically.
1949 * If the counter is enabled, someone has beaten us to rebalancing.
1950 * Drop the lock and try again in the fast path....
1952 if (!(xfs_icsb_counter_disabled(mp, field))) {
1953 xfs_icsb_unlock(mp);
1958 * The counter is currently disabled. Because we are
1959 * running atomically here, we know a rebalance cannot
1960 * be in progress. Hence we can go straight to operating
1961 * on the global superblock. We do not call xfs_mod_incore_sb()
1962 * here even though we need to get the m_sb_lock. Doing so
1963 * will cause us to re-enter this function and deadlock.
1964 * Hence we get the m_sb_lock ourselves and then call
1965 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1966 * directly on the global counters.
1968 spin_lock(&mp->m_sb_lock);
1969 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1970 spin_unlock(&mp->m_sb_lock);
1973 * Now that we've modified the global superblock, we
1974 * may be able to re-enable the distributed counters
1975 * (e.g. lots of space just got freed). After that
1979 xfs_icsb_balance_counter(mp, field, 0);
1980 xfs_icsb_unlock(mp);
1984 xfs_icsb_unlock_cntr(icsbp);
1988 * We may have multiple threads here if multiple per-cpu
1989 * counters run dry at the same time. This will mean we can
1990 * do more balances than strictly necessary but it is not
1991 * the common slowpath case.
1996 * running atomically.
1998 * This will leave the counter in the correct state for future
1999 * accesses. After the rebalance, we simply try again and our retry
2000 * will either succeed through the fast path or slow path without
2001 * another balance operation being required.
2003 xfs_icsb_balance_counter(mp, field, delta);
2004 xfs_icsb_unlock(mp);