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);
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));
327 xfs_sb_quota_from_disk(sbp);
330 * If we haven't validated the superblock, do so now before we try
331 * to check the sector size and reread the superblock appropriately.
333 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
335 xfs_warn(mp, "Invalid superblock magic number");
341 * We must be able to do sector-sized and sector-aligned IO.
343 if (sector_size > sbp->sb_sectsize) {
345 xfs_warn(mp, "device supports %u byte sectors (not %u)",
346 sector_size, sbp->sb_sectsize);
351 if (buf_ops == NULL) {
353 * Re-read the superblock so the buffer is correctly sized,
354 * and properly verified.
357 sector_size = sbp->sb_sectsize;
358 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
362 /* Initialize per-cpu counters */
363 xfs_icsb_reinit_counters(mp);
365 /* no need to be quiet anymore, so reset the buf ops */
366 bp->b_ops = &xfs_sb_buf_ops;
378 * Update alignment values based on mount options and sb values
381 xfs_update_alignment(xfs_mount_t *mp)
383 xfs_sb_t *sbp = &(mp->m_sb);
387 * If stripe unit and stripe width are not multiples
388 * of the fs blocksize turn off alignment.
390 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
391 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
393 "alignment check failed: sunit/swidth vs. blocksize(%d)",
395 return XFS_ERROR(EINVAL);
398 * Convert the stripe unit and width to FSBs.
400 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
401 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
403 "alignment check failed: sunit/swidth vs. agsize(%d)",
405 return XFS_ERROR(EINVAL);
406 } else if (mp->m_dalign) {
407 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
410 "alignment check failed: sunit(%d) less than bsize(%d)",
411 mp->m_dalign, sbp->sb_blocksize);
412 return XFS_ERROR(EINVAL);
417 * Update superblock with new values
420 if (xfs_sb_version_hasdalign(sbp)) {
421 if (sbp->sb_unit != mp->m_dalign) {
422 sbp->sb_unit = mp->m_dalign;
423 mp->m_update_flags |= XFS_SB_UNIT;
425 if (sbp->sb_width != mp->m_swidth) {
426 sbp->sb_width = mp->m_swidth;
427 mp->m_update_flags |= XFS_SB_WIDTH;
431 "cannot change alignment: superblock does not support data alignment");
432 return XFS_ERROR(EINVAL);
434 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
435 xfs_sb_version_hasdalign(&mp->m_sb)) {
436 mp->m_dalign = sbp->sb_unit;
437 mp->m_swidth = sbp->sb_width;
444 * Set the maximum inode count for this filesystem
447 xfs_set_maxicount(xfs_mount_t *mp)
449 xfs_sb_t *sbp = &(mp->m_sb);
452 if (sbp->sb_imax_pct) {
454 * Make sure the maximum inode count is a multiple
455 * of the units we allocate inodes in.
457 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
459 do_div(icount, mp->m_ialloc_blks);
460 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
468 * Set the default minimum read and write sizes unless
469 * already specified in a mount option.
470 * We use smaller I/O sizes when the file system
471 * is being used for NFS service (wsync mount option).
474 xfs_set_rw_sizes(xfs_mount_t *mp)
476 xfs_sb_t *sbp = &(mp->m_sb);
477 int readio_log, writeio_log;
479 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
480 if (mp->m_flags & XFS_MOUNT_WSYNC) {
481 readio_log = XFS_WSYNC_READIO_LOG;
482 writeio_log = XFS_WSYNC_WRITEIO_LOG;
484 readio_log = XFS_READIO_LOG_LARGE;
485 writeio_log = XFS_WRITEIO_LOG_LARGE;
488 readio_log = mp->m_readio_log;
489 writeio_log = mp->m_writeio_log;
492 if (sbp->sb_blocklog > readio_log) {
493 mp->m_readio_log = sbp->sb_blocklog;
495 mp->m_readio_log = readio_log;
497 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
498 if (sbp->sb_blocklog > writeio_log) {
499 mp->m_writeio_log = sbp->sb_blocklog;
501 mp->m_writeio_log = writeio_log;
503 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
507 * precalculate the low space thresholds for dynamic speculative preallocation.
510 xfs_set_low_space_thresholds(
511 struct xfs_mount *mp)
515 for (i = 0; i < XFS_LOWSP_MAX; i++) {
516 __uint64_t space = mp->m_sb.sb_dblocks;
519 mp->m_low_space[i] = space * (i + 1);
525 * Set whether we're using inode alignment.
528 xfs_set_inoalignment(xfs_mount_t *mp)
530 if (xfs_sb_version_hasalign(&mp->m_sb) &&
531 mp->m_sb.sb_inoalignmt >=
532 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
533 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
535 mp->m_inoalign_mask = 0;
537 * If we are using stripe alignment, check whether
538 * the stripe unit is a multiple of the inode alignment
540 if (mp->m_dalign && mp->m_inoalign_mask &&
541 !(mp->m_dalign & mp->m_inoalign_mask))
542 mp->m_sinoalign = mp->m_dalign;
548 * Check that the data (and log if separate) is an ok size.
551 xfs_check_sizes(xfs_mount_t *mp)
556 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
557 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
558 xfs_warn(mp, "filesystem size mismatch detected");
559 return XFS_ERROR(EFBIG);
561 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
562 d - XFS_FSS_TO_BB(mp, 1),
563 XFS_FSS_TO_BB(mp, 1), 0, NULL);
565 xfs_warn(mp, "last sector read failed");
570 if (mp->m_logdev_targp != mp->m_ddev_targp) {
571 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
572 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
573 xfs_warn(mp, "log size mismatch detected");
574 return XFS_ERROR(EFBIG);
576 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
577 d - XFS_FSB_TO_BB(mp, 1),
578 XFS_FSB_TO_BB(mp, 1), 0, NULL);
580 xfs_warn(mp, "log device read failed");
589 * Clear the quotaflags in memory and in the superblock.
592 xfs_mount_reset_sbqflags(
593 struct xfs_mount *mp)
596 struct xfs_trans *tp;
601 * It is OK to look at sb_qflags here in mount path,
604 if (mp->m_sb.sb_qflags == 0)
606 spin_lock(&mp->m_sb_lock);
607 mp->m_sb.sb_qflags = 0;
608 spin_unlock(&mp->m_sb_lock);
611 * If the fs is readonly, let the incore superblock run
612 * with quotas off but don't flush the update out to disk
614 if (mp->m_flags & XFS_MOUNT_RDONLY)
617 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
618 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
620 xfs_trans_cancel(tp, 0);
621 xfs_alert(mp, "%s: Superblock update failed!", __func__);
625 xfs_mod_sb(tp, XFS_SB_QFLAGS);
626 return xfs_trans_commit(tp, 0);
630 xfs_default_resblks(xfs_mount_t *mp)
635 * We default to 5% or 8192 fsbs of space reserved, whichever is
636 * smaller. This is intended to cover concurrent allocation
637 * transactions when we initially hit enospc. These each require a 4
638 * block reservation. Hence by default we cover roughly 2000 concurrent
639 * allocation reservations.
641 resblks = mp->m_sb.sb_dblocks;
643 resblks = min_t(__uint64_t, resblks, 8192);
648 * This function does the following on an initial mount of a file system:
649 * - reads the superblock from disk and init the mount struct
650 * - if we're a 32-bit kernel, do a size check on the superblock
651 * so we don't mount terabyte filesystems
652 * - init mount struct realtime fields
653 * - allocate inode hash table for fs
654 * - init directory manager
655 * - perform recovery and init the log manager
661 xfs_sb_t *sbp = &(mp->m_sb);
668 xfs_sb_mount_common(mp, sbp);
671 * Check for a mismatched features2 values. Older kernels
672 * read & wrote into the wrong sb offset for sb_features2
673 * on some platforms due to xfs_sb_t not being 64bit size aligned
674 * when sb_features2 was added, which made older superblock
675 * reading/writing routines swap it as a 64-bit value.
677 * For backwards compatibility, we make both slots equal.
679 * If we detect a mismatched field, we OR the set bits into the
680 * existing features2 field in case it has already been modified; we
681 * don't want to lose any features. We then update the bad location
682 * with the ORed value so that older kernels will see any features2
683 * flags, and mark the two fields as needing updates once the
684 * transaction subsystem is online.
686 if (xfs_sb_has_mismatched_features2(sbp)) {
687 xfs_warn(mp, "correcting sb_features alignment problem");
688 sbp->sb_features2 |= sbp->sb_bad_features2;
689 sbp->sb_bad_features2 = sbp->sb_features2;
690 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
693 * Re-check for ATTR2 in case it was found in bad_features2
696 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
697 !(mp->m_flags & XFS_MOUNT_NOATTR2))
698 mp->m_flags |= XFS_MOUNT_ATTR2;
701 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
702 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
703 xfs_sb_version_removeattr2(&mp->m_sb);
704 mp->m_update_flags |= XFS_SB_FEATURES2;
706 /* update sb_versionnum for the clearing of the morebits */
707 if (!sbp->sb_features2)
708 mp->m_update_flags |= XFS_SB_VERSIONNUM;
711 /* always use v2 inodes by default now */
712 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
713 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
714 mp->m_update_flags |= XFS_SB_VERSIONNUM;
718 * Check if sb_agblocks is aligned at stripe boundary
719 * If sb_agblocks is NOT aligned turn off m_dalign since
720 * allocator alignment is within an ag, therefore ag has
721 * to be aligned at stripe boundary.
723 error = xfs_update_alignment(mp);
727 xfs_alloc_compute_maxlevels(mp);
728 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
729 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
730 xfs_ialloc_compute_maxlevels(mp);
732 xfs_set_maxicount(mp);
734 error = xfs_uuid_mount(mp);
739 * Set the minimum read and write sizes
741 xfs_set_rw_sizes(mp);
743 /* set the low space thresholds for dynamic preallocation */
744 xfs_set_low_space_thresholds(mp);
747 * Set the inode cluster size.
748 * This may still be overridden by the file system
749 * block size if it is larger than the chosen cluster size.
751 * For v5 filesystems, scale the cluster size with the inode size to
752 * keep a constant ratio of inode per cluster buffer, but only if mkfs
753 * has set the inode alignment value appropriately for larger cluster
756 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
757 if (xfs_sb_version_hascrc(&mp->m_sb)) {
758 int new_size = mp->m_inode_cluster_size;
760 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
761 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
762 mp->m_inode_cluster_size = new_size;
766 * Set inode alignment fields
768 xfs_set_inoalignment(mp);
771 * Check that the data (and log if separate) is an ok size.
773 error = xfs_check_sizes(mp);
775 goto out_remove_uuid;
778 * Initialize realtime fields in the mount structure
780 error = xfs_rtmount_init(mp);
782 xfs_warn(mp, "RT mount failed");
783 goto out_remove_uuid;
787 * Copies the low order bits of the timestamp and the randomly
788 * set "sequence" number out of a UUID.
790 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
792 mp->m_dmevmask = 0; /* not persistent; set after each mount */
794 error = xfs_da_mount(mp);
796 xfs_warn(mp, "Failed dir/attr init: %d", error);
797 goto out_remove_uuid;
801 * Initialize the precomputed transaction reservations values.
806 * Allocate and initialize the per-ag data.
808 spin_lock_init(&mp->m_perag_lock);
809 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
810 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
812 xfs_warn(mp, "Failed per-ag init: %d", error);
816 if (!sbp->sb_logblocks) {
817 xfs_warn(mp, "no log defined");
818 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
819 error = XFS_ERROR(EFSCORRUPTED);
824 * log's mount-time initialization. Perform 1st part recovery if needed
826 error = xfs_log_mount(mp, mp->m_logdev_targp,
827 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
828 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
830 xfs_warn(mp, "log mount failed");
835 * Now the log is mounted, we know if it was an unclean shutdown or
836 * not. If it was, with the first phase of recovery has completed, we
837 * have consistent AG blocks on disk. We have not recovered EFIs yet,
838 * but they are recovered transactionally in the second recovery phase
841 * Hence we can safely re-initialise incore superblock counters from
842 * the per-ag data. These may not be correct if the filesystem was not
843 * cleanly unmounted, so we need to wait for recovery to finish before
846 * If the filesystem was cleanly unmounted, then we can trust the
847 * values in the superblock to be correct and we don't need to do
850 * If we are currently making the filesystem, the initialisation will
851 * fail as the perag data is in an undefined state.
853 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
854 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
855 !mp->m_sb.sb_inprogress) {
856 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
862 * Get and sanity-check the root inode.
863 * Save the pointer to it in the mount structure.
865 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
867 xfs_warn(mp, "failed to read root inode");
868 goto out_log_dealloc;
873 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
874 xfs_warn(mp, "corrupted root inode %llu: not a directory",
875 (unsigned long long)rip->i_ino);
876 xfs_iunlock(rip, XFS_ILOCK_EXCL);
877 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
879 error = XFS_ERROR(EFSCORRUPTED);
882 mp->m_rootip = rip; /* save it */
884 xfs_iunlock(rip, XFS_ILOCK_EXCL);
887 * Initialize realtime inode pointers in the mount structure
889 error = xfs_rtmount_inodes(mp);
892 * Free up the root inode.
894 xfs_warn(mp, "failed to read RT inodes");
899 * If this is a read-only mount defer the superblock updates until
900 * the next remount into writeable mode. Otherwise we would never
901 * perform the update e.g. for the root filesystem.
903 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
904 error = xfs_mount_log_sb(mp, mp->m_update_flags);
906 xfs_warn(mp, "failed to write sb changes");
912 * Initialise the XFS quota management subsystem for this mount
914 if (XFS_IS_QUOTA_RUNNING(mp)) {
915 error = xfs_qm_newmount(mp, "amount, "aflags);
919 ASSERT(!XFS_IS_QUOTA_ON(mp));
922 * If a file system had quotas running earlier, but decided to
923 * mount without -o uquota/pquota/gquota options, revoke the
924 * quotachecked license.
926 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
927 xfs_notice(mp, "resetting quota flags");
928 error = xfs_mount_reset_sbqflags(mp);
935 * Finish recovering the file system. This part needed to be
936 * delayed until after the root and real-time bitmap inodes
937 * were consistently read in.
939 error = xfs_log_mount_finish(mp);
941 xfs_warn(mp, "log mount finish failed");
946 * Complete the quota initialisation, post-log-replay component.
949 ASSERT(mp->m_qflags == 0);
950 mp->m_qflags = quotaflags;
952 xfs_qm_mount_quotas(mp);
956 * Now we are mounted, reserve a small amount of unused space for
957 * privileged transactions. This is needed so that transaction
958 * space required for critical operations can dip into this pool
959 * when at ENOSPC. This is needed for operations like create with
960 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
961 * are not allowed to use this reserved space.
963 * This may drive us straight to ENOSPC on mount, but that implies
964 * we were already there on the last unmount. Warn if this occurs.
966 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
967 resblks = xfs_default_resblks(mp);
968 error = xfs_reserve_blocks(mp, &resblks, NULL);
971 "Unable to allocate reserve blocks. Continuing without reserve pool.");
977 xfs_rtunmount_inodes(mp);
983 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
984 xfs_wait_buftarg(mp->m_logdev_targp);
985 xfs_wait_buftarg(mp->m_ddev_targp);
991 xfs_uuid_unmount(mp);
997 * This flushes out the inodes,dquots and the superblock, unmounts the
998 * log and makes sure that incore structures are freed.
1002 struct xfs_mount *mp)
1007 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1009 xfs_qm_unmount_quotas(mp);
1010 xfs_rtunmount_inodes(mp);
1011 IRELE(mp->m_rootip);
1014 * We can potentially deadlock here if we have an inode cluster
1015 * that has been freed has its buffer still pinned in memory because
1016 * the transaction is still sitting in a iclog. The stale inodes
1017 * on that buffer will have their flush locks held until the
1018 * transaction hits the disk and the callbacks run. the inode
1019 * flush takes the flush lock unconditionally and with nothing to
1020 * push out the iclog we will never get that unlocked. hence we
1021 * need to force the log first.
1023 xfs_log_force(mp, XFS_LOG_SYNC);
1026 * Flush all pending changes from the AIL.
1028 xfs_ail_push_all_sync(mp->m_ail);
1031 * And reclaim all inodes. At this point there should be no dirty
1032 * inodes and none should be pinned or locked, but use synchronous
1033 * reclaim just to be sure. We can stop background inode reclaim
1034 * here as well if it is still running.
1036 cancel_delayed_work_sync(&mp->m_reclaim_work);
1037 xfs_reclaim_inodes(mp, SYNC_WAIT);
1042 * Unreserve any blocks we have so that when we unmount we don't account
1043 * the reserved free space as used. This is really only necessary for
1044 * lazy superblock counting because it trusts the incore superblock
1045 * counters to be absolutely correct on clean unmount.
1047 * We don't bother correcting this elsewhere for lazy superblock
1048 * counting because on mount of an unclean filesystem we reconstruct the
1049 * correct counter value and this is irrelevant.
1051 * For non-lazy counter filesystems, this doesn't matter at all because
1052 * we only every apply deltas to the superblock and hence the incore
1053 * value does not matter....
1056 error = xfs_reserve_blocks(mp, &resblks, NULL);
1058 xfs_warn(mp, "Unable to free reserved block pool. "
1059 "Freespace may not be correct on next mount.");
1061 error = xfs_log_sbcount(mp);
1063 xfs_warn(mp, "Unable to update superblock counters. "
1064 "Freespace may not be correct on next mount.");
1066 xfs_log_unmount(mp);
1068 xfs_uuid_unmount(mp);
1071 xfs_errortag_clearall(mp, 0);
1077 xfs_fs_writable(xfs_mount_t *mp)
1079 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1080 (mp->m_flags & XFS_MOUNT_RDONLY));
1086 * Sync the superblock counters to disk.
1088 * Note this code can be called during the process of freezing, so
1089 * we may need to use the transaction allocator which does not
1090 * block when the transaction subsystem is in its frozen state.
1093 xfs_log_sbcount(xfs_mount_t *mp)
1098 if (!xfs_fs_writable(mp))
1101 xfs_icsb_sync_counters(mp, 0);
1104 * we don't need to do this if we are updating the superblock
1105 * counters on every modification.
1107 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1110 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1111 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1113 xfs_trans_cancel(tp, 0);
1117 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1118 xfs_trans_set_sync(tp);
1119 error = xfs_trans_commit(tp, 0);
1124 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1125 * a delta to a specified field in the in-core superblock. Simply
1126 * switch on the field indicated and apply the delta to that field.
1127 * Fields are not allowed to dip below zero, so if the delta would
1128 * do this do not apply it and return EINVAL.
1130 * The m_sb_lock must be held when this routine is called.
1133 xfs_mod_incore_sb_unlocked(
1135 xfs_sb_field_t field,
1139 int scounter; /* short counter for 32 bit fields */
1140 long long lcounter; /* long counter for 64 bit fields */
1141 long long res_used, rem;
1144 * With the in-core superblock spin lock held, switch
1145 * on the indicated field. Apply the delta to the
1146 * proper field. If the fields value would dip below
1147 * 0, then do not apply the delta and return EINVAL.
1150 case XFS_SBS_ICOUNT:
1151 lcounter = (long long)mp->m_sb.sb_icount;
1155 return XFS_ERROR(EINVAL);
1157 mp->m_sb.sb_icount = lcounter;
1160 lcounter = (long long)mp->m_sb.sb_ifree;
1164 return XFS_ERROR(EINVAL);
1166 mp->m_sb.sb_ifree = lcounter;
1168 case XFS_SBS_FDBLOCKS:
1169 lcounter = (long long)
1170 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1171 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1173 if (delta > 0) { /* Putting blocks back */
1174 if (res_used > delta) {
1175 mp->m_resblks_avail += delta;
1177 rem = delta - res_used;
1178 mp->m_resblks_avail = mp->m_resblks;
1181 } else { /* Taking blocks away */
1183 if (lcounter >= 0) {
1184 mp->m_sb.sb_fdblocks = lcounter +
1185 XFS_ALLOC_SET_ASIDE(mp);
1190 * We are out of blocks, use any available reserved
1191 * blocks if were allowed to.
1194 return XFS_ERROR(ENOSPC);
1196 lcounter = (long long)mp->m_resblks_avail + delta;
1197 if (lcounter >= 0) {
1198 mp->m_resblks_avail = lcounter;
1201 printk_once(KERN_WARNING
1202 "Filesystem \"%s\": reserve blocks depleted! "
1203 "Consider increasing reserve pool size.",
1205 return XFS_ERROR(ENOSPC);
1208 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1210 case XFS_SBS_FREXTENTS:
1211 lcounter = (long long)mp->m_sb.sb_frextents;
1214 return XFS_ERROR(ENOSPC);
1216 mp->m_sb.sb_frextents = lcounter;
1218 case XFS_SBS_DBLOCKS:
1219 lcounter = (long long)mp->m_sb.sb_dblocks;
1223 return XFS_ERROR(EINVAL);
1225 mp->m_sb.sb_dblocks = lcounter;
1227 case XFS_SBS_AGCOUNT:
1228 scounter = mp->m_sb.sb_agcount;
1232 return XFS_ERROR(EINVAL);
1234 mp->m_sb.sb_agcount = scounter;
1236 case XFS_SBS_IMAX_PCT:
1237 scounter = mp->m_sb.sb_imax_pct;
1241 return XFS_ERROR(EINVAL);
1243 mp->m_sb.sb_imax_pct = scounter;
1245 case XFS_SBS_REXTSIZE:
1246 scounter = mp->m_sb.sb_rextsize;
1250 return XFS_ERROR(EINVAL);
1252 mp->m_sb.sb_rextsize = scounter;
1254 case XFS_SBS_RBMBLOCKS:
1255 scounter = mp->m_sb.sb_rbmblocks;
1259 return XFS_ERROR(EINVAL);
1261 mp->m_sb.sb_rbmblocks = scounter;
1263 case XFS_SBS_RBLOCKS:
1264 lcounter = (long long)mp->m_sb.sb_rblocks;
1268 return XFS_ERROR(EINVAL);
1270 mp->m_sb.sb_rblocks = lcounter;
1272 case XFS_SBS_REXTENTS:
1273 lcounter = (long long)mp->m_sb.sb_rextents;
1277 return XFS_ERROR(EINVAL);
1279 mp->m_sb.sb_rextents = lcounter;
1281 case XFS_SBS_REXTSLOG:
1282 scounter = mp->m_sb.sb_rextslog;
1286 return XFS_ERROR(EINVAL);
1288 mp->m_sb.sb_rextslog = scounter;
1292 return XFS_ERROR(EINVAL);
1297 * xfs_mod_incore_sb() is used to change a field in the in-core
1298 * superblock structure by the specified delta. This modification
1299 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1300 * routine to do the work.
1304 struct xfs_mount *mp,
1305 xfs_sb_field_t field,
1311 #ifdef HAVE_PERCPU_SB
1312 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1314 spin_lock(&mp->m_sb_lock);
1315 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1316 spin_unlock(&mp->m_sb_lock);
1322 * Change more than one field in the in-core superblock structure at a time.
1324 * The fields and changes to those fields are specified in the array of
1325 * xfs_mod_sb structures passed in. Either all of the specified deltas
1326 * will be applied or none of them will. If any modified field dips below 0,
1327 * then all modifications will be backed out and EINVAL will be returned.
1329 * Note that this function may not be used for the superblock values that
1330 * are tracked with the in-memory per-cpu counters - a direct call to
1331 * xfs_icsb_modify_counters is required for these.
1334 xfs_mod_incore_sb_batch(
1335 struct xfs_mount *mp,
1344 * Loop through the array of mod structures and apply each individually.
1345 * If any fail, then back out all those which have already been applied.
1346 * Do all of this within the scope of the m_sb_lock so that all of the
1347 * changes will be atomic.
1349 spin_lock(&mp->m_sb_lock);
1350 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1351 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1352 msbp->msb_field > XFS_SBS_FDBLOCKS);
1354 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1355 msbp->msb_delta, rsvd);
1359 spin_unlock(&mp->m_sb_lock);
1363 while (--msbp >= msb) {
1364 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1365 -msbp->msb_delta, rsvd);
1368 spin_unlock(&mp->m_sb_lock);
1373 * xfs_getsb() is called to obtain the buffer for the superblock.
1374 * The buffer is returned locked and read in from disk.
1375 * The buffer should be released with a call to xfs_brelse().
1377 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1378 * the superblock buffer if it can be locked without sleeping.
1379 * If it can't then we'll return NULL.
1383 struct xfs_mount *mp,
1386 struct xfs_buf *bp = mp->m_sb_bp;
1388 if (!xfs_buf_trylock(bp)) {
1389 if (flags & XBF_TRYLOCK)
1395 ASSERT(XFS_BUF_ISDONE(bp));
1400 * Used to free the superblock along various error paths.
1404 struct xfs_mount *mp)
1406 struct xfs_buf *bp = mp->m_sb_bp;
1414 * Used to log changes to the superblock unit and width fields which could
1415 * be altered by the mount options, as well as any potential sb_features2
1416 * fixup. Only the first superblock is updated.
1426 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1427 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1428 XFS_SB_VERSIONNUM));
1430 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1431 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1433 xfs_trans_cancel(tp, 0);
1436 xfs_mod_sb(tp, fields);
1437 error = xfs_trans_commit(tp, 0);
1442 * If the underlying (data/log/rt) device is readonly, there are some
1443 * operations that cannot proceed.
1446 xfs_dev_is_read_only(
1447 struct xfs_mount *mp,
1450 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1451 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1452 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1453 xfs_notice(mp, "%s required on read-only device.", message);
1454 xfs_notice(mp, "write access unavailable, cannot proceed.");
1460 #ifdef HAVE_PERCPU_SB
1462 * Per-cpu incore superblock counters
1464 * Simple concept, difficult implementation
1466 * Basically, replace the incore superblock counters with a distributed per cpu
1467 * counter for contended fields (e.g. free block count).
1469 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1470 * hence needs to be accurately read when we are running low on space. Hence
1471 * there is a method to enable and disable the per-cpu counters based on how
1472 * much "stuff" is available in them.
1474 * Basically, a counter is enabled if there is enough free resource to justify
1475 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1476 * ENOSPC), then we disable the counters to synchronise all callers and
1477 * re-distribute the available resources.
1479 * If, once we redistributed the available resources, we still get a failure,
1480 * we disable the per-cpu counter and go through the slow path.
1482 * The slow path is the current xfs_mod_incore_sb() function. This means that
1483 * when we disable a per-cpu counter, we need to drain its resources back to
1484 * the global superblock. We do this after disabling the counter to prevent
1485 * more threads from queueing up on the counter.
1487 * Essentially, this means that we still need a lock in the fast path to enable
1488 * synchronisation between the global counters and the per-cpu counters. This
1489 * is not a problem because the lock will be local to a CPU almost all the time
1490 * and have little contention except when we get to ENOSPC conditions.
1492 * Basically, this lock becomes a barrier that enables us to lock out the fast
1493 * path while we do things like enabling and disabling counters and
1494 * synchronising the counters.
1498 * 1. m_sb_lock before picking up per-cpu locks
1499 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1500 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1501 * 4. modifying per-cpu counters requires holding per-cpu lock
1502 * 5. modifying global counters requires holding m_sb_lock
1503 * 6. enabling or disabling a counter requires holding the m_sb_lock
1504 * and _none_ of the per-cpu locks.
1506 * Disabled counters are only ever re-enabled by a balance operation
1507 * that results in more free resources per CPU than a given threshold.
1508 * To ensure counters don't remain disabled, they are rebalanced when
1509 * the global resource goes above a higher threshold (i.e. some hysteresis
1510 * is present to prevent thrashing).
1513 #ifdef CONFIG_HOTPLUG_CPU
1515 * hot-plug CPU notifier support.
1517 * We need a notifier per filesystem as we need to be able to identify
1518 * the filesystem to balance the counters out. This is achieved by
1519 * having a notifier block embedded in the xfs_mount_t and doing pointer
1520 * magic to get the mount pointer from the notifier block address.
1523 xfs_icsb_cpu_notify(
1524 struct notifier_block *nfb,
1525 unsigned long action,
1528 xfs_icsb_cnts_t *cntp;
1531 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1532 cntp = (xfs_icsb_cnts_t *)
1533 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1535 case CPU_UP_PREPARE:
1536 case CPU_UP_PREPARE_FROZEN:
1537 /* Easy Case - initialize the area and locks, and
1538 * then rebalance when online does everything else for us. */
1539 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1542 case CPU_ONLINE_FROZEN:
1544 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1545 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1546 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1547 xfs_icsb_unlock(mp);
1550 case CPU_DEAD_FROZEN:
1551 /* Disable all the counters, then fold the dead cpu's
1552 * count into the total on the global superblock and
1553 * re-enable the counters. */
1555 spin_lock(&mp->m_sb_lock);
1556 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1557 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1558 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1560 mp->m_sb.sb_icount += cntp->icsb_icount;
1561 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1562 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1564 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1566 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1567 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1568 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1569 spin_unlock(&mp->m_sb_lock);
1570 xfs_icsb_unlock(mp);
1576 #endif /* CONFIG_HOTPLUG_CPU */
1579 xfs_icsb_init_counters(
1582 xfs_icsb_cnts_t *cntp;
1585 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1586 if (mp->m_sb_cnts == NULL)
1589 for_each_online_cpu(i) {
1590 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1591 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1594 mutex_init(&mp->m_icsb_mutex);
1597 * start with all counters disabled so that the
1598 * initial balance kicks us off correctly
1600 mp->m_icsb_counters = -1;
1602 #ifdef CONFIG_HOTPLUG_CPU
1603 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1604 mp->m_icsb_notifier.priority = 0;
1605 register_hotcpu_notifier(&mp->m_icsb_notifier);
1606 #endif /* CONFIG_HOTPLUG_CPU */
1612 xfs_icsb_reinit_counters(
1617 * start with all counters disabled so that the
1618 * initial balance kicks us off correctly
1620 mp->m_icsb_counters = -1;
1621 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1622 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1623 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1624 xfs_icsb_unlock(mp);
1628 xfs_icsb_destroy_counters(
1631 if (mp->m_sb_cnts) {
1632 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1633 free_percpu(mp->m_sb_cnts);
1635 mutex_destroy(&mp->m_icsb_mutex);
1640 xfs_icsb_cnts_t *icsbp)
1642 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1648 xfs_icsb_unlock_cntr(
1649 xfs_icsb_cnts_t *icsbp)
1651 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1656 xfs_icsb_lock_all_counters(
1659 xfs_icsb_cnts_t *cntp;
1662 for_each_online_cpu(i) {
1663 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1664 xfs_icsb_lock_cntr(cntp);
1669 xfs_icsb_unlock_all_counters(
1672 xfs_icsb_cnts_t *cntp;
1675 for_each_online_cpu(i) {
1676 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1677 xfs_icsb_unlock_cntr(cntp);
1684 xfs_icsb_cnts_t *cnt,
1687 xfs_icsb_cnts_t *cntp;
1690 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1692 if (!(flags & XFS_ICSB_LAZY_COUNT))
1693 xfs_icsb_lock_all_counters(mp);
1695 for_each_online_cpu(i) {
1696 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1697 cnt->icsb_icount += cntp->icsb_icount;
1698 cnt->icsb_ifree += cntp->icsb_ifree;
1699 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1702 if (!(flags & XFS_ICSB_LAZY_COUNT))
1703 xfs_icsb_unlock_all_counters(mp);
1707 xfs_icsb_counter_disabled(
1709 xfs_sb_field_t field)
1711 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1712 return test_bit(field, &mp->m_icsb_counters);
1716 xfs_icsb_disable_counter(
1718 xfs_sb_field_t field)
1720 xfs_icsb_cnts_t cnt;
1722 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1725 * If we are already disabled, then there is nothing to do
1726 * here. We check before locking all the counters to avoid
1727 * the expensive lock operation when being called in the
1728 * slow path and the counter is already disabled. This is
1729 * safe because the only time we set or clear this state is under
1732 if (xfs_icsb_counter_disabled(mp, field))
1735 xfs_icsb_lock_all_counters(mp);
1736 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1737 /* drain back to superblock */
1739 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1741 case XFS_SBS_ICOUNT:
1742 mp->m_sb.sb_icount = cnt.icsb_icount;
1745 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1747 case XFS_SBS_FDBLOCKS:
1748 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1755 xfs_icsb_unlock_all_counters(mp);
1759 xfs_icsb_enable_counter(
1761 xfs_sb_field_t field,
1765 xfs_icsb_cnts_t *cntp;
1768 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1770 xfs_icsb_lock_all_counters(mp);
1771 for_each_online_cpu(i) {
1772 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1774 case XFS_SBS_ICOUNT:
1775 cntp->icsb_icount = count + resid;
1778 cntp->icsb_ifree = count + resid;
1780 case XFS_SBS_FDBLOCKS:
1781 cntp->icsb_fdblocks = count + resid;
1789 clear_bit(field, &mp->m_icsb_counters);
1790 xfs_icsb_unlock_all_counters(mp);
1794 xfs_icsb_sync_counters_locked(
1798 xfs_icsb_cnts_t cnt;
1800 xfs_icsb_count(mp, &cnt, flags);
1802 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1803 mp->m_sb.sb_icount = cnt.icsb_icount;
1804 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1805 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1806 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1807 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1811 * Accurate update of per-cpu counters to incore superblock
1814 xfs_icsb_sync_counters(
1818 spin_lock(&mp->m_sb_lock);
1819 xfs_icsb_sync_counters_locked(mp, flags);
1820 spin_unlock(&mp->m_sb_lock);
1824 * Balance and enable/disable counters as necessary.
1826 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1827 * chosen to be the same number as single on disk allocation chunk per CPU, and
1828 * free blocks is something far enough zero that we aren't going thrash when we
1829 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1830 * prevent looping endlessly when xfs_alloc_space asks for more than will
1831 * be distributed to a single CPU but each CPU has enough blocks to be
1834 * Note that we can be called when counters are already disabled.
1835 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1836 * prevent locking every per-cpu counter needlessly.
1839 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1840 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1841 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1843 xfs_icsb_balance_counter_locked(
1845 xfs_sb_field_t field,
1848 uint64_t count, resid;
1849 int weight = num_online_cpus();
1850 uint64_t min = (uint64_t)min_per_cpu;
1852 /* disable counter and sync counter */
1853 xfs_icsb_disable_counter(mp, field);
1855 /* update counters - first CPU gets residual*/
1857 case XFS_SBS_ICOUNT:
1858 count = mp->m_sb.sb_icount;
1859 resid = do_div(count, weight);
1860 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1864 count = mp->m_sb.sb_ifree;
1865 resid = do_div(count, weight);
1866 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1869 case XFS_SBS_FDBLOCKS:
1870 count = mp->m_sb.sb_fdblocks;
1871 resid = do_div(count, weight);
1872 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
1877 count = resid = 0; /* quiet, gcc */
1881 xfs_icsb_enable_counter(mp, field, count, resid);
1885 xfs_icsb_balance_counter(
1887 xfs_sb_field_t fields,
1890 spin_lock(&mp->m_sb_lock);
1891 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1892 spin_unlock(&mp->m_sb_lock);
1896 xfs_icsb_modify_counters(
1898 xfs_sb_field_t field,
1902 xfs_icsb_cnts_t *icsbp;
1903 long long lcounter; /* long counter for 64 bit fields */
1909 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1912 * if the counter is disabled, go to slow path
1914 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1916 xfs_icsb_lock_cntr(icsbp);
1917 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1918 xfs_icsb_unlock_cntr(icsbp);
1923 case XFS_SBS_ICOUNT:
1924 lcounter = icsbp->icsb_icount;
1926 if (unlikely(lcounter < 0))
1927 goto balance_counter;
1928 icsbp->icsb_icount = lcounter;
1932 lcounter = icsbp->icsb_ifree;
1934 if (unlikely(lcounter < 0))
1935 goto balance_counter;
1936 icsbp->icsb_ifree = lcounter;
1939 case XFS_SBS_FDBLOCKS:
1940 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1942 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1944 if (unlikely(lcounter < 0))
1945 goto balance_counter;
1946 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1952 xfs_icsb_unlock_cntr(icsbp);
1960 * serialise with a mutex so we don't burn lots of cpu on
1961 * the superblock lock. We still need to hold the superblock
1962 * lock, however, when we modify the global structures.
1967 * Now running atomically.
1969 * If the counter is enabled, someone has beaten us to rebalancing.
1970 * Drop the lock and try again in the fast path....
1972 if (!(xfs_icsb_counter_disabled(mp, field))) {
1973 xfs_icsb_unlock(mp);
1978 * The counter is currently disabled. Because we are
1979 * running atomically here, we know a rebalance cannot
1980 * be in progress. Hence we can go straight to operating
1981 * on the global superblock. We do not call xfs_mod_incore_sb()
1982 * here even though we need to get the m_sb_lock. Doing so
1983 * will cause us to re-enter this function and deadlock.
1984 * Hence we get the m_sb_lock ourselves and then call
1985 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1986 * directly on the global counters.
1988 spin_lock(&mp->m_sb_lock);
1989 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1990 spin_unlock(&mp->m_sb_lock);
1993 * Now that we've modified the global superblock, we
1994 * may be able to re-enable the distributed counters
1995 * (e.g. lots of space just got freed). After that
1999 xfs_icsb_balance_counter(mp, field, 0);
2000 xfs_icsb_unlock(mp);
2004 xfs_icsb_unlock_cntr(icsbp);
2008 * We may have multiple threads here if multiple per-cpu
2009 * counters run dry at the same time. This will mean we can
2010 * do more balances than strictly necessary but it is not
2011 * the common slowpath case.
2016 * running atomically.
2018 * This will leave the counter in the correct state for future
2019 * accesses. After the rebalance, we simply try again and our retry
2020 * will either succeed through the fast path or slow path without
2021 * another balance operation being required.
2023 xfs_icsb_balance_counter(mp, field, delta);
2024 xfs_icsb_unlock(mp);