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);
286 ASSERT(mp->m_sb_bp == NULL);
287 ASSERT(mp->m_ddev_targp != NULL);
290 * Allocate a (locked) buffer to hold the superblock.
291 * This will be kept around at all times to optimize
292 * access to the superblock.
294 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
297 bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
298 BTOBB(sector_size), 0,
299 loud ? &xfs_sb_buf_ops
300 : &xfs_sb_quiet_buf_ops);
303 xfs_warn(mp, "SB buffer read failed");
309 xfs_warn(mp, "SB validate failed with error %d.", error);
314 * Initialize the mount structure from the superblock.
316 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
317 xfs_sb_quota_from_disk(&mp->m_sb);
320 * We must be able to do sector-sized and sector-aligned IO.
322 if (sector_size > sbp->sb_sectsize) {
324 xfs_warn(mp, "device supports %u byte sectors (not %u)",
325 sector_size, sbp->sb_sectsize);
331 * If device sector size is smaller than the superblock size,
332 * re-read the superblock so the buffer is correctly sized.
334 if (sector_size < sbp->sb_sectsize) {
336 sector_size = sbp->sb_sectsize;
340 /* Initialize per-cpu counters */
341 xfs_icsb_reinit_counters(mp);
343 /* no need to be quiet anymore, so reset the buf ops */
344 bp->b_ops = &xfs_sb_buf_ops;
356 * Update alignment values based on mount options and sb values
359 xfs_update_alignment(xfs_mount_t *mp)
361 xfs_sb_t *sbp = &(mp->m_sb);
365 * If stripe unit and stripe width are not multiples
366 * of the fs blocksize turn off alignment.
368 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
369 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
371 "alignment check failed: sunit/swidth vs. blocksize(%d)",
373 return XFS_ERROR(EINVAL);
376 * Convert the stripe unit and width to FSBs.
378 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
379 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
381 "alignment check failed: sunit/swidth vs. agsize(%d)",
383 return XFS_ERROR(EINVAL);
384 } else if (mp->m_dalign) {
385 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
388 "alignment check failed: sunit(%d) less than bsize(%d)",
389 mp->m_dalign, sbp->sb_blocksize);
390 return XFS_ERROR(EINVAL);
395 * Update superblock with new values
398 if (xfs_sb_version_hasdalign(sbp)) {
399 if (sbp->sb_unit != mp->m_dalign) {
400 sbp->sb_unit = mp->m_dalign;
401 mp->m_update_flags |= XFS_SB_UNIT;
403 if (sbp->sb_width != mp->m_swidth) {
404 sbp->sb_width = mp->m_swidth;
405 mp->m_update_flags |= XFS_SB_WIDTH;
409 "cannot change alignment: superblock does not support data alignment");
410 return XFS_ERROR(EINVAL);
412 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
413 xfs_sb_version_hasdalign(&mp->m_sb)) {
414 mp->m_dalign = sbp->sb_unit;
415 mp->m_swidth = sbp->sb_width;
422 * Set the maximum inode count for this filesystem
425 xfs_set_maxicount(xfs_mount_t *mp)
427 xfs_sb_t *sbp = &(mp->m_sb);
430 if (sbp->sb_imax_pct) {
432 * Make sure the maximum inode count is a multiple
433 * of the units we allocate inodes in.
435 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
437 do_div(icount, mp->m_ialloc_blks);
438 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
446 * Set the default minimum read and write sizes unless
447 * already specified in a mount option.
448 * We use smaller I/O sizes when the file system
449 * is being used for NFS service (wsync mount option).
452 xfs_set_rw_sizes(xfs_mount_t *mp)
454 xfs_sb_t *sbp = &(mp->m_sb);
455 int readio_log, writeio_log;
457 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
458 if (mp->m_flags & XFS_MOUNT_WSYNC) {
459 readio_log = XFS_WSYNC_READIO_LOG;
460 writeio_log = XFS_WSYNC_WRITEIO_LOG;
462 readio_log = XFS_READIO_LOG_LARGE;
463 writeio_log = XFS_WRITEIO_LOG_LARGE;
466 readio_log = mp->m_readio_log;
467 writeio_log = mp->m_writeio_log;
470 if (sbp->sb_blocklog > readio_log) {
471 mp->m_readio_log = sbp->sb_blocklog;
473 mp->m_readio_log = readio_log;
475 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
476 if (sbp->sb_blocklog > writeio_log) {
477 mp->m_writeio_log = sbp->sb_blocklog;
479 mp->m_writeio_log = writeio_log;
481 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
485 * precalculate the low space thresholds for dynamic speculative preallocation.
488 xfs_set_low_space_thresholds(
489 struct xfs_mount *mp)
493 for (i = 0; i < XFS_LOWSP_MAX; i++) {
494 __uint64_t space = mp->m_sb.sb_dblocks;
497 mp->m_low_space[i] = space * (i + 1);
503 * Set whether we're using inode alignment.
506 xfs_set_inoalignment(xfs_mount_t *mp)
508 if (xfs_sb_version_hasalign(&mp->m_sb) &&
509 mp->m_sb.sb_inoalignmt >=
510 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
511 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
513 mp->m_inoalign_mask = 0;
515 * If we are using stripe alignment, check whether
516 * the stripe unit is a multiple of the inode alignment
518 if (mp->m_dalign && mp->m_inoalign_mask &&
519 !(mp->m_dalign & mp->m_inoalign_mask))
520 mp->m_sinoalign = mp->m_dalign;
526 * Check that the data (and log if separate) is an ok size.
529 xfs_check_sizes(xfs_mount_t *mp)
534 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
535 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
536 xfs_warn(mp, "filesystem size mismatch detected");
537 return XFS_ERROR(EFBIG);
539 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
540 d - XFS_FSS_TO_BB(mp, 1),
541 XFS_FSS_TO_BB(mp, 1), 0, NULL);
543 xfs_warn(mp, "last sector read failed");
548 if (mp->m_logdev_targp != mp->m_ddev_targp) {
549 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
550 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
551 xfs_warn(mp, "log size mismatch detected");
552 return XFS_ERROR(EFBIG);
554 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
555 d - XFS_FSB_TO_BB(mp, 1),
556 XFS_FSB_TO_BB(mp, 1), 0, NULL);
558 xfs_warn(mp, "log device read failed");
567 * Clear the quotaflags in memory and in the superblock.
570 xfs_mount_reset_sbqflags(
571 struct xfs_mount *mp)
574 struct xfs_trans *tp;
579 * It is OK to look at sb_qflags here in mount path,
582 if (mp->m_sb.sb_qflags == 0)
584 spin_lock(&mp->m_sb_lock);
585 mp->m_sb.sb_qflags = 0;
586 spin_unlock(&mp->m_sb_lock);
589 * If the fs is readonly, let the incore superblock run
590 * with quotas off but don't flush the update out to disk
592 if (mp->m_flags & XFS_MOUNT_RDONLY)
595 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
596 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
598 xfs_trans_cancel(tp, 0);
599 xfs_alert(mp, "%s: Superblock update failed!", __func__);
603 xfs_mod_sb(tp, XFS_SB_QFLAGS);
604 return xfs_trans_commit(tp, 0);
608 xfs_default_resblks(xfs_mount_t *mp)
613 * We default to 5% or 8192 fsbs of space reserved, whichever is
614 * smaller. This is intended to cover concurrent allocation
615 * transactions when we initially hit enospc. These each require a 4
616 * block reservation. Hence by default we cover roughly 2000 concurrent
617 * allocation reservations.
619 resblks = mp->m_sb.sb_dblocks;
621 resblks = min_t(__uint64_t, resblks, 8192);
626 * This function does the following on an initial mount of a file system:
627 * - reads the superblock from disk and init the mount struct
628 * - if we're a 32-bit kernel, do a size check on the superblock
629 * so we don't mount terabyte filesystems
630 * - init mount struct realtime fields
631 * - allocate inode hash table for fs
632 * - init directory manager
633 * - perform recovery and init the log manager
639 xfs_sb_t *sbp = &(mp->m_sb);
646 xfs_sb_mount_common(mp, sbp);
649 * Check for a mismatched features2 values. Older kernels
650 * read & wrote into the wrong sb offset for sb_features2
651 * on some platforms due to xfs_sb_t not being 64bit size aligned
652 * when sb_features2 was added, which made older superblock
653 * reading/writing routines swap it as a 64-bit value.
655 * For backwards compatibility, we make both slots equal.
657 * If we detect a mismatched field, we OR the set bits into the
658 * existing features2 field in case it has already been modified; we
659 * don't want to lose any features. We then update the bad location
660 * with the ORed value so that older kernels will see any features2
661 * flags, and mark the two fields as needing updates once the
662 * transaction subsystem is online.
664 if (xfs_sb_has_mismatched_features2(sbp)) {
665 xfs_warn(mp, "correcting sb_features alignment problem");
666 sbp->sb_features2 |= sbp->sb_bad_features2;
667 sbp->sb_bad_features2 = sbp->sb_features2;
668 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
671 * Re-check for ATTR2 in case it was found in bad_features2
674 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
675 !(mp->m_flags & XFS_MOUNT_NOATTR2))
676 mp->m_flags |= XFS_MOUNT_ATTR2;
679 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
680 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
681 xfs_sb_version_removeattr2(&mp->m_sb);
682 mp->m_update_flags |= XFS_SB_FEATURES2;
684 /* update sb_versionnum for the clearing of the morebits */
685 if (!sbp->sb_features2)
686 mp->m_update_flags |= XFS_SB_VERSIONNUM;
690 * Check if sb_agblocks is aligned at stripe boundary
691 * If sb_agblocks is NOT aligned turn off m_dalign since
692 * allocator alignment is within an ag, therefore ag has
693 * to be aligned at stripe boundary.
695 error = xfs_update_alignment(mp);
699 xfs_alloc_compute_maxlevels(mp);
700 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
701 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
702 xfs_ialloc_compute_maxlevels(mp);
704 xfs_set_maxicount(mp);
706 error = xfs_uuid_mount(mp);
711 * Set the minimum read and write sizes
713 xfs_set_rw_sizes(mp);
715 /* set the low space thresholds for dynamic preallocation */
716 xfs_set_low_space_thresholds(mp);
719 * Set the inode cluster size.
720 * This may still be overridden by the file system
721 * block size if it is larger than the chosen cluster size.
723 * For v5 filesystems, scale the cluster size with the inode size to
724 * keep a constant ratio of inode per cluster buffer, but only if mkfs
725 * has set the inode alignment value appropriately for larger cluster
728 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
729 if (xfs_sb_version_hascrc(&mp->m_sb)) {
730 int new_size = mp->m_inode_cluster_size;
732 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
733 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
734 mp->m_inode_cluster_size = new_size;
735 xfs_info(mp, "Using inode cluster size of %d bytes",
736 mp->m_inode_cluster_size);
740 * Set inode alignment fields
742 xfs_set_inoalignment(mp);
745 * Check that the data (and log if separate) is an ok size.
747 error = xfs_check_sizes(mp);
749 goto out_remove_uuid;
752 * Initialize realtime fields in the mount structure
754 error = xfs_rtmount_init(mp);
756 xfs_warn(mp, "RT mount failed");
757 goto out_remove_uuid;
761 * Copies the low order bits of the timestamp and the randomly
762 * set "sequence" number out of a UUID.
764 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
766 mp->m_dmevmask = 0; /* not persistent; set after each mount */
771 * Initialize the attribute manager's entries.
773 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
776 * Initialize the precomputed transaction reservations values.
781 * Allocate and initialize the per-ag data.
783 spin_lock_init(&mp->m_perag_lock);
784 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
785 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
787 xfs_warn(mp, "Failed per-ag init: %d", error);
788 goto out_remove_uuid;
791 if (!sbp->sb_logblocks) {
792 xfs_warn(mp, "no log defined");
793 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
794 error = XFS_ERROR(EFSCORRUPTED);
799 * log's mount-time initialization. Perform 1st part recovery if needed
801 error = xfs_log_mount(mp, mp->m_logdev_targp,
802 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
803 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
805 xfs_warn(mp, "log mount failed");
810 * Now the log is mounted, we know if it was an unclean shutdown or
811 * not. If it was, with the first phase of recovery has completed, we
812 * have consistent AG blocks on disk. We have not recovered EFIs yet,
813 * but they are recovered transactionally in the second recovery phase
816 * Hence we can safely re-initialise incore superblock counters from
817 * the per-ag data. These may not be correct if the filesystem was not
818 * cleanly unmounted, so we need to wait for recovery to finish before
821 * If the filesystem was cleanly unmounted, then we can trust the
822 * values in the superblock to be correct and we don't need to do
825 * If we are currently making the filesystem, the initialisation will
826 * fail as the perag data is in an undefined state.
828 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
829 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
830 !mp->m_sb.sb_inprogress) {
831 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
837 * Get and sanity-check the root inode.
838 * Save the pointer to it in the mount structure.
840 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
842 xfs_warn(mp, "failed to read root inode");
843 goto out_log_dealloc;
848 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
849 xfs_warn(mp, "corrupted root inode %llu: not a directory",
850 (unsigned long long)rip->i_ino);
851 xfs_iunlock(rip, XFS_ILOCK_EXCL);
852 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
854 error = XFS_ERROR(EFSCORRUPTED);
857 mp->m_rootip = rip; /* save it */
859 xfs_iunlock(rip, XFS_ILOCK_EXCL);
862 * Initialize realtime inode pointers in the mount structure
864 error = xfs_rtmount_inodes(mp);
867 * Free up the root inode.
869 xfs_warn(mp, "failed to read RT inodes");
874 * If this is a read-only mount defer the superblock updates until
875 * the next remount into writeable mode. Otherwise we would never
876 * perform the update e.g. for the root filesystem.
878 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
879 error = xfs_mount_log_sb(mp, mp->m_update_flags);
881 xfs_warn(mp, "failed to write sb changes");
887 * Initialise the XFS quota management subsystem for this mount
889 if (XFS_IS_QUOTA_RUNNING(mp)) {
890 error = xfs_qm_newmount(mp, "amount, "aflags);
894 ASSERT(!XFS_IS_QUOTA_ON(mp));
897 * If a file system had quotas running earlier, but decided to
898 * mount without -o uquota/pquota/gquota options, revoke the
899 * quotachecked license.
901 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
902 xfs_notice(mp, "resetting quota flags");
903 error = xfs_mount_reset_sbqflags(mp);
910 * Finish recovering the file system. This part needed to be
911 * delayed until after the root and real-time bitmap inodes
912 * were consistently read in.
914 error = xfs_log_mount_finish(mp);
916 xfs_warn(mp, "log mount finish failed");
921 * Complete the quota initialisation, post-log-replay component.
924 ASSERT(mp->m_qflags == 0);
925 mp->m_qflags = quotaflags;
927 xfs_qm_mount_quotas(mp);
931 * Now we are mounted, reserve a small amount of unused space for
932 * privileged transactions. This is needed so that transaction
933 * space required for critical operations can dip into this pool
934 * when at ENOSPC. This is needed for operations like create with
935 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
936 * are not allowed to use this reserved space.
938 * This may drive us straight to ENOSPC on mount, but that implies
939 * we were already there on the last unmount. Warn if this occurs.
941 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
942 resblks = xfs_default_resblks(mp);
943 error = xfs_reserve_blocks(mp, &resblks, NULL);
946 "Unable to allocate reserve blocks. Continuing without reserve pool.");
952 xfs_rtunmount_inodes(mp);
958 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
959 xfs_wait_buftarg(mp->m_logdev_targp);
960 xfs_wait_buftarg(mp->m_ddev_targp);
964 xfs_uuid_unmount(mp);
970 * This flushes out the inodes,dquots and the superblock, unmounts the
971 * log and makes sure that incore structures are freed.
975 struct xfs_mount *mp)
980 cancel_delayed_work_sync(&mp->m_eofblocks_work);
982 xfs_qm_unmount_quotas(mp);
983 xfs_rtunmount_inodes(mp);
987 * We can potentially deadlock here if we have an inode cluster
988 * that has been freed has its buffer still pinned in memory because
989 * the transaction is still sitting in a iclog. The stale inodes
990 * on that buffer will have their flush locks held until the
991 * transaction hits the disk and the callbacks run. the inode
992 * flush takes the flush lock unconditionally and with nothing to
993 * push out the iclog we will never get that unlocked. hence we
994 * need to force the log first.
996 xfs_log_force(mp, XFS_LOG_SYNC);
999 * Flush all pending changes from the AIL.
1001 xfs_ail_push_all_sync(mp->m_ail);
1004 * And reclaim all inodes. At this point there should be no dirty
1005 * inodes and none should be pinned or locked, but use synchronous
1006 * reclaim just to be sure. We can stop background inode reclaim
1007 * here as well if it is still running.
1009 cancel_delayed_work_sync(&mp->m_reclaim_work);
1010 xfs_reclaim_inodes(mp, SYNC_WAIT);
1015 * Unreserve any blocks we have so that when we unmount we don't account
1016 * the reserved free space as used. This is really only necessary for
1017 * lazy superblock counting because it trusts the incore superblock
1018 * counters to be absolutely correct on clean unmount.
1020 * We don't bother correcting this elsewhere for lazy superblock
1021 * counting because on mount of an unclean filesystem we reconstruct the
1022 * correct counter value and this is irrelevant.
1024 * For non-lazy counter filesystems, this doesn't matter at all because
1025 * we only every apply deltas to the superblock and hence the incore
1026 * value does not matter....
1029 error = xfs_reserve_blocks(mp, &resblks, NULL);
1031 xfs_warn(mp, "Unable to free reserved block pool. "
1032 "Freespace may not be correct on next mount.");
1034 error = xfs_log_sbcount(mp);
1036 xfs_warn(mp, "Unable to update superblock counters. "
1037 "Freespace may not be correct on next mount.");
1039 xfs_log_unmount(mp);
1040 xfs_uuid_unmount(mp);
1043 xfs_errortag_clearall(mp, 0);
1049 xfs_fs_writable(xfs_mount_t *mp)
1051 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1052 (mp->m_flags & XFS_MOUNT_RDONLY));
1058 * Sync the superblock counters to disk.
1060 * Note this code can be called during the process of freezing, so
1061 * we may need to use the transaction allocator which does not
1062 * block when the transaction subsystem is in its frozen state.
1065 xfs_log_sbcount(xfs_mount_t *mp)
1070 if (!xfs_fs_writable(mp))
1073 xfs_icsb_sync_counters(mp, 0);
1076 * we don't need to do this if we are updating the superblock
1077 * counters on every modification.
1079 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1082 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1083 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1085 xfs_trans_cancel(tp, 0);
1089 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1090 xfs_trans_set_sync(tp);
1091 error = xfs_trans_commit(tp, 0);
1096 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1097 * a delta to a specified field in the in-core superblock. Simply
1098 * switch on the field indicated and apply the delta to that field.
1099 * Fields are not allowed to dip below zero, so if the delta would
1100 * do this do not apply it and return EINVAL.
1102 * The m_sb_lock must be held when this routine is called.
1105 xfs_mod_incore_sb_unlocked(
1107 xfs_sb_field_t field,
1111 int scounter; /* short counter for 32 bit fields */
1112 long long lcounter; /* long counter for 64 bit fields */
1113 long long res_used, rem;
1116 * With the in-core superblock spin lock held, switch
1117 * on the indicated field. Apply the delta to the
1118 * proper field. If the fields value would dip below
1119 * 0, then do not apply the delta and return EINVAL.
1122 case XFS_SBS_ICOUNT:
1123 lcounter = (long long)mp->m_sb.sb_icount;
1127 return XFS_ERROR(EINVAL);
1129 mp->m_sb.sb_icount = lcounter;
1132 lcounter = (long long)mp->m_sb.sb_ifree;
1136 return XFS_ERROR(EINVAL);
1138 mp->m_sb.sb_ifree = lcounter;
1140 case XFS_SBS_FDBLOCKS:
1141 lcounter = (long long)
1142 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1143 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1145 if (delta > 0) { /* Putting blocks back */
1146 if (res_used > delta) {
1147 mp->m_resblks_avail += delta;
1149 rem = delta - res_used;
1150 mp->m_resblks_avail = mp->m_resblks;
1153 } else { /* Taking blocks away */
1155 if (lcounter >= 0) {
1156 mp->m_sb.sb_fdblocks = lcounter +
1157 XFS_ALLOC_SET_ASIDE(mp);
1162 * We are out of blocks, use any available reserved
1163 * blocks if were allowed to.
1166 return XFS_ERROR(ENOSPC);
1168 lcounter = (long long)mp->m_resblks_avail + delta;
1169 if (lcounter >= 0) {
1170 mp->m_resblks_avail = lcounter;
1173 printk_once(KERN_WARNING
1174 "Filesystem \"%s\": reserve blocks depleted! "
1175 "Consider increasing reserve pool size.",
1177 return XFS_ERROR(ENOSPC);
1180 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1182 case XFS_SBS_FREXTENTS:
1183 lcounter = (long long)mp->m_sb.sb_frextents;
1186 return XFS_ERROR(ENOSPC);
1188 mp->m_sb.sb_frextents = lcounter;
1190 case XFS_SBS_DBLOCKS:
1191 lcounter = (long long)mp->m_sb.sb_dblocks;
1195 return XFS_ERROR(EINVAL);
1197 mp->m_sb.sb_dblocks = lcounter;
1199 case XFS_SBS_AGCOUNT:
1200 scounter = mp->m_sb.sb_agcount;
1204 return XFS_ERROR(EINVAL);
1206 mp->m_sb.sb_agcount = scounter;
1208 case XFS_SBS_IMAX_PCT:
1209 scounter = mp->m_sb.sb_imax_pct;
1213 return XFS_ERROR(EINVAL);
1215 mp->m_sb.sb_imax_pct = scounter;
1217 case XFS_SBS_REXTSIZE:
1218 scounter = mp->m_sb.sb_rextsize;
1222 return XFS_ERROR(EINVAL);
1224 mp->m_sb.sb_rextsize = scounter;
1226 case XFS_SBS_RBMBLOCKS:
1227 scounter = mp->m_sb.sb_rbmblocks;
1231 return XFS_ERROR(EINVAL);
1233 mp->m_sb.sb_rbmblocks = scounter;
1235 case XFS_SBS_RBLOCKS:
1236 lcounter = (long long)mp->m_sb.sb_rblocks;
1240 return XFS_ERROR(EINVAL);
1242 mp->m_sb.sb_rblocks = lcounter;
1244 case XFS_SBS_REXTENTS:
1245 lcounter = (long long)mp->m_sb.sb_rextents;
1249 return XFS_ERROR(EINVAL);
1251 mp->m_sb.sb_rextents = lcounter;
1253 case XFS_SBS_REXTSLOG:
1254 scounter = mp->m_sb.sb_rextslog;
1258 return XFS_ERROR(EINVAL);
1260 mp->m_sb.sb_rextslog = scounter;
1264 return XFS_ERROR(EINVAL);
1269 * xfs_mod_incore_sb() is used to change a field in the in-core
1270 * superblock structure by the specified delta. This modification
1271 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1272 * routine to do the work.
1276 struct xfs_mount *mp,
1277 xfs_sb_field_t field,
1283 #ifdef HAVE_PERCPU_SB
1284 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1286 spin_lock(&mp->m_sb_lock);
1287 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1288 spin_unlock(&mp->m_sb_lock);
1294 * Change more than one field in the in-core superblock structure at a time.
1296 * The fields and changes to those fields are specified in the array of
1297 * xfs_mod_sb structures passed in. Either all of the specified deltas
1298 * will be applied or none of them will. If any modified field dips below 0,
1299 * then all modifications will be backed out and EINVAL will be returned.
1301 * Note that this function may not be used for the superblock values that
1302 * are tracked with the in-memory per-cpu counters - a direct call to
1303 * xfs_icsb_modify_counters is required for these.
1306 xfs_mod_incore_sb_batch(
1307 struct xfs_mount *mp,
1316 * Loop through the array of mod structures and apply each individually.
1317 * If any fail, then back out all those which have already been applied.
1318 * Do all of this within the scope of the m_sb_lock so that all of the
1319 * changes will be atomic.
1321 spin_lock(&mp->m_sb_lock);
1322 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1323 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1324 msbp->msb_field > XFS_SBS_FDBLOCKS);
1326 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1327 msbp->msb_delta, rsvd);
1331 spin_unlock(&mp->m_sb_lock);
1335 while (--msbp >= msb) {
1336 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1337 -msbp->msb_delta, rsvd);
1340 spin_unlock(&mp->m_sb_lock);
1345 * xfs_getsb() is called to obtain the buffer for the superblock.
1346 * The buffer is returned locked and read in from disk.
1347 * The buffer should be released with a call to xfs_brelse().
1349 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1350 * the superblock buffer if it can be locked without sleeping.
1351 * If it can't then we'll return NULL.
1355 struct xfs_mount *mp,
1358 struct xfs_buf *bp = mp->m_sb_bp;
1360 if (!xfs_buf_trylock(bp)) {
1361 if (flags & XBF_TRYLOCK)
1367 ASSERT(XFS_BUF_ISDONE(bp));
1372 * Used to free the superblock along various error paths.
1376 struct xfs_mount *mp)
1378 struct xfs_buf *bp = mp->m_sb_bp;
1386 * Used to log changes to the superblock unit and width fields which could
1387 * be altered by the mount options, as well as any potential sb_features2
1388 * fixup. Only the first superblock is updated.
1398 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1399 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1400 XFS_SB_VERSIONNUM));
1402 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1403 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
1405 xfs_trans_cancel(tp, 0);
1408 xfs_mod_sb(tp, fields);
1409 error = xfs_trans_commit(tp, 0);
1414 * If the underlying (data/log/rt) device is readonly, there are some
1415 * operations that cannot proceed.
1418 xfs_dev_is_read_only(
1419 struct xfs_mount *mp,
1422 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1423 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1424 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1425 xfs_notice(mp, "%s required on read-only device.", message);
1426 xfs_notice(mp, "write access unavailable, cannot proceed.");
1432 #ifdef HAVE_PERCPU_SB
1434 * Per-cpu incore superblock counters
1436 * Simple concept, difficult implementation
1438 * Basically, replace the incore superblock counters with a distributed per cpu
1439 * counter for contended fields (e.g. free block count).
1441 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1442 * hence needs to be accurately read when we are running low on space. Hence
1443 * there is a method to enable and disable the per-cpu counters based on how
1444 * much "stuff" is available in them.
1446 * Basically, a counter is enabled if there is enough free resource to justify
1447 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1448 * ENOSPC), then we disable the counters to synchronise all callers and
1449 * re-distribute the available resources.
1451 * If, once we redistributed the available resources, we still get a failure,
1452 * we disable the per-cpu counter and go through the slow path.
1454 * The slow path is the current xfs_mod_incore_sb() function. This means that
1455 * when we disable a per-cpu counter, we need to drain its resources back to
1456 * the global superblock. We do this after disabling the counter to prevent
1457 * more threads from queueing up on the counter.
1459 * Essentially, this means that we still need a lock in the fast path to enable
1460 * synchronisation between the global counters and the per-cpu counters. This
1461 * is not a problem because the lock will be local to a CPU almost all the time
1462 * and have little contention except when we get to ENOSPC conditions.
1464 * Basically, this lock becomes a barrier that enables us to lock out the fast
1465 * path while we do things like enabling and disabling counters and
1466 * synchronising the counters.
1470 * 1. m_sb_lock before picking up per-cpu locks
1471 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1472 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1473 * 4. modifying per-cpu counters requires holding per-cpu lock
1474 * 5. modifying global counters requires holding m_sb_lock
1475 * 6. enabling or disabling a counter requires holding the m_sb_lock
1476 * and _none_ of the per-cpu locks.
1478 * Disabled counters are only ever re-enabled by a balance operation
1479 * that results in more free resources per CPU than a given threshold.
1480 * To ensure counters don't remain disabled, they are rebalanced when
1481 * the global resource goes above a higher threshold (i.e. some hysteresis
1482 * is present to prevent thrashing).
1485 #ifdef CONFIG_HOTPLUG_CPU
1487 * hot-plug CPU notifier support.
1489 * We need a notifier per filesystem as we need to be able to identify
1490 * the filesystem to balance the counters out. This is achieved by
1491 * having a notifier block embedded in the xfs_mount_t and doing pointer
1492 * magic to get the mount pointer from the notifier block address.
1495 xfs_icsb_cpu_notify(
1496 struct notifier_block *nfb,
1497 unsigned long action,
1500 xfs_icsb_cnts_t *cntp;
1503 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1504 cntp = (xfs_icsb_cnts_t *)
1505 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1507 case CPU_UP_PREPARE:
1508 case CPU_UP_PREPARE_FROZEN:
1509 /* Easy Case - initialize the area and locks, and
1510 * then rebalance when online does everything else for us. */
1511 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1514 case CPU_ONLINE_FROZEN:
1516 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1517 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1518 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1519 xfs_icsb_unlock(mp);
1522 case CPU_DEAD_FROZEN:
1523 /* Disable all the counters, then fold the dead cpu's
1524 * count into the total on the global superblock and
1525 * re-enable the counters. */
1527 spin_lock(&mp->m_sb_lock);
1528 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1529 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1530 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1532 mp->m_sb.sb_icount += cntp->icsb_icount;
1533 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1534 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1536 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1538 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1539 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1540 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1541 spin_unlock(&mp->m_sb_lock);
1542 xfs_icsb_unlock(mp);
1548 #endif /* CONFIG_HOTPLUG_CPU */
1551 xfs_icsb_init_counters(
1554 xfs_icsb_cnts_t *cntp;
1557 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1558 if (mp->m_sb_cnts == NULL)
1561 for_each_online_cpu(i) {
1562 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1563 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1566 mutex_init(&mp->m_icsb_mutex);
1569 * start with all counters disabled so that the
1570 * initial balance kicks us off correctly
1572 mp->m_icsb_counters = -1;
1574 #ifdef CONFIG_HOTPLUG_CPU
1575 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1576 mp->m_icsb_notifier.priority = 0;
1577 register_hotcpu_notifier(&mp->m_icsb_notifier);
1578 #endif /* CONFIG_HOTPLUG_CPU */
1584 xfs_icsb_reinit_counters(
1589 * start with all counters disabled so that the
1590 * initial balance kicks us off correctly
1592 mp->m_icsb_counters = -1;
1593 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1594 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1595 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1596 xfs_icsb_unlock(mp);
1600 xfs_icsb_destroy_counters(
1603 if (mp->m_sb_cnts) {
1604 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1605 free_percpu(mp->m_sb_cnts);
1607 mutex_destroy(&mp->m_icsb_mutex);
1612 xfs_icsb_cnts_t *icsbp)
1614 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1620 xfs_icsb_unlock_cntr(
1621 xfs_icsb_cnts_t *icsbp)
1623 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1628 xfs_icsb_lock_all_counters(
1631 xfs_icsb_cnts_t *cntp;
1634 for_each_online_cpu(i) {
1635 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1636 xfs_icsb_lock_cntr(cntp);
1641 xfs_icsb_unlock_all_counters(
1644 xfs_icsb_cnts_t *cntp;
1647 for_each_online_cpu(i) {
1648 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1649 xfs_icsb_unlock_cntr(cntp);
1656 xfs_icsb_cnts_t *cnt,
1659 xfs_icsb_cnts_t *cntp;
1662 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1664 if (!(flags & XFS_ICSB_LAZY_COUNT))
1665 xfs_icsb_lock_all_counters(mp);
1667 for_each_online_cpu(i) {
1668 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1669 cnt->icsb_icount += cntp->icsb_icount;
1670 cnt->icsb_ifree += cntp->icsb_ifree;
1671 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1674 if (!(flags & XFS_ICSB_LAZY_COUNT))
1675 xfs_icsb_unlock_all_counters(mp);
1679 xfs_icsb_counter_disabled(
1681 xfs_sb_field_t field)
1683 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1684 return test_bit(field, &mp->m_icsb_counters);
1688 xfs_icsb_disable_counter(
1690 xfs_sb_field_t field)
1692 xfs_icsb_cnts_t cnt;
1694 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1697 * If we are already disabled, then there is nothing to do
1698 * here. We check before locking all the counters to avoid
1699 * the expensive lock operation when being called in the
1700 * slow path and the counter is already disabled. This is
1701 * safe because the only time we set or clear this state is under
1704 if (xfs_icsb_counter_disabled(mp, field))
1707 xfs_icsb_lock_all_counters(mp);
1708 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1709 /* drain back to superblock */
1711 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1713 case XFS_SBS_ICOUNT:
1714 mp->m_sb.sb_icount = cnt.icsb_icount;
1717 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1719 case XFS_SBS_FDBLOCKS:
1720 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1727 xfs_icsb_unlock_all_counters(mp);
1731 xfs_icsb_enable_counter(
1733 xfs_sb_field_t field,
1737 xfs_icsb_cnts_t *cntp;
1740 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1742 xfs_icsb_lock_all_counters(mp);
1743 for_each_online_cpu(i) {
1744 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1746 case XFS_SBS_ICOUNT:
1747 cntp->icsb_icount = count + resid;
1750 cntp->icsb_ifree = count + resid;
1752 case XFS_SBS_FDBLOCKS:
1753 cntp->icsb_fdblocks = count + resid;
1761 clear_bit(field, &mp->m_icsb_counters);
1762 xfs_icsb_unlock_all_counters(mp);
1766 xfs_icsb_sync_counters_locked(
1770 xfs_icsb_cnts_t cnt;
1772 xfs_icsb_count(mp, &cnt, flags);
1774 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1775 mp->m_sb.sb_icount = cnt.icsb_icount;
1776 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1777 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1778 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1779 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1783 * Accurate update of per-cpu counters to incore superblock
1786 xfs_icsb_sync_counters(
1790 spin_lock(&mp->m_sb_lock);
1791 xfs_icsb_sync_counters_locked(mp, flags);
1792 spin_unlock(&mp->m_sb_lock);
1796 * Balance and enable/disable counters as necessary.
1798 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1799 * chosen to be the same number as single on disk allocation chunk per CPU, and
1800 * free blocks is something far enough zero that we aren't going thrash when we
1801 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1802 * prevent looping endlessly when xfs_alloc_space asks for more than will
1803 * be distributed to a single CPU but each CPU has enough blocks to be
1806 * Note that we can be called when counters are already disabled.
1807 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1808 * prevent locking every per-cpu counter needlessly.
1811 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1812 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1813 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1815 xfs_icsb_balance_counter_locked(
1817 xfs_sb_field_t field,
1820 uint64_t count, resid;
1821 int weight = num_online_cpus();
1822 uint64_t min = (uint64_t)min_per_cpu;
1824 /* disable counter and sync counter */
1825 xfs_icsb_disable_counter(mp, field);
1827 /* update counters - first CPU gets residual*/
1829 case XFS_SBS_ICOUNT:
1830 count = mp->m_sb.sb_icount;
1831 resid = do_div(count, weight);
1832 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1836 count = mp->m_sb.sb_ifree;
1837 resid = do_div(count, weight);
1838 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
1841 case XFS_SBS_FDBLOCKS:
1842 count = mp->m_sb.sb_fdblocks;
1843 resid = do_div(count, weight);
1844 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
1849 count = resid = 0; /* quiet, gcc */
1853 xfs_icsb_enable_counter(mp, field, count, resid);
1857 xfs_icsb_balance_counter(
1859 xfs_sb_field_t fields,
1862 spin_lock(&mp->m_sb_lock);
1863 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1864 spin_unlock(&mp->m_sb_lock);
1868 xfs_icsb_modify_counters(
1870 xfs_sb_field_t field,
1874 xfs_icsb_cnts_t *icsbp;
1875 long long lcounter; /* long counter for 64 bit fields */
1881 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1884 * if the counter is disabled, go to slow path
1886 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1888 xfs_icsb_lock_cntr(icsbp);
1889 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1890 xfs_icsb_unlock_cntr(icsbp);
1895 case XFS_SBS_ICOUNT:
1896 lcounter = icsbp->icsb_icount;
1898 if (unlikely(lcounter < 0))
1899 goto balance_counter;
1900 icsbp->icsb_icount = lcounter;
1904 lcounter = icsbp->icsb_ifree;
1906 if (unlikely(lcounter < 0))
1907 goto balance_counter;
1908 icsbp->icsb_ifree = lcounter;
1911 case XFS_SBS_FDBLOCKS:
1912 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1914 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1916 if (unlikely(lcounter < 0))
1917 goto balance_counter;
1918 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1924 xfs_icsb_unlock_cntr(icsbp);
1932 * serialise with a mutex so we don't burn lots of cpu on
1933 * the superblock lock. We still need to hold the superblock
1934 * lock, however, when we modify the global structures.
1939 * Now running atomically.
1941 * If the counter is enabled, someone has beaten us to rebalancing.
1942 * Drop the lock and try again in the fast path....
1944 if (!(xfs_icsb_counter_disabled(mp, field))) {
1945 xfs_icsb_unlock(mp);
1950 * The counter is currently disabled. Because we are
1951 * running atomically here, we know a rebalance cannot
1952 * be in progress. Hence we can go straight to operating
1953 * on the global superblock. We do not call xfs_mod_incore_sb()
1954 * here even though we need to get the m_sb_lock. Doing so
1955 * will cause us to re-enter this function and deadlock.
1956 * Hence we get the m_sb_lock ourselves and then call
1957 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1958 * directly on the global counters.
1960 spin_lock(&mp->m_sb_lock);
1961 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1962 spin_unlock(&mp->m_sb_lock);
1965 * Now that we've modified the global superblock, we
1966 * may be able to re-enable the distributed counters
1967 * (e.g. lots of space just got freed). After that
1971 xfs_icsb_balance_counter(mp, field, 0);
1972 xfs_icsb_unlock(mp);
1976 xfs_icsb_unlock_cntr(icsbp);
1980 * We may have multiple threads here if multiple per-cpu
1981 * counters run dry at the same time. This will mean we can
1982 * do more balances than strictly necessary but it is not
1983 * the common slowpath case.
1988 * running atomically.
1990 * This will leave the counter in the correct state for future
1991 * accesses. After the rebalance, we simply try again and our retry
1992 * will either succeed through the fast path or slow path without
1993 * another balance operation being required.
1995 xfs_icsb_balance_counter(mp, field, delta);
1996 xfs_icsb_unlock(mp);