1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_rtalloc.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_priv.h"
24 #include "xfs_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_fsops.h"
27 #include "xfs_icache.h"
28 #include "xfs_sysfs.h"
29 #include "xfs_rmap_btree.h"
30 #include "xfs_refcount_btree.h"
31 #include "xfs_reflink.h"
32 #include "xfs_extent_busy.h"
33 #include "xfs_health.h"
34 #include "xfs_trace.h"
36 static DEFINE_MUTEX(xfs_uuid_table_mutex);
37 static int xfs_uuid_table_size;
38 static uuid_t *xfs_uuid_table;
41 xfs_uuid_table_free(void)
43 if (xfs_uuid_table_size == 0)
45 kmem_free(xfs_uuid_table);
46 xfs_uuid_table = NULL;
47 xfs_uuid_table_size = 0;
51 * See if the UUID is unique among mounted XFS filesystems.
52 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
58 uuid_t *uuid = &mp->m_sb.sb_uuid;
61 /* Publish UUID in struct super_block */
62 uuid_copy(&mp->m_super->s_uuid, uuid);
64 if (mp->m_flags & XFS_MOUNT_NOUUID)
67 if (uuid_is_null(uuid)) {
68 xfs_warn(mp, "Filesystem has null UUID - can't mount");
72 mutex_lock(&xfs_uuid_table_mutex);
73 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
74 if (uuid_is_null(&xfs_uuid_table[i])) {
78 if (uuid_equal(uuid, &xfs_uuid_table[i]))
83 xfs_uuid_table = krealloc(xfs_uuid_table,
84 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
85 GFP_KERNEL | __GFP_NOFAIL);
86 hole = xfs_uuid_table_size++;
88 xfs_uuid_table[hole] = *uuid;
89 mutex_unlock(&xfs_uuid_table_mutex);
94 mutex_unlock(&xfs_uuid_table_mutex);
95 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
101 struct xfs_mount *mp)
103 uuid_t *uuid = &mp->m_sb.sb_uuid;
106 if (mp->m_flags & XFS_MOUNT_NOUUID)
109 mutex_lock(&xfs_uuid_table_mutex);
110 for (i = 0; i < xfs_uuid_table_size; i++) {
111 if (uuid_is_null(&xfs_uuid_table[i]))
113 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
115 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
118 ASSERT(i < xfs_uuid_table_size);
119 mutex_unlock(&xfs_uuid_table_mutex);
125 struct rcu_head *head)
127 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
129 ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
130 ASSERT(atomic_read(&pag->pag_ref) == 0);
135 * Free up the per-ag resources associated with the mount structure.
142 struct xfs_perag *pag;
144 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
145 spin_lock(&mp->m_perag_lock);
146 pag = radix_tree_delete(&mp->m_perag_tree, agno);
147 spin_unlock(&mp->m_perag_lock);
149 ASSERT(atomic_read(&pag->pag_ref) == 0);
150 cancel_delayed_work_sync(&pag->pag_blockgc_work);
151 xfs_iunlink_destroy(pag);
152 xfs_buf_hash_destroy(pag);
153 call_rcu(&pag->rcu_head, __xfs_free_perag);
158 * Check size of device based on the (data/realtime) block count.
159 * Note: this check is used by the growfs code as well as mount.
162 xfs_sb_validate_fsb_count(
166 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
167 ASSERT(sbp->sb_blocklog >= BBSHIFT);
169 /* Limited by ULONG_MAX of page cache index */
170 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
176 xfs_initialize_perag(
178 xfs_agnumber_t agcount,
179 xfs_agnumber_t *maxagi)
181 xfs_agnumber_t index;
182 xfs_agnumber_t first_initialised = NULLAGNUMBER;
187 * Walk the current per-ag tree so we don't try to initialise AGs
188 * that already exist (growfs case). Allocate and insert all the
189 * AGs we don't find ready for initialisation.
191 for (index = 0; index < agcount; index++) {
192 pag = xfs_perag_get(mp, index);
198 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
201 goto out_unwind_new_pags;
203 pag->pag_agno = index;
205 spin_lock_init(&pag->pag_ici_lock);
206 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
207 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
209 error = xfs_buf_hash_init(pag);
212 init_waitqueue_head(&pag->pagb_wait);
213 spin_lock_init(&pag->pagb_lock);
215 pag->pagb_tree = RB_ROOT;
217 error = radix_tree_preload(GFP_NOFS);
219 goto out_hash_destroy;
221 spin_lock(&mp->m_perag_lock);
222 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
224 spin_unlock(&mp->m_perag_lock);
225 radix_tree_preload_end();
227 goto out_hash_destroy;
229 spin_unlock(&mp->m_perag_lock);
230 radix_tree_preload_end();
231 /* first new pag is fully initialized */
232 if (first_initialised == NULLAGNUMBER)
233 first_initialised = index;
234 error = xfs_iunlink_init(pag);
236 goto out_hash_destroy;
237 spin_lock_init(&pag->pag_state_lock);
240 index = xfs_set_inode_alloc(mp, agcount);
245 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
249 xfs_buf_hash_destroy(pag);
253 /* unwind any prior newly initialized pags */
254 for (index = first_initialised; index < agcount; index++) {
255 pag = radix_tree_delete(&mp->m_perag_tree, index);
258 xfs_buf_hash_destroy(pag);
259 xfs_iunlink_destroy(pag);
268 * Does the initial read of the superblock.
272 struct xfs_mount *mp,
275 unsigned int sector_size;
277 struct xfs_sb *sbp = &mp->m_sb;
279 int loud = !(flags & XFS_MFSI_QUIET);
280 const struct xfs_buf_ops *buf_ops;
282 ASSERT(mp->m_sb_bp == NULL);
283 ASSERT(mp->m_ddev_targp != NULL);
286 * For the initial read, we must guess at the sector
287 * size based on the block device. It's enough to
288 * get the sb_sectsize out of the superblock and
289 * then reread with the proper length.
290 * We don't verify it yet, because it may not be complete.
292 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
296 * Allocate a (locked) buffer to hold the superblock. This will be kept
297 * around at all times to optimize access to the superblock. Therefore,
298 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
302 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
303 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
307 xfs_warn(mp, "SB validate failed with error %d.", error);
308 /* bad CRC means corrupted metadata */
309 if (error == -EFSBADCRC)
310 error = -EFSCORRUPTED;
315 * Initialize the mount structure from the superblock.
317 xfs_sb_from_disk(sbp, bp->b_addr);
320 * If we haven't validated the superblock, do so now before we try
321 * to check the sector size and reread the superblock appropriately.
323 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
325 xfs_warn(mp, "Invalid superblock magic number");
331 * We must be able to do sector-sized and sector-aligned IO.
333 if (sector_size > sbp->sb_sectsize) {
335 xfs_warn(mp, "device supports %u byte sectors (not %u)",
336 sector_size, sbp->sb_sectsize);
341 if (buf_ops == NULL) {
343 * Re-read the superblock so the buffer is correctly sized,
344 * and properly verified.
347 sector_size = sbp->sb_sectsize;
348 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
352 xfs_reinit_percpu_counters(mp);
354 /* no need to be quiet anymore, so reset the buf ops */
355 bp->b_ops = &xfs_sb_buf_ops;
367 * If the sunit/swidth change would move the precomputed root inode value, we
368 * must reject the ondisk change because repair will stumble over that.
369 * However, we allow the mount to proceed because we never rejected this
370 * combination before. Returns true to update the sb, false otherwise.
373 xfs_check_new_dalign(
374 struct xfs_mount *mp,
378 struct xfs_sb *sbp = &mp->m_sb;
381 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
382 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
384 if (sbp->sb_rootino == calc_ino) {
390 "Cannot change stripe alignment; would require moving root inode.");
393 * XXX: Next time we add a new incompat feature, this should start
394 * returning -EINVAL to fail the mount. Until then, spit out a warning
395 * that we're ignoring the administrator's instructions.
397 xfs_warn(mp, "Skipping superblock stripe alignment update.");
403 * If we were provided with new sunit/swidth values as mount options, make sure
404 * that they pass basic alignment and superblock feature checks, and convert
405 * them into the same units (FSB) that everything else expects. This step
406 * /must/ be done before computing the inode geometry.
409 xfs_validate_new_dalign(
410 struct xfs_mount *mp)
412 if (mp->m_dalign == 0)
416 * If stripe unit and stripe width are not multiples
417 * of the fs blocksize turn off alignment.
419 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
420 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
422 "alignment check failed: sunit/swidth vs. blocksize(%d)",
423 mp->m_sb.sb_blocksize);
427 * Convert the stripe unit and width to FSBs.
429 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
430 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
432 "alignment check failed: sunit/swidth vs. agsize(%d)",
433 mp->m_sb.sb_agblocks);
435 } else if (mp->m_dalign) {
436 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
439 "alignment check failed: sunit(%d) less than bsize(%d)",
440 mp->m_dalign, mp->m_sb.sb_blocksize);
445 if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
447 "cannot change alignment: superblock does not support data alignment");
454 /* Update alignment values based on mount options and sb values. */
456 xfs_update_alignment(
457 struct xfs_mount *mp)
459 struct xfs_sb *sbp = &mp->m_sb;
465 if (sbp->sb_unit == mp->m_dalign &&
466 sbp->sb_width == mp->m_swidth)
469 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
470 if (error || !update_sb)
473 sbp->sb_unit = mp->m_dalign;
474 sbp->sb_width = mp->m_swidth;
475 mp->m_update_sb = true;
476 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
477 xfs_sb_version_hasdalign(&mp->m_sb)) {
478 mp->m_dalign = sbp->sb_unit;
479 mp->m_swidth = sbp->sb_width;
486 * precalculate the low space thresholds for dynamic speculative preallocation.
489 xfs_set_low_space_thresholds(
490 struct xfs_mount *mp)
494 for (i = 0; i < XFS_LOWSP_MAX; i++) {
495 uint64_t space = mp->m_sb.sb_dblocks;
498 mp->m_low_space[i] = space * (i + 1);
503 * Check that the data (and log if separate) is an ok size.
507 struct xfs_mount *mp)
513 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
514 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
515 xfs_warn(mp, "filesystem size mismatch detected");
518 error = xfs_buf_read_uncached(mp->m_ddev_targp,
519 d - XFS_FSS_TO_BB(mp, 1),
520 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
522 xfs_warn(mp, "last sector read failed");
527 if (mp->m_logdev_targp == mp->m_ddev_targp)
530 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
531 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
532 xfs_warn(mp, "log size mismatch detected");
535 error = xfs_buf_read_uncached(mp->m_logdev_targp,
536 d - XFS_FSB_TO_BB(mp, 1),
537 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
539 xfs_warn(mp, "log device read failed");
547 * Clear the quotaflags in memory and in the superblock.
550 xfs_mount_reset_sbqflags(
551 struct xfs_mount *mp)
555 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
556 if (mp->m_sb.sb_qflags == 0)
558 spin_lock(&mp->m_sb_lock);
559 mp->m_sb.sb_qflags = 0;
560 spin_unlock(&mp->m_sb_lock);
562 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
565 return xfs_sync_sb(mp, false);
569 xfs_default_resblks(xfs_mount_t *mp)
574 * We default to 5% or 8192 fsbs of space reserved, whichever is
575 * smaller. This is intended to cover concurrent allocation
576 * transactions when we initially hit enospc. These each require a 4
577 * block reservation. Hence by default we cover roughly 2000 concurrent
578 * allocation reservations.
580 resblks = mp->m_sb.sb_dblocks;
582 resblks = min_t(uint64_t, resblks, 8192);
586 /* Ensure the summary counts are correct. */
588 xfs_check_summary_counts(
589 struct xfs_mount *mp)
592 * The AG0 superblock verifier rejects in-progress filesystems,
593 * so we should never see the flag set this far into mounting.
595 if (mp->m_sb.sb_inprogress) {
596 xfs_err(mp, "sb_inprogress set after log recovery??");
598 return -EFSCORRUPTED;
602 * Now the log is mounted, we know if it was an unclean shutdown or
603 * not. If it was, with the first phase of recovery has completed, we
604 * have consistent AG blocks on disk. We have not recovered EFIs yet,
605 * but they are recovered transactionally in the second recovery phase
608 * If the log was clean when we mounted, we can check the summary
609 * counters. If any of them are obviously incorrect, we can recompute
610 * them from the AGF headers in the next step.
612 if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
613 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
614 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
615 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
616 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
619 * We can safely re-initialise incore superblock counters from the
620 * per-ag data. These may not be correct if the filesystem was not
621 * cleanly unmounted, so we waited for recovery to finish before doing
624 * If the filesystem was cleanly unmounted or the previous check did
625 * not flag anything weird, then we can trust the values in the
626 * superblock to be correct and we don't need to do anything here.
627 * Otherwise, recalculate the summary counters.
629 if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
630 XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
631 !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
634 return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
638 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
639 * internal inode structures can be sitting in the CIL and AIL at this point,
640 * so we need to unpin them, write them back and/or reclaim them before unmount
643 * An inode cluster that has been freed can have its buffer still pinned in
644 * memory because the transaction is still sitting in a iclog. The stale inodes
645 * on that buffer will be pinned to the buffer until the transaction hits the
646 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
647 * may never see the pinned buffer, so nothing will push out the iclog and
650 * Hence we need to force the log to unpin everything first. However, log
651 * forces don't wait for the discards they issue to complete, so we have to
652 * explicitly wait for them to complete here as well.
654 * Then we can tell the world we are unmounting so that error handling knows
655 * that the filesystem is going away and we should error out anything that we
656 * have been retrying in the background. This will prevent never-ending
657 * retries in AIL pushing from hanging the unmount.
659 * Finally, we can push the AIL to clean all the remaining dirty objects, then
660 * reclaim the remaining inodes that are still in memory at this point in time.
663 xfs_unmount_flush_inodes(
664 struct xfs_mount *mp)
666 xfs_log_force(mp, XFS_LOG_SYNC);
667 xfs_extent_busy_wait_all(mp);
668 flush_workqueue(xfs_discard_wq);
670 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
672 xfs_ail_push_all_sync(mp->m_ail);
673 cancel_delayed_work_sync(&mp->m_reclaim_work);
674 xfs_reclaim_inodes(mp);
675 xfs_health_unmount(mp);
679 xfs_mount_setup_inode_geom(
680 struct xfs_mount *mp)
682 struct xfs_ino_geometry *igeo = M_IGEO(mp);
684 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
685 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
687 xfs_ialloc_setup_geometry(mp);
691 * This function does the following on an initial mount of a file system:
692 * - reads the superblock from disk and init the mount struct
693 * - if we're a 32-bit kernel, do a size check on the superblock
694 * so we don't mount terabyte filesystems
695 * - init mount struct realtime fields
696 * - allocate inode hash table for fs
697 * - init directory manager
698 * - perform recovery and init the log manager
702 struct xfs_mount *mp)
704 struct xfs_sb *sbp = &(mp->m_sb);
705 struct xfs_inode *rip;
706 struct xfs_ino_geometry *igeo = M_IGEO(mp);
712 xfs_sb_mount_common(mp, sbp);
715 * Check for a mismatched features2 values. Older kernels read & wrote
716 * into the wrong sb offset for sb_features2 on some platforms due to
717 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
718 * which made older superblock reading/writing routines swap it as a
721 * For backwards compatibility, we make both slots equal.
723 * If we detect a mismatched field, we OR the set bits into the existing
724 * features2 field in case it has already been modified; we don't want
725 * to lose any features. We then update the bad location with the ORed
726 * value so that older kernels will see any features2 flags. The
727 * superblock writeback code ensures the new sb_features2 is copied to
728 * sb_bad_features2 before it is logged or written to disk.
730 if (xfs_sb_has_mismatched_features2(sbp)) {
731 xfs_warn(mp, "correcting sb_features alignment problem");
732 sbp->sb_features2 |= sbp->sb_bad_features2;
733 mp->m_update_sb = true;
736 * Re-check for ATTR2 in case it was found in bad_features2
739 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
740 !(mp->m_flags & XFS_MOUNT_NOATTR2))
741 mp->m_flags |= XFS_MOUNT_ATTR2;
744 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
745 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
746 xfs_sb_version_removeattr2(&mp->m_sb);
747 mp->m_update_sb = true;
749 /* update sb_versionnum for the clearing of the morebits */
750 if (!sbp->sb_features2)
751 mp->m_update_sb = true;
754 /* always use v2 inodes by default now */
755 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
756 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
757 mp->m_update_sb = true;
761 * If we were given new sunit/swidth options, do some basic validation
762 * checks and convert the incore dalign and swidth values to the
763 * same units (FSB) that everything else uses. This /must/ happen
764 * before computing the inode geometry.
766 error = xfs_validate_new_dalign(mp);
770 xfs_alloc_compute_maxlevels(mp);
771 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
772 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
773 xfs_mount_setup_inode_geom(mp);
774 xfs_rmapbt_compute_maxlevels(mp);
775 xfs_refcountbt_compute_maxlevels(mp);
778 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
779 * is NOT aligned turn off m_dalign since allocator alignment is within
780 * an ag, therefore ag has to be aligned at stripe boundary. Note that
781 * we must compute the free space and rmap btree geometry before doing
784 error = xfs_update_alignment(mp);
788 /* enable fail_at_unmount as default */
789 mp->m_fail_unmount = true;
791 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
792 NULL, mp->m_super->s_id);
796 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
797 &mp->m_kobj, "stats");
799 goto out_remove_sysfs;
801 error = xfs_error_sysfs_init(mp);
805 error = xfs_errortag_init(mp);
807 goto out_remove_error_sysfs;
809 error = xfs_uuid_mount(mp);
811 goto out_remove_errortag;
814 * Update the preferred write size based on the information from the
815 * on-disk superblock.
817 mp->m_allocsize_log =
818 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
819 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
821 /* set the low space thresholds for dynamic preallocation */
822 xfs_set_low_space_thresholds(mp);
825 * If enabled, sparse inode chunk alignment is expected to match the
826 * cluster size. Full inode chunk alignment must match the chunk size,
827 * but that is checked on sb read verification...
829 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
830 mp->m_sb.sb_spino_align !=
831 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
833 "Sparse inode block alignment (%u) must match cluster size (%llu).",
834 mp->m_sb.sb_spino_align,
835 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
837 goto out_remove_uuid;
841 * Check that the data (and log if separate) is an ok size.
843 error = xfs_check_sizes(mp);
845 goto out_remove_uuid;
848 * Initialize realtime fields in the mount structure
850 error = xfs_rtmount_init(mp);
852 xfs_warn(mp, "RT mount failed");
853 goto out_remove_uuid;
857 * Copies the low order bits of the timestamp and the randomly
858 * set "sequence" number out of a UUID.
861 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
862 get_unaligned_be16(&sbp->sb_uuid.b[4]);
863 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
865 error = xfs_da_mount(mp);
867 xfs_warn(mp, "Failed dir/attr init: %d", error);
868 goto out_remove_uuid;
872 * Initialize the precomputed transaction reservations values.
877 * Allocate and initialize the per-ag data.
879 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
881 xfs_warn(mp, "Failed per-ag init: %d", error);
885 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
886 xfs_warn(mp, "no log defined");
887 error = -EFSCORRUPTED;
892 * Log's mount-time initialization. The first part of recovery can place
893 * some items on the AIL, to be handled when recovery is finished or
896 error = xfs_log_mount(mp, mp->m_logdev_targp,
897 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
898 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
900 xfs_warn(mp, "log mount failed");
904 /* Make sure the summary counts are ok. */
905 error = xfs_check_summary_counts(mp);
907 goto out_log_dealloc;
910 * Get and sanity-check the root inode.
911 * Save the pointer to it in the mount structure.
913 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
914 XFS_ILOCK_EXCL, &rip);
917 "Failed to read root inode 0x%llx, error %d",
918 sbp->sb_rootino, -error);
919 goto out_log_dealloc;
924 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
925 xfs_warn(mp, "corrupted root inode %llu: not a directory",
926 (unsigned long long)rip->i_ino);
927 xfs_iunlock(rip, XFS_ILOCK_EXCL);
928 error = -EFSCORRUPTED;
931 mp->m_rootip = rip; /* save it */
933 xfs_iunlock(rip, XFS_ILOCK_EXCL);
936 * Initialize realtime inode pointers in the mount structure
938 error = xfs_rtmount_inodes(mp);
941 * Free up the root inode.
943 xfs_warn(mp, "failed to read RT inodes");
948 * If this is a read-only mount defer the superblock updates until
949 * the next remount into writeable mode. Otherwise we would never
950 * perform the update e.g. for the root filesystem.
952 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
953 error = xfs_sync_sb(mp, false);
955 xfs_warn(mp, "failed to write sb changes");
961 * Initialise the XFS quota management subsystem for this mount
963 if (XFS_IS_QUOTA_RUNNING(mp)) {
964 error = xfs_qm_newmount(mp, "amount, "aflags);
968 ASSERT(!XFS_IS_QUOTA_ON(mp));
971 * If a file system had quotas running earlier, but decided to
972 * mount without -o uquota/pquota/gquota options, revoke the
973 * quotachecked license.
975 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
976 xfs_notice(mp, "resetting quota flags");
977 error = xfs_mount_reset_sbqflags(mp);
984 * Finish recovering the file system. This part needed to be delayed
985 * until after the root and real-time bitmap inodes were consistently
988 error = xfs_log_mount_finish(mp);
990 xfs_warn(mp, "log mount finish failed");
995 * Now the log is fully replayed, we can transition to full read-only
996 * mode for read-only mounts. This will sync all the metadata and clean
997 * the log so that the recovery we just performed does not have to be
998 * replayed again on the next mount.
1000 * We use the same quiesce mechanism as the rw->ro remount, as they are
1001 * semantically identical operations.
1003 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
1009 * Complete the quota initialisation, post-log-replay component.
1012 ASSERT(mp->m_qflags == 0);
1013 mp->m_qflags = quotaflags;
1015 xfs_qm_mount_quotas(mp);
1019 * Now we are mounted, reserve a small amount of unused space for
1020 * privileged transactions. This is needed so that transaction
1021 * space required for critical operations can dip into this pool
1022 * when at ENOSPC. This is needed for operations like create with
1023 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1024 * are not allowed to use this reserved space.
1026 * This may drive us straight to ENOSPC on mount, but that implies
1027 * we were already there on the last unmount. Warn if this occurs.
1029 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1030 resblks = xfs_default_resblks(mp);
1031 error = xfs_reserve_blocks(mp, &resblks, NULL);
1034 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1036 /* Recover any CoW blocks that never got remapped. */
1037 error = xfs_reflink_recover_cow(mp);
1040 "Error %d recovering leftover CoW allocations.", error);
1041 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1045 /* Reserve AG blocks for future btree expansion. */
1046 error = xfs_fs_reserve_ag_blocks(mp);
1047 if (error && error != -ENOSPC)
1054 xfs_fs_unreserve_ag_blocks(mp);
1056 xfs_qm_unmount_quotas(mp);
1058 xfs_rtunmount_inodes(mp);
1061 /* Clean out dquots that might be in memory after quotacheck. */
1064 * Flush all inode reclamation work and flush the log.
1065 * We have to do this /after/ rtunmount and qm_unmount because those
1066 * two will have scheduled delayed reclaim for the rt/quota inodes.
1068 * This is slightly different from the unmountfs call sequence
1069 * because we could be tearing down a partially set up mount. In
1070 * particular, if log_mount_finish fails we bail out without calling
1071 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1074 xfs_unmount_flush_inodes(mp);
1076 xfs_log_mount_cancel(mp);
1078 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1079 xfs_buftarg_drain(mp->m_logdev_targp);
1080 xfs_buftarg_drain(mp->m_ddev_targp);
1086 xfs_uuid_unmount(mp);
1087 out_remove_errortag:
1088 xfs_errortag_del(mp);
1089 out_remove_error_sysfs:
1090 xfs_error_sysfs_del(mp);
1092 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1094 xfs_sysfs_del(&mp->m_kobj);
1100 * This flushes out the inodes,dquots and the superblock, unmounts the
1101 * log and makes sure that incore structures are freed.
1105 struct xfs_mount *mp)
1110 xfs_blockgc_stop(mp);
1111 xfs_fs_unreserve_ag_blocks(mp);
1112 xfs_qm_unmount_quotas(mp);
1113 xfs_rtunmount_inodes(mp);
1114 xfs_irele(mp->m_rootip);
1116 xfs_unmount_flush_inodes(mp);
1121 * Unreserve any blocks we have so that when we unmount we don't account
1122 * the reserved free space as used. This is really only necessary for
1123 * lazy superblock counting because it trusts the incore superblock
1124 * counters to be absolutely correct on clean unmount.
1126 * We don't bother correcting this elsewhere for lazy superblock
1127 * counting because on mount of an unclean filesystem we reconstruct the
1128 * correct counter value and this is irrelevant.
1130 * For non-lazy counter filesystems, this doesn't matter at all because
1131 * we only every apply deltas to the superblock and hence the incore
1132 * value does not matter....
1135 error = xfs_reserve_blocks(mp, &resblks, NULL);
1137 xfs_warn(mp, "Unable to free reserved block pool. "
1138 "Freespace may not be correct on next mount.");
1140 xfs_log_unmount(mp);
1142 xfs_uuid_unmount(mp);
1145 xfs_errortag_clearall(mp);
1149 xfs_errortag_del(mp);
1150 xfs_error_sysfs_del(mp);
1151 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1152 xfs_sysfs_del(&mp->m_kobj);
1156 * Determine whether modifications can proceed. The caller specifies the minimum
1157 * freeze level for which modifications should not be allowed. This allows
1158 * certain operations to proceed while the freeze sequence is in progress, if
1163 struct xfs_mount *mp,
1166 ASSERT(level > SB_UNFROZEN);
1167 if ((mp->m_super->s_writers.frozen >= level) ||
1168 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1175 * Deltas for the block count can vary from 1 to very large, but lock contention
1176 * only occurs on frequent small block count updates such as in the delayed
1177 * allocation path for buffered writes (page a time updates). Hence we set
1178 * a large batch count (1024) to minimise global counter updates except when
1179 * we get near to ENOSPC and we have to be very accurate with our updates.
1181 #define XFS_FDBLOCKS_BATCH 1024
1184 struct xfs_mount *mp,
1195 * If the reserve pool is depleted, put blocks back into it
1196 * first. Most of the time the pool is full.
1198 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1199 percpu_counter_add(&mp->m_fdblocks, delta);
1203 spin_lock(&mp->m_sb_lock);
1204 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1206 if (res_used > delta) {
1207 mp->m_resblks_avail += delta;
1210 mp->m_resblks_avail = mp->m_resblks;
1211 percpu_counter_add(&mp->m_fdblocks, delta);
1213 spin_unlock(&mp->m_sb_lock);
1218 * Taking blocks away, need to be more accurate the closer we
1221 * If the counter has a value of less than 2 * max batch size,
1222 * then make everything serialise as we are real close to
1225 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1226 XFS_FDBLOCKS_BATCH) < 0)
1229 batch = XFS_FDBLOCKS_BATCH;
1232 * Set aside allocbt blocks because these blocks are tracked as free
1233 * space but not available for allocation. Technically this means that a
1234 * single reservation cannot consume all remaining free space, but the
1235 * ratio of allocbt blocks to usable free blocks should be rather small.
1236 * The tradeoff without this is that filesystems that maintain high
1237 * perag block reservations can over reserve physical block availability
1238 * and fail physical allocation, which leads to much more serious
1239 * problems (i.e. transaction abort, pagecache discards, etc.) than
1240 * slightly premature -ENOSPC.
1242 set_aside = mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks);
1243 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1244 if (__percpu_counter_compare(&mp->m_fdblocks, set_aside,
1245 XFS_FDBLOCKS_BATCH) >= 0) {
1251 * lock up the sb for dipping into reserves before releasing the space
1252 * that took us to ENOSPC.
1254 spin_lock(&mp->m_sb_lock);
1255 percpu_counter_add(&mp->m_fdblocks, -delta);
1257 goto fdblocks_enospc;
1259 lcounter = (long long)mp->m_resblks_avail + delta;
1260 if (lcounter >= 0) {
1261 mp->m_resblks_avail = lcounter;
1262 spin_unlock(&mp->m_sb_lock);
1266 "Reserve blocks depleted! Consider increasing reserve pool size.");
1269 spin_unlock(&mp->m_sb_lock);
1275 struct xfs_mount *mp,
1281 spin_lock(&mp->m_sb_lock);
1282 lcounter = mp->m_sb.sb_frextents + delta;
1286 mp->m_sb.sb_frextents = lcounter;
1287 spin_unlock(&mp->m_sb_lock);
1292 * Used to free the superblock along various error paths.
1296 struct xfs_mount *mp)
1298 struct xfs_buf *bp = mp->m_sb_bp;
1306 * If the underlying (data/log/rt) device is readonly, there are some
1307 * operations that cannot proceed.
1310 xfs_dev_is_read_only(
1311 struct xfs_mount *mp,
1314 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1315 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1316 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1317 xfs_notice(mp, "%s required on read-only device.", message);
1318 xfs_notice(mp, "write access unavailable, cannot proceed.");
1324 /* Force the summary counters to be recalculated at next mount. */
1326 xfs_force_summary_recalc(
1327 struct xfs_mount *mp)
1329 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1332 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1336 * Update the in-core delayed block counter.
1338 * We prefer to update the counter without having to take a spinlock for every
1339 * counter update (i.e. batching). Each change to delayed allocation
1340 * reservations can change can easily exceed the default percpu counter
1341 * batching, so we use a larger batch factor here.
1343 * Note that we don't currently have any callers requiring fast summation
1344 * (e.g. percpu_counter_read) so we can use a big batch value here.
1346 #define XFS_DELALLOC_BATCH (4096)
1349 struct xfs_mount *mp,
1352 percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1353 XFS_DELALLOC_BATCH);