1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 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"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
17 #include "xfs_alloc.h"
18 #include "xfs_fsops.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
22 #include "xfs_log_priv.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_mru_cache.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_icache.h"
28 #include "xfs_trace.h"
29 #include "xfs_icreate_item.h"
30 #include "xfs_filestream.h"
31 #include "xfs_quota.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_ondisk.h"
34 #include "xfs_rmap_item.h"
35 #include "xfs_refcount_item.h"
36 #include "xfs_bmap_item.h"
37 #include "xfs_reflink.h"
38 #include "xfs_pwork.h"
41 #include <linux/magic.h>
42 #include <linux/fs_context.h>
43 #include <linux/fs_parser.h>
45 static const struct super_operations xfs_super_operations;
47 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
49 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
59 xfs_mount_set_dax_mode(
61 enum xfs_dax_mode mode)
65 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
68 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
69 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
72 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
73 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
78 static const struct constant_table dax_param_enums[] = {
79 {"inode", XFS_DAX_INODE },
80 {"always", XFS_DAX_ALWAYS },
81 {"never", XFS_DAX_NEVER },
86 * Table driven mount option parser.
89 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
90 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
91 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
92 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
93 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
94 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
95 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
96 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
97 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
100 static const struct fs_parameter_spec xfs_fs_parameters[] = {
101 fsparam_u32("logbufs", Opt_logbufs),
102 fsparam_string("logbsize", Opt_logbsize),
103 fsparam_string("logdev", Opt_logdev),
104 fsparam_string("rtdev", Opt_rtdev),
105 fsparam_flag("wsync", Opt_wsync),
106 fsparam_flag("noalign", Opt_noalign),
107 fsparam_flag("swalloc", Opt_swalloc),
108 fsparam_u32("sunit", Opt_sunit),
109 fsparam_u32("swidth", Opt_swidth),
110 fsparam_flag("nouuid", Opt_nouuid),
111 fsparam_flag("grpid", Opt_grpid),
112 fsparam_flag("nogrpid", Opt_nogrpid),
113 fsparam_flag("bsdgroups", Opt_bsdgroups),
114 fsparam_flag("sysvgroups", Opt_sysvgroups),
115 fsparam_string("allocsize", Opt_allocsize),
116 fsparam_flag("norecovery", Opt_norecovery),
117 fsparam_flag("inode64", Opt_inode64),
118 fsparam_flag("inode32", Opt_inode32),
119 fsparam_flag("ikeep", Opt_ikeep),
120 fsparam_flag("noikeep", Opt_noikeep),
121 fsparam_flag("largeio", Opt_largeio),
122 fsparam_flag("nolargeio", Opt_nolargeio),
123 fsparam_flag("attr2", Opt_attr2),
124 fsparam_flag("noattr2", Opt_noattr2),
125 fsparam_flag("filestreams", Opt_filestreams),
126 fsparam_flag("quota", Opt_quota),
127 fsparam_flag("noquota", Opt_noquota),
128 fsparam_flag("usrquota", Opt_usrquota),
129 fsparam_flag("grpquota", Opt_grpquota),
130 fsparam_flag("prjquota", Opt_prjquota),
131 fsparam_flag("uquota", Opt_uquota),
132 fsparam_flag("gquota", Opt_gquota),
133 fsparam_flag("pquota", Opt_pquota),
134 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
135 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
136 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
137 fsparam_flag("qnoenforce", Opt_qnoenforce),
138 fsparam_flag("discard", Opt_discard),
139 fsparam_flag("nodiscard", Opt_nodiscard),
140 fsparam_flag("dax", Opt_dax),
141 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
145 struct proc_xfs_info {
155 static struct proc_xfs_info xfs_info_set[] = {
156 /* the few simple ones we can get from the mount struct */
157 { XFS_MOUNT_IKEEP, ",ikeep" },
158 { XFS_MOUNT_WSYNC, ",wsync" },
159 { XFS_MOUNT_NOALIGN, ",noalign" },
160 { XFS_MOUNT_SWALLOC, ",swalloc" },
161 { XFS_MOUNT_NOUUID, ",nouuid" },
162 { XFS_MOUNT_NORECOVERY, ",norecovery" },
163 { XFS_MOUNT_ATTR2, ",attr2" },
164 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
165 { XFS_MOUNT_GRPID, ",grpid" },
166 { XFS_MOUNT_DISCARD, ",discard" },
167 { XFS_MOUNT_LARGEIO, ",largeio" },
168 { XFS_MOUNT_DAX_ALWAYS, ",dax=always" },
169 { XFS_MOUNT_DAX_NEVER, ",dax=never" },
172 struct xfs_mount *mp = XFS_M(root->d_sb);
173 struct proc_xfs_info *xfs_infop;
175 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
176 if (mp->m_flags & xfs_infop->flag)
177 seq_puts(m, xfs_infop->str);
180 seq_printf(m, ",inode%d",
181 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
183 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
184 seq_printf(m, ",allocsize=%dk",
185 (1 << mp->m_allocsize_log) >> 10);
187 if (mp->m_logbufs > 0)
188 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
189 if (mp->m_logbsize > 0)
190 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
193 seq_show_option(m, "logdev", mp->m_logname);
195 seq_show_option(m, "rtdev", mp->m_rtname);
197 if (mp->m_dalign > 0)
198 seq_printf(m, ",sunit=%d",
199 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
200 if (mp->m_swidth > 0)
201 seq_printf(m, ",swidth=%d",
202 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
204 if (mp->m_qflags & XFS_UQUOTA_ACCT) {
205 if (mp->m_qflags & XFS_UQUOTA_ENFD)
206 seq_puts(m, ",usrquota");
208 seq_puts(m, ",uqnoenforce");
211 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
212 if (mp->m_qflags & XFS_PQUOTA_ENFD)
213 seq_puts(m, ",prjquota");
215 seq_puts(m, ",pqnoenforce");
217 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
218 if (mp->m_qflags & XFS_GQUOTA_ENFD)
219 seq_puts(m, ",grpquota");
221 seq_puts(m, ",gqnoenforce");
224 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
225 seq_puts(m, ",noquota");
231 * Set parameters for inode allocation heuristics, taking into account
232 * filesystem size and inode32/inode64 mount options; i.e. specifically
233 * whether or not XFS_MOUNT_SMALL_INUMS is set.
235 * Inode allocation patterns are altered only if inode32 is requested
236 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
237 * If altered, XFS_MOUNT_32BITINODES is set as well.
239 * An agcount independent of that in the mount structure is provided
240 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
241 * to the potentially higher ag count.
243 * Returns the maximum AG index which may contain inodes.
247 struct xfs_mount *mp,
248 xfs_agnumber_t agcount)
250 xfs_agnumber_t index;
251 xfs_agnumber_t maxagi = 0;
252 xfs_sb_t *sbp = &mp->m_sb;
253 xfs_agnumber_t max_metadata;
258 * Calculate how much should be reserved for inodes to meet
259 * the max inode percentage. Used only for inode32.
261 if (M_IGEO(mp)->maxicount) {
264 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
266 icount += sbp->sb_agblocks - 1;
267 do_div(icount, sbp->sb_agblocks);
268 max_metadata = icount;
270 max_metadata = agcount;
273 /* Get the last possible inode in the filesystem */
274 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
275 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
278 * If user asked for no more than 32-bit inodes, and the fs is
279 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
280 * the allocator to accommodate the request.
282 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
283 mp->m_flags |= XFS_MOUNT_32BITINODES;
285 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
287 for (index = 0; index < agcount; index++) {
288 struct xfs_perag *pag;
290 ino = XFS_AGINO_TO_INO(mp, index, agino);
292 pag = xfs_perag_get(mp, index);
294 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
295 if (ino > XFS_MAXINUMBER_32) {
296 pag->pagi_inodeok = 0;
297 pag->pagf_metadata = 0;
299 pag->pagi_inodeok = 1;
301 if (index < max_metadata)
302 pag->pagf_metadata = 1;
304 pag->pagf_metadata = 0;
307 pag->pagi_inodeok = 1;
308 pag->pagf_metadata = 0;
314 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
321 struct block_device **bdevp)
325 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
327 if (IS_ERR(*bdevp)) {
328 error = PTR_ERR(*bdevp);
329 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
337 struct block_device *bdev)
340 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
345 struct xfs_mount *mp)
347 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
349 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
350 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
351 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
353 xfs_free_buftarg(mp->m_logdev_targp);
354 xfs_blkdev_put(logdev);
355 fs_put_dax(dax_logdev);
357 if (mp->m_rtdev_targp) {
358 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
359 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
361 xfs_free_buftarg(mp->m_rtdev_targp);
362 xfs_blkdev_put(rtdev);
363 fs_put_dax(dax_rtdev);
365 xfs_free_buftarg(mp->m_ddev_targp);
366 fs_put_dax(dax_ddev);
370 * The file system configurations are:
371 * (1) device (partition) with data and internal log
372 * (2) logical volume with data and log subvolumes.
373 * (3) logical volume with data, log, and realtime subvolumes.
375 * We only have to handle opening the log and realtime volumes here if
376 * they are present. The data subvolume has already been opened by
377 * get_sb_bdev() and is stored in sb->s_bdev.
381 struct xfs_mount *mp)
383 struct block_device *ddev = mp->m_super->s_bdev;
384 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
385 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
386 struct block_device *logdev = NULL, *rtdev = NULL;
390 * Open real time and log devices - order is important.
393 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
396 dax_logdev = fs_dax_get_by_bdev(logdev);
400 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
402 goto out_close_logdev;
404 if (rtdev == ddev || rtdev == logdev) {
406 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
408 goto out_close_rtdev;
410 dax_rtdev = fs_dax_get_by_bdev(rtdev);
414 * Setup xfs_mount buffer target pointers
417 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
418 if (!mp->m_ddev_targp)
419 goto out_close_rtdev;
422 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
423 if (!mp->m_rtdev_targp)
424 goto out_free_ddev_targ;
427 if (logdev && logdev != ddev) {
428 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
429 if (!mp->m_logdev_targp)
430 goto out_free_rtdev_targ;
432 mp->m_logdev_targp = mp->m_ddev_targp;
438 if (mp->m_rtdev_targp)
439 xfs_free_buftarg(mp->m_rtdev_targp);
441 xfs_free_buftarg(mp->m_ddev_targp);
443 xfs_blkdev_put(rtdev);
444 fs_put_dax(dax_rtdev);
446 if (logdev && logdev != ddev) {
447 xfs_blkdev_put(logdev);
448 fs_put_dax(dax_logdev);
451 fs_put_dax(dax_ddev);
456 * Setup xfs_mount buffer target pointers based on superblock
460 struct xfs_mount *mp)
464 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
468 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
469 unsigned int log_sector_size = BBSIZE;
471 if (xfs_sb_version_hassector(&mp->m_sb))
472 log_sector_size = mp->m_sb.sb_logsectsize;
473 error = xfs_setsize_buftarg(mp->m_logdev_targp,
478 if (mp->m_rtdev_targp) {
479 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
480 mp->m_sb.sb_sectsize);
489 xfs_init_mount_workqueues(
490 struct xfs_mount *mp)
492 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
493 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
494 1, mp->m_super->s_id);
495 if (!mp->m_buf_workqueue)
498 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
499 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
500 0, mp->m_super->s_id);
501 if (!mp->m_unwritten_workqueue)
502 goto out_destroy_buf;
504 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
505 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_UNBOUND),
506 0, mp->m_super->s_id);
507 if (!mp->m_cil_workqueue)
508 goto out_destroy_unwritten;
510 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
511 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
512 0, mp->m_super->s_id);
513 if (!mp->m_reclaim_workqueue)
514 goto out_destroy_cil;
516 mp->m_gc_workqueue = alloc_workqueue("xfs-gc/%s",
517 WQ_SYSFS | WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM,
518 0, mp->m_super->s_id);
519 if (!mp->m_gc_workqueue)
520 goto out_destroy_reclaim;
522 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
523 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
524 if (!mp->m_sync_workqueue)
525 goto out_destroy_eofb;
530 destroy_workqueue(mp->m_gc_workqueue);
532 destroy_workqueue(mp->m_reclaim_workqueue);
534 destroy_workqueue(mp->m_cil_workqueue);
535 out_destroy_unwritten:
536 destroy_workqueue(mp->m_unwritten_workqueue);
538 destroy_workqueue(mp->m_buf_workqueue);
544 xfs_destroy_mount_workqueues(
545 struct xfs_mount *mp)
547 destroy_workqueue(mp->m_sync_workqueue);
548 destroy_workqueue(mp->m_gc_workqueue);
549 destroy_workqueue(mp->m_reclaim_workqueue);
550 destroy_workqueue(mp->m_cil_workqueue);
551 destroy_workqueue(mp->m_unwritten_workqueue);
552 destroy_workqueue(mp->m_buf_workqueue);
556 xfs_flush_inodes_worker(
557 struct work_struct *work)
559 struct xfs_mount *mp = container_of(work, struct xfs_mount,
560 m_flush_inodes_work);
561 struct super_block *sb = mp->m_super;
563 if (down_read_trylock(&sb->s_umount)) {
565 up_read(&sb->s_umount);
570 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
571 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
572 * for IO to complete so that we effectively throttle multiple callers to the
573 * rate at which IO is completing.
577 struct xfs_mount *mp)
580 * If flush_work() returns true then that means we waited for a flush
581 * which was already in progress. Don't bother running another scan.
583 if (flush_work(&mp->m_flush_inodes_work))
586 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
587 flush_work(&mp->m_flush_inodes_work);
590 /* Catch misguided souls that try to use this interface on XFS */
591 STATIC struct inode *
593 struct super_block *sb)
602 struct xfs_inode *ip,
605 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
606 struct xfs_bmbt_irec got;
607 struct xfs_iext_cursor icur;
609 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
612 if (isnullstartblock(got.br_startblock)) {
613 xfs_warn(ip->i_mount,
614 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
616 whichfork == XFS_DATA_FORK ? "data" : "cow",
617 got.br_startoff, got.br_blockcount);
619 } while (xfs_iext_next_extent(ifp, &icur, &got));
622 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
626 * Now that the generic code is guaranteed not to be accessing
627 * the linux inode, we can inactivate and reclaim the inode.
630 xfs_fs_destroy_inode(
633 struct xfs_inode *ip = XFS_I(inode);
635 trace_xfs_destroy_inode(ip);
637 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
638 XFS_STATS_INC(ip->i_mount, vn_rele);
639 XFS_STATS_INC(ip->i_mount, vn_remove);
643 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
644 xfs_check_delalloc(ip, XFS_DATA_FORK);
645 xfs_check_delalloc(ip, XFS_COW_FORK);
649 XFS_STATS_INC(ip->i_mount, vn_reclaim);
652 * We should never get here with one of the reclaim flags already set.
654 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
655 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
658 * We always use background reclaim here because even if the inode is
659 * clean, it still may be under IO and hence we have wait for IO
660 * completion to occur before we can reclaim the inode. The background
661 * reclaim path handles this more efficiently than we can here, so
662 * simply let background reclaim tear down all inodes.
664 xfs_inode_mark_reclaimable(ip);
672 struct xfs_inode *ip = XFS_I(inode);
673 struct xfs_mount *mp = ip->i_mount;
674 struct xfs_trans *tp;
676 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
678 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
681 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
683 xfs_ilock(ip, XFS_ILOCK_EXCL);
684 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
685 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
686 xfs_trans_commit(tp);
690 * Slab object creation initialisation for the XFS inode.
691 * This covers only the idempotent fields in the XFS inode;
692 * all other fields need to be initialised on allocation
693 * from the slab. This avoids the need to repeatedly initialise
694 * fields in the xfs inode that left in the initialise state
695 * when freeing the inode.
698 xfs_fs_inode_init_once(
701 struct xfs_inode *ip = inode;
703 memset(ip, 0, sizeof(struct xfs_inode));
706 inode_init_once(VFS_I(ip));
709 atomic_set(&ip->i_pincount, 0);
710 spin_lock_init(&ip->i_flags_lock);
712 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
713 "xfsino", ip->i_ino);
714 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
715 "xfsino", ip->i_ino);
719 * We do an unlocked check for XFS_IDONTCACHE here because we are already
720 * serialised against cache hits here via the inode->i_lock and igrab() in
721 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
722 * racing with us, and it avoids needing to grab a spinlock here for every inode
723 * we drop the final reference on.
729 struct xfs_inode *ip = XFS_I(inode);
732 * If this unlinked inode is in the middle of recovery, don't
733 * drop the inode just yet; log recovery will take care of
734 * that. See the comment for this inode flag.
736 if (ip->i_flags & XFS_IRECOVERY) {
737 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
741 return generic_drop_inode(inode);
746 struct xfs_mount *mp)
749 kfree(mp->m_logname);
755 struct super_block *sb,
758 struct xfs_mount *mp = XFS_M(sb);
761 * Doing anything during the async pass would be counterproductive.
766 xfs_log_force(mp, XFS_LOG_SYNC);
769 * The disk must be active because we're syncing.
770 * We schedule log work now (now that the disk is
771 * active) instead of later (when it might not be).
773 flush_delayed_work(&mp->m_log->l_work);
781 struct dentry *dentry,
782 struct kstatfs *statp)
784 struct xfs_mount *mp = XFS_M(dentry->d_sb);
785 xfs_sb_t *sbp = &mp->m_sb;
786 struct xfs_inode *ip = XFS_I(d_inode(dentry));
787 uint64_t fakeinos, id;
794 statp->f_type = XFS_SUPER_MAGIC;
795 statp->f_namelen = MAXNAMELEN - 1;
797 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
798 statp->f_fsid = u64_to_fsid(id);
800 icount = percpu_counter_sum(&mp->m_icount);
801 ifree = percpu_counter_sum(&mp->m_ifree);
802 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
804 spin_lock(&mp->m_sb_lock);
805 statp->f_bsize = sbp->sb_blocksize;
806 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
807 statp->f_blocks = sbp->sb_dblocks - lsize;
808 spin_unlock(&mp->m_sb_lock);
810 /* make sure statp->f_bfree does not underflow */
811 statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
812 statp->f_bavail = statp->f_bfree;
814 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
815 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
816 if (M_IGEO(mp)->maxicount)
817 statp->f_files = min_t(typeof(statp->f_files),
819 M_IGEO(mp)->maxicount);
821 /* If sb_icount overshot maxicount, report actual allocation */
822 statp->f_files = max_t(typeof(statp->f_files),
826 /* make sure statp->f_ffree does not underflow */
827 ffree = statp->f_files - (icount - ifree);
828 statp->f_ffree = max_t(int64_t, ffree, 0);
831 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
832 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
833 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
834 xfs_qm_statvfs(ip, statp);
836 if (XFS_IS_REALTIME_MOUNT(mp) &&
837 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
838 statp->f_blocks = sbp->sb_rblocks;
839 statp->f_bavail = statp->f_bfree =
840 sbp->sb_frextents * sbp->sb_rextsize;
847 xfs_save_resvblks(struct xfs_mount *mp)
849 uint64_t resblks = 0;
851 mp->m_resblks_save = mp->m_resblks;
852 xfs_reserve_blocks(mp, &resblks, NULL);
856 xfs_restore_resvblks(struct xfs_mount *mp)
860 if (mp->m_resblks_save) {
861 resblks = mp->m_resblks_save;
862 mp->m_resblks_save = 0;
864 resblks = xfs_default_resblks(mp);
866 xfs_reserve_blocks(mp, &resblks, NULL);
870 * Second stage of a freeze. The data is already frozen so we only
871 * need to take care of the metadata. Once that's done sync the superblock
872 * to the log to dirty it in case of a crash while frozen. This ensures that we
873 * will recover the unlinked inode lists on the next mount.
877 struct super_block *sb)
879 struct xfs_mount *mp = XFS_M(sb);
884 * The filesystem is now frozen far enough that memory reclaim
885 * cannot safely operate on the filesystem. Hence we need to
886 * set a GFP_NOFS context here to avoid recursion deadlocks.
888 flags = memalloc_nofs_save();
889 xfs_blockgc_stop(mp);
890 xfs_save_resvblks(mp);
891 ret = xfs_log_quiesce(mp);
892 memalloc_nofs_restore(flags);
898 struct super_block *sb)
900 struct xfs_mount *mp = XFS_M(sb);
902 xfs_restore_resvblks(mp);
903 xfs_log_work_queue(mp);
904 xfs_blockgc_start(mp);
909 * This function fills in xfs_mount_t fields based on mount args.
910 * Note: the superblock _has_ now been read in.
914 struct xfs_mount *mp)
916 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
918 /* Fail a mount where the logbuf is smaller than the log stripe */
919 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
920 if (mp->m_logbsize <= 0 &&
921 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
922 mp->m_logbsize = mp->m_sb.sb_logsunit;
923 } else if (mp->m_logbsize > 0 &&
924 mp->m_logbsize < mp->m_sb.sb_logsunit) {
926 "logbuf size must be greater than or equal to log stripe size");
930 /* Fail a mount if the logbuf is larger than 32K */
931 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
933 "logbuf size for version 1 logs must be 16K or 32K");
939 * V5 filesystems always use attr2 format for attributes.
941 if (xfs_sb_version_hascrc(&mp->m_sb) &&
942 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
943 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
944 "attr2 is always enabled for V5 filesystems.");
949 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
950 * told by noattr2 to turn it off
952 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
953 !(mp->m_flags & XFS_MOUNT_NOATTR2))
954 mp->m_flags |= XFS_MOUNT_ATTR2;
957 * prohibit r/w mounts of read-only filesystems
959 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
961 "cannot mount a read-only filesystem as read-write");
965 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
966 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
967 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
969 "Super block does not support project and group quota together");
977 xfs_init_percpu_counters(
978 struct xfs_mount *mp)
982 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
986 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
990 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
994 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1001 percpu_counter_destroy(&mp->m_fdblocks);
1003 percpu_counter_destroy(&mp->m_ifree);
1005 percpu_counter_destroy(&mp->m_icount);
1010 xfs_reinit_percpu_counters(
1011 struct xfs_mount *mp)
1013 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1014 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1015 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1019 xfs_destroy_percpu_counters(
1020 struct xfs_mount *mp)
1022 percpu_counter_destroy(&mp->m_icount);
1023 percpu_counter_destroy(&mp->m_ifree);
1024 percpu_counter_destroy(&mp->m_fdblocks);
1025 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1026 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1027 percpu_counter_destroy(&mp->m_delalloc_blks);
1032 struct super_block *sb)
1034 struct xfs_mount *mp = XFS_M(sb);
1036 /* if ->fill_super failed, we have no mount to tear down */
1040 xfs_notice(mp, "Unmounting Filesystem");
1041 xfs_filestream_unmount(mp);
1045 free_percpu(mp->m_stats.xs_stats);
1046 xfs_destroy_percpu_counters(mp);
1047 xfs_destroy_mount_workqueues(mp);
1048 xfs_close_devices(mp);
1050 sb->s_fs_info = NULL;
1055 xfs_fs_nr_cached_objects(
1056 struct super_block *sb,
1057 struct shrink_control *sc)
1059 /* Paranoia: catch incorrect calls during mount setup or teardown */
1060 if (WARN_ON_ONCE(!sb->s_fs_info))
1062 return xfs_reclaim_inodes_count(XFS_M(sb));
1066 xfs_fs_free_cached_objects(
1067 struct super_block *sb,
1068 struct shrink_control *sc)
1070 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1073 static const struct super_operations xfs_super_operations = {
1074 .alloc_inode = xfs_fs_alloc_inode,
1075 .destroy_inode = xfs_fs_destroy_inode,
1076 .dirty_inode = xfs_fs_dirty_inode,
1077 .drop_inode = xfs_fs_drop_inode,
1078 .put_super = xfs_fs_put_super,
1079 .sync_fs = xfs_fs_sync_fs,
1080 .freeze_fs = xfs_fs_freeze,
1081 .unfreeze_fs = xfs_fs_unfreeze,
1082 .statfs = xfs_fs_statfs,
1083 .show_options = xfs_fs_show_options,
1084 .nr_cached_objects = xfs_fs_nr_cached_objects,
1085 .free_cached_objects = xfs_fs_free_cached_objects,
1094 int last, shift_left_factor = 0, _res;
1098 value = kstrdup(s, GFP_KERNEL);
1102 last = strlen(value) - 1;
1103 if (value[last] == 'K' || value[last] == 'k') {
1104 shift_left_factor = 10;
1107 if (value[last] == 'M' || value[last] == 'm') {
1108 shift_left_factor = 20;
1111 if (value[last] == 'G' || value[last] == 'g') {
1112 shift_left_factor = 30;
1116 if (kstrtoint(value, base, &_res))
1119 *res = _res << shift_left_factor;
1124 xfs_fs_warn_deprecated(
1125 struct fs_context *fc,
1126 struct fs_parameter *param,
1130 /* Don't print the warning if reconfiguring and current mount point
1131 * already had the flag set
1133 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1134 !!(XFS_M(fc->root->d_sb)->m_flags & flag) == value)
1136 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1140 * Set mount state from a mount option.
1142 * NOTE: mp->m_super is NULL here!
1146 struct fs_context *fc,
1147 struct fs_parameter *param)
1149 struct xfs_mount *parsing_mp = fc->s_fs_info;
1150 struct fs_parse_result result;
1154 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1160 parsing_mp->m_logbufs = result.uint_32;
1163 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1167 kfree(parsing_mp->m_logname);
1168 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1169 if (!parsing_mp->m_logname)
1173 kfree(parsing_mp->m_rtname);
1174 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1175 if (!parsing_mp->m_rtname)
1179 if (suffix_kstrtoint(param->string, 10, &size))
1181 parsing_mp->m_allocsize_log = ffs(size) - 1;
1182 parsing_mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1186 parsing_mp->m_flags |= XFS_MOUNT_GRPID;
1189 case Opt_sysvgroups:
1190 parsing_mp->m_flags &= ~XFS_MOUNT_GRPID;
1193 parsing_mp->m_flags |= XFS_MOUNT_WSYNC;
1195 case Opt_norecovery:
1196 parsing_mp->m_flags |= XFS_MOUNT_NORECOVERY;
1199 parsing_mp->m_flags |= XFS_MOUNT_NOALIGN;
1202 parsing_mp->m_flags |= XFS_MOUNT_SWALLOC;
1205 parsing_mp->m_dalign = result.uint_32;
1208 parsing_mp->m_swidth = result.uint_32;
1211 parsing_mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1214 parsing_mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1217 parsing_mp->m_flags |= XFS_MOUNT_NOUUID;
1220 parsing_mp->m_flags |= XFS_MOUNT_LARGEIO;
1223 parsing_mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1225 case Opt_filestreams:
1226 parsing_mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1229 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1230 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1231 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1236 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1239 case Opt_qnoenforce:
1240 case Opt_uqnoenforce:
1241 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1242 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1246 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1249 case Opt_pqnoenforce:
1250 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1251 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1255 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1258 case Opt_gqnoenforce:
1259 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1260 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1263 parsing_mp->m_flags |= XFS_MOUNT_DISCARD;
1266 parsing_mp->m_flags &= ~XFS_MOUNT_DISCARD;
1268 #ifdef CONFIG_FS_DAX
1270 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1273 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1276 /* Following mount options will be removed in September 2025 */
1278 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, true);
1279 parsing_mp->m_flags |= XFS_MOUNT_IKEEP;
1282 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, false);
1283 parsing_mp->m_flags &= ~XFS_MOUNT_IKEEP;
1286 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_ATTR2, true);
1287 parsing_mp->m_flags |= XFS_MOUNT_ATTR2;
1290 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_NOATTR2, true);
1291 parsing_mp->m_flags &= ~XFS_MOUNT_ATTR2;
1292 parsing_mp->m_flags |= XFS_MOUNT_NOATTR2;
1295 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1303 xfs_fs_validate_params(
1304 struct xfs_mount *mp)
1307 * no recovery flag requires a read-only mount
1309 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1310 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1311 xfs_warn(mp, "no-recovery mounts must be read-only.");
1315 if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1316 (mp->m_dalign || mp->m_swidth)) {
1318 "sunit and swidth options incompatible with the noalign option");
1322 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1323 xfs_warn(mp, "quota support not available in this kernel.");
1327 if ((mp->m_dalign && !mp->m_swidth) ||
1328 (!mp->m_dalign && mp->m_swidth)) {
1329 xfs_warn(mp, "sunit and swidth must be specified together");
1333 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1335 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1336 mp->m_swidth, mp->m_dalign);
1340 if (mp->m_logbufs != -1 &&
1341 mp->m_logbufs != 0 &&
1342 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1343 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1344 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1345 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1349 if (mp->m_logbsize != -1 &&
1350 mp->m_logbsize != 0 &&
1351 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1352 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1353 !is_power_of_2(mp->m_logbsize))) {
1355 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1360 if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1361 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1362 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1363 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1364 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1373 struct super_block *sb,
1374 struct fs_context *fc)
1376 struct xfs_mount *mp = sb->s_fs_info;
1378 int flags = 0, error;
1382 error = xfs_fs_validate_params(mp);
1384 goto out_free_names;
1386 sb_min_blocksize(sb, BBSIZE);
1387 sb->s_xattr = xfs_xattr_handlers;
1388 sb->s_export_op = &xfs_export_operations;
1389 #ifdef CONFIG_XFS_QUOTA
1390 sb->s_qcop = &xfs_quotactl_operations;
1391 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1393 sb->s_op = &xfs_super_operations;
1396 * Delay mount work if the debug hook is set. This is debug
1397 * instrumention to coordinate simulation of xfs mount failures with
1398 * VFS superblock operations
1400 if (xfs_globals.mount_delay) {
1401 xfs_notice(mp, "Delaying mount for %d seconds.",
1402 xfs_globals.mount_delay);
1403 msleep(xfs_globals.mount_delay * 1000);
1406 if (fc->sb_flags & SB_SILENT)
1407 flags |= XFS_MFSI_QUIET;
1409 error = xfs_open_devices(mp);
1411 goto out_free_names;
1413 error = xfs_init_mount_workqueues(mp);
1415 goto out_close_devices;
1417 error = xfs_init_percpu_counters(mp);
1419 goto out_destroy_workqueues;
1421 /* Allocate stats memory before we do operations that might use it */
1422 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1423 if (!mp->m_stats.xs_stats) {
1425 goto out_destroy_counters;
1428 error = xfs_readsb(mp, flags);
1430 goto out_free_stats;
1432 error = xfs_finish_flags(mp);
1436 error = xfs_setup_devices(mp);
1440 /* V4 support is undergoing deprecation. */
1441 if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1442 #ifdef CONFIG_XFS_SUPPORT_V4
1444 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1447 "Deprecated V4 format (crc=0) not supported by kernel.");
1453 /* Filesystem claims it needs repair, so refuse the mount. */
1454 if (xfs_sb_version_needsrepair(&mp->m_sb)) {
1455 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1456 error = -EFSCORRUPTED;
1461 * Don't touch the filesystem if a user tool thinks it owns the primary
1462 * superblock. mkfs doesn't clear the flag from secondary supers, so
1463 * we don't check them at all.
1465 if (mp->m_sb.sb_inprogress) {
1466 xfs_warn(mp, "Offline file system operation in progress!");
1467 error = -EFSCORRUPTED;
1472 * Until this is fixed only page-sized or smaller data blocks work.
1474 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1476 "File system with blocksize %d bytes. "
1477 "Only pagesize (%ld) or less will currently work.",
1478 mp->m_sb.sb_blocksize, PAGE_SIZE);
1483 /* Ensure this filesystem fits in the page cache limits */
1484 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1485 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1487 "file system too large to be mounted on this system.");
1493 * XFS block mappings use 54 bits to store the logical block offset.
1494 * This should suffice to handle the maximum file size that the VFS
1495 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1496 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1497 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1498 * to check this assertion.
1500 * Avoid integer overflow by comparing the maximum bmbt offset to the
1501 * maximum pagecache offset in units of fs blocks.
1503 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1505 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1506 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1512 error = xfs_filestream_mount(mp);
1517 * we must configure the block size in the superblock before we run the
1518 * full mount process as the mount process can lookup and cache inodes.
1520 sb->s_magic = XFS_SUPER_MAGIC;
1521 sb->s_blocksize = mp->m_sb.sb_blocksize;
1522 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1523 sb->s_maxbytes = MAX_LFS_FILESIZE;
1524 sb->s_max_links = XFS_MAXLINK;
1525 sb->s_time_gran = 1;
1526 if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1527 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1528 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1530 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1531 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1533 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1534 sb->s_iflags |= SB_I_CGROUPWB;
1536 set_posix_acl_flag(sb);
1538 /* version 5 superblocks support inode version counters. */
1539 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1540 sb->s_flags |= SB_I_VERSION;
1542 if (xfs_sb_version_hasbigtime(&mp->m_sb))
1544 "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1546 if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1547 bool rtdev_is_dax = false, datadev_is_dax;
1550 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1552 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1554 if (mp->m_rtdev_targp)
1555 rtdev_is_dax = bdev_dax_supported(
1556 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1557 if (!rtdev_is_dax && !datadev_is_dax) {
1559 "DAX unsupported by block device. Turning off DAX.");
1560 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1562 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1564 "DAX and reflink cannot be used together!");
1566 goto out_filestream_unmount;
1570 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1571 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1573 if (!blk_queue_discard(q)) {
1574 xfs_warn(mp, "mounting with \"discard\" option, but "
1575 "the device does not support discard");
1576 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1580 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1581 if (mp->m_sb.sb_rblocks) {
1583 "reflink not compatible with realtime device!");
1585 goto out_filestream_unmount;
1588 if (xfs_globals.always_cow) {
1589 xfs_info(mp, "using DEBUG-only always_cow mode.");
1590 mp->m_always_cow = true;
1594 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1596 "reverse mapping btree not compatible with realtime device!");
1598 goto out_filestream_unmount;
1601 if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1603 "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1605 error = xfs_mountfs(mp);
1607 goto out_filestream_unmount;
1609 root = igrab(VFS_I(mp->m_rootip));
1614 sb->s_root = d_make_root(root);
1622 out_filestream_unmount:
1623 xfs_filestream_unmount(mp);
1627 free_percpu(mp->m_stats.xs_stats);
1628 out_destroy_counters:
1629 xfs_destroy_percpu_counters(mp);
1630 out_destroy_workqueues:
1631 xfs_destroy_mount_workqueues(mp);
1633 xfs_close_devices(mp);
1635 sb->s_fs_info = NULL;
1640 xfs_filestream_unmount(mp);
1647 struct fs_context *fc)
1649 return get_tree_bdev(fc, xfs_fs_fill_super);
1654 struct xfs_mount *mp)
1656 struct xfs_sb *sbp = &mp->m_sb;
1659 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1661 "ro->rw transition prohibited on norecovery mount");
1665 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1666 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1668 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1669 (sbp->sb_features_ro_compat &
1670 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1674 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1677 * If this is the first remount to writeable state we might have some
1678 * superblock changes to update.
1680 if (mp->m_update_sb) {
1681 error = xfs_sync_sb(mp, false);
1683 xfs_warn(mp, "failed to write sb changes");
1686 mp->m_update_sb = false;
1690 * Fill out the reserve pool if it is empty. Use the stashed value if
1691 * it is non-zero, otherwise go with the default.
1693 xfs_restore_resvblks(mp);
1694 xfs_log_work_queue(mp);
1696 /* Recover any CoW blocks that never got remapped. */
1697 error = xfs_reflink_recover_cow(mp);
1700 "Error %d recovering leftover CoW allocations.", error);
1701 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1704 xfs_blockgc_start(mp);
1706 /* Create the per-AG metadata reservation pool .*/
1707 error = xfs_fs_reserve_ag_blocks(mp);
1708 if (error && error != -ENOSPC)
1716 struct xfs_mount *mp)
1721 * Cancel background eofb scanning so it cannot race with the final
1722 * log force+buftarg wait and deadlock the remount.
1724 xfs_blockgc_stop(mp);
1726 /* Get rid of any leftover CoW reservations... */
1727 error = xfs_blockgc_free_space(mp, NULL);
1729 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1733 /* Free the per-AG metadata reservation pool. */
1734 error = xfs_fs_unreserve_ag_blocks(mp);
1736 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1741 * Before we sync the metadata, we need to free up the reserve block
1742 * pool so that the used block count in the superblock on disk is
1743 * correct at the end of the remount. Stash the current* reserve pool
1744 * size so that if we get remounted rw, we can return it to the same
1747 xfs_save_resvblks(mp);
1750 mp->m_flags |= XFS_MOUNT_RDONLY;
1756 * Logically we would return an error here to prevent users from believing
1757 * they might have changed mount options using remount which can't be changed.
1759 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1760 * arguments in some cases so we can't blindly reject options, but have to
1761 * check for each specified option if it actually differs from the currently
1762 * set option and only reject it if that's the case.
1764 * Until that is implemented we return success for every remount request, and
1765 * silently ignore all options that we can't actually change.
1769 struct fs_context *fc)
1771 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1772 struct xfs_mount *new_mp = fc->s_fs_info;
1773 xfs_sb_t *sbp = &mp->m_sb;
1774 int flags = fc->sb_flags;
1777 /* version 5 superblocks always support version counters. */
1778 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1779 fc->sb_flags |= SB_I_VERSION;
1781 error = xfs_fs_validate_params(new_mp);
1785 sync_filesystem(mp->m_super);
1787 /* inode32 -> inode64 */
1788 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1789 !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1790 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1791 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1794 /* inode64 -> inode32 */
1795 if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1796 (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1797 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1798 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1802 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1803 error = xfs_remount_rw(mp);
1809 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1810 error = xfs_remount_ro(mp);
1818 static void xfs_fs_free(
1819 struct fs_context *fc)
1821 struct xfs_mount *mp = fc->s_fs_info;
1824 * mp is stored in the fs_context when it is initialized.
1825 * mp is transferred to the superblock on a successful mount,
1826 * but if an error occurs before the transfer we have to free
1833 static const struct fs_context_operations xfs_context_ops = {
1834 .parse_param = xfs_fs_parse_param,
1835 .get_tree = xfs_fs_get_tree,
1836 .reconfigure = xfs_fs_reconfigure,
1837 .free = xfs_fs_free,
1840 static int xfs_init_fs_context(
1841 struct fs_context *fc)
1843 struct xfs_mount *mp;
1845 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1849 spin_lock_init(&mp->m_sb_lock);
1850 spin_lock_init(&mp->m_agirotor_lock);
1851 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1852 spin_lock_init(&mp->m_perag_lock);
1853 mutex_init(&mp->m_growlock);
1854 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1855 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1856 mp->m_kobj.kobject.kset = xfs_kset;
1858 * We don't create the finobt per-ag space reservation until after log
1859 * recovery, so we must set this to true so that an ifree transaction
1860 * started during log recovery will not depend on space reservations
1861 * for finobt expansion.
1863 mp->m_finobt_nores = true;
1866 * These can be overridden by the mount option parsing.
1869 mp->m_logbsize = -1;
1870 mp->m_allocsize_log = 16; /* 64k */
1873 * Copy binary VFS mount flags we are interested in.
1875 if (fc->sb_flags & SB_RDONLY)
1876 mp->m_flags |= XFS_MOUNT_RDONLY;
1877 if (fc->sb_flags & SB_DIRSYNC)
1878 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1879 if (fc->sb_flags & SB_SYNCHRONOUS)
1880 mp->m_flags |= XFS_MOUNT_WSYNC;
1883 fc->ops = &xfs_context_ops;
1888 static struct file_system_type xfs_fs_type = {
1889 .owner = THIS_MODULE,
1891 .init_fs_context = xfs_init_fs_context,
1892 .parameters = xfs_fs_parameters,
1893 .kill_sb = kill_block_super,
1894 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1896 MODULE_ALIAS_FS("xfs");
1899 xfs_init_zones(void)
1901 xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1902 sizeof(struct xlog_ticket),
1904 if (!xfs_log_ticket_zone)
1907 xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1908 sizeof(struct xfs_extent_free_item),
1910 if (!xfs_bmap_free_item_zone)
1911 goto out_destroy_log_ticket_zone;
1913 xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1914 sizeof(struct xfs_btree_cur),
1916 if (!xfs_btree_cur_zone)
1917 goto out_destroy_bmap_free_item_zone;
1919 xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1920 sizeof(struct xfs_da_state),
1922 if (!xfs_da_state_zone)
1923 goto out_destroy_btree_cur_zone;
1925 xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1926 sizeof(struct xfs_ifork),
1928 if (!xfs_ifork_zone)
1929 goto out_destroy_da_state_zone;
1931 xfs_trans_zone = kmem_cache_create("xfs_trans",
1932 sizeof(struct xfs_trans),
1934 if (!xfs_trans_zone)
1935 goto out_destroy_ifork_zone;
1939 * The size of the zone allocated buf log item is the maximum
1940 * size possible under XFS. This wastes a little bit of memory,
1941 * but it is much faster.
1943 xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1944 sizeof(struct xfs_buf_log_item),
1946 if (!xfs_buf_item_zone)
1947 goto out_destroy_trans_zone;
1949 xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1950 (sizeof(struct xfs_efd_log_item) +
1951 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1952 sizeof(struct xfs_extent)),
1955 goto out_destroy_buf_item_zone;
1957 xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1958 (sizeof(struct xfs_efi_log_item) +
1959 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1960 sizeof(struct xfs_extent)),
1963 goto out_destroy_efd_zone;
1965 xfs_inode_zone = kmem_cache_create("xfs_inode",
1966 sizeof(struct xfs_inode), 0,
1967 (SLAB_HWCACHE_ALIGN |
1968 SLAB_RECLAIM_ACCOUNT |
1969 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1970 xfs_fs_inode_init_once);
1971 if (!xfs_inode_zone)
1972 goto out_destroy_efi_zone;
1974 xfs_ili_zone = kmem_cache_create("xfs_ili",
1975 sizeof(struct xfs_inode_log_item), 0,
1976 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1979 goto out_destroy_inode_zone;
1981 xfs_icreate_zone = kmem_cache_create("xfs_icr",
1982 sizeof(struct xfs_icreate_item),
1984 if (!xfs_icreate_zone)
1985 goto out_destroy_ili_zone;
1987 xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1988 sizeof(struct xfs_rud_log_item),
1991 goto out_destroy_icreate_zone;
1993 xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1994 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1997 goto out_destroy_rud_zone;
1999 xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2000 sizeof(struct xfs_cud_log_item),
2003 goto out_destroy_rui_zone;
2005 xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2006 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2009 goto out_destroy_cud_zone;
2011 xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2012 sizeof(struct xfs_bud_log_item),
2015 goto out_destroy_cui_zone;
2017 xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2018 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2021 goto out_destroy_bud_zone;
2025 out_destroy_bud_zone:
2026 kmem_cache_destroy(xfs_bud_zone);
2027 out_destroy_cui_zone:
2028 kmem_cache_destroy(xfs_cui_zone);
2029 out_destroy_cud_zone:
2030 kmem_cache_destroy(xfs_cud_zone);
2031 out_destroy_rui_zone:
2032 kmem_cache_destroy(xfs_rui_zone);
2033 out_destroy_rud_zone:
2034 kmem_cache_destroy(xfs_rud_zone);
2035 out_destroy_icreate_zone:
2036 kmem_cache_destroy(xfs_icreate_zone);
2037 out_destroy_ili_zone:
2038 kmem_cache_destroy(xfs_ili_zone);
2039 out_destroy_inode_zone:
2040 kmem_cache_destroy(xfs_inode_zone);
2041 out_destroy_efi_zone:
2042 kmem_cache_destroy(xfs_efi_zone);
2043 out_destroy_efd_zone:
2044 kmem_cache_destroy(xfs_efd_zone);
2045 out_destroy_buf_item_zone:
2046 kmem_cache_destroy(xfs_buf_item_zone);
2047 out_destroy_trans_zone:
2048 kmem_cache_destroy(xfs_trans_zone);
2049 out_destroy_ifork_zone:
2050 kmem_cache_destroy(xfs_ifork_zone);
2051 out_destroy_da_state_zone:
2052 kmem_cache_destroy(xfs_da_state_zone);
2053 out_destroy_btree_cur_zone:
2054 kmem_cache_destroy(xfs_btree_cur_zone);
2055 out_destroy_bmap_free_item_zone:
2056 kmem_cache_destroy(xfs_bmap_free_item_zone);
2057 out_destroy_log_ticket_zone:
2058 kmem_cache_destroy(xfs_log_ticket_zone);
2064 xfs_destroy_zones(void)
2067 * Make sure all delayed rcu free are flushed before we
2071 kmem_cache_destroy(xfs_bui_zone);
2072 kmem_cache_destroy(xfs_bud_zone);
2073 kmem_cache_destroy(xfs_cui_zone);
2074 kmem_cache_destroy(xfs_cud_zone);
2075 kmem_cache_destroy(xfs_rui_zone);
2076 kmem_cache_destroy(xfs_rud_zone);
2077 kmem_cache_destroy(xfs_icreate_zone);
2078 kmem_cache_destroy(xfs_ili_zone);
2079 kmem_cache_destroy(xfs_inode_zone);
2080 kmem_cache_destroy(xfs_efi_zone);
2081 kmem_cache_destroy(xfs_efd_zone);
2082 kmem_cache_destroy(xfs_buf_item_zone);
2083 kmem_cache_destroy(xfs_trans_zone);
2084 kmem_cache_destroy(xfs_ifork_zone);
2085 kmem_cache_destroy(xfs_da_state_zone);
2086 kmem_cache_destroy(xfs_btree_cur_zone);
2087 kmem_cache_destroy(xfs_bmap_free_item_zone);
2088 kmem_cache_destroy(xfs_log_ticket_zone);
2092 xfs_init_workqueues(void)
2095 * The allocation workqueue can be used in memory reclaim situations
2096 * (writepage path), and parallelism is only limited by the number of
2097 * AGs in all the filesystems mounted. Hence use the default large
2098 * max_active value for this workqueue.
2100 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2101 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2105 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2107 if (!xfs_discard_wq)
2108 goto out_free_alloc_wq;
2112 destroy_workqueue(xfs_alloc_wq);
2117 xfs_destroy_workqueues(void)
2119 destroy_workqueue(xfs_discard_wq);
2120 destroy_workqueue(xfs_alloc_wq);
2128 xfs_check_ondisk_structs();
2130 printk(KERN_INFO XFS_VERSION_STRING " with "
2131 XFS_BUILD_OPTIONS " enabled\n");
2135 error = xfs_init_zones();
2139 error = xfs_init_workqueues();
2141 goto out_destroy_zones;
2143 error = xfs_mru_cache_init();
2145 goto out_destroy_wq;
2147 error = xfs_buf_init();
2149 goto out_mru_cache_uninit;
2151 error = xfs_init_procfs();
2153 goto out_buf_terminate;
2155 error = xfs_sysctl_register();
2157 goto out_cleanup_procfs;
2159 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2162 goto out_sysctl_unregister;
2165 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2167 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2168 if (!xfsstats.xs_stats) {
2170 goto out_kset_unregister;
2173 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2176 goto out_free_stats;
2179 xfs_dbg_kobj.kobject.kset = xfs_kset;
2180 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2182 goto out_remove_stats_kobj;
2185 error = xfs_qm_init();
2187 goto out_remove_dbg_kobj;
2189 error = register_filesystem(&xfs_fs_type);
2196 out_remove_dbg_kobj:
2198 xfs_sysfs_del(&xfs_dbg_kobj);
2199 out_remove_stats_kobj:
2201 xfs_sysfs_del(&xfsstats.xs_kobj);
2203 free_percpu(xfsstats.xs_stats);
2204 out_kset_unregister:
2205 kset_unregister(xfs_kset);
2206 out_sysctl_unregister:
2207 xfs_sysctl_unregister();
2209 xfs_cleanup_procfs();
2211 xfs_buf_terminate();
2212 out_mru_cache_uninit:
2213 xfs_mru_cache_uninit();
2215 xfs_destroy_workqueues();
2217 xfs_destroy_zones();
2226 unregister_filesystem(&xfs_fs_type);
2228 xfs_sysfs_del(&xfs_dbg_kobj);
2230 xfs_sysfs_del(&xfsstats.xs_kobj);
2231 free_percpu(xfsstats.xs_stats);
2232 kset_unregister(xfs_kset);
2233 xfs_sysctl_unregister();
2234 xfs_cleanup_procfs();
2235 xfs_buf_terminate();
2236 xfs_mru_cache_uninit();
2237 xfs_destroy_workqueues();
2238 xfs_destroy_zones();
2239 xfs_uuid_table_free();
2242 module_init(init_xfs_fs);
2243 module_exit(exit_xfs_fs);
2245 MODULE_AUTHOR("Silicon Graphics, Inc.");
2246 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2247 MODULE_LICENSE("GPL");