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"
40 #include "xfs_defer.h"
42 #include <linux/magic.h>
43 #include <linux/fs_context.h>
44 #include <linux/fs_parser.h>
46 static const struct super_operations xfs_super_operations;
48 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
50 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
53 #ifdef CONFIG_HOTPLUG_CPU
54 static LIST_HEAD(xfs_mount_list);
55 static DEFINE_SPINLOCK(xfs_mount_list_lock);
57 static inline void xfs_mount_list_add(struct xfs_mount *mp)
59 spin_lock(&xfs_mount_list_lock);
60 list_add(&mp->m_mount_list, &xfs_mount_list);
61 spin_unlock(&xfs_mount_list_lock);
64 static inline void xfs_mount_list_del(struct xfs_mount *mp)
66 spin_lock(&xfs_mount_list_lock);
67 list_del(&mp->m_mount_list);
68 spin_unlock(&xfs_mount_list_lock);
70 #else /* !CONFIG_HOTPLUG_CPU */
71 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
72 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
82 xfs_mount_set_dax_mode(
84 enum xfs_dax_mode mode)
88 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
91 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
92 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
95 mp->m_features |= XFS_FEAT_DAX_NEVER;
96 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
101 static const struct constant_table dax_param_enums[] = {
102 {"inode", XFS_DAX_INODE },
103 {"always", XFS_DAX_ALWAYS },
104 {"never", XFS_DAX_NEVER },
109 * Table driven mount option parser.
112 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
113 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
114 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
115 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
116 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
117 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
118 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
119 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
120 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
123 static const struct fs_parameter_spec xfs_fs_parameters[] = {
124 fsparam_u32("logbufs", Opt_logbufs),
125 fsparam_string("logbsize", Opt_logbsize),
126 fsparam_string("logdev", Opt_logdev),
127 fsparam_string("rtdev", Opt_rtdev),
128 fsparam_flag("wsync", Opt_wsync),
129 fsparam_flag("noalign", Opt_noalign),
130 fsparam_flag("swalloc", Opt_swalloc),
131 fsparam_u32("sunit", Opt_sunit),
132 fsparam_u32("swidth", Opt_swidth),
133 fsparam_flag("nouuid", Opt_nouuid),
134 fsparam_flag("grpid", Opt_grpid),
135 fsparam_flag("nogrpid", Opt_nogrpid),
136 fsparam_flag("bsdgroups", Opt_bsdgroups),
137 fsparam_flag("sysvgroups", Opt_sysvgroups),
138 fsparam_string("allocsize", Opt_allocsize),
139 fsparam_flag("norecovery", Opt_norecovery),
140 fsparam_flag("inode64", Opt_inode64),
141 fsparam_flag("inode32", Opt_inode32),
142 fsparam_flag("ikeep", Opt_ikeep),
143 fsparam_flag("noikeep", Opt_noikeep),
144 fsparam_flag("largeio", Opt_largeio),
145 fsparam_flag("nolargeio", Opt_nolargeio),
146 fsparam_flag("attr2", Opt_attr2),
147 fsparam_flag("noattr2", Opt_noattr2),
148 fsparam_flag("filestreams", Opt_filestreams),
149 fsparam_flag("quota", Opt_quota),
150 fsparam_flag("noquota", Opt_noquota),
151 fsparam_flag("usrquota", Opt_usrquota),
152 fsparam_flag("grpquota", Opt_grpquota),
153 fsparam_flag("prjquota", Opt_prjquota),
154 fsparam_flag("uquota", Opt_uquota),
155 fsparam_flag("gquota", Opt_gquota),
156 fsparam_flag("pquota", Opt_pquota),
157 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
158 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
159 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
160 fsparam_flag("qnoenforce", Opt_qnoenforce),
161 fsparam_flag("discard", Opt_discard),
162 fsparam_flag("nodiscard", Opt_nodiscard),
163 fsparam_flag("dax", Opt_dax),
164 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
168 struct proc_xfs_info {
178 static struct proc_xfs_info xfs_info_set[] = {
179 /* the few simple ones we can get from the mount struct */
180 { XFS_FEAT_IKEEP, ",ikeep" },
181 { XFS_FEAT_WSYNC, ",wsync" },
182 { XFS_FEAT_NOALIGN, ",noalign" },
183 { XFS_FEAT_SWALLOC, ",swalloc" },
184 { XFS_FEAT_NOUUID, ",nouuid" },
185 { XFS_FEAT_NORECOVERY, ",norecovery" },
186 { XFS_FEAT_ATTR2, ",attr2" },
187 { XFS_FEAT_FILESTREAMS, ",filestreams" },
188 { XFS_FEAT_GRPID, ",grpid" },
189 { XFS_FEAT_DISCARD, ",discard" },
190 { XFS_FEAT_LARGE_IOSIZE, ",largeio" },
191 { XFS_FEAT_DAX_ALWAYS, ",dax=always" },
192 { XFS_FEAT_DAX_NEVER, ",dax=never" },
195 struct xfs_mount *mp = XFS_M(root->d_sb);
196 struct proc_xfs_info *xfs_infop;
198 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
199 if (mp->m_features & xfs_infop->flag)
200 seq_puts(m, xfs_infop->str);
203 seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
205 if (xfs_has_allocsize(mp))
206 seq_printf(m, ",allocsize=%dk",
207 (1 << mp->m_allocsize_log) >> 10);
209 if (mp->m_logbufs > 0)
210 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
211 if (mp->m_logbsize > 0)
212 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
215 seq_show_option(m, "logdev", mp->m_logname);
217 seq_show_option(m, "rtdev", mp->m_rtname);
219 if (mp->m_dalign > 0)
220 seq_printf(m, ",sunit=%d",
221 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
222 if (mp->m_swidth > 0)
223 seq_printf(m, ",swidth=%d",
224 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
226 if (mp->m_qflags & XFS_UQUOTA_ENFD)
227 seq_puts(m, ",usrquota");
228 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
229 seq_puts(m, ",uqnoenforce");
231 if (mp->m_qflags & XFS_PQUOTA_ENFD)
232 seq_puts(m, ",prjquota");
233 else if (mp->m_qflags & XFS_PQUOTA_ACCT)
234 seq_puts(m, ",pqnoenforce");
236 if (mp->m_qflags & XFS_GQUOTA_ENFD)
237 seq_puts(m, ",grpquota");
238 else if (mp->m_qflags & XFS_GQUOTA_ACCT)
239 seq_puts(m, ",gqnoenforce");
241 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
242 seq_puts(m, ",noquota");
248 * Set parameters for inode allocation heuristics, taking into account
249 * filesystem size and inode32/inode64 mount options; i.e. specifically
250 * whether or not XFS_FEAT_SMALL_INUMS is set.
252 * Inode allocation patterns are altered only if inode32 is requested
253 * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
254 * If altered, XFS_OPSTATE_INODE32 is set as well.
256 * An agcount independent of that in the mount structure is provided
257 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
258 * to the potentially higher ag count.
260 * Returns the maximum AG index which may contain inodes.
264 struct xfs_mount *mp,
265 xfs_agnumber_t agcount)
267 xfs_agnumber_t index;
268 xfs_agnumber_t maxagi = 0;
269 xfs_sb_t *sbp = &mp->m_sb;
270 xfs_agnumber_t max_metadata;
275 * Calculate how much should be reserved for inodes to meet
276 * the max inode percentage. Used only for inode32.
278 if (M_IGEO(mp)->maxicount) {
281 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
283 icount += sbp->sb_agblocks - 1;
284 do_div(icount, sbp->sb_agblocks);
285 max_metadata = icount;
287 max_metadata = agcount;
290 /* Get the last possible inode in the filesystem */
291 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
292 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
295 * If user asked for no more than 32-bit inodes, and the fs is
296 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
297 * the allocator to accommodate the request.
299 if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
300 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
302 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
304 for (index = 0; index < agcount; index++) {
305 struct xfs_perag *pag;
307 ino = XFS_AGINO_TO_INO(mp, index, agino);
309 pag = xfs_perag_get(mp, index);
311 if (xfs_is_inode32(mp)) {
312 if (ino > XFS_MAXINUMBER_32) {
313 pag->pagi_inodeok = 0;
314 pag->pagf_metadata = 0;
316 pag->pagi_inodeok = 1;
318 if (index < max_metadata)
319 pag->pagf_metadata = 1;
321 pag->pagf_metadata = 0;
324 pag->pagi_inodeok = 1;
325 pag->pagf_metadata = 0;
331 return xfs_is_inode32(mp) ? maxagi : agcount;
336 struct super_block *sb,
337 struct xfs_buftarg *bt)
339 return dax_supported(bt->bt_daxdev, bt->bt_bdev, sb->s_blocksize, 0,
340 bdev_nr_sectors(bt->bt_bdev));
347 struct block_device **bdevp)
351 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
353 if (IS_ERR(*bdevp)) {
354 error = PTR_ERR(*bdevp);
355 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
363 struct block_device *bdev)
366 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
371 struct xfs_mount *mp)
373 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
375 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
376 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
377 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
379 xfs_free_buftarg(mp->m_logdev_targp);
380 xfs_blkdev_put(logdev);
381 fs_put_dax(dax_logdev);
383 if (mp->m_rtdev_targp) {
384 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
385 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
387 xfs_free_buftarg(mp->m_rtdev_targp);
388 xfs_blkdev_put(rtdev);
389 fs_put_dax(dax_rtdev);
391 xfs_free_buftarg(mp->m_ddev_targp);
392 fs_put_dax(dax_ddev);
396 * The file system configurations are:
397 * (1) device (partition) with data and internal log
398 * (2) logical volume with data and log subvolumes.
399 * (3) logical volume with data, log, and realtime subvolumes.
401 * We only have to handle opening the log and realtime volumes here if
402 * they are present. The data subvolume has already been opened by
403 * get_sb_bdev() and is stored in sb->s_bdev.
407 struct xfs_mount *mp)
409 struct block_device *ddev = mp->m_super->s_bdev;
410 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
411 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
412 struct block_device *logdev = NULL, *rtdev = NULL;
416 * Open real time and log devices - order is important.
419 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
422 dax_logdev = fs_dax_get_by_bdev(logdev);
426 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
428 goto out_close_logdev;
430 if (rtdev == ddev || rtdev == logdev) {
432 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
434 goto out_close_rtdev;
436 dax_rtdev = fs_dax_get_by_bdev(rtdev);
440 * Setup xfs_mount buffer target pointers
443 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
444 if (!mp->m_ddev_targp)
445 goto out_close_rtdev;
448 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
449 if (!mp->m_rtdev_targp)
450 goto out_free_ddev_targ;
453 if (logdev && logdev != ddev) {
454 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
455 if (!mp->m_logdev_targp)
456 goto out_free_rtdev_targ;
458 mp->m_logdev_targp = mp->m_ddev_targp;
464 if (mp->m_rtdev_targp)
465 xfs_free_buftarg(mp->m_rtdev_targp);
467 xfs_free_buftarg(mp->m_ddev_targp);
469 xfs_blkdev_put(rtdev);
470 fs_put_dax(dax_rtdev);
472 if (logdev && logdev != ddev) {
473 xfs_blkdev_put(logdev);
474 fs_put_dax(dax_logdev);
477 fs_put_dax(dax_ddev);
482 * Setup xfs_mount buffer target pointers based on superblock
486 struct xfs_mount *mp)
490 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
494 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
495 unsigned int log_sector_size = BBSIZE;
497 if (xfs_has_sector(mp))
498 log_sector_size = mp->m_sb.sb_logsectsize;
499 error = xfs_setsize_buftarg(mp->m_logdev_targp,
504 if (mp->m_rtdev_targp) {
505 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
506 mp->m_sb.sb_sectsize);
515 xfs_init_mount_workqueues(
516 struct xfs_mount *mp)
518 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
519 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
520 1, mp->m_super->s_id);
521 if (!mp->m_buf_workqueue)
524 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
525 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
526 0, mp->m_super->s_id);
527 if (!mp->m_unwritten_workqueue)
528 goto out_destroy_buf;
530 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
531 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
532 0, mp->m_super->s_id);
533 if (!mp->m_reclaim_workqueue)
534 goto out_destroy_unwritten;
536 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
537 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
538 0, mp->m_super->s_id);
539 if (!mp->m_blockgc_wq)
540 goto out_destroy_reclaim;
542 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
543 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
544 1, mp->m_super->s_id);
545 if (!mp->m_inodegc_wq)
546 goto out_destroy_blockgc;
548 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
549 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
550 if (!mp->m_sync_workqueue)
551 goto out_destroy_inodegc;
556 destroy_workqueue(mp->m_inodegc_wq);
558 destroy_workqueue(mp->m_blockgc_wq);
560 destroy_workqueue(mp->m_reclaim_workqueue);
561 out_destroy_unwritten:
562 destroy_workqueue(mp->m_unwritten_workqueue);
564 destroy_workqueue(mp->m_buf_workqueue);
570 xfs_destroy_mount_workqueues(
571 struct xfs_mount *mp)
573 destroy_workqueue(mp->m_sync_workqueue);
574 destroy_workqueue(mp->m_blockgc_wq);
575 destroy_workqueue(mp->m_inodegc_wq);
576 destroy_workqueue(mp->m_reclaim_workqueue);
577 destroy_workqueue(mp->m_unwritten_workqueue);
578 destroy_workqueue(mp->m_buf_workqueue);
582 xfs_flush_inodes_worker(
583 struct work_struct *work)
585 struct xfs_mount *mp = container_of(work, struct xfs_mount,
586 m_flush_inodes_work);
587 struct super_block *sb = mp->m_super;
589 if (down_read_trylock(&sb->s_umount)) {
591 up_read(&sb->s_umount);
596 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
597 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
598 * for IO to complete so that we effectively throttle multiple callers to the
599 * rate at which IO is completing.
603 struct xfs_mount *mp)
606 * If flush_work() returns true then that means we waited for a flush
607 * which was already in progress. Don't bother running another scan.
609 if (flush_work(&mp->m_flush_inodes_work))
612 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
613 flush_work(&mp->m_flush_inodes_work);
616 /* Catch misguided souls that try to use this interface on XFS */
617 STATIC struct inode *
619 struct super_block *sb)
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);
640 xfs_inode_mark_reclaimable(ip);
648 struct xfs_inode *ip = XFS_I(inode);
649 struct xfs_mount *mp = ip->i_mount;
650 struct xfs_trans *tp;
652 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
654 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
657 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
659 xfs_ilock(ip, XFS_ILOCK_EXCL);
660 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
661 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
662 xfs_trans_commit(tp);
666 * Slab object creation initialisation for the XFS inode.
667 * This covers only the idempotent fields in the XFS inode;
668 * all other fields need to be initialised on allocation
669 * from the slab. This avoids the need to repeatedly initialise
670 * fields in the xfs inode that left in the initialise state
671 * when freeing the inode.
674 xfs_fs_inode_init_once(
677 struct xfs_inode *ip = inode;
679 memset(ip, 0, sizeof(struct xfs_inode));
682 inode_init_once(VFS_I(ip));
685 atomic_set(&ip->i_pincount, 0);
686 spin_lock_init(&ip->i_flags_lock);
688 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
689 "xfsino", ip->i_ino);
693 * We do an unlocked check for XFS_IDONTCACHE here because we are already
694 * serialised against cache hits here via the inode->i_lock and igrab() in
695 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
696 * racing with us, and it avoids needing to grab a spinlock here for every inode
697 * we drop the final reference on.
703 struct xfs_inode *ip = XFS_I(inode);
706 * If this unlinked inode is in the middle of recovery, don't
707 * drop the inode just yet; log recovery will take care of
708 * that. See the comment for this inode flag.
710 if (ip->i_flags & XFS_IRECOVERY) {
711 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
715 return generic_drop_inode(inode);
720 struct xfs_mount *mp)
723 kfree(mp->m_logname);
729 struct super_block *sb,
732 struct xfs_mount *mp = XFS_M(sb);
734 trace_xfs_fs_sync_fs(mp, __return_address);
737 * Doing anything during the async pass would be counterproductive.
742 xfs_log_force(mp, XFS_LOG_SYNC);
745 * The disk must be active because we're syncing.
746 * We schedule log work now (now that the disk is
747 * active) instead of later (when it might not be).
749 flush_delayed_work(&mp->m_log->l_work);
753 * If we are called with page faults frozen out, it means we are about
754 * to freeze the transaction subsystem. Take the opportunity to shut
755 * down inodegc because once SB_FREEZE_FS is set it's too late to
756 * prevent inactivation races with freeze. The fs doesn't get called
757 * again by the freezing process until after SB_FREEZE_FS has been set,
758 * so it's now or never. Same logic applies to speculative allocation
759 * garbage collection.
761 * We don't care if this is a normal syncfs call that does this or
762 * freeze that does this - we can run this multiple times without issue
763 * and we won't race with a restart because a restart can only occur
764 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
766 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
767 xfs_inodegc_stop(mp);
768 xfs_blockgc_stop(mp);
776 struct dentry *dentry,
777 struct kstatfs *statp)
779 struct xfs_mount *mp = XFS_M(dentry->d_sb);
780 xfs_sb_t *sbp = &mp->m_sb;
781 struct xfs_inode *ip = XFS_I(d_inode(dentry));
782 uint64_t fakeinos, id;
789 /* Wait for whatever inactivations are in progress. */
790 xfs_inodegc_flush(mp);
792 statp->f_type = XFS_SUPER_MAGIC;
793 statp->f_namelen = MAXNAMELEN - 1;
795 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
796 statp->f_fsid = u64_to_fsid(id);
798 icount = percpu_counter_sum(&mp->m_icount);
799 ifree = percpu_counter_sum(&mp->m_ifree);
800 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
802 spin_lock(&mp->m_sb_lock);
803 statp->f_bsize = sbp->sb_blocksize;
804 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
805 statp->f_blocks = sbp->sb_dblocks - lsize;
806 spin_unlock(&mp->m_sb_lock);
808 /* make sure statp->f_bfree does not underflow */
809 statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
810 statp->f_bavail = statp->f_bfree;
812 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
813 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
814 if (M_IGEO(mp)->maxicount)
815 statp->f_files = min_t(typeof(statp->f_files),
817 M_IGEO(mp)->maxicount);
819 /* If sb_icount overshot maxicount, report actual allocation */
820 statp->f_files = max_t(typeof(statp->f_files),
824 /* make sure statp->f_ffree does not underflow */
825 ffree = statp->f_files - (icount - ifree);
826 statp->f_ffree = max_t(int64_t, ffree, 0);
829 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
830 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
831 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
832 xfs_qm_statvfs(ip, statp);
834 if (XFS_IS_REALTIME_MOUNT(mp) &&
835 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
836 statp->f_blocks = sbp->sb_rblocks;
837 statp->f_bavail = statp->f_bfree =
838 sbp->sb_frextents * sbp->sb_rextsize;
845 xfs_save_resvblks(struct xfs_mount *mp)
847 uint64_t resblks = 0;
849 mp->m_resblks_save = mp->m_resblks;
850 xfs_reserve_blocks(mp, &resblks, NULL);
854 xfs_restore_resvblks(struct xfs_mount *mp)
858 if (mp->m_resblks_save) {
859 resblks = mp->m_resblks_save;
860 mp->m_resblks_save = 0;
862 resblks = xfs_default_resblks(mp);
864 xfs_reserve_blocks(mp, &resblks, NULL);
868 * Second stage of a freeze. The data is already frozen so we only
869 * need to take care of the metadata. Once that's done sync the superblock
870 * to the log to dirty it in case of a crash while frozen. This ensures that we
871 * will recover the unlinked inode lists on the next mount.
875 struct super_block *sb)
877 struct xfs_mount *mp = XFS_M(sb);
882 * The filesystem is now frozen far enough that memory reclaim
883 * cannot safely operate on the filesystem. Hence we need to
884 * set a GFP_NOFS context here to avoid recursion deadlocks.
886 flags = memalloc_nofs_save();
887 xfs_save_resvblks(mp);
888 ret = xfs_log_quiesce(mp);
889 memalloc_nofs_restore(flags);
892 * For read-write filesystems, we need to restart the inodegc on error
893 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
894 * going to be run to restart it now. We are at SB_FREEZE_FS level
895 * here, so we can restart safely without racing with a stop in
898 if (ret && !xfs_is_readonly(mp)) {
899 xfs_blockgc_start(mp);
900 xfs_inodegc_start(mp);
908 struct super_block *sb)
910 struct xfs_mount *mp = XFS_M(sb);
912 xfs_restore_resvblks(mp);
913 xfs_log_work_queue(mp);
916 * Don't reactivate the inodegc worker on a readonly filesystem because
917 * inodes are sent directly to reclaim. Don't reactivate the blockgc
918 * worker because there are no speculative preallocations on a readonly
921 if (!xfs_is_readonly(mp)) {
922 xfs_blockgc_start(mp);
923 xfs_inodegc_start(mp);
930 * This function fills in xfs_mount_t fields based on mount args.
931 * Note: the superblock _has_ now been read in.
935 struct xfs_mount *mp)
937 /* Fail a mount where the logbuf is smaller than the log stripe */
938 if (xfs_has_logv2(mp)) {
939 if (mp->m_logbsize <= 0 &&
940 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
941 mp->m_logbsize = mp->m_sb.sb_logsunit;
942 } else if (mp->m_logbsize > 0 &&
943 mp->m_logbsize < mp->m_sb.sb_logsunit) {
945 "logbuf size must be greater than or equal to log stripe size");
949 /* Fail a mount if the logbuf is larger than 32K */
950 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
952 "logbuf size for version 1 logs must be 16K or 32K");
958 * V5 filesystems always use attr2 format for attributes.
960 if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
961 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
962 "attr2 is always enabled for V5 filesystems.");
967 * prohibit r/w mounts of read-only filesystems
969 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
971 "cannot mount a read-only filesystem as read-write");
975 if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
976 (mp->m_qflags & XFS_PQUOTA_ACCT) &&
977 !xfs_has_pquotino(mp)) {
979 "Super block does not support project and group quota together");
987 xfs_init_percpu_counters(
988 struct xfs_mount *mp)
992 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
996 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1000 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1004 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1011 percpu_counter_destroy(&mp->m_fdblocks);
1013 percpu_counter_destroy(&mp->m_ifree);
1015 percpu_counter_destroy(&mp->m_icount);
1020 xfs_reinit_percpu_counters(
1021 struct xfs_mount *mp)
1023 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1024 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1025 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1029 xfs_destroy_percpu_counters(
1030 struct xfs_mount *mp)
1032 percpu_counter_destroy(&mp->m_icount);
1033 percpu_counter_destroy(&mp->m_ifree);
1034 percpu_counter_destroy(&mp->m_fdblocks);
1035 ASSERT(xfs_is_shutdown(mp) ||
1036 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1037 percpu_counter_destroy(&mp->m_delalloc_blks);
1041 xfs_inodegc_init_percpu(
1042 struct xfs_mount *mp)
1044 struct xfs_inodegc *gc;
1047 mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1051 for_each_possible_cpu(cpu) {
1052 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1053 init_llist_head(&gc->list);
1055 INIT_WORK(&gc->work, xfs_inodegc_worker);
1061 xfs_inodegc_free_percpu(
1062 struct xfs_mount *mp)
1066 free_percpu(mp->m_inodegc);
1071 struct super_block *sb)
1073 struct xfs_mount *mp = XFS_M(sb);
1075 /* if ->fill_super failed, we have no mount to tear down */
1079 xfs_notice(mp, "Unmounting Filesystem");
1080 xfs_filestream_unmount(mp);
1084 free_percpu(mp->m_stats.xs_stats);
1085 xfs_mount_list_del(mp);
1086 xfs_inodegc_free_percpu(mp);
1087 xfs_destroy_percpu_counters(mp);
1088 xfs_destroy_mount_workqueues(mp);
1089 xfs_close_devices(mp);
1091 sb->s_fs_info = NULL;
1096 xfs_fs_nr_cached_objects(
1097 struct super_block *sb,
1098 struct shrink_control *sc)
1100 /* Paranoia: catch incorrect calls during mount setup or teardown */
1101 if (WARN_ON_ONCE(!sb->s_fs_info))
1103 return xfs_reclaim_inodes_count(XFS_M(sb));
1107 xfs_fs_free_cached_objects(
1108 struct super_block *sb,
1109 struct shrink_control *sc)
1111 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1114 static const struct super_operations xfs_super_operations = {
1115 .alloc_inode = xfs_fs_alloc_inode,
1116 .destroy_inode = xfs_fs_destroy_inode,
1117 .dirty_inode = xfs_fs_dirty_inode,
1118 .drop_inode = xfs_fs_drop_inode,
1119 .put_super = xfs_fs_put_super,
1120 .sync_fs = xfs_fs_sync_fs,
1121 .freeze_fs = xfs_fs_freeze,
1122 .unfreeze_fs = xfs_fs_unfreeze,
1123 .statfs = xfs_fs_statfs,
1124 .show_options = xfs_fs_show_options,
1125 .nr_cached_objects = xfs_fs_nr_cached_objects,
1126 .free_cached_objects = xfs_fs_free_cached_objects,
1135 int last, shift_left_factor = 0, _res;
1139 value = kstrdup(s, GFP_KERNEL);
1143 last = strlen(value) - 1;
1144 if (value[last] == 'K' || value[last] == 'k') {
1145 shift_left_factor = 10;
1148 if (value[last] == 'M' || value[last] == 'm') {
1149 shift_left_factor = 20;
1152 if (value[last] == 'G' || value[last] == 'g') {
1153 shift_left_factor = 30;
1157 if (kstrtoint(value, base, &_res))
1160 *res = _res << shift_left_factor;
1165 xfs_fs_warn_deprecated(
1166 struct fs_context *fc,
1167 struct fs_parameter *param,
1171 /* Don't print the warning if reconfiguring and current mount point
1172 * already had the flag set
1174 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1175 !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1177 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1181 * Set mount state from a mount option.
1183 * NOTE: mp->m_super is NULL here!
1187 struct fs_context *fc,
1188 struct fs_parameter *param)
1190 struct xfs_mount *parsing_mp = fc->s_fs_info;
1191 struct fs_parse_result result;
1195 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1201 parsing_mp->m_logbufs = result.uint_32;
1204 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1208 kfree(parsing_mp->m_logname);
1209 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1210 if (!parsing_mp->m_logname)
1214 kfree(parsing_mp->m_rtname);
1215 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1216 if (!parsing_mp->m_rtname)
1220 if (suffix_kstrtoint(param->string, 10, &size))
1222 parsing_mp->m_allocsize_log = ffs(size) - 1;
1223 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1227 parsing_mp->m_features |= XFS_FEAT_GRPID;
1230 case Opt_sysvgroups:
1231 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1234 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1236 case Opt_norecovery:
1237 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1240 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1243 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1246 parsing_mp->m_dalign = result.uint_32;
1249 parsing_mp->m_swidth = result.uint_32;
1252 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1255 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1258 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1261 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1264 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1266 case Opt_filestreams:
1267 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1270 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1271 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1276 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1278 case Opt_qnoenforce:
1279 case Opt_uqnoenforce:
1280 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1281 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1285 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1287 case Opt_pqnoenforce:
1288 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1289 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1293 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1295 case Opt_gqnoenforce:
1296 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1297 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1300 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1303 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1305 #ifdef CONFIG_FS_DAX
1307 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1310 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1313 /* Following mount options will be removed in September 2025 */
1315 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1316 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1319 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1320 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1323 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1324 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1327 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1328 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1331 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1339 xfs_fs_validate_params(
1340 struct xfs_mount *mp)
1342 /* No recovery flag requires a read-only mount */
1343 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1344 xfs_warn(mp, "no-recovery mounts must be read-only.");
1349 * We have not read the superblock at this point, so only the attr2
1350 * mount option can set the attr2 feature by this stage.
1352 if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1353 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1358 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1360 "sunit and swidth options incompatible with the noalign option");
1364 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1365 xfs_warn(mp, "quota support not available in this kernel.");
1369 if ((mp->m_dalign && !mp->m_swidth) ||
1370 (!mp->m_dalign && mp->m_swidth)) {
1371 xfs_warn(mp, "sunit and swidth must be specified together");
1375 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1377 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1378 mp->m_swidth, mp->m_dalign);
1382 if (mp->m_logbufs != -1 &&
1383 mp->m_logbufs != 0 &&
1384 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1385 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1386 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1387 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1391 if (mp->m_logbsize != -1 &&
1392 mp->m_logbsize != 0 &&
1393 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1394 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1395 !is_power_of_2(mp->m_logbsize))) {
1397 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1402 if (xfs_has_allocsize(mp) &&
1403 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1404 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1405 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1406 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1415 struct super_block *sb,
1416 struct fs_context *fc)
1418 struct xfs_mount *mp = sb->s_fs_info;
1420 int flags = 0, error;
1424 error = xfs_fs_validate_params(mp);
1426 goto out_free_names;
1428 sb_min_blocksize(sb, BBSIZE);
1429 sb->s_xattr = xfs_xattr_handlers;
1430 sb->s_export_op = &xfs_export_operations;
1431 #ifdef CONFIG_XFS_QUOTA
1432 sb->s_qcop = &xfs_quotactl_operations;
1433 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1435 sb->s_op = &xfs_super_operations;
1438 * Delay mount work if the debug hook is set. This is debug
1439 * instrumention to coordinate simulation of xfs mount failures with
1440 * VFS superblock operations
1442 if (xfs_globals.mount_delay) {
1443 xfs_notice(mp, "Delaying mount for %d seconds.",
1444 xfs_globals.mount_delay);
1445 msleep(xfs_globals.mount_delay * 1000);
1448 if (fc->sb_flags & SB_SILENT)
1449 flags |= XFS_MFSI_QUIET;
1451 error = xfs_open_devices(mp);
1453 goto out_free_names;
1455 error = xfs_init_mount_workqueues(mp);
1457 goto out_close_devices;
1459 error = xfs_init_percpu_counters(mp);
1461 goto out_destroy_workqueues;
1463 error = xfs_inodegc_init_percpu(mp);
1465 goto out_destroy_counters;
1468 * All percpu data structures requiring cleanup when a cpu goes offline
1469 * must be allocated before adding this @mp to the cpu-dead handler's
1472 xfs_mount_list_add(mp);
1474 /* Allocate stats memory before we do operations that might use it */
1475 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1476 if (!mp->m_stats.xs_stats) {
1478 goto out_destroy_inodegc;
1481 error = xfs_readsb(mp, flags);
1483 goto out_free_stats;
1485 error = xfs_finish_flags(mp);
1489 error = xfs_setup_devices(mp);
1493 /* V4 support is undergoing deprecation. */
1494 if (!xfs_has_crc(mp)) {
1495 #ifdef CONFIG_XFS_SUPPORT_V4
1497 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1500 "Deprecated V4 format (crc=0) not supported by kernel.");
1506 /* Filesystem claims it needs repair, so refuse the mount. */
1507 if (xfs_has_needsrepair(mp)) {
1508 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1509 error = -EFSCORRUPTED;
1514 * Don't touch the filesystem if a user tool thinks it owns the primary
1515 * superblock. mkfs doesn't clear the flag from secondary supers, so
1516 * we don't check them at all.
1518 if (mp->m_sb.sb_inprogress) {
1519 xfs_warn(mp, "Offline file system operation in progress!");
1520 error = -EFSCORRUPTED;
1525 * Until this is fixed only page-sized or smaller data blocks work.
1527 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1529 "File system with blocksize %d bytes. "
1530 "Only pagesize (%ld) or less will currently work.",
1531 mp->m_sb.sb_blocksize, PAGE_SIZE);
1536 /* Ensure this filesystem fits in the page cache limits */
1537 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1538 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1540 "file system too large to be mounted on this system.");
1546 * XFS block mappings use 54 bits to store the logical block offset.
1547 * This should suffice to handle the maximum file size that the VFS
1548 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1549 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1550 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1551 * to check this assertion.
1553 * Avoid integer overflow by comparing the maximum bmbt offset to the
1554 * maximum pagecache offset in units of fs blocks.
1556 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1558 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1559 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1565 error = xfs_filestream_mount(mp);
1570 * we must configure the block size in the superblock before we run the
1571 * full mount process as the mount process can lookup and cache inodes.
1573 sb->s_magic = XFS_SUPER_MAGIC;
1574 sb->s_blocksize = mp->m_sb.sb_blocksize;
1575 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1576 sb->s_maxbytes = MAX_LFS_FILESIZE;
1577 sb->s_max_links = XFS_MAXLINK;
1578 sb->s_time_gran = 1;
1579 if (xfs_has_bigtime(mp)) {
1580 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1581 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1583 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1584 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1586 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1587 sb->s_iflags |= SB_I_CGROUPWB;
1589 set_posix_acl_flag(sb);
1591 /* version 5 superblocks support inode version counters. */
1592 if (xfs_has_crc(mp))
1593 sb->s_flags |= SB_I_VERSION;
1595 if (xfs_has_dax_always(mp)) {
1596 bool rtdev_is_dax = false, datadev_is_dax;
1599 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1601 datadev_is_dax = xfs_buftarg_is_dax(sb, mp->m_ddev_targp);
1602 if (mp->m_rtdev_targp)
1603 rtdev_is_dax = xfs_buftarg_is_dax(sb,
1605 if (!rtdev_is_dax && !datadev_is_dax) {
1607 "DAX unsupported by block device. Turning off DAX.");
1608 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1610 if (xfs_has_reflink(mp)) {
1612 "DAX and reflink cannot be used together!");
1614 goto out_filestream_unmount;
1618 if (xfs_has_discard(mp)) {
1619 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1621 if (!blk_queue_discard(q)) {
1622 xfs_warn(mp, "mounting with \"discard\" option, but "
1623 "the device does not support discard");
1624 mp->m_features &= ~XFS_FEAT_DISCARD;
1628 if (xfs_has_reflink(mp)) {
1629 if (mp->m_sb.sb_rblocks) {
1631 "reflink not compatible with realtime device!");
1633 goto out_filestream_unmount;
1636 if (xfs_globals.always_cow) {
1637 xfs_info(mp, "using DEBUG-only always_cow mode.");
1638 mp->m_always_cow = true;
1642 if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1644 "reverse mapping btree not compatible with realtime device!");
1646 goto out_filestream_unmount;
1649 error = xfs_mountfs(mp);
1651 goto out_filestream_unmount;
1653 root = igrab(VFS_I(mp->m_rootip));
1658 sb->s_root = d_make_root(root);
1666 out_filestream_unmount:
1667 xfs_filestream_unmount(mp);
1671 free_percpu(mp->m_stats.xs_stats);
1672 out_destroy_inodegc:
1673 xfs_mount_list_del(mp);
1674 xfs_inodegc_free_percpu(mp);
1675 out_destroy_counters:
1676 xfs_destroy_percpu_counters(mp);
1677 out_destroy_workqueues:
1678 xfs_destroy_mount_workqueues(mp);
1680 xfs_close_devices(mp);
1682 sb->s_fs_info = NULL;
1687 xfs_filestream_unmount(mp);
1694 struct fs_context *fc)
1696 return get_tree_bdev(fc, xfs_fs_fill_super);
1701 struct xfs_mount *mp)
1703 struct xfs_sb *sbp = &mp->m_sb;
1706 if (xfs_has_norecovery(mp)) {
1708 "ro->rw transition prohibited on norecovery mount");
1712 if (xfs_sb_is_v5(sbp) &&
1713 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1715 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1716 (sbp->sb_features_ro_compat &
1717 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1721 clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1724 * If this is the first remount to writeable state we might have some
1725 * superblock changes to update.
1727 if (mp->m_update_sb) {
1728 error = xfs_sync_sb(mp, false);
1730 xfs_warn(mp, "failed to write sb changes");
1733 mp->m_update_sb = false;
1737 * Fill out the reserve pool if it is empty. Use the stashed value if
1738 * it is non-zero, otherwise go with the default.
1740 xfs_restore_resvblks(mp);
1741 xfs_log_work_queue(mp);
1743 /* Recover any CoW blocks that never got remapped. */
1744 error = xfs_reflink_recover_cow(mp);
1747 "Error %d recovering leftover CoW allocations.", error);
1748 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1751 xfs_blockgc_start(mp);
1753 /* Create the per-AG metadata reservation pool .*/
1754 error = xfs_fs_reserve_ag_blocks(mp);
1755 if (error && error != -ENOSPC)
1758 /* Re-enable the background inode inactivation worker. */
1759 xfs_inodegc_start(mp);
1766 struct xfs_mount *mp)
1768 struct xfs_icwalk icw = {
1769 .icw_flags = XFS_ICWALK_FLAG_SYNC,
1774 * Cancel background eofb scanning so it cannot race with the final
1775 * log force+buftarg wait and deadlock the remount.
1777 xfs_blockgc_stop(mp);
1780 * Clear out all remaining COW staging extents and speculative post-EOF
1781 * preallocations so that we don't leave inodes requiring inactivation
1782 * cleanups during reclaim on a read-only mount. We must process every
1783 * cached inode, so this requires a synchronous cache scan.
1785 error = xfs_blockgc_free_space(mp, &icw);
1787 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1792 * Stop the inodegc background worker. xfs_fs_reconfigure already
1793 * flushed all pending inodegc work when it sync'd the filesystem.
1794 * The VFS holds s_umount, so we know that inodes cannot enter
1795 * xfs_fs_destroy_inode during a remount operation. In readonly mode
1796 * we send inodes straight to reclaim, so no inodes will be queued.
1798 xfs_inodegc_stop(mp);
1800 /* Free the per-AG metadata reservation pool. */
1801 error = xfs_fs_unreserve_ag_blocks(mp);
1803 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1808 * Before we sync the metadata, we need to free up the reserve block
1809 * pool so that the used block count in the superblock on disk is
1810 * correct at the end of the remount. Stash the current* reserve pool
1811 * size so that if we get remounted rw, we can return it to the same
1814 xfs_save_resvblks(mp);
1817 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1823 * Logically we would return an error here to prevent users from believing
1824 * they might have changed mount options using remount which can't be changed.
1826 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1827 * arguments in some cases so we can't blindly reject options, but have to
1828 * check for each specified option if it actually differs from the currently
1829 * set option and only reject it if that's the case.
1831 * Until that is implemented we return success for every remount request, and
1832 * silently ignore all options that we can't actually change.
1836 struct fs_context *fc)
1838 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1839 struct xfs_mount *new_mp = fc->s_fs_info;
1840 int flags = fc->sb_flags;
1843 /* version 5 superblocks always support version counters. */
1844 if (xfs_has_crc(mp))
1845 fc->sb_flags |= SB_I_VERSION;
1847 error = xfs_fs_validate_params(new_mp);
1851 sync_filesystem(mp->m_super);
1853 /* inode32 -> inode64 */
1854 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1855 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1856 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1859 /* inode64 -> inode32 */
1860 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1861 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1862 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1866 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1867 error = xfs_remount_rw(mp);
1873 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1874 error = xfs_remount_ro(mp);
1882 static void xfs_fs_free(
1883 struct fs_context *fc)
1885 struct xfs_mount *mp = fc->s_fs_info;
1888 * mp is stored in the fs_context when it is initialized.
1889 * mp is transferred to the superblock on a successful mount,
1890 * but if an error occurs before the transfer we have to free
1897 static const struct fs_context_operations xfs_context_ops = {
1898 .parse_param = xfs_fs_parse_param,
1899 .get_tree = xfs_fs_get_tree,
1900 .reconfigure = xfs_fs_reconfigure,
1901 .free = xfs_fs_free,
1904 static int xfs_init_fs_context(
1905 struct fs_context *fc)
1907 struct xfs_mount *mp;
1909 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1913 spin_lock_init(&mp->m_sb_lock);
1914 spin_lock_init(&mp->m_agirotor_lock);
1915 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1916 spin_lock_init(&mp->m_perag_lock);
1917 mutex_init(&mp->m_growlock);
1918 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1919 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1920 mp->m_kobj.kobject.kset = xfs_kset;
1922 * We don't create the finobt per-ag space reservation until after log
1923 * recovery, so we must set this to true so that an ifree transaction
1924 * started during log recovery will not depend on space reservations
1925 * for finobt expansion.
1927 mp->m_finobt_nores = true;
1930 * These can be overridden by the mount option parsing.
1933 mp->m_logbsize = -1;
1934 mp->m_allocsize_log = 16; /* 64k */
1937 * Copy binary VFS mount flags we are interested in.
1939 if (fc->sb_flags & SB_RDONLY)
1940 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1941 if (fc->sb_flags & SB_DIRSYNC)
1942 mp->m_features |= XFS_FEAT_DIRSYNC;
1943 if (fc->sb_flags & SB_SYNCHRONOUS)
1944 mp->m_features |= XFS_FEAT_WSYNC;
1947 fc->ops = &xfs_context_ops;
1952 static struct file_system_type xfs_fs_type = {
1953 .owner = THIS_MODULE,
1955 .init_fs_context = xfs_init_fs_context,
1956 .parameters = xfs_fs_parameters,
1957 .kill_sb = kill_block_super,
1958 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1960 MODULE_ALIAS_FS("xfs");
1963 xfs_init_caches(void)
1967 xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1968 sizeof(struct xlog_ticket),
1970 if (!xfs_log_ticket_cache)
1973 error = xfs_btree_init_cur_caches();
1975 goto out_destroy_log_ticket_cache;
1977 error = xfs_defer_init_item_caches();
1979 goto out_destroy_btree_cur_cache;
1981 xfs_da_state_cache = kmem_cache_create("xfs_da_state",
1982 sizeof(struct xfs_da_state),
1984 if (!xfs_da_state_cache)
1985 goto out_destroy_defer_item_cache;
1987 xfs_ifork_cache = kmem_cache_create("xfs_ifork",
1988 sizeof(struct xfs_ifork),
1990 if (!xfs_ifork_cache)
1991 goto out_destroy_da_state_cache;
1993 xfs_trans_cache = kmem_cache_create("xfs_trans",
1994 sizeof(struct xfs_trans),
1996 if (!xfs_trans_cache)
1997 goto out_destroy_ifork_cache;
2001 * The size of the cache-allocated buf log item is the maximum
2002 * size possible under XFS. This wastes a little bit of memory,
2003 * but it is much faster.
2005 xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2006 sizeof(struct xfs_buf_log_item),
2008 if (!xfs_buf_item_cache)
2009 goto out_destroy_trans_cache;
2011 xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2012 (sizeof(struct xfs_efd_log_item) +
2013 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2014 sizeof(struct xfs_extent)),
2017 goto out_destroy_buf_item_cache;
2019 xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2020 (sizeof(struct xfs_efi_log_item) +
2021 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2022 sizeof(struct xfs_extent)),
2025 goto out_destroy_efd_cache;
2027 xfs_inode_cache = kmem_cache_create("xfs_inode",
2028 sizeof(struct xfs_inode), 0,
2029 (SLAB_HWCACHE_ALIGN |
2030 SLAB_RECLAIM_ACCOUNT |
2031 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2032 xfs_fs_inode_init_once);
2033 if (!xfs_inode_cache)
2034 goto out_destroy_efi_cache;
2036 xfs_ili_cache = kmem_cache_create("xfs_ili",
2037 sizeof(struct xfs_inode_log_item), 0,
2038 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2041 goto out_destroy_inode_cache;
2043 xfs_icreate_cache = kmem_cache_create("xfs_icr",
2044 sizeof(struct xfs_icreate_item),
2046 if (!xfs_icreate_cache)
2047 goto out_destroy_ili_cache;
2049 xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2050 sizeof(struct xfs_rud_log_item),
2053 goto out_destroy_icreate_cache;
2055 xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2056 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2059 goto out_destroy_rud_cache;
2061 xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2062 sizeof(struct xfs_cud_log_item),
2065 goto out_destroy_rui_cache;
2067 xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2068 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2071 goto out_destroy_cud_cache;
2073 xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2074 sizeof(struct xfs_bud_log_item),
2077 goto out_destroy_cui_cache;
2079 xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2080 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2083 goto out_destroy_bud_cache;
2087 out_destroy_bud_cache:
2088 kmem_cache_destroy(xfs_bud_cache);
2089 out_destroy_cui_cache:
2090 kmem_cache_destroy(xfs_cui_cache);
2091 out_destroy_cud_cache:
2092 kmem_cache_destroy(xfs_cud_cache);
2093 out_destroy_rui_cache:
2094 kmem_cache_destroy(xfs_rui_cache);
2095 out_destroy_rud_cache:
2096 kmem_cache_destroy(xfs_rud_cache);
2097 out_destroy_icreate_cache:
2098 kmem_cache_destroy(xfs_icreate_cache);
2099 out_destroy_ili_cache:
2100 kmem_cache_destroy(xfs_ili_cache);
2101 out_destroy_inode_cache:
2102 kmem_cache_destroy(xfs_inode_cache);
2103 out_destroy_efi_cache:
2104 kmem_cache_destroy(xfs_efi_cache);
2105 out_destroy_efd_cache:
2106 kmem_cache_destroy(xfs_efd_cache);
2107 out_destroy_buf_item_cache:
2108 kmem_cache_destroy(xfs_buf_item_cache);
2109 out_destroy_trans_cache:
2110 kmem_cache_destroy(xfs_trans_cache);
2111 out_destroy_ifork_cache:
2112 kmem_cache_destroy(xfs_ifork_cache);
2113 out_destroy_da_state_cache:
2114 kmem_cache_destroy(xfs_da_state_cache);
2115 out_destroy_defer_item_cache:
2116 xfs_defer_destroy_item_caches();
2117 out_destroy_btree_cur_cache:
2118 xfs_btree_destroy_cur_caches();
2119 out_destroy_log_ticket_cache:
2120 kmem_cache_destroy(xfs_log_ticket_cache);
2126 xfs_destroy_caches(void)
2129 * Make sure all delayed rcu free are flushed before we
2133 kmem_cache_destroy(xfs_bui_cache);
2134 kmem_cache_destroy(xfs_bud_cache);
2135 kmem_cache_destroy(xfs_cui_cache);
2136 kmem_cache_destroy(xfs_cud_cache);
2137 kmem_cache_destroy(xfs_rui_cache);
2138 kmem_cache_destroy(xfs_rud_cache);
2139 kmem_cache_destroy(xfs_icreate_cache);
2140 kmem_cache_destroy(xfs_ili_cache);
2141 kmem_cache_destroy(xfs_inode_cache);
2142 kmem_cache_destroy(xfs_efi_cache);
2143 kmem_cache_destroy(xfs_efd_cache);
2144 kmem_cache_destroy(xfs_buf_item_cache);
2145 kmem_cache_destroy(xfs_trans_cache);
2146 kmem_cache_destroy(xfs_ifork_cache);
2147 kmem_cache_destroy(xfs_da_state_cache);
2148 xfs_defer_destroy_item_caches();
2149 xfs_btree_destroy_cur_caches();
2150 kmem_cache_destroy(xfs_log_ticket_cache);
2154 xfs_init_workqueues(void)
2157 * The allocation workqueue can be used in memory reclaim situations
2158 * (writepage path), and parallelism is only limited by the number of
2159 * AGs in all the filesystems mounted. Hence use the default large
2160 * max_active value for this workqueue.
2162 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2163 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2167 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2169 if (!xfs_discard_wq)
2170 goto out_free_alloc_wq;
2174 destroy_workqueue(xfs_alloc_wq);
2179 xfs_destroy_workqueues(void)
2181 destroy_workqueue(xfs_discard_wq);
2182 destroy_workqueue(xfs_alloc_wq);
2185 #ifdef CONFIG_HOTPLUG_CPU
2190 struct xfs_mount *mp, *n;
2192 spin_lock(&xfs_mount_list_lock);
2193 list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2194 spin_unlock(&xfs_mount_list_lock);
2195 xfs_inodegc_cpu_dead(mp, cpu);
2196 spin_lock(&xfs_mount_list_lock);
2198 spin_unlock(&xfs_mount_list_lock);
2203 xfs_cpu_hotplug_init(void)
2207 error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2211 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2217 xfs_cpu_hotplug_destroy(void)
2219 cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2222 #else /* !CONFIG_HOTPLUG_CPU */
2223 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2224 static inline void xfs_cpu_hotplug_destroy(void) {}
2232 xfs_check_ondisk_structs();
2234 printk(KERN_INFO XFS_VERSION_STRING " with "
2235 XFS_BUILD_OPTIONS " enabled\n");
2239 error = xfs_cpu_hotplug_init();
2243 error = xfs_init_caches();
2245 goto out_destroy_hp;
2247 error = xfs_init_workqueues();
2249 goto out_destroy_caches;
2251 error = xfs_mru_cache_init();
2253 goto out_destroy_wq;
2255 error = xfs_buf_init();
2257 goto out_mru_cache_uninit;
2259 error = xfs_init_procfs();
2261 goto out_buf_terminate;
2263 error = xfs_sysctl_register();
2265 goto out_cleanup_procfs;
2267 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2270 goto out_sysctl_unregister;
2273 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2275 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2276 if (!xfsstats.xs_stats) {
2278 goto out_kset_unregister;
2281 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2284 goto out_free_stats;
2287 xfs_dbg_kobj.kobject.kset = xfs_kset;
2288 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2290 goto out_remove_stats_kobj;
2293 error = xfs_qm_init();
2295 goto out_remove_dbg_kobj;
2297 error = register_filesystem(&xfs_fs_type);
2304 out_remove_dbg_kobj:
2306 xfs_sysfs_del(&xfs_dbg_kobj);
2307 out_remove_stats_kobj:
2309 xfs_sysfs_del(&xfsstats.xs_kobj);
2311 free_percpu(xfsstats.xs_stats);
2312 out_kset_unregister:
2313 kset_unregister(xfs_kset);
2314 out_sysctl_unregister:
2315 xfs_sysctl_unregister();
2317 xfs_cleanup_procfs();
2319 xfs_buf_terminate();
2320 out_mru_cache_uninit:
2321 xfs_mru_cache_uninit();
2323 xfs_destroy_workqueues();
2325 xfs_destroy_caches();
2327 xfs_cpu_hotplug_destroy();
2336 unregister_filesystem(&xfs_fs_type);
2338 xfs_sysfs_del(&xfs_dbg_kobj);
2340 xfs_sysfs_del(&xfsstats.xs_kobj);
2341 free_percpu(xfsstats.xs_stats);
2342 kset_unregister(xfs_kset);
2343 xfs_sysctl_unregister();
2344 xfs_cleanup_procfs();
2345 xfs_buf_terminate();
2346 xfs_mru_cache_uninit();
2347 xfs_destroy_workqueues();
2348 xfs_destroy_caches();
2349 xfs_uuid_table_free();
2350 xfs_cpu_hotplug_destroy();
2353 module_init(init_xfs_fs);
2354 module_exit(exit_xfs_fs);
2356 MODULE_AUTHOR("Silicon Graphics, Inc.");
2357 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2358 MODULE_LICENSE("GPL");