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;
335 xfs_setup_dax_always(
336 struct xfs_mount *mp)
338 if (!mp->m_ddev_targp->bt_daxdev &&
339 (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
341 "DAX unsupported by block device. Turning off DAX.");
345 if (mp->m_super->s_blocksize != PAGE_SIZE) {
347 "DAX not supported for blocksize. Turning off DAX.");
351 if (xfs_has_reflink(mp)) {
352 xfs_alert(mp, "DAX and reflink cannot be used together!");
356 xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
360 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
368 struct block_device **bdevp)
372 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
374 if (IS_ERR(*bdevp)) {
375 error = PTR_ERR(*bdevp);
376 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
384 struct block_device *bdev)
387 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
392 struct xfs_mount *mp)
394 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
395 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
397 xfs_free_buftarg(mp->m_logdev_targp);
398 xfs_blkdev_put(logdev);
400 if (mp->m_rtdev_targp) {
401 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
403 xfs_free_buftarg(mp->m_rtdev_targp);
404 xfs_blkdev_put(rtdev);
406 xfs_free_buftarg(mp->m_ddev_targp);
410 * The file system configurations are:
411 * (1) device (partition) with data and internal log
412 * (2) logical volume with data and log subvolumes.
413 * (3) logical volume with data, log, and realtime subvolumes.
415 * We only have to handle opening the log and realtime volumes here if
416 * they are present. The data subvolume has already been opened by
417 * get_sb_bdev() and is stored in sb->s_bdev.
421 struct xfs_mount *mp)
423 struct block_device *ddev = mp->m_super->s_bdev;
424 struct block_device *logdev = NULL, *rtdev = NULL;
428 * Open real time and log devices - order is important.
431 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
437 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
439 goto out_close_logdev;
441 if (rtdev == ddev || rtdev == logdev) {
443 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
445 goto out_close_rtdev;
450 * Setup xfs_mount buffer target pointers
453 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
454 if (!mp->m_ddev_targp)
455 goto out_close_rtdev;
458 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
459 if (!mp->m_rtdev_targp)
460 goto out_free_ddev_targ;
463 if (logdev && logdev != ddev) {
464 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
465 if (!mp->m_logdev_targp)
466 goto out_free_rtdev_targ;
468 mp->m_logdev_targp = mp->m_ddev_targp;
474 if (mp->m_rtdev_targp)
475 xfs_free_buftarg(mp->m_rtdev_targp);
477 xfs_free_buftarg(mp->m_ddev_targp);
479 xfs_blkdev_put(rtdev);
481 if (logdev && logdev != ddev)
482 xfs_blkdev_put(logdev);
487 * Setup xfs_mount buffer target pointers based on superblock
491 struct xfs_mount *mp)
495 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
499 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
500 unsigned int log_sector_size = BBSIZE;
502 if (xfs_has_sector(mp))
503 log_sector_size = mp->m_sb.sb_logsectsize;
504 error = xfs_setsize_buftarg(mp->m_logdev_targp,
509 if (mp->m_rtdev_targp) {
510 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
511 mp->m_sb.sb_sectsize);
520 xfs_init_mount_workqueues(
521 struct xfs_mount *mp)
523 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
524 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
525 1, mp->m_super->s_id);
526 if (!mp->m_buf_workqueue)
529 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
530 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
531 0, mp->m_super->s_id);
532 if (!mp->m_unwritten_workqueue)
533 goto out_destroy_buf;
535 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
536 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
537 0, mp->m_super->s_id);
538 if (!mp->m_reclaim_workqueue)
539 goto out_destroy_unwritten;
541 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
542 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
543 0, mp->m_super->s_id);
544 if (!mp->m_blockgc_wq)
545 goto out_destroy_reclaim;
547 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
548 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
549 1, mp->m_super->s_id);
550 if (!mp->m_inodegc_wq)
551 goto out_destroy_blockgc;
553 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
554 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
555 if (!mp->m_sync_workqueue)
556 goto out_destroy_inodegc;
561 destroy_workqueue(mp->m_inodegc_wq);
563 destroy_workqueue(mp->m_blockgc_wq);
565 destroy_workqueue(mp->m_reclaim_workqueue);
566 out_destroy_unwritten:
567 destroy_workqueue(mp->m_unwritten_workqueue);
569 destroy_workqueue(mp->m_buf_workqueue);
575 xfs_destroy_mount_workqueues(
576 struct xfs_mount *mp)
578 destroy_workqueue(mp->m_sync_workqueue);
579 destroy_workqueue(mp->m_blockgc_wq);
580 destroy_workqueue(mp->m_inodegc_wq);
581 destroy_workqueue(mp->m_reclaim_workqueue);
582 destroy_workqueue(mp->m_unwritten_workqueue);
583 destroy_workqueue(mp->m_buf_workqueue);
587 xfs_flush_inodes_worker(
588 struct work_struct *work)
590 struct xfs_mount *mp = container_of(work, struct xfs_mount,
591 m_flush_inodes_work);
592 struct super_block *sb = mp->m_super;
594 if (down_read_trylock(&sb->s_umount)) {
596 up_read(&sb->s_umount);
601 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
602 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
603 * for IO to complete so that we effectively throttle multiple callers to the
604 * rate at which IO is completing.
608 struct xfs_mount *mp)
611 * If flush_work() returns true then that means we waited for a flush
612 * which was already in progress. Don't bother running another scan.
614 if (flush_work(&mp->m_flush_inodes_work))
617 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
618 flush_work(&mp->m_flush_inodes_work);
621 /* Catch misguided souls that try to use this interface on XFS */
622 STATIC struct inode *
624 struct super_block *sb)
631 * Now that the generic code is guaranteed not to be accessing
632 * the linux inode, we can inactivate and reclaim the inode.
635 xfs_fs_destroy_inode(
638 struct xfs_inode *ip = XFS_I(inode);
640 trace_xfs_destroy_inode(ip);
642 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
643 XFS_STATS_INC(ip->i_mount, vn_rele);
644 XFS_STATS_INC(ip->i_mount, vn_remove);
645 xfs_inode_mark_reclaimable(ip);
653 struct xfs_inode *ip = XFS_I(inode);
654 struct xfs_mount *mp = ip->i_mount;
655 struct xfs_trans *tp;
657 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
659 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
662 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
664 xfs_ilock(ip, XFS_ILOCK_EXCL);
665 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
666 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
667 xfs_trans_commit(tp);
671 * Slab object creation initialisation for the XFS inode.
672 * This covers only the idempotent fields in the XFS inode;
673 * all other fields need to be initialised on allocation
674 * from the slab. This avoids the need to repeatedly initialise
675 * fields in the xfs inode that left in the initialise state
676 * when freeing the inode.
679 xfs_fs_inode_init_once(
682 struct xfs_inode *ip = inode;
684 memset(ip, 0, sizeof(struct xfs_inode));
687 inode_init_once(VFS_I(ip));
690 atomic_set(&ip->i_pincount, 0);
691 spin_lock_init(&ip->i_flags_lock);
693 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
694 "xfsino", ip->i_ino);
698 * We do an unlocked check for XFS_IDONTCACHE here because we are already
699 * serialised against cache hits here via the inode->i_lock and igrab() in
700 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
701 * racing with us, and it avoids needing to grab a spinlock here for every inode
702 * we drop the final reference on.
708 struct xfs_inode *ip = XFS_I(inode);
711 * If this unlinked inode is in the middle of recovery, don't
712 * drop the inode just yet; log recovery will take care of
713 * that. See the comment for this inode flag.
715 if (ip->i_flags & XFS_IRECOVERY) {
716 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
720 return generic_drop_inode(inode);
725 struct xfs_mount *mp)
728 kfree(mp->m_logname);
734 struct super_block *sb,
737 struct xfs_mount *mp = XFS_M(sb);
740 trace_xfs_fs_sync_fs(mp, __return_address);
743 * Doing anything during the async pass would be counterproductive.
748 error = xfs_log_force(mp, XFS_LOG_SYNC);
754 * The disk must be active because we're syncing.
755 * We schedule log work now (now that the disk is
756 * active) instead of later (when it might not be).
758 flush_delayed_work(&mp->m_log->l_work);
762 * If we are called with page faults frozen out, it means we are about
763 * to freeze the transaction subsystem. Take the opportunity to shut
764 * down inodegc because once SB_FREEZE_FS is set it's too late to
765 * prevent inactivation races with freeze. The fs doesn't get called
766 * again by the freezing process until after SB_FREEZE_FS has been set,
767 * so it's now or never. Same logic applies to speculative allocation
768 * garbage collection.
770 * We don't care if this is a normal syncfs call that does this or
771 * freeze that does this - we can run this multiple times without issue
772 * and we won't race with a restart because a restart can only occur
773 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
775 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
776 xfs_inodegc_stop(mp);
777 xfs_blockgc_stop(mp);
785 struct dentry *dentry,
786 struct kstatfs *statp)
788 struct xfs_mount *mp = XFS_M(dentry->d_sb);
789 xfs_sb_t *sbp = &mp->m_sb;
790 struct xfs_inode *ip = XFS_I(d_inode(dentry));
791 uint64_t fakeinos, id;
798 /* Wait for whatever inactivations are in progress. */
799 xfs_inodegc_flush(mp);
801 statp->f_type = XFS_SUPER_MAGIC;
802 statp->f_namelen = MAXNAMELEN - 1;
804 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
805 statp->f_fsid = u64_to_fsid(id);
807 icount = percpu_counter_sum(&mp->m_icount);
808 ifree = percpu_counter_sum(&mp->m_ifree);
809 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
811 spin_lock(&mp->m_sb_lock);
812 statp->f_bsize = sbp->sb_blocksize;
813 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
814 statp->f_blocks = sbp->sb_dblocks - lsize;
815 spin_unlock(&mp->m_sb_lock);
817 /* make sure statp->f_bfree does not underflow */
818 statp->f_bfree = max_t(int64_t, 0,
819 fdblocks - xfs_fdblocks_unavailable(mp));
820 statp->f_bavail = statp->f_bfree;
822 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
823 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
824 if (M_IGEO(mp)->maxicount)
825 statp->f_files = min_t(typeof(statp->f_files),
827 M_IGEO(mp)->maxicount);
829 /* If sb_icount overshot maxicount, report actual allocation */
830 statp->f_files = max_t(typeof(statp->f_files),
834 /* make sure statp->f_ffree does not underflow */
835 ffree = statp->f_files - (icount - ifree);
836 statp->f_ffree = max_t(int64_t, ffree, 0);
839 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
840 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
841 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
842 xfs_qm_statvfs(ip, statp);
844 if (XFS_IS_REALTIME_MOUNT(mp) &&
845 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
846 statp->f_blocks = sbp->sb_rblocks;
847 statp->f_bavail = statp->f_bfree =
848 sbp->sb_frextents * sbp->sb_rextsize;
855 xfs_save_resvblks(struct xfs_mount *mp)
857 uint64_t resblks = 0;
859 mp->m_resblks_save = mp->m_resblks;
860 xfs_reserve_blocks(mp, &resblks, NULL);
864 xfs_restore_resvblks(struct xfs_mount *mp)
868 if (mp->m_resblks_save) {
869 resblks = mp->m_resblks_save;
870 mp->m_resblks_save = 0;
872 resblks = xfs_default_resblks(mp);
874 xfs_reserve_blocks(mp, &resblks, NULL);
878 * Second stage of a freeze. The data is already frozen so we only
879 * need to take care of the metadata. Once that's done sync the superblock
880 * to the log to dirty it in case of a crash while frozen. This ensures that we
881 * will recover the unlinked inode lists on the next mount.
885 struct super_block *sb)
887 struct xfs_mount *mp = XFS_M(sb);
892 * The filesystem is now frozen far enough that memory reclaim
893 * cannot safely operate on the filesystem. Hence we need to
894 * set a GFP_NOFS context here to avoid recursion deadlocks.
896 flags = memalloc_nofs_save();
897 xfs_save_resvblks(mp);
898 ret = xfs_log_quiesce(mp);
899 memalloc_nofs_restore(flags);
902 * For read-write filesystems, we need to restart the inodegc on error
903 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
904 * going to be run to restart it now. We are at SB_FREEZE_FS level
905 * here, so we can restart safely without racing with a stop in
908 if (ret && !xfs_is_readonly(mp)) {
909 xfs_blockgc_start(mp);
910 xfs_inodegc_start(mp);
918 struct super_block *sb)
920 struct xfs_mount *mp = XFS_M(sb);
922 xfs_restore_resvblks(mp);
923 xfs_log_work_queue(mp);
926 * Don't reactivate the inodegc worker on a readonly filesystem because
927 * inodes are sent directly to reclaim. Don't reactivate the blockgc
928 * worker because there are no speculative preallocations on a readonly
931 if (!xfs_is_readonly(mp)) {
932 xfs_blockgc_start(mp);
933 xfs_inodegc_start(mp);
940 * This function fills in xfs_mount_t fields based on mount args.
941 * Note: the superblock _has_ now been read in.
945 struct xfs_mount *mp)
947 /* Fail a mount where the logbuf is smaller than the log stripe */
948 if (xfs_has_logv2(mp)) {
949 if (mp->m_logbsize <= 0 &&
950 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
951 mp->m_logbsize = mp->m_sb.sb_logsunit;
952 } else if (mp->m_logbsize > 0 &&
953 mp->m_logbsize < mp->m_sb.sb_logsunit) {
955 "logbuf size must be greater than or equal to log stripe size");
959 /* Fail a mount if the logbuf is larger than 32K */
960 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
962 "logbuf size for version 1 logs must be 16K or 32K");
968 * V5 filesystems always use attr2 format for attributes.
970 if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
971 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
972 "attr2 is always enabled for V5 filesystems.");
977 * prohibit r/w mounts of read-only filesystems
979 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
981 "cannot mount a read-only filesystem as read-write");
985 if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
986 (mp->m_qflags & XFS_PQUOTA_ACCT) &&
987 !xfs_has_pquotino(mp)) {
989 "Super block does not support project and group quota together");
997 xfs_init_percpu_counters(
998 struct xfs_mount *mp)
1002 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1006 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1010 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1014 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1021 percpu_counter_destroy(&mp->m_fdblocks);
1023 percpu_counter_destroy(&mp->m_ifree);
1025 percpu_counter_destroy(&mp->m_icount);
1030 xfs_reinit_percpu_counters(
1031 struct xfs_mount *mp)
1033 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1034 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1035 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1039 xfs_destroy_percpu_counters(
1040 struct xfs_mount *mp)
1042 percpu_counter_destroy(&mp->m_icount);
1043 percpu_counter_destroy(&mp->m_ifree);
1044 percpu_counter_destroy(&mp->m_fdblocks);
1045 ASSERT(xfs_is_shutdown(mp) ||
1046 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1047 percpu_counter_destroy(&mp->m_delalloc_blks);
1051 xfs_inodegc_init_percpu(
1052 struct xfs_mount *mp)
1054 struct xfs_inodegc *gc;
1057 mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1061 for_each_possible_cpu(cpu) {
1062 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1063 init_llist_head(&gc->list);
1065 INIT_WORK(&gc->work, xfs_inodegc_worker);
1071 xfs_inodegc_free_percpu(
1072 struct xfs_mount *mp)
1076 free_percpu(mp->m_inodegc);
1081 struct super_block *sb)
1083 struct xfs_mount *mp = XFS_M(sb);
1085 /* if ->fill_super failed, we have no mount to tear down */
1089 xfs_notice(mp, "Unmounting Filesystem");
1090 xfs_filestream_unmount(mp);
1094 free_percpu(mp->m_stats.xs_stats);
1095 xfs_mount_list_del(mp);
1096 xfs_inodegc_free_percpu(mp);
1097 xfs_destroy_percpu_counters(mp);
1098 xfs_destroy_mount_workqueues(mp);
1099 xfs_close_devices(mp);
1101 sb->s_fs_info = NULL;
1106 xfs_fs_nr_cached_objects(
1107 struct super_block *sb,
1108 struct shrink_control *sc)
1110 /* Paranoia: catch incorrect calls during mount setup or teardown */
1111 if (WARN_ON_ONCE(!sb->s_fs_info))
1113 return xfs_reclaim_inodes_count(XFS_M(sb));
1117 xfs_fs_free_cached_objects(
1118 struct super_block *sb,
1119 struct shrink_control *sc)
1121 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1124 static const struct super_operations xfs_super_operations = {
1125 .alloc_inode = xfs_fs_alloc_inode,
1126 .destroy_inode = xfs_fs_destroy_inode,
1127 .dirty_inode = xfs_fs_dirty_inode,
1128 .drop_inode = xfs_fs_drop_inode,
1129 .put_super = xfs_fs_put_super,
1130 .sync_fs = xfs_fs_sync_fs,
1131 .freeze_fs = xfs_fs_freeze,
1132 .unfreeze_fs = xfs_fs_unfreeze,
1133 .statfs = xfs_fs_statfs,
1134 .show_options = xfs_fs_show_options,
1135 .nr_cached_objects = xfs_fs_nr_cached_objects,
1136 .free_cached_objects = xfs_fs_free_cached_objects,
1145 int last, shift_left_factor = 0, _res;
1149 value = kstrdup(s, GFP_KERNEL);
1153 last = strlen(value) - 1;
1154 if (value[last] == 'K' || value[last] == 'k') {
1155 shift_left_factor = 10;
1158 if (value[last] == 'M' || value[last] == 'm') {
1159 shift_left_factor = 20;
1162 if (value[last] == 'G' || value[last] == 'g') {
1163 shift_left_factor = 30;
1167 if (kstrtoint(value, base, &_res))
1170 *res = _res << shift_left_factor;
1175 xfs_fs_warn_deprecated(
1176 struct fs_context *fc,
1177 struct fs_parameter *param,
1181 /* Don't print the warning if reconfiguring and current mount point
1182 * already had the flag set
1184 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1185 !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1187 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1191 * Set mount state from a mount option.
1193 * NOTE: mp->m_super is NULL here!
1197 struct fs_context *fc,
1198 struct fs_parameter *param)
1200 struct xfs_mount *parsing_mp = fc->s_fs_info;
1201 struct fs_parse_result result;
1205 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1211 parsing_mp->m_logbufs = result.uint_32;
1214 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1218 kfree(parsing_mp->m_logname);
1219 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1220 if (!parsing_mp->m_logname)
1224 kfree(parsing_mp->m_rtname);
1225 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1226 if (!parsing_mp->m_rtname)
1230 if (suffix_kstrtoint(param->string, 10, &size))
1232 parsing_mp->m_allocsize_log = ffs(size) - 1;
1233 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1237 parsing_mp->m_features |= XFS_FEAT_GRPID;
1240 case Opt_sysvgroups:
1241 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1244 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1246 case Opt_norecovery:
1247 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1250 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1253 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1256 parsing_mp->m_dalign = result.uint_32;
1259 parsing_mp->m_swidth = result.uint_32;
1262 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1265 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1268 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1271 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1274 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1276 case Opt_filestreams:
1277 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1280 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1281 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1286 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1288 case Opt_qnoenforce:
1289 case Opt_uqnoenforce:
1290 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1291 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1295 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1297 case Opt_pqnoenforce:
1298 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1299 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1303 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1305 case Opt_gqnoenforce:
1306 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1307 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1310 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1313 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1315 #ifdef CONFIG_FS_DAX
1317 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1320 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1323 /* Following mount options will be removed in September 2025 */
1325 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1326 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1329 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1330 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1333 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1334 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1337 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1338 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1341 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1349 xfs_fs_validate_params(
1350 struct xfs_mount *mp)
1352 /* No recovery flag requires a read-only mount */
1353 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1354 xfs_warn(mp, "no-recovery mounts must be read-only.");
1359 * We have not read the superblock at this point, so only the attr2
1360 * mount option can set the attr2 feature by this stage.
1362 if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1363 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1368 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1370 "sunit and swidth options incompatible with the noalign option");
1374 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1375 xfs_warn(mp, "quota support not available in this kernel.");
1379 if ((mp->m_dalign && !mp->m_swidth) ||
1380 (!mp->m_dalign && mp->m_swidth)) {
1381 xfs_warn(mp, "sunit and swidth must be specified together");
1385 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1387 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1388 mp->m_swidth, mp->m_dalign);
1392 if (mp->m_logbufs != -1 &&
1393 mp->m_logbufs != 0 &&
1394 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1395 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1396 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1397 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1401 if (mp->m_logbsize != -1 &&
1402 mp->m_logbsize != 0 &&
1403 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1404 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1405 !is_power_of_2(mp->m_logbsize))) {
1407 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1412 if (xfs_has_allocsize(mp) &&
1413 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1414 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1415 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1416 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1425 struct super_block *sb,
1426 struct fs_context *fc)
1428 struct xfs_mount *mp = sb->s_fs_info;
1430 int flags = 0, error;
1434 error = xfs_fs_validate_params(mp);
1436 goto out_free_names;
1438 sb_min_blocksize(sb, BBSIZE);
1439 sb->s_xattr = xfs_xattr_handlers;
1440 sb->s_export_op = &xfs_export_operations;
1441 #ifdef CONFIG_XFS_QUOTA
1442 sb->s_qcop = &xfs_quotactl_operations;
1443 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1445 sb->s_op = &xfs_super_operations;
1448 * Delay mount work if the debug hook is set. This is debug
1449 * instrumention to coordinate simulation of xfs mount failures with
1450 * VFS superblock operations
1452 if (xfs_globals.mount_delay) {
1453 xfs_notice(mp, "Delaying mount for %d seconds.",
1454 xfs_globals.mount_delay);
1455 msleep(xfs_globals.mount_delay * 1000);
1458 if (fc->sb_flags & SB_SILENT)
1459 flags |= XFS_MFSI_QUIET;
1461 error = xfs_open_devices(mp);
1463 goto out_free_names;
1465 error = xfs_init_mount_workqueues(mp);
1467 goto out_close_devices;
1469 error = xfs_init_percpu_counters(mp);
1471 goto out_destroy_workqueues;
1473 error = xfs_inodegc_init_percpu(mp);
1475 goto out_destroy_counters;
1478 * All percpu data structures requiring cleanup when a cpu goes offline
1479 * must be allocated before adding this @mp to the cpu-dead handler's
1482 xfs_mount_list_add(mp);
1484 /* Allocate stats memory before we do operations that might use it */
1485 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1486 if (!mp->m_stats.xs_stats) {
1488 goto out_destroy_inodegc;
1491 error = xfs_readsb(mp, flags);
1493 goto out_free_stats;
1495 error = xfs_finish_flags(mp);
1499 error = xfs_setup_devices(mp);
1503 /* V4 support is undergoing deprecation. */
1504 if (!xfs_has_crc(mp)) {
1505 #ifdef CONFIG_XFS_SUPPORT_V4
1507 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1510 "Deprecated V4 format (crc=0) not supported by kernel.");
1516 /* Filesystem claims it needs repair, so refuse the mount. */
1517 if (xfs_has_needsrepair(mp)) {
1518 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1519 error = -EFSCORRUPTED;
1524 * Don't touch the filesystem if a user tool thinks it owns the primary
1525 * superblock. mkfs doesn't clear the flag from secondary supers, so
1526 * we don't check them at all.
1528 if (mp->m_sb.sb_inprogress) {
1529 xfs_warn(mp, "Offline file system operation in progress!");
1530 error = -EFSCORRUPTED;
1535 * Until this is fixed only page-sized or smaller data blocks work.
1537 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1539 "File system with blocksize %d bytes. "
1540 "Only pagesize (%ld) or less will currently work.",
1541 mp->m_sb.sb_blocksize, PAGE_SIZE);
1546 /* Ensure this filesystem fits in the page cache limits */
1547 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1548 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1550 "file system too large to be mounted on this system.");
1556 * XFS block mappings use 54 bits to store the logical block offset.
1557 * This should suffice to handle the maximum file size that the VFS
1558 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1559 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1560 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1561 * to check this assertion.
1563 * Avoid integer overflow by comparing the maximum bmbt offset to the
1564 * maximum pagecache offset in units of fs blocks.
1566 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1568 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1569 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1575 error = xfs_filestream_mount(mp);
1580 * we must configure the block size in the superblock before we run the
1581 * full mount process as the mount process can lookup and cache inodes.
1583 sb->s_magic = XFS_SUPER_MAGIC;
1584 sb->s_blocksize = mp->m_sb.sb_blocksize;
1585 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1586 sb->s_maxbytes = MAX_LFS_FILESIZE;
1587 sb->s_max_links = XFS_MAXLINK;
1588 sb->s_time_gran = 1;
1589 if (xfs_has_bigtime(mp)) {
1590 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1591 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1593 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1594 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1596 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1597 sb->s_iflags |= SB_I_CGROUPWB;
1599 set_posix_acl_flag(sb);
1601 /* version 5 superblocks support inode version counters. */
1602 if (xfs_has_crc(mp))
1603 sb->s_flags |= SB_I_VERSION;
1605 if (xfs_has_dax_always(mp)) {
1606 error = xfs_setup_dax_always(mp);
1608 goto out_filestream_unmount;
1611 if (xfs_has_discard(mp)) {
1612 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1614 if (!blk_queue_discard(q)) {
1615 xfs_warn(mp, "mounting with \"discard\" option, but "
1616 "the device does not support discard");
1617 mp->m_features &= ~XFS_FEAT_DISCARD;
1621 if (xfs_has_reflink(mp)) {
1622 if (mp->m_sb.sb_rblocks) {
1624 "reflink not compatible with realtime device!");
1626 goto out_filestream_unmount;
1629 if (xfs_globals.always_cow) {
1630 xfs_info(mp, "using DEBUG-only always_cow mode.");
1631 mp->m_always_cow = true;
1635 if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1637 "reverse mapping btree not compatible with realtime device!");
1639 goto out_filestream_unmount;
1642 error = xfs_mountfs(mp);
1644 goto out_filestream_unmount;
1646 root = igrab(VFS_I(mp->m_rootip));
1651 sb->s_root = d_make_root(root);
1659 out_filestream_unmount:
1660 xfs_filestream_unmount(mp);
1664 free_percpu(mp->m_stats.xs_stats);
1665 out_destroy_inodegc:
1666 xfs_mount_list_del(mp);
1667 xfs_inodegc_free_percpu(mp);
1668 out_destroy_counters:
1669 xfs_destroy_percpu_counters(mp);
1670 out_destroy_workqueues:
1671 xfs_destroy_mount_workqueues(mp);
1673 xfs_close_devices(mp);
1675 sb->s_fs_info = NULL;
1680 xfs_filestream_unmount(mp);
1687 struct fs_context *fc)
1689 return get_tree_bdev(fc, xfs_fs_fill_super);
1694 struct xfs_mount *mp)
1696 struct xfs_sb *sbp = &mp->m_sb;
1699 if (xfs_has_norecovery(mp)) {
1701 "ro->rw transition prohibited on norecovery mount");
1705 if (xfs_sb_is_v5(sbp) &&
1706 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1708 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1709 (sbp->sb_features_ro_compat &
1710 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1714 clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1717 * If this is the first remount to writeable state we might have some
1718 * superblock changes to update.
1720 if (mp->m_update_sb) {
1721 error = xfs_sync_sb(mp, false);
1723 xfs_warn(mp, "failed to write sb changes");
1726 mp->m_update_sb = false;
1730 * Fill out the reserve pool if it is empty. Use the stashed value if
1731 * it is non-zero, otherwise go with the default.
1733 xfs_restore_resvblks(mp);
1734 xfs_log_work_queue(mp);
1735 xfs_blockgc_start(mp);
1737 /* Create the per-AG metadata reservation pool .*/
1738 error = xfs_fs_reserve_ag_blocks(mp);
1739 if (error && error != -ENOSPC)
1742 /* Re-enable the background inode inactivation worker. */
1743 xfs_inodegc_start(mp);
1750 struct xfs_mount *mp)
1752 struct xfs_icwalk icw = {
1753 .icw_flags = XFS_ICWALK_FLAG_SYNC,
1757 /* Flush all the dirty data to disk. */
1758 error = sync_filesystem(mp->m_super);
1763 * Cancel background eofb scanning so it cannot race with the final
1764 * log force+buftarg wait and deadlock the remount.
1766 xfs_blockgc_stop(mp);
1769 * Clear out all remaining COW staging extents and speculative post-EOF
1770 * preallocations so that we don't leave inodes requiring inactivation
1771 * cleanups during reclaim on a read-only mount. We must process every
1772 * cached inode, so this requires a synchronous cache scan.
1774 error = xfs_blockgc_free_space(mp, &icw);
1776 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1781 * Stop the inodegc background worker. xfs_fs_reconfigure already
1782 * flushed all pending inodegc work when it sync'd the filesystem.
1783 * The VFS holds s_umount, so we know that inodes cannot enter
1784 * xfs_fs_destroy_inode during a remount operation. In readonly mode
1785 * we send inodes straight to reclaim, so no inodes will be queued.
1787 xfs_inodegc_stop(mp);
1789 /* Free the per-AG metadata reservation pool. */
1790 error = xfs_fs_unreserve_ag_blocks(mp);
1792 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1797 * Before we sync the metadata, we need to free up the reserve block
1798 * pool so that the used block count in the superblock on disk is
1799 * correct at the end of the remount. Stash the current* reserve pool
1800 * size so that if we get remounted rw, we can return it to the same
1803 xfs_save_resvblks(mp);
1806 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1812 * Logically we would return an error here to prevent users from believing
1813 * they might have changed mount options using remount which can't be changed.
1815 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1816 * arguments in some cases so we can't blindly reject options, but have to
1817 * check for each specified option if it actually differs from the currently
1818 * set option and only reject it if that's the case.
1820 * Until that is implemented we return success for every remount request, and
1821 * silently ignore all options that we can't actually change.
1825 struct fs_context *fc)
1827 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1828 struct xfs_mount *new_mp = fc->s_fs_info;
1829 int flags = fc->sb_flags;
1832 /* version 5 superblocks always support version counters. */
1833 if (xfs_has_crc(mp))
1834 fc->sb_flags |= SB_I_VERSION;
1836 error = xfs_fs_validate_params(new_mp);
1840 /* inode32 -> inode64 */
1841 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1842 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1843 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1846 /* inode64 -> inode32 */
1847 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1848 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1849 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1853 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1854 error = xfs_remount_rw(mp);
1860 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1861 error = xfs_remount_ro(mp);
1869 static void xfs_fs_free(
1870 struct fs_context *fc)
1872 struct xfs_mount *mp = fc->s_fs_info;
1875 * mp is stored in the fs_context when it is initialized.
1876 * mp is transferred to the superblock on a successful mount,
1877 * but if an error occurs before the transfer we have to free
1884 static const struct fs_context_operations xfs_context_ops = {
1885 .parse_param = xfs_fs_parse_param,
1886 .get_tree = xfs_fs_get_tree,
1887 .reconfigure = xfs_fs_reconfigure,
1888 .free = xfs_fs_free,
1891 static int xfs_init_fs_context(
1892 struct fs_context *fc)
1894 struct xfs_mount *mp;
1896 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1900 spin_lock_init(&mp->m_sb_lock);
1901 spin_lock_init(&mp->m_agirotor_lock);
1902 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1903 spin_lock_init(&mp->m_perag_lock);
1904 mutex_init(&mp->m_growlock);
1905 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1906 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1907 mp->m_kobj.kobject.kset = xfs_kset;
1909 * We don't create the finobt per-ag space reservation until after log
1910 * recovery, so we must set this to true so that an ifree transaction
1911 * started during log recovery will not depend on space reservations
1912 * for finobt expansion.
1914 mp->m_finobt_nores = true;
1917 * These can be overridden by the mount option parsing.
1920 mp->m_logbsize = -1;
1921 mp->m_allocsize_log = 16; /* 64k */
1924 * Copy binary VFS mount flags we are interested in.
1926 if (fc->sb_flags & SB_RDONLY)
1927 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1928 if (fc->sb_flags & SB_DIRSYNC)
1929 mp->m_features |= XFS_FEAT_DIRSYNC;
1930 if (fc->sb_flags & SB_SYNCHRONOUS)
1931 mp->m_features |= XFS_FEAT_WSYNC;
1934 fc->ops = &xfs_context_ops;
1939 static struct file_system_type xfs_fs_type = {
1940 .owner = THIS_MODULE,
1942 .init_fs_context = xfs_init_fs_context,
1943 .parameters = xfs_fs_parameters,
1944 .kill_sb = kill_block_super,
1945 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1947 MODULE_ALIAS_FS("xfs");
1950 xfs_init_caches(void)
1954 xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1955 sizeof(struct xlog_ticket),
1957 if (!xfs_log_ticket_cache)
1960 error = xfs_btree_init_cur_caches();
1962 goto out_destroy_log_ticket_cache;
1964 error = xfs_defer_init_item_caches();
1966 goto out_destroy_btree_cur_cache;
1968 xfs_da_state_cache = kmem_cache_create("xfs_da_state",
1969 sizeof(struct xfs_da_state),
1971 if (!xfs_da_state_cache)
1972 goto out_destroy_defer_item_cache;
1974 xfs_ifork_cache = kmem_cache_create("xfs_ifork",
1975 sizeof(struct xfs_ifork),
1977 if (!xfs_ifork_cache)
1978 goto out_destroy_da_state_cache;
1980 xfs_trans_cache = kmem_cache_create("xfs_trans",
1981 sizeof(struct xfs_trans),
1983 if (!xfs_trans_cache)
1984 goto out_destroy_ifork_cache;
1988 * The size of the cache-allocated buf log item is the maximum
1989 * size possible under XFS. This wastes a little bit of memory,
1990 * but it is much faster.
1992 xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
1993 sizeof(struct xfs_buf_log_item),
1995 if (!xfs_buf_item_cache)
1996 goto out_destroy_trans_cache;
1998 xfs_efd_cache = kmem_cache_create("xfs_efd_item",
1999 (sizeof(struct xfs_efd_log_item) +
2000 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2001 sizeof(struct xfs_extent)),
2004 goto out_destroy_buf_item_cache;
2006 xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2007 (sizeof(struct xfs_efi_log_item) +
2008 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2009 sizeof(struct xfs_extent)),
2012 goto out_destroy_efd_cache;
2014 xfs_inode_cache = kmem_cache_create("xfs_inode",
2015 sizeof(struct xfs_inode), 0,
2016 (SLAB_HWCACHE_ALIGN |
2017 SLAB_RECLAIM_ACCOUNT |
2018 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2019 xfs_fs_inode_init_once);
2020 if (!xfs_inode_cache)
2021 goto out_destroy_efi_cache;
2023 xfs_ili_cache = kmem_cache_create("xfs_ili",
2024 sizeof(struct xfs_inode_log_item), 0,
2025 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2028 goto out_destroy_inode_cache;
2030 xfs_icreate_cache = kmem_cache_create("xfs_icr",
2031 sizeof(struct xfs_icreate_item),
2033 if (!xfs_icreate_cache)
2034 goto out_destroy_ili_cache;
2036 xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2037 sizeof(struct xfs_rud_log_item),
2040 goto out_destroy_icreate_cache;
2042 xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2043 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2046 goto out_destroy_rud_cache;
2048 xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2049 sizeof(struct xfs_cud_log_item),
2052 goto out_destroy_rui_cache;
2054 xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2055 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2058 goto out_destroy_cud_cache;
2060 xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2061 sizeof(struct xfs_bud_log_item),
2064 goto out_destroy_cui_cache;
2066 xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2067 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2070 goto out_destroy_bud_cache;
2074 out_destroy_bud_cache:
2075 kmem_cache_destroy(xfs_bud_cache);
2076 out_destroy_cui_cache:
2077 kmem_cache_destroy(xfs_cui_cache);
2078 out_destroy_cud_cache:
2079 kmem_cache_destroy(xfs_cud_cache);
2080 out_destroy_rui_cache:
2081 kmem_cache_destroy(xfs_rui_cache);
2082 out_destroy_rud_cache:
2083 kmem_cache_destroy(xfs_rud_cache);
2084 out_destroy_icreate_cache:
2085 kmem_cache_destroy(xfs_icreate_cache);
2086 out_destroy_ili_cache:
2087 kmem_cache_destroy(xfs_ili_cache);
2088 out_destroy_inode_cache:
2089 kmem_cache_destroy(xfs_inode_cache);
2090 out_destroy_efi_cache:
2091 kmem_cache_destroy(xfs_efi_cache);
2092 out_destroy_efd_cache:
2093 kmem_cache_destroy(xfs_efd_cache);
2094 out_destroy_buf_item_cache:
2095 kmem_cache_destroy(xfs_buf_item_cache);
2096 out_destroy_trans_cache:
2097 kmem_cache_destroy(xfs_trans_cache);
2098 out_destroy_ifork_cache:
2099 kmem_cache_destroy(xfs_ifork_cache);
2100 out_destroy_da_state_cache:
2101 kmem_cache_destroy(xfs_da_state_cache);
2102 out_destroy_defer_item_cache:
2103 xfs_defer_destroy_item_caches();
2104 out_destroy_btree_cur_cache:
2105 xfs_btree_destroy_cur_caches();
2106 out_destroy_log_ticket_cache:
2107 kmem_cache_destroy(xfs_log_ticket_cache);
2113 xfs_destroy_caches(void)
2116 * Make sure all delayed rcu free are flushed before we
2120 kmem_cache_destroy(xfs_bui_cache);
2121 kmem_cache_destroy(xfs_bud_cache);
2122 kmem_cache_destroy(xfs_cui_cache);
2123 kmem_cache_destroy(xfs_cud_cache);
2124 kmem_cache_destroy(xfs_rui_cache);
2125 kmem_cache_destroy(xfs_rud_cache);
2126 kmem_cache_destroy(xfs_icreate_cache);
2127 kmem_cache_destroy(xfs_ili_cache);
2128 kmem_cache_destroy(xfs_inode_cache);
2129 kmem_cache_destroy(xfs_efi_cache);
2130 kmem_cache_destroy(xfs_efd_cache);
2131 kmem_cache_destroy(xfs_buf_item_cache);
2132 kmem_cache_destroy(xfs_trans_cache);
2133 kmem_cache_destroy(xfs_ifork_cache);
2134 kmem_cache_destroy(xfs_da_state_cache);
2135 xfs_defer_destroy_item_caches();
2136 xfs_btree_destroy_cur_caches();
2137 kmem_cache_destroy(xfs_log_ticket_cache);
2141 xfs_init_workqueues(void)
2144 * The allocation workqueue can be used in memory reclaim situations
2145 * (writepage path), and parallelism is only limited by the number of
2146 * AGs in all the filesystems mounted. Hence use the default large
2147 * max_active value for this workqueue.
2149 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2150 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2154 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2156 if (!xfs_discard_wq)
2157 goto out_free_alloc_wq;
2161 destroy_workqueue(xfs_alloc_wq);
2166 xfs_destroy_workqueues(void)
2168 destroy_workqueue(xfs_discard_wq);
2169 destroy_workqueue(xfs_alloc_wq);
2172 #ifdef CONFIG_HOTPLUG_CPU
2177 struct xfs_mount *mp, *n;
2179 spin_lock(&xfs_mount_list_lock);
2180 list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2181 spin_unlock(&xfs_mount_list_lock);
2182 xfs_inodegc_cpu_dead(mp, cpu);
2183 spin_lock(&xfs_mount_list_lock);
2185 spin_unlock(&xfs_mount_list_lock);
2190 xfs_cpu_hotplug_init(void)
2194 error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2198 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2204 xfs_cpu_hotplug_destroy(void)
2206 cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2209 #else /* !CONFIG_HOTPLUG_CPU */
2210 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2211 static inline void xfs_cpu_hotplug_destroy(void) {}
2219 xfs_check_ondisk_structs();
2221 printk(KERN_INFO XFS_VERSION_STRING " with "
2222 XFS_BUILD_OPTIONS " enabled\n");
2226 error = xfs_cpu_hotplug_init();
2230 error = xfs_init_caches();
2232 goto out_destroy_hp;
2234 error = xfs_init_workqueues();
2236 goto out_destroy_caches;
2238 error = xfs_mru_cache_init();
2240 goto out_destroy_wq;
2242 error = xfs_buf_init();
2244 goto out_mru_cache_uninit;
2246 error = xfs_init_procfs();
2248 goto out_buf_terminate;
2250 error = xfs_sysctl_register();
2252 goto out_cleanup_procfs;
2254 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2257 goto out_sysctl_unregister;
2260 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2262 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2263 if (!xfsstats.xs_stats) {
2265 goto out_kset_unregister;
2268 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2271 goto out_free_stats;
2274 xfs_dbg_kobj.kobject.kset = xfs_kset;
2275 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2277 goto out_remove_stats_kobj;
2280 error = xfs_qm_init();
2282 goto out_remove_dbg_kobj;
2284 error = register_filesystem(&xfs_fs_type);
2291 out_remove_dbg_kobj:
2293 xfs_sysfs_del(&xfs_dbg_kobj);
2294 out_remove_stats_kobj:
2296 xfs_sysfs_del(&xfsstats.xs_kobj);
2298 free_percpu(xfsstats.xs_stats);
2299 out_kset_unregister:
2300 kset_unregister(xfs_kset);
2301 out_sysctl_unregister:
2302 xfs_sysctl_unregister();
2304 xfs_cleanup_procfs();
2306 xfs_buf_terminate();
2307 out_mru_cache_uninit:
2308 xfs_mru_cache_uninit();
2310 xfs_destroy_workqueues();
2312 xfs_destroy_caches();
2314 xfs_cpu_hotplug_destroy();
2323 unregister_filesystem(&xfs_fs_type);
2325 xfs_sysfs_del(&xfs_dbg_kobj);
2327 xfs_sysfs_del(&xfsstats.xs_kobj);
2328 free_percpu(xfsstats.xs_stats);
2329 kset_unregister(xfs_kset);
2330 xfs_sysctl_unregister();
2331 xfs_cleanup_procfs();
2332 xfs_buf_terminate();
2333 xfs_mru_cache_uninit();
2334 xfs_destroy_workqueues();
2335 xfs_destroy_caches();
2336 xfs_uuid_table_free();
2337 xfs_cpu_hotplug_destroy();
2340 module_init(init_xfs_fs);
2341 module_exit(exit_xfs_fs);
2343 MODULE_AUTHOR("Silicon Graphics, Inc.");
2344 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2345 MODULE_LICENSE("GPL");