1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
17 #include "xfs_alloc.h"
18 #include "xfs_fsops.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
22 #include "xfs_log_priv.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_mru_cache.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_icache.h"
28 #include "xfs_trace.h"
29 #include "xfs_icreate_item.h"
30 #include "xfs_filestream.h"
31 #include "xfs_quota.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_ondisk.h"
34 #include "xfs_rmap_item.h"
35 #include "xfs_refcount_item.h"
36 #include "xfs_bmap_item.h"
37 #include "xfs_reflink.h"
38 #include "xfs_pwork.h"
41 #include <linux/magic.h>
42 #include <linux/fs_context.h>
43 #include <linux/fs_parser.h>
45 static const struct super_operations xfs_super_operations;
47 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
49 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
52 #ifdef CONFIG_HOTPLUG_CPU
53 static LIST_HEAD(xfs_mount_list);
54 static DEFINE_SPINLOCK(xfs_mount_list_lock);
56 static inline void xfs_mount_list_add(struct xfs_mount *mp)
58 spin_lock(&xfs_mount_list_lock);
59 list_add(&mp->m_mount_list, &xfs_mount_list);
60 spin_unlock(&xfs_mount_list_lock);
63 static inline void xfs_mount_list_del(struct xfs_mount *mp)
65 spin_lock(&xfs_mount_list_lock);
66 list_del(&mp->m_mount_list);
67 spin_unlock(&xfs_mount_list_lock);
69 #else /* !CONFIG_HOTPLUG_CPU */
70 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
71 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
81 xfs_mount_set_dax_mode(
83 enum xfs_dax_mode mode)
87 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
90 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
91 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
94 mp->m_features |= XFS_FEAT_DAX_NEVER;
95 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
100 static const struct constant_table dax_param_enums[] = {
101 {"inode", XFS_DAX_INODE },
102 {"always", XFS_DAX_ALWAYS },
103 {"never", XFS_DAX_NEVER },
108 * Table driven mount option parser.
111 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
112 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
113 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
114 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
115 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
116 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
117 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
118 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
119 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
122 static const struct fs_parameter_spec xfs_fs_parameters[] = {
123 fsparam_u32("logbufs", Opt_logbufs),
124 fsparam_string("logbsize", Opt_logbsize),
125 fsparam_string("logdev", Opt_logdev),
126 fsparam_string("rtdev", Opt_rtdev),
127 fsparam_flag("wsync", Opt_wsync),
128 fsparam_flag("noalign", Opt_noalign),
129 fsparam_flag("swalloc", Opt_swalloc),
130 fsparam_u32("sunit", Opt_sunit),
131 fsparam_u32("swidth", Opt_swidth),
132 fsparam_flag("nouuid", Opt_nouuid),
133 fsparam_flag("grpid", Opt_grpid),
134 fsparam_flag("nogrpid", Opt_nogrpid),
135 fsparam_flag("bsdgroups", Opt_bsdgroups),
136 fsparam_flag("sysvgroups", Opt_sysvgroups),
137 fsparam_string("allocsize", Opt_allocsize),
138 fsparam_flag("norecovery", Opt_norecovery),
139 fsparam_flag("inode64", Opt_inode64),
140 fsparam_flag("inode32", Opt_inode32),
141 fsparam_flag("ikeep", Opt_ikeep),
142 fsparam_flag("noikeep", Opt_noikeep),
143 fsparam_flag("largeio", Opt_largeio),
144 fsparam_flag("nolargeio", Opt_nolargeio),
145 fsparam_flag("attr2", Opt_attr2),
146 fsparam_flag("noattr2", Opt_noattr2),
147 fsparam_flag("filestreams", Opt_filestreams),
148 fsparam_flag("quota", Opt_quota),
149 fsparam_flag("noquota", Opt_noquota),
150 fsparam_flag("usrquota", Opt_usrquota),
151 fsparam_flag("grpquota", Opt_grpquota),
152 fsparam_flag("prjquota", Opt_prjquota),
153 fsparam_flag("uquota", Opt_uquota),
154 fsparam_flag("gquota", Opt_gquota),
155 fsparam_flag("pquota", Opt_pquota),
156 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
157 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
158 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
159 fsparam_flag("qnoenforce", Opt_qnoenforce),
160 fsparam_flag("discard", Opt_discard),
161 fsparam_flag("nodiscard", Opt_nodiscard),
162 fsparam_flag("dax", Opt_dax),
163 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
167 struct proc_xfs_info {
177 static struct proc_xfs_info xfs_info_set[] = {
178 /* the few simple ones we can get from the mount struct */
179 { XFS_FEAT_IKEEP, ",ikeep" },
180 { XFS_FEAT_WSYNC, ",wsync" },
181 { XFS_FEAT_NOALIGN, ",noalign" },
182 { XFS_FEAT_SWALLOC, ",swalloc" },
183 { XFS_FEAT_NOUUID, ",nouuid" },
184 { XFS_FEAT_NORECOVERY, ",norecovery" },
185 { XFS_FEAT_ATTR2, ",attr2" },
186 { XFS_FEAT_FILESTREAMS, ",filestreams" },
187 { XFS_FEAT_GRPID, ",grpid" },
188 { XFS_FEAT_DISCARD, ",discard" },
189 { XFS_FEAT_LARGE_IOSIZE, ",largeio" },
190 { XFS_FEAT_DAX_ALWAYS, ",dax=always" },
191 { XFS_FEAT_DAX_NEVER, ",dax=never" },
194 struct xfs_mount *mp = XFS_M(root->d_sb);
195 struct proc_xfs_info *xfs_infop;
197 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
198 if (mp->m_features & xfs_infop->flag)
199 seq_puts(m, xfs_infop->str);
202 seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
204 if (xfs_has_allocsize(mp))
205 seq_printf(m, ",allocsize=%dk",
206 (1 << mp->m_allocsize_log) >> 10);
208 if (mp->m_logbufs > 0)
209 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
210 if (mp->m_logbsize > 0)
211 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
214 seq_show_option(m, "logdev", mp->m_logname);
216 seq_show_option(m, "rtdev", mp->m_rtname);
218 if (mp->m_dalign > 0)
219 seq_printf(m, ",sunit=%d",
220 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
221 if (mp->m_swidth > 0)
222 seq_printf(m, ",swidth=%d",
223 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
225 if (mp->m_qflags & XFS_UQUOTA_ENFD)
226 seq_puts(m, ",usrquota");
227 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
228 seq_puts(m, ",uqnoenforce");
230 if (mp->m_qflags & XFS_PQUOTA_ENFD)
231 seq_puts(m, ",prjquota");
232 else if (mp->m_qflags & XFS_PQUOTA_ACCT)
233 seq_puts(m, ",pqnoenforce");
235 if (mp->m_qflags & XFS_GQUOTA_ENFD)
236 seq_puts(m, ",grpquota");
237 else if (mp->m_qflags & XFS_GQUOTA_ACCT)
238 seq_puts(m, ",gqnoenforce");
240 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
241 seq_puts(m, ",noquota");
247 * Set parameters for inode allocation heuristics, taking into account
248 * filesystem size and inode32/inode64 mount options; i.e. specifically
249 * whether or not XFS_FEAT_SMALL_INUMS is set.
251 * Inode allocation patterns are altered only if inode32 is requested
252 * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
253 * If altered, XFS_OPSTATE_INODE32 is set as well.
255 * An agcount independent of that in the mount structure is provided
256 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
257 * to the potentially higher ag count.
259 * Returns the maximum AG index which may contain inodes.
263 struct xfs_mount *mp,
264 xfs_agnumber_t agcount)
266 xfs_agnumber_t index;
267 xfs_agnumber_t maxagi = 0;
268 xfs_sb_t *sbp = &mp->m_sb;
269 xfs_agnumber_t max_metadata;
274 * Calculate how much should be reserved for inodes to meet
275 * the max inode percentage. Used only for inode32.
277 if (M_IGEO(mp)->maxicount) {
280 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
282 icount += sbp->sb_agblocks - 1;
283 do_div(icount, sbp->sb_agblocks);
284 max_metadata = icount;
286 max_metadata = agcount;
289 /* Get the last possible inode in the filesystem */
290 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
291 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
294 * If user asked for no more than 32-bit inodes, and the fs is
295 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
296 * the allocator to accommodate the request.
298 if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
299 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
301 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
303 for (index = 0; index < agcount; index++) {
304 struct xfs_perag *pag;
306 ino = XFS_AGINO_TO_INO(mp, index, agino);
308 pag = xfs_perag_get(mp, index);
310 if (xfs_is_inode32(mp)) {
311 if (ino > XFS_MAXINUMBER_32) {
312 pag->pagi_inodeok = 0;
313 pag->pagf_metadata = 0;
315 pag->pagi_inodeok = 1;
317 if (index < max_metadata)
318 pag->pagf_metadata = 1;
320 pag->pagf_metadata = 0;
323 pag->pagi_inodeok = 1;
324 pag->pagf_metadata = 0;
330 return xfs_is_inode32(mp) ? maxagi : agcount;
337 struct block_device **bdevp)
341 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
343 if (IS_ERR(*bdevp)) {
344 error = PTR_ERR(*bdevp);
345 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
353 struct block_device *bdev)
356 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
361 struct xfs_mount *mp)
363 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
365 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
366 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
367 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
369 xfs_free_buftarg(mp->m_logdev_targp);
370 xfs_blkdev_put(logdev);
371 fs_put_dax(dax_logdev);
373 if (mp->m_rtdev_targp) {
374 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
375 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
377 xfs_free_buftarg(mp->m_rtdev_targp);
378 xfs_blkdev_put(rtdev);
379 fs_put_dax(dax_rtdev);
381 xfs_free_buftarg(mp->m_ddev_targp);
382 fs_put_dax(dax_ddev);
386 * The file system configurations are:
387 * (1) device (partition) with data and internal log
388 * (2) logical volume with data and log subvolumes.
389 * (3) logical volume with data, log, and realtime subvolumes.
391 * We only have to handle opening the log and realtime volumes here if
392 * they are present. The data subvolume has already been opened by
393 * get_sb_bdev() and is stored in sb->s_bdev.
397 struct xfs_mount *mp)
399 struct block_device *ddev = mp->m_super->s_bdev;
400 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
401 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
402 struct block_device *logdev = NULL, *rtdev = NULL;
406 * Open real time and log devices - order is important.
409 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
412 dax_logdev = fs_dax_get_by_bdev(logdev);
416 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
418 goto out_close_logdev;
420 if (rtdev == ddev || rtdev == logdev) {
422 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
424 goto out_close_rtdev;
426 dax_rtdev = fs_dax_get_by_bdev(rtdev);
430 * Setup xfs_mount buffer target pointers
433 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
434 if (!mp->m_ddev_targp)
435 goto out_close_rtdev;
438 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
439 if (!mp->m_rtdev_targp)
440 goto out_free_ddev_targ;
443 if (logdev && logdev != ddev) {
444 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
445 if (!mp->m_logdev_targp)
446 goto out_free_rtdev_targ;
448 mp->m_logdev_targp = mp->m_ddev_targp;
454 if (mp->m_rtdev_targp)
455 xfs_free_buftarg(mp->m_rtdev_targp);
457 xfs_free_buftarg(mp->m_ddev_targp);
459 xfs_blkdev_put(rtdev);
460 fs_put_dax(dax_rtdev);
462 if (logdev && logdev != ddev) {
463 xfs_blkdev_put(logdev);
464 fs_put_dax(dax_logdev);
467 fs_put_dax(dax_ddev);
472 * Setup xfs_mount buffer target pointers based on superblock
476 struct xfs_mount *mp)
480 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
484 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
485 unsigned int log_sector_size = BBSIZE;
487 if (xfs_has_sector(mp))
488 log_sector_size = mp->m_sb.sb_logsectsize;
489 error = xfs_setsize_buftarg(mp->m_logdev_targp,
494 if (mp->m_rtdev_targp) {
495 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
496 mp->m_sb.sb_sectsize);
505 xfs_init_mount_workqueues(
506 struct xfs_mount *mp)
508 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
509 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
510 1, mp->m_super->s_id);
511 if (!mp->m_buf_workqueue)
514 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
515 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
516 0, mp->m_super->s_id);
517 if (!mp->m_unwritten_workqueue)
518 goto out_destroy_buf;
520 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
521 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
522 0, mp->m_super->s_id);
523 if (!mp->m_reclaim_workqueue)
524 goto out_destroy_unwritten;
526 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
527 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
528 0, mp->m_super->s_id);
529 if (!mp->m_blockgc_wq)
530 goto out_destroy_reclaim;
532 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
533 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
534 1, mp->m_super->s_id);
535 if (!mp->m_inodegc_wq)
536 goto out_destroy_blockgc;
538 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
539 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
540 if (!mp->m_sync_workqueue)
541 goto out_destroy_inodegc;
546 destroy_workqueue(mp->m_inodegc_wq);
548 destroy_workqueue(mp->m_blockgc_wq);
550 destroy_workqueue(mp->m_reclaim_workqueue);
551 out_destroy_unwritten:
552 destroy_workqueue(mp->m_unwritten_workqueue);
554 destroy_workqueue(mp->m_buf_workqueue);
560 xfs_destroy_mount_workqueues(
561 struct xfs_mount *mp)
563 destroy_workqueue(mp->m_sync_workqueue);
564 destroy_workqueue(mp->m_blockgc_wq);
565 destroy_workqueue(mp->m_inodegc_wq);
566 destroy_workqueue(mp->m_reclaim_workqueue);
567 destroy_workqueue(mp->m_unwritten_workqueue);
568 destroy_workqueue(mp->m_buf_workqueue);
572 xfs_flush_inodes_worker(
573 struct work_struct *work)
575 struct xfs_mount *mp = container_of(work, struct xfs_mount,
576 m_flush_inodes_work);
577 struct super_block *sb = mp->m_super;
579 if (down_read_trylock(&sb->s_umount)) {
581 up_read(&sb->s_umount);
586 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
587 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
588 * for IO to complete so that we effectively throttle multiple callers to the
589 * rate at which IO is completing.
593 struct xfs_mount *mp)
596 * If flush_work() returns true then that means we waited for a flush
597 * which was already in progress. Don't bother running another scan.
599 if (flush_work(&mp->m_flush_inodes_work))
602 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
603 flush_work(&mp->m_flush_inodes_work);
606 /* Catch misguided souls that try to use this interface on XFS */
607 STATIC struct inode *
609 struct super_block *sb)
616 * Now that the generic code is guaranteed not to be accessing
617 * the linux inode, we can inactivate and reclaim the inode.
620 xfs_fs_destroy_inode(
623 struct xfs_inode *ip = XFS_I(inode);
625 trace_xfs_destroy_inode(ip);
627 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
628 XFS_STATS_INC(ip->i_mount, vn_rele);
629 XFS_STATS_INC(ip->i_mount, vn_remove);
630 xfs_inode_mark_reclaimable(ip);
638 struct xfs_inode *ip = XFS_I(inode);
639 struct xfs_mount *mp = ip->i_mount;
640 struct xfs_trans *tp;
642 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
644 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
647 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
649 xfs_ilock(ip, XFS_ILOCK_EXCL);
650 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
651 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
652 xfs_trans_commit(tp);
656 * Slab object creation initialisation for the XFS inode.
657 * This covers only the idempotent fields in the XFS inode;
658 * all other fields need to be initialised on allocation
659 * from the slab. This avoids the need to repeatedly initialise
660 * fields in the xfs inode that left in the initialise state
661 * when freeing the inode.
664 xfs_fs_inode_init_once(
667 struct xfs_inode *ip = inode;
669 memset(ip, 0, sizeof(struct xfs_inode));
672 inode_init_once(VFS_I(ip));
675 atomic_set(&ip->i_pincount, 0);
676 spin_lock_init(&ip->i_flags_lock);
678 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
679 "xfsino", ip->i_ino);
683 * We do an unlocked check for XFS_IDONTCACHE here because we are already
684 * serialised against cache hits here via the inode->i_lock and igrab() in
685 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
686 * racing with us, and it avoids needing to grab a spinlock here for every inode
687 * we drop the final reference on.
693 struct xfs_inode *ip = XFS_I(inode);
696 * If this unlinked inode is in the middle of recovery, don't
697 * drop the inode just yet; log recovery will take care of
698 * that. See the comment for this inode flag.
700 if (ip->i_flags & XFS_IRECOVERY) {
701 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
705 return generic_drop_inode(inode);
710 struct xfs_mount *mp)
713 kfree(mp->m_logname);
719 struct super_block *sb,
722 struct xfs_mount *mp = XFS_M(sb);
724 trace_xfs_fs_sync_fs(mp, __return_address);
727 * Doing anything during the async pass would be counterproductive.
732 xfs_log_force(mp, XFS_LOG_SYNC);
735 * The disk must be active because we're syncing.
736 * We schedule log work now (now that the disk is
737 * active) instead of later (when it might not be).
739 flush_delayed_work(&mp->m_log->l_work);
743 * If we are called with page faults frozen out, it means we are about
744 * to freeze the transaction subsystem. Take the opportunity to shut
745 * down inodegc because once SB_FREEZE_FS is set it's too late to
746 * prevent inactivation races with freeze. The fs doesn't get called
747 * again by the freezing process until after SB_FREEZE_FS has been set,
748 * so it's now or never. Same logic applies to speculative allocation
749 * garbage collection.
751 * We don't care if this is a normal syncfs call that does this or
752 * freeze that does this - we can run this multiple times without issue
753 * and we won't race with a restart because a restart can only occur
754 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
756 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
757 xfs_inodegc_stop(mp);
758 xfs_blockgc_stop(mp);
766 struct dentry *dentry,
767 struct kstatfs *statp)
769 struct xfs_mount *mp = XFS_M(dentry->d_sb);
770 xfs_sb_t *sbp = &mp->m_sb;
771 struct xfs_inode *ip = XFS_I(d_inode(dentry));
772 uint64_t fakeinos, id;
779 /* Wait for whatever inactivations are in progress. */
780 xfs_inodegc_flush(mp);
782 statp->f_type = XFS_SUPER_MAGIC;
783 statp->f_namelen = MAXNAMELEN - 1;
785 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
786 statp->f_fsid = u64_to_fsid(id);
788 icount = percpu_counter_sum(&mp->m_icount);
789 ifree = percpu_counter_sum(&mp->m_ifree);
790 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
792 spin_lock(&mp->m_sb_lock);
793 statp->f_bsize = sbp->sb_blocksize;
794 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
795 statp->f_blocks = sbp->sb_dblocks - lsize;
796 spin_unlock(&mp->m_sb_lock);
798 /* make sure statp->f_bfree does not underflow */
799 statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
800 statp->f_bavail = statp->f_bfree;
802 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
803 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
804 if (M_IGEO(mp)->maxicount)
805 statp->f_files = min_t(typeof(statp->f_files),
807 M_IGEO(mp)->maxicount);
809 /* If sb_icount overshot maxicount, report actual allocation */
810 statp->f_files = max_t(typeof(statp->f_files),
814 /* make sure statp->f_ffree does not underflow */
815 ffree = statp->f_files - (icount - ifree);
816 statp->f_ffree = max_t(int64_t, ffree, 0);
819 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
820 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
821 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
822 xfs_qm_statvfs(ip, statp);
824 if (XFS_IS_REALTIME_MOUNT(mp) &&
825 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
826 statp->f_blocks = sbp->sb_rblocks;
827 statp->f_bavail = statp->f_bfree =
828 sbp->sb_frextents * sbp->sb_rextsize;
835 xfs_save_resvblks(struct xfs_mount *mp)
837 uint64_t resblks = 0;
839 mp->m_resblks_save = mp->m_resblks;
840 xfs_reserve_blocks(mp, &resblks, NULL);
844 xfs_restore_resvblks(struct xfs_mount *mp)
848 if (mp->m_resblks_save) {
849 resblks = mp->m_resblks_save;
850 mp->m_resblks_save = 0;
852 resblks = xfs_default_resblks(mp);
854 xfs_reserve_blocks(mp, &resblks, NULL);
858 * Second stage of a freeze. The data is already frozen so we only
859 * need to take care of the metadata. Once that's done sync the superblock
860 * to the log to dirty it in case of a crash while frozen. This ensures that we
861 * will recover the unlinked inode lists on the next mount.
865 struct super_block *sb)
867 struct xfs_mount *mp = XFS_M(sb);
872 * The filesystem is now frozen far enough that memory reclaim
873 * cannot safely operate on the filesystem. Hence we need to
874 * set a GFP_NOFS context here to avoid recursion deadlocks.
876 flags = memalloc_nofs_save();
877 xfs_save_resvblks(mp);
878 ret = xfs_log_quiesce(mp);
879 memalloc_nofs_restore(flags);
882 * For read-write filesystems, we need to restart the inodegc on error
883 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
884 * going to be run to restart it now. We are at SB_FREEZE_FS level
885 * here, so we can restart safely without racing with a stop in
888 if (ret && !xfs_is_readonly(mp)) {
889 xfs_blockgc_start(mp);
890 xfs_inodegc_start(mp);
898 struct super_block *sb)
900 struct xfs_mount *mp = XFS_M(sb);
902 xfs_restore_resvblks(mp);
903 xfs_log_work_queue(mp);
906 * Don't reactivate the inodegc worker on a readonly filesystem because
907 * inodes are sent directly to reclaim. Don't reactivate the blockgc
908 * worker because there are no speculative preallocations on a readonly
911 if (!xfs_is_readonly(mp)) {
912 xfs_blockgc_start(mp);
913 xfs_inodegc_start(mp);
920 * This function fills in xfs_mount_t fields based on mount args.
921 * Note: the superblock _has_ now been read in.
925 struct xfs_mount *mp)
927 /* Fail a mount where the logbuf is smaller than the log stripe */
928 if (xfs_has_logv2(mp)) {
929 if (mp->m_logbsize <= 0 &&
930 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
931 mp->m_logbsize = mp->m_sb.sb_logsunit;
932 } else if (mp->m_logbsize > 0 &&
933 mp->m_logbsize < mp->m_sb.sb_logsunit) {
935 "logbuf size must be greater than or equal to log stripe size");
939 /* Fail a mount if the logbuf is larger than 32K */
940 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
942 "logbuf size for version 1 logs must be 16K or 32K");
948 * V5 filesystems always use attr2 format for attributes.
950 if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
951 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
952 "attr2 is always enabled for V5 filesystems.");
957 * prohibit r/w mounts of read-only filesystems
959 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
961 "cannot mount a read-only filesystem as read-write");
965 if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
966 (mp->m_qflags & XFS_PQUOTA_ACCT) &&
967 !xfs_has_pquotino(mp)) {
969 "Super block does not support project and group quota together");
977 xfs_init_percpu_counters(
978 struct xfs_mount *mp)
982 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
986 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
990 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
994 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1001 percpu_counter_destroy(&mp->m_fdblocks);
1003 percpu_counter_destroy(&mp->m_ifree);
1005 percpu_counter_destroy(&mp->m_icount);
1010 xfs_reinit_percpu_counters(
1011 struct xfs_mount *mp)
1013 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1014 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1015 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1019 xfs_destroy_percpu_counters(
1020 struct xfs_mount *mp)
1022 percpu_counter_destroy(&mp->m_icount);
1023 percpu_counter_destroy(&mp->m_ifree);
1024 percpu_counter_destroy(&mp->m_fdblocks);
1025 ASSERT(xfs_is_shutdown(mp) ||
1026 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1027 percpu_counter_destroy(&mp->m_delalloc_blks);
1031 xfs_inodegc_init_percpu(
1032 struct xfs_mount *mp)
1034 struct xfs_inodegc *gc;
1037 mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1041 for_each_possible_cpu(cpu) {
1042 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1043 init_llist_head(&gc->list);
1045 INIT_WORK(&gc->work, xfs_inodegc_worker);
1051 xfs_inodegc_free_percpu(
1052 struct xfs_mount *mp)
1056 free_percpu(mp->m_inodegc);
1061 struct super_block *sb)
1063 struct xfs_mount *mp = XFS_M(sb);
1065 /* if ->fill_super failed, we have no mount to tear down */
1069 xfs_notice(mp, "Unmounting Filesystem");
1070 xfs_filestream_unmount(mp);
1074 free_percpu(mp->m_stats.xs_stats);
1075 xfs_mount_list_del(mp);
1076 xfs_inodegc_free_percpu(mp);
1077 xfs_destroy_percpu_counters(mp);
1078 xfs_destroy_mount_workqueues(mp);
1079 xfs_close_devices(mp);
1081 sb->s_fs_info = NULL;
1086 xfs_fs_nr_cached_objects(
1087 struct super_block *sb,
1088 struct shrink_control *sc)
1090 /* Paranoia: catch incorrect calls during mount setup or teardown */
1091 if (WARN_ON_ONCE(!sb->s_fs_info))
1093 return xfs_reclaim_inodes_count(XFS_M(sb));
1097 xfs_fs_free_cached_objects(
1098 struct super_block *sb,
1099 struct shrink_control *sc)
1101 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1104 static const struct super_operations xfs_super_operations = {
1105 .alloc_inode = xfs_fs_alloc_inode,
1106 .destroy_inode = xfs_fs_destroy_inode,
1107 .dirty_inode = xfs_fs_dirty_inode,
1108 .drop_inode = xfs_fs_drop_inode,
1109 .put_super = xfs_fs_put_super,
1110 .sync_fs = xfs_fs_sync_fs,
1111 .freeze_fs = xfs_fs_freeze,
1112 .unfreeze_fs = xfs_fs_unfreeze,
1113 .statfs = xfs_fs_statfs,
1114 .show_options = xfs_fs_show_options,
1115 .nr_cached_objects = xfs_fs_nr_cached_objects,
1116 .free_cached_objects = xfs_fs_free_cached_objects,
1125 int last, shift_left_factor = 0, _res;
1129 value = kstrdup(s, GFP_KERNEL);
1133 last = strlen(value) - 1;
1134 if (value[last] == 'K' || value[last] == 'k') {
1135 shift_left_factor = 10;
1138 if (value[last] == 'M' || value[last] == 'm') {
1139 shift_left_factor = 20;
1142 if (value[last] == 'G' || value[last] == 'g') {
1143 shift_left_factor = 30;
1147 if (kstrtoint(value, base, &_res))
1150 *res = _res << shift_left_factor;
1155 xfs_fs_warn_deprecated(
1156 struct fs_context *fc,
1157 struct fs_parameter *param,
1161 /* Don't print the warning if reconfiguring and current mount point
1162 * already had the flag set
1164 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1165 !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1167 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1171 * Set mount state from a mount option.
1173 * NOTE: mp->m_super is NULL here!
1177 struct fs_context *fc,
1178 struct fs_parameter *param)
1180 struct xfs_mount *parsing_mp = fc->s_fs_info;
1181 struct fs_parse_result result;
1185 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1191 parsing_mp->m_logbufs = result.uint_32;
1194 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1198 kfree(parsing_mp->m_logname);
1199 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1200 if (!parsing_mp->m_logname)
1204 kfree(parsing_mp->m_rtname);
1205 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1206 if (!parsing_mp->m_rtname)
1210 if (suffix_kstrtoint(param->string, 10, &size))
1212 parsing_mp->m_allocsize_log = ffs(size) - 1;
1213 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1217 parsing_mp->m_features |= XFS_FEAT_GRPID;
1220 case Opt_sysvgroups:
1221 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1224 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1226 case Opt_norecovery:
1227 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1230 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1233 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1236 parsing_mp->m_dalign = result.uint_32;
1239 parsing_mp->m_swidth = result.uint_32;
1242 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1245 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1248 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1251 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1254 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1256 case Opt_filestreams:
1257 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1260 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1261 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1266 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1268 case Opt_qnoenforce:
1269 case Opt_uqnoenforce:
1270 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1271 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1275 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1277 case Opt_pqnoenforce:
1278 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1279 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1283 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1285 case Opt_gqnoenforce:
1286 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1287 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1290 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1293 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1295 #ifdef CONFIG_FS_DAX
1297 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1300 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1303 /* Following mount options will be removed in September 2025 */
1305 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1306 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1309 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1310 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1313 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1314 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1317 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1318 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1321 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1329 xfs_fs_validate_params(
1330 struct xfs_mount *mp)
1332 /* No recovery flag requires a read-only mount */
1333 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1334 xfs_warn(mp, "no-recovery mounts must be read-only.");
1339 * We have not read the superblock at this point, so only the attr2
1340 * mount option can set the attr2 feature by this stage.
1342 if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1343 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1348 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1350 "sunit and swidth options incompatible with the noalign option");
1354 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1355 xfs_warn(mp, "quota support not available in this kernel.");
1359 if ((mp->m_dalign && !mp->m_swidth) ||
1360 (!mp->m_dalign && mp->m_swidth)) {
1361 xfs_warn(mp, "sunit and swidth must be specified together");
1365 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1367 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1368 mp->m_swidth, mp->m_dalign);
1372 if (mp->m_logbufs != -1 &&
1373 mp->m_logbufs != 0 &&
1374 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1375 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1376 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1377 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1381 if (mp->m_logbsize != -1 &&
1382 mp->m_logbsize != 0 &&
1383 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1384 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1385 !is_power_of_2(mp->m_logbsize))) {
1387 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1392 if (xfs_has_allocsize(mp) &&
1393 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1394 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1395 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1396 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1405 struct super_block *sb,
1406 struct fs_context *fc)
1408 struct xfs_mount *mp = sb->s_fs_info;
1410 int flags = 0, error;
1414 error = xfs_fs_validate_params(mp);
1416 goto out_free_names;
1418 sb_min_blocksize(sb, BBSIZE);
1419 sb->s_xattr = xfs_xattr_handlers;
1420 sb->s_export_op = &xfs_export_operations;
1421 #ifdef CONFIG_XFS_QUOTA
1422 sb->s_qcop = &xfs_quotactl_operations;
1423 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1425 sb->s_op = &xfs_super_operations;
1428 * Delay mount work if the debug hook is set. This is debug
1429 * instrumention to coordinate simulation of xfs mount failures with
1430 * VFS superblock operations
1432 if (xfs_globals.mount_delay) {
1433 xfs_notice(mp, "Delaying mount for %d seconds.",
1434 xfs_globals.mount_delay);
1435 msleep(xfs_globals.mount_delay * 1000);
1438 if (fc->sb_flags & SB_SILENT)
1439 flags |= XFS_MFSI_QUIET;
1441 error = xfs_open_devices(mp);
1443 goto out_free_names;
1445 error = xfs_init_mount_workqueues(mp);
1447 goto out_close_devices;
1449 error = xfs_init_percpu_counters(mp);
1451 goto out_destroy_workqueues;
1453 error = xfs_inodegc_init_percpu(mp);
1455 goto out_destroy_counters;
1458 * All percpu data structures requiring cleanup when a cpu goes offline
1459 * must be allocated before adding this @mp to the cpu-dead handler's
1462 xfs_mount_list_add(mp);
1464 /* Allocate stats memory before we do operations that might use it */
1465 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1466 if (!mp->m_stats.xs_stats) {
1468 goto out_destroy_inodegc;
1471 error = xfs_readsb(mp, flags);
1473 goto out_free_stats;
1475 error = xfs_finish_flags(mp);
1479 error = xfs_setup_devices(mp);
1483 /* V4 support is undergoing deprecation. */
1484 if (!xfs_has_crc(mp)) {
1485 #ifdef CONFIG_XFS_SUPPORT_V4
1487 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1490 "Deprecated V4 format (crc=0) not supported by kernel.");
1496 /* Filesystem claims it needs repair, so refuse the mount. */
1497 if (xfs_has_needsrepair(mp)) {
1498 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1499 error = -EFSCORRUPTED;
1504 * Don't touch the filesystem if a user tool thinks it owns the primary
1505 * superblock. mkfs doesn't clear the flag from secondary supers, so
1506 * we don't check them at all.
1508 if (mp->m_sb.sb_inprogress) {
1509 xfs_warn(mp, "Offline file system operation in progress!");
1510 error = -EFSCORRUPTED;
1515 * Until this is fixed only page-sized or smaller data blocks work.
1517 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1519 "File system with blocksize %d bytes. "
1520 "Only pagesize (%ld) or less will currently work.",
1521 mp->m_sb.sb_blocksize, PAGE_SIZE);
1526 /* Ensure this filesystem fits in the page cache limits */
1527 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1528 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1530 "file system too large to be mounted on this system.");
1536 * XFS block mappings use 54 bits to store the logical block offset.
1537 * This should suffice to handle the maximum file size that the VFS
1538 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1539 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1540 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1541 * to check this assertion.
1543 * Avoid integer overflow by comparing the maximum bmbt offset to the
1544 * maximum pagecache offset in units of fs blocks.
1546 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1548 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1549 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1555 error = xfs_filestream_mount(mp);
1560 * we must configure the block size in the superblock before we run the
1561 * full mount process as the mount process can lookup and cache inodes.
1563 sb->s_magic = XFS_SUPER_MAGIC;
1564 sb->s_blocksize = mp->m_sb.sb_blocksize;
1565 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1566 sb->s_maxbytes = MAX_LFS_FILESIZE;
1567 sb->s_max_links = XFS_MAXLINK;
1568 sb->s_time_gran = 1;
1569 if (xfs_has_bigtime(mp)) {
1570 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1571 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1573 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1574 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1576 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1577 sb->s_iflags |= SB_I_CGROUPWB;
1579 set_posix_acl_flag(sb);
1581 /* version 5 superblocks support inode version counters. */
1582 if (xfs_has_crc(mp))
1583 sb->s_flags |= SB_I_VERSION;
1585 if (xfs_has_dax_always(mp)) {
1586 bool rtdev_is_dax = false, datadev_is_dax;
1589 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1591 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1593 if (mp->m_rtdev_targp)
1594 rtdev_is_dax = bdev_dax_supported(
1595 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1596 if (!rtdev_is_dax && !datadev_is_dax) {
1598 "DAX unsupported by block device. Turning off DAX.");
1599 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1601 if (xfs_has_reflink(mp)) {
1603 "DAX and reflink cannot be used together!");
1605 goto out_filestream_unmount;
1609 if (xfs_has_discard(mp)) {
1610 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1612 if (!blk_queue_discard(q)) {
1613 xfs_warn(mp, "mounting with \"discard\" option, but "
1614 "the device does not support discard");
1615 mp->m_features &= ~XFS_FEAT_DISCARD;
1619 if (xfs_has_reflink(mp)) {
1620 if (mp->m_sb.sb_rblocks) {
1622 "reflink not compatible with realtime device!");
1624 goto out_filestream_unmount;
1627 if (xfs_globals.always_cow) {
1628 xfs_info(mp, "using DEBUG-only always_cow mode.");
1629 mp->m_always_cow = true;
1633 if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1635 "reverse mapping btree not compatible with realtime device!");
1637 goto out_filestream_unmount;
1640 error = xfs_mountfs(mp);
1642 goto out_filestream_unmount;
1644 root = igrab(VFS_I(mp->m_rootip));
1649 sb->s_root = d_make_root(root);
1657 out_filestream_unmount:
1658 xfs_filestream_unmount(mp);
1662 free_percpu(mp->m_stats.xs_stats);
1663 out_destroy_inodegc:
1664 xfs_mount_list_del(mp);
1665 xfs_inodegc_free_percpu(mp);
1666 out_destroy_counters:
1667 xfs_destroy_percpu_counters(mp);
1668 out_destroy_workqueues:
1669 xfs_destroy_mount_workqueues(mp);
1671 xfs_close_devices(mp);
1673 sb->s_fs_info = NULL;
1678 xfs_filestream_unmount(mp);
1685 struct fs_context *fc)
1687 return get_tree_bdev(fc, xfs_fs_fill_super);
1692 struct xfs_mount *mp)
1694 struct xfs_sb *sbp = &mp->m_sb;
1697 if (xfs_has_norecovery(mp)) {
1699 "ro->rw transition prohibited on norecovery mount");
1703 if (xfs_sb_is_v5(sbp) &&
1704 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1706 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1707 (sbp->sb_features_ro_compat &
1708 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1712 clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1715 * If this is the first remount to writeable state we might have some
1716 * superblock changes to update.
1718 if (mp->m_update_sb) {
1719 error = xfs_sync_sb(mp, false);
1721 xfs_warn(mp, "failed to write sb changes");
1724 mp->m_update_sb = false;
1728 * Fill out the reserve pool if it is empty. Use the stashed value if
1729 * it is non-zero, otherwise go with the default.
1731 xfs_restore_resvblks(mp);
1732 xfs_log_work_queue(mp);
1734 /* Recover any CoW blocks that never got remapped. */
1735 error = xfs_reflink_recover_cow(mp);
1738 "Error %d recovering leftover CoW allocations.", error);
1739 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1742 xfs_blockgc_start(mp);
1744 /* Create the per-AG metadata reservation pool .*/
1745 error = xfs_fs_reserve_ag_blocks(mp);
1746 if (error && error != -ENOSPC)
1749 /* Re-enable the background inode inactivation worker. */
1750 xfs_inodegc_start(mp);
1757 struct xfs_mount *mp)
1762 * Cancel background eofb scanning so it cannot race with the final
1763 * log force+buftarg wait and deadlock the remount.
1765 xfs_blockgc_stop(mp);
1767 /* Get rid of any leftover CoW reservations... */
1768 error = xfs_blockgc_free_space(mp, NULL);
1770 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1775 * Stop the inodegc background worker. xfs_fs_reconfigure already
1776 * flushed all pending inodegc work when it sync'd the filesystem.
1777 * The VFS holds s_umount, so we know that inodes cannot enter
1778 * xfs_fs_destroy_inode during a remount operation. In readonly mode
1779 * we send inodes straight to reclaim, so no inodes will be queued.
1781 xfs_inodegc_stop(mp);
1783 /* Free the per-AG metadata reservation pool. */
1784 error = xfs_fs_unreserve_ag_blocks(mp);
1786 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1791 * Before we sync the metadata, we need to free up the reserve block
1792 * pool so that the used block count in the superblock on disk is
1793 * correct at the end of the remount. Stash the current* reserve pool
1794 * size so that if we get remounted rw, we can return it to the same
1797 xfs_save_resvblks(mp);
1800 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1806 * Logically we would return an error here to prevent users from believing
1807 * they might have changed mount options using remount which can't be changed.
1809 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1810 * arguments in some cases so we can't blindly reject options, but have to
1811 * check for each specified option if it actually differs from the currently
1812 * set option and only reject it if that's the case.
1814 * Until that is implemented we return success for every remount request, and
1815 * silently ignore all options that we can't actually change.
1819 struct fs_context *fc)
1821 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1822 struct xfs_mount *new_mp = fc->s_fs_info;
1823 int flags = fc->sb_flags;
1826 /* version 5 superblocks always support version counters. */
1827 if (xfs_has_crc(mp))
1828 fc->sb_flags |= SB_I_VERSION;
1830 error = xfs_fs_validate_params(new_mp);
1834 sync_filesystem(mp->m_super);
1836 /* inode32 -> inode64 */
1837 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1838 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1839 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1842 /* inode64 -> inode32 */
1843 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1844 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1845 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1849 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1850 error = xfs_remount_rw(mp);
1856 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1857 error = xfs_remount_ro(mp);
1865 static void xfs_fs_free(
1866 struct fs_context *fc)
1868 struct xfs_mount *mp = fc->s_fs_info;
1871 * mp is stored in the fs_context when it is initialized.
1872 * mp is transferred to the superblock on a successful mount,
1873 * but if an error occurs before the transfer we have to free
1880 static const struct fs_context_operations xfs_context_ops = {
1881 .parse_param = xfs_fs_parse_param,
1882 .get_tree = xfs_fs_get_tree,
1883 .reconfigure = xfs_fs_reconfigure,
1884 .free = xfs_fs_free,
1887 static int xfs_init_fs_context(
1888 struct fs_context *fc)
1890 struct xfs_mount *mp;
1892 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1896 spin_lock_init(&mp->m_sb_lock);
1897 spin_lock_init(&mp->m_agirotor_lock);
1898 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1899 spin_lock_init(&mp->m_perag_lock);
1900 mutex_init(&mp->m_growlock);
1901 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1902 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1903 mp->m_kobj.kobject.kset = xfs_kset;
1905 * We don't create the finobt per-ag space reservation until after log
1906 * recovery, so we must set this to true so that an ifree transaction
1907 * started during log recovery will not depend on space reservations
1908 * for finobt expansion.
1910 mp->m_finobt_nores = true;
1913 * These can be overridden by the mount option parsing.
1916 mp->m_logbsize = -1;
1917 mp->m_allocsize_log = 16; /* 64k */
1920 * Copy binary VFS mount flags we are interested in.
1922 if (fc->sb_flags & SB_RDONLY)
1923 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1924 if (fc->sb_flags & SB_DIRSYNC)
1925 mp->m_features |= XFS_FEAT_DIRSYNC;
1926 if (fc->sb_flags & SB_SYNCHRONOUS)
1927 mp->m_features |= XFS_FEAT_WSYNC;
1930 fc->ops = &xfs_context_ops;
1935 static struct file_system_type xfs_fs_type = {
1936 .owner = THIS_MODULE,
1938 .init_fs_context = xfs_init_fs_context,
1939 .parameters = xfs_fs_parameters,
1940 .kill_sb = kill_block_super,
1941 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1943 MODULE_ALIAS_FS("xfs");
1946 xfs_init_zones(void)
1948 xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1949 sizeof(struct xlog_ticket),
1951 if (!xfs_log_ticket_zone)
1954 xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1955 sizeof(struct xfs_extent_free_item),
1957 if (!xfs_bmap_free_item_zone)
1958 goto out_destroy_log_ticket_zone;
1960 xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1961 sizeof(struct xfs_btree_cur),
1963 if (!xfs_btree_cur_zone)
1964 goto out_destroy_bmap_free_item_zone;
1966 xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1967 sizeof(struct xfs_da_state),
1969 if (!xfs_da_state_zone)
1970 goto out_destroy_btree_cur_zone;
1972 xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1973 sizeof(struct xfs_ifork),
1975 if (!xfs_ifork_zone)
1976 goto out_destroy_da_state_zone;
1978 xfs_trans_zone = kmem_cache_create("xfs_trans",
1979 sizeof(struct xfs_trans),
1981 if (!xfs_trans_zone)
1982 goto out_destroy_ifork_zone;
1986 * The size of the zone allocated buf log item is the maximum
1987 * size possible under XFS. This wastes a little bit of memory,
1988 * but it is much faster.
1990 xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1991 sizeof(struct xfs_buf_log_item),
1993 if (!xfs_buf_item_zone)
1994 goto out_destroy_trans_zone;
1996 xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1997 (sizeof(struct xfs_efd_log_item) +
1998 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1999 sizeof(struct xfs_extent)),
2002 goto out_destroy_buf_item_zone;
2004 xfs_efi_zone = kmem_cache_create("xfs_efi_item",
2005 (sizeof(struct xfs_efi_log_item) +
2006 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2007 sizeof(struct xfs_extent)),
2010 goto out_destroy_efd_zone;
2012 xfs_inode_zone = kmem_cache_create("xfs_inode",
2013 sizeof(struct xfs_inode), 0,
2014 (SLAB_HWCACHE_ALIGN |
2015 SLAB_RECLAIM_ACCOUNT |
2016 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2017 xfs_fs_inode_init_once);
2018 if (!xfs_inode_zone)
2019 goto out_destroy_efi_zone;
2021 xfs_ili_zone = kmem_cache_create("xfs_ili",
2022 sizeof(struct xfs_inode_log_item), 0,
2023 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2026 goto out_destroy_inode_zone;
2028 xfs_icreate_zone = kmem_cache_create("xfs_icr",
2029 sizeof(struct xfs_icreate_item),
2031 if (!xfs_icreate_zone)
2032 goto out_destroy_ili_zone;
2034 xfs_rud_zone = kmem_cache_create("xfs_rud_item",
2035 sizeof(struct xfs_rud_log_item),
2038 goto out_destroy_icreate_zone;
2040 xfs_rui_zone = kmem_cache_create("xfs_rui_item",
2041 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2044 goto out_destroy_rud_zone;
2046 xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2047 sizeof(struct xfs_cud_log_item),
2050 goto out_destroy_rui_zone;
2052 xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2053 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2056 goto out_destroy_cud_zone;
2058 xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2059 sizeof(struct xfs_bud_log_item),
2062 goto out_destroy_cui_zone;
2064 xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2065 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2068 goto out_destroy_bud_zone;
2072 out_destroy_bud_zone:
2073 kmem_cache_destroy(xfs_bud_zone);
2074 out_destroy_cui_zone:
2075 kmem_cache_destroy(xfs_cui_zone);
2076 out_destroy_cud_zone:
2077 kmem_cache_destroy(xfs_cud_zone);
2078 out_destroy_rui_zone:
2079 kmem_cache_destroy(xfs_rui_zone);
2080 out_destroy_rud_zone:
2081 kmem_cache_destroy(xfs_rud_zone);
2082 out_destroy_icreate_zone:
2083 kmem_cache_destroy(xfs_icreate_zone);
2084 out_destroy_ili_zone:
2085 kmem_cache_destroy(xfs_ili_zone);
2086 out_destroy_inode_zone:
2087 kmem_cache_destroy(xfs_inode_zone);
2088 out_destroy_efi_zone:
2089 kmem_cache_destroy(xfs_efi_zone);
2090 out_destroy_efd_zone:
2091 kmem_cache_destroy(xfs_efd_zone);
2092 out_destroy_buf_item_zone:
2093 kmem_cache_destroy(xfs_buf_item_zone);
2094 out_destroy_trans_zone:
2095 kmem_cache_destroy(xfs_trans_zone);
2096 out_destroy_ifork_zone:
2097 kmem_cache_destroy(xfs_ifork_zone);
2098 out_destroy_da_state_zone:
2099 kmem_cache_destroy(xfs_da_state_zone);
2100 out_destroy_btree_cur_zone:
2101 kmem_cache_destroy(xfs_btree_cur_zone);
2102 out_destroy_bmap_free_item_zone:
2103 kmem_cache_destroy(xfs_bmap_free_item_zone);
2104 out_destroy_log_ticket_zone:
2105 kmem_cache_destroy(xfs_log_ticket_zone);
2111 xfs_destroy_zones(void)
2114 * Make sure all delayed rcu free are flushed before we
2118 kmem_cache_destroy(xfs_bui_zone);
2119 kmem_cache_destroy(xfs_bud_zone);
2120 kmem_cache_destroy(xfs_cui_zone);
2121 kmem_cache_destroy(xfs_cud_zone);
2122 kmem_cache_destroy(xfs_rui_zone);
2123 kmem_cache_destroy(xfs_rud_zone);
2124 kmem_cache_destroy(xfs_icreate_zone);
2125 kmem_cache_destroy(xfs_ili_zone);
2126 kmem_cache_destroy(xfs_inode_zone);
2127 kmem_cache_destroy(xfs_efi_zone);
2128 kmem_cache_destroy(xfs_efd_zone);
2129 kmem_cache_destroy(xfs_buf_item_zone);
2130 kmem_cache_destroy(xfs_trans_zone);
2131 kmem_cache_destroy(xfs_ifork_zone);
2132 kmem_cache_destroy(xfs_da_state_zone);
2133 kmem_cache_destroy(xfs_btree_cur_zone);
2134 kmem_cache_destroy(xfs_bmap_free_item_zone);
2135 kmem_cache_destroy(xfs_log_ticket_zone);
2139 xfs_init_workqueues(void)
2142 * The allocation workqueue can be used in memory reclaim situations
2143 * (writepage path), and parallelism is only limited by the number of
2144 * AGs in all the filesystems mounted. Hence use the default large
2145 * max_active value for this workqueue.
2147 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2148 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2152 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2154 if (!xfs_discard_wq)
2155 goto out_free_alloc_wq;
2159 destroy_workqueue(xfs_alloc_wq);
2164 xfs_destroy_workqueues(void)
2166 destroy_workqueue(xfs_discard_wq);
2167 destroy_workqueue(xfs_alloc_wq);
2170 #ifdef CONFIG_HOTPLUG_CPU
2175 struct xfs_mount *mp, *n;
2177 spin_lock(&xfs_mount_list_lock);
2178 list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2179 spin_unlock(&xfs_mount_list_lock);
2180 xfs_inodegc_cpu_dead(mp, cpu);
2181 spin_lock(&xfs_mount_list_lock);
2183 spin_unlock(&xfs_mount_list_lock);
2188 xfs_cpu_hotplug_init(void)
2192 error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2196 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2202 xfs_cpu_hotplug_destroy(void)
2204 cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2207 #else /* !CONFIG_HOTPLUG_CPU */
2208 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2209 static inline void xfs_cpu_hotplug_destroy(void) {}
2217 xfs_check_ondisk_structs();
2219 printk(KERN_INFO XFS_VERSION_STRING " with "
2220 XFS_BUILD_OPTIONS " enabled\n");
2224 error = xfs_cpu_hotplug_init();
2228 error = xfs_init_zones();
2230 goto out_destroy_hp;
2232 error = xfs_init_workqueues();
2234 goto out_destroy_zones;
2236 error = xfs_mru_cache_init();
2238 goto out_destroy_wq;
2240 error = xfs_buf_init();
2242 goto out_mru_cache_uninit;
2244 error = xfs_init_procfs();
2246 goto out_buf_terminate;
2248 error = xfs_sysctl_register();
2250 goto out_cleanup_procfs;
2252 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2255 goto out_sysctl_unregister;
2258 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2260 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2261 if (!xfsstats.xs_stats) {
2263 goto out_kset_unregister;
2266 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2269 goto out_free_stats;
2272 xfs_dbg_kobj.kobject.kset = xfs_kset;
2273 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2275 goto out_remove_stats_kobj;
2278 error = xfs_qm_init();
2280 goto out_remove_dbg_kobj;
2282 error = register_filesystem(&xfs_fs_type);
2289 out_remove_dbg_kobj:
2291 xfs_sysfs_del(&xfs_dbg_kobj);
2292 out_remove_stats_kobj:
2294 xfs_sysfs_del(&xfsstats.xs_kobj);
2296 free_percpu(xfsstats.xs_stats);
2297 out_kset_unregister:
2298 kset_unregister(xfs_kset);
2299 out_sysctl_unregister:
2300 xfs_sysctl_unregister();
2302 xfs_cleanup_procfs();
2304 xfs_buf_terminate();
2305 out_mru_cache_uninit:
2306 xfs_mru_cache_uninit();
2308 xfs_destroy_workqueues();
2310 xfs_destroy_zones();
2312 xfs_cpu_hotplug_destroy();
2321 unregister_filesystem(&xfs_fs_type);
2323 xfs_sysfs_del(&xfs_dbg_kobj);
2325 xfs_sysfs_del(&xfsstats.xs_kobj);
2326 free_percpu(xfsstats.xs_stats);
2327 kset_unregister(xfs_kset);
2328 xfs_sysctl_unregister();
2329 xfs_cleanup_procfs();
2330 xfs_buf_terminate();
2331 xfs_mru_cache_uninit();
2332 xfs_destroy_workqueues();
2333 xfs_destroy_zones();
2334 xfs_uuid_table_free();
2335 xfs_cpu_hotplug_destroy();
2338 module_init(init_xfs_fs);
2339 module_exit(exit_xfs_fs);
2341 MODULE_AUTHOR("Silicon Graphics, Inc.");
2342 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2343 MODULE_LICENSE("GPL");