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"
39 #include <linux/magic.h>
40 #include <linux/fs_context.h>
41 #include <linux/fs_parser.h>
43 static const struct super_operations xfs_super_operations;
45 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
47 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
57 xfs_mount_set_dax_mode(
59 enum xfs_dax_mode mode)
63 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
66 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
67 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
70 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
71 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
76 static const struct constant_table dax_param_enums[] = {
77 {"inode", XFS_DAX_INODE },
78 {"always", XFS_DAX_ALWAYS },
79 {"never", XFS_DAX_NEVER },
84 * Table driven mount option parser.
87 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
88 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
89 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
90 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
91 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
92 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
93 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
94 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
95 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
98 static const struct fs_parameter_spec xfs_fs_parameters[] = {
99 fsparam_u32("logbufs", Opt_logbufs),
100 fsparam_string("logbsize", Opt_logbsize),
101 fsparam_string("logdev", Opt_logdev),
102 fsparam_string("rtdev", Opt_rtdev),
103 fsparam_flag("wsync", Opt_wsync),
104 fsparam_flag("noalign", Opt_noalign),
105 fsparam_flag("swalloc", Opt_swalloc),
106 fsparam_u32("sunit", Opt_sunit),
107 fsparam_u32("swidth", Opt_swidth),
108 fsparam_flag("nouuid", Opt_nouuid),
109 fsparam_flag("grpid", Opt_grpid),
110 fsparam_flag("nogrpid", Opt_nogrpid),
111 fsparam_flag("bsdgroups", Opt_bsdgroups),
112 fsparam_flag("sysvgroups", Opt_sysvgroups),
113 fsparam_string("allocsize", Opt_allocsize),
114 fsparam_flag("norecovery", Opt_norecovery),
115 fsparam_flag("inode64", Opt_inode64),
116 fsparam_flag("inode32", Opt_inode32),
117 fsparam_flag("ikeep", Opt_ikeep),
118 fsparam_flag("noikeep", Opt_noikeep),
119 fsparam_flag("largeio", Opt_largeio),
120 fsparam_flag("nolargeio", Opt_nolargeio),
121 fsparam_flag("attr2", Opt_attr2),
122 fsparam_flag("noattr2", Opt_noattr2),
123 fsparam_flag("filestreams", Opt_filestreams),
124 fsparam_flag("quota", Opt_quota),
125 fsparam_flag("noquota", Opt_noquota),
126 fsparam_flag("usrquota", Opt_usrquota),
127 fsparam_flag("grpquota", Opt_grpquota),
128 fsparam_flag("prjquota", Opt_prjquota),
129 fsparam_flag("uquota", Opt_uquota),
130 fsparam_flag("gquota", Opt_gquota),
131 fsparam_flag("pquota", Opt_pquota),
132 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
133 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
134 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
135 fsparam_flag("qnoenforce", Opt_qnoenforce),
136 fsparam_flag("discard", Opt_discard),
137 fsparam_flag("nodiscard", Opt_nodiscard),
138 fsparam_flag("dax", Opt_dax),
139 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
143 struct proc_xfs_info {
153 static struct proc_xfs_info xfs_info_set[] = {
154 /* the few simple ones we can get from the mount struct */
155 { XFS_MOUNT_IKEEP, ",ikeep" },
156 { XFS_MOUNT_WSYNC, ",wsync" },
157 { XFS_MOUNT_NOALIGN, ",noalign" },
158 { XFS_MOUNT_SWALLOC, ",swalloc" },
159 { XFS_MOUNT_NOUUID, ",nouuid" },
160 { XFS_MOUNT_NORECOVERY, ",norecovery" },
161 { XFS_MOUNT_ATTR2, ",attr2" },
162 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
163 { XFS_MOUNT_GRPID, ",grpid" },
164 { XFS_MOUNT_DISCARD, ",discard" },
165 { XFS_MOUNT_LARGEIO, ",largeio" },
166 { XFS_MOUNT_DAX_ALWAYS, ",dax=always" },
167 { XFS_MOUNT_DAX_NEVER, ",dax=never" },
170 struct xfs_mount *mp = XFS_M(root->d_sb);
171 struct proc_xfs_info *xfs_infop;
173 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
174 if (mp->m_flags & xfs_infop->flag)
175 seq_puts(m, xfs_infop->str);
178 seq_printf(m, ",inode%d",
179 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
181 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
182 seq_printf(m, ",allocsize=%dk",
183 (1 << mp->m_allocsize_log) >> 10);
185 if (mp->m_logbufs > 0)
186 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
187 if (mp->m_logbsize > 0)
188 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
191 seq_show_option(m, "logdev", mp->m_logname);
193 seq_show_option(m, "rtdev", mp->m_rtname);
195 if (mp->m_dalign > 0)
196 seq_printf(m, ",sunit=%d",
197 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
198 if (mp->m_swidth > 0)
199 seq_printf(m, ",swidth=%d",
200 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
202 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
203 seq_puts(m, ",usrquota");
204 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
205 seq_puts(m, ",uqnoenforce");
207 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
208 if (mp->m_qflags & XFS_PQUOTA_ENFD)
209 seq_puts(m, ",prjquota");
211 seq_puts(m, ",pqnoenforce");
213 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
214 if (mp->m_qflags & XFS_GQUOTA_ENFD)
215 seq_puts(m, ",grpquota");
217 seq_puts(m, ",gqnoenforce");
220 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
221 seq_puts(m, ",noquota");
227 * Set parameters for inode allocation heuristics, taking into account
228 * filesystem size and inode32/inode64 mount options; i.e. specifically
229 * whether or not XFS_MOUNT_SMALL_INUMS is set.
231 * Inode allocation patterns are altered only if inode32 is requested
232 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
233 * If altered, XFS_MOUNT_32BITINODES is set as well.
235 * An agcount independent of that in the mount structure is provided
236 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
237 * to the potentially higher ag count.
239 * Returns the maximum AG index which may contain inodes.
243 struct xfs_mount *mp,
244 xfs_agnumber_t agcount)
246 xfs_agnumber_t index;
247 xfs_agnumber_t maxagi = 0;
248 xfs_sb_t *sbp = &mp->m_sb;
249 xfs_agnumber_t max_metadata;
254 * Calculate how much should be reserved for inodes to meet
255 * the max inode percentage. Used only for inode32.
257 if (M_IGEO(mp)->maxicount) {
260 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
262 icount += sbp->sb_agblocks - 1;
263 do_div(icount, sbp->sb_agblocks);
264 max_metadata = icount;
266 max_metadata = agcount;
269 /* Get the last possible inode in the filesystem */
270 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
271 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
274 * If user asked for no more than 32-bit inodes, and the fs is
275 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
276 * the allocator to accommodate the request.
278 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
279 mp->m_flags |= XFS_MOUNT_32BITINODES;
281 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
283 for (index = 0; index < agcount; index++) {
284 struct xfs_perag *pag;
286 ino = XFS_AGINO_TO_INO(mp, index, agino);
288 pag = xfs_perag_get(mp, index);
290 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
291 if (ino > XFS_MAXINUMBER_32) {
292 pag->pagi_inodeok = 0;
293 pag->pagf_metadata = 0;
295 pag->pagi_inodeok = 1;
297 if (index < max_metadata)
298 pag->pagf_metadata = 1;
300 pag->pagf_metadata = 0;
303 pag->pagi_inodeok = 1;
304 pag->pagf_metadata = 0;
310 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
317 struct block_device **bdevp)
321 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
323 if (IS_ERR(*bdevp)) {
324 error = PTR_ERR(*bdevp);
325 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
333 struct block_device *bdev)
336 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
340 xfs_blkdev_issue_flush(
341 xfs_buftarg_t *buftarg)
343 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS);
348 struct xfs_mount *mp)
350 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
352 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
353 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
354 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
356 xfs_free_buftarg(mp->m_logdev_targp);
357 xfs_blkdev_put(logdev);
358 fs_put_dax(dax_logdev);
360 if (mp->m_rtdev_targp) {
361 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
362 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
364 xfs_free_buftarg(mp->m_rtdev_targp);
365 xfs_blkdev_put(rtdev);
366 fs_put_dax(dax_rtdev);
368 xfs_free_buftarg(mp->m_ddev_targp);
369 fs_put_dax(dax_ddev);
373 * The file system configurations are:
374 * (1) device (partition) with data and internal log
375 * (2) logical volume with data and log subvolumes.
376 * (3) logical volume with data, log, and realtime subvolumes.
378 * We only have to handle opening the log and realtime volumes here if
379 * they are present. The data subvolume has already been opened by
380 * get_sb_bdev() and is stored in sb->s_bdev.
384 struct xfs_mount *mp)
386 struct block_device *ddev = mp->m_super->s_bdev;
387 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
388 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
389 struct block_device *logdev = NULL, *rtdev = NULL;
393 * Open real time and log devices - order is important.
396 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
399 dax_logdev = fs_dax_get_by_bdev(logdev);
403 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
405 goto out_close_logdev;
407 if (rtdev == ddev || rtdev == logdev) {
409 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
411 goto out_close_rtdev;
413 dax_rtdev = fs_dax_get_by_bdev(rtdev);
417 * Setup xfs_mount buffer target pointers
420 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
421 if (!mp->m_ddev_targp)
422 goto out_close_rtdev;
425 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
426 if (!mp->m_rtdev_targp)
427 goto out_free_ddev_targ;
430 if (logdev && logdev != ddev) {
431 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
432 if (!mp->m_logdev_targp)
433 goto out_free_rtdev_targ;
435 mp->m_logdev_targp = mp->m_ddev_targp;
441 if (mp->m_rtdev_targp)
442 xfs_free_buftarg(mp->m_rtdev_targp);
444 xfs_free_buftarg(mp->m_ddev_targp);
446 xfs_blkdev_put(rtdev);
447 fs_put_dax(dax_rtdev);
449 if (logdev && logdev != ddev) {
450 xfs_blkdev_put(logdev);
451 fs_put_dax(dax_logdev);
454 fs_put_dax(dax_ddev);
459 * Setup xfs_mount buffer target pointers based on superblock
463 struct xfs_mount *mp)
467 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
471 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
472 unsigned int log_sector_size = BBSIZE;
474 if (xfs_sb_version_hassector(&mp->m_sb))
475 log_sector_size = mp->m_sb.sb_logsectsize;
476 error = xfs_setsize_buftarg(mp->m_logdev_targp,
481 if (mp->m_rtdev_targp) {
482 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
483 mp->m_sb.sb_sectsize);
492 xfs_init_mount_workqueues(
493 struct xfs_mount *mp)
495 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
496 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_super->s_id);
497 if (!mp->m_buf_workqueue)
500 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
501 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
502 if (!mp->m_unwritten_workqueue)
503 goto out_destroy_buf;
505 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
506 WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
507 0, mp->m_super->s_id);
508 if (!mp->m_cil_workqueue)
509 goto out_destroy_unwritten;
511 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
512 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
513 if (!mp->m_reclaim_workqueue)
514 goto out_destroy_cil;
516 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
517 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
518 if (!mp->m_eofblocks_workqueue)
519 goto out_destroy_reclaim;
521 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
523 if (!mp->m_sync_workqueue)
524 goto out_destroy_eofb;
529 destroy_workqueue(mp->m_eofblocks_workqueue);
531 destroy_workqueue(mp->m_reclaim_workqueue);
533 destroy_workqueue(mp->m_cil_workqueue);
534 out_destroy_unwritten:
535 destroy_workqueue(mp->m_unwritten_workqueue);
537 destroy_workqueue(mp->m_buf_workqueue);
543 xfs_destroy_mount_workqueues(
544 struct xfs_mount *mp)
546 destroy_workqueue(mp->m_sync_workqueue);
547 destroy_workqueue(mp->m_eofblocks_workqueue);
548 destroy_workqueue(mp->m_reclaim_workqueue);
549 destroy_workqueue(mp->m_cil_workqueue);
550 destroy_workqueue(mp->m_unwritten_workqueue);
551 destroy_workqueue(mp->m_buf_workqueue);
555 xfs_flush_inodes_worker(
556 struct work_struct *work)
558 struct xfs_mount *mp = container_of(work, struct xfs_mount,
559 m_flush_inodes_work);
560 struct super_block *sb = mp->m_super;
562 if (down_read_trylock(&sb->s_umount)) {
564 up_read(&sb->s_umount);
569 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
570 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
571 * for IO to complete so that we effectively throttle multiple callers to the
572 * rate at which IO is completing.
576 struct xfs_mount *mp)
579 * If flush_work() returns true then that means we waited for a flush
580 * which was already in progress. Don't bother running another scan.
582 if (flush_work(&mp->m_flush_inodes_work))
585 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
586 flush_work(&mp->m_flush_inodes_work);
589 /* Catch misguided souls that try to use this interface on XFS */
590 STATIC struct inode *
592 struct super_block *sb)
601 struct xfs_inode *ip,
604 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
605 struct xfs_bmbt_irec got;
606 struct xfs_iext_cursor icur;
608 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
611 if (isnullstartblock(got.br_startblock)) {
612 xfs_warn(ip->i_mount,
613 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
615 whichfork == XFS_DATA_FORK ? "data" : "cow",
616 got.br_startoff, got.br_blockcount);
618 } while (xfs_iext_next_extent(ifp, &icur, &got));
621 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
625 * Now that the generic code is guaranteed not to be accessing
626 * the linux inode, we can inactivate and reclaim the inode.
629 xfs_fs_destroy_inode(
632 struct xfs_inode *ip = XFS_I(inode);
634 trace_xfs_destroy_inode(ip);
636 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
637 XFS_STATS_INC(ip->i_mount, vn_rele);
638 XFS_STATS_INC(ip->i_mount, vn_remove);
642 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
643 xfs_check_delalloc(ip, XFS_DATA_FORK);
644 xfs_check_delalloc(ip, XFS_COW_FORK);
648 XFS_STATS_INC(ip->i_mount, vn_reclaim);
651 * We should never get here with one of the reclaim flags already set.
653 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
654 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
657 * We always use background reclaim here because even if the inode is
658 * clean, it still may be under IO and hence we have wait for IO
659 * completion to occur before we can reclaim the inode. The background
660 * reclaim path handles this more efficiently than we can here, so
661 * simply let background reclaim tear down all inodes.
663 xfs_inode_set_reclaim_tag(ip);
671 struct xfs_inode *ip = XFS_I(inode);
672 struct xfs_mount *mp = ip->i_mount;
673 struct xfs_trans *tp;
675 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
677 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
680 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
682 xfs_ilock(ip, XFS_ILOCK_EXCL);
683 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
684 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
685 xfs_trans_commit(tp);
689 * Slab object creation initialisation for the XFS inode.
690 * This covers only the idempotent fields in the XFS inode;
691 * all other fields need to be initialised on allocation
692 * from the slab. This avoids the need to repeatedly initialise
693 * fields in the xfs inode that left in the initialise state
694 * when freeing the inode.
697 xfs_fs_inode_init_once(
700 struct xfs_inode *ip = inode;
702 memset(ip, 0, sizeof(struct xfs_inode));
705 inode_init_once(VFS_I(ip));
708 atomic_set(&ip->i_pincount, 0);
709 spin_lock_init(&ip->i_flags_lock);
711 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
712 "xfsino", ip->i_ino);
713 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
714 "xfsino", ip->i_ino);
718 * We do an unlocked check for XFS_IDONTCACHE here because we are already
719 * serialised against cache hits here via the inode->i_lock and igrab() in
720 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
721 * racing with us, and it avoids needing to grab a spinlock here for every inode
722 * we drop the final reference on.
728 struct xfs_inode *ip = XFS_I(inode);
731 * If this unlinked inode is in the middle of recovery, don't
732 * drop the inode just yet; log recovery will take care of
733 * that. See the comment for this inode flag.
735 if (ip->i_flags & XFS_IRECOVERY) {
736 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
740 return generic_drop_inode(inode);
745 struct xfs_mount *mp)
748 kfree(mp->m_logname);
754 struct super_block *sb,
757 struct xfs_mount *mp = XFS_M(sb);
760 * Doing anything during the async pass would be counterproductive.
765 xfs_log_force(mp, XFS_LOG_SYNC);
768 * The disk must be active because we're syncing.
769 * We schedule log work now (now that the disk is
770 * active) instead of later (when it might not be).
772 flush_delayed_work(&mp->m_log->l_work);
780 struct dentry *dentry,
781 struct kstatfs *statp)
783 struct xfs_mount *mp = XFS_M(dentry->d_sb);
784 xfs_sb_t *sbp = &mp->m_sb;
785 struct xfs_inode *ip = XFS_I(d_inode(dentry));
786 uint64_t fakeinos, id;
793 statp->f_type = XFS_SUPER_MAGIC;
794 statp->f_namelen = MAXNAMELEN - 1;
796 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
797 statp->f_fsid.val[0] = (u32)id;
798 statp->f_fsid.val[1] = (u32)(id >> 32);
800 icount = percpu_counter_sum(&mp->m_icount);
801 ifree = percpu_counter_sum(&mp->m_ifree);
802 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
804 spin_lock(&mp->m_sb_lock);
805 statp->f_bsize = sbp->sb_blocksize;
806 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
807 statp->f_blocks = sbp->sb_dblocks - lsize;
808 spin_unlock(&mp->m_sb_lock);
810 /* make sure statp->f_bfree does not underflow */
811 statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
812 statp->f_bavail = statp->f_bfree;
814 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
815 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
816 if (M_IGEO(mp)->maxicount)
817 statp->f_files = min_t(typeof(statp->f_files),
819 M_IGEO(mp)->maxicount);
821 /* If sb_icount overshot maxicount, report actual allocation */
822 statp->f_files = max_t(typeof(statp->f_files),
826 /* make sure statp->f_ffree does not underflow */
827 ffree = statp->f_files - (icount - ifree);
828 statp->f_ffree = max_t(int64_t, ffree, 0);
831 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
832 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
833 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
834 xfs_qm_statvfs(ip, statp);
836 if (XFS_IS_REALTIME_MOUNT(mp) &&
837 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
838 statp->f_blocks = sbp->sb_rblocks;
839 statp->f_bavail = statp->f_bfree =
840 sbp->sb_frextents * sbp->sb_rextsize;
847 xfs_save_resvblks(struct xfs_mount *mp)
849 uint64_t resblks = 0;
851 mp->m_resblks_save = mp->m_resblks;
852 xfs_reserve_blocks(mp, &resblks, NULL);
856 xfs_restore_resvblks(struct xfs_mount *mp)
860 if (mp->m_resblks_save) {
861 resblks = mp->m_resblks_save;
862 mp->m_resblks_save = 0;
864 resblks = xfs_default_resblks(mp);
866 xfs_reserve_blocks(mp, &resblks, NULL);
870 * Trigger writeback of all the dirty metadata in the file system.
872 * This ensures that the metadata is written to their location on disk rather
873 * than just existing in transactions in the log. This means after a quiesce
874 * there is no log replay required to write the inodes to disk - this is the
875 * primary difference between a sync and a quiesce.
877 * We cancel log work early here to ensure all transactions the log worker may
878 * run have finished before we clean up and log the superblock and write an
879 * unmount record. The unfreeze process is responsible for restarting the log
884 struct xfs_mount *mp)
888 cancel_delayed_work_sync(&mp->m_log->l_work);
890 /* force the log to unpin objects from the now complete transactions */
891 xfs_log_force(mp, XFS_LOG_SYNC);
894 /* Push the superblock and write an unmount record */
895 error = xfs_log_sbcount(mp);
897 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
898 "Frozen image may not be consistent.");
903 * Second stage of a freeze. The data is already frozen so we only
904 * need to take care of the metadata. Once that's done sync the superblock
905 * to the log to dirty it in case of a crash while frozen. This ensures that we
906 * will recover the unlinked inode lists on the next mount.
910 struct super_block *sb)
912 struct xfs_mount *mp = XFS_M(sb);
917 * The filesystem is now frozen far enough that memory reclaim
918 * cannot safely operate on the filesystem. Hence we need to
919 * set a GFP_NOFS context here to avoid recursion deadlocks.
921 flags = memalloc_nofs_save();
922 xfs_stop_block_reaping(mp);
923 xfs_save_resvblks(mp);
924 xfs_quiesce_attr(mp);
925 ret = xfs_sync_sb(mp, true);
926 memalloc_nofs_restore(flags);
932 struct super_block *sb)
934 struct xfs_mount *mp = XFS_M(sb);
936 xfs_restore_resvblks(mp);
937 xfs_log_work_queue(mp);
938 xfs_start_block_reaping(mp);
943 * This function fills in xfs_mount_t fields based on mount args.
944 * Note: the superblock _has_ now been read in.
948 struct xfs_mount *mp)
950 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
952 /* Fail a mount where the logbuf is smaller than the log stripe */
953 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
954 if (mp->m_logbsize <= 0 &&
955 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
956 mp->m_logbsize = mp->m_sb.sb_logsunit;
957 } else if (mp->m_logbsize > 0 &&
958 mp->m_logbsize < mp->m_sb.sb_logsunit) {
960 "logbuf size must be greater than or equal to log stripe size");
964 /* Fail a mount if the logbuf is larger than 32K */
965 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
967 "logbuf size for version 1 logs must be 16K or 32K");
973 * V5 filesystems always use attr2 format for attributes.
975 if (xfs_sb_version_hascrc(&mp->m_sb) &&
976 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
977 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
978 "attr2 is always enabled for V5 filesystems.");
983 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
984 * told by noattr2 to turn it off
986 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
987 !(mp->m_flags & XFS_MOUNT_NOATTR2))
988 mp->m_flags |= XFS_MOUNT_ATTR2;
991 * prohibit r/w mounts of read-only filesystems
993 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
995 "cannot mount a read-only filesystem as read-write");
999 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1000 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1001 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1003 "Super block does not support project and group quota together");
1011 xfs_init_percpu_counters(
1012 struct xfs_mount *mp)
1016 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1020 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1024 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1028 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1035 percpu_counter_destroy(&mp->m_fdblocks);
1037 percpu_counter_destroy(&mp->m_ifree);
1039 percpu_counter_destroy(&mp->m_icount);
1044 xfs_reinit_percpu_counters(
1045 struct xfs_mount *mp)
1047 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1048 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1049 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1053 xfs_destroy_percpu_counters(
1054 struct xfs_mount *mp)
1056 percpu_counter_destroy(&mp->m_icount);
1057 percpu_counter_destroy(&mp->m_ifree);
1058 percpu_counter_destroy(&mp->m_fdblocks);
1059 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1060 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1061 percpu_counter_destroy(&mp->m_delalloc_blks);
1066 struct super_block *sb)
1068 struct xfs_mount *mp = XFS_M(sb);
1070 /* if ->fill_super failed, we have no mount to tear down */
1074 xfs_notice(mp, "Unmounting Filesystem");
1075 xfs_filestream_unmount(mp);
1079 free_percpu(mp->m_stats.xs_stats);
1080 xfs_destroy_percpu_counters(mp);
1081 xfs_destroy_mount_workqueues(mp);
1082 xfs_close_devices(mp);
1084 sb->s_fs_info = NULL;
1089 xfs_fs_nr_cached_objects(
1090 struct super_block *sb,
1091 struct shrink_control *sc)
1093 /* Paranoia: catch incorrect calls during mount setup or teardown */
1094 if (WARN_ON_ONCE(!sb->s_fs_info))
1096 return xfs_reclaim_inodes_count(XFS_M(sb));
1100 xfs_fs_free_cached_objects(
1101 struct super_block *sb,
1102 struct shrink_control *sc)
1104 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1107 static const struct super_operations xfs_super_operations = {
1108 .alloc_inode = xfs_fs_alloc_inode,
1109 .destroy_inode = xfs_fs_destroy_inode,
1110 .dirty_inode = xfs_fs_dirty_inode,
1111 .drop_inode = xfs_fs_drop_inode,
1112 .put_super = xfs_fs_put_super,
1113 .sync_fs = xfs_fs_sync_fs,
1114 .freeze_fs = xfs_fs_freeze,
1115 .unfreeze_fs = xfs_fs_unfreeze,
1116 .statfs = xfs_fs_statfs,
1117 .show_options = xfs_fs_show_options,
1118 .nr_cached_objects = xfs_fs_nr_cached_objects,
1119 .free_cached_objects = xfs_fs_free_cached_objects,
1128 int last, shift_left_factor = 0, _res;
1132 value = kstrdup(s, GFP_KERNEL);
1136 last = strlen(value) - 1;
1137 if (value[last] == 'K' || value[last] == 'k') {
1138 shift_left_factor = 10;
1141 if (value[last] == 'M' || value[last] == 'm') {
1142 shift_left_factor = 20;
1145 if (value[last] == 'G' || value[last] == 'g') {
1146 shift_left_factor = 30;
1150 if (kstrtoint(value, base, &_res))
1153 *res = _res << shift_left_factor;
1158 * Set mount state from a mount option.
1160 * NOTE: mp->m_super is NULL here!
1164 struct fs_context *fc,
1165 struct fs_parameter *param)
1167 struct xfs_mount *mp = fc->s_fs_info;
1168 struct fs_parse_result result;
1172 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1178 mp->m_logbufs = result.uint_32;
1181 if (suffix_kstrtoint(param->string, 10, &mp->m_logbsize))
1185 kfree(mp->m_logname);
1186 mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1191 kfree(mp->m_rtname);
1192 mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1197 if (suffix_kstrtoint(param->string, 10, &size))
1199 mp->m_allocsize_log = ffs(size) - 1;
1200 mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1204 mp->m_flags |= XFS_MOUNT_GRPID;
1207 case Opt_sysvgroups:
1208 mp->m_flags &= ~XFS_MOUNT_GRPID;
1211 mp->m_flags |= XFS_MOUNT_WSYNC;
1213 case Opt_norecovery:
1214 mp->m_flags |= XFS_MOUNT_NORECOVERY;
1217 mp->m_flags |= XFS_MOUNT_NOALIGN;
1220 mp->m_flags |= XFS_MOUNT_SWALLOC;
1223 mp->m_dalign = result.uint_32;
1226 mp->m_swidth = result.uint_32;
1229 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1232 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1235 mp->m_flags |= XFS_MOUNT_NOUUID;
1238 mp->m_flags |= XFS_MOUNT_LARGEIO;
1241 mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1243 case Opt_filestreams:
1244 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1247 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1248 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1249 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1254 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1257 case Opt_qnoenforce:
1258 case Opt_uqnoenforce:
1259 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1260 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1264 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1267 case Opt_pqnoenforce:
1268 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1269 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1273 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1276 case Opt_gqnoenforce:
1277 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1278 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1281 mp->m_flags |= XFS_MOUNT_DISCARD;
1284 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1286 #ifdef CONFIG_FS_DAX
1288 xfs_mount_set_dax_mode(mp, XFS_DAX_ALWAYS);
1291 xfs_mount_set_dax_mode(mp, result.uint_32);
1294 /* Following mount options will be removed in September 2025 */
1296 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1297 mp->m_flags |= XFS_MOUNT_IKEEP;
1300 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1301 mp->m_flags &= ~XFS_MOUNT_IKEEP;
1304 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1305 mp->m_flags |= XFS_MOUNT_ATTR2;
1308 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1309 mp->m_flags &= ~XFS_MOUNT_ATTR2;
1310 mp->m_flags |= XFS_MOUNT_NOATTR2;
1313 xfs_warn(mp, "unknown mount option [%s].", param->key);
1321 xfs_fc_validate_params(
1322 struct xfs_mount *mp)
1325 * no recovery flag requires a read-only mount
1327 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1328 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1329 xfs_warn(mp, "no-recovery mounts must be read-only.");
1333 if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1334 (mp->m_dalign || mp->m_swidth)) {
1336 "sunit and swidth options incompatible with the noalign option");
1340 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1341 xfs_warn(mp, "quota support not available in this kernel.");
1345 if ((mp->m_dalign && !mp->m_swidth) ||
1346 (!mp->m_dalign && mp->m_swidth)) {
1347 xfs_warn(mp, "sunit and swidth must be specified together");
1351 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1353 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1354 mp->m_swidth, mp->m_dalign);
1358 if (mp->m_logbufs != -1 &&
1359 mp->m_logbufs != 0 &&
1360 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1361 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1362 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1363 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1367 if (mp->m_logbsize != -1 &&
1368 mp->m_logbsize != 0 &&
1369 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1370 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1371 !is_power_of_2(mp->m_logbsize))) {
1373 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1378 if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1379 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1380 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1381 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1382 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1391 struct super_block *sb,
1392 struct fs_context *fc)
1394 struct xfs_mount *mp = sb->s_fs_info;
1396 int flags = 0, error;
1400 error = xfs_fc_validate_params(mp);
1402 goto out_free_names;
1404 sb_min_blocksize(sb, BBSIZE);
1405 sb->s_xattr = xfs_xattr_handlers;
1406 sb->s_export_op = &xfs_export_operations;
1407 #ifdef CONFIG_XFS_QUOTA
1408 sb->s_qcop = &xfs_quotactl_operations;
1409 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1411 sb->s_op = &xfs_super_operations;
1414 * Delay mount work if the debug hook is set. This is debug
1415 * instrumention to coordinate simulation of xfs mount failures with
1416 * VFS superblock operations
1418 if (xfs_globals.mount_delay) {
1419 xfs_notice(mp, "Delaying mount for %d seconds.",
1420 xfs_globals.mount_delay);
1421 msleep(xfs_globals.mount_delay * 1000);
1424 if (fc->sb_flags & SB_SILENT)
1425 flags |= XFS_MFSI_QUIET;
1427 error = xfs_open_devices(mp);
1429 goto out_free_names;
1431 error = xfs_init_mount_workqueues(mp);
1433 goto out_close_devices;
1435 error = xfs_init_percpu_counters(mp);
1437 goto out_destroy_workqueues;
1439 /* Allocate stats memory before we do operations that might use it */
1440 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1441 if (!mp->m_stats.xs_stats) {
1443 goto out_destroy_counters;
1446 error = xfs_readsb(mp, flags);
1448 goto out_free_stats;
1450 error = xfs_finish_flags(mp);
1454 error = xfs_setup_devices(mp);
1458 /* V4 support is undergoing deprecation. */
1459 if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1460 #ifdef CONFIG_XFS_SUPPORT_V4
1462 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1465 "Deprecated V4 format (crc=0) not supported by kernel.");
1472 * XFS block mappings use 54 bits to store the logical block offset.
1473 * This should suffice to handle the maximum file size that the VFS
1474 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1475 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1476 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1477 * to check this assertion.
1479 * Avoid integer overflow by comparing the maximum bmbt offset to the
1480 * maximum pagecache offset in units of fs blocks.
1482 if (XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE) > XFS_MAX_FILEOFF) {
1484 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1485 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1491 error = xfs_filestream_mount(mp);
1496 * we must configure the block size in the superblock before we run the
1497 * full mount process as the mount process can lookup and cache inodes.
1499 sb->s_magic = XFS_SUPER_MAGIC;
1500 sb->s_blocksize = mp->m_sb.sb_blocksize;
1501 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1502 sb->s_maxbytes = MAX_LFS_FILESIZE;
1503 sb->s_max_links = XFS_MAXLINK;
1504 sb->s_time_gran = 1;
1505 if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1506 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1507 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1509 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1510 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1512 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1513 sb->s_iflags |= SB_I_CGROUPWB;
1515 set_posix_acl_flag(sb);
1517 /* version 5 superblocks support inode version counters. */
1518 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1519 sb->s_flags |= SB_I_VERSION;
1521 if (xfs_sb_version_hasbigtime(&mp->m_sb))
1523 "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1525 if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1526 bool rtdev_is_dax = false, datadev_is_dax;
1529 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1531 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1533 if (mp->m_rtdev_targp)
1534 rtdev_is_dax = bdev_dax_supported(
1535 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1536 if (!rtdev_is_dax && !datadev_is_dax) {
1538 "DAX unsupported by block device. Turning off DAX.");
1539 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1541 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1543 "DAX and reflink cannot be used together!");
1545 goto out_filestream_unmount;
1549 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1550 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1552 if (!blk_queue_discard(q)) {
1553 xfs_warn(mp, "mounting with \"discard\" option, but "
1554 "the device does not support discard");
1555 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1559 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1560 if (mp->m_sb.sb_rblocks) {
1562 "reflink not compatible with realtime device!");
1564 goto out_filestream_unmount;
1567 if (xfs_globals.always_cow) {
1568 xfs_info(mp, "using DEBUG-only always_cow mode.");
1569 mp->m_always_cow = true;
1573 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1575 "reverse mapping btree not compatible with realtime device!");
1577 goto out_filestream_unmount;
1580 if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1582 "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1584 error = xfs_mountfs(mp);
1586 goto out_filestream_unmount;
1588 root = igrab(VFS_I(mp->m_rootip));
1593 sb->s_root = d_make_root(root);
1601 out_filestream_unmount:
1602 xfs_filestream_unmount(mp);
1606 free_percpu(mp->m_stats.xs_stats);
1607 out_destroy_counters:
1608 xfs_destroy_percpu_counters(mp);
1609 out_destroy_workqueues:
1610 xfs_destroy_mount_workqueues(mp);
1612 xfs_close_devices(mp);
1614 sb->s_fs_info = NULL;
1619 xfs_filestream_unmount(mp);
1626 struct fs_context *fc)
1628 return get_tree_bdev(fc, xfs_fc_fill_super);
1633 struct xfs_mount *mp)
1635 struct xfs_sb *sbp = &mp->m_sb;
1638 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1640 "ro->rw transition prohibited on norecovery mount");
1644 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1645 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1647 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1648 (sbp->sb_features_ro_compat &
1649 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1653 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1656 * If this is the first remount to writeable state we might have some
1657 * superblock changes to update.
1659 if (mp->m_update_sb) {
1660 error = xfs_sync_sb(mp, false);
1662 xfs_warn(mp, "failed to write sb changes");
1665 mp->m_update_sb = false;
1669 * Fill out the reserve pool if it is empty. Use the stashed value if
1670 * it is non-zero, otherwise go with the default.
1672 xfs_restore_resvblks(mp);
1673 xfs_log_work_queue(mp);
1675 /* Recover any CoW blocks that never got remapped. */
1676 error = xfs_reflink_recover_cow(mp);
1679 "Error %d recovering leftover CoW allocations.", error);
1680 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1683 xfs_start_block_reaping(mp);
1685 /* Create the per-AG metadata reservation pool .*/
1686 error = xfs_fs_reserve_ag_blocks(mp);
1687 if (error && error != -ENOSPC)
1695 struct xfs_mount *mp)
1700 * Cancel background eofb scanning so it cannot race with the final
1701 * log force+buftarg wait and deadlock the remount.
1703 xfs_stop_block_reaping(mp);
1705 /* Get rid of any leftover CoW reservations... */
1706 error = xfs_icache_free_cowblocks(mp, NULL);
1708 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1712 /* Free the per-AG metadata reservation pool. */
1713 error = xfs_fs_unreserve_ag_blocks(mp);
1715 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1720 * Before we sync the metadata, we need to free up the reserve block
1721 * pool so that the used block count in the superblock on disk is
1722 * correct at the end of the remount. Stash the current* reserve pool
1723 * size so that if we get remounted rw, we can return it to the same
1726 xfs_save_resvblks(mp);
1728 xfs_quiesce_attr(mp);
1729 mp->m_flags |= XFS_MOUNT_RDONLY;
1735 * Logically we would return an error here to prevent users from believing
1736 * they might have changed mount options using remount which can't be changed.
1738 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1739 * arguments in some cases so we can't blindly reject options, but have to
1740 * check for each specified option if it actually differs from the currently
1741 * set option and only reject it if that's the case.
1743 * Until that is implemented we return success for every remount request, and
1744 * silently ignore all options that we can't actually change.
1748 struct fs_context *fc)
1750 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1751 struct xfs_mount *new_mp = fc->s_fs_info;
1752 xfs_sb_t *sbp = &mp->m_sb;
1753 int flags = fc->sb_flags;
1756 /* version 5 superblocks always support version counters. */
1757 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1758 fc->sb_flags |= SB_I_VERSION;
1760 error = xfs_fc_validate_params(new_mp);
1764 sync_filesystem(mp->m_super);
1766 /* inode32 -> inode64 */
1767 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1768 !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1769 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1770 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1773 /* inode64 -> inode32 */
1774 if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1775 (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1776 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1777 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1781 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1782 error = xfs_remount_rw(mp);
1788 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1789 error = xfs_remount_ro(mp);
1797 static void xfs_fc_free(
1798 struct fs_context *fc)
1800 struct xfs_mount *mp = fc->s_fs_info;
1803 * mp is stored in the fs_context when it is initialized.
1804 * mp is transferred to the superblock on a successful mount,
1805 * but if an error occurs before the transfer we have to free
1812 static const struct fs_context_operations xfs_context_ops = {
1813 .parse_param = xfs_fc_parse_param,
1814 .get_tree = xfs_fc_get_tree,
1815 .reconfigure = xfs_fc_reconfigure,
1816 .free = xfs_fc_free,
1819 static int xfs_init_fs_context(
1820 struct fs_context *fc)
1822 struct xfs_mount *mp;
1824 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1828 spin_lock_init(&mp->m_sb_lock);
1829 spin_lock_init(&mp->m_agirotor_lock);
1830 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1831 spin_lock_init(&mp->m_perag_lock);
1832 mutex_init(&mp->m_growlock);
1833 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1834 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1835 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1836 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1837 mp->m_kobj.kobject.kset = xfs_kset;
1839 * We don't create the finobt per-ag space reservation until after log
1840 * recovery, so we must set this to true so that an ifree transaction
1841 * started during log recovery will not depend on space reservations
1842 * for finobt expansion.
1844 mp->m_finobt_nores = true;
1847 * These can be overridden by the mount option parsing.
1850 mp->m_logbsize = -1;
1851 mp->m_allocsize_log = 16; /* 64k */
1854 * Copy binary VFS mount flags we are interested in.
1856 if (fc->sb_flags & SB_RDONLY)
1857 mp->m_flags |= XFS_MOUNT_RDONLY;
1858 if (fc->sb_flags & SB_DIRSYNC)
1859 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1860 if (fc->sb_flags & SB_SYNCHRONOUS)
1861 mp->m_flags |= XFS_MOUNT_WSYNC;
1864 fc->ops = &xfs_context_ops;
1869 static struct file_system_type xfs_fs_type = {
1870 .owner = THIS_MODULE,
1872 .init_fs_context = xfs_init_fs_context,
1873 .parameters = xfs_fs_parameters,
1874 .kill_sb = kill_block_super,
1875 .fs_flags = FS_REQUIRES_DEV,
1877 MODULE_ALIAS_FS("xfs");
1880 xfs_init_zones(void)
1882 xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1883 sizeof(struct xlog_ticket),
1885 if (!xfs_log_ticket_zone)
1888 xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1889 sizeof(struct xfs_extent_free_item),
1891 if (!xfs_bmap_free_item_zone)
1892 goto out_destroy_log_ticket_zone;
1894 xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1895 sizeof(struct xfs_btree_cur),
1897 if (!xfs_btree_cur_zone)
1898 goto out_destroy_bmap_free_item_zone;
1900 xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1901 sizeof(struct xfs_da_state),
1903 if (!xfs_da_state_zone)
1904 goto out_destroy_btree_cur_zone;
1906 xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1907 sizeof(struct xfs_ifork),
1909 if (!xfs_ifork_zone)
1910 goto out_destroy_da_state_zone;
1912 xfs_trans_zone = kmem_cache_create("xf_trans",
1913 sizeof(struct xfs_trans),
1915 if (!xfs_trans_zone)
1916 goto out_destroy_ifork_zone;
1920 * The size of the zone allocated buf log item is the maximum
1921 * size possible under XFS. This wastes a little bit of memory,
1922 * but it is much faster.
1924 xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1925 sizeof(struct xfs_buf_log_item),
1927 if (!xfs_buf_item_zone)
1928 goto out_destroy_trans_zone;
1930 xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1931 (sizeof(struct xfs_efd_log_item) +
1932 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1933 sizeof(struct xfs_extent)),
1936 goto out_destroy_buf_item_zone;
1938 xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1939 (sizeof(struct xfs_efi_log_item) +
1940 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1941 sizeof(struct xfs_extent)),
1944 goto out_destroy_efd_zone;
1946 xfs_inode_zone = kmem_cache_create("xfs_inode",
1947 sizeof(struct xfs_inode), 0,
1948 (SLAB_HWCACHE_ALIGN |
1949 SLAB_RECLAIM_ACCOUNT |
1950 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1951 xfs_fs_inode_init_once);
1952 if (!xfs_inode_zone)
1953 goto out_destroy_efi_zone;
1955 xfs_ili_zone = kmem_cache_create("xfs_ili",
1956 sizeof(struct xfs_inode_log_item), 0,
1957 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1960 goto out_destroy_inode_zone;
1962 xfs_icreate_zone = kmem_cache_create("xfs_icr",
1963 sizeof(struct xfs_icreate_item),
1965 if (!xfs_icreate_zone)
1966 goto out_destroy_ili_zone;
1968 xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1969 sizeof(struct xfs_rud_log_item),
1972 goto out_destroy_icreate_zone;
1974 xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1975 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1978 goto out_destroy_rud_zone;
1980 xfs_cud_zone = kmem_cache_create("xfs_cud_item",
1981 sizeof(struct xfs_cud_log_item),
1984 goto out_destroy_rui_zone;
1986 xfs_cui_zone = kmem_cache_create("xfs_cui_item",
1987 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1990 goto out_destroy_cud_zone;
1992 xfs_bud_zone = kmem_cache_create("xfs_bud_item",
1993 sizeof(struct xfs_bud_log_item),
1996 goto out_destroy_cui_zone;
1998 xfs_bui_zone = kmem_cache_create("xfs_bui_item",
1999 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2002 goto out_destroy_bud_zone;
2006 out_destroy_bud_zone:
2007 kmem_cache_destroy(xfs_bud_zone);
2008 out_destroy_cui_zone:
2009 kmem_cache_destroy(xfs_cui_zone);
2010 out_destroy_cud_zone:
2011 kmem_cache_destroy(xfs_cud_zone);
2012 out_destroy_rui_zone:
2013 kmem_cache_destroy(xfs_rui_zone);
2014 out_destroy_rud_zone:
2015 kmem_cache_destroy(xfs_rud_zone);
2016 out_destroy_icreate_zone:
2017 kmem_cache_destroy(xfs_icreate_zone);
2018 out_destroy_ili_zone:
2019 kmem_cache_destroy(xfs_ili_zone);
2020 out_destroy_inode_zone:
2021 kmem_cache_destroy(xfs_inode_zone);
2022 out_destroy_efi_zone:
2023 kmem_cache_destroy(xfs_efi_zone);
2024 out_destroy_efd_zone:
2025 kmem_cache_destroy(xfs_efd_zone);
2026 out_destroy_buf_item_zone:
2027 kmem_cache_destroy(xfs_buf_item_zone);
2028 out_destroy_trans_zone:
2029 kmem_cache_destroy(xfs_trans_zone);
2030 out_destroy_ifork_zone:
2031 kmem_cache_destroy(xfs_ifork_zone);
2032 out_destroy_da_state_zone:
2033 kmem_cache_destroy(xfs_da_state_zone);
2034 out_destroy_btree_cur_zone:
2035 kmem_cache_destroy(xfs_btree_cur_zone);
2036 out_destroy_bmap_free_item_zone:
2037 kmem_cache_destroy(xfs_bmap_free_item_zone);
2038 out_destroy_log_ticket_zone:
2039 kmem_cache_destroy(xfs_log_ticket_zone);
2045 xfs_destroy_zones(void)
2048 * Make sure all delayed rcu free are flushed before we
2052 kmem_cache_destroy(xfs_bui_zone);
2053 kmem_cache_destroy(xfs_bud_zone);
2054 kmem_cache_destroy(xfs_cui_zone);
2055 kmem_cache_destroy(xfs_cud_zone);
2056 kmem_cache_destroy(xfs_rui_zone);
2057 kmem_cache_destroy(xfs_rud_zone);
2058 kmem_cache_destroy(xfs_icreate_zone);
2059 kmem_cache_destroy(xfs_ili_zone);
2060 kmem_cache_destroy(xfs_inode_zone);
2061 kmem_cache_destroy(xfs_efi_zone);
2062 kmem_cache_destroy(xfs_efd_zone);
2063 kmem_cache_destroy(xfs_buf_item_zone);
2064 kmem_cache_destroy(xfs_trans_zone);
2065 kmem_cache_destroy(xfs_ifork_zone);
2066 kmem_cache_destroy(xfs_da_state_zone);
2067 kmem_cache_destroy(xfs_btree_cur_zone);
2068 kmem_cache_destroy(xfs_bmap_free_item_zone);
2069 kmem_cache_destroy(xfs_log_ticket_zone);
2073 xfs_init_workqueues(void)
2076 * The allocation workqueue can be used in memory reclaim situations
2077 * (writepage path), and parallelism is only limited by the number of
2078 * AGs in all the filesystems mounted. Hence use the default large
2079 * max_active value for this workqueue.
2081 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2082 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2086 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2087 if (!xfs_discard_wq)
2088 goto out_free_alloc_wq;
2092 destroy_workqueue(xfs_alloc_wq);
2097 xfs_destroy_workqueues(void)
2099 destroy_workqueue(xfs_discard_wq);
2100 destroy_workqueue(xfs_alloc_wq);
2108 xfs_check_ondisk_structs();
2110 printk(KERN_INFO XFS_VERSION_STRING " with "
2111 XFS_BUILD_OPTIONS " enabled\n");
2115 error = xfs_init_zones();
2119 error = xfs_init_workqueues();
2121 goto out_destroy_zones;
2123 error = xfs_mru_cache_init();
2125 goto out_destroy_wq;
2127 error = xfs_buf_init();
2129 goto out_mru_cache_uninit;
2131 error = xfs_init_procfs();
2133 goto out_buf_terminate;
2135 error = xfs_sysctl_register();
2137 goto out_cleanup_procfs;
2139 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2142 goto out_sysctl_unregister;
2145 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2147 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2148 if (!xfsstats.xs_stats) {
2150 goto out_kset_unregister;
2153 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2156 goto out_free_stats;
2159 xfs_dbg_kobj.kobject.kset = xfs_kset;
2160 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2162 goto out_remove_stats_kobj;
2165 error = xfs_qm_init();
2167 goto out_remove_dbg_kobj;
2169 error = register_filesystem(&xfs_fs_type);
2176 out_remove_dbg_kobj:
2178 xfs_sysfs_del(&xfs_dbg_kobj);
2179 out_remove_stats_kobj:
2181 xfs_sysfs_del(&xfsstats.xs_kobj);
2183 free_percpu(xfsstats.xs_stats);
2184 out_kset_unregister:
2185 kset_unregister(xfs_kset);
2186 out_sysctl_unregister:
2187 xfs_sysctl_unregister();
2189 xfs_cleanup_procfs();
2191 xfs_buf_terminate();
2192 out_mru_cache_uninit:
2193 xfs_mru_cache_uninit();
2195 xfs_destroy_workqueues();
2197 xfs_destroy_zones();
2206 unregister_filesystem(&xfs_fs_type);
2208 xfs_sysfs_del(&xfs_dbg_kobj);
2210 xfs_sysfs_del(&xfsstats.xs_kobj);
2211 free_percpu(xfsstats.xs_stats);
2212 kset_unregister(xfs_kset);
2213 xfs_sysctl_unregister();
2214 xfs_cleanup_procfs();
2215 xfs_buf_terminate();
2216 xfs_mru_cache_uninit();
2217 xfs_destroy_workqueues();
2218 xfs_destroy_zones();
2219 xfs_uuid_table_free();
2222 module_init(init_xfs_fs);
2223 module_exit(exit_xfs_fs);
2225 MODULE_AUTHOR("Silicon Graphics, Inc.");
2226 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2227 MODULE_LICENSE("GPL");