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 */
51 * Table driven mount option parser.
54 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
55 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
56 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
57 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
58 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
59 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
60 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
61 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
62 Opt_discard, Opt_nodiscard, Opt_dax,
65 static const struct fs_parameter_spec xfs_fs_parameters[] = {
66 fsparam_u32("logbufs", Opt_logbufs),
67 fsparam_string("logbsize", Opt_logbsize),
68 fsparam_string("logdev", Opt_logdev),
69 fsparam_string("rtdev", Opt_rtdev),
70 fsparam_flag("wsync", Opt_wsync),
71 fsparam_flag("noalign", Opt_noalign),
72 fsparam_flag("swalloc", Opt_swalloc),
73 fsparam_u32("sunit", Opt_sunit),
74 fsparam_u32("swidth", Opt_swidth),
75 fsparam_flag("nouuid", Opt_nouuid),
76 fsparam_flag("grpid", Opt_grpid),
77 fsparam_flag("nogrpid", Opt_nogrpid),
78 fsparam_flag("bsdgroups", Opt_bsdgroups),
79 fsparam_flag("sysvgroups", Opt_sysvgroups),
80 fsparam_string("allocsize", Opt_allocsize),
81 fsparam_flag("norecovery", Opt_norecovery),
82 fsparam_flag("inode64", Opt_inode64),
83 fsparam_flag("inode32", Opt_inode32),
84 fsparam_flag("ikeep", Opt_ikeep),
85 fsparam_flag("noikeep", Opt_noikeep),
86 fsparam_flag("largeio", Opt_largeio),
87 fsparam_flag("nolargeio", Opt_nolargeio),
88 fsparam_flag("attr2", Opt_attr2),
89 fsparam_flag("noattr2", Opt_noattr2),
90 fsparam_flag("filestreams", Opt_filestreams),
91 fsparam_flag("quota", Opt_quota),
92 fsparam_flag("noquota", Opt_noquota),
93 fsparam_flag("usrquota", Opt_usrquota),
94 fsparam_flag("grpquota", Opt_grpquota),
95 fsparam_flag("prjquota", Opt_prjquota),
96 fsparam_flag("uquota", Opt_uquota),
97 fsparam_flag("gquota", Opt_gquota),
98 fsparam_flag("pquota", Opt_pquota),
99 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
100 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
101 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
102 fsparam_flag("qnoenforce", Opt_qnoenforce),
103 fsparam_flag("discard", Opt_discard),
104 fsparam_flag("nodiscard", Opt_nodiscard),
105 fsparam_flag("dax", Opt_dax),
109 struct proc_xfs_info {
119 static struct proc_xfs_info xfs_info_set[] = {
120 /* the few simple ones we can get from the mount struct */
121 { XFS_MOUNT_IKEEP, ",ikeep" },
122 { XFS_MOUNT_WSYNC, ",wsync" },
123 { XFS_MOUNT_NOALIGN, ",noalign" },
124 { XFS_MOUNT_SWALLOC, ",swalloc" },
125 { XFS_MOUNT_NOUUID, ",nouuid" },
126 { XFS_MOUNT_NORECOVERY, ",norecovery" },
127 { XFS_MOUNT_ATTR2, ",attr2" },
128 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
129 { XFS_MOUNT_GRPID, ",grpid" },
130 { XFS_MOUNT_DISCARD, ",discard" },
131 { XFS_MOUNT_LARGEIO, ",largeio" },
132 { XFS_MOUNT_DAX, ",dax" },
135 struct xfs_mount *mp = XFS_M(root->d_sb);
136 struct proc_xfs_info *xfs_infop;
138 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
139 if (mp->m_flags & xfs_infop->flag)
140 seq_puts(m, xfs_infop->str);
143 seq_printf(m, ",inode%d",
144 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
146 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
147 seq_printf(m, ",allocsize=%dk",
148 (1 << mp->m_allocsize_log) >> 10);
150 if (mp->m_logbufs > 0)
151 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
152 if (mp->m_logbsize > 0)
153 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
156 seq_show_option(m, "logdev", mp->m_logname);
158 seq_show_option(m, "rtdev", mp->m_rtname);
160 if (mp->m_dalign > 0)
161 seq_printf(m, ",sunit=%d",
162 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
163 if (mp->m_swidth > 0)
164 seq_printf(m, ",swidth=%d",
165 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
167 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
168 seq_puts(m, ",usrquota");
169 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
170 seq_puts(m, ",uqnoenforce");
172 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
173 if (mp->m_qflags & XFS_PQUOTA_ENFD)
174 seq_puts(m, ",prjquota");
176 seq_puts(m, ",pqnoenforce");
178 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
179 if (mp->m_qflags & XFS_GQUOTA_ENFD)
180 seq_puts(m, ",grpquota");
182 seq_puts(m, ",gqnoenforce");
185 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
186 seq_puts(m, ",noquota");
192 * Set parameters for inode allocation heuristics, taking into account
193 * filesystem size and inode32/inode64 mount options; i.e. specifically
194 * whether or not XFS_MOUNT_SMALL_INUMS is set.
196 * Inode allocation patterns are altered only if inode32 is requested
197 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
198 * If altered, XFS_MOUNT_32BITINODES is set as well.
200 * An agcount independent of that in the mount structure is provided
201 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
202 * to the potentially higher ag count.
204 * Returns the maximum AG index which may contain inodes.
208 struct xfs_mount *mp,
209 xfs_agnumber_t agcount)
211 xfs_agnumber_t index;
212 xfs_agnumber_t maxagi = 0;
213 xfs_sb_t *sbp = &mp->m_sb;
214 xfs_agnumber_t max_metadata;
219 * Calculate how much should be reserved for inodes to meet
220 * the max inode percentage. Used only for inode32.
222 if (M_IGEO(mp)->maxicount) {
225 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
227 icount += sbp->sb_agblocks - 1;
228 do_div(icount, sbp->sb_agblocks);
229 max_metadata = icount;
231 max_metadata = agcount;
234 /* Get the last possible inode in the filesystem */
235 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
236 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
239 * If user asked for no more than 32-bit inodes, and the fs is
240 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
241 * the allocator to accommodate the request.
243 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
244 mp->m_flags |= XFS_MOUNT_32BITINODES;
246 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
248 for (index = 0; index < agcount; index++) {
249 struct xfs_perag *pag;
251 ino = XFS_AGINO_TO_INO(mp, index, agino);
253 pag = xfs_perag_get(mp, index);
255 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
256 if (ino > XFS_MAXINUMBER_32) {
257 pag->pagi_inodeok = 0;
258 pag->pagf_metadata = 0;
260 pag->pagi_inodeok = 1;
262 if (index < max_metadata)
263 pag->pagf_metadata = 1;
265 pag->pagf_metadata = 0;
268 pag->pagi_inodeok = 1;
269 pag->pagf_metadata = 0;
275 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
282 struct block_device **bdevp)
286 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
288 if (IS_ERR(*bdevp)) {
289 error = PTR_ERR(*bdevp);
290 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
298 struct block_device *bdev)
301 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
305 xfs_blkdev_issue_flush(
306 xfs_buftarg_t *buftarg)
308 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
313 struct xfs_mount *mp)
315 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
317 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
318 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
319 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
321 xfs_free_buftarg(mp->m_logdev_targp);
322 xfs_blkdev_put(logdev);
323 fs_put_dax(dax_logdev);
325 if (mp->m_rtdev_targp) {
326 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
327 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
329 xfs_free_buftarg(mp->m_rtdev_targp);
330 xfs_blkdev_put(rtdev);
331 fs_put_dax(dax_rtdev);
333 xfs_free_buftarg(mp->m_ddev_targp);
334 fs_put_dax(dax_ddev);
338 * The file system configurations are:
339 * (1) device (partition) with data and internal log
340 * (2) logical volume with data and log subvolumes.
341 * (3) logical volume with data, log, and realtime subvolumes.
343 * We only have to handle opening the log and realtime volumes here if
344 * they are present. The data subvolume has already been opened by
345 * get_sb_bdev() and is stored in sb->s_bdev.
349 struct xfs_mount *mp)
351 struct block_device *ddev = mp->m_super->s_bdev;
352 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
353 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
354 struct block_device *logdev = NULL, *rtdev = NULL;
358 * Open real time and log devices - order is important.
361 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
364 dax_logdev = fs_dax_get_by_bdev(logdev);
368 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
370 goto out_close_logdev;
372 if (rtdev == ddev || rtdev == logdev) {
374 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
376 goto out_close_rtdev;
378 dax_rtdev = fs_dax_get_by_bdev(rtdev);
382 * Setup xfs_mount buffer target pointers
385 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
386 if (!mp->m_ddev_targp)
387 goto out_close_rtdev;
390 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
391 if (!mp->m_rtdev_targp)
392 goto out_free_ddev_targ;
395 if (logdev && logdev != ddev) {
396 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
397 if (!mp->m_logdev_targp)
398 goto out_free_rtdev_targ;
400 mp->m_logdev_targp = mp->m_ddev_targp;
406 if (mp->m_rtdev_targp)
407 xfs_free_buftarg(mp->m_rtdev_targp);
409 xfs_free_buftarg(mp->m_ddev_targp);
411 xfs_blkdev_put(rtdev);
412 fs_put_dax(dax_rtdev);
414 if (logdev && logdev != ddev) {
415 xfs_blkdev_put(logdev);
416 fs_put_dax(dax_logdev);
419 fs_put_dax(dax_ddev);
424 * Setup xfs_mount buffer target pointers based on superblock
428 struct xfs_mount *mp)
432 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
436 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
437 unsigned int log_sector_size = BBSIZE;
439 if (xfs_sb_version_hassector(&mp->m_sb))
440 log_sector_size = mp->m_sb.sb_logsectsize;
441 error = xfs_setsize_buftarg(mp->m_logdev_targp,
446 if (mp->m_rtdev_targp) {
447 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
448 mp->m_sb.sb_sectsize);
457 xfs_init_mount_workqueues(
458 struct xfs_mount *mp)
460 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
461 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_super->s_id);
462 if (!mp->m_buf_workqueue)
465 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
466 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
467 if (!mp->m_unwritten_workqueue)
468 goto out_destroy_buf;
470 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
471 WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
472 0, mp->m_super->s_id);
473 if (!mp->m_cil_workqueue)
474 goto out_destroy_unwritten;
476 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
477 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
478 if (!mp->m_reclaim_workqueue)
479 goto out_destroy_cil;
481 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
482 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
483 if (!mp->m_eofblocks_workqueue)
484 goto out_destroy_reclaim;
486 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
488 if (!mp->m_sync_workqueue)
489 goto out_destroy_eofb;
494 destroy_workqueue(mp->m_eofblocks_workqueue);
496 destroy_workqueue(mp->m_reclaim_workqueue);
498 destroy_workqueue(mp->m_cil_workqueue);
499 out_destroy_unwritten:
500 destroy_workqueue(mp->m_unwritten_workqueue);
502 destroy_workqueue(mp->m_buf_workqueue);
508 xfs_destroy_mount_workqueues(
509 struct xfs_mount *mp)
511 destroy_workqueue(mp->m_sync_workqueue);
512 destroy_workqueue(mp->m_eofblocks_workqueue);
513 destroy_workqueue(mp->m_reclaim_workqueue);
514 destroy_workqueue(mp->m_cil_workqueue);
515 destroy_workqueue(mp->m_unwritten_workqueue);
516 destroy_workqueue(mp->m_buf_workqueue);
520 xfs_flush_inodes_worker(
521 struct work_struct *work)
523 struct xfs_mount *mp = container_of(work, struct xfs_mount,
524 m_flush_inodes_work);
525 struct super_block *sb = mp->m_super;
527 if (down_read_trylock(&sb->s_umount)) {
529 up_read(&sb->s_umount);
534 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
535 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
536 * for IO to complete so that we effectively throttle multiple callers to the
537 * rate at which IO is completing.
541 struct xfs_mount *mp)
544 * If flush_work() returns true then that means we waited for a flush
545 * which was already in progress. Don't bother running another scan.
547 if (flush_work(&mp->m_flush_inodes_work))
550 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
551 flush_work(&mp->m_flush_inodes_work);
554 /* Catch misguided souls that try to use this interface on XFS */
555 STATIC struct inode *
557 struct super_block *sb)
566 struct xfs_inode *ip,
569 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
570 struct xfs_bmbt_irec got;
571 struct xfs_iext_cursor icur;
573 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
576 if (isnullstartblock(got.br_startblock)) {
577 xfs_warn(ip->i_mount,
578 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
580 whichfork == XFS_DATA_FORK ? "data" : "cow",
581 got.br_startoff, got.br_blockcount);
583 } while (xfs_iext_next_extent(ifp, &icur, &got));
586 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
590 * Now that the generic code is guaranteed not to be accessing
591 * the linux inode, we can inactivate and reclaim the inode.
594 xfs_fs_destroy_inode(
597 struct xfs_inode *ip = XFS_I(inode);
599 trace_xfs_destroy_inode(ip);
601 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
602 XFS_STATS_INC(ip->i_mount, vn_rele);
603 XFS_STATS_INC(ip->i_mount, vn_remove);
607 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
608 xfs_check_delalloc(ip, XFS_DATA_FORK);
609 xfs_check_delalloc(ip, XFS_COW_FORK);
613 XFS_STATS_INC(ip->i_mount, vn_reclaim);
616 * We should never get here with one of the reclaim flags already set.
618 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
619 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
622 * We always use background reclaim here because even if the
623 * inode is clean, it still may be under IO and hence we have
624 * to take the flush lock. The background reclaim path handles
625 * this more efficiently than we can here, so simply let background
626 * reclaim tear down all inodes.
628 xfs_inode_set_reclaim_tag(ip);
636 struct xfs_inode *ip = XFS_I(inode);
637 struct xfs_mount *mp = ip->i_mount;
638 struct xfs_trans *tp;
640 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
642 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
645 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
647 xfs_ilock(ip, XFS_ILOCK_EXCL);
648 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
649 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
650 xfs_trans_commit(tp);
654 * Slab object creation initialisation for the XFS inode.
655 * This covers only the idempotent fields in the XFS inode;
656 * all other fields need to be initialised on allocation
657 * from the slab. This avoids the need to repeatedly initialise
658 * fields in the xfs inode that left in the initialise state
659 * when freeing the inode.
662 xfs_fs_inode_init_once(
665 struct xfs_inode *ip = inode;
667 memset(ip, 0, sizeof(struct xfs_inode));
670 inode_init_once(VFS_I(ip));
673 atomic_set(&ip->i_pincount, 0);
674 spin_lock_init(&ip->i_flags_lock);
676 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
677 "xfsino", ip->i_ino);
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(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
705 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
710 struct xfs_mount *mp)
713 kfree(mp->m_logname);
719 struct super_block *sb,
722 struct xfs_mount *mp = XFS_M(sb);
725 * Doing anything during the async pass would be counterproductive.
730 xfs_log_force(mp, XFS_LOG_SYNC);
733 * The disk must be active because we're syncing.
734 * We schedule log work now (now that the disk is
735 * active) instead of later (when it might not be).
737 flush_delayed_work(&mp->m_log->l_work);
745 struct dentry *dentry,
746 struct kstatfs *statp)
748 struct xfs_mount *mp = XFS_M(dentry->d_sb);
749 xfs_sb_t *sbp = &mp->m_sb;
750 struct xfs_inode *ip = XFS_I(d_inode(dentry));
751 uint64_t fakeinos, id;
758 statp->f_type = XFS_SUPER_MAGIC;
759 statp->f_namelen = MAXNAMELEN - 1;
761 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
762 statp->f_fsid.val[0] = (u32)id;
763 statp->f_fsid.val[1] = (u32)(id >> 32);
765 icount = percpu_counter_sum(&mp->m_icount);
766 ifree = percpu_counter_sum(&mp->m_ifree);
767 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
769 spin_lock(&mp->m_sb_lock);
770 statp->f_bsize = sbp->sb_blocksize;
771 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
772 statp->f_blocks = sbp->sb_dblocks - lsize;
773 spin_unlock(&mp->m_sb_lock);
775 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
776 statp->f_bavail = statp->f_bfree;
778 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
779 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
780 if (M_IGEO(mp)->maxicount)
781 statp->f_files = min_t(typeof(statp->f_files),
783 M_IGEO(mp)->maxicount);
785 /* If sb_icount overshot maxicount, report actual allocation */
786 statp->f_files = max_t(typeof(statp->f_files),
790 /* make sure statp->f_ffree does not underflow */
791 ffree = statp->f_files - (icount - ifree);
792 statp->f_ffree = max_t(int64_t, ffree, 0);
795 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
796 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
797 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
798 xfs_qm_statvfs(ip, statp);
800 if (XFS_IS_REALTIME_MOUNT(mp) &&
801 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
802 statp->f_blocks = sbp->sb_rblocks;
803 statp->f_bavail = statp->f_bfree =
804 sbp->sb_frextents * sbp->sb_rextsize;
811 xfs_save_resvblks(struct xfs_mount *mp)
813 uint64_t resblks = 0;
815 mp->m_resblks_save = mp->m_resblks;
816 xfs_reserve_blocks(mp, &resblks, NULL);
820 xfs_restore_resvblks(struct xfs_mount *mp)
824 if (mp->m_resblks_save) {
825 resblks = mp->m_resblks_save;
826 mp->m_resblks_save = 0;
828 resblks = xfs_default_resblks(mp);
830 xfs_reserve_blocks(mp, &resblks, NULL);
834 * Trigger writeback of all the dirty metadata in the file system.
836 * This ensures that the metadata is written to their location on disk rather
837 * than just existing in transactions in the log. This means after a quiesce
838 * there is no log replay required to write the inodes to disk - this is the
839 * primary difference between a sync and a quiesce.
841 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
842 * it is started again when appropriate.
846 struct xfs_mount *mp)
850 /* wait for all modifications to complete */
851 while (atomic_read(&mp->m_active_trans) > 0)
854 /* force the log to unpin objects from the now complete transactions */
855 xfs_log_force(mp, XFS_LOG_SYNC);
857 /* reclaim inodes to do any IO before the freeze completes */
858 xfs_reclaim_inodes(mp, 0);
859 xfs_reclaim_inodes(mp, SYNC_WAIT);
861 /* Push the superblock and write an unmount record */
862 error = xfs_log_sbcount(mp);
864 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
865 "Frozen image may not be consistent.");
867 * Just warn here till VFS can correctly support
868 * read-only remount without racing.
870 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
876 * Second stage of a freeze. The data is already frozen so we only
877 * need to take care of the metadata. Once that's done sync the superblock
878 * to the log to dirty it in case of a crash while frozen. This ensures that we
879 * will recover the unlinked inode lists on the next mount.
883 struct super_block *sb)
885 struct xfs_mount *mp = XFS_M(sb);
887 xfs_stop_block_reaping(mp);
888 xfs_save_resvblks(mp);
889 xfs_quiesce_attr(mp);
890 return xfs_sync_sb(mp, true);
895 struct super_block *sb)
897 struct xfs_mount *mp = XFS_M(sb);
899 xfs_restore_resvblks(mp);
900 xfs_log_work_queue(mp);
901 xfs_start_block_reaping(mp);
906 * This function fills in xfs_mount_t fields based on mount args.
907 * Note: the superblock _has_ now been read in.
911 struct xfs_mount *mp)
913 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
915 /* Fail a mount where the logbuf is smaller than the log stripe */
916 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
917 if (mp->m_logbsize <= 0 &&
918 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
919 mp->m_logbsize = mp->m_sb.sb_logsunit;
920 } else if (mp->m_logbsize > 0 &&
921 mp->m_logbsize < mp->m_sb.sb_logsunit) {
923 "logbuf size must be greater than or equal to log stripe size");
927 /* Fail a mount if the logbuf is larger than 32K */
928 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
930 "logbuf size for version 1 logs must be 16K or 32K");
936 * V5 filesystems always use attr2 format for attributes.
938 if (xfs_sb_version_hascrc(&mp->m_sb) &&
939 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
940 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
941 "attr2 is always enabled for V5 filesystems.");
946 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
947 * told by noattr2 to turn it off
949 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
950 !(mp->m_flags & XFS_MOUNT_NOATTR2))
951 mp->m_flags |= XFS_MOUNT_ATTR2;
954 * prohibit r/w mounts of read-only filesystems
956 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
958 "cannot mount a read-only filesystem as read-write");
962 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
963 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
964 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
966 "Super block does not support project and group quota together");
974 xfs_init_percpu_counters(
975 struct xfs_mount *mp)
979 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
983 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
987 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
991 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
998 percpu_counter_destroy(&mp->m_fdblocks);
1000 percpu_counter_destroy(&mp->m_ifree);
1002 percpu_counter_destroy(&mp->m_icount);
1007 xfs_reinit_percpu_counters(
1008 struct xfs_mount *mp)
1010 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1011 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1012 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1016 xfs_destroy_percpu_counters(
1017 struct xfs_mount *mp)
1019 percpu_counter_destroy(&mp->m_icount);
1020 percpu_counter_destroy(&mp->m_ifree);
1021 percpu_counter_destroy(&mp->m_fdblocks);
1022 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1023 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1024 percpu_counter_destroy(&mp->m_delalloc_blks);
1029 struct super_block *sb)
1031 struct xfs_mount *mp = XFS_M(sb);
1033 /* if ->fill_super failed, we have no mount to tear down */
1037 xfs_notice(mp, "Unmounting Filesystem");
1038 xfs_filestream_unmount(mp);
1042 free_percpu(mp->m_stats.xs_stats);
1043 xfs_destroy_percpu_counters(mp);
1044 xfs_destroy_mount_workqueues(mp);
1045 xfs_close_devices(mp);
1047 sb->s_fs_info = NULL;
1052 xfs_fs_nr_cached_objects(
1053 struct super_block *sb,
1054 struct shrink_control *sc)
1056 /* Paranoia: catch incorrect calls during mount setup or teardown */
1057 if (WARN_ON_ONCE(!sb->s_fs_info))
1059 return xfs_reclaim_inodes_count(XFS_M(sb));
1063 xfs_fs_free_cached_objects(
1064 struct super_block *sb,
1065 struct shrink_control *sc)
1067 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1070 static const struct super_operations xfs_super_operations = {
1071 .alloc_inode = xfs_fs_alloc_inode,
1072 .destroy_inode = xfs_fs_destroy_inode,
1073 .dirty_inode = xfs_fs_dirty_inode,
1074 .drop_inode = xfs_fs_drop_inode,
1075 .put_super = xfs_fs_put_super,
1076 .sync_fs = xfs_fs_sync_fs,
1077 .freeze_fs = xfs_fs_freeze,
1078 .unfreeze_fs = xfs_fs_unfreeze,
1079 .statfs = xfs_fs_statfs,
1080 .show_options = xfs_fs_show_options,
1081 .nr_cached_objects = xfs_fs_nr_cached_objects,
1082 .free_cached_objects = xfs_fs_free_cached_objects,
1091 int last, shift_left_factor = 0, _res;
1095 value = kstrdup(s, GFP_KERNEL);
1099 last = strlen(value) - 1;
1100 if (value[last] == 'K' || value[last] == 'k') {
1101 shift_left_factor = 10;
1104 if (value[last] == 'M' || value[last] == 'm') {
1105 shift_left_factor = 20;
1108 if (value[last] == 'G' || value[last] == 'g') {
1109 shift_left_factor = 30;
1113 if (kstrtoint(value, base, &_res))
1116 *res = _res << shift_left_factor;
1121 * Set mount state from a mount option.
1123 * NOTE: mp->m_super is NULL here!
1127 struct fs_context *fc,
1128 struct fs_parameter *param)
1130 struct xfs_mount *mp = fc->s_fs_info;
1131 struct fs_parse_result result;
1135 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1141 mp->m_logbufs = result.uint_32;
1144 if (suffix_kstrtoint(param->string, 10, &mp->m_logbsize))
1148 kfree(mp->m_logname);
1149 mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1154 kfree(mp->m_rtname);
1155 mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1160 if (suffix_kstrtoint(param->string, 10, &size))
1162 mp->m_allocsize_log = ffs(size) - 1;
1163 mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1167 mp->m_flags |= XFS_MOUNT_GRPID;
1170 case Opt_sysvgroups:
1171 mp->m_flags &= ~XFS_MOUNT_GRPID;
1174 mp->m_flags |= XFS_MOUNT_WSYNC;
1176 case Opt_norecovery:
1177 mp->m_flags |= XFS_MOUNT_NORECOVERY;
1180 mp->m_flags |= XFS_MOUNT_NOALIGN;
1183 mp->m_flags |= XFS_MOUNT_SWALLOC;
1186 mp->m_dalign = result.uint_32;
1189 mp->m_swidth = result.uint_32;
1192 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1195 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1198 mp->m_flags |= XFS_MOUNT_NOUUID;
1201 mp->m_flags |= XFS_MOUNT_IKEEP;
1204 mp->m_flags &= ~XFS_MOUNT_IKEEP;
1207 mp->m_flags |= XFS_MOUNT_LARGEIO;
1210 mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1213 mp->m_flags |= XFS_MOUNT_ATTR2;
1216 mp->m_flags &= ~XFS_MOUNT_ATTR2;
1217 mp->m_flags |= XFS_MOUNT_NOATTR2;
1219 case Opt_filestreams:
1220 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1223 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1224 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1225 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1230 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1233 case Opt_qnoenforce:
1234 case Opt_uqnoenforce:
1235 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1236 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1240 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1243 case Opt_pqnoenforce:
1244 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1245 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1249 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1252 case Opt_gqnoenforce:
1253 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1254 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1257 mp->m_flags |= XFS_MOUNT_DISCARD;
1260 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1262 #ifdef CONFIG_FS_DAX
1264 mp->m_flags |= XFS_MOUNT_DAX;
1268 xfs_warn(mp, "unknown mount option [%s].", param->key);
1276 xfs_fc_validate_params(
1277 struct xfs_mount *mp)
1280 * no recovery flag requires a read-only mount
1282 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1283 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1284 xfs_warn(mp, "no-recovery mounts must be read-only.");
1288 if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1289 (mp->m_dalign || mp->m_swidth)) {
1291 "sunit and swidth options incompatible with the noalign option");
1295 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1296 xfs_warn(mp, "quota support not available in this kernel.");
1300 if ((mp->m_dalign && !mp->m_swidth) ||
1301 (!mp->m_dalign && mp->m_swidth)) {
1302 xfs_warn(mp, "sunit and swidth must be specified together");
1306 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1308 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1309 mp->m_swidth, mp->m_dalign);
1313 if (mp->m_logbufs != -1 &&
1314 mp->m_logbufs != 0 &&
1315 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1316 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1317 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1318 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1322 if (mp->m_logbsize != -1 &&
1323 mp->m_logbsize != 0 &&
1324 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1325 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1326 !is_power_of_2(mp->m_logbsize))) {
1328 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1333 if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1334 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1335 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1336 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1337 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1346 struct super_block *sb,
1347 struct fs_context *fc)
1349 struct xfs_mount *mp = sb->s_fs_info;
1351 int flags = 0, error;
1355 error = xfs_fc_validate_params(mp);
1357 goto out_free_names;
1359 sb_min_blocksize(sb, BBSIZE);
1360 sb->s_xattr = xfs_xattr_handlers;
1361 sb->s_export_op = &xfs_export_operations;
1362 #ifdef CONFIG_XFS_QUOTA
1363 sb->s_qcop = &xfs_quotactl_operations;
1364 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1366 sb->s_op = &xfs_super_operations;
1369 * Delay mount work if the debug hook is set. This is debug
1370 * instrumention to coordinate simulation of xfs mount failures with
1371 * VFS superblock operations
1373 if (xfs_globals.mount_delay) {
1374 xfs_notice(mp, "Delaying mount for %d seconds.",
1375 xfs_globals.mount_delay);
1376 msleep(xfs_globals.mount_delay * 1000);
1379 if (fc->sb_flags & SB_SILENT)
1380 flags |= XFS_MFSI_QUIET;
1382 error = xfs_open_devices(mp);
1384 goto out_free_names;
1386 error = xfs_init_mount_workqueues(mp);
1388 goto out_close_devices;
1390 error = xfs_init_percpu_counters(mp);
1392 goto out_destroy_workqueues;
1394 /* Allocate stats memory before we do operations that might use it */
1395 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1396 if (!mp->m_stats.xs_stats) {
1398 goto out_destroy_counters;
1401 error = xfs_readsb(mp, flags);
1403 goto out_free_stats;
1405 error = xfs_finish_flags(mp);
1409 error = xfs_setup_devices(mp);
1414 * XFS block mappings use 54 bits to store the logical block offset.
1415 * This should suffice to handle the maximum file size that the VFS
1416 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1417 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1418 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1419 * to check this assertion.
1421 * Avoid integer overflow by comparing the maximum bmbt offset to the
1422 * maximum pagecache offset in units of fs blocks.
1424 if (XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE) > XFS_MAX_FILEOFF) {
1426 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1427 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1433 error = xfs_filestream_mount(mp);
1438 * we must configure the block size in the superblock before we run the
1439 * full mount process as the mount process can lookup and cache inodes.
1441 sb->s_magic = XFS_SUPER_MAGIC;
1442 sb->s_blocksize = mp->m_sb.sb_blocksize;
1443 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1444 sb->s_maxbytes = MAX_LFS_FILESIZE;
1445 sb->s_max_links = XFS_MAXLINK;
1446 sb->s_time_gran = 1;
1447 sb->s_time_min = S32_MIN;
1448 sb->s_time_max = S32_MAX;
1449 sb->s_iflags |= SB_I_CGROUPWB;
1451 set_posix_acl_flag(sb);
1453 /* version 5 superblocks support inode version counters. */
1454 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1455 sb->s_flags |= SB_I_VERSION;
1457 if (mp->m_flags & XFS_MOUNT_DAX) {
1458 bool rtdev_is_dax = false, datadev_is_dax;
1461 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1463 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1465 if (mp->m_rtdev_targp)
1466 rtdev_is_dax = bdev_dax_supported(
1467 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1468 if (!rtdev_is_dax && !datadev_is_dax) {
1470 "DAX unsupported by block device. Turning off DAX.");
1471 mp->m_flags &= ~XFS_MOUNT_DAX;
1473 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1475 "DAX and reflink cannot be used together!");
1477 goto out_filestream_unmount;
1481 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1482 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1484 if (!blk_queue_discard(q)) {
1485 xfs_warn(mp, "mounting with \"discard\" option, but "
1486 "the device does not support discard");
1487 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1491 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1492 if (mp->m_sb.sb_rblocks) {
1494 "reflink not compatible with realtime device!");
1496 goto out_filestream_unmount;
1499 if (xfs_globals.always_cow) {
1500 xfs_info(mp, "using DEBUG-only always_cow mode.");
1501 mp->m_always_cow = true;
1505 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1507 "reverse mapping btree not compatible with realtime device!");
1509 goto out_filestream_unmount;
1512 error = xfs_mountfs(mp);
1514 goto out_filestream_unmount;
1516 root = igrab(VFS_I(mp->m_rootip));
1521 sb->s_root = d_make_root(root);
1529 out_filestream_unmount:
1530 xfs_filestream_unmount(mp);
1534 free_percpu(mp->m_stats.xs_stats);
1535 out_destroy_counters:
1536 xfs_destroy_percpu_counters(mp);
1537 out_destroy_workqueues:
1538 xfs_destroy_mount_workqueues(mp);
1540 xfs_close_devices(mp);
1542 sb->s_fs_info = NULL;
1547 xfs_filestream_unmount(mp);
1554 struct fs_context *fc)
1556 return get_tree_bdev(fc, xfs_fc_fill_super);
1561 struct xfs_mount *mp)
1563 struct xfs_sb *sbp = &mp->m_sb;
1566 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1568 "ro->rw transition prohibited on norecovery mount");
1572 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1573 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1575 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1576 (sbp->sb_features_ro_compat &
1577 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1581 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1584 * If this is the first remount to writeable state we might have some
1585 * superblock changes to update.
1587 if (mp->m_update_sb) {
1588 error = xfs_sync_sb(mp, false);
1590 xfs_warn(mp, "failed to write sb changes");
1593 mp->m_update_sb = false;
1597 * Fill out the reserve pool if it is empty. Use the stashed value if
1598 * it is non-zero, otherwise go with the default.
1600 xfs_restore_resvblks(mp);
1601 xfs_log_work_queue(mp);
1603 /* Recover any CoW blocks that never got remapped. */
1604 error = xfs_reflink_recover_cow(mp);
1607 "Error %d recovering leftover CoW allocations.", error);
1608 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1611 xfs_start_block_reaping(mp);
1613 /* Create the per-AG metadata reservation pool .*/
1614 error = xfs_fs_reserve_ag_blocks(mp);
1615 if (error && error != -ENOSPC)
1623 struct xfs_mount *mp)
1628 * Cancel background eofb scanning so it cannot race with the final
1629 * log force+buftarg wait and deadlock the remount.
1631 xfs_stop_block_reaping(mp);
1633 /* Get rid of any leftover CoW reservations... */
1634 error = xfs_icache_free_cowblocks(mp, NULL);
1636 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1640 /* Free the per-AG metadata reservation pool. */
1641 error = xfs_fs_unreserve_ag_blocks(mp);
1643 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1648 * Before we sync the metadata, we need to free up the reserve block
1649 * pool so that the used block count in the superblock on disk is
1650 * correct at the end of the remount. Stash the current* reserve pool
1651 * size so that if we get remounted rw, we can return it to the same
1654 xfs_save_resvblks(mp);
1656 xfs_quiesce_attr(mp);
1657 mp->m_flags |= XFS_MOUNT_RDONLY;
1663 * Logically we would return an error here to prevent users from believing
1664 * they might have changed mount options using remount which can't be changed.
1666 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1667 * arguments in some cases so we can't blindly reject options, but have to
1668 * check for each specified option if it actually differs from the currently
1669 * set option and only reject it if that's the case.
1671 * Until that is implemented we return success for every remount request, and
1672 * silently ignore all options that we can't actually change.
1676 struct fs_context *fc)
1678 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1679 struct xfs_mount *new_mp = fc->s_fs_info;
1680 xfs_sb_t *sbp = &mp->m_sb;
1681 int flags = fc->sb_flags;
1684 error = xfs_fc_validate_params(new_mp);
1688 sync_filesystem(mp->m_super);
1690 /* inode32 -> inode64 */
1691 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1692 !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1693 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1694 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1697 /* inode64 -> inode32 */
1698 if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1699 (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1700 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1701 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1705 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1706 error = xfs_remount_rw(mp);
1712 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1713 error = xfs_remount_ro(mp);
1721 static void xfs_fc_free(
1722 struct fs_context *fc)
1724 struct xfs_mount *mp = fc->s_fs_info;
1727 * mp is stored in the fs_context when it is initialized.
1728 * mp is transferred to the superblock on a successful mount,
1729 * but if an error occurs before the transfer we have to free
1736 static const struct fs_context_operations xfs_context_ops = {
1737 .parse_param = xfs_fc_parse_param,
1738 .get_tree = xfs_fc_get_tree,
1739 .reconfigure = xfs_fc_reconfigure,
1740 .free = xfs_fc_free,
1743 static int xfs_init_fs_context(
1744 struct fs_context *fc)
1746 struct xfs_mount *mp;
1748 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1752 spin_lock_init(&mp->m_sb_lock);
1753 spin_lock_init(&mp->m_agirotor_lock);
1754 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1755 spin_lock_init(&mp->m_perag_lock);
1756 mutex_init(&mp->m_growlock);
1757 atomic_set(&mp->m_active_trans, 0);
1758 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1759 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1760 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1761 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1762 mp->m_kobj.kobject.kset = xfs_kset;
1764 * We don't create the finobt per-ag space reservation until after log
1765 * recovery, so we must set this to true so that an ifree transaction
1766 * started during log recovery will not depend on space reservations
1767 * for finobt expansion.
1769 mp->m_finobt_nores = true;
1772 * These can be overridden by the mount option parsing.
1775 mp->m_logbsize = -1;
1776 mp->m_allocsize_log = 16; /* 64k */
1779 * Copy binary VFS mount flags we are interested in.
1781 if (fc->sb_flags & SB_RDONLY)
1782 mp->m_flags |= XFS_MOUNT_RDONLY;
1783 if (fc->sb_flags & SB_DIRSYNC)
1784 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1785 if (fc->sb_flags & SB_SYNCHRONOUS)
1786 mp->m_flags |= XFS_MOUNT_WSYNC;
1789 fc->ops = &xfs_context_ops;
1794 static struct file_system_type xfs_fs_type = {
1795 .owner = THIS_MODULE,
1797 .init_fs_context = xfs_init_fs_context,
1798 .parameters = xfs_fs_parameters,
1799 .kill_sb = kill_block_super,
1800 .fs_flags = FS_REQUIRES_DEV,
1802 MODULE_ALIAS_FS("xfs");
1805 xfs_init_zones(void)
1807 xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1808 sizeof(struct xlog_ticket),
1810 if (!xfs_log_ticket_zone)
1813 xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1814 sizeof(struct xfs_extent_free_item),
1816 if (!xfs_bmap_free_item_zone)
1817 goto out_destroy_log_ticket_zone;
1819 xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1820 sizeof(struct xfs_btree_cur),
1822 if (!xfs_btree_cur_zone)
1823 goto out_destroy_bmap_free_item_zone;
1825 xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1826 sizeof(struct xfs_da_state),
1828 if (!xfs_da_state_zone)
1829 goto out_destroy_btree_cur_zone;
1831 xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1832 sizeof(struct xfs_ifork),
1834 if (!xfs_ifork_zone)
1835 goto out_destroy_da_state_zone;
1837 xfs_trans_zone = kmem_cache_create("xf_trans",
1838 sizeof(struct xfs_trans),
1840 if (!xfs_trans_zone)
1841 goto out_destroy_ifork_zone;
1845 * The size of the zone allocated buf log item is the maximum
1846 * size possible under XFS. This wastes a little bit of memory,
1847 * but it is much faster.
1849 xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1850 sizeof(struct xfs_buf_log_item),
1852 if (!xfs_buf_item_zone)
1853 goto out_destroy_trans_zone;
1855 xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1856 (sizeof(struct xfs_efd_log_item) +
1857 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1858 sizeof(struct xfs_extent)),
1861 goto out_destroy_buf_item_zone;
1863 xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1864 (sizeof(struct xfs_efi_log_item) +
1865 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1866 sizeof(struct xfs_extent)),
1869 goto out_destroy_efd_zone;
1871 xfs_inode_zone = kmem_cache_create("xfs_inode",
1872 sizeof(struct xfs_inode), 0,
1873 (SLAB_HWCACHE_ALIGN |
1874 SLAB_RECLAIM_ACCOUNT |
1875 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1876 xfs_fs_inode_init_once);
1877 if (!xfs_inode_zone)
1878 goto out_destroy_efi_zone;
1880 xfs_ili_zone = kmem_cache_create("xfs_ili",
1881 sizeof(struct xfs_inode_log_item), 0,
1882 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1885 goto out_destroy_inode_zone;
1887 xfs_icreate_zone = kmem_cache_create("xfs_icr",
1888 sizeof(struct xfs_icreate_item),
1890 if (!xfs_icreate_zone)
1891 goto out_destroy_ili_zone;
1893 xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1894 sizeof(struct xfs_rud_log_item),
1897 goto out_destroy_icreate_zone;
1899 xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1900 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1903 goto out_destroy_rud_zone;
1905 xfs_cud_zone = kmem_cache_create("xfs_cud_item",
1906 sizeof(struct xfs_cud_log_item),
1909 goto out_destroy_rui_zone;
1911 xfs_cui_zone = kmem_cache_create("xfs_cui_item",
1912 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1915 goto out_destroy_cud_zone;
1917 xfs_bud_zone = kmem_cache_create("xfs_bud_item",
1918 sizeof(struct xfs_bud_log_item),
1921 goto out_destroy_cui_zone;
1923 xfs_bui_zone = kmem_cache_create("xfs_bui_item",
1924 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1927 goto out_destroy_bud_zone;
1931 out_destroy_bud_zone:
1932 kmem_cache_destroy(xfs_bud_zone);
1933 out_destroy_cui_zone:
1934 kmem_cache_destroy(xfs_cui_zone);
1935 out_destroy_cud_zone:
1936 kmem_cache_destroy(xfs_cud_zone);
1937 out_destroy_rui_zone:
1938 kmem_cache_destroy(xfs_rui_zone);
1939 out_destroy_rud_zone:
1940 kmem_cache_destroy(xfs_rud_zone);
1941 out_destroy_icreate_zone:
1942 kmem_cache_destroy(xfs_icreate_zone);
1943 out_destroy_ili_zone:
1944 kmem_cache_destroy(xfs_ili_zone);
1945 out_destroy_inode_zone:
1946 kmem_cache_destroy(xfs_inode_zone);
1947 out_destroy_efi_zone:
1948 kmem_cache_destroy(xfs_efi_zone);
1949 out_destroy_efd_zone:
1950 kmem_cache_destroy(xfs_efd_zone);
1951 out_destroy_buf_item_zone:
1952 kmem_cache_destroy(xfs_buf_item_zone);
1953 out_destroy_trans_zone:
1954 kmem_cache_destroy(xfs_trans_zone);
1955 out_destroy_ifork_zone:
1956 kmem_cache_destroy(xfs_ifork_zone);
1957 out_destroy_da_state_zone:
1958 kmem_cache_destroy(xfs_da_state_zone);
1959 out_destroy_btree_cur_zone:
1960 kmem_cache_destroy(xfs_btree_cur_zone);
1961 out_destroy_bmap_free_item_zone:
1962 kmem_cache_destroy(xfs_bmap_free_item_zone);
1963 out_destroy_log_ticket_zone:
1964 kmem_cache_destroy(xfs_log_ticket_zone);
1970 xfs_destroy_zones(void)
1973 * Make sure all delayed rcu free are flushed before we
1977 kmem_cache_destroy(xfs_bui_zone);
1978 kmem_cache_destroy(xfs_bud_zone);
1979 kmem_cache_destroy(xfs_cui_zone);
1980 kmem_cache_destroy(xfs_cud_zone);
1981 kmem_cache_destroy(xfs_rui_zone);
1982 kmem_cache_destroy(xfs_rud_zone);
1983 kmem_cache_destroy(xfs_icreate_zone);
1984 kmem_cache_destroy(xfs_ili_zone);
1985 kmem_cache_destroy(xfs_inode_zone);
1986 kmem_cache_destroy(xfs_efi_zone);
1987 kmem_cache_destroy(xfs_efd_zone);
1988 kmem_cache_destroy(xfs_buf_item_zone);
1989 kmem_cache_destroy(xfs_trans_zone);
1990 kmem_cache_destroy(xfs_ifork_zone);
1991 kmem_cache_destroy(xfs_da_state_zone);
1992 kmem_cache_destroy(xfs_btree_cur_zone);
1993 kmem_cache_destroy(xfs_bmap_free_item_zone);
1994 kmem_cache_destroy(xfs_log_ticket_zone);
1998 xfs_init_workqueues(void)
2001 * The allocation workqueue can be used in memory reclaim situations
2002 * (writepage path), and parallelism is only limited by the number of
2003 * AGs in all the filesystems mounted. Hence use the default large
2004 * max_active value for this workqueue.
2006 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2007 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2011 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2012 if (!xfs_discard_wq)
2013 goto out_free_alloc_wq;
2017 destroy_workqueue(xfs_alloc_wq);
2022 xfs_destroy_workqueues(void)
2024 destroy_workqueue(xfs_discard_wq);
2025 destroy_workqueue(xfs_alloc_wq);
2033 xfs_check_ondisk_structs();
2035 printk(KERN_INFO XFS_VERSION_STRING " with "
2036 XFS_BUILD_OPTIONS " enabled\n");
2040 error = xfs_init_zones();
2044 error = xfs_init_workqueues();
2046 goto out_destroy_zones;
2048 error = xfs_mru_cache_init();
2050 goto out_destroy_wq;
2052 error = xfs_buf_init();
2054 goto out_mru_cache_uninit;
2056 error = xfs_init_procfs();
2058 goto out_buf_terminate;
2060 error = xfs_sysctl_register();
2062 goto out_cleanup_procfs;
2064 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2067 goto out_sysctl_unregister;
2070 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2072 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2073 if (!xfsstats.xs_stats) {
2075 goto out_kset_unregister;
2078 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2081 goto out_free_stats;
2084 xfs_dbg_kobj.kobject.kset = xfs_kset;
2085 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2087 goto out_remove_stats_kobj;
2090 error = xfs_qm_init();
2092 goto out_remove_dbg_kobj;
2094 error = register_filesystem(&xfs_fs_type);
2101 out_remove_dbg_kobj:
2103 xfs_sysfs_del(&xfs_dbg_kobj);
2104 out_remove_stats_kobj:
2106 xfs_sysfs_del(&xfsstats.xs_kobj);
2108 free_percpu(xfsstats.xs_stats);
2109 out_kset_unregister:
2110 kset_unregister(xfs_kset);
2111 out_sysctl_unregister:
2112 xfs_sysctl_unregister();
2114 xfs_cleanup_procfs();
2116 xfs_buf_terminate();
2117 out_mru_cache_uninit:
2118 xfs_mru_cache_uninit();
2120 xfs_destroy_workqueues();
2122 xfs_destroy_zones();
2131 unregister_filesystem(&xfs_fs_type);
2133 xfs_sysfs_del(&xfs_dbg_kobj);
2135 xfs_sysfs_del(&xfsstats.xs_kobj);
2136 free_percpu(xfsstats.xs_stats);
2137 kset_unregister(xfs_kset);
2138 xfs_sysctl_unregister();
2139 xfs_cleanup_procfs();
2140 xfs_buf_terminate();
2141 xfs_mru_cache_uninit();
2142 xfs_destroy_workqueues();
2143 xfs_destroy_zones();
2144 xfs_uuid_table_free();
2147 module_init(init_xfs_fs);
2148 module_exit(exit_xfs_fs);
2150 MODULE_AUTHOR("Silicon Graphics, Inc.");
2151 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2152 MODULE_LICENSE("GPL");