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_da_format.h"
15 #include "xfs_inode.h"
16 #include "xfs_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_error.h"
20 #include "xfs_fsops.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
24 #include "xfs_log_priv.h"
25 #include "xfs_da_btree.h"
27 #include "xfs_extfree_item.h"
28 #include "xfs_mru_cache.h"
29 #include "xfs_inode_item.h"
30 #include "xfs_icache.h"
31 #include "xfs_trace.h"
32 #include "xfs_icreate_item.h"
33 #include "xfs_filestream.h"
34 #include "xfs_quota.h"
35 #include "xfs_sysfs.h"
36 #include "xfs_ondisk.h"
37 #include "xfs_rmap_item.h"
38 #include "xfs_refcount_item.h"
39 #include "xfs_bmap_item.h"
40 #include "xfs_reflink.h"
42 #include <linux/namei.h>
43 #include <linux/dax.h>
44 #include <linux/init.h>
45 #include <linux/slab.h>
46 #include <linux/mount.h>
47 #include <linux/mempool.h>
48 #include <linux/writeback.h>
49 #include <linux/kthread.h>
50 #include <linux/freezer.h>
51 #include <linux/parser.h>
53 static const struct super_operations xfs_super_operations;
54 struct bio_set xfs_ioend_bioset;
56 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
58 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
62 * Table driven mount option parser.
65 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
66 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
67 Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
68 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
69 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
70 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
71 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
72 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
73 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
76 static const match_table_t tokens = {
77 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */
78 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */
79 {Opt_logdev, "logdev=%s"}, /* log device */
80 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */
81 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */
82 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */
83 {Opt_noalign, "noalign"}, /* turn off stripe alignment */
84 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */
85 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */
86 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */
87 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */
88 {Opt_mtpt, "mtpt"}, /* filesystem mount point */
89 {Opt_grpid, "grpid"}, /* group-ID from parent directory */
90 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */
91 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */
92 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */
93 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
94 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */
95 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */
96 {Opt_inode32, "inode32"}, /* inode allocation limited to
97 * XFS_MAXINUMBER_32 */
98 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */
99 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */
100 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */
101 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes
103 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */
104 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */
105 {Opt_filestreams,"filestreams"},/* use filestreams allocator */
106 {Opt_quota, "quota"}, /* disk quotas (user) */
107 {Opt_noquota, "noquota"}, /* no quotas */
108 {Opt_usrquota, "usrquota"}, /* user quota enabled */
109 {Opt_grpquota, "grpquota"}, /* group quota enabled */
110 {Opt_prjquota, "prjquota"}, /* project quota enabled */
111 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */
112 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */
113 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */
114 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
115 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
116 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
117 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
118 {Opt_discard, "discard"}, /* Discard unused blocks */
119 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */
120 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */
126 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
128 int last, shift_left_factor = 0, _res;
132 value = match_strdup(s);
136 last = strlen(value) - 1;
137 if (value[last] == 'K' || value[last] == 'k') {
138 shift_left_factor = 10;
141 if (value[last] == 'M' || value[last] == 'm') {
142 shift_left_factor = 20;
145 if (value[last] == 'G' || value[last] == 'g') {
146 shift_left_factor = 30;
150 if (kstrtoint(value, base, &_res))
153 *res = _res << shift_left_factor;
158 * This function fills in xfs_mount_t fields based on mount args.
159 * Note: the superblock has _not_ yet been read in.
161 * Note that this function leaks the various device name allocations on
162 * failure. The caller takes care of them.
164 * *sb is const because this is also used to test options on the remount
165 * path, and we don't want this to have any side effects at remount time.
166 * Today this function does not change *sb, but just to future-proof...
170 struct xfs_mount *mp,
173 const struct super_block *sb = mp->m_super;
175 substring_t args[MAX_OPT_ARGS];
179 uint8_t iosizelog = 0;
182 * set up the mount name first so all the errors will refer to the
185 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
188 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
191 * Copy binary VFS mount flags we are interested in.
194 mp->m_flags |= XFS_MOUNT_RDONLY;
195 if (sb->s_flags & SB_DIRSYNC)
196 mp->m_flags |= XFS_MOUNT_DIRSYNC;
197 if (sb->s_flags & SB_SYNCHRONOUS)
198 mp->m_flags |= XFS_MOUNT_WSYNC;
201 * Set some default flags that could be cleared by the mount option
204 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
207 * These can be overridden by the mount option parsing.
215 while ((p = strsep(&options, ",")) != NULL) {
221 token = match_token(p, tokens, args);
224 if (match_int(args, &mp->m_logbufs))
228 if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
232 kfree(mp->m_logname);
233 mp->m_logname = match_strdup(args);
238 xfs_warn(mp, "%s option not allowed on this system", p);
242 mp->m_rtname = match_strdup(args);
248 if (suffix_kstrtoint(args, 10, &iosize))
250 iosizelog = ffs(iosize) - 1;
254 mp->m_flags |= XFS_MOUNT_GRPID;
258 mp->m_flags &= ~XFS_MOUNT_GRPID;
261 mp->m_flags |= XFS_MOUNT_WSYNC;
264 mp->m_flags |= XFS_MOUNT_NORECOVERY;
267 mp->m_flags |= XFS_MOUNT_NOALIGN;
270 mp->m_flags |= XFS_MOUNT_SWALLOC;
273 if (match_int(args, &dsunit))
277 if (match_int(args, &dswidth))
281 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
284 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
287 mp->m_flags |= XFS_MOUNT_NOUUID;
290 mp->m_flags |= XFS_MOUNT_IKEEP;
293 mp->m_flags &= ~XFS_MOUNT_IKEEP;
296 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
299 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
302 mp->m_flags |= XFS_MOUNT_ATTR2;
305 mp->m_flags &= ~XFS_MOUNT_ATTR2;
306 mp->m_flags |= XFS_MOUNT_NOATTR2;
308 case Opt_filestreams:
309 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
312 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
313 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
314 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
319 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
323 case Opt_uqnoenforce:
324 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
325 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
329 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
332 case Opt_pqnoenforce:
333 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
334 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
338 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
341 case Opt_gqnoenforce:
342 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
343 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
346 mp->m_flags |= XFS_MOUNT_DISCARD;
349 mp->m_flags &= ~XFS_MOUNT_DISCARD;
353 mp->m_flags |= XFS_MOUNT_DAX;
357 xfs_warn(mp, "unknown mount option [%s].", p);
363 * no recovery flag requires a read-only mount
365 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
366 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
367 xfs_warn(mp, "no-recovery mounts must be read-only.");
371 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
373 "sunit and swidth options incompatible with the noalign option");
377 #ifndef CONFIG_XFS_QUOTA
378 if (XFS_IS_QUOTA_RUNNING(mp)) {
379 xfs_warn(mp, "quota support not available in this kernel.");
384 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
385 xfs_warn(mp, "sunit and swidth must be specified together");
389 if (dsunit && (dswidth % dsunit != 0)) {
391 "stripe width (%d) must be a multiple of the stripe unit (%d)",
397 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
399 * At this point the superblock has not been read
400 * in, therefore we do not know the block size.
401 * Before the mount call ends we will convert
404 mp->m_dalign = dsunit;
405 mp->m_swidth = dswidth;
408 if (mp->m_logbufs != -1 &&
409 mp->m_logbufs != 0 &&
410 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
411 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
412 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
413 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
416 if (mp->m_logbsize != -1 &&
417 mp->m_logbsize != 0 &&
418 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
419 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
420 !is_power_of_2(mp->m_logbsize))) {
422 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
428 if (iosizelog > XFS_MAX_IO_LOG ||
429 iosizelog < XFS_MIN_IO_LOG) {
430 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
431 iosizelog, XFS_MIN_IO_LOG,
436 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
437 mp->m_readio_log = iosizelog;
438 mp->m_writeio_log = iosizelog;
444 struct proc_xfs_info {
451 struct xfs_mount *mp,
454 static struct proc_xfs_info xfs_info_set[] = {
455 /* the few simple ones we can get from the mount struct */
456 { XFS_MOUNT_IKEEP, ",ikeep" },
457 { XFS_MOUNT_WSYNC, ",wsync" },
458 { XFS_MOUNT_NOALIGN, ",noalign" },
459 { XFS_MOUNT_SWALLOC, ",swalloc" },
460 { XFS_MOUNT_NOUUID, ",nouuid" },
461 { XFS_MOUNT_NORECOVERY, ",norecovery" },
462 { XFS_MOUNT_ATTR2, ",attr2" },
463 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
464 { XFS_MOUNT_GRPID, ",grpid" },
465 { XFS_MOUNT_DISCARD, ",discard" },
466 { XFS_MOUNT_SMALL_INUMS, ",inode32" },
467 { XFS_MOUNT_DAX, ",dax" },
470 static struct proc_xfs_info xfs_info_unset[] = {
471 /* the few simple ones we can get from the mount struct */
472 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" },
473 { XFS_MOUNT_SMALL_INUMS, ",inode64" },
476 struct proc_xfs_info *xfs_infop;
478 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
479 if (mp->m_flags & xfs_infop->flag)
480 seq_puts(m, xfs_infop->str);
482 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
483 if (!(mp->m_flags & xfs_infop->flag))
484 seq_puts(m, xfs_infop->str);
487 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
488 seq_printf(m, ",allocsize=%dk",
489 (int)(1 << mp->m_writeio_log) >> 10);
491 if (mp->m_logbufs > 0)
492 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
493 if (mp->m_logbsize > 0)
494 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
497 seq_show_option(m, "logdev", mp->m_logname);
499 seq_show_option(m, "rtdev", mp->m_rtname);
501 if (mp->m_dalign > 0)
502 seq_printf(m, ",sunit=%d",
503 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
504 if (mp->m_swidth > 0)
505 seq_printf(m, ",swidth=%d",
506 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
508 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
509 seq_puts(m, ",usrquota");
510 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
511 seq_puts(m, ",uqnoenforce");
513 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
514 if (mp->m_qflags & XFS_PQUOTA_ENFD)
515 seq_puts(m, ",prjquota");
517 seq_puts(m, ",pqnoenforce");
519 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
520 if (mp->m_qflags & XFS_GQUOTA_ENFD)
521 seq_puts(m, ",grpquota");
523 seq_puts(m, ",gqnoenforce");
526 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
527 seq_puts(m, ",noquota");
533 unsigned int blockshift)
535 unsigned int pagefactor = 1;
536 unsigned int bitshift = BITS_PER_LONG - 1;
538 /* Figure out maximum filesize, on Linux this can depend on
539 * the filesystem blocksize (on 32 bit platforms).
540 * __block_write_begin does this in an [unsigned] long...
541 * page->index << (PAGE_SHIFT - bbits)
542 * So, for page sized blocks (4K on 32 bit platforms),
543 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
544 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
545 * but for smaller blocksizes it is less (bbits = log2 bsize).
546 * Note1: get_block_t takes a long (implicit cast from above)
547 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
548 * can optionally convert the [unsigned] long from above into
549 * an [unsigned] long long.
552 #if BITS_PER_LONG == 32
553 # if defined(CONFIG_LBDAF)
554 ASSERT(sizeof(sector_t) == 8);
555 pagefactor = PAGE_SIZE;
556 bitshift = BITS_PER_LONG;
558 pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
562 return (((uint64_t)pagefactor) << bitshift) - 1;
566 * Set parameters for inode allocation heuristics, taking into account
567 * filesystem size and inode32/inode64 mount options; i.e. specifically
568 * whether or not XFS_MOUNT_SMALL_INUMS is set.
570 * Inode allocation patterns are altered only if inode32 is requested
571 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
572 * If altered, XFS_MOUNT_32BITINODES is set as well.
574 * An agcount independent of that in the mount structure is provided
575 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
576 * to the potentially higher ag count.
578 * Returns the maximum AG index which may contain inodes.
582 struct xfs_mount *mp,
583 xfs_agnumber_t agcount)
585 xfs_agnumber_t index;
586 xfs_agnumber_t maxagi = 0;
587 xfs_sb_t *sbp = &mp->m_sb;
588 xfs_agnumber_t max_metadata;
593 * Calculate how much should be reserved for inodes to meet
594 * the max inode percentage. Used only for inode32.
596 if (mp->m_maxicount) {
599 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
601 icount += sbp->sb_agblocks - 1;
602 do_div(icount, sbp->sb_agblocks);
603 max_metadata = icount;
605 max_metadata = agcount;
608 /* Get the last possible inode in the filesystem */
609 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
610 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
613 * If user asked for no more than 32-bit inodes, and the fs is
614 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
615 * the allocator to accommodate the request.
617 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
618 mp->m_flags |= XFS_MOUNT_32BITINODES;
620 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
622 for (index = 0; index < agcount; index++) {
623 struct xfs_perag *pag;
625 ino = XFS_AGINO_TO_INO(mp, index, agino);
627 pag = xfs_perag_get(mp, index);
629 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
630 if (ino > XFS_MAXINUMBER_32) {
631 pag->pagi_inodeok = 0;
632 pag->pagf_metadata = 0;
634 pag->pagi_inodeok = 1;
636 if (index < max_metadata)
637 pag->pagf_metadata = 1;
639 pag->pagf_metadata = 0;
642 pag->pagi_inodeok = 1;
643 pag->pagf_metadata = 0;
649 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
656 struct block_device **bdevp)
660 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
662 if (IS_ERR(*bdevp)) {
663 error = PTR_ERR(*bdevp);
664 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
672 struct block_device *bdev)
675 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
679 xfs_blkdev_issue_flush(
680 xfs_buftarg_t *buftarg)
682 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
687 struct xfs_mount *mp)
689 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
691 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
692 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
693 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
695 xfs_free_buftarg(mp->m_logdev_targp);
696 xfs_blkdev_put(logdev);
697 fs_put_dax(dax_logdev);
699 if (mp->m_rtdev_targp) {
700 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
701 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
703 xfs_free_buftarg(mp->m_rtdev_targp);
704 xfs_blkdev_put(rtdev);
705 fs_put_dax(dax_rtdev);
707 xfs_free_buftarg(mp->m_ddev_targp);
708 fs_put_dax(dax_ddev);
712 * The file system configurations are:
713 * (1) device (partition) with data and internal log
714 * (2) logical volume with data and log subvolumes.
715 * (3) logical volume with data, log, and realtime subvolumes.
717 * We only have to handle opening the log and realtime volumes here if
718 * they are present. The data subvolume has already been opened by
719 * get_sb_bdev() and is stored in sb->s_bdev.
723 struct xfs_mount *mp)
725 struct block_device *ddev = mp->m_super->s_bdev;
726 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
727 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
728 struct block_device *logdev = NULL, *rtdev = NULL;
732 * Open real time and log devices - order is important.
735 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
738 dax_logdev = fs_dax_get_by_bdev(logdev);
742 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
744 goto out_close_logdev;
746 if (rtdev == ddev || rtdev == logdev) {
748 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
750 goto out_close_rtdev;
752 dax_rtdev = fs_dax_get_by_bdev(rtdev);
756 * Setup xfs_mount buffer target pointers
759 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
760 if (!mp->m_ddev_targp)
761 goto out_close_rtdev;
764 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
765 if (!mp->m_rtdev_targp)
766 goto out_free_ddev_targ;
769 if (logdev && logdev != ddev) {
770 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
771 if (!mp->m_logdev_targp)
772 goto out_free_rtdev_targ;
774 mp->m_logdev_targp = mp->m_ddev_targp;
780 if (mp->m_rtdev_targp)
781 xfs_free_buftarg(mp->m_rtdev_targp);
783 xfs_free_buftarg(mp->m_ddev_targp);
785 xfs_blkdev_put(rtdev);
786 fs_put_dax(dax_rtdev);
788 if (logdev && logdev != ddev) {
789 xfs_blkdev_put(logdev);
790 fs_put_dax(dax_logdev);
793 fs_put_dax(dax_ddev);
798 * Setup xfs_mount buffer target pointers based on superblock
802 struct xfs_mount *mp)
806 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
810 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
811 unsigned int log_sector_size = BBSIZE;
813 if (xfs_sb_version_hassector(&mp->m_sb))
814 log_sector_size = mp->m_sb.sb_logsectsize;
815 error = xfs_setsize_buftarg(mp->m_logdev_targp,
820 if (mp->m_rtdev_targp) {
821 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
822 mp->m_sb.sb_sectsize);
831 xfs_init_mount_workqueues(
832 struct xfs_mount *mp)
834 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
835 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
836 if (!mp->m_buf_workqueue)
839 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
840 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
841 if (!mp->m_data_workqueue)
842 goto out_destroy_buf;
844 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
845 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
846 if (!mp->m_unwritten_workqueue)
847 goto out_destroy_data_iodone_queue;
849 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
850 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
851 if (!mp->m_cil_workqueue)
852 goto out_destroy_unwritten;
854 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
855 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
856 if (!mp->m_reclaim_workqueue)
857 goto out_destroy_cil;
859 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
860 WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
862 if (!mp->m_log_workqueue)
863 goto out_destroy_reclaim;
865 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
866 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
867 if (!mp->m_eofblocks_workqueue)
868 goto out_destroy_log;
870 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
872 if (!mp->m_sync_workqueue)
873 goto out_destroy_eofb;
878 destroy_workqueue(mp->m_eofblocks_workqueue);
880 destroy_workqueue(mp->m_log_workqueue);
882 destroy_workqueue(mp->m_reclaim_workqueue);
884 destroy_workqueue(mp->m_cil_workqueue);
885 out_destroy_unwritten:
886 destroy_workqueue(mp->m_unwritten_workqueue);
887 out_destroy_data_iodone_queue:
888 destroy_workqueue(mp->m_data_workqueue);
890 destroy_workqueue(mp->m_buf_workqueue);
896 xfs_destroy_mount_workqueues(
897 struct xfs_mount *mp)
899 destroy_workqueue(mp->m_sync_workqueue);
900 destroy_workqueue(mp->m_eofblocks_workqueue);
901 destroy_workqueue(mp->m_log_workqueue);
902 destroy_workqueue(mp->m_reclaim_workqueue);
903 destroy_workqueue(mp->m_cil_workqueue);
904 destroy_workqueue(mp->m_data_workqueue);
905 destroy_workqueue(mp->m_unwritten_workqueue);
906 destroy_workqueue(mp->m_buf_workqueue);
910 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
911 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
912 * for IO to complete so that we effectively throttle multiple callers to the
913 * rate at which IO is completing.
917 struct xfs_mount *mp)
919 struct super_block *sb = mp->m_super;
921 if (down_read_trylock(&sb->s_umount)) {
923 up_read(&sb->s_umount);
927 /* Catch misguided souls that try to use this interface on XFS */
928 STATIC struct inode *
930 struct super_block *sb)
937 * Now that the generic code is guaranteed not to be accessing
938 * the linux inode, we can inactivate and reclaim the inode.
941 xfs_fs_destroy_inode(
944 struct xfs_inode *ip = XFS_I(inode);
946 trace_xfs_destroy_inode(ip);
948 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
949 XFS_STATS_INC(ip->i_mount, vn_rele);
950 XFS_STATS_INC(ip->i_mount, vn_remove);
954 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
955 XFS_STATS_INC(ip->i_mount, vn_reclaim);
958 * We should never get here with one of the reclaim flags already set.
960 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
961 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
964 * We always use background reclaim here because even if the
965 * inode is clean, it still may be under IO and hence we have
966 * to take the flush lock. The background reclaim path handles
967 * this more efficiently than we can here, so simply let background
968 * reclaim tear down all inodes.
970 xfs_inode_set_reclaim_tag(ip);
978 struct xfs_inode *ip = XFS_I(inode);
979 struct xfs_mount *mp = ip->i_mount;
980 struct xfs_trans *tp;
982 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
984 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
987 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
989 xfs_ilock(ip, XFS_ILOCK_EXCL);
990 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
991 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
992 xfs_trans_commit(tp);
996 * Slab object creation initialisation for the XFS inode.
997 * This covers only the idempotent fields in the XFS inode;
998 * all other fields need to be initialised on allocation
999 * from the slab. This avoids the need to repeatedly initialise
1000 * fields in the xfs inode that left in the initialise state
1001 * when freeing the inode.
1004 xfs_fs_inode_init_once(
1007 struct xfs_inode *ip = inode;
1009 memset(ip, 0, sizeof(struct xfs_inode));
1012 inode_init_once(VFS_I(ip));
1015 atomic_set(&ip->i_pincount, 0);
1016 spin_lock_init(&ip->i_flags_lock);
1018 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1019 "xfsino", ip->i_ino);
1020 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1021 "xfsino", ip->i_ino);
1025 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1026 * serialised against cache hits here via the inode->i_lock and igrab() in
1027 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1028 * racing with us, and it avoids needing to grab a spinlock here for every inode
1029 * we drop the final reference on.
1033 struct inode *inode)
1035 struct xfs_inode *ip = XFS_I(inode);
1038 * If this unlinked inode is in the middle of recovery, don't
1039 * drop the inode just yet; log recovery will take care of
1040 * that. See the comment for this inode flag.
1042 if (ip->i_flags & XFS_IRECOVERY) {
1043 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1047 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1052 struct xfs_mount *mp)
1054 kfree(mp->m_fsname);
1055 kfree(mp->m_rtname);
1056 kfree(mp->m_logname);
1061 struct super_block *sb,
1064 struct xfs_mount *mp = XFS_M(sb);
1067 * Doing anything during the async pass would be counterproductive.
1072 xfs_log_force(mp, XFS_LOG_SYNC);
1075 * The disk must be active because we're syncing.
1076 * We schedule log work now (now that the disk is
1077 * active) instead of later (when it might not be).
1079 flush_delayed_work(&mp->m_log->l_work);
1087 struct dentry *dentry,
1088 struct kstatfs *statp)
1090 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1091 xfs_sb_t *sbp = &mp->m_sb;
1092 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1093 uint64_t fakeinos, id;
1100 statp->f_type = XFS_SB_MAGIC;
1101 statp->f_namelen = MAXNAMELEN - 1;
1103 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1104 statp->f_fsid.val[0] = (u32)id;
1105 statp->f_fsid.val[1] = (u32)(id >> 32);
1107 icount = percpu_counter_sum(&mp->m_icount);
1108 ifree = percpu_counter_sum(&mp->m_ifree);
1109 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1111 spin_lock(&mp->m_sb_lock);
1112 statp->f_bsize = sbp->sb_blocksize;
1113 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1114 statp->f_blocks = sbp->sb_dblocks - lsize;
1115 spin_unlock(&mp->m_sb_lock);
1117 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1118 statp->f_bavail = statp->f_bfree;
1120 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1121 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1122 if (mp->m_maxicount)
1123 statp->f_files = min_t(typeof(statp->f_files),
1127 /* If sb_icount overshot maxicount, report actual allocation */
1128 statp->f_files = max_t(typeof(statp->f_files),
1132 /* make sure statp->f_ffree does not underflow */
1133 ffree = statp->f_files - (icount - ifree);
1134 statp->f_ffree = max_t(int64_t, ffree, 0);
1137 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1138 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1139 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1140 xfs_qm_statvfs(ip, statp);
1142 if (XFS_IS_REALTIME_MOUNT(mp) &&
1143 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1144 statp->f_blocks = sbp->sb_rblocks;
1145 statp->f_bavail = statp->f_bfree =
1146 sbp->sb_frextents * sbp->sb_rextsize;
1153 xfs_save_resvblks(struct xfs_mount *mp)
1155 uint64_t resblks = 0;
1157 mp->m_resblks_save = mp->m_resblks;
1158 xfs_reserve_blocks(mp, &resblks, NULL);
1162 xfs_restore_resvblks(struct xfs_mount *mp)
1166 if (mp->m_resblks_save) {
1167 resblks = mp->m_resblks_save;
1168 mp->m_resblks_save = 0;
1170 resblks = xfs_default_resblks(mp);
1172 xfs_reserve_blocks(mp, &resblks, NULL);
1176 * Trigger writeback of all the dirty metadata in the file system.
1178 * This ensures that the metadata is written to their location on disk rather
1179 * than just existing in transactions in the log. This means after a quiesce
1180 * there is no log replay required to write the inodes to disk - this is the
1181 * primary difference between a sync and a quiesce.
1183 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1184 * it is started again when appropriate.
1188 struct xfs_mount *mp)
1192 /* wait for all modifications to complete */
1193 while (atomic_read(&mp->m_active_trans) > 0)
1196 /* force the log to unpin objects from the now complete transactions */
1197 xfs_log_force(mp, XFS_LOG_SYNC);
1199 /* reclaim inodes to do any IO before the freeze completes */
1200 xfs_reclaim_inodes(mp, 0);
1201 xfs_reclaim_inodes(mp, SYNC_WAIT);
1203 /* Push the superblock and write an unmount record */
1204 error = xfs_log_sbcount(mp);
1206 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1207 "Frozen image may not be consistent.");
1209 * Just warn here till VFS can correctly support
1210 * read-only remount without racing.
1212 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1214 xfs_log_quiesce(mp);
1218 xfs_test_remount_options(
1219 struct super_block *sb,
1223 struct xfs_mount *tmp_mp;
1225 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1229 tmp_mp->m_super = sb;
1230 error = xfs_parseargs(tmp_mp, options);
1231 xfs_free_fsname(tmp_mp);
1239 struct super_block *sb,
1243 struct xfs_mount *mp = XFS_M(sb);
1244 xfs_sb_t *sbp = &mp->m_sb;
1245 substring_t args[MAX_OPT_ARGS];
1249 /* First, check for complete junk; i.e. invalid options */
1250 error = xfs_test_remount_options(sb, options);
1254 sync_filesystem(sb);
1255 while ((p = strsep(&options, ",")) != NULL) {
1261 token = match_token(p, tokens, args);
1264 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1265 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1268 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1269 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1273 * Logically we would return an error here to prevent
1274 * users from believing they might have changed
1275 * mount options using remount which can't be changed.
1277 * But unfortunately mount(8) adds all options from
1278 * mtab and fstab to the mount arguments in some cases
1279 * so we can't blindly reject options, but have to
1280 * check for each specified option if it actually
1281 * differs from the currently set option and only
1282 * reject it if that's the case.
1284 * Until that is implemented we return success for
1285 * every remount request, and silently ignore all
1286 * options that we can't actually change.
1290 "mount option \"%s\" not supported for remount", p);
1299 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1300 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1302 "ro->rw transition prohibited on norecovery mount");
1306 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1307 xfs_sb_has_ro_compat_feature(sbp,
1308 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1310 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1311 (sbp->sb_features_ro_compat &
1312 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1316 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1319 * If this is the first remount to writeable state we
1320 * might have some superblock changes to update.
1322 if (mp->m_update_sb) {
1323 error = xfs_sync_sb(mp, false);
1325 xfs_warn(mp, "failed to write sb changes");
1328 mp->m_update_sb = false;
1332 * Fill out the reserve pool if it is empty. Use the stashed
1333 * value if it is non-zero, otherwise go with the default.
1335 xfs_restore_resvblks(mp);
1336 xfs_log_work_queue(mp);
1338 /* Recover any CoW blocks that never got remapped. */
1339 error = xfs_reflink_recover_cow(mp);
1342 "Error %d recovering leftover CoW allocations.", error);
1343 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1346 xfs_icache_enable_reclaim(mp);
1348 /* Create the per-AG metadata reservation pool .*/
1349 error = xfs_fs_reserve_ag_blocks(mp);
1350 if (error && error != -ENOSPC)
1355 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1357 * Cancel background eofb scanning so it cannot race with the
1358 * final log force+buftarg wait and deadlock the remount.
1360 xfs_icache_disable_reclaim(mp);
1362 /* Get rid of any leftover CoW reservations... */
1363 error = xfs_icache_free_cowblocks(mp, NULL);
1365 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1369 /* Free the per-AG metadata reservation pool. */
1370 error = xfs_fs_unreserve_ag_blocks(mp);
1372 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1377 * Before we sync the metadata, we need to free up the reserve
1378 * block pool so that the used block count in the superblock on
1379 * disk is correct at the end of the remount. Stash the current
1380 * reserve pool size so that if we get remounted rw, we can
1381 * return it to the same size.
1383 xfs_save_resvblks(mp);
1385 xfs_quiesce_attr(mp);
1386 mp->m_flags |= XFS_MOUNT_RDONLY;
1393 * Second stage of a freeze. The data is already frozen so we only
1394 * need to take care of the metadata. Once that's done sync the superblock
1395 * to the log to dirty it in case of a crash while frozen. This ensures that we
1396 * will recover the unlinked inode lists on the next mount.
1400 struct super_block *sb)
1402 struct xfs_mount *mp = XFS_M(sb);
1404 xfs_icache_disable_reclaim(mp);
1405 xfs_save_resvblks(mp);
1406 xfs_quiesce_attr(mp);
1407 return xfs_sync_sb(mp, true);
1412 struct super_block *sb)
1414 struct xfs_mount *mp = XFS_M(sb);
1416 xfs_restore_resvblks(mp);
1417 xfs_log_work_queue(mp);
1418 xfs_icache_enable_reclaim(mp);
1423 xfs_fs_show_options(
1425 struct dentry *root)
1427 return xfs_showargs(XFS_M(root->d_sb), m);
1431 * This function fills in xfs_mount_t fields based on mount args.
1432 * Note: the superblock _has_ now been read in.
1436 struct xfs_mount *mp)
1438 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1440 /* Fail a mount where the logbuf is smaller than the log stripe */
1441 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1442 if (mp->m_logbsize <= 0 &&
1443 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1444 mp->m_logbsize = mp->m_sb.sb_logsunit;
1445 } else if (mp->m_logbsize > 0 &&
1446 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1448 "logbuf size must be greater than or equal to log stripe size");
1452 /* Fail a mount if the logbuf is larger than 32K */
1453 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1455 "logbuf size for version 1 logs must be 16K or 32K");
1461 * V5 filesystems always use attr2 format for attributes.
1463 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1464 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1465 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1466 "attr2 is always enabled for V5 filesystems.");
1471 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1472 * told by noattr2 to turn it off
1474 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1475 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1476 mp->m_flags |= XFS_MOUNT_ATTR2;
1479 * prohibit r/w mounts of read-only filesystems
1481 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1483 "cannot mount a read-only filesystem as read-write");
1487 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1488 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1489 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1491 "Super block does not support project and group quota together");
1499 xfs_init_percpu_counters(
1500 struct xfs_mount *mp)
1504 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1508 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1512 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1519 percpu_counter_destroy(&mp->m_ifree);
1521 percpu_counter_destroy(&mp->m_icount);
1526 xfs_reinit_percpu_counters(
1527 struct xfs_mount *mp)
1529 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1530 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1531 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1535 xfs_destroy_percpu_counters(
1536 struct xfs_mount *mp)
1538 percpu_counter_destroy(&mp->m_icount);
1539 percpu_counter_destroy(&mp->m_ifree);
1540 percpu_counter_destroy(&mp->m_fdblocks);
1543 static struct xfs_mount *
1545 struct super_block *sb)
1547 struct xfs_mount *mp;
1549 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1554 spin_lock_init(&mp->m_sb_lock);
1555 spin_lock_init(&mp->m_agirotor_lock);
1556 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1557 spin_lock_init(&mp->m_perag_lock);
1558 mutex_init(&mp->m_growlock);
1559 atomic_set(&mp->m_active_trans, 0);
1560 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1561 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1562 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1563 mp->m_kobj.kobject.kset = xfs_kset;
1570 struct super_block *sb,
1575 struct xfs_mount *mp = NULL;
1576 int flags = 0, error = -ENOMEM;
1579 * allocate mp and do all low-level struct initializations before we
1580 * attach it to the super
1582 mp = xfs_mount_alloc(sb);
1587 error = xfs_parseargs(mp, (char *)data);
1589 goto out_free_fsname;
1591 sb_min_blocksize(sb, BBSIZE);
1592 sb->s_xattr = xfs_xattr_handlers;
1593 sb->s_export_op = &xfs_export_operations;
1594 #ifdef CONFIG_XFS_QUOTA
1595 sb->s_qcop = &xfs_quotactl_operations;
1596 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1598 sb->s_op = &xfs_super_operations;
1601 * Delay mount work if the debug hook is set. This is debug
1602 * instrumention to coordinate simulation of xfs mount failures with
1603 * VFS superblock operations
1605 if (xfs_globals.mount_delay) {
1606 xfs_notice(mp, "Delaying mount for %d seconds.",
1607 xfs_globals.mount_delay);
1608 msleep(xfs_globals.mount_delay * 1000);
1612 flags |= XFS_MFSI_QUIET;
1614 error = xfs_open_devices(mp);
1616 goto out_free_fsname;
1618 error = xfs_init_mount_workqueues(mp);
1620 goto out_close_devices;
1622 error = xfs_init_percpu_counters(mp);
1624 goto out_destroy_workqueues;
1626 /* Allocate stats memory before we do operations that might use it */
1627 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1628 if (!mp->m_stats.xs_stats) {
1630 goto out_destroy_counters;
1633 error = xfs_readsb(mp, flags);
1635 goto out_free_stats;
1637 error = xfs_finish_flags(mp);
1641 error = xfs_setup_devices(mp);
1645 error = xfs_filestream_mount(mp);
1650 * we must configure the block size in the superblock before we run the
1651 * full mount process as the mount process can lookup and cache inodes.
1653 sb->s_magic = XFS_SB_MAGIC;
1654 sb->s_blocksize = mp->m_sb.sb_blocksize;
1655 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1656 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1657 sb->s_max_links = XFS_MAXLINK;
1658 sb->s_time_gran = 1;
1659 set_posix_acl_flag(sb);
1661 /* version 5 superblocks support inode version counters. */
1662 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1663 sb->s_flags |= SB_I_VERSION;
1665 if (mp->m_flags & XFS_MOUNT_DAX) {
1666 bool rtdev_is_dax = false, datadev_is_dax;
1669 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1671 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1673 if (mp->m_rtdev_targp)
1674 rtdev_is_dax = bdev_dax_supported(
1675 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1676 if (!rtdev_is_dax && !datadev_is_dax) {
1678 "DAX unsupported by block device. Turning off DAX.");
1679 mp->m_flags &= ~XFS_MOUNT_DAX;
1681 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1683 "DAX and reflink cannot be used together!");
1685 goto out_filestream_unmount;
1689 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1690 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1692 if (!blk_queue_discard(q)) {
1693 xfs_warn(mp, "mounting with \"discard\" option, but "
1694 "the device does not support discard");
1695 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1699 if (xfs_sb_version_hasreflink(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1701 "reflink not compatible with realtime device!");
1703 goto out_filestream_unmount;
1706 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1708 "reverse mapping btree not compatible with realtime device!");
1710 goto out_filestream_unmount;
1713 error = xfs_mountfs(mp);
1715 goto out_filestream_unmount;
1717 root = igrab(VFS_I(mp->m_rootip));
1722 sb->s_root = d_make_root(root);
1730 out_filestream_unmount:
1731 xfs_filestream_unmount(mp);
1735 free_percpu(mp->m_stats.xs_stats);
1736 out_destroy_counters:
1737 xfs_destroy_percpu_counters(mp);
1738 out_destroy_workqueues:
1739 xfs_destroy_mount_workqueues(mp);
1741 xfs_close_devices(mp);
1743 sb->s_fs_info = NULL;
1744 xfs_free_fsname(mp);
1750 xfs_filestream_unmount(mp);
1757 struct super_block *sb)
1759 struct xfs_mount *mp = XFS_M(sb);
1761 /* if ->fill_super failed, we have no mount to tear down */
1765 xfs_notice(mp, "Unmounting Filesystem");
1766 xfs_filestream_unmount(mp);
1770 free_percpu(mp->m_stats.xs_stats);
1771 xfs_destroy_percpu_counters(mp);
1772 xfs_destroy_mount_workqueues(mp);
1773 xfs_close_devices(mp);
1775 sb->s_fs_info = NULL;
1776 xfs_free_fsname(mp);
1780 STATIC struct dentry *
1782 struct file_system_type *fs_type,
1784 const char *dev_name,
1787 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1791 xfs_fs_nr_cached_objects(
1792 struct super_block *sb,
1793 struct shrink_control *sc)
1795 /* Paranoia: catch incorrect calls during mount setup or teardown */
1796 if (WARN_ON_ONCE(!sb->s_fs_info))
1798 return xfs_reclaim_inodes_count(XFS_M(sb));
1802 xfs_fs_free_cached_objects(
1803 struct super_block *sb,
1804 struct shrink_control *sc)
1806 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1809 static const struct super_operations xfs_super_operations = {
1810 .alloc_inode = xfs_fs_alloc_inode,
1811 .destroy_inode = xfs_fs_destroy_inode,
1812 .dirty_inode = xfs_fs_dirty_inode,
1813 .drop_inode = xfs_fs_drop_inode,
1814 .put_super = xfs_fs_put_super,
1815 .sync_fs = xfs_fs_sync_fs,
1816 .freeze_fs = xfs_fs_freeze,
1817 .unfreeze_fs = xfs_fs_unfreeze,
1818 .statfs = xfs_fs_statfs,
1819 .remount_fs = xfs_fs_remount,
1820 .show_options = xfs_fs_show_options,
1821 .nr_cached_objects = xfs_fs_nr_cached_objects,
1822 .free_cached_objects = xfs_fs_free_cached_objects,
1825 static struct file_system_type xfs_fs_type = {
1826 .owner = THIS_MODULE,
1828 .mount = xfs_fs_mount,
1829 .kill_sb = kill_block_super,
1830 .fs_flags = FS_REQUIRES_DEV,
1832 MODULE_ALIAS_FS("xfs");
1835 xfs_init_zones(void)
1837 if (bioset_init(&xfs_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
1838 offsetof(struct xfs_ioend, io_inline_bio),
1842 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1844 if (!xfs_log_ticket_zone)
1845 goto out_free_ioend_bioset;
1847 xfs_bmap_free_item_zone = kmem_zone_init(
1848 sizeof(struct xfs_extent_free_item),
1849 "xfs_bmap_free_item");
1850 if (!xfs_bmap_free_item_zone)
1851 goto out_destroy_log_ticket_zone;
1853 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1855 if (!xfs_btree_cur_zone)
1856 goto out_destroy_bmap_free_item_zone;
1858 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1860 if (!xfs_da_state_zone)
1861 goto out_destroy_btree_cur_zone;
1863 xfs_ifork_zone = kmem_zone_init(sizeof(struct xfs_ifork), "xfs_ifork");
1864 if (!xfs_ifork_zone)
1865 goto out_destroy_da_state_zone;
1867 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1868 if (!xfs_trans_zone)
1869 goto out_destroy_ifork_zone;
1873 * The size of the zone allocated buf log item is the maximum
1874 * size possible under XFS. This wastes a little bit of memory,
1875 * but it is much faster.
1877 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1879 if (!xfs_buf_item_zone)
1880 goto out_destroy_trans_zone;
1882 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1883 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1884 sizeof(xfs_extent_t))), "xfs_efd_item");
1886 goto out_destroy_buf_item_zone;
1888 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1889 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1890 sizeof(xfs_extent_t))), "xfs_efi_item");
1892 goto out_destroy_efd_zone;
1895 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1896 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1897 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1898 if (!xfs_inode_zone)
1899 goto out_destroy_efi_zone;
1902 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1903 KM_ZONE_SPREAD, NULL);
1905 goto out_destroy_inode_zone;
1906 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1908 if (!xfs_icreate_zone)
1909 goto out_destroy_ili_zone;
1911 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1914 goto out_destroy_icreate_zone;
1916 xfs_rui_zone = kmem_zone_init(
1917 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1920 goto out_destroy_rud_zone;
1922 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1925 goto out_destroy_rui_zone;
1927 xfs_cui_zone = kmem_zone_init(
1928 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1931 goto out_destroy_cud_zone;
1933 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1936 goto out_destroy_cui_zone;
1938 xfs_bui_zone = kmem_zone_init(
1939 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1942 goto out_destroy_bud_zone;
1946 out_destroy_bud_zone:
1947 kmem_zone_destroy(xfs_bud_zone);
1948 out_destroy_cui_zone:
1949 kmem_zone_destroy(xfs_cui_zone);
1950 out_destroy_cud_zone:
1951 kmem_zone_destroy(xfs_cud_zone);
1952 out_destroy_rui_zone:
1953 kmem_zone_destroy(xfs_rui_zone);
1954 out_destroy_rud_zone:
1955 kmem_zone_destroy(xfs_rud_zone);
1956 out_destroy_icreate_zone:
1957 kmem_zone_destroy(xfs_icreate_zone);
1958 out_destroy_ili_zone:
1959 kmem_zone_destroy(xfs_ili_zone);
1960 out_destroy_inode_zone:
1961 kmem_zone_destroy(xfs_inode_zone);
1962 out_destroy_efi_zone:
1963 kmem_zone_destroy(xfs_efi_zone);
1964 out_destroy_efd_zone:
1965 kmem_zone_destroy(xfs_efd_zone);
1966 out_destroy_buf_item_zone:
1967 kmem_zone_destroy(xfs_buf_item_zone);
1968 out_destroy_trans_zone:
1969 kmem_zone_destroy(xfs_trans_zone);
1970 out_destroy_ifork_zone:
1971 kmem_zone_destroy(xfs_ifork_zone);
1972 out_destroy_da_state_zone:
1973 kmem_zone_destroy(xfs_da_state_zone);
1974 out_destroy_btree_cur_zone:
1975 kmem_zone_destroy(xfs_btree_cur_zone);
1976 out_destroy_bmap_free_item_zone:
1977 kmem_zone_destroy(xfs_bmap_free_item_zone);
1978 out_destroy_log_ticket_zone:
1979 kmem_zone_destroy(xfs_log_ticket_zone);
1980 out_free_ioend_bioset:
1981 bioset_exit(&xfs_ioend_bioset);
1987 xfs_destroy_zones(void)
1990 * Make sure all delayed rcu free are flushed before we
1994 kmem_zone_destroy(xfs_bui_zone);
1995 kmem_zone_destroy(xfs_bud_zone);
1996 kmem_zone_destroy(xfs_cui_zone);
1997 kmem_zone_destroy(xfs_cud_zone);
1998 kmem_zone_destroy(xfs_rui_zone);
1999 kmem_zone_destroy(xfs_rud_zone);
2000 kmem_zone_destroy(xfs_icreate_zone);
2001 kmem_zone_destroy(xfs_ili_zone);
2002 kmem_zone_destroy(xfs_inode_zone);
2003 kmem_zone_destroy(xfs_efi_zone);
2004 kmem_zone_destroy(xfs_efd_zone);
2005 kmem_zone_destroy(xfs_buf_item_zone);
2006 kmem_zone_destroy(xfs_trans_zone);
2007 kmem_zone_destroy(xfs_ifork_zone);
2008 kmem_zone_destroy(xfs_da_state_zone);
2009 kmem_zone_destroy(xfs_btree_cur_zone);
2010 kmem_zone_destroy(xfs_bmap_free_item_zone);
2011 kmem_zone_destroy(xfs_log_ticket_zone);
2012 bioset_exit(&xfs_ioend_bioset);
2016 xfs_init_workqueues(void)
2019 * The allocation workqueue can be used in memory reclaim situations
2020 * (writepage path), and parallelism is only limited by the number of
2021 * AGs in all the filesystems mounted. Hence use the default large
2022 * max_active value for this workqueue.
2024 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2025 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2029 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2030 if (!xfs_discard_wq)
2031 goto out_free_alloc_wq;
2035 destroy_workqueue(xfs_alloc_wq);
2040 xfs_destroy_workqueues(void)
2042 destroy_workqueue(xfs_discard_wq);
2043 destroy_workqueue(xfs_alloc_wq);
2051 xfs_check_ondisk_structs();
2053 printk(KERN_INFO XFS_VERSION_STRING " with "
2054 XFS_BUILD_OPTIONS " enabled\n");
2056 xfs_extent_free_init_defer_op();
2057 xfs_rmap_update_init_defer_op();
2058 xfs_refcount_update_init_defer_op();
2059 xfs_bmap_update_init_defer_op();
2063 error = xfs_init_zones();
2067 error = xfs_init_workqueues();
2069 goto out_destroy_zones;
2071 error = xfs_mru_cache_init();
2073 goto out_destroy_wq;
2075 error = xfs_buf_init();
2077 goto out_mru_cache_uninit;
2079 error = xfs_init_procfs();
2081 goto out_buf_terminate;
2083 error = xfs_sysctl_register();
2085 goto out_cleanup_procfs;
2087 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2090 goto out_sysctl_unregister;
2093 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2095 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2096 if (!xfsstats.xs_stats) {
2098 goto out_kset_unregister;
2101 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2104 goto out_free_stats;
2107 xfs_dbg_kobj.kobject.kset = xfs_kset;
2108 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2110 goto out_remove_stats_kobj;
2113 error = xfs_qm_init();
2115 goto out_remove_dbg_kobj;
2117 error = register_filesystem(&xfs_fs_type);
2124 out_remove_dbg_kobj:
2126 xfs_sysfs_del(&xfs_dbg_kobj);
2127 out_remove_stats_kobj:
2129 xfs_sysfs_del(&xfsstats.xs_kobj);
2131 free_percpu(xfsstats.xs_stats);
2132 out_kset_unregister:
2133 kset_unregister(xfs_kset);
2134 out_sysctl_unregister:
2135 xfs_sysctl_unregister();
2137 xfs_cleanup_procfs();
2139 xfs_buf_terminate();
2140 out_mru_cache_uninit:
2141 xfs_mru_cache_uninit();
2143 xfs_destroy_workqueues();
2145 xfs_destroy_zones();
2154 unregister_filesystem(&xfs_fs_type);
2156 xfs_sysfs_del(&xfs_dbg_kobj);
2158 xfs_sysfs_del(&xfsstats.xs_kobj);
2159 free_percpu(xfsstats.xs_stats);
2160 kset_unregister(xfs_kset);
2161 xfs_sysctl_unregister();
2162 xfs_cleanup_procfs();
2163 xfs_buf_terminate();
2164 xfs_mru_cache_uninit();
2165 xfs_destroy_workqueues();
2166 xfs_destroy_zones();
2167 xfs_uuid_table_free();
2170 module_init(init_xfs_fs);
2171 module_exit(exit_xfs_fs);
2173 MODULE_AUTHOR("Silicon Graphics, Inc.");
2174 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2175 MODULE_LICENSE("GPL");