2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
30 #include "xfs_alloc.h"
31 #include "xfs_error.h"
32 #include "xfs_fsops.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
36 #include "xfs_log_priv.h"
37 #include "xfs_da_btree.h"
39 #include "xfs_extfree_item.h"
40 #include "xfs_mru_cache.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_icache.h"
43 #include "xfs_trace.h"
44 #include "xfs_icreate_item.h"
45 #include "xfs_filestream.h"
46 #include "xfs_quota.h"
47 #include "xfs_sysfs.h"
48 #include "xfs_ondisk.h"
49 #include "xfs_rmap_item.h"
50 #include "xfs_refcount_item.h"
51 #include "xfs_bmap_item.h"
52 #include "xfs_reflink.h"
54 #include <linux/namei.h>
55 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mount.h>
58 #include <linux/mempool.h>
59 #include <linux/writeback.h>
60 #include <linux/kthread.h>
61 #include <linux/freezer.h>
62 #include <linux/parser.h>
64 static const struct super_operations xfs_super_operations;
65 struct bio_set *xfs_ioend_bioset;
67 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
69 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
73 * Table driven mount option parser.
76 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
77 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
78 Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
79 Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
80 Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
81 Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
82 Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
83 Opt_uquota, Opt_gquota, Opt_pquota,
84 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
85 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
88 static const match_table_t tokens = {
89 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */
90 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */
91 {Opt_logdev, "logdev=%s"}, /* log device */
92 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */
93 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */
94 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */
95 {Opt_noalign, "noalign"}, /* turn off stripe alignment */
96 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */
97 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */
98 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */
99 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */
100 {Opt_mtpt, "mtpt"}, /* filesystem mount point */
101 {Opt_grpid, "grpid"}, /* group-ID from parent directory */
102 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */
103 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */
104 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */
105 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
106 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */
107 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */
108 {Opt_inode32, "inode32"}, /* inode allocation limited to
109 * XFS_MAXINUMBER_32 */
110 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */
111 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */
112 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */
113 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes
115 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */
116 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */
117 {Opt_filestreams,"filestreams"},/* use filestreams allocator */
118 {Opt_quota, "quota"}, /* disk quotas (user) */
119 {Opt_noquota, "noquota"}, /* no quotas */
120 {Opt_usrquota, "usrquota"}, /* user quota enabled */
121 {Opt_grpquota, "grpquota"}, /* group quota enabled */
122 {Opt_prjquota, "prjquota"}, /* project quota enabled */
123 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */
124 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */
125 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */
126 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
127 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
128 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
129 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
130 {Opt_discard, "discard"}, /* Discard unused blocks */
131 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */
133 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */
135 /* Deprecated mount options scheduled for removal */
136 {Opt_barrier, "barrier"}, /* use writer barriers for log write and
137 * unwritten extent conversion */
138 {Opt_nobarrier, "nobarrier"}, /* .. disable */
145 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
147 int last, shift_left_factor = 0, _res;
151 value = match_strdup(s);
155 last = strlen(value) - 1;
156 if (value[last] == 'K' || value[last] == 'k') {
157 shift_left_factor = 10;
160 if (value[last] == 'M' || value[last] == 'm') {
161 shift_left_factor = 20;
164 if (value[last] == 'G' || value[last] == 'g') {
165 shift_left_factor = 30;
169 if (kstrtoint(value, base, &_res))
172 *res = _res << shift_left_factor;
177 * This function fills in xfs_mount_t fields based on mount args.
178 * Note: the superblock has _not_ yet been read in.
180 * Note that this function leaks the various device name allocations on
181 * failure. The caller takes care of them.
183 * *sb is const because this is also used to test options on the remount
184 * path, and we don't want this to have any side effects at remount time.
185 * Today this function does not change *sb, but just to future-proof...
189 struct xfs_mount *mp,
192 const struct super_block *sb = mp->m_super;
194 substring_t args[MAX_OPT_ARGS];
198 __uint8_t iosizelog = 0;
201 * set up the mount name first so all the errors will refer to the
204 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
207 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
210 * Copy binary VFS mount flags we are interested in.
212 if (sb->s_flags & MS_RDONLY)
213 mp->m_flags |= XFS_MOUNT_RDONLY;
214 if (sb->s_flags & MS_DIRSYNC)
215 mp->m_flags |= XFS_MOUNT_DIRSYNC;
216 if (sb->s_flags & MS_SYNCHRONOUS)
217 mp->m_flags |= XFS_MOUNT_WSYNC;
220 * Set some default flags that could be cleared by the mount option
223 mp->m_flags |= XFS_MOUNT_BARRIER;
224 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
227 * These can be overridden by the mount option parsing.
235 while ((p = strsep(&options, ",")) != NULL) {
241 token = match_token(p, tokens, args);
244 if (match_int(args, &mp->m_logbufs))
248 if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
252 mp->m_logname = match_strdup(args);
257 xfs_warn(mp, "%s option not allowed on this system", p);
260 mp->m_rtname = match_strdup(args);
266 if (suffix_kstrtoint(args, 10, &iosize))
268 iosizelog = ffs(iosize) - 1;
272 mp->m_flags |= XFS_MOUNT_GRPID;
276 mp->m_flags &= ~XFS_MOUNT_GRPID;
279 mp->m_flags |= XFS_MOUNT_WSYNC;
282 mp->m_flags |= XFS_MOUNT_NORECOVERY;
285 mp->m_flags |= XFS_MOUNT_NOALIGN;
288 mp->m_flags |= XFS_MOUNT_SWALLOC;
291 if (match_int(args, &dsunit))
295 if (match_int(args, &dswidth))
299 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
302 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
305 mp->m_flags |= XFS_MOUNT_NOUUID;
308 mp->m_flags |= XFS_MOUNT_IKEEP;
311 mp->m_flags &= ~XFS_MOUNT_IKEEP;
314 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
317 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
320 mp->m_flags |= XFS_MOUNT_ATTR2;
323 mp->m_flags &= ~XFS_MOUNT_ATTR2;
324 mp->m_flags |= XFS_MOUNT_NOATTR2;
326 case Opt_filestreams:
327 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
330 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
331 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
332 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
337 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
341 case Opt_uqnoenforce:
342 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
343 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
347 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
350 case Opt_pqnoenforce:
351 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
352 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
356 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
359 case Opt_gqnoenforce:
360 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
361 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
364 mp->m_flags |= XFS_MOUNT_DISCARD;
367 mp->m_flags &= ~XFS_MOUNT_DISCARD;
371 mp->m_flags |= XFS_MOUNT_DAX;
375 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
376 mp->m_flags |= XFS_MOUNT_BARRIER;
379 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
380 mp->m_flags &= ~XFS_MOUNT_BARRIER;
383 xfs_warn(mp, "unknown mount option [%s].", p);
389 * no recovery flag requires a read-only mount
391 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
392 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
393 xfs_warn(mp, "no-recovery mounts must be read-only.");
397 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
399 "sunit and swidth options incompatible with the noalign option");
403 #ifndef CONFIG_XFS_QUOTA
404 if (XFS_IS_QUOTA_RUNNING(mp)) {
405 xfs_warn(mp, "quota support not available in this kernel.");
410 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
411 xfs_warn(mp, "sunit and swidth must be specified together");
415 if (dsunit && (dswidth % dsunit != 0)) {
417 "stripe width (%d) must be a multiple of the stripe unit (%d)",
423 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
425 * At this point the superblock has not been read
426 * in, therefore we do not know the block size.
427 * Before the mount call ends we will convert
430 mp->m_dalign = dsunit;
431 mp->m_swidth = dswidth;
434 if (mp->m_logbufs != -1 &&
435 mp->m_logbufs != 0 &&
436 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
437 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
438 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
439 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
442 if (mp->m_logbsize != -1 &&
443 mp->m_logbsize != 0 &&
444 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
445 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
446 !is_power_of_2(mp->m_logbsize))) {
448 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
454 if (iosizelog > XFS_MAX_IO_LOG ||
455 iosizelog < XFS_MIN_IO_LOG) {
456 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
457 iosizelog, XFS_MIN_IO_LOG,
462 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
463 mp->m_readio_log = iosizelog;
464 mp->m_writeio_log = iosizelog;
470 struct proc_xfs_info {
477 struct xfs_mount *mp,
480 static struct proc_xfs_info xfs_info_set[] = {
481 /* the few simple ones we can get from the mount struct */
482 { XFS_MOUNT_IKEEP, ",ikeep" },
483 { XFS_MOUNT_WSYNC, ",wsync" },
484 { XFS_MOUNT_NOALIGN, ",noalign" },
485 { XFS_MOUNT_SWALLOC, ",swalloc" },
486 { XFS_MOUNT_NOUUID, ",nouuid" },
487 { XFS_MOUNT_NORECOVERY, ",norecovery" },
488 { XFS_MOUNT_ATTR2, ",attr2" },
489 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
490 { XFS_MOUNT_GRPID, ",grpid" },
491 { XFS_MOUNT_DISCARD, ",discard" },
492 { XFS_MOUNT_SMALL_INUMS, ",inode32" },
493 { XFS_MOUNT_DAX, ",dax" },
496 static struct proc_xfs_info xfs_info_unset[] = {
497 /* the few simple ones we can get from the mount struct */
498 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" },
499 { XFS_MOUNT_BARRIER, ",nobarrier" },
500 { XFS_MOUNT_SMALL_INUMS, ",inode64" },
503 struct proc_xfs_info *xfs_infop;
505 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
506 if (mp->m_flags & xfs_infop->flag)
507 seq_puts(m, xfs_infop->str);
509 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
510 if (!(mp->m_flags & xfs_infop->flag))
511 seq_puts(m, xfs_infop->str);
514 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
515 seq_printf(m, ",allocsize=%dk",
516 (int)(1 << mp->m_writeio_log) >> 10);
518 if (mp->m_logbufs > 0)
519 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
520 if (mp->m_logbsize > 0)
521 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
524 seq_show_option(m, "logdev", mp->m_logname);
526 seq_show_option(m, "rtdev", mp->m_rtname);
528 if (mp->m_dalign > 0)
529 seq_printf(m, ",sunit=%d",
530 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
531 if (mp->m_swidth > 0)
532 seq_printf(m, ",swidth=%d",
533 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
535 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
536 seq_puts(m, ",usrquota");
537 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
538 seq_puts(m, ",uqnoenforce");
540 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
541 if (mp->m_qflags & XFS_PQUOTA_ENFD)
542 seq_puts(m, ",prjquota");
544 seq_puts(m, ",pqnoenforce");
546 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
547 if (mp->m_qflags & XFS_GQUOTA_ENFD)
548 seq_puts(m, ",grpquota");
550 seq_puts(m, ",gqnoenforce");
553 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
554 seq_puts(m, ",noquota");
560 unsigned int blockshift)
562 unsigned int pagefactor = 1;
563 unsigned int bitshift = BITS_PER_LONG - 1;
565 /* Figure out maximum filesize, on Linux this can depend on
566 * the filesystem blocksize (on 32 bit platforms).
567 * __block_write_begin does this in an [unsigned] long...
568 * page->index << (PAGE_SHIFT - bbits)
569 * So, for page sized blocks (4K on 32 bit platforms),
570 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
571 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
572 * but for smaller blocksizes it is less (bbits = log2 bsize).
573 * Note1: get_block_t takes a long (implicit cast from above)
574 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
575 * can optionally convert the [unsigned] long from above into
576 * an [unsigned] long long.
579 #if BITS_PER_LONG == 32
580 # if defined(CONFIG_LBDAF)
581 ASSERT(sizeof(sector_t) == 8);
582 pagefactor = PAGE_SIZE;
583 bitshift = BITS_PER_LONG;
585 pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
589 return (((__uint64_t)pagefactor) << bitshift) - 1;
593 * Set parameters for inode allocation heuristics, taking into account
594 * filesystem size and inode32/inode64 mount options; i.e. specifically
595 * whether or not XFS_MOUNT_SMALL_INUMS is set.
597 * Inode allocation patterns are altered only if inode32 is requested
598 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
599 * If altered, XFS_MOUNT_32BITINODES is set as well.
601 * An agcount independent of that in the mount structure is provided
602 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
603 * to the potentially higher ag count.
605 * Returns the maximum AG index which may contain inodes.
609 struct xfs_mount *mp,
610 xfs_agnumber_t agcount)
612 xfs_agnumber_t index;
613 xfs_agnumber_t maxagi = 0;
614 xfs_sb_t *sbp = &mp->m_sb;
615 xfs_agnumber_t max_metadata;
620 * Calculate how much should be reserved for inodes to meet
621 * the max inode percentage. Used only for inode32.
623 if (mp->m_maxicount) {
626 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
628 icount += sbp->sb_agblocks - 1;
629 do_div(icount, sbp->sb_agblocks);
630 max_metadata = icount;
632 max_metadata = agcount;
635 /* Get the last possible inode in the filesystem */
636 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
637 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
640 * If user asked for no more than 32-bit inodes, and the fs is
641 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
642 * the allocator to accommodate the request.
644 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
645 mp->m_flags |= XFS_MOUNT_32BITINODES;
647 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
649 for (index = 0; index < agcount; index++) {
650 struct xfs_perag *pag;
652 ino = XFS_AGINO_TO_INO(mp, index, agino);
654 pag = xfs_perag_get(mp, index);
656 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
657 if (ino > XFS_MAXINUMBER_32) {
658 pag->pagi_inodeok = 0;
659 pag->pagf_metadata = 0;
661 pag->pagi_inodeok = 1;
663 if (index < max_metadata)
664 pag->pagf_metadata = 1;
666 pag->pagf_metadata = 0;
669 pag->pagi_inodeok = 1;
670 pag->pagf_metadata = 0;
676 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
683 struct block_device **bdevp)
687 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
689 if (IS_ERR(*bdevp)) {
690 error = PTR_ERR(*bdevp);
691 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
699 struct block_device *bdev)
702 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
706 xfs_blkdev_issue_flush(
707 xfs_buftarg_t *buftarg)
709 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
714 struct xfs_mount *mp)
716 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
717 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
718 xfs_free_buftarg(mp, mp->m_logdev_targp);
719 xfs_blkdev_put(logdev);
721 if (mp->m_rtdev_targp) {
722 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
723 xfs_free_buftarg(mp, mp->m_rtdev_targp);
724 xfs_blkdev_put(rtdev);
726 xfs_free_buftarg(mp, mp->m_ddev_targp);
730 * The file system configurations are:
731 * (1) device (partition) with data and internal log
732 * (2) logical volume with data and log subvolumes.
733 * (3) logical volume with data, log, and realtime subvolumes.
735 * We only have to handle opening the log and realtime volumes here if
736 * they are present. The data subvolume has already been opened by
737 * get_sb_bdev() and is stored in sb->s_bdev.
741 struct xfs_mount *mp)
743 struct block_device *ddev = mp->m_super->s_bdev;
744 struct block_device *logdev = NULL, *rtdev = NULL;
748 * Open real time and log devices - order is important.
751 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
757 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
759 goto out_close_logdev;
761 if (rtdev == ddev || rtdev == logdev) {
763 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
765 goto out_close_rtdev;
770 * Setup xfs_mount buffer target pointers
773 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
774 if (!mp->m_ddev_targp)
775 goto out_close_rtdev;
778 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
779 if (!mp->m_rtdev_targp)
780 goto out_free_ddev_targ;
783 if (logdev && logdev != ddev) {
784 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
785 if (!mp->m_logdev_targp)
786 goto out_free_rtdev_targ;
788 mp->m_logdev_targp = mp->m_ddev_targp;
794 if (mp->m_rtdev_targp)
795 xfs_free_buftarg(mp, mp->m_rtdev_targp);
797 xfs_free_buftarg(mp, mp->m_ddev_targp);
799 xfs_blkdev_put(rtdev);
801 if (logdev && logdev != ddev)
802 xfs_blkdev_put(logdev);
808 * Setup xfs_mount buffer target pointers based on superblock
812 struct xfs_mount *mp)
816 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
820 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
821 unsigned int log_sector_size = BBSIZE;
823 if (xfs_sb_version_hassector(&mp->m_sb))
824 log_sector_size = mp->m_sb.sb_logsectsize;
825 error = xfs_setsize_buftarg(mp->m_logdev_targp,
830 if (mp->m_rtdev_targp) {
831 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
832 mp->m_sb.sb_sectsize);
841 xfs_init_mount_workqueues(
842 struct xfs_mount *mp)
844 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
845 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
846 if (!mp->m_buf_workqueue)
849 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
850 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
851 if (!mp->m_data_workqueue)
852 goto out_destroy_buf;
854 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
855 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
856 if (!mp->m_unwritten_workqueue)
857 goto out_destroy_data_iodone_queue;
859 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
860 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
861 if (!mp->m_cil_workqueue)
862 goto out_destroy_unwritten;
864 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
865 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
866 if (!mp->m_reclaim_workqueue)
867 goto out_destroy_cil;
869 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
870 WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
872 if (!mp->m_log_workqueue)
873 goto out_destroy_reclaim;
875 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
876 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
877 if (!mp->m_eofblocks_workqueue)
878 goto out_destroy_log;
880 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
882 if (!mp->m_sync_workqueue)
883 goto out_destroy_eofb;
888 destroy_workqueue(mp->m_eofblocks_workqueue);
890 destroy_workqueue(mp->m_log_workqueue);
892 destroy_workqueue(mp->m_reclaim_workqueue);
894 destroy_workqueue(mp->m_cil_workqueue);
895 out_destroy_unwritten:
896 destroy_workqueue(mp->m_unwritten_workqueue);
897 out_destroy_data_iodone_queue:
898 destroy_workqueue(mp->m_data_workqueue);
900 destroy_workqueue(mp->m_buf_workqueue);
906 xfs_destroy_mount_workqueues(
907 struct xfs_mount *mp)
909 destroy_workqueue(mp->m_sync_workqueue);
910 destroy_workqueue(mp->m_eofblocks_workqueue);
911 destroy_workqueue(mp->m_log_workqueue);
912 destroy_workqueue(mp->m_reclaim_workqueue);
913 destroy_workqueue(mp->m_cil_workqueue);
914 destroy_workqueue(mp->m_data_workqueue);
915 destroy_workqueue(mp->m_unwritten_workqueue);
916 destroy_workqueue(mp->m_buf_workqueue);
920 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
921 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
922 * for IO to complete so that we effectively throttle multiple callers to the
923 * rate at which IO is completing.
927 struct xfs_mount *mp)
929 struct super_block *sb = mp->m_super;
931 if (down_read_trylock(&sb->s_umount)) {
933 up_read(&sb->s_umount);
937 /* Catch misguided souls that try to use this interface on XFS */
938 STATIC struct inode *
940 struct super_block *sb)
947 * Now that the generic code is guaranteed not to be accessing
948 * the linux inode, we can inactivate and reclaim the inode.
951 xfs_fs_destroy_inode(
954 struct xfs_inode *ip = XFS_I(inode);
957 trace_xfs_destroy_inode(ip);
959 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
960 XFS_STATS_INC(ip->i_mount, vn_rele);
961 XFS_STATS_INC(ip->i_mount, vn_remove);
963 if (xfs_is_reflink_inode(ip)) {
964 error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
965 if (error && !XFS_FORCED_SHUTDOWN(ip->i_mount))
966 xfs_warn(ip->i_mount,
967 "Error %d while evicting CoW blocks for inode %llu.",
973 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
974 XFS_STATS_INC(ip->i_mount, vn_reclaim);
977 * We should never get here with one of the reclaim flags already set.
979 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
980 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
983 * We always use background reclaim here because even if the
984 * inode is clean, it still may be under IO and hence we have
985 * to take the flush lock. The background reclaim path handles
986 * this more efficiently than we can here, so simply let background
987 * reclaim tear down all inodes.
989 xfs_inode_set_reclaim_tag(ip);
993 * Slab object creation initialisation for the XFS inode.
994 * This covers only the idempotent fields in the XFS inode;
995 * all other fields need to be initialised on allocation
996 * from the slab. This avoids the need to repeatedly initialise
997 * fields in the xfs inode that left in the initialise state
998 * when freeing the inode.
1001 xfs_fs_inode_init_once(
1004 struct xfs_inode *ip = inode;
1006 memset(ip, 0, sizeof(struct xfs_inode));
1009 inode_init_once(VFS_I(ip));
1012 atomic_set(&ip->i_pincount, 0);
1013 spin_lock_init(&ip->i_flags_lock);
1015 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1016 "xfsino", ip->i_ino);
1017 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1018 "xfsino", ip->i_ino);
1022 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1023 * serialised against cache hits here via the inode->i_lock and igrab() in
1024 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1025 * racing with us, and it avoids needing to grab a spinlock here for every inode
1026 * we drop the final reference on.
1030 struct inode *inode)
1032 struct xfs_inode *ip = XFS_I(inode);
1035 * If this unlinked inode is in the middle of recovery, don't
1036 * drop the inode just yet; log recovery will take care of
1037 * that. See the comment for this inode flag.
1039 if (ip->i_flags & XFS_IRECOVERY) {
1040 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1044 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1049 struct xfs_mount *mp)
1051 kfree(mp->m_fsname);
1052 kfree(mp->m_rtname);
1053 kfree(mp->m_logname);
1058 struct super_block *sb,
1061 struct xfs_mount *mp = XFS_M(sb);
1064 * Doing anything during the async pass would be counterproductive.
1069 xfs_log_force(mp, XFS_LOG_SYNC);
1072 * The disk must be active because we're syncing.
1073 * We schedule log work now (now that the disk is
1074 * active) instead of later (when it might not be).
1076 flush_delayed_work(&mp->m_log->l_work);
1084 struct dentry *dentry,
1085 struct kstatfs *statp)
1087 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1088 xfs_sb_t *sbp = &mp->m_sb;
1089 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1090 __uint64_t fakeinos, id;
1093 __uint64_t fdblocks;
1097 statp->f_type = XFS_SB_MAGIC;
1098 statp->f_namelen = MAXNAMELEN - 1;
1100 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1101 statp->f_fsid.val[0] = (u32)id;
1102 statp->f_fsid.val[1] = (u32)(id >> 32);
1104 icount = percpu_counter_sum(&mp->m_icount);
1105 ifree = percpu_counter_sum(&mp->m_ifree);
1106 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1108 spin_lock(&mp->m_sb_lock);
1109 statp->f_bsize = sbp->sb_blocksize;
1110 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1111 statp->f_blocks = sbp->sb_dblocks - lsize;
1112 spin_unlock(&mp->m_sb_lock);
1114 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1115 statp->f_bavail = statp->f_bfree;
1117 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1118 statp->f_files = MIN(icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1119 if (mp->m_maxicount)
1120 statp->f_files = min_t(typeof(statp->f_files),
1124 /* If sb_icount overshot maxicount, report actual allocation */
1125 statp->f_files = max_t(typeof(statp->f_files),
1129 /* make sure statp->f_ffree does not underflow */
1130 ffree = statp->f_files - (icount - ifree);
1131 statp->f_ffree = max_t(__int64_t, ffree, 0);
1134 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1135 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1136 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1137 xfs_qm_statvfs(ip, statp);
1142 xfs_save_resvblks(struct xfs_mount *mp)
1144 __uint64_t resblks = 0;
1146 mp->m_resblks_save = mp->m_resblks;
1147 xfs_reserve_blocks(mp, &resblks, NULL);
1151 xfs_restore_resvblks(struct xfs_mount *mp)
1155 if (mp->m_resblks_save) {
1156 resblks = mp->m_resblks_save;
1157 mp->m_resblks_save = 0;
1159 resblks = xfs_default_resblks(mp);
1161 xfs_reserve_blocks(mp, &resblks, NULL);
1165 * Trigger writeback of all the dirty metadata in the file system.
1167 * This ensures that the metadata is written to their location on disk rather
1168 * than just existing in transactions in the log. This means after a quiesce
1169 * there is no log replay required to write the inodes to disk - this is the
1170 * primary difference between a sync and a quiesce.
1172 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1173 * it is started again when appropriate.
1177 struct xfs_mount *mp)
1181 /* wait for all modifications to complete */
1182 while (atomic_read(&mp->m_active_trans) > 0)
1185 /* force the log to unpin objects from the now complete transactions */
1186 xfs_log_force(mp, XFS_LOG_SYNC);
1188 /* reclaim inodes to do any IO before the freeze completes */
1189 xfs_reclaim_inodes(mp, 0);
1190 xfs_reclaim_inodes(mp, SYNC_WAIT);
1192 /* Push the superblock and write an unmount record */
1193 error = xfs_log_sbcount(mp);
1195 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1196 "Frozen image may not be consistent.");
1198 * Just warn here till VFS can correctly support
1199 * read-only remount without racing.
1201 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1203 xfs_log_quiesce(mp);
1207 xfs_test_remount_options(
1208 struct super_block *sb,
1209 struct xfs_mount *mp,
1213 struct xfs_mount *tmp_mp;
1215 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1219 tmp_mp->m_super = sb;
1220 error = xfs_parseargs(tmp_mp, options);
1221 xfs_free_fsname(tmp_mp);
1229 struct super_block *sb,
1233 struct xfs_mount *mp = XFS_M(sb);
1234 xfs_sb_t *sbp = &mp->m_sb;
1235 substring_t args[MAX_OPT_ARGS];
1239 /* First, check for complete junk; i.e. invalid options */
1240 error = xfs_test_remount_options(sb, mp, options);
1244 sync_filesystem(sb);
1245 while ((p = strsep(&options, ",")) != NULL) {
1251 token = match_token(p, tokens, args);
1254 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1255 mp->m_flags |= XFS_MOUNT_BARRIER;
1258 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1259 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1262 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1263 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1266 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1267 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1271 * Logically we would return an error here to prevent
1272 * users from believing they might have changed
1273 * mount options using remount which can't be changed.
1275 * But unfortunately mount(8) adds all options from
1276 * mtab and fstab to the mount arguments in some cases
1277 * so we can't blindly reject options, but have to
1278 * check for each specified option if it actually
1279 * differs from the currently set option and only
1280 * reject it if that's the case.
1282 * Until that is implemented we return success for
1283 * every remount request, and silently ignore all
1284 * options that we can't actually change.
1288 "mount option \"%s\" not supported for remount", p);
1297 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1298 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1300 "ro->rw transition prohibited on norecovery mount");
1304 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1305 xfs_sb_has_ro_compat_feature(sbp,
1306 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1308 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1309 (sbp->sb_features_ro_compat &
1310 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1314 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1317 * If this is the first remount to writeable state we
1318 * might have some superblock changes to update.
1320 if (mp->m_update_sb) {
1321 error = xfs_sync_sb(mp, false);
1323 xfs_warn(mp, "failed to write sb changes");
1326 mp->m_update_sb = false;
1330 * Fill out the reserve pool if it is empty. Use the stashed
1331 * value if it is non-zero, otherwise go with the default.
1333 xfs_restore_resvblks(mp);
1334 xfs_log_work_queue(mp);
1335 xfs_queue_eofblocks(mp);
1337 /* Recover any CoW blocks that never got remapped. */
1338 error = xfs_reflink_recover_cow(mp);
1341 "Error %d recovering leftover CoW allocations.", error);
1342 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1346 /* Create the per-AG metadata reservation pool .*/
1347 error = xfs_fs_reserve_ag_blocks(mp);
1348 if (error && error != -ENOSPC)
1353 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1354 /* Free the per-AG metadata reservation pool. */
1355 error = xfs_fs_unreserve_ag_blocks(mp);
1357 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1362 * Before we sync the metadata, we need to free up the reserve
1363 * block pool so that the used block count in the superblock on
1364 * disk is correct at the end of the remount. Stash the current
1365 * reserve pool size so that if we get remounted rw, we can
1366 * return it to the same size.
1368 xfs_save_resvblks(mp);
1371 * Cancel background eofb scanning so it cannot race with the
1372 * final log force+buftarg wait and deadlock the remount.
1374 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1376 xfs_quiesce_attr(mp);
1377 mp->m_flags |= XFS_MOUNT_RDONLY;
1384 * Second stage of a freeze. The data is already frozen so we only
1385 * need to take care of the metadata. Once that's done sync the superblock
1386 * to the log to dirty it in case of a crash while frozen. This ensures that we
1387 * will recover the unlinked inode lists on the next mount.
1391 struct super_block *sb)
1393 struct xfs_mount *mp = XFS_M(sb);
1395 xfs_save_resvblks(mp);
1396 xfs_quiesce_attr(mp);
1397 return xfs_sync_sb(mp, true);
1402 struct super_block *sb)
1404 struct xfs_mount *mp = XFS_M(sb);
1406 xfs_restore_resvblks(mp);
1407 xfs_log_work_queue(mp);
1412 xfs_fs_show_options(
1414 struct dentry *root)
1416 return xfs_showargs(XFS_M(root->d_sb), m);
1420 * This function fills in xfs_mount_t fields based on mount args.
1421 * Note: the superblock _has_ now been read in.
1425 struct xfs_mount *mp)
1427 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1429 /* Fail a mount where the logbuf is smaller than the log stripe */
1430 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1431 if (mp->m_logbsize <= 0 &&
1432 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1433 mp->m_logbsize = mp->m_sb.sb_logsunit;
1434 } else if (mp->m_logbsize > 0 &&
1435 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1437 "logbuf size must be greater than or equal to log stripe size");
1441 /* Fail a mount if the logbuf is larger than 32K */
1442 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1444 "logbuf size for version 1 logs must be 16K or 32K");
1450 * V5 filesystems always use attr2 format for attributes.
1452 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1453 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1454 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1455 "attr2 is always enabled for V5 filesystems.");
1460 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1461 * told by noattr2 to turn it off
1463 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1464 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1465 mp->m_flags |= XFS_MOUNT_ATTR2;
1468 * prohibit r/w mounts of read-only filesystems
1470 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1472 "cannot mount a read-only filesystem as read-write");
1476 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1477 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1478 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1480 "Super block does not support project and group quota together");
1488 xfs_init_percpu_counters(
1489 struct xfs_mount *mp)
1493 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1497 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1501 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1508 percpu_counter_destroy(&mp->m_ifree);
1510 percpu_counter_destroy(&mp->m_icount);
1515 xfs_reinit_percpu_counters(
1516 struct xfs_mount *mp)
1518 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1519 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1520 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1524 xfs_destroy_percpu_counters(
1525 struct xfs_mount *mp)
1527 percpu_counter_destroy(&mp->m_icount);
1528 percpu_counter_destroy(&mp->m_ifree);
1529 percpu_counter_destroy(&mp->m_fdblocks);
1534 struct super_block *sb,
1539 struct xfs_mount *mp = NULL;
1540 int flags = 0, error = -ENOMEM;
1542 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1546 spin_lock_init(&mp->m_sb_lock);
1547 mutex_init(&mp->m_growlock);
1548 atomic_set(&mp->m_active_trans, 0);
1549 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1550 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1551 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1552 mp->m_kobj.kobject.kset = xfs_kset;
1557 error = xfs_parseargs(mp, (char *)data);
1559 goto out_free_fsname;
1561 sb_min_blocksize(sb, BBSIZE);
1562 sb->s_xattr = xfs_xattr_handlers;
1563 sb->s_export_op = &xfs_export_operations;
1564 #ifdef CONFIG_XFS_QUOTA
1565 sb->s_qcop = &xfs_quotactl_operations;
1566 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1568 sb->s_op = &xfs_super_operations;
1571 flags |= XFS_MFSI_QUIET;
1573 error = xfs_open_devices(mp);
1575 goto out_free_fsname;
1577 error = xfs_init_mount_workqueues(mp);
1579 goto out_close_devices;
1581 error = xfs_init_percpu_counters(mp);
1583 goto out_destroy_workqueues;
1585 /* Allocate stats memory before we do operations that might use it */
1586 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1587 if (!mp->m_stats.xs_stats) {
1589 goto out_destroy_counters;
1592 error = xfs_readsb(mp, flags);
1594 goto out_free_stats;
1596 error = xfs_finish_flags(mp);
1600 error = xfs_setup_devices(mp);
1604 error = xfs_filestream_mount(mp);
1609 * we must configure the block size in the superblock before we run the
1610 * full mount process as the mount process can lookup and cache inodes.
1612 sb->s_magic = XFS_SB_MAGIC;
1613 sb->s_blocksize = mp->m_sb.sb_blocksize;
1614 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1615 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1616 sb->s_max_links = XFS_MAXLINK;
1617 sb->s_time_gran = 1;
1618 set_posix_acl_flag(sb);
1620 /* version 5 superblocks support inode version counters. */
1621 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1622 sb->s_flags |= MS_I_VERSION;
1624 if (mp->m_flags & XFS_MOUNT_DAX) {
1626 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1628 error = bdev_dax_supported(sb, sb->s_blocksize);
1631 "DAX unsupported by block device. Turning off DAX.");
1632 mp->m_flags &= ~XFS_MOUNT_DAX;
1634 if (xfs_sb_version_hasreflink(&mp->m_sb))
1636 "DAX and reflink have not been tested together!");
1639 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1640 if (mp->m_sb.sb_rblocks) {
1642 "EXPERIMENTAL reverse mapping btree not compatible with realtime device!");
1644 goto out_filestream_unmount;
1647 "EXPERIMENTAL reverse mapping btree feature enabled. Use at your own risk!");
1650 if (xfs_sb_version_hasreflink(&mp->m_sb))
1652 "EXPERIMENTAL reflink feature enabled. Use at your own risk!");
1654 error = xfs_mountfs(mp);
1656 goto out_filestream_unmount;
1658 root = igrab(VFS_I(mp->m_rootip));
1663 sb->s_root = d_make_root(root);
1671 out_filestream_unmount:
1672 xfs_filestream_unmount(mp);
1676 free_percpu(mp->m_stats.xs_stats);
1677 out_destroy_counters:
1678 xfs_destroy_percpu_counters(mp);
1679 out_destroy_workqueues:
1680 xfs_destroy_mount_workqueues(mp);
1682 xfs_close_devices(mp);
1684 xfs_free_fsname(mp);
1690 xfs_filestream_unmount(mp);
1697 struct super_block *sb)
1699 struct xfs_mount *mp = XFS_M(sb);
1701 xfs_notice(mp, "Unmounting Filesystem");
1702 xfs_filestream_unmount(mp);
1706 free_percpu(mp->m_stats.xs_stats);
1707 xfs_destroy_percpu_counters(mp);
1708 xfs_destroy_mount_workqueues(mp);
1709 xfs_close_devices(mp);
1710 xfs_free_fsname(mp);
1714 STATIC struct dentry *
1716 struct file_system_type *fs_type,
1718 const char *dev_name,
1721 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1725 xfs_fs_nr_cached_objects(
1726 struct super_block *sb,
1727 struct shrink_control *sc)
1729 return xfs_reclaim_inodes_count(XFS_M(sb));
1733 xfs_fs_free_cached_objects(
1734 struct super_block *sb,
1735 struct shrink_control *sc)
1737 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1740 static const struct super_operations xfs_super_operations = {
1741 .alloc_inode = xfs_fs_alloc_inode,
1742 .destroy_inode = xfs_fs_destroy_inode,
1743 .drop_inode = xfs_fs_drop_inode,
1744 .put_super = xfs_fs_put_super,
1745 .sync_fs = xfs_fs_sync_fs,
1746 .freeze_fs = xfs_fs_freeze,
1747 .unfreeze_fs = xfs_fs_unfreeze,
1748 .statfs = xfs_fs_statfs,
1749 .remount_fs = xfs_fs_remount,
1750 .show_options = xfs_fs_show_options,
1751 .nr_cached_objects = xfs_fs_nr_cached_objects,
1752 .free_cached_objects = xfs_fs_free_cached_objects,
1755 static struct file_system_type xfs_fs_type = {
1756 .owner = THIS_MODULE,
1758 .mount = xfs_fs_mount,
1759 .kill_sb = kill_block_super,
1760 .fs_flags = FS_REQUIRES_DEV,
1762 MODULE_ALIAS_FS("xfs");
1765 xfs_init_zones(void)
1767 xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1768 offsetof(struct xfs_ioend, io_inline_bio));
1769 if (!xfs_ioend_bioset)
1772 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1774 if (!xfs_log_ticket_zone)
1775 goto out_free_ioend_bioset;
1777 xfs_bmap_free_item_zone = kmem_zone_init(
1778 sizeof(struct xfs_extent_free_item),
1779 "xfs_bmap_free_item");
1780 if (!xfs_bmap_free_item_zone)
1781 goto out_destroy_log_ticket_zone;
1783 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1785 if (!xfs_btree_cur_zone)
1786 goto out_destroy_bmap_free_item_zone;
1788 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1790 if (!xfs_da_state_zone)
1791 goto out_destroy_btree_cur_zone;
1793 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1794 if (!xfs_ifork_zone)
1795 goto out_destroy_da_state_zone;
1797 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1798 if (!xfs_trans_zone)
1799 goto out_destroy_ifork_zone;
1801 xfs_log_item_desc_zone =
1802 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1803 "xfs_log_item_desc");
1804 if (!xfs_log_item_desc_zone)
1805 goto out_destroy_trans_zone;
1808 * The size of the zone allocated buf log item is the maximum
1809 * size possible under XFS. This wastes a little bit of memory,
1810 * but it is much faster.
1812 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1814 if (!xfs_buf_item_zone)
1815 goto out_destroy_log_item_desc_zone;
1817 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1818 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1819 sizeof(xfs_extent_t))), "xfs_efd_item");
1821 goto out_destroy_buf_item_zone;
1823 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1824 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1825 sizeof(xfs_extent_t))), "xfs_efi_item");
1827 goto out_destroy_efd_zone;
1830 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1831 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1832 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1833 if (!xfs_inode_zone)
1834 goto out_destroy_efi_zone;
1837 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1838 KM_ZONE_SPREAD, NULL);
1840 goto out_destroy_inode_zone;
1841 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1843 if (!xfs_icreate_zone)
1844 goto out_destroy_ili_zone;
1846 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1849 goto out_destroy_icreate_zone;
1851 xfs_rui_zone = kmem_zone_init(
1852 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1855 goto out_destroy_rud_zone;
1857 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1860 goto out_destroy_rui_zone;
1862 xfs_cui_zone = kmem_zone_init(
1863 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1866 goto out_destroy_cud_zone;
1868 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1871 goto out_destroy_cui_zone;
1873 xfs_bui_zone = kmem_zone_init(
1874 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1877 goto out_destroy_bud_zone;
1881 out_destroy_bud_zone:
1882 kmem_zone_destroy(xfs_bud_zone);
1883 out_destroy_cui_zone:
1884 kmem_zone_destroy(xfs_cui_zone);
1885 out_destroy_cud_zone:
1886 kmem_zone_destroy(xfs_cud_zone);
1887 out_destroy_rui_zone:
1888 kmem_zone_destroy(xfs_rui_zone);
1889 out_destroy_rud_zone:
1890 kmem_zone_destroy(xfs_rud_zone);
1891 out_destroy_icreate_zone:
1892 kmem_zone_destroy(xfs_icreate_zone);
1893 out_destroy_ili_zone:
1894 kmem_zone_destroy(xfs_ili_zone);
1895 out_destroy_inode_zone:
1896 kmem_zone_destroy(xfs_inode_zone);
1897 out_destroy_efi_zone:
1898 kmem_zone_destroy(xfs_efi_zone);
1899 out_destroy_efd_zone:
1900 kmem_zone_destroy(xfs_efd_zone);
1901 out_destroy_buf_item_zone:
1902 kmem_zone_destroy(xfs_buf_item_zone);
1903 out_destroy_log_item_desc_zone:
1904 kmem_zone_destroy(xfs_log_item_desc_zone);
1905 out_destroy_trans_zone:
1906 kmem_zone_destroy(xfs_trans_zone);
1907 out_destroy_ifork_zone:
1908 kmem_zone_destroy(xfs_ifork_zone);
1909 out_destroy_da_state_zone:
1910 kmem_zone_destroy(xfs_da_state_zone);
1911 out_destroy_btree_cur_zone:
1912 kmem_zone_destroy(xfs_btree_cur_zone);
1913 out_destroy_bmap_free_item_zone:
1914 kmem_zone_destroy(xfs_bmap_free_item_zone);
1915 out_destroy_log_ticket_zone:
1916 kmem_zone_destroy(xfs_log_ticket_zone);
1917 out_free_ioend_bioset:
1918 bioset_free(xfs_ioend_bioset);
1924 xfs_destroy_zones(void)
1927 * Make sure all delayed rcu free are flushed before we
1931 kmem_zone_destroy(xfs_bui_zone);
1932 kmem_zone_destroy(xfs_bud_zone);
1933 kmem_zone_destroy(xfs_cui_zone);
1934 kmem_zone_destroy(xfs_cud_zone);
1935 kmem_zone_destroy(xfs_rui_zone);
1936 kmem_zone_destroy(xfs_rud_zone);
1937 kmem_zone_destroy(xfs_icreate_zone);
1938 kmem_zone_destroy(xfs_ili_zone);
1939 kmem_zone_destroy(xfs_inode_zone);
1940 kmem_zone_destroy(xfs_efi_zone);
1941 kmem_zone_destroy(xfs_efd_zone);
1942 kmem_zone_destroy(xfs_buf_item_zone);
1943 kmem_zone_destroy(xfs_log_item_desc_zone);
1944 kmem_zone_destroy(xfs_trans_zone);
1945 kmem_zone_destroy(xfs_ifork_zone);
1946 kmem_zone_destroy(xfs_da_state_zone);
1947 kmem_zone_destroy(xfs_btree_cur_zone);
1948 kmem_zone_destroy(xfs_bmap_free_item_zone);
1949 kmem_zone_destroy(xfs_log_ticket_zone);
1950 bioset_free(xfs_ioend_bioset);
1954 xfs_init_workqueues(void)
1957 * The allocation workqueue can be used in memory reclaim situations
1958 * (writepage path), and parallelism is only limited by the number of
1959 * AGs in all the filesystems mounted. Hence use the default large
1960 * max_active value for this workqueue.
1962 xfs_alloc_wq = alloc_workqueue("xfsalloc",
1963 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1967 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
1968 if (!xfs_discard_wq)
1969 goto out_free_alloc_wq;
1973 destroy_workqueue(xfs_alloc_wq);
1978 xfs_destroy_workqueues(void)
1980 destroy_workqueue(xfs_discard_wq);
1981 destroy_workqueue(xfs_alloc_wq);
1989 xfs_check_ondisk_structs();
1991 printk(KERN_INFO XFS_VERSION_STRING " with "
1992 XFS_BUILD_OPTIONS " enabled\n");
1994 xfs_extent_free_init_defer_op();
1995 xfs_rmap_update_init_defer_op();
1996 xfs_refcount_update_init_defer_op();
1997 xfs_bmap_update_init_defer_op();
2001 error = xfs_init_zones();
2005 error = xfs_init_workqueues();
2007 goto out_destroy_zones;
2009 error = xfs_mru_cache_init();
2011 goto out_destroy_wq;
2013 error = xfs_buf_init();
2015 goto out_mru_cache_uninit;
2017 error = xfs_init_procfs();
2019 goto out_buf_terminate;
2021 error = xfs_sysctl_register();
2023 goto out_cleanup_procfs;
2025 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2028 goto out_sysctl_unregister;
2031 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2033 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2034 if (!xfsstats.xs_stats) {
2036 goto out_kset_unregister;
2039 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2042 goto out_free_stats;
2045 xfs_dbg_kobj.kobject.kset = xfs_kset;
2046 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2048 goto out_remove_stats_kobj;
2051 error = xfs_qm_init();
2053 goto out_remove_dbg_kobj;
2055 error = register_filesystem(&xfs_fs_type);
2062 out_remove_dbg_kobj:
2064 xfs_sysfs_del(&xfs_dbg_kobj);
2065 out_remove_stats_kobj:
2067 xfs_sysfs_del(&xfsstats.xs_kobj);
2069 free_percpu(xfsstats.xs_stats);
2070 out_kset_unregister:
2071 kset_unregister(xfs_kset);
2072 out_sysctl_unregister:
2073 xfs_sysctl_unregister();
2075 xfs_cleanup_procfs();
2077 xfs_buf_terminate();
2078 out_mru_cache_uninit:
2079 xfs_mru_cache_uninit();
2081 xfs_destroy_workqueues();
2083 xfs_destroy_zones();
2092 unregister_filesystem(&xfs_fs_type);
2094 xfs_sysfs_del(&xfs_dbg_kobj);
2096 xfs_sysfs_del(&xfsstats.xs_kobj);
2097 free_percpu(xfsstats.xs_stats);
2098 kset_unregister(xfs_kset);
2099 xfs_sysctl_unregister();
2100 xfs_cleanup_procfs();
2101 xfs_buf_terminate();
2102 xfs_mru_cache_uninit();
2103 xfs_destroy_workqueues();
2104 xfs_destroy_zones();
2105 xfs_uuid_table_free();
2108 module_init(init_xfs_fs);
2109 module_exit(exit_xfs_fs);
2111 MODULE_AUTHOR("Silicon Graphics, Inc.");
2112 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2113 MODULE_LICENSE("GPL");