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_barrier, Opt_nobarrier,
69 Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
70 Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
71 Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
72 Opt_uquota, Opt_gquota, Opt_pquota,
73 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
74 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
77 static const match_table_t tokens = {
78 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */
79 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */
80 {Opt_logdev, "logdev=%s"}, /* log device */
81 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */
82 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */
83 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */
84 {Opt_noalign, "noalign"}, /* turn off stripe alignment */
85 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */
86 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */
87 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */
88 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */
89 {Opt_mtpt, "mtpt"}, /* filesystem mount point */
90 {Opt_grpid, "grpid"}, /* group-ID from parent directory */
91 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */
92 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */
93 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */
94 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
95 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */
96 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */
97 {Opt_inode32, "inode32"}, /* inode allocation limited to
98 * XFS_MAXINUMBER_32 */
99 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */
100 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */
101 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */
102 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes
104 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */
105 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */
106 {Opt_filestreams,"filestreams"},/* use filestreams allocator */
107 {Opt_quota, "quota"}, /* disk quotas (user) */
108 {Opt_noquota, "noquota"}, /* no quotas */
109 {Opt_usrquota, "usrquota"}, /* user quota enabled */
110 {Opt_grpquota, "grpquota"}, /* group quota enabled */
111 {Opt_prjquota, "prjquota"}, /* project quota enabled */
112 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */
113 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */
114 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */
115 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
116 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
117 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
118 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
119 {Opt_discard, "discard"}, /* Discard unused blocks */
120 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */
122 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */
124 /* Deprecated mount options scheduled for removal */
125 {Opt_barrier, "barrier"}, /* use writer barriers for log write and
126 * unwritten extent conversion */
127 {Opt_nobarrier, "nobarrier"}, /* .. disable */
134 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
136 int last, shift_left_factor = 0, _res;
140 value = match_strdup(s);
144 last = strlen(value) - 1;
145 if (value[last] == 'K' || value[last] == 'k') {
146 shift_left_factor = 10;
149 if (value[last] == 'M' || value[last] == 'm') {
150 shift_left_factor = 20;
153 if (value[last] == 'G' || value[last] == 'g') {
154 shift_left_factor = 30;
158 if (kstrtoint(value, base, &_res))
161 *res = _res << shift_left_factor;
166 * This function fills in xfs_mount_t fields based on mount args.
167 * Note: the superblock has _not_ yet been read in.
169 * Note that this function leaks the various device name allocations on
170 * failure. The caller takes care of them.
172 * *sb is const because this is also used to test options on the remount
173 * path, and we don't want this to have any side effects at remount time.
174 * Today this function does not change *sb, but just to future-proof...
178 struct xfs_mount *mp,
181 const struct super_block *sb = mp->m_super;
183 substring_t args[MAX_OPT_ARGS];
187 uint8_t iosizelog = 0;
190 * set up the mount name first so all the errors will refer to the
193 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
196 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
199 * Copy binary VFS mount flags we are interested in.
202 mp->m_flags |= XFS_MOUNT_RDONLY;
203 if (sb->s_flags & SB_DIRSYNC)
204 mp->m_flags |= XFS_MOUNT_DIRSYNC;
205 if (sb->s_flags & SB_SYNCHRONOUS)
206 mp->m_flags |= XFS_MOUNT_WSYNC;
209 * Set some default flags that could be cleared by the mount option
212 mp->m_flags |= XFS_MOUNT_BARRIER;
213 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
216 * These can be overridden by the mount option parsing.
224 while ((p = strsep(&options, ",")) != NULL) {
230 token = match_token(p, tokens, args);
233 if (match_int(args, &mp->m_logbufs))
237 if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
241 kfree(mp->m_logname);
242 mp->m_logname = match_strdup(args);
247 xfs_warn(mp, "%s option not allowed on this system", p);
251 mp->m_rtname = match_strdup(args);
257 if (suffix_kstrtoint(args, 10, &iosize))
259 iosizelog = ffs(iosize) - 1;
263 mp->m_flags |= XFS_MOUNT_GRPID;
267 mp->m_flags &= ~XFS_MOUNT_GRPID;
270 mp->m_flags |= XFS_MOUNT_WSYNC;
273 mp->m_flags |= XFS_MOUNT_NORECOVERY;
276 mp->m_flags |= XFS_MOUNT_NOALIGN;
279 mp->m_flags |= XFS_MOUNT_SWALLOC;
282 if (match_int(args, &dsunit))
286 if (match_int(args, &dswidth))
290 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
293 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
296 mp->m_flags |= XFS_MOUNT_NOUUID;
299 mp->m_flags |= XFS_MOUNT_IKEEP;
302 mp->m_flags &= ~XFS_MOUNT_IKEEP;
305 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
308 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
311 mp->m_flags |= XFS_MOUNT_ATTR2;
314 mp->m_flags &= ~XFS_MOUNT_ATTR2;
315 mp->m_flags |= XFS_MOUNT_NOATTR2;
317 case Opt_filestreams:
318 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
321 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
322 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
323 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
328 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
332 case Opt_uqnoenforce:
333 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
334 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
338 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
341 case Opt_pqnoenforce:
342 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
343 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
347 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
350 case Opt_gqnoenforce:
351 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
352 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
355 mp->m_flags |= XFS_MOUNT_DISCARD;
358 mp->m_flags &= ~XFS_MOUNT_DISCARD;
362 mp->m_flags |= XFS_MOUNT_DAX;
366 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
367 mp->m_flags |= XFS_MOUNT_BARRIER;
370 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
371 mp->m_flags &= ~XFS_MOUNT_BARRIER;
374 xfs_warn(mp, "unknown mount option [%s].", p);
380 * no recovery flag requires a read-only mount
382 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
383 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
384 xfs_warn(mp, "no-recovery mounts must be read-only.");
388 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
390 "sunit and swidth options incompatible with the noalign option");
394 #ifndef CONFIG_XFS_QUOTA
395 if (XFS_IS_QUOTA_RUNNING(mp)) {
396 xfs_warn(mp, "quota support not available in this kernel.");
401 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
402 xfs_warn(mp, "sunit and swidth must be specified together");
406 if (dsunit && (dswidth % dsunit != 0)) {
408 "stripe width (%d) must be a multiple of the stripe unit (%d)",
414 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
416 * At this point the superblock has not been read
417 * in, therefore we do not know the block size.
418 * Before the mount call ends we will convert
421 mp->m_dalign = dsunit;
422 mp->m_swidth = dswidth;
425 if (mp->m_logbufs != -1 &&
426 mp->m_logbufs != 0 &&
427 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
428 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
429 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
430 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
433 if (mp->m_logbsize != -1 &&
434 mp->m_logbsize != 0 &&
435 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
436 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
437 !is_power_of_2(mp->m_logbsize))) {
439 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
445 if (iosizelog > XFS_MAX_IO_LOG ||
446 iosizelog < XFS_MIN_IO_LOG) {
447 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
448 iosizelog, XFS_MIN_IO_LOG,
453 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
454 mp->m_readio_log = iosizelog;
455 mp->m_writeio_log = iosizelog;
461 struct proc_xfs_info {
468 struct xfs_mount *mp,
471 static struct proc_xfs_info xfs_info_set[] = {
472 /* the few simple ones we can get from the mount struct */
473 { XFS_MOUNT_IKEEP, ",ikeep" },
474 { XFS_MOUNT_WSYNC, ",wsync" },
475 { XFS_MOUNT_NOALIGN, ",noalign" },
476 { XFS_MOUNT_SWALLOC, ",swalloc" },
477 { XFS_MOUNT_NOUUID, ",nouuid" },
478 { XFS_MOUNT_NORECOVERY, ",norecovery" },
479 { XFS_MOUNT_ATTR2, ",attr2" },
480 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
481 { XFS_MOUNT_GRPID, ",grpid" },
482 { XFS_MOUNT_DISCARD, ",discard" },
483 { XFS_MOUNT_SMALL_INUMS, ",inode32" },
484 { XFS_MOUNT_DAX, ",dax" },
487 static struct proc_xfs_info xfs_info_unset[] = {
488 /* the few simple ones we can get from the mount struct */
489 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" },
490 { XFS_MOUNT_BARRIER, ",nobarrier" },
491 { XFS_MOUNT_SMALL_INUMS, ",inode64" },
494 struct proc_xfs_info *xfs_infop;
496 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
497 if (mp->m_flags & xfs_infop->flag)
498 seq_puts(m, xfs_infop->str);
500 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
501 if (!(mp->m_flags & xfs_infop->flag))
502 seq_puts(m, xfs_infop->str);
505 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
506 seq_printf(m, ",allocsize=%dk",
507 (int)(1 << mp->m_writeio_log) >> 10);
509 if (mp->m_logbufs > 0)
510 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
511 if (mp->m_logbsize > 0)
512 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
515 seq_show_option(m, "logdev", mp->m_logname);
517 seq_show_option(m, "rtdev", mp->m_rtname);
519 if (mp->m_dalign > 0)
520 seq_printf(m, ",sunit=%d",
521 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
522 if (mp->m_swidth > 0)
523 seq_printf(m, ",swidth=%d",
524 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
526 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
527 seq_puts(m, ",usrquota");
528 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
529 seq_puts(m, ",uqnoenforce");
531 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
532 if (mp->m_qflags & XFS_PQUOTA_ENFD)
533 seq_puts(m, ",prjquota");
535 seq_puts(m, ",pqnoenforce");
537 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
538 if (mp->m_qflags & XFS_GQUOTA_ENFD)
539 seq_puts(m, ",grpquota");
541 seq_puts(m, ",gqnoenforce");
544 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
545 seq_puts(m, ",noquota");
551 unsigned int blockshift)
553 unsigned int pagefactor = 1;
554 unsigned int bitshift = BITS_PER_LONG - 1;
556 /* Figure out maximum filesize, on Linux this can depend on
557 * the filesystem blocksize (on 32 bit platforms).
558 * __block_write_begin does this in an [unsigned] long...
559 * page->index << (PAGE_SHIFT - bbits)
560 * So, for page sized blocks (4K on 32 bit platforms),
561 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
562 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
563 * but for smaller blocksizes it is less (bbits = log2 bsize).
564 * Note1: get_block_t takes a long (implicit cast from above)
565 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
566 * can optionally convert the [unsigned] long from above into
567 * an [unsigned] long long.
570 #if BITS_PER_LONG == 32
571 # if defined(CONFIG_LBDAF)
572 ASSERT(sizeof(sector_t) == 8);
573 pagefactor = PAGE_SIZE;
574 bitshift = BITS_PER_LONG;
576 pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
580 return (((uint64_t)pagefactor) << bitshift) - 1;
584 * Set parameters for inode allocation heuristics, taking into account
585 * filesystem size and inode32/inode64 mount options; i.e. specifically
586 * whether or not XFS_MOUNT_SMALL_INUMS is set.
588 * Inode allocation patterns are altered only if inode32 is requested
589 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
590 * If altered, XFS_MOUNT_32BITINODES is set as well.
592 * An agcount independent of that in the mount structure is provided
593 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
594 * to the potentially higher ag count.
596 * Returns the maximum AG index which may contain inodes.
600 struct xfs_mount *mp,
601 xfs_agnumber_t agcount)
603 xfs_agnumber_t index;
604 xfs_agnumber_t maxagi = 0;
605 xfs_sb_t *sbp = &mp->m_sb;
606 xfs_agnumber_t max_metadata;
611 * Calculate how much should be reserved for inodes to meet
612 * the max inode percentage. Used only for inode32.
614 if (mp->m_maxicount) {
617 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
619 icount += sbp->sb_agblocks - 1;
620 do_div(icount, sbp->sb_agblocks);
621 max_metadata = icount;
623 max_metadata = agcount;
626 /* Get the last possible inode in the filesystem */
627 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
628 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
631 * If user asked for no more than 32-bit inodes, and the fs is
632 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
633 * the allocator to accommodate the request.
635 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
636 mp->m_flags |= XFS_MOUNT_32BITINODES;
638 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
640 for (index = 0; index < agcount; index++) {
641 struct xfs_perag *pag;
643 ino = XFS_AGINO_TO_INO(mp, index, agino);
645 pag = xfs_perag_get(mp, index);
647 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
648 if (ino > XFS_MAXINUMBER_32) {
649 pag->pagi_inodeok = 0;
650 pag->pagf_metadata = 0;
652 pag->pagi_inodeok = 1;
654 if (index < max_metadata)
655 pag->pagf_metadata = 1;
657 pag->pagf_metadata = 0;
660 pag->pagi_inodeok = 1;
661 pag->pagf_metadata = 0;
667 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
674 struct block_device **bdevp)
678 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
680 if (IS_ERR(*bdevp)) {
681 error = PTR_ERR(*bdevp);
682 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
690 struct block_device *bdev)
693 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
697 xfs_blkdev_issue_flush(
698 xfs_buftarg_t *buftarg)
700 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
705 struct xfs_mount *mp)
707 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
709 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
710 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
711 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
713 xfs_free_buftarg(mp->m_logdev_targp);
714 xfs_blkdev_put(logdev);
715 fs_put_dax(dax_logdev);
717 if (mp->m_rtdev_targp) {
718 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
719 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
721 xfs_free_buftarg(mp->m_rtdev_targp);
722 xfs_blkdev_put(rtdev);
723 fs_put_dax(dax_rtdev);
725 xfs_free_buftarg(mp->m_ddev_targp);
726 fs_put_dax(dax_ddev);
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 dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
745 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
746 struct block_device *logdev = NULL, *rtdev = NULL;
750 * Open real time and log devices - order is important.
753 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
756 dax_logdev = fs_dax_get_by_bdev(logdev);
760 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
762 goto out_close_logdev;
764 if (rtdev == ddev || rtdev == logdev) {
766 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
768 goto out_close_rtdev;
770 dax_rtdev = fs_dax_get_by_bdev(rtdev);
774 * Setup xfs_mount buffer target pointers
777 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
778 if (!mp->m_ddev_targp)
779 goto out_close_rtdev;
782 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
783 if (!mp->m_rtdev_targp)
784 goto out_free_ddev_targ;
787 if (logdev && logdev != ddev) {
788 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
789 if (!mp->m_logdev_targp)
790 goto out_free_rtdev_targ;
792 mp->m_logdev_targp = mp->m_ddev_targp;
798 if (mp->m_rtdev_targp)
799 xfs_free_buftarg(mp->m_rtdev_targp);
801 xfs_free_buftarg(mp->m_ddev_targp);
803 xfs_blkdev_put(rtdev);
804 fs_put_dax(dax_rtdev);
806 if (logdev && logdev != ddev) {
807 xfs_blkdev_put(logdev);
808 fs_put_dax(dax_logdev);
811 fs_put_dax(dax_ddev);
816 * Setup xfs_mount buffer target pointers based on superblock
820 struct xfs_mount *mp)
824 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
828 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
829 unsigned int log_sector_size = BBSIZE;
831 if (xfs_sb_version_hassector(&mp->m_sb))
832 log_sector_size = mp->m_sb.sb_logsectsize;
833 error = xfs_setsize_buftarg(mp->m_logdev_targp,
838 if (mp->m_rtdev_targp) {
839 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
840 mp->m_sb.sb_sectsize);
849 xfs_init_mount_workqueues(
850 struct xfs_mount *mp)
852 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
853 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
854 if (!mp->m_buf_workqueue)
857 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
858 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
859 if (!mp->m_data_workqueue)
860 goto out_destroy_buf;
862 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
863 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
864 if (!mp->m_unwritten_workqueue)
865 goto out_destroy_data_iodone_queue;
867 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
868 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
869 if (!mp->m_cil_workqueue)
870 goto out_destroy_unwritten;
872 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
873 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
874 if (!mp->m_reclaim_workqueue)
875 goto out_destroy_cil;
877 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
878 WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
880 if (!mp->m_log_workqueue)
881 goto out_destroy_reclaim;
883 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
884 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
885 if (!mp->m_eofblocks_workqueue)
886 goto out_destroy_log;
888 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
890 if (!mp->m_sync_workqueue)
891 goto out_destroy_eofb;
896 destroy_workqueue(mp->m_eofblocks_workqueue);
898 destroy_workqueue(mp->m_log_workqueue);
900 destroy_workqueue(mp->m_reclaim_workqueue);
902 destroy_workqueue(mp->m_cil_workqueue);
903 out_destroy_unwritten:
904 destroy_workqueue(mp->m_unwritten_workqueue);
905 out_destroy_data_iodone_queue:
906 destroy_workqueue(mp->m_data_workqueue);
908 destroy_workqueue(mp->m_buf_workqueue);
914 xfs_destroy_mount_workqueues(
915 struct xfs_mount *mp)
917 destroy_workqueue(mp->m_sync_workqueue);
918 destroy_workqueue(mp->m_eofblocks_workqueue);
919 destroy_workqueue(mp->m_log_workqueue);
920 destroy_workqueue(mp->m_reclaim_workqueue);
921 destroy_workqueue(mp->m_cil_workqueue);
922 destroy_workqueue(mp->m_data_workqueue);
923 destroy_workqueue(mp->m_unwritten_workqueue);
924 destroy_workqueue(mp->m_buf_workqueue);
928 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
929 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
930 * for IO to complete so that we effectively throttle multiple callers to the
931 * rate at which IO is completing.
935 struct xfs_mount *mp)
937 struct super_block *sb = mp->m_super;
939 if (down_read_trylock(&sb->s_umount)) {
941 up_read(&sb->s_umount);
945 /* Catch misguided souls that try to use this interface on XFS */
946 STATIC struct inode *
948 struct super_block *sb)
955 * Now that the generic code is guaranteed not to be accessing
956 * the linux inode, we can inactivate and reclaim the inode.
959 xfs_fs_destroy_inode(
962 struct xfs_inode *ip = XFS_I(inode);
964 trace_xfs_destroy_inode(ip);
966 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
967 XFS_STATS_INC(ip->i_mount, vn_rele);
968 XFS_STATS_INC(ip->i_mount, vn_remove);
972 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
973 XFS_STATS_INC(ip->i_mount, vn_reclaim);
976 * We should never get here with one of the reclaim flags already set.
978 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
979 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
982 * We always use background reclaim here because even if the
983 * inode is clean, it still may be under IO and hence we have
984 * to take the flush lock. The background reclaim path handles
985 * this more efficiently than we can here, so simply let background
986 * reclaim tear down all inodes.
988 xfs_inode_set_reclaim_tag(ip);
996 struct xfs_inode *ip = XFS_I(inode);
997 struct xfs_mount *mp = ip->i_mount;
998 struct xfs_trans *tp;
1000 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
1002 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
1005 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
1007 xfs_ilock(ip, XFS_ILOCK_EXCL);
1008 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1009 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1010 xfs_trans_commit(tp);
1014 * Slab object creation initialisation for the XFS inode.
1015 * This covers only the idempotent fields in the XFS inode;
1016 * all other fields need to be initialised on allocation
1017 * from the slab. This avoids the need to repeatedly initialise
1018 * fields in the xfs inode that left in the initialise state
1019 * when freeing the inode.
1022 xfs_fs_inode_init_once(
1025 struct xfs_inode *ip = inode;
1027 memset(ip, 0, sizeof(struct xfs_inode));
1030 inode_init_once(VFS_I(ip));
1033 atomic_set(&ip->i_pincount, 0);
1034 spin_lock_init(&ip->i_flags_lock);
1036 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1037 "xfsino", ip->i_ino);
1038 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1039 "xfsino", ip->i_ino);
1043 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1044 * serialised against cache hits here via the inode->i_lock and igrab() in
1045 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1046 * racing with us, and it avoids needing to grab a spinlock here for every inode
1047 * we drop the final reference on.
1051 struct inode *inode)
1053 struct xfs_inode *ip = XFS_I(inode);
1056 * If this unlinked inode is in the middle of recovery, don't
1057 * drop the inode just yet; log recovery will take care of
1058 * that. See the comment for this inode flag.
1060 if (ip->i_flags & XFS_IRECOVERY) {
1061 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1065 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1070 struct xfs_mount *mp)
1072 kfree(mp->m_fsname);
1073 kfree(mp->m_rtname);
1074 kfree(mp->m_logname);
1079 struct super_block *sb,
1082 struct xfs_mount *mp = XFS_M(sb);
1085 * Doing anything during the async pass would be counterproductive.
1090 xfs_log_force(mp, XFS_LOG_SYNC);
1093 * The disk must be active because we're syncing.
1094 * We schedule log work now (now that the disk is
1095 * active) instead of later (when it might not be).
1097 flush_delayed_work(&mp->m_log->l_work);
1105 struct dentry *dentry,
1106 struct kstatfs *statp)
1108 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1109 xfs_sb_t *sbp = &mp->m_sb;
1110 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1111 uint64_t fakeinos, id;
1118 statp->f_type = XFS_SB_MAGIC;
1119 statp->f_namelen = MAXNAMELEN - 1;
1121 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1122 statp->f_fsid.val[0] = (u32)id;
1123 statp->f_fsid.val[1] = (u32)(id >> 32);
1125 icount = percpu_counter_sum(&mp->m_icount);
1126 ifree = percpu_counter_sum(&mp->m_ifree);
1127 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1129 spin_lock(&mp->m_sb_lock);
1130 statp->f_bsize = sbp->sb_blocksize;
1131 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1132 statp->f_blocks = sbp->sb_dblocks - lsize;
1133 spin_unlock(&mp->m_sb_lock);
1135 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1136 statp->f_bavail = statp->f_bfree;
1138 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1139 statp->f_files = MIN(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1140 if (mp->m_maxicount)
1141 statp->f_files = min_t(typeof(statp->f_files),
1145 /* If sb_icount overshot maxicount, report actual allocation */
1146 statp->f_files = max_t(typeof(statp->f_files),
1150 /* make sure statp->f_ffree does not underflow */
1151 ffree = statp->f_files - (icount - ifree);
1152 statp->f_ffree = max_t(int64_t, ffree, 0);
1155 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1156 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1157 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1158 xfs_qm_statvfs(ip, statp);
1160 if (XFS_IS_REALTIME_MOUNT(mp) &&
1161 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1162 statp->f_blocks = sbp->sb_rblocks;
1163 statp->f_bavail = statp->f_bfree =
1164 sbp->sb_frextents * sbp->sb_rextsize;
1171 xfs_save_resvblks(struct xfs_mount *mp)
1173 uint64_t resblks = 0;
1175 mp->m_resblks_save = mp->m_resblks;
1176 xfs_reserve_blocks(mp, &resblks, NULL);
1180 xfs_restore_resvblks(struct xfs_mount *mp)
1184 if (mp->m_resblks_save) {
1185 resblks = mp->m_resblks_save;
1186 mp->m_resblks_save = 0;
1188 resblks = xfs_default_resblks(mp);
1190 xfs_reserve_blocks(mp, &resblks, NULL);
1194 * Trigger writeback of all the dirty metadata in the file system.
1196 * This ensures that the metadata is written to their location on disk rather
1197 * than just existing in transactions in the log. This means after a quiesce
1198 * there is no log replay required to write the inodes to disk - this is the
1199 * primary difference between a sync and a quiesce.
1201 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1202 * it is started again when appropriate.
1206 struct xfs_mount *mp)
1210 /* wait for all modifications to complete */
1211 while (atomic_read(&mp->m_active_trans) > 0)
1214 /* force the log to unpin objects from the now complete transactions */
1215 xfs_log_force(mp, XFS_LOG_SYNC);
1217 /* reclaim inodes to do any IO before the freeze completes */
1218 xfs_reclaim_inodes(mp, 0);
1219 xfs_reclaim_inodes(mp, SYNC_WAIT);
1221 /* Push the superblock and write an unmount record */
1222 error = xfs_log_sbcount(mp);
1224 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1225 "Frozen image may not be consistent.");
1227 * Just warn here till VFS can correctly support
1228 * read-only remount without racing.
1230 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1232 xfs_log_quiesce(mp);
1236 xfs_test_remount_options(
1237 struct super_block *sb,
1241 struct xfs_mount *tmp_mp;
1243 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1247 tmp_mp->m_super = sb;
1248 error = xfs_parseargs(tmp_mp, options);
1249 xfs_free_fsname(tmp_mp);
1257 struct super_block *sb,
1261 struct xfs_mount *mp = XFS_M(sb);
1262 xfs_sb_t *sbp = &mp->m_sb;
1263 substring_t args[MAX_OPT_ARGS];
1267 /* First, check for complete junk; i.e. invalid options */
1268 error = xfs_test_remount_options(sb, options);
1272 sync_filesystem(sb);
1273 while ((p = strsep(&options, ",")) != NULL) {
1279 token = match_token(p, tokens, args);
1282 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1283 mp->m_flags |= XFS_MOUNT_BARRIER;
1286 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1287 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1290 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1291 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1294 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1295 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1299 * Logically we would return an error here to prevent
1300 * users from believing they might have changed
1301 * mount options using remount which can't be changed.
1303 * But unfortunately mount(8) adds all options from
1304 * mtab and fstab to the mount arguments in some cases
1305 * so we can't blindly reject options, but have to
1306 * check for each specified option if it actually
1307 * differs from the currently set option and only
1308 * reject it if that's the case.
1310 * Until that is implemented we return success for
1311 * every remount request, and silently ignore all
1312 * options that we can't actually change.
1316 "mount option \"%s\" not supported for remount", p);
1325 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1326 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1328 "ro->rw transition prohibited on norecovery mount");
1332 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1333 xfs_sb_has_ro_compat_feature(sbp,
1334 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1336 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1337 (sbp->sb_features_ro_compat &
1338 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1342 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1345 * If this is the first remount to writeable state we
1346 * might have some superblock changes to update.
1348 if (mp->m_update_sb) {
1349 error = xfs_sync_sb(mp, false);
1351 xfs_warn(mp, "failed to write sb changes");
1354 mp->m_update_sb = false;
1358 * Fill out the reserve pool if it is empty. Use the stashed
1359 * value if it is non-zero, otherwise go with the default.
1361 xfs_restore_resvblks(mp);
1362 xfs_log_work_queue(mp);
1364 /* Recover any CoW blocks that never got remapped. */
1365 error = xfs_reflink_recover_cow(mp);
1368 "Error %d recovering leftover CoW allocations.", error);
1369 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1372 xfs_icache_enable_reclaim(mp);
1374 /* Create the per-AG metadata reservation pool .*/
1375 error = xfs_fs_reserve_ag_blocks(mp);
1376 if (error && error != -ENOSPC)
1381 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1383 * Cancel background eofb scanning so it cannot race with the
1384 * final log force+buftarg wait and deadlock the remount.
1386 xfs_icache_disable_reclaim(mp);
1388 /* Get rid of any leftover CoW reservations... */
1389 error = xfs_icache_free_cowblocks(mp, NULL);
1391 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1395 /* Free the per-AG metadata reservation pool. */
1396 error = xfs_fs_unreserve_ag_blocks(mp);
1398 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1403 * Before we sync the metadata, we need to free up the reserve
1404 * block pool so that the used block count in the superblock on
1405 * disk is correct at the end of the remount. Stash the current
1406 * reserve pool size so that if we get remounted rw, we can
1407 * return it to the same size.
1409 xfs_save_resvblks(mp);
1411 xfs_quiesce_attr(mp);
1412 mp->m_flags |= XFS_MOUNT_RDONLY;
1419 * Second stage of a freeze. The data is already frozen so we only
1420 * need to take care of the metadata. Once that's done sync the superblock
1421 * to the log to dirty it in case of a crash while frozen. This ensures that we
1422 * will recover the unlinked inode lists on the next mount.
1426 struct super_block *sb)
1428 struct xfs_mount *mp = XFS_M(sb);
1430 xfs_icache_disable_reclaim(mp);
1431 xfs_save_resvblks(mp);
1432 xfs_quiesce_attr(mp);
1433 return xfs_sync_sb(mp, true);
1438 struct super_block *sb)
1440 struct xfs_mount *mp = XFS_M(sb);
1442 xfs_restore_resvblks(mp);
1443 xfs_log_work_queue(mp);
1444 xfs_icache_enable_reclaim(mp);
1449 xfs_fs_show_options(
1451 struct dentry *root)
1453 return xfs_showargs(XFS_M(root->d_sb), m);
1457 * This function fills in xfs_mount_t fields based on mount args.
1458 * Note: the superblock _has_ now been read in.
1462 struct xfs_mount *mp)
1464 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1466 /* Fail a mount where the logbuf is smaller than the log stripe */
1467 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1468 if (mp->m_logbsize <= 0 &&
1469 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1470 mp->m_logbsize = mp->m_sb.sb_logsunit;
1471 } else if (mp->m_logbsize > 0 &&
1472 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1474 "logbuf size must be greater than or equal to log stripe size");
1478 /* Fail a mount if the logbuf is larger than 32K */
1479 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1481 "logbuf size for version 1 logs must be 16K or 32K");
1487 * V5 filesystems always use attr2 format for attributes.
1489 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1490 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1491 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1492 "attr2 is always enabled for V5 filesystems.");
1497 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1498 * told by noattr2 to turn it off
1500 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1501 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1502 mp->m_flags |= XFS_MOUNT_ATTR2;
1505 * prohibit r/w mounts of read-only filesystems
1507 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1509 "cannot mount a read-only filesystem as read-write");
1513 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1514 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1515 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1517 "Super block does not support project and group quota together");
1525 xfs_init_percpu_counters(
1526 struct xfs_mount *mp)
1530 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1534 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1538 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1545 percpu_counter_destroy(&mp->m_ifree);
1547 percpu_counter_destroy(&mp->m_icount);
1552 xfs_reinit_percpu_counters(
1553 struct xfs_mount *mp)
1555 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1556 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1557 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1561 xfs_destroy_percpu_counters(
1562 struct xfs_mount *mp)
1564 percpu_counter_destroy(&mp->m_icount);
1565 percpu_counter_destroy(&mp->m_ifree);
1566 percpu_counter_destroy(&mp->m_fdblocks);
1569 static struct xfs_mount *
1571 struct super_block *sb)
1573 struct xfs_mount *mp;
1575 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1580 spin_lock_init(&mp->m_sb_lock);
1581 spin_lock_init(&mp->m_agirotor_lock);
1582 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1583 spin_lock_init(&mp->m_perag_lock);
1584 mutex_init(&mp->m_growlock);
1585 atomic_set(&mp->m_active_trans, 0);
1586 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1587 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1588 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1589 mp->m_kobj.kobject.kset = xfs_kset;
1596 struct super_block *sb,
1601 struct xfs_mount *mp = NULL;
1602 int flags = 0, error = -ENOMEM;
1605 * allocate mp and do all low-level struct initializations before we
1606 * attach it to the super
1608 mp = xfs_mount_alloc(sb);
1613 error = xfs_parseargs(mp, (char *)data);
1615 goto out_free_fsname;
1617 sb_min_blocksize(sb, BBSIZE);
1618 sb->s_xattr = xfs_xattr_handlers;
1619 sb->s_export_op = &xfs_export_operations;
1620 #ifdef CONFIG_XFS_QUOTA
1621 sb->s_qcop = &xfs_quotactl_operations;
1622 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1624 sb->s_op = &xfs_super_operations;
1627 * Delay mount work if the debug hook is set. This is debug
1628 * instrumention to coordinate simulation of xfs mount failures with
1629 * VFS superblock operations
1631 if (xfs_globals.mount_delay) {
1632 xfs_notice(mp, "Delaying mount for %d seconds.",
1633 xfs_globals.mount_delay);
1634 msleep(xfs_globals.mount_delay * 1000);
1638 flags |= XFS_MFSI_QUIET;
1640 error = xfs_open_devices(mp);
1642 goto out_free_fsname;
1644 error = xfs_init_mount_workqueues(mp);
1646 goto out_close_devices;
1648 error = xfs_init_percpu_counters(mp);
1650 goto out_destroy_workqueues;
1652 /* Allocate stats memory before we do operations that might use it */
1653 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1654 if (!mp->m_stats.xs_stats) {
1656 goto out_destroy_counters;
1659 error = xfs_readsb(mp, flags);
1661 goto out_free_stats;
1663 error = xfs_finish_flags(mp);
1667 error = xfs_setup_devices(mp);
1671 error = xfs_filestream_mount(mp);
1676 * we must configure the block size in the superblock before we run the
1677 * full mount process as the mount process can lookup and cache inodes.
1679 sb->s_magic = XFS_SB_MAGIC;
1680 sb->s_blocksize = mp->m_sb.sb_blocksize;
1681 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1682 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1683 sb->s_max_links = XFS_MAXLINK;
1684 sb->s_time_gran = 1;
1685 set_posix_acl_flag(sb);
1687 /* version 5 superblocks support inode version counters. */
1688 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1689 sb->s_flags |= SB_I_VERSION;
1691 if (mp->m_flags & XFS_MOUNT_DAX) {
1692 bool rtdev_is_dax = false, datadev_is_dax;
1695 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1697 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1699 if (mp->m_rtdev_targp)
1700 rtdev_is_dax = bdev_dax_supported(
1701 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1702 if (!rtdev_is_dax && !datadev_is_dax) {
1704 "DAX unsupported by block device. Turning off DAX.");
1705 mp->m_flags &= ~XFS_MOUNT_DAX;
1707 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1709 "DAX and reflink cannot be used together!");
1711 goto out_filestream_unmount;
1715 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1716 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1718 if (!blk_queue_discard(q)) {
1719 xfs_warn(mp, "mounting with \"discard\" option, but "
1720 "the device does not support discard");
1721 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1725 if (xfs_sb_version_hasreflink(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1727 "reflink not compatible with realtime device!");
1729 goto out_filestream_unmount;
1732 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1734 "reverse mapping btree not compatible with realtime device!");
1736 goto out_filestream_unmount;
1739 error = xfs_mountfs(mp);
1741 goto out_filestream_unmount;
1743 root = igrab(VFS_I(mp->m_rootip));
1748 sb->s_root = d_make_root(root);
1756 out_filestream_unmount:
1757 xfs_filestream_unmount(mp);
1761 free_percpu(mp->m_stats.xs_stats);
1762 out_destroy_counters:
1763 xfs_destroy_percpu_counters(mp);
1764 out_destroy_workqueues:
1765 xfs_destroy_mount_workqueues(mp);
1767 xfs_close_devices(mp);
1769 sb->s_fs_info = NULL;
1770 xfs_free_fsname(mp);
1776 xfs_filestream_unmount(mp);
1783 struct super_block *sb)
1785 struct xfs_mount *mp = XFS_M(sb);
1787 /* if ->fill_super failed, we have no mount to tear down */
1791 xfs_notice(mp, "Unmounting Filesystem");
1792 xfs_filestream_unmount(mp);
1796 free_percpu(mp->m_stats.xs_stats);
1797 xfs_destroy_percpu_counters(mp);
1798 xfs_destroy_mount_workqueues(mp);
1799 xfs_close_devices(mp);
1801 sb->s_fs_info = NULL;
1802 xfs_free_fsname(mp);
1806 STATIC struct dentry *
1808 struct file_system_type *fs_type,
1810 const char *dev_name,
1813 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1817 xfs_fs_nr_cached_objects(
1818 struct super_block *sb,
1819 struct shrink_control *sc)
1821 /* Paranoia: catch incorrect calls during mount setup or teardown */
1822 if (WARN_ON_ONCE(!sb->s_fs_info))
1824 return xfs_reclaim_inodes_count(XFS_M(sb));
1828 xfs_fs_free_cached_objects(
1829 struct super_block *sb,
1830 struct shrink_control *sc)
1832 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1835 static const struct super_operations xfs_super_operations = {
1836 .alloc_inode = xfs_fs_alloc_inode,
1837 .destroy_inode = xfs_fs_destroy_inode,
1838 .dirty_inode = xfs_fs_dirty_inode,
1839 .drop_inode = xfs_fs_drop_inode,
1840 .put_super = xfs_fs_put_super,
1841 .sync_fs = xfs_fs_sync_fs,
1842 .freeze_fs = xfs_fs_freeze,
1843 .unfreeze_fs = xfs_fs_unfreeze,
1844 .statfs = xfs_fs_statfs,
1845 .remount_fs = xfs_fs_remount,
1846 .show_options = xfs_fs_show_options,
1847 .nr_cached_objects = xfs_fs_nr_cached_objects,
1848 .free_cached_objects = xfs_fs_free_cached_objects,
1851 static struct file_system_type xfs_fs_type = {
1852 .owner = THIS_MODULE,
1854 .mount = xfs_fs_mount,
1855 .kill_sb = kill_block_super,
1856 .fs_flags = FS_REQUIRES_DEV,
1858 MODULE_ALIAS_FS("xfs");
1861 xfs_init_zones(void)
1863 xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1864 offsetof(struct xfs_ioend, io_inline_bio),
1866 if (!xfs_ioend_bioset)
1869 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1871 if (!xfs_log_ticket_zone)
1872 goto out_free_ioend_bioset;
1874 xfs_bmap_free_item_zone = kmem_zone_init(
1875 sizeof(struct xfs_extent_free_item),
1876 "xfs_bmap_free_item");
1877 if (!xfs_bmap_free_item_zone)
1878 goto out_destroy_log_ticket_zone;
1880 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1882 if (!xfs_btree_cur_zone)
1883 goto out_destroy_bmap_free_item_zone;
1885 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1887 if (!xfs_da_state_zone)
1888 goto out_destroy_btree_cur_zone;
1890 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1891 if (!xfs_ifork_zone)
1892 goto out_destroy_da_state_zone;
1894 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1895 if (!xfs_trans_zone)
1896 goto out_destroy_ifork_zone;
1900 * The size of the zone allocated buf log item is the maximum
1901 * size possible under XFS. This wastes a little bit of memory,
1902 * but it is much faster.
1904 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1906 if (!xfs_buf_item_zone)
1907 goto out_destroy_trans_zone;
1909 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1910 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1911 sizeof(xfs_extent_t))), "xfs_efd_item");
1913 goto out_destroy_buf_item_zone;
1915 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1916 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1917 sizeof(xfs_extent_t))), "xfs_efi_item");
1919 goto out_destroy_efd_zone;
1922 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1923 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1924 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1925 if (!xfs_inode_zone)
1926 goto out_destroy_efi_zone;
1929 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1930 KM_ZONE_SPREAD, NULL);
1932 goto out_destroy_inode_zone;
1933 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1935 if (!xfs_icreate_zone)
1936 goto out_destroy_ili_zone;
1938 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1941 goto out_destroy_icreate_zone;
1943 xfs_rui_zone = kmem_zone_init(
1944 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1947 goto out_destroy_rud_zone;
1949 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1952 goto out_destroy_rui_zone;
1954 xfs_cui_zone = kmem_zone_init(
1955 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1958 goto out_destroy_cud_zone;
1960 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1963 goto out_destroy_cui_zone;
1965 xfs_bui_zone = kmem_zone_init(
1966 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1969 goto out_destroy_bud_zone;
1973 out_destroy_bud_zone:
1974 kmem_zone_destroy(xfs_bud_zone);
1975 out_destroy_cui_zone:
1976 kmem_zone_destroy(xfs_cui_zone);
1977 out_destroy_cud_zone:
1978 kmem_zone_destroy(xfs_cud_zone);
1979 out_destroy_rui_zone:
1980 kmem_zone_destroy(xfs_rui_zone);
1981 out_destroy_rud_zone:
1982 kmem_zone_destroy(xfs_rud_zone);
1983 out_destroy_icreate_zone:
1984 kmem_zone_destroy(xfs_icreate_zone);
1985 out_destroy_ili_zone:
1986 kmem_zone_destroy(xfs_ili_zone);
1987 out_destroy_inode_zone:
1988 kmem_zone_destroy(xfs_inode_zone);
1989 out_destroy_efi_zone:
1990 kmem_zone_destroy(xfs_efi_zone);
1991 out_destroy_efd_zone:
1992 kmem_zone_destroy(xfs_efd_zone);
1993 out_destroy_buf_item_zone:
1994 kmem_zone_destroy(xfs_buf_item_zone);
1995 out_destroy_trans_zone:
1996 kmem_zone_destroy(xfs_trans_zone);
1997 out_destroy_ifork_zone:
1998 kmem_zone_destroy(xfs_ifork_zone);
1999 out_destroy_da_state_zone:
2000 kmem_zone_destroy(xfs_da_state_zone);
2001 out_destroy_btree_cur_zone:
2002 kmem_zone_destroy(xfs_btree_cur_zone);
2003 out_destroy_bmap_free_item_zone:
2004 kmem_zone_destroy(xfs_bmap_free_item_zone);
2005 out_destroy_log_ticket_zone:
2006 kmem_zone_destroy(xfs_log_ticket_zone);
2007 out_free_ioend_bioset:
2008 bioset_free(xfs_ioend_bioset);
2014 xfs_destroy_zones(void)
2017 * Make sure all delayed rcu free are flushed before we
2021 kmem_zone_destroy(xfs_bui_zone);
2022 kmem_zone_destroy(xfs_bud_zone);
2023 kmem_zone_destroy(xfs_cui_zone);
2024 kmem_zone_destroy(xfs_cud_zone);
2025 kmem_zone_destroy(xfs_rui_zone);
2026 kmem_zone_destroy(xfs_rud_zone);
2027 kmem_zone_destroy(xfs_icreate_zone);
2028 kmem_zone_destroy(xfs_ili_zone);
2029 kmem_zone_destroy(xfs_inode_zone);
2030 kmem_zone_destroy(xfs_efi_zone);
2031 kmem_zone_destroy(xfs_efd_zone);
2032 kmem_zone_destroy(xfs_buf_item_zone);
2033 kmem_zone_destroy(xfs_trans_zone);
2034 kmem_zone_destroy(xfs_ifork_zone);
2035 kmem_zone_destroy(xfs_da_state_zone);
2036 kmem_zone_destroy(xfs_btree_cur_zone);
2037 kmem_zone_destroy(xfs_bmap_free_item_zone);
2038 kmem_zone_destroy(xfs_log_ticket_zone);
2039 bioset_free(xfs_ioend_bioset);
2043 xfs_init_workqueues(void)
2046 * The allocation workqueue can be used in memory reclaim situations
2047 * (writepage path), and parallelism is only limited by the number of
2048 * AGs in all the filesystems mounted. Hence use the default large
2049 * max_active value for this workqueue.
2051 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2052 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2056 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2057 if (!xfs_discard_wq)
2058 goto out_free_alloc_wq;
2062 destroy_workqueue(xfs_alloc_wq);
2067 xfs_destroy_workqueues(void)
2069 destroy_workqueue(xfs_discard_wq);
2070 destroy_workqueue(xfs_alloc_wq);
2078 xfs_check_ondisk_structs();
2080 printk(KERN_INFO XFS_VERSION_STRING " with "
2081 XFS_BUILD_OPTIONS " enabled\n");
2083 xfs_extent_free_init_defer_op();
2084 xfs_rmap_update_init_defer_op();
2085 xfs_refcount_update_init_defer_op();
2086 xfs_bmap_update_init_defer_op();
2090 error = xfs_init_zones();
2094 error = xfs_init_workqueues();
2096 goto out_destroy_zones;
2098 error = xfs_mru_cache_init();
2100 goto out_destroy_wq;
2102 error = xfs_buf_init();
2104 goto out_mru_cache_uninit;
2106 error = xfs_init_procfs();
2108 goto out_buf_terminate;
2110 error = xfs_sysctl_register();
2112 goto out_cleanup_procfs;
2114 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2117 goto out_sysctl_unregister;
2120 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2122 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2123 if (!xfsstats.xs_stats) {
2125 goto out_kset_unregister;
2128 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2131 goto out_free_stats;
2134 xfs_dbg_kobj.kobject.kset = xfs_kset;
2135 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2137 goto out_remove_stats_kobj;
2140 error = xfs_qm_init();
2142 goto out_remove_dbg_kobj;
2144 error = register_filesystem(&xfs_fs_type);
2151 out_remove_dbg_kobj:
2153 xfs_sysfs_del(&xfs_dbg_kobj);
2154 out_remove_stats_kobj:
2156 xfs_sysfs_del(&xfsstats.xs_kobj);
2158 free_percpu(xfsstats.xs_stats);
2159 out_kset_unregister:
2160 kset_unregister(xfs_kset);
2161 out_sysctl_unregister:
2162 xfs_sysctl_unregister();
2164 xfs_cleanup_procfs();
2166 xfs_buf_terminate();
2167 out_mru_cache_uninit:
2168 xfs_mru_cache_uninit();
2170 xfs_destroy_workqueues();
2172 xfs_destroy_zones();
2181 unregister_filesystem(&xfs_fs_type);
2183 xfs_sysfs_del(&xfs_dbg_kobj);
2185 xfs_sysfs_del(&xfsstats.xs_kobj);
2186 free_percpu(xfsstats.xs_stats);
2187 kset_unregister(xfs_kset);
2188 xfs_sysctl_unregister();
2189 xfs_cleanup_procfs();
2190 xfs_buf_terminate();
2191 xfs_mru_cache_uninit();
2192 xfs_destroy_workqueues();
2193 xfs_destroy_zones();
2194 xfs_uuid_table_free();
2197 module_init(init_xfs_fs);
2198 module_exit(exit_xfs_fs);
2200 MODULE_AUTHOR("Silicon Graphics, Inc.");
2201 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2202 MODULE_LICENSE("GPL");