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
23 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_ialloc.h"
38 #include "xfs_rtalloc.h"
39 #include "xfs_error.h"
40 #include "xfs_itable.h"
41 #include "xfs_fsops.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_utils.h"
45 #include "xfs_vnodeops.h"
46 #include "xfs_log_priv.h"
47 #include "xfs_trans_priv.h"
48 #include "xfs_filestream.h"
49 #include "xfs_da_btree.h"
50 #include "xfs_extfree_item.h"
51 #include "xfs_mru_cache.h"
52 #include "xfs_inode_item.h"
54 #include "xfs_trace.h"
56 #include <linux/namei.h>
57 #include <linux/init.h>
58 #include <linux/slab.h>
59 #include <linux/mount.h>
60 #include <linux/mempool.h>
61 #include <linux/writeback.h>
62 #include <linux/kthread.h>
63 #include <linux/freezer.h>
64 #include <linux/parser.h>
66 static const struct super_operations xfs_super_operations;
67 static kmem_zone_t *xfs_ioend_zone;
68 mempool_t *xfs_ioend_pool;
70 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
71 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
72 #define MNTOPT_LOGDEV "logdev" /* log device */
73 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
74 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
75 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
76 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
77 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
78 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
79 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
80 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
81 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
82 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
83 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
84 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
85 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
86 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
87 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
88 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
89 * unwritten extent conversion */
90 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
91 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
92 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
93 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
94 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
95 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
97 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
98 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
99 #define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
100 #define MNTOPT_QUOTA "quota" /* disk quotas (user) */
101 #define MNTOPT_NOQUOTA "noquota" /* no quotas */
102 #define MNTOPT_USRQUOTA "usrquota" /* user quota enabled */
103 #define MNTOPT_GRPQUOTA "grpquota" /* group quota enabled */
104 #define MNTOPT_PRJQUOTA "prjquota" /* project quota enabled */
105 #define MNTOPT_UQUOTA "uquota" /* user quota (IRIX variant) */
106 #define MNTOPT_GQUOTA "gquota" /* group quota (IRIX variant) */
107 #define MNTOPT_PQUOTA "pquota" /* project quota (IRIX variant) */
108 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
109 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
110 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
111 #define MNTOPT_QUOTANOENF "qnoenforce" /* same as uqnoenforce */
112 #define MNTOPT_DELAYLOG "delaylog" /* Delayed logging enabled */
113 #define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed logging disabled */
114 #define MNTOPT_DISCARD "discard" /* Discard unused blocks */
115 #define MNTOPT_NODISCARD "nodiscard" /* Do not discard unused blocks */
118 * Table driven mount option parser.
120 * Currently only used for remount, but it will be used for mount
121 * in the future, too.
124 Opt_barrier, Opt_nobarrier, Opt_err
127 static const match_table_t tokens = {
128 {Opt_barrier, "barrier"},
129 {Opt_nobarrier, "nobarrier"},
135 suffix_strtoul(char *s, char **endp, unsigned int base)
137 int last, shift_left_factor = 0;
140 last = strlen(value) - 1;
141 if (value[last] == 'K' || value[last] == 'k') {
142 shift_left_factor = 10;
145 if (value[last] == 'M' || value[last] == 'm') {
146 shift_left_factor = 20;
149 if (value[last] == 'G' || value[last] == 'g') {
150 shift_left_factor = 30;
154 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
158 * This function fills in xfs_mount_t fields based on mount args.
159 * Note: the superblock has _not_ yet been read in.
161 * Note that this function leaks the various device name allocations on
162 * failure. The caller takes care of them.
166 struct xfs_mount *mp,
169 struct super_block *sb = mp->m_super;
170 char *this_char, *value, *eov;
174 __uint8_t iosizelog = 0;
177 * set up the mount name first so all the errors will refer to the
180 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
183 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
186 * Copy binary VFS mount flags we are interested in.
188 if (sb->s_flags & MS_RDONLY)
189 mp->m_flags |= XFS_MOUNT_RDONLY;
190 if (sb->s_flags & MS_DIRSYNC)
191 mp->m_flags |= XFS_MOUNT_DIRSYNC;
192 if (sb->s_flags & MS_SYNCHRONOUS)
193 mp->m_flags |= XFS_MOUNT_WSYNC;
196 * Set some default flags that could be cleared by the mount option
199 mp->m_flags |= XFS_MOUNT_BARRIER;
200 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
201 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
202 mp->m_flags |= XFS_MOUNT_DELAYLOG;
205 * These can be overridden by the mount option parsing.
213 while ((this_char = strsep(&options, ",")) != NULL) {
216 if ((value = strchr(this_char, '=')) != NULL)
219 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
220 if (!value || !*value) {
221 xfs_warn(mp, "%s option requires an argument",
225 mp->m_logbufs = simple_strtoul(value, &eov, 10);
226 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
227 if (!value || !*value) {
228 xfs_warn(mp, "%s option requires an argument",
232 mp->m_logbsize = suffix_strtoul(value, &eov, 10);
233 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
234 if (!value || !*value) {
235 xfs_warn(mp, "%s option requires an argument",
239 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
242 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
243 xfs_warn(mp, "%s option not allowed on this system",
246 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
247 if (!value || !*value) {
248 xfs_warn(mp, "%s option requires an argument",
252 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
255 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
256 if (!value || !*value) {
257 xfs_warn(mp, "%s option requires an argument",
261 iosize = simple_strtoul(value, &eov, 10);
262 iosizelog = ffs(iosize) - 1;
263 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
264 if (!value || !*value) {
265 xfs_warn(mp, "%s option requires an argument",
269 iosize = suffix_strtoul(value, &eov, 10);
270 iosizelog = ffs(iosize) - 1;
271 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
272 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
273 mp->m_flags |= XFS_MOUNT_GRPID;
274 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
275 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
276 mp->m_flags &= ~XFS_MOUNT_GRPID;
277 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
278 mp->m_flags |= XFS_MOUNT_WSYNC;
279 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
280 mp->m_flags |= XFS_MOUNT_NORECOVERY;
281 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
282 mp->m_flags |= XFS_MOUNT_NOALIGN;
283 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
284 mp->m_flags |= XFS_MOUNT_SWALLOC;
285 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
286 if (!value || !*value) {
287 xfs_warn(mp, "%s option requires an argument",
291 dsunit = simple_strtoul(value, &eov, 10);
292 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
293 if (!value || !*value) {
294 xfs_warn(mp, "%s option requires an argument",
298 dswidth = simple_strtoul(value, &eov, 10);
299 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
300 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
302 xfs_warn(mp, "%s option not allowed on this system",
306 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
307 mp->m_flags |= XFS_MOUNT_NOUUID;
308 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
309 mp->m_flags |= XFS_MOUNT_BARRIER;
310 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
311 mp->m_flags &= ~XFS_MOUNT_BARRIER;
312 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
313 mp->m_flags |= XFS_MOUNT_IKEEP;
314 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
315 mp->m_flags &= ~XFS_MOUNT_IKEEP;
316 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
317 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
318 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
319 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
320 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
321 mp->m_flags |= XFS_MOUNT_ATTR2;
322 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
323 mp->m_flags &= ~XFS_MOUNT_ATTR2;
324 mp->m_flags |= XFS_MOUNT_NOATTR2;
325 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
326 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
327 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
328 mp->m_qflags &= ~(XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
329 XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
330 XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
331 XFS_UQUOTA_ENFD | XFS_OQUOTA_ENFD);
332 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
333 !strcmp(this_char, MNTOPT_UQUOTA) ||
334 !strcmp(this_char, MNTOPT_USRQUOTA)) {
335 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
337 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
338 !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
339 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
340 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
341 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
342 !strcmp(this_char, MNTOPT_PRJQUOTA)) {
343 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
345 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
346 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
347 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
348 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
349 !strcmp(this_char, MNTOPT_GRPQUOTA)) {
350 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
352 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
353 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
354 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
355 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
356 mp->m_flags |= XFS_MOUNT_DELAYLOG;
357 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
358 mp->m_flags &= ~XFS_MOUNT_DELAYLOG;
359 } else if (!strcmp(this_char, MNTOPT_DISCARD)) {
360 mp->m_flags |= XFS_MOUNT_DISCARD;
361 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
362 mp->m_flags &= ~XFS_MOUNT_DISCARD;
363 } else if (!strcmp(this_char, "ihashsize")) {
365 "ihashsize no longer used, option is deprecated.");
366 } else if (!strcmp(this_char, "osyncisdsync")) {
368 "osyncisdsync has no effect, option is deprecated.");
369 } else if (!strcmp(this_char, "osyncisosync")) {
371 "osyncisosync has no effect, option is deprecated.");
372 } else if (!strcmp(this_char, "irixsgid")) {
374 "irixsgid is now a sysctl(2) variable, option is deprecated.");
376 xfs_warn(mp, "unknown mount option [%s].", this_char);
382 * no recovery flag requires a read-only mount
384 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
385 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
386 xfs_warn(mp, "no-recovery mounts must be read-only.");
390 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
392 "sunit and swidth options incompatible with the noalign option");
396 if ((mp->m_flags & XFS_MOUNT_DISCARD) &&
397 !(mp->m_flags & XFS_MOUNT_DELAYLOG)) {
399 "the discard option is incompatible with the nodelaylog 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 ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
411 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
412 xfs_warn(mp, "cannot mount with both project and group quota");
416 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
417 xfs_warn(mp, "sunit and swidth must be specified together");
421 if (dsunit && (dswidth % dsunit != 0)) {
423 "stripe width (%d) must be a multiple of the stripe unit (%d)",
429 if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
431 * At this point the superblock has not been read
432 * in, therefore we do not know the block size.
433 * Before the mount call ends we will convert
437 mp->m_dalign = dsunit;
438 mp->m_flags |= XFS_MOUNT_RETERR;
442 mp->m_swidth = dswidth;
445 if (mp->m_logbufs != -1 &&
446 mp->m_logbufs != 0 &&
447 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
448 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
449 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
450 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
451 return XFS_ERROR(EINVAL);
453 if (mp->m_logbsize != -1 &&
454 mp->m_logbsize != 0 &&
455 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
456 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
457 !is_power_of_2(mp->m_logbsize))) {
459 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
461 return XFS_ERROR(EINVAL);
465 if (iosizelog > XFS_MAX_IO_LOG ||
466 iosizelog < XFS_MIN_IO_LOG) {
467 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
468 iosizelog, XFS_MIN_IO_LOG,
470 return XFS_ERROR(EINVAL);
473 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
474 mp->m_readio_log = iosizelog;
475 mp->m_writeio_log = iosizelog;
481 struct proc_xfs_info {
488 struct xfs_mount *mp,
491 static struct proc_xfs_info xfs_info_set[] = {
492 /* the few simple ones we can get from the mount struct */
493 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP },
494 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
495 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
496 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
497 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
498 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
499 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 },
500 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
501 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
502 { XFS_MOUNT_DELAYLOG, "," MNTOPT_DELAYLOG },
503 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD },
506 static struct proc_xfs_info xfs_info_unset[] = {
507 /* the few simple ones we can get from the mount struct */
508 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO },
509 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER },
510 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE },
513 struct proc_xfs_info *xfs_infop;
515 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
516 if (mp->m_flags & xfs_infop->flag)
517 seq_puts(m, xfs_infop->str);
519 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
520 if (!(mp->m_flags & xfs_infop->flag))
521 seq_puts(m, xfs_infop->str);
524 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
525 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
526 (int)(1 << mp->m_writeio_log) >> 10);
528 if (mp->m_logbufs > 0)
529 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
530 if (mp->m_logbsize > 0)
531 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
534 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
536 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
538 if (mp->m_dalign > 0)
539 seq_printf(m, "," MNTOPT_SUNIT "=%d",
540 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
541 if (mp->m_swidth > 0)
542 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
543 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
545 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
546 seq_puts(m, "," MNTOPT_USRQUOTA);
547 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
548 seq_puts(m, "," MNTOPT_UQUOTANOENF);
550 /* Either project or group quotas can be active, not both */
552 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
553 if (mp->m_qflags & XFS_OQUOTA_ENFD)
554 seq_puts(m, "," MNTOPT_PRJQUOTA);
556 seq_puts(m, "," MNTOPT_PQUOTANOENF);
557 } else if (mp->m_qflags & XFS_GQUOTA_ACCT) {
558 if (mp->m_qflags & XFS_OQUOTA_ENFD)
559 seq_puts(m, "," MNTOPT_GRPQUOTA);
561 seq_puts(m, "," MNTOPT_GQUOTANOENF);
564 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
565 seq_puts(m, "," MNTOPT_NOQUOTA);
571 unsigned int blockshift)
573 unsigned int pagefactor = 1;
574 unsigned int bitshift = BITS_PER_LONG - 1;
576 /* Figure out maximum filesize, on Linux this can depend on
577 * the filesystem blocksize (on 32 bit platforms).
578 * __block_write_begin does this in an [unsigned] long...
579 * page->index << (PAGE_CACHE_SHIFT - bbits)
580 * So, for page sized blocks (4K on 32 bit platforms),
581 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
582 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
583 * but for smaller blocksizes it is less (bbits = log2 bsize).
584 * Note1: get_block_t takes a long (implicit cast from above)
585 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
586 * can optionally convert the [unsigned] long from above into
587 * an [unsigned] long long.
590 #if BITS_PER_LONG == 32
591 # if defined(CONFIG_LBDAF)
592 ASSERT(sizeof(sector_t) == 8);
593 pagefactor = PAGE_CACHE_SIZE;
594 bitshift = BITS_PER_LONG;
596 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
600 return (((__uint64_t)pagefactor) << bitshift) - 1;
607 struct block_device **bdevp)
611 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
613 if (IS_ERR(*bdevp)) {
614 error = PTR_ERR(*bdevp);
615 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
623 struct block_device *bdev)
626 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
630 xfs_blkdev_issue_flush(
631 xfs_buftarg_t *buftarg)
633 blkdev_issue_flush(buftarg->bt_bdev, GFP_KERNEL, NULL);
638 struct xfs_mount *mp)
640 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
641 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
642 xfs_free_buftarg(mp, mp->m_logdev_targp);
643 xfs_blkdev_put(logdev);
645 if (mp->m_rtdev_targp) {
646 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
647 xfs_free_buftarg(mp, mp->m_rtdev_targp);
648 xfs_blkdev_put(rtdev);
650 xfs_free_buftarg(mp, mp->m_ddev_targp);
654 * The file system configurations are:
655 * (1) device (partition) with data and internal log
656 * (2) logical volume with data and log subvolumes.
657 * (3) logical volume with data, log, and realtime subvolumes.
659 * We only have to handle opening the log and realtime volumes here if
660 * they are present. The data subvolume has already been opened by
661 * get_sb_bdev() and is stored in sb->s_bdev.
665 struct xfs_mount *mp)
667 struct block_device *ddev = mp->m_super->s_bdev;
668 struct block_device *logdev = NULL, *rtdev = NULL;
672 * Open real time and log devices - order is important.
675 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
681 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
683 goto out_close_logdev;
685 if (rtdev == ddev || rtdev == logdev) {
687 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
689 goto out_close_rtdev;
694 * Setup xfs_mount buffer target pointers
697 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
698 if (!mp->m_ddev_targp)
699 goto out_close_rtdev;
702 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
704 if (!mp->m_rtdev_targp)
705 goto out_free_ddev_targ;
708 if (logdev && logdev != ddev) {
709 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
711 if (!mp->m_logdev_targp)
712 goto out_free_rtdev_targ;
714 mp->m_logdev_targp = mp->m_ddev_targp;
720 if (mp->m_rtdev_targp)
721 xfs_free_buftarg(mp, mp->m_rtdev_targp);
723 xfs_free_buftarg(mp, mp->m_ddev_targp);
726 xfs_blkdev_put(rtdev);
728 if (logdev && logdev != ddev)
729 xfs_blkdev_put(logdev);
735 * Setup xfs_mount buffer target pointers based on superblock
739 struct xfs_mount *mp)
743 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
744 mp->m_sb.sb_sectsize);
748 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
749 unsigned int log_sector_size = BBSIZE;
751 if (xfs_sb_version_hassector(&mp->m_sb))
752 log_sector_size = mp->m_sb.sb_logsectsize;
753 error = xfs_setsize_buftarg(mp->m_logdev_targp,
754 mp->m_sb.sb_blocksize,
759 if (mp->m_rtdev_targp) {
760 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
761 mp->m_sb.sb_blocksize,
762 mp->m_sb.sb_sectsize);
770 /* Catch misguided souls that try to use this interface on XFS */
771 STATIC struct inode *
773 struct super_block *sb)
780 * Now that the generic code is guaranteed not to be accessing
781 * the linux inode, we can reclaim the inode.
784 xfs_fs_destroy_inode(
787 struct xfs_inode *ip = XFS_I(inode);
789 trace_xfs_destroy_inode(ip);
791 XFS_STATS_INC(vn_reclaim);
793 /* bad inode, get out here ASAP */
794 if (is_bad_inode(inode))
799 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
802 * We should never get here with one of the reclaim flags already set.
804 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
805 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
808 * We always use background reclaim here because even if the
809 * inode is clean, it still may be under IO and hence we have
810 * to take the flush lock. The background reclaim path handles
811 * this more efficiently than we can here, so simply let background
812 * reclaim tear down all inodes.
815 xfs_inode_set_reclaim_tag(ip);
819 * Slab object creation initialisation for the XFS inode.
820 * This covers only the idempotent fields in the XFS inode;
821 * all other fields need to be initialised on allocation
822 * from the slab. This avoids the need to repeatedly initialise
823 * fields in the xfs inode that left in the initialise state
824 * when freeing the inode.
827 xfs_fs_inode_init_once(
830 struct xfs_inode *ip = inode;
832 memset(ip, 0, sizeof(struct xfs_inode));
835 inode_init_once(VFS_I(ip));
838 atomic_set(&ip->i_iocount, 0);
839 atomic_set(&ip->i_pincount, 0);
840 spin_lock_init(&ip->i_flags_lock);
841 init_waitqueue_head(&ip->i_ipin_wait);
843 * Because we want to use a counting completion, complete
844 * the flush completion once to allow a single access to
845 * the flush completion without blocking.
847 init_completion(&ip->i_flush);
848 complete(&ip->i_flush);
850 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
851 "xfsino", ip->i_ino);
855 * Dirty the XFS inode when mark_inode_dirty_sync() is called so that
856 * we catch unlogged VFS level updates to the inode.
858 * We need the barrier() to maintain correct ordering between unlogged
859 * updates and the transaction commit code that clears the i_update_core
860 * field. This requires all updates to be completed before marking the
869 XFS_I(inode)->i_update_core = 1;
874 struct xfs_inode *ip)
876 struct xfs_mount *mp = ip->i_mount;
877 struct xfs_trans *tp;
880 xfs_iunlock(ip, XFS_ILOCK_SHARED);
881 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
882 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
885 xfs_trans_cancel(tp, 0);
886 /* we need to return with the lock hold shared */
887 xfs_ilock(ip, XFS_ILOCK_SHARED);
891 xfs_ilock(ip, XFS_ILOCK_EXCL);
894 * Note - it's possible that we might have pushed ourselves out of the
895 * way during trans_reserve which would flush the inode. But there's
896 * no guarantee that the inode buffer has actually gone out yet (it's
897 * delwri). Plus the buffer could be pinned anyway if it's part of
898 * an inode in another recent transaction. So we play it safe and
899 * fire off the transaction anyway.
901 xfs_trans_ijoin(tp, ip);
902 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
903 error = xfs_trans_commit(tp, 0);
904 xfs_ilock_demote(ip, XFS_ILOCK_EXCL);
912 struct writeback_control *wbc)
914 struct xfs_inode *ip = XFS_I(inode);
915 struct xfs_mount *mp = ip->i_mount;
918 trace_xfs_write_inode(ip);
920 if (XFS_FORCED_SHUTDOWN(mp))
921 return XFS_ERROR(EIO);
923 if (wbc->sync_mode == WB_SYNC_ALL) {
925 * Make sure the inode has made it it into the log. Instead
926 * of forcing it all the way to stable storage using a
927 * synchronous transaction we let the log force inside the
928 * ->sync_fs call do that for thus, which reduces the number
929 * of synchronous log foces dramatically.
932 xfs_ilock(ip, XFS_ILOCK_SHARED);
933 if (ip->i_update_core) {
934 error = xfs_log_inode(ip);
940 * We make this non-blocking if the inode is contended, return
941 * EAGAIN to indicate to the caller that they did not succeed.
942 * This prevents the flush path from blocking on inodes inside
943 * another operation right now, they get caught later by
946 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
949 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip))
953 * Now we have the flush lock and the inode is not pinned, we
954 * can check if the inode is really clean as we know that
955 * there are no pending transaction completions, it is not
956 * waiting on the delayed write queue and there is no IO in
959 if (xfs_inode_clean(ip)) {
964 error = xfs_iflush(ip, SYNC_TRYLOCK);
968 xfs_iunlock(ip, XFS_ILOCK_SHARED);
971 * if we failed to write out the inode then mark
972 * it dirty again so we'll try again later.
975 xfs_mark_inode_dirty_sync(ip);
983 xfs_inode_t *ip = XFS_I(inode);
985 trace_xfs_evict_inode(ip);
987 truncate_inode_pages(&inode->i_data, 0);
988 end_writeback(inode);
989 XFS_STATS_INC(vn_rele);
990 XFS_STATS_INC(vn_remove);
991 XFS_STATS_DEC(vn_active);
994 * The iolock is used by the file system to coordinate reads,
995 * writes, and block truncates. Up to this point the lock
996 * protected concurrent accesses by users of the inode. But
997 * from here forward we're doing some final processing of the
998 * inode because we're done with it, and although we reuse the
999 * iolock for protection it is really a distinct lock class
1000 * (in the lockdep sense) from before. To keep lockdep happy
1001 * (and basically indicate what we are doing), we explicitly
1002 * re-init the iolock here.
1004 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
1005 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
1006 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
1007 &xfs_iolock_reclaimable, "xfs_iolock_reclaimable");
1014 struct xfs_mount *mp)
1016 kfree(mp->m_fsname);
1017 kfree(mp->m_rtname);
1018 kfree(mp->m_logname);
1023 struct super_block *sb)
1025 struct xfs_mount *mp = XFS_M(sb);
1028 * Unregister the memory shrinker before we tear down the mount
1029 * structure so we don't have memory reclaim racing with us here.
1031 xfs_inode_shrinker_unregister(mp);
1035 * Blow away any referenced inode in the filestreams cache.
1036 * This can and will cause log traffic as inodes go inactive
1039 xfs_filestream_unmount(mp);
1041 XFS_bflush(mp->m_ddev_targp);
1045 xfs_icsb_destroy_counters(mp);
1046 xfs_close_devices(mp);
1047 xfs_free_fsname(mp);
1053 struct super_block *sb,
1056 struct xfs_mount *mp = XFS_M(sb);
1060 * Not much we can do for the first async pass. Writing out the
1061 * superblock would be counter-productive as we are going to redirty
1062 * when writing out other data and metadata (and writing out a single
1063 * block is quite fast anyway).
1065 * Try to asynchronously kick off quota syncing at least.
1068 xfs_qm_sync(mp, SYNC_TRYLOCK);
1072 error = xfs_quiesce_data(mp);
1078 * The disk must be active because we're syncing.
1079 * We schedule xfssyncd now (now that the disk is
1080 * active) instead of later (when it might not be).
1082 flush_delayed_work_sync(&mp->m_sync_work);
1090 struct dentry *dentry,
1091 struct kstatfs *statp)
1093 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1094 xfs_sb_t *sbp = &mp->m_sb;
1095 struct xfs_inode *ip = XFS_I(dentry->d_inode);
1096 __uint64_t fakeinos, id;
1100 statp->f_type = XFS_SB_MAGIC;
1101 statp->f_namelen = MAXNAMELEN - 1;
1103 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1104 statp->f_fsid.val[0] = (u32)id;
1105 statp->f_fsid.val[1] = (u32)(id >> 32);
1107 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1109 spin_lock(&mp->m_sb_lock);
1110 statp->f_bsize = sbp->sb_blocksize;
1111 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1112 statp->f_blocks = sbp->sb_dblocks - lsize;
1113 statp->f_bfree = statp->f_bavail =
1114 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1115 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1117 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1118 if (mp->m_maxicount)
1119 statp->f_files = min_t(typeof(statp->f_files),
1123 /* make sure statp->f_ffree does not underflow */
1124 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1125 statp->f_ffree = max_t(__int64_t, ffree, 0);
1127 spin_unlock(&mp->m_sb_lock);
1129 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) ||
1130 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) ==
1131 (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
1132 xfs_qm_statvfs(ip, statp);
1137 xfs_save_resvblks(struct xfs_mount *mp)
1139 __uint64_t resblks = 0;
1141 mp->m_resblks_save = mp->m_resblks;
1142 xfs_reserve_blocks(mp, &resblks, NULL);
1146 xfs_restore_resvblks(struct xfs_mount *mp)
1150 if (mp->m_resblks_save) {
1151 resblks = mp->m_resblks_save;
1152 mp->m_resblks_save = 0;
1154 resblks = xfs_default_resblks(mp);
1156 xfs_reserve_blocks(mp, &resblks, NULL);
1161 struct super_block *sb,
1165 struct xfs_mount *mp = XFS_M(sb);
1166 substring_t args[MAX_OPT_ARGS];
1170 while ((p = strsep(&options, ",")) != NULL) {
1176 token = match_token(p, tokens, args);
1179 mp->m_flags |= XFS_MOUNT_BARRIER;
1182 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1186 * Logically we would return an error here to prevent
1187 * users from believing they might have changed
1188 * mount options using remount which can't be changed.
1190 * But unfortunately mount(8) adds all options from
1191 * mtab and fstab to the mount arguments in some cases
1192 * so we can't blindly reject options, but have to
1193 * check for each specified option if it actually
1194 * differs from the currently set option and only
1195 * reject it if that's the case.
1197 * Until that is implemented we return success for
1198 * every remount request, and silently ignore all
1199 * options that we can't actually change.
1203 "mount option \"%s\" not supported for remount\n", p);
1212 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1213 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1216 * If this is the first remount to writeable state we
1217 * might have some superblock changes to update.
1219 if (mp->m_update_flags) {
1220 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1222 xfs_warn(mp, "failed to write sb changes");
1225 mp->m_update_flags = 0;
1229 * Fill out the reserve pool if it is empty. Use the stashed
1230 * value if it is non-zero, otherwise go with the default.
1232 xfs_restore_resvblks(mp);
1236 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1238 * After we have synced the data but before we sync the
1239 * metadata, we need to free up the reserve block pool so that
1240 * the used block count in the superblock on disk is correct at
1241 * the end of the remount. Stash the current reserve pool size
1242 * so that if we get remounted rw, we can return it to the same
1246 xfs_quiesce_data(mp);
1247 xfs_save_resvblks(mp);
1248 xfs_quiesce_attr(mp);
1249 mp->m_flags |= XFS_MOUNT_RDONLY;
1256 * Second stage of a freeze. The data is already frozen so we only
1257 * need to take care of the metadata. Once that's done write a dummy
1258 * record to dirty the log in case of a crash while frozen.
1262 struct super_block *sb)
1264 struct xfs_mount *mp = XFS_M(sb);
1266 xfs_save_resvblks(mp);
1267 xfs_quiesce_attr(mp);
1268 return -xfs_fs_log_dummy(mp);
1273 struct super_block *sb)
1275 struct xfs_mount *mp = XFS_M(sb);
1277 xfs_restore_resvblks(mp);
1282 xfs_fs_show_options(
1284 struct vfsmount *mnt)
1286 return -xfs_showargs(XFS_M(mnt->mnt_sb), m);
1290 * This function fills in xfs_mount_t fields based on mount args.
1291 * Note: the superblock _has_ now been read in.
1295 struct xfs_mount *mp)
1297 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1299 /* Fail a mount where the logbuf is smaller than the log stripe */
1300 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1301 if (mp->m_logbsize <= 0 &&
1302 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1303 mp->m_logbsize = mp->m_sb.sb_logsunit;
1304 } else if (mp->m_logbsize > 0 &&
1305 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1307 "logbuf size must be greater than or equal to log stripe size");
1308 return XFS_ERROR(EINVAL);
1311 /* Fail a mount if the logbuf is larger than 32K */
1312 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1314 "logbuf size for version 1 logs must be 16K or 32K");
1315 return XFS_ERROR(EINVAL);
1320 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1321 * told by noattr2 to turn it off
1323 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1324 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1325 mp->m_flags |= XFS_MOUNT_ATTR2;
1328 * prohibit r/w mounts of read-only filesystems
1330 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1332 "cannot mount a read-only filesystem as read-write");
1333 return XFS_ERROR(EROFS);
1341 struct super_block *sb,
1346 struct xfs_mount *mp = NULL;
1347 int flags = 0, error = ENOMEM;
1349 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1353 spin_lock_init(&mp->m_sb_lock);
1354 mutex_init(&mp->m_growlock);
1355 atomic_set(&mp->m_active_trans, 0);
1360 error = xfs_parseargs(mp, (char *)data);
1362 goto out_free_fsname;
1364 sb_min_blocksize(sb, BBSIZE);
1365 sb->s_xattr = xfs_xattr_handlers;
1366 sb->s_export_op = &xfs_export_operations;
1367 #ifdef CONFIG_XFS_QUOTA
1368 sb->s_qcop = &xfs_quotactl_operations;
1370 sb->s_op = &xfs_super_operations;
1373 flags |= XFS_MFSI_QUIET;
1375 error = xfs_open_devices(mp);
1377 goto out_free_fsname;
1379 error = xfs_icsb_init_counters(mp);
1381 goto out_close_devices;
1383 error = xfs_readsb(mp, flags);
1385 goto out_destroy_counters;
1387 error = xfs_finish_flags(mp);
1391 error = xfs_setup_devices(mp);
1395 error = xfs_filestream_mount(mp);
1400 * we must configure the block size in the superblock before we run the
1401 * full mount process as the mount process can lookup and cache inodes.
1402 * For the same reason we must also initialise the syncd and register
1403 * the inode cache shrinker so that inodes can be reclaimed during
1404 * operations like a quotacheck that iterate all inodes in the
1407 sb->s_magic = XFS_SB_MAGIC;
1408 sb->s_blocksize = mp->m_sb.sb_blocksize;
1409 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1410 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1411 sb->s_time_gran = 1;
1412 set_posix_acl_flag(sb);
1414 xfs_inode_shrinker_register(mp);
1416 error = xfs_mountfs(mp);
1418 goto out_filestream_unmount;
1420 error = xfs_syncd_init(mp);
1424 root = igrab(VFS_I(mp->m_rootip));
1427 goto out_syncd_stop;
1429 if (is_bad_inode(root)) {
1431 goto out_syncd_stop;
1433 sb->s_root = d_alloc_root(root);
1441 out_filestream_unmount:
1442 xfs_inode_shrinker_unregister(mp);
1443 xfs_filestream_unmount(mp);
1446 out_destroy_counters:
1447 xfs_icsb_destroy_counters(mp);
1449 xfs_close_devices(mp);
1451 xfs_free_fsname(mp);
1461 xfs_inode_shrinker_unregister(mp);
1464 * Blow away any referenced inode in the filestreams cache.
1465 * This can and will cause log traffic as inodes go inactive
1468 xfs_filestream_unmount(mp);
1470 XFS_bflush(mp->m_ddev_targp);
1476 STATIC struct dentry *
1478 struct file_system_type *fs_type,
1480 const char *dev_name,
1483 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1486 static const struct super_operations xfs_super_operations = {
1487 .alloc_inode = xfs_fs_alloc_inode,
1488 .destroy_inode = xfs_fs_destroy_inode,
1489 .dirty_inode = xfs_fs_dirty_inode,
1490 .write_inode = xfs_fs_write_inode,
1491 .evict_inode = xfs_fs_evict_inode,
1492 .put_super = xfs_fs_put_super,
1493 .sync_fs = xfs_fs_sync_fs,
1494 .freeze_fs = xfs_fs_freeze,
1495 .unfreeze_fs = xfs_fs_unfreeze,
1496 .statfs = xfs_fs_statfs,
1497 .remount_fs = xfs_fs_remount,
1498 .show_options = xfs_fs_show_options,
1501 static struct file_system_type xfs_fs_type = {
1502 .owner = THIS_MODULE,
1504 .mount = xfs_fs_mount,
1505 .kill_sb = kill_block_super,
1506 .fs_flags = FS_REQUIRES_DEV,
1510 xfs_init_zones(void)
1513 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1514 if (!xfs_ioend_zone)
1517 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1519 if (!xfs_ioend_pool)
1520 goto out_destroy_ioend_zone;
1522 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1524 if (!xfs_log_ticket_zone)
1525 goto out_destroy_ioend_pool;
1527 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1528 "xfs_bmap_free_item");
1529 if (!xfs_bmap_free_item_zone)
1530 goto out_destroy_log_ticket_zone;
1532 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1534 if (!xfs_btree_cur_zone)
1535 goto out_destroy_bmap_free_item_zone;
1537 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1539 if (!xfs_da_state_zone)
1540 goto out_destroy_btree_cur_zone;
1542 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
1543 if (!xfs_dabuf_zone)
1544 goto out_destroy_da_state_zone;
1546 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1547 if (!xfs_ifork_zone)
1548 goto out_destroy_dabuf_zone;
1550 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1551 if (!xfs_trans_zone)
1552 goto out_destroy_ifork_zone;
1554 xfs_log_item_desc_zone =
1555 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1556 "xfs_log_item_desc");
1557 if (!xfs_log_item_desc_zone)
1558 goto out_destroy_trans_zone;
1561 * The size of the zone allocated buf log item is the maximum
1562 * size possible under XFS. This wastes a little bit of memory,
1563 * but it is much faster.
1565 xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) +
1566 (((XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK) /
1567 NBWORD) * sizeof(int))), "xfs_buf_item");
1568 if (!xfs_buf_item_zone)
1569 goto out_destroy_log_item_desc_zone;
1571 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1572 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1573 sizeof(xfs_extent_t))), "xfs_efd_item");
1575 goto out_destroy_buf_item_zone;
1577 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1578 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1579 sizeof(xfs_extent_t))), "xfs_efi_item");
1581 goto out_destroy_efd_zone;
1584 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1585 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1586 xfs_fs_inode_init_once);
1587 if (!xfs_inode_zone)
1588 goto out_destroy_efi_zone;
1591 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1592 KM_ZONE_SPREAD, NULL);
1594 goto out_destroy_inode_zone;
1598 out_destroy_inode_zone:
1599 kmem_zone_destroy(xfs_inode_zone);
1600 out_destroy_efi_zone:
1601 kmem_zone_destroy(xfs_efi_zone);
1602 out_destroy_efd_zone:
1603 kmem_zone_destroy(xfs_efd_zone);
1604 out_destroy_buf_item_zone:
1605 kmem_zone_destroy(xfs_buf_item_zone);
1606 out_destroy_log_item_desc_zone:
1607 kmem_zone_destroy(xfs_log_item_desc_zone);
1608 out_destroy_trans_zone:
1609 kmem_zone_destroy(xfs_trans_zone);
1610 out_destroy_ifork_zone:
1611 kmem_zone_destroy(xfs_ifork_zone);
1612 out_destroy_dabuf_zone:
1613 kmem_zone_destroy(xfs_dabuf_zone);
1614 out_destroy_da_state_zone:
1615 kmem_zone_destroy(xfs_da_state_zone);
1616 out_destroy_btree_cur_zone:
1617 kmem_zone_destroy(xfs_btree_cur_zone);
1618 out_destroy_bmap_free_item_zone:
1619 kmem_zone_destroy(xfs_bmap_free_item_zone);
1620 out_destroy_log_ticket_zone:
1621 kmem_zone_destroy(xfs_log_ticket_zone);
1622 out_destroy_ioend_pool:
1623 mempool_destroy(xfs_ioend_pool);
1624 out_destroy_ioend_zone:
1625 kmem_zone_destroy(xfs_ioend_zone);
1631 xfs_destroy_zones(void)
1633 kmem_zone_destroy(xfs_ili_zone);
1634 kmem_zone_destroy(xfs_inode_zone);
1635 kmem_zone_destroy(xfs_efi_zone);
1636 kmem_zone_destroy(xfs_efd_zone);
1637 kmem_zone_destroy(xfs_buf_item_zone);
1638 kmem_zone_destroy(xfs_log_item_desc_zone);
1639 kmem_zone_destroy(xfs_trans_zone);
1640 kmem_zone_destroy(xfs_ifork_zone);
1641 kmem_zone_destroy(xfs_dabuf_zone);
1642 kmem_zone_destroy(xfs_da_state_zone);
1643 kmem_zone_destroy(xfs_btree_cur_zone);
1644 kmem_zone_destroy(xfs_bmap_free_item_zone);
1645 kmem_zone_destroy(xfs_log_ticket_zone);
1646 mempool_destroy(xfs_ioend_pool);
1647 kmem_zone_destroy(xfs_ioend_zone);
1652 xfs_init_workqueues(void)
1655 * max_active is set to 8 to give enough concurency to allow
1656 * multiple work operations on each CPU to run. This allows multiple
1657 * filesystems to be running sync work concurrently, and scales with
1658 * the number of CPUs in the system.
1660 xfs_syncd_wq = alloc_workqueue("xfssyncd", WQ_CPU_INTENSIVE, 8);
1664 xfs_ail_wq = alloc_workqueue("xfsail", WQ_CPU_INTENSIVE, 8);
1666 goto out_destroy_syncd;
1671 destroy_workqueue(xfs_syncd_wq);
1677 xfs_destroy_workqueues(void)
1679 destroy_workqueue(xfs_ail_wq);
1680 destroy_workqueue(xfs_syncd_wq);
1688 printk(KERN_INFO XFS_VERSION_STRING " with "
1689 XFS_BUILD_OPTIONS " enabled\n");
1694 error = xfs_init_zones();
1698 error = xfs_init_workqueues();
1700 goto out_destroy_zones;
1702 error = xfs_mru_cache_init();
1704 goto out_destroy_wq;
1706 error = xfs_filestream_init();
1708 goto out_mru_cache_uninit;
1710 error = xfs_buf_init();
1712 goto out_filestream_uninit;
1714 error = xfs_init_procfs();
1716 goto out_buf_terminate;
1718 error = xfs_sysctl_register();
1720 goto out_cleanup_procfs;
1724 error = register_filesystem(&xfs_fs_type);
1726 goto out_sysctl_unregister;
1729 out_sysctl_unregister:
1730 xfs_sysctl_unregister();
1732 xfs_cleanup_procfs();
1734 xfs_buf_terminate();
1735 out_filestream_uninit:
1736 xfs_filestream_uninit();
1737 out_mru_cache_uninit:
1738 xfs_mru_cache_uninit();
1740 xfs_destroy_workqueues();
1742 xfs_destroy_zones();
1751 unregister_filesystem(&xfs_fs_type);
1752 xfs_sysctl_unregister();
1753 xfs_cleanup_procfs();
1754 xfs_buf_terminate();
1755 xfs_filestream_uninit();
1756 xfs_mru_cache_uninit();
1757 xfs_destroy_workqueues();
1758 xfs_destroy_zones();
1761 module_init(init_xfs_fs);
1762 module_exit(exit_xfs_fs);
1764 MODULE_AUTHOR("Silicon Graphics, Inc.");
1765 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1766 MODULE_LICENSE("GPL");