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
22 #include "xfs_trans.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_itable.h"
40 #include "xfs_fsops.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_utils.h"
44 #include "xfs_vnodeops.h"
45 #include "xfs_log_priv.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_filestream.h"
48 #include "xfs_da_btree.h"
49 #include "xfs_extfree_item.h"
50 #include "xfs_mru_cache.h"
51 #include "xfs_inode_item.h"
53 #include "xfs_trace.h"
55 #include <linux/namei.h>
56 #include <linux/init.h>
57 #include <linux/slab.h>
58 #include <linux/mount.h>
59 #include <linux/mempool.h>
60 #include <linux/writeback.h>
61 #include <linux/kthread.h>
62 #include <linux/freezer.h>
63 #include <linux/parser.h>
65 static const struct super_operations xfs_super_operations;
66 static kmem_zone_t *xfs_ioend_zone;
67 mempool_t *xfs_ioend_pool;
69 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
70 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
71 #define MNTOPT_LOGDEV "logdev" /* log device */
72 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
73 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
74 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
75 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
76 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
77 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
78 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
79 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
80 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
81 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
82 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
83 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
84 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
85 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
86 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
87 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
88 * unwritten extent conversion */
89 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
90 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
91 #define MNTOPT_32BITINODE "inode32" /* inode allocation limited to
92 * XFS_MAXINUMBER_32 */
93 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
94 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
95 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
96 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
98 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
99 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
100 #define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
101 #define MNTOPT_QUOTA "quota" /* disk quotas (user) */
102 #define MNTOPT_NOQUOTA "noquota" /* no quotas */
103 #define MNTOPT_USRQUOTA "usrquota" /* user quota enabled */
104 #define MNTOPT_GRPQUOTA "grpquota" /* group quota enabled */
105 #define MNTOPT_PRJQUOTA "prjquota" /* project quota enabled */
106 #define MNTOPT_UQUOTA "uquota" /* user quota (IRIX variant) */
107 #define MNTOPT_GQUOTA "gquota" /* group quota (IRIX variant) */
108 #define MNTOPT_PQUOTA "pquota" /* project quota (IRIX variant) */
109 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
110 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
111 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
112 #define MNTOPT_QUOTANOENF "qnoenforce" /* same as uqnoenforce */
113 #define MNTOPT_DELAYLOG "delaylog" /* Delayed logging enabled */
114 #define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed logging disabled */
115 #define MNTOPT_DISCARD "discard" /* Discard unused blocks */
116 #define MNTOPT_NODISCARD "nodiscard" /* Do not discard unused blocks */
119 * Table driven mount option parser.
121 * Currently only used for remount, but it will be used for mount
122 * in the future, too.
125 Opt_barrier, Opt_nobarrier, Opt_inode64, Opt_err
128 static const match_table_t tokens = {
129 {Opt_barrier, "barrier"},
130 {Opt_nobarrier, "nobarrier"},
131 {Opt_inode64, "inode64"},
137 suffix_strtoul(char *s, char **endp, unsigned int base)
139 int last, shift_left_factor = 0;
142 last = strlen(value) - 1;
143 if (value[last] == 'K' || value[last] == 'k') {
144 shift_left_factor = 10;
147 if (value[last] == 'M' || value[last] == 'm') {
148 shift_left_factor = 20;
151 if (value[last] == 'G' || value[last] == 'g') {
152 shift_left_factor = 30;
156 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
160 * This function fills in xfs_mount_t fields based on mount args.
161 * Note: the superblock has _not_ yet been read in.
163 * Note that this function leaks the various device name allocations on
164 * failure. The caller takes care of them.
168 struct xfs_mount *mp,
171 struct super_block *sb = mp->m_super;
172 char *this_char, *value, *eov;
176 __uint8_t iosizelog = 0;
179 * set up the mount name first so all the errors will refer to the
182 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
185 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
188 * Copy binary VFS mount flags we are interested in.
190 if (sb->s_flags & MS_RDONLY)
191 mp->m_flags |= XFS_MOUNT_RDONLY;
192 if (sb->s_flags & MS_DIRSYNC)
193 mp->m_flags |= XFS_MOUNT_DIRSYNC;
194 if (sb->s_flags & MS_SYNCHRONOUS)
195 mp->m_flags |= XFS_MOUNT_WSYNC;
198 * Set some default flags that could be cleared by the mount option
201 mp->m_flags |= XFS_MOUNT_BARRIER;
202 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
204 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
208 * These can be overridden by the mount option parsing.
216 while ((this_char = strsep(&options, ",")) != NULL) {
219 if ((value = strchr(this_char, '=')) != NULL)
222 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
223 if (!value || !*value) {
224 xfs_warn(mp, "%s option requires an argument",
228 mp->m_logbufs = simple_strtoul(value, &eov, 10);
229 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
230 if (!value || !*value) {
231 xfs_warn(mp, "%s option requires an argument",
235 mp->m_logbsize = suffix_strtoul(value, &eov, 10);
236 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
237 if (!value || !*value) {
238 xfs_warn(mp, "%s option requires an argument",
242 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
245 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
246 xfs_warn(mp, "%s option not allowed on this system",
249 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
250 if (!value || !*value) {
251 xfs_warn(mp, "%s option requires an argument",
255 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
258 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
259 if (!value || !*value) {
260 xfs_warn(mp, "%s option requires an argument",
264 iosize = simple_strtoul(value, &eov, 10);
265 iosizelog = ffs(iosize) - 1;
266 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
267 if (!value || !*value) {
268 xfs_warn(mp, "%s option requires an argument",
272 iosize = suffix_strtoul(value, &eov, 10);
273 iosizelog = ffs(iosize) - 1;
274 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
275 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
276 mp->m_flags |= XFS_MOUNT_GRPID;
277 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
278 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
279 mp->m_flags &= ~XFS_MOUNT_GRPID;
280 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
281 mp->m_flags |= XFS_MOUNT_WSYNC;
282 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
283 mp->m_flags |= XFS_MOUNT_NORECOVERY;
284 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
285 mp->m_flags |= XFS_MOUNT_NOALIGN;
286 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
287 mp->m_flags |= XFS_MOUNT_SWALLOC;
288 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
289 if (!value || !*value) {
290 xfs_warn(mp, "%s option requires an argument",
294 dsunit = simple_strtoul(value, &eov, 10);
295 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
296 if (!value || !*value) {
297 xfs_warn(mp, "%s option requires an argument",
301 dswidth = simple_strtoul(value, &eov, 10);
302 } else if (!strcmp(this_char, MNTOPT_32BITINODE)) {
303 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
304 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
305 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
307 xfs_warn(mp, "%s option not allowed on this system",
311 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
312 mp->m_flags |= XFS_MOUNT_NOUUID;
313 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
314 mp->m_flags |= XFS_MOUNT_BARRIER;
315 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
316 mp->m_flags &= ~XFS_MOUNT_BARRIER;
317 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
318 mp->m_flags |= XFS_MOUNT_IKEEP;
319 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
320 mp->m_flags &= ~XFS_MOUNT_IKEEP;
321 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
322 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
323 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
324 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
325 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
326 mp->m_flags |= XFS_MOUNT_ATTR2;
327 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
328 mp->m_flags &= ~XFS_MOUNT_ATTR2;
329 mp->m_flags |= XFS_MOUNT_NOATTR2;
330 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
331 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
332 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
333 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
334 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
335 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
336 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
337 !strcmp(this_char, MNTOPT_UQUOTA) ||
338 !strcmp(this_char, MNTOPT_USRQUOTA)) {
339 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
341 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
342 !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
343 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
344 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
345 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
346 !strcmp(this_char, MNTOPT_PRJQUOTA)) {
347 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
349 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
350 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
351 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
352 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
353 !strcmp(this_char, MNTOPT_GRPQUOTA)) {
354 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
356 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
357 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
358 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
359 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
361 "delaylog is the default now, option is deprecated.");
362 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
364 "nodelaylog support has been removed, option is deprecated.");
365 } else if (!strcmp(this_char, MNTOPT_DISCARD)) {
366 mp->m_flags |= XFS_MOUNT_DISCARD;
367 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
368 mp->m_flags &= ~XFS_MOUNT_DISCARD;
369 } else if (!strcmp(this_char, "ihashsize")) {
371 "ihashsize no longer used, option is deprecated.");
372 } else if (!strcmp(this_char, "osyncisdsync")) {
374 "osyncisdsync has no effect, option is deprecated.");
375 } else if (!strcmp(this_char, "osyncisosync")) {
377 "osyncisosync has no effect, option is deprecated.");
378 } else if (!strcmp(this_char, "irixsgid")) {
380 "irixsgid is now a sysctl(2) variable, option is deprecated.");
382 xfs_warn(mp, "unknown mount option [%s].", this_char);
388 * no recovery flag requires a read-only mount
390 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
391 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
392 xfs_warn(mp, "no-recovery mounts must be read-only.");
396 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
398 "sunit and swidth options incompatible with the noalign option");
402 #ifndef CONFIG_XFS_QUOTA
403 if (XFS_IS_QUOTA_RUNNING(mp)) {
404 xfs_warn(mp, "quota support not available in this kernel.");
409 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
410 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
411 xfs_warn(mp, "cannot mount with both project and group quota");
415 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
416 xfs_warn(mp, "sunit and swidth must be specified together");
420 if (dsunit && (dswidth % dsunit != 0)) {
422 "stripe width (%d) must be a multiple of the stripe unit (%d)",
428 if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
430 * At this point the superblock has not been read
431 * in, therefore we do not know the block size.
432 * Before the mount call ends we will convert
436 mp->m_dalign = dsunit;
437 mp->m_flags |= XFS_MOUNT_RETERR;
441 mp->m_swidth = dswidth;
444 if (mp->m_logbufs != -1 &&
445 mp->m_logbufs != 0 &&
446 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
447 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
448 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
449 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
450 return XFS_ERROR(EINVAL);
452 if (mp->m_logbsize != -1 &&
453 mp->m_logbsize != 0 &&
454 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
455 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
456 !is_power_of_2(mp->m_logbsize))) {
458 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
460 return XFS_ERROR(EINVAL);
464 if (iosizelog > XFS_MAX_IO_LOG ||
465 iosizelog < XFS_MIN_IO_LOG) {
466 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
467 iosizelog, XFS_MIN_IO_LOG,
469 return XFS_ERROR(EINVAL);
472 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
473 mp->m_readio_log = iosizelog;
474 mp->m_writeio_log = iosizelog;
480 struct proc_xfs_info {
487 struct xfs_mount *mp,
490 static struct proc_xfs_info xfs_info_set[] = {
491 /* the few simple ones we can get from the mount struct */
492 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP },
493 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
494 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
495 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
496 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
497 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
498 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 },
499 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
500 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
501 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD },
502 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_32BITINODE },
505 static struct proc_xfs_info xfs_info_unset[] = {
506 /* the few simple ones we can get from the mount struct */
507 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO },
508 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER },
509 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE },
512 struct proc_xfs_info *xfs_infop;
514 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
515 if (mp->m_flags & xfs_infop->flag)
516 seq_puts(m, xfs_infop->str);
518 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
519 if (!(mp->m_flags & xfs_infop->flag))
520 seq_puts(m, xfs_infop->str);
523 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
524 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
525 (int)(1 << mp->m_writeio_log) >> 10);
527 if (mp->m_logbufs > 0)
528 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
529 if (mp->m_logbsize > 0)
530 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
533 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
535 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
537 if (mp->m_dalign > 0)
538 seq_printf(m, "," MNTOPT_SUNIT "=%d",
539 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
540 if (mp->m_swidth > 0)
541 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
542 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
544 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
545 seq_puts(m, "," MNTOPT_USRQUOTA);
546 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
547 seq_puts(m, "," MNTOPT_UQUOTANOENF);
549 /* Either project or group quotas can be active, not both */
551 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
552 if (mp->m_qflags & XFS_OQUOTA_ENFD)
553 seq_puts(m, "," MNTOPT_PRJQUOTA);
555 seq_puts(m, "," MNTOPT_PQUOTANOENF);
556 } else if (mp->m_qflags & XFS_GQUOTA_ACCT) {
557 if (mp->m_qflags & XFS_OQUOTA_ENFD)
558 seq_puts(m, "," MNTOPT_GRPQUOTA);
560 seq_puts(m, "," MNTOPT_GQUOTANOENF);
563 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
564 seq_puts(m, "," MNTOPT_NOQUOTA);
570 unsigned int blockshift)
572 unsigned int pagefactor = 1;
573 unsigned int bitshift = BITS_PER_LONG - 1;
575 /* Figure out maximum filesize, on Linux this can depend on
576 * the filesystem blocksize (on 32 bit platforms).
577 * __block_write_begin does this in an [unsigned] long...
578 * page->index << (PAGE_CACHE_SHIFT - bbits)
579 * So, for page sized blocks (4K on 32 bit platforms),
580 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
581 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
582 * but for smaller blocksizes it is less (bbits = log2 bsize).
583 * Note1: get_block_t takes a long (implicit cast from above)
584 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
585 * can optionally convert the [unsigned] long from above into
586 * an [unsigned] long long.
589 #if BITS_PER_LONG == 32
590 # if defined(CONFIG_LBDAF)
591 ASSERT(sizeof(sector_t) == 8);
592 pagefactor = PAGE_CACHE_SIZE;
593 bitshift = BITS_PER_LONG;
595 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
599 return (((__uint64_t)pagefactor) << bitshift) - 1;
603 xfs_set_inode32(struct xfs_mount *mp)
605 xfs_agnumber_t index = 0;
606 xfs_sb_t *sbp = &mp->m_sb;
607 xfs_agnumber_t max_metadata;
608 xfs_agino_t agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks -1, 0);
609 xfs_ino_t ino = XFS_AGINO_TO_INO(mp, sbp->sb_agcount -1, agino);
612 /* Calculate how much should be reserved for inodes to meet
613 * the max inode percentage.
615 if (mp->m_maxicount) {
618 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
620 icount += sbp->sb_agblocks - 1;
621 do_div(icount, sbp->sb_agblocks);
622 max_metadata = icount;
624 max_metadata = sbp->sb_agcount;
627 for (index = 0; index < sbp->sb_agcount; index++) {
628 ino = XFS_AGINO_TO_INO(mp, index, agino);
629 if (ino > XFS_MAXINUMBER_32) {
634 pag = xfs_perag_get(mp, index);
635 pag->pagi_inodeok = 1;
636 if (index < max_metadata)
637 pag->pagf_metadata = 1;
644 xfs_set_inode64(struct xfs_mount *mp)
646 xfs_agnumber_t index = 0;
648 for (index = 0; index < mp->m_sb.sb_agcount; index++) {
649 struct xfs_perag *pag;
651 pag = xfs_perag_get(mp, index);
652 pag->pagi_inodeok = 1;
653 pag->pagf_metadata = 0;
657 /* There is no need for lock protection on m_flags,
658 * the rw_semaphore of the VFS superblock is locked
659 * during mount/umount/remount operations, so this is
660 * enough to avoid concurency on the m_flags field
662 mp->m_flags &= ~(XFS_MOUNT_32BITINODES |
663 XFS_MOUNT_SMALL_INUMS);
671 struct block_device **bdevp)
675 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
677 if (IS_ERR(*bdevp)) {
678 error = PTR_ERR(*bdevp);
679 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
687 struct block_device *bdev)
690 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
694 xfs_blkdev_issue_flush(
695 xfs_buftarg_t *buftarg)
697 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
702 struct xfs_mount *mp)
704 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
705 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
706 xfs_free_buftarg(mp, mp->m_logdev_targp);
707 xfs_blkdev_put(logdev);
709 if (mp->m_rtdev_targp) {
710 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
711 xfs_free_buftarg(mp, mp->m_rtdev_targp);
712 xfs_blkdev_put(rtdev);
714 xfs_free_buftarg(mp, mp->m_ddev_targp);
718 * The file system configurations are:
719 * (1) device (partition) with data and internal log
720 * (2) logical volume with data and log subvolumes.
721 * (3) logical volume with data, log, and realtime subvolumes.
723 * We only have to handle opening the log and realtime volumes here if
724 * they are present. The data subvolume has already been opened by
725 * get_sb_bdev() and is stored in sb->s_bdev.
729 struct xfs_mount *mp)
731 struct block_device *ddev = mp->m_super->s_bdev;
732 struct block_device *logdev = NULL, *rtdev = NULL;
736 * Open real time and log devices - order is important.
739 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
745 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
747 goto out_close_logdev;
749 if (rtdev == ddev || rtdev == logdev) {
751 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
753 goto out_close_rtdev;
758 * Setup xfs_mount buffer target pointers
761 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
762 if (!mp->m_ddev_targp)
763 goto out_close_rtdev;
766 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
768 if (!mp->m_rtdev_targp)
769 goto out_free_ddev_targ;
772 if (logdev && logdev != ddev) {
773 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
775 if (!mp->m_logdev_targp)
776 goto out_free_rtdev_targ;
778 mp->m_logdev_targp = mp->m_ddev_targp;
784 if (mp->m_rtdev_targp)
785 xfs_free_buftarg(mp, mp->m_rtdev_targp);
787 xfs_free_buftarg(mp, mp->m_ddev_targp);
790 xfs_blkdev_put(rtdev);
792 if (logdev && logdev != ddev)
793 xfs_blkdev_put(logdev);
799 * Setup xfs_mount buffer target pointers based on superblock
803 struct xfs_mount *mp)
807 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
808 mp->m_sb.sb_sectsize);
812 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
813 unsigned int log_sector_size = BBSIZE;
815 if (xfs_sb_version_hassector(&mp->m_sb))
816 log_sector_size = mp->m_sb.sb_logsectsize;
817 error = xfs_setsize_buftarg(mp->m_logdev_targp,
818 mp->m_sb.sb_blocksize,
823 if (mp->m_rtdev_targp) {
824 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
825 mp->m_sb.sb_blocksize,
826 mp->m_sb.sb_sectsize);
835 xfs_init_mount_workqueues(
836 struct xfs_mount *mp)
838 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
839 WQ_MEM_RECLAIM, 0, mp->m_fsname);
840 if (!mp->m_data_workqueue)
843 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
844 WQ_MEM_RECLAIM, 0, mp->m_fsname);
845 if (!mp->m_unwritten_workqueue)
846 goto out_destroy_data_iodone_queue;
848 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
849 WQ_MEM_RECLAIM, 0, mp->m_fsname);
850 if (!mp->m_cil_workqueue)
851 goto out_destroy_unwritten;
854 out_destroy_unwritten:
855 destroy_workqueue(mp->m_unwritten_workqueue);
856 out_destroy_data_iodone_queue:
857 destroy_workqueue(mp->m_data_workqueue);
863 xfs_destroy_mount_workqueues(
864 struct xfs_mount *mp)
866 destroy_workqueue(mp->m_cil_workqueue);
867 destroy_workqueue(mp->m_data_workqueue);
868 destroy_workqueue(mp->m_unwritten_workqueue);
871 /* Catch misguided souls that try to use this interface on XFS */
872 STATIC struct inode *
874 struct super_block *sb)
881 * Now that the generic code is guaranteed not to be accessing
882 * the linux inode, we can reclaim the inode.
885 xfs_fs_destroy_inode(
888 struct xfs_inode *ip = XFS_I(inode);
890 trace_xfs_destroy_inode(ip);
892 XFS_STATS_INC(vn_reclaim);
894 /* bad inode, get out here ASAP */
895 if (is_bad_inode(inode))
898 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
901 * We should never get here with one of the reclaim flags already set.
903 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
904 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
907 * We always use background reclaim here because even if the
908 * inode is clean, it still may be under IO and hence we have
909 * to take the flush lock. The background reclaim path handles
910 * this more efficiently than we can here, so simply let background
911 * reclaim tear down all inodes.
914 xfs_inode_set_reclaim_tag(ip);
918 * Slab object creation initialisation for the XFS inode.
919 * This covers only the idempotent fields in the XFS inode;
920 * all other fields need to be initialised on allocation
921 * from the slab. This avoids the need to repeatedly initialise
922 * fields in the xfs inode that left in the initialise state
923 * when freeing the inode.
926 xfs_fs_inode_init_once(
929 struct xfs_inode *ip = inode;
931 memset(ip, 0, sizeof(struct xfs_inode));
934 inode_init_once(VFS_I(ip));
937 atomic_set(&ip->i_pincount, 0);
938 spin_lock_init(&ip->i_flags_lock);
940 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
941 "xfsino", ip->i_ino);
948 xfs_inode_t *ip = XFS_I(inode);
950 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
952 trace_xfs_evict_inode(ip);
954 truncate_inode_pages(&inode->i_data, 0);
956 XFS_STATS_INC(vn_rele);
957 XFS_STATS_INC(vn_remove);
958 XFS_STATS_DEC(vn_active);
964 * We do an unlocked check for XFS_IDONTCACHE here because we are already
965 * serialised against cache hits here via the inode->i_lock and igrab() in
966 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
967 * racing with us, and it avoids needing to grab a spinlock here for every inode
968 * we drop the final reference on.
974 struct xfs_inode *ip = XFS_I(inode);
976 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
981 struct xfs_mount *mp)
985 kfree(mp->m_logname);
990 struct super_block *sb)
992 struct xfs_mount *mp = XFS_M(sb);
994 xfs_filestream_unmount(mp);
995 cancel_delayed_work_sync(&mp->m_sync_work);
999 xfs_icsb_destroy_counters(mp);
1000 xfs_destroy_mount_workqueues(mp);
1001 xfs_close_devices(mp);
1002 xfs_free_fsname(mp);
1008 struct super_block *sb,
1011 struct xfs_mount *mp = XFS_M(sb);
1015 * Doing anything during the async pass would be counterproductive.
1020 error = xfs_quiesce_data(mp);
1026 * The disk must be active because we're syncing.
1027 * We schedule xfssyncd now (now that the disk is
1028 * active) instead of later (when it might not be).
1030 flush_delayed_work_sync(&mp->m_sync_work);
1038 struct dentry *dentry,
1039 struct kstatfs *statp)
1041 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1042 xfs_sb_t *sbp = &mp->m_sb;
1043 struct xfs_inode *ip = XFS_I(dentry->d_inode);
1044 __uint64_t fakeinos, id;
1048 statp->f_type = XFS_SB_MAGIC;
1049 statp->f_namelen = MAXNAMELEN - 1;
1051 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1052 statp->f_fsid.val[0] = (u32)id;
1053 statp->f_fsid.val[1] = (u32)(id >> 32);
1055 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1057 spin_lock(&mp->m_sb_lock);
1058 statp->f_bsize = sbp->sb_blocksize;
1059 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1060 statp->f_blocks = sbp->sb_dblocks - lsize;
1061 statp->f_bfree = statp->f_bavail =
1062 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1063 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1065 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1066 if (mp->m_maxicount)
1067 statp->f_files = min_t(typeof(statp->f_files),
1071 /* make sure statp->f_ffree does not underflow */
1072 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1073 statp->f_ffree = max_t(__int64_t, ffree, 0);
1075 spin_unlock(&mp->m_sb_lock);
1077 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1078 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) ==
1079 (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
1080 xfs_qm_statvfs(ip, statp);
1085 xfs_save_resvblks(struct xfs_mount *mp)
1087 __uint64_t resblks = 0;
1089 mp->m_resblks_save = mp->m_resblks;
1090 xfs_reserve_blocks(mp, &resblks, NULL);
1094 xfs_restore_resvblks(struct xfs_mount *mp)
1098 if (mp->m_resblks_save) {
1099 resblks = mp->m_resblks_save;
1100 mp->m_resblks_save = 0;
1102 resblks = xfs_default_resblks(mp);
1104 xfs_reserve_blocks(mp, &resblks, NULL);
1109 struct super_block *sb,
1113 struct xfs_mount *mp = XFS_M(sb);
1114 substring_t args[MAX_OPT_ARGS];
1118 while ((p = strsep(&options, ",")) != NULL) {
1124 token = match_token(p, tokens, args);
1127 mp->m_flags |= XFS_MOUNT_BARRIER;
1130 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1133 xfs_set_inode64(mp);
1137 * Logically we would return an error here to prevent
1138 * users from believing they might have changed
1139 * mount options using remount which can't be changed.
1141 * But unfortunately mount(8) adds all options from
1142 * mtab and fstab to the mount arguments in some cases
1143 * so we can't blindly reject options, but have to
1144 * check for each specified option if it actually
1145 * differs from the currently set option and only
1146 * reject it if that's the case.
1148 * Until that is implemented we return success for
1149 * every remount request, and silently ignore all
1150 * options that we can't actually change.
1154 "mount option \"%s\" not supported for remount\n", p);
1163 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1164 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1167 * If this is the first remount to writeable state we
1168 * might have some superblock changes to update.
1170 if (mp->m_update_flags) {
1171 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1173 xfs_warn(mp, "failed to write sb changes");
1176 mp->m_update_flags = 0;
1180 * Fill out the reserve pool if it is empty. Use the stashed
1181 * value if it is non-zero, otherwise go with the default.
1183 xfs_restore_resvblks(mp);
1187 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1189 * After we have synced the data but before we sync the
1190 * metadata, we need to free up the reserve block pool so that
1191 * the used block count in the superblock on disk is correct at
1192 * the end of the remount. Stash the current reserve pool size
1193 * so that if we get remounted rw, we can return it to the same
1197 xfs_quiesce_data(mp);
1198 xfs_save_resvblks(mp);
1199 xfs_quiesce_attr(mp);
1200 mp->m_flags |= XFS_MOUNT_RDONLY;
1207 * Second stage of a freeze. The data is already frozen so we only
1208 * need to take care of the metadata. Once that's done write a dummy
1209 * record to dirty the log in case of a crash while frozen.
1213 struct super_block *sb)
1215 struct xfs_mount *mp = XFS_M(sb);
1217 xfs_save_resvblks(mp);
1218 xfs_quiesce_attr(mp);
1219 return -xfs_fs_log_dummy(mp);
1224 struct super_block *sb)
1226 struct xfs_mount *mp = XFS_M(sb);
1228 xfs_restore_resvblks(mp);
1233 xfs_fs_show_options(
1235 struct dentry *root)
1237 return -xfs_showargs(XFS_M(root->d_sb), m);
1241 * This function fills in xfs_mount_t fields based on mount args.
1242 * Note: the superblock _has_ now been read in.
1246 struct xfs_mount *mp)
1248 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1250 /* Fail a mount where the logbuf is smaller than the log stripe */
1251 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1252 if (mp->m_logbsize <= 0 &&
1253 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1254 mp->m_logbsize = mp->m_sb.sb_logsunit;
1255 } else if (mp->m_logbsize > 0 &&
1256 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1258 "logbuf size must be greater than or equal to log stripe size");
1259 return XFS_ERROR(EINVAL);
1262 /* Fail a mount if the logbuf is larger than 32K */
1263 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1265 "logbuf size for version 1 logs must be 16K or 32K");
1266 return XFS_ERROR(EINVAL);
1271 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1272 * told by noattr2 to turn it off
1274 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1275 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1276 mp->m_flags |= XFS_MOUNT_ATTR2;
1279 * prohibit r/w mounts of read-only filesystems
1281 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1283 "cannot mount a read-only filesystem as read-write");
1284 return XFS_ERROR(EROFS);
1292 struct super_block *sb,
1297 struct xfs_mount *mp = NULL;
1298 int flags = 0, error = ENOMEM;
1300 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1304 spin_lock_init(&mp->m_sb_lock);
1305 mutex_init(&mp->m_growlock);
1306 atomic_set(&mp->m_active_trans, 0);
1311 error = xfs_parseargs(mp, (char *)data);
1313 goto out_free_fsname;
1315 sb_min_blocksize(sb, BBSIZE);
1316 sb->s_xattr = xfs_xattr_handlers;
1317 sb->s_export_op = &xfs_export_operations;
1318 #ifdef CONFIG_XFS_QUOTA
1319 sb->s_qcop = &xfs_quotactl_operations;
1321 sb->s_op = &xfs_super_operations;
1324 flags |= XFS_MFSI_QUIET;
1326 error = xfs_open_devices(mp);
1328 goto out_free_fsname;
1330 error = xfs_init_mount_workqueues(mp);
1332 goto out_close_devices;
1334 error = xfs_icsb_init_counters(mp);
1336 goto out_destroy_workqueues;
1338 error = xfs_readsb(mp, flags);
1340 goto out_destroy_counters;
1342 error = xfs_finish_flags(mp);
1346 error = xfs_setup_devices(mp);
1350 error = xfs_filestream_mount(mp);
1355 * we must configure the block size in the superblock before we run the
1356 * full mount process as the mount process can lookup and cache inodes.
1357 * For the same reason we must also initialise the syncd and register
1358 * the inode cache shrinker so that inodes can be reclaimed during
1359 * operations like a quotacheck that iterate all inodes in the
1362 sb->s_magic = XFS_SB_MAGIC;
1363 sb->s_blocksize = mp->m_sb.sb_blocksize;
1364 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1365 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1366 sb->s_max_links = XFS_MAXLINK;
1367 sb->s_time_gran = 1;
1368 set_posix_acl_flag(sb);
1370 error = xfs_syncd_init(mp);
1372 goto out_filestream_unmount;
1374 error = xfs_mountfs(mp);
1376 goto out_syncd_stop;
1378 root = igrab(VFS_I(mp->m_rootip));
1383 if (is_bad_inode(root)) {
1387 sb->s_root = d_make_root(root);
1396 out_filestream_unmount:
1397 xfs_filestream_unmount(mp);
1400 out_destroy_counters:
1401 xfs_icsb_destroy_counters(mp);
1402 out_destroy_workqueues:
1403 xfs_destroy_mount_workqueues(mp);
1405 xfs_close_devices(mp);
1407 xfs_free_fsname(mp);
1413 xfs_filestream_unmount(mp);
1419 STATIC struct dentry *
1421 struct file_system_type *fs_type,
1423 const char *dev_name,
1426 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1430 xfs_fs_nr_cached_objects(
1431 struct super_block *sb)
1433 return xfs_reclaim_inodes_count(XFS_M(sb));
1437 xfs_fs_free_cached_objects(
1438 struct super_block *sb,
1441 xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1444 static const struct super_operations xfs_super_operations = {
1445 .alloc_inode = xfs_fs_alloc_inode,
1446 .destroy_inode = xfs_fs_destroy_inode,
1447 .evict_inode = xfs_fs_evict_inode,
1448 .drop_inode = xfs_fs_drop_inode,
1449 .put_super = xfs_fs_put_super,
1450 .sync_fs = xfs_fs_sync_fs,
1451 .freeze_fs = xfs_fs_freeze,
1452 .unfreeze_fs = xfs_fs_unfreeze,
1453 .statfs = xfs_fs_statfs,
1454 .remount_fs = xfs_fs_remount,
1455 .show_options = xfs_fs_show_options,
1456 .nr_cached_objects = xfs_fs_nr_cached_objects,
1457 .free_cached_objects = xfs_fs_free_cached_objects,
1460 static struct file_system_type xfs_fs_type = {
1461 .owner = THIS_MODULE,
1463 .mount = xfs_fs_mount,
1464 .kill_sb = kill_block_super,
1465 .fs_flags = FS_REQUIRES_DEV,
1469 xfs_init_zones(void)
1472 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1473 if (!xfs_ioend_zone)
1476 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1478 if (!xfs_ioend_pool)
1479 goto out_destroy_ioend_zone;
1481 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1483 if (!xfs_log_ticket_zone)
1484 goto out_destroy_ioend_pool;
1486 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1487 "xfs_bmap_free_item");
1488 if (!xfs_bmap_free_item_zone)
1489 goto out_destroy_log_ticket_zone;
1491 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1493 if (!xfs_btree_cur_zone)
1494 goto out_destroy_bmap_free_item_zone;
1496 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1498 if (!xfs_da_state_zone)
1499 goto out_destroy_btree_cur_zone;
1501 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1502 if (!xfs_ifork_zone)
1503 goto out_destroy_da_state_zone;
1505 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1506 if (!xfs_trans_zone)
1507 goto out_destroy_ifork_zone;
1509 xfs_log_item_desc_zone =
1510 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1511 "xfs_log_item_desc");
1512 if (!xfs_log_item_desc_zone)
1513 goto out_destroy_trans_zone;
1516 * The size of the zone allocated buf log item is the maximum
1517 * size possible under XFS. This wastes a little bit of memory,
1518 * but it is much faster.
1520 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1522 if (!xfs_buf_item_zone)
1523 goto out_destroy_log_item_desc_zone;
1525 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1526 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1527 sizeof(xfs_extent_t))), "xfs_efd_item");
1529 goto out_destroy_buf_item_zone;
1531 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1532 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1533 sizeof(xfs_extent_t))), "xfs_efi_item");
1535 goto out_destroy_efd_zone;
1538 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1539 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1540 xfs_fs_inode_init_once);
1541 if (!xfs_inode_zone)
1542 goto out_destroy_efi_zone;
1545 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1546 KM_ZONE_SPREAD, NULL);
1548 goto out_destroy_inode_zone;
1552 out_destroy_inode_zone:
1553 kmem_zone_destroy(xfs_inode_zone);
1554 out_destroy_efi_zone:
1555 kmem_zone_destroy(xfs_efi_zone);
1556 out_destroy_efd_zone:
1557 kmem_zone_destroy(xfs_efd_zone);
1558 out_destroy_buf_item_zone:
1559 kmem_zone_destroy(xfs_buf_item_zone);
1560 out_destroy_log_item_desc_zone:
1561 kmem_zone_destroy(xfs_log_item_desc_zone);
1562 out_destroy_trans_zone:
1563 kmem_zone_destroy(xfs_trans_zone);
1564 out_destroy_ifork_zone:
1565 kmem_zone_destroy(xfs_ifork_zone);
1566 out_destroy_da_state_zone:
1567 kmem_zone_destroy(xfs_da_state_zone);
1568 out_destroy_btree_cur_zone:
1569 kmem_zone_destroy(xfs_btree_cur_zone);
1570 out_destroy_bmap_free_item_zone:
1571 kmem_zone_destroy(xfs_bmap_free_item_zone);
1572 out_destroy_log_ticket_zone:
1573 kmem_zone_destroy(xfs_log_ticket_zone);
1574 out_destroy_ioend_pool:
1575 mempool_destroy(xfs_ioend_pool);
1576 out_destroy_ioend_zone:
1577 kmem_zone_destroy(xfs_ioend_zone);
1583 xfs_destroy_zones(void)
1585 kmem_zone_destroy(xfs_ili_zone);
1586 kmem_zone_destroy(xfs_inode_zone);
1587 kmem_zone_destroy(xfs_efi_zone);
1588 kmem_zone_destroy(xfs_efd_zone);
1589 kmem_zone_destroy(xfs_buf_item_zone);
1590 kmem_zone_destroy(xfs_log_item_desc_zone);
1591 kmem_zone_destroy(xfs_trans_zone);
1592 kmem_zone_destroy(xfs_ifork_zone);
1593 kmem_zone_destroy(xfs_da_state_zone);
1594 kmem_zone_destroy(xfs_btree_cur_zone);
1595 kmem_zone_destroy(xfs_bmap_free_item_zone);
1596 kmem_zone_destroy(xfs_log_ticket_zone);
1597 mempool_destroy(xfs_ioend_pool);
1598 kmem_zone_destroy(xfs_ioend_zone);
1603 xfs_init_workqueues(void)
1606 * We never want to the same work item to run twice, reclaiming inodes
1607 * or idling the log is not going to get any faster by multiple CPUs
1608 * competing for ressources. Use the default large max_active value
1609 * so that even lots of filesystems can perform these task in parallel.
1611 xfs_syncd_wq = alloc_workqueue("xfssyncd", WQ_NON_REENTRANT, 0);
1616 * The allocation workqueue can be used in memory reclaim situations
1617 * (writepage path), and parallelism is only limited by the number of
1618 * AGs in all the filesystems mounted. Hence use the default large
1619 * max_active value for this workqueue.
1621 xfs_alloc_wq = alloc_workqueue("xfsalloc", WQ_MEM_RECLAIM, 0);
1623 goto out_destroy_syncd;
1628 destroy_workqueue(xfs_syncd_wq);
1633 xfs_destroy_workqueues(void)
1635 destroy_workqueue(xfs_alloc_wq);
1636 destroy_workqueue(xfs_syncd_wq);
1644 printk(KERN_INFO XFS_VERSION_STRING " with "
1645 XFS_BUILD_OPTIONS " enabled\n");
1649 error = xfs_init_zones();
1653 error = xfs_init_workqueues();
1655 goto out_destroy_zones;
1657 error = xfs_mru_cache_init();
1659 goto out_destroy_wq;
1661 error = xfs_filestream_init();
1663 goto out_mru_cache_uninit;
1665 error = xfs_buf_init();
1667 goto out_filestream_uninit;
1669 error = xfs_init_procfs();
1671 goto out_buf_terminate;
1673 error = xfs_sysctl_register();
1675 goto out_cleanup_procfs;
1677 error = xfs_qm_init();
1679 goto out_sysctl_unregister;
1681 error = register_filesystem(&xfs_fs_type);
1688 out_sysctl_unregister:
1689 xfs_sysctl_unregister();
1691 xfs_cleanup_procfs();
1693 xfs_buf_terminate();
1694 out_filestream_uninit:
1695 xfs_filestream_uninit();
1696 out_mru_cache_uninit:
1697 xfs_mru_cache_uninit();
1699 xfs_destroy_workqueues();
1701 xfs_destroy_zones();
1710 unregister_filesystem(&xfs_fs_type);
1711 xfs_sysctl_unregister();
1712 xfs_cleanup_procfs();
1713 xfs_buf_terminate();
1714 xfs_filestream_uninit();
1715 xfs_mru_cache_uninit();
1716 xfs_destroy_workqueues();
1717 xfs_destroy_zones();
1720 module_init(init_xfs_fs);
1721 module_exit(exit_xfs_fs);
1723 MODULE_AUTHOR("Silicon Graphics, Inc.");
1724 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1725 MODULE_LICENSE("GPL");