?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-2.6.git] / fs / xfs / xfs_super.c
blob3033ba5e9762f19609a7f86d5117f64eb70f7bd4
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18
19 #include "xfs.h"
20 #include "xfs_log.h"
21 #include "xfs_inum.h"
22 #include "xfs_trans.h"
23 #include "xfs_sb.h"
24 #include "xfs_ag.h"
25 #include "xfs_dir2.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"
36 #include "xfs_bmap.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_itable.h"
40 #include "xfs_fsops.h"
41 #include "xfs_attr.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"
52 #include "xfs_icache.h"
53 #include "xfs_trace.h"
54
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>
64
65 static const struct super_operations xfs_super_operations;
66 static kmem_zone_t *xfs_ioend_zone;
67 mempool_t *xfs_ioend_pool;
68
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
97 * in stat(). */
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 */
117
118 /*
119 * Table driven mount option parser.
120 *
121 * Currently only used for remount, but it will be used for mount
122 * in the future, too.
123 */
124 enum {
125 Opt_barrier,
126 Opt_nobarrier,
127 Opt_inode64,
128 Opt_inode32,
129 Opt_err
130 };
131
132 static const match_table_t tokens = {
133 {Opt_barrier, "barrier"},
134 {Opt_nobarrier, "nobarrier"},
135 {Opt_inode64, "inode64"},
136 {Opt_inode32, "inode32"},
137 {Opt_err, NULL}
138 };
139
140
141 STATIC unsigned long
142 suffix_kstrtoint(char *s, unsigned int base, int *res)
143 {
144 int last, shift_left_factor = 0, _res;
145 char *value = s;
146
147 last = strlen(value) - 1;
148 if (value[last] == 'K' || value[last] == 'k') {
149 shift_left_factor = 10;
150 value[last] = '\0';
151 }
152 if (value[last] == 'M' || value[last] == 'm') {
153 shift_left_factor = 20;
154 value[last] = '\0';
155 }
156 if (value[last] == 'G' || value[last] == 'g') {
157 shift_left_factor = 30;
158 value[last] = '\0';
159 }
160
161 if (kstrtoint(s, base, &_res))
162 return -EINVAL;
163 *res = _res << shift_left_factor;
164 return 0;
165 }
166
167 /*
168 * This function fills in xfs_mount_t fields based on mount args.
169 * Note: the superblock has _not_ yet been read in.
170 *
171 * Note that this function leaks the various device name allocations on
172 * failure. The caller takes care of them.
173 */
174 STATIC int
175 xfs_parseargs(
176 struct xfs_mount *mp,
177 char *options)
178 {
179 struct super_block *sb = mp->m_super;
180 char *this_char, *value;
181 int dsunit = 0;
182 int dswidth = 0;
183 int iosize = 0;
184 __uint8_t iosizelog = 0;
185
186 /*
187 * set up the mount name first so all the errors will refer to the
188 * correct device.
189 */
190 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
191 if (!mp->m_fsname)
192 return ENOMEM;
193 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
194
195 /*
196 * Copy binary VFS mount flags we are interested in.
197 */
198 if (sb->s_flags & MS_RDONLY)
199 mp->m_flags |= XFS_MOUNT_RDONLY;
200 if (sb->s_flags & MS_DIRSYNC)
201 mp->m_flags |= XFS_MOUNT_DIRSYNC;
202 if (sb->s_flags & MS_SYNCHRONOUS)
203 mp->m_flags |= XFS_MOUNT_WSYNC;
204
205 /*
206 * Set some default flags that could be cleared by the mount option
207 * parsing.
208 */
209 mp->m_flags |= XFS_MOUNT_BARRIER;
210 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
211 #if !XFS_BIG_INUMS
212 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
213 #endif
214
215 /*
216 * These can be overridden by the mount option parsing.
217 */
218 mp->m_logbufs = -1;
219 mp->m_logbsize = -1;
220
221 if (!options)
222 goto done;
223
224 while ((this_char = strsep(&options, ",")) != NULL) {
225 if (!*this_char)
226 continue;
227 if ((value = strchr(this_char, '=')) != NULL)
228 *value++ = 0;
229
230 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
231 if (!value || !*value) {
232 xfs_warn(mp, "%s option requires an argument",
233 this_char);
234 return EINVAL;
235 }
236 if (kstrtoint(value, 10, &mp->m_logbufs))
237 return EINVAL;
238 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
239 if (!value || !*value) {
240 xfs_warn(mp, "%s option requires an argument",
241 this_char);
242 return EINVAL;
243 }
244 if (suffix_kstrtoint(value, 10, &mp->m_logbsize))
245 return EINVAL;
246 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
247 if (!value || !*value) {
248 xfs_warn(mp, "%s option requires an argument",
249 this_char);
250 return EINVAL;
251 }
252 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
253 if (!mp->m_logname)
254 return ENOMEM;
255 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
256 xfs_warn(mp, "%s option not allowed on this system",
257 this_char);
258 return EINVAL;
259 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
260 if (!value || !*value) {
261 xfs_warn(mp, "%s option requires an argument",
262 this_char);
263 return EINVAL;
264 }
265 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
266 if (!mp->m_rtname)
267 return ENOMEM;
268 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
269 if (!value || !*value) {
270 xfs_warn(mp, "%s option requires an argument",
271 this_char);
272 return EINVAL;
273 }
274 if (kstrtoint(value, 10, &iosize))
275 return EINVAL;
276 iosizelog = ffs(iosize) - 1;
277 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
278 if (!value || !*value) {
279 xfs_warn(mp, "%s option requires an argument",
280 this_char);
281 return EINVAL;
282 }
283 if (suffix_kstrtoint(value, 10, &iosize))
284 return EINVAL;
285 iosizelog = ffs(iosize) - 1;
286 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
287 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
288 mp->m_flags |= XFS_MOUNT_GRPID;
289 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
290 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
291 mp->m_flags &= ~XFS_MOUNT_GRPID;
292 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
293 mp->m_flags |= XFS_MOUNT_WSYNC;
294 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
295 mp->m_flags |= XFS_MOUNT_NORECOVERY;
296 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
297 mp->m_flags |= XFS_MOUNT_NOALIGN;
298 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
299 mp->m_flags |= XFS_MOUNT_SWALLOC;
300 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
301 if (!value || !*value) {
302 xfs_warn(mp, "%s option requires an argument",
303 this_char);
304 return EINVAL;
305 }
306 if (kstrtoint(value, 10, &dsunit))
307 return EINVAL;
308 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
309 if (!value || !*value) {
310 xfs_warn(mp, "%s option requires an argument",
311 this_char);
312 return EINVAL;
313 }
314 if (kstrtoint(value, 10, &dswidth))
315 return EINVAL;
316 } else if (!strcmp(this_char, MNTOPT_32BITINODE)) {
317 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
318 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
319 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
320 #if !XFS_BIG_INUMS
321 xfs_warn(mp, "%s option not allowed on this system",
322 this_char);
323 return EINVAL;
324 #endif
325 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
326 mp->m_flags |= XFS_MOUNT_NOUUID;
327 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
328 mp->m_flags |= XFS_MOUNT_BARRIER;
329 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
330 mp->m_flags &= ~XFS_MOUNT_BARRIER;
331 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
332 mp->m_flags |= XFS_MOUNT_IKEEP;
333 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
334 mp->m_flags &= ~XFS_MOUNT_IKEEP;
335 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
336 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
337 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
338 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
339 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
340 mp->m_flags |= XFS_MOUNT_ATTR2;
341 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
342 mp->m_flags &= ~XFS_MOUNT_ATTR2;
343 mp->m_flags |= XFS_MOUNT_NOATTR2;
344 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
345 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
346 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
347 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
348 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
349 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
350 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
351 !strcmp(this_char, MNTOPT_UQUOTA) ||
352 !strcmp(this_char, MNTOPT_USRQUOTA)) {
353 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
354 XFS_UQUOTA_ENFD);
355 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
356 !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
357 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
358 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
359 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
360 !strcmp(this_char, MNTOPT_PRJQUOTA)) {
361 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
362 XFS_OQUOTA_ENFD);
363 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
364 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
365 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
366 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
367 !strcmp(this_char, MNTOPT_GRPQUOTA)) {
368 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
369 XFS_OQUOTA_ENFD);
370 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
371 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
372 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
373 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
374 xfs_warn(mp,
375 "delaylog is the default now, option is deprecated.");
376 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
377 xfs_warn(mp,
378 "nodelaylog support has been removed, option is deprecated.");
379 } else if (!strcmp(this_char, MNTOPT_DISCARD)) {
380 mp->m_flags |= XFS_MOUNT_DISCARD;
381 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
382 mp->m_flags &= ~XFS_MOUNT_DISCARD;
383 } else if (!strcmp(this_char, "ihashsize")) {
384 xfs_warn(mp,
385 "ihashsize no longer used, option is deprecated.");
386 } else if (!strcmp(this_char, "osyncisdsync")) {
387 xfs_warn(mp,
388 "osyncisdsync has no effect, option is deprecated.");
389 } else if (!strcmp(this_char, "osyncisosync")) {
390 xfs_warn(mp,
391 "osyncisosync has no effect, option is deprecated.");
392 } else if (!strcmp(this_char, "irixsgid")) {
393 xfs_warn(mp,
394 "irixsgid is now a sysctl(2) variable, option is deprecated.");
395 } else {
396 xfs_warn(mp, "unknown mount option [%s].", this_char);
397 return EINVAL;
398 }
399 }
400
401 /*
402 * no recovery flag requires a read-only mount
403 */
404 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
405 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
406 xfs_warn(mp, "no-recovery mounts must be read-only.");
407 return EINVAL;
408 }
409
410 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
411 xfs_warn(mp,
412 "sunit and swidth options incompatible with the noalign option");
413 return EINVAL;
414 }
415
416 #ifndef CONFIG_XFS_QUOTA
417 if (XFS_IS_QUOTA_RUNNING(mp)) {
418 xfs_warn(mp, "quota support not available in this kernel.");
419 return EINVAL;
420 }
421 #endif
422
423 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
424 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
425 xfs_warn(mp, "cannot mount with both project and group quota");
426 return EINVAL;
427 }
428
429 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
430 xfs_warn(mp, "sunit and swidth must be specified together");
431 return EINVAL;
432 }
433
434 if (dsunit && (dswidth % dsunit != 0)) {
435 xfs_warn(mp,
436 "stripe width (%d) must be a multiple of the stripe unit (%d)",
437 dswidth, dsunit);
438 return EINVAL;
439 }
440
441 done:
442 if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
443 /*
444 * At this point the superblock has not been read
445 * in, therefore we do not know the block size.
446 * Before the mount call ends we will convert
447 * these to FSBs.
448 */
449 if (dsunit) {
450 mp->m_dalign = dsunit;
451 mp->m_flags |= XFS_MOUNT_RETERR;
452 }
453
454 if (dswidth)
455 mp->m_swidth = dswidth;
456 }
457
458 if (mp->m_logbufs != -1 &&
459 mp->m_logbufs != 0 &&
460 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
461 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
462 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
463 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
464 return XFS_ERROR(EINVAL);
465 }
466 if (mp->m_logbsize != -1 &&
467 mp->m_logbsize != 0 &&
468 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
469 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
470 !is_power_of_2(mp->m_logbsize))) {
471 xfs_warn(mp,
472 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
473 mp->m_logbsize);
474 return XFS_ERROR(EINVAL);
475 }
476
477 if (iosizelog) {
478 if (iosizelog > XFS_MAX_IO_LOG ||
479 iosizelog < XFS_MIN_IO_LOG) {
480 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
481 iosizelog, XFS_MIN_IO_LOG,
482 XFS_MAX_IO_LOG);
483 return XFS_ERROR(EINVAL);
484 }
485
486 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
487 mp->m_readio_log = iosizelog;
488 mp->m_writeio_log = iosizelog;
489 }
490
491 return 0;
492 }
493
494 struct proc_xfs_info {
495 int flag;
496 char *str;
497 };
498
499 STATIC int
500 xfs_showargs(
501 struct xfs_mount *mp,
502 struct seq_file *m)
503 {
504 static struct proc_xfs_info xfs_info_set[] = {
505 /* the few simple ones we can get from the mount struct */
506 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP },
507 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
508 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
509 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
510 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
511 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
512 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 },
513 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
514 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
515 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD },
516 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_32BITINODE },
517 { 0, NULL }
518 };
519 static struct proc_xfs_info xfs_info_unset[] = {
520 /* the few simple ones we can get from the mount struct */
521 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO },
522 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER },
523 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE },
524 { 0, NULL }
525 };
526 struct proc_xfs_info *xfs_infop;
527
528 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
529 if (mp->m_flags & xfs_infop->flag)
530 seq_puts(m, xfs_infop->str);
531 }
532 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
533 if (!(mp->m_flags & xfs_infop->flag))
534 seq_puts(m, xfs_infop->str);
535 }
536
537 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
538 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
539 (int)(1 << mp->m_writeio_log) >> 10);
540
541 if (mp->m_logbufs > 0)
542 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
543 if (mp->m_logbsize > 0)
544 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
545
546 if (mp->m_logname)
547 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
548 if (mp->m_rtname)
549 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
550
551 if (mp->m_dalign > 0)
552 seq_printf(m, "," MNTOPT_SUNIT "=%d",
553 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
554 if (mp->m_swidth > 0)
555 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
556 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
557
558 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
559 seq_puts(m, "," MNTOPT_USRQUOTA);
560 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
561 seq_puts(m, "," MNTOPT_UQUOTANOENF);
562
563 /* Either project or group quotas can be active, not both */
564
565 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
566 if (mp->m_qflags & XFS_OQUOTA_ENFD)
567 seq_puts(m, "," MNTOPT_PRJQUOTA);
568 else
569 seq_puts(m, "," MNTOPT_PQUOTANOENF);
570 } else if (mp->m_qflags & XFS_GQUOTA_ACCT) {
571 if (mp->m_qflags & XFS_OQUOTA_ENFD)
572 seq_puts(m, "," MNTOPT_GRPQUOTA);
573 else
574 seq_puts(m, "," MNTOPT_GQUOTANOENF);
575 }
576
577 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
578 seq_puts(m, "," MNTOPT_NOQUOTA);
579
580 return 0;
581 }
582 __uint64_t
583 xfs_max_file_offset(
584 unsigned int blockshift)
585 {
586 unsigned int pagefactor = 1;
587 unsigned int bitshift = BITS_PER_LONG - 1;
588
589 /* Figure out maximum filesize, on Linux this can depend on
590 * the filesystem blocksize (on 32 bit platforms).
591 * __block_write_begin does this in an [unsigned] long...
592 * page->index << (PAGE_CACHE_SHIFT - bbits)
593 * So, for page sized blocks (4K on 32 bit platforms),
594 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
595 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
596 * but for smaller blocksizes it is less (bbits = log2 bsize).
597 * Note1: get_block_t takes a long (implicit cast from above)
598 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
599 * can optionally convert the [unsigned] long from above into
600 * an [unsigned] long long.
601 */
602
603 #if BITS_PER_LONG == 32
604 # if defined(CONFIG_LBDAF)
605 ASSERT(sizeof(sector_t) == 8);
606 pagefactor = PAGE_CACHE_SIZE;
607 bitshift = BITS_PER_LONG;
608 # else
609 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
610 # endif
611 #endif
612
613 return (((__uint64_t)pagefactor) << bitshift) - 1;
614 }
615
616 xfs_agnumber_t
617 xfs_set_inode32(struct xfs_mount *mp)
618 {
619 xfs_agnumber_t index = 0;
620 xfs_agnumber_t maxagi = 0;
621 xfs_sb_t *sbp = &mp->m_sb;
622 xfs_agnumber_t max_metadata;
623 xfs_agino_t agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks -1, 0);
624 xfs_ino_t ino = XFS_AGINO_TO_INO(mp, sbp->sb_agcount -1, agino);
625 xfs_perag_t *pag;
626
627 /* Calculate how much should be reserved for inodes to meet
628 * the max inode percentage.
629 */
630 if (mp->m_maxicount) {
631 __uint64_t icount;
632
633 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
634 do_div(icount, 100);
635 icount += sbp->sb_agblocks - 1;
636 do_div(icount, sbp->sb_agblocks);
637 max_metadata = icount;
638 } else {
639 max_metadata = sbp->sb_agcount;
640 }
641
642 for (index = 0; index < sbp->sb_agcount; index++) {
643 ino = XFS_AGINO_TO_INO(mp, index, agino);
644
645 if (ino > XFS_MAXINUMBER_32) {
646 pag = xfs_perag_get(mp, index);
647 pag->pagi_inodeok = 0;
648 pag->pagf_metadata = 0;
649 xfs_perag_put(pag);
650 continue;
651 }
652
653 pag = xfs_perag_get(mp, index);
654 pag->pagi_inodeok = 1;
655 maxagi++;
656 if (index < max_metadata)
657 pag->pagf_metadata = 1;
658 xfs_perag_put(pag);
659 }
660 mp->m_flags |= (XFS_MOUNT_32BITINODES |
661 XFS_MOUNT_SMALL_INUMS);
662
663 return maxagi;
664 }
665
666 xfs_agnumber_t
667 xfs_set_inode64(struct xfs_mount *mp)
668 {
669 xfs_agnumber_t index = 0;
670
671 for (index = 0; index < mp->m_sb.sb_agcount; index++) {
672 struct xfs_perag *pag;
673
674 pag = xfs_perag_get(mp, index);
675 pag->pagi_inodeok = 1;
676 pag->pagf_metadata = 0;
677 xfs_perag_put(pag);
678 }
679
680 /* There is no need for lock protection on m_flags,
681 * the rw_semaphore of the VFS superblock is locked
682 * during mount/umount/remount operations, so this is
683 * enough to avoid concurency on the m_flags field
684 */
685 mp->m_flags &= ~(XFS_MOUNT_32BITINODES |
686 XFS_MOUNT_SMALL_INUMS);
687 return index;
688 }
689
690 STATIC int
691 xfs_blkdev_get(
692 xfs_mount_t *mp,
693 const char *name,
694 struct block_device **bdevp)
695 {
696 int error = 0;
697
698 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
699 mp);
700 if (IS_ERR(*bdevp)) {
701 error = PTR_ERR(*bdevp);
702 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
703 }
704
705 return -error;
706 }
707
708 STATIC void
709 xfs_blkdev_put(
710 struct block_device *bdev)
711 {
712 if (bdev)
713 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
714 }
715
716 void
717 xfs_blkdev_issue_flush(
718 xfs_buftarg_t *buftarg)
719 {
720 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
721 }
722
723 STATIC void
724 xfs_close_devices(
725 struct xfs_mount *mp)
726 {
727 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
728 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
729 xfs_free_buftarg(mp, mp->m_logdev_targp);
730 xfs_blkdev_put(logdev);
731 }
732 if (mp->m_rtdev_targp) {
733 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
734 xfs_free_buftarg(mp, mp->m_rtdev_targp);
735 xfs_blkdev_put(rtdev);
736 }
737 xfs_free_buftarg(mp, mp->m_ddev_targp);
738 }
739
740 /*
741 * The file system configurations are:
742 * (1) device (partition) with data and internal log
743 * (2) logical volume with data and log subvolumes.
744 * (3) logical volume with data, log, and realtime subvolumes.
745 *
746 * We only have to handle opening the log and realtime volumes here if
747 * they are present. The data subvolume has already been opened by
748 * get_sb_bdev() and is stored in sb->s_bdev.
749 */
750 STATIC int
751 xfs_open_devices(
752 struct xfs_mount *mp)
753 {
754 struct block_device *ddev = mp->m_super->s_bdev;
755 struct block_device *logdev = NULL, *rtdev = NULL;
756 int error;
757
758 /*
759 * Open real time and log devices - order is important.
760 */
761 if (mp->m_logname) {
762 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
763 if (error)
764 goto out;
765 }
766
767 if (mp->m_rtname) {
768 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
769 if (error)
770 goto out_close_logdev;
771
772 if (rtdev == ddev || rtdev == logdev) {
773 xfs_warn(mp,
774 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
775 error = EINVAL;
776 goto out_close_rtdev;
777 }
778 }
779
780 /*
781 * Setup xfs_mount buffer target pointers
782 */
783 error = ENOMEM;
784 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
785 if (!mp->m_ddev_targp)
786 goto out_close_rtdev;
787
788 if (rtdev) {
789 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
790 mp->m_fsname);
791 if (!mp->m_rtdev_targp)
792 goto out_free_ddev_targ;
793 }
794
795 if (logdev && logdev != ddev) {
796 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
797 mp->m_fsname);
798 if (!mp->m_logdev_targp)
799 goto out_free_rtdev_targ;
800 } else {
801 mp->m_logdev_targp = mp->m_ddev_targp;
802 }
803
804 return 0;
805
806 out_free_rtdev_targ:
807 if (mp->m_rtdev_targp)
808 xfs_free_buftarg(mp, mp->m_rtdev_targp);
809 out_free_ddev_targ:
810 xfs_free_buftarg(mp, mp->m_ddev_targp);
811 out_close_rtdev:
812 if (rtdev)
813 xfs_blkdev_put(rtdev);
814 out_close_logdev:
815 if (logdev && logdev != ddev)
816 xfs_blkdev_put(logdev);
817 out:
818 return error;
819 }
820
821 /*
822 * Setup xfs_mount buffer target pointers based on superblock
823 */
824 STATIC int
825 xfs_setup_devices(
826 struct xfs_mount *mp)
827 {
828 int error;
829
830 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
831 mp->m_sb.sb_sectsize);
832 if (error)
833 return error;
834
835 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
836 unsigned int log_sector_size = BBSIZE;
837
838 if (xfs_sb_version_hassector(&mp->m_sb))
839 log_sector_size = mp->m_sb.sb_logsectsize;
840 error = xfs_setsize_buftarg(mp->m_logdev_targp,
841 mp->m_sb.sb_blocksize,
842 log_sector_size);
843 if (error)
844 return error;
845 }
846 if (mp->m_rtdev_targp) {
847 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
848 mp->m_sb.sb_blocksize,
849 mp->m_sb.sb_sectsize);
850 if (error)
851 return error;
852 }
853
854 return 0;
855 }
856
857 STATIC int
858 xfs_init_mount_workqueues(
859 struct xfs_mount *mp)
860 {
861 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
862 WQ_MEM_RECLAIM, 0, mp->m_fsname);
863 if (!mp->m_data_workqueue)
864 goto out;
865
866 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
867 WQ_MEM_RECLAIM, 0, mp->m_fsname);
868 if (!mp->m_unwritten_workqueue)
869 goto out_destroy_data_iodone_queue;
870
871 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
872 WQ_MEM_RECLAIM, 0, mp->m_fsname);
873 if (!mp->m_cil_workqueue)
874 goto out_destroy_unwritten;
875
876 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
877 WQ_NON_REENTRANT, 0, mp->m_fsname);
878 if (!mp->m_reclaim_workqueue)
879 goto out_destroy_cil;
880
881 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
882 WQ_NON_REENTRANT, 0, mp->m_fsname);
883 if (!mp->m_log_workqueue)
884 goto out_destroy_reclaim;
885
886 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
887 WQ_NON_REENTRANT, 0, mp->m_fsname);
888 if (!mp->m_eofblocks_workqueue)
889 goto out_destroy_log;
890
891 return 0;
892
893 out_destroy_log:
894 destroy_workqueue(mp->m_log_workqueue);
895 out_destroy_reclaim:
896 destroy_workqueue(mp->m_reclaim_workqueue);
897 out_destroy_cil:
898 destroy_workqueue(mp->m_cil_workqueue);
899 out_destroy_unwritten:
900 destroy_workqueue(mp->m_unwritten_workqueue);
901 out_destroy_data_iodone_queue:
902 destroy_workqueue(mp->m_data_workqueue);
903 out:
904 return -ENOMEM;
905 }
906
907 STATIC void
908 xfs_destroy_mount_workqueues(
909 struct xfs_mount *mp)
910 {
911 destroy_workqueue(mp->m_eofblocks_workqueue);
912 destroy_workqueue(mp->m_log_workqueue);
913 destroy_workqueue(mp->m_reclaim_workqueue);
914 destroy_workqueue(mp->m_cil_workqueue);
915 destroy_workqueue(mp->m_data_workqueue);
916 destroy_workqueue(mp->m_unwritten_workqueue);
917 }
918
919 /*
920 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
921 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
922 * for IO to complete so that we effectively throttle multiple callers to the
923 * rate at which IO is completing.
924 */
925 void
926 xfs_flush_inodes(
927 struct xfs_mount *mp)
928 {
929 struct super_block *sb = mp->m_super;
930
931 if (down_read_trylock(&sb->s_umount)) {
932 sync_inodes_sb(sb);
933 up_read(&sb->s_umount);
934 }
935 }
936
937 /* Catch misguided souls that try to use this interface on XFS */
938 STATIC struct inode *
939 xfs_fs_alloc_inode(
940 struct super_block *sb)
941 {
942 BUG();
943 return NULL;
944 }
945
946 /*
947 * Now that the generic code is guaranteed not to be accessing
948 * the linux inode, we can reclaim the inode.
949 */
950 STATIC void
951 xfs_fs_destroy_inode(
952 struct inode *inode)
953 {
954 struct xfs_inode *ip = XFS_I(inode);
955
956 trace_xfs_destroy_inode(ip);
957
958 XFS_STATS_INC(vn_reclaim);
959
960 /* bad inode, get out here ASAP */
961 if (is_bad_inode(inode))
962 goto out_reclaim;
963
964 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
965
966 /*
967 * We should never get here with one of the reclaim flags already set.
968 */
969 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
970 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
971
972 /*
973 * We always use background reclaim here because even if the
974 * inode is clean, it still may be under IO and hence we have
975 * to take the flush lock. The background reclaim path handles
976 * this more efficiently than we can here, so simply let background
977 * reclaim tear down all inodes.
978 */
979 out_reclaim:
980 xfs_inode_set_reclaim_tag(ip);
981 }
982
983 /*
984 * Slab object creation initialisation for the XFS inode.
985 * This covers only the idempotent fields in the XFS inode;
986 * all other fields need to be initialised on allocation
987 * from the slab. This avoids the need to repeatedly initialise
988 * fields in the xfs inode that left in the initialise state
989 * when freeing the inode.
990 */
991 STATIC void
992 xfs_fs_inode_init_once(
993 void *inode)
994 {
995 struct xfs_inode *ip = inode;
996
997 memset(ip, 0, sizeof(struct xfs_inode));
998
999 /* vfs inode */
1000 inode_init_once(VFS_I(ip));
1001
1002 /* xfs inode */
1003 atomic_set(&ip->i_pincount, 0);
1004 spin_lock_init(&ip->i_flags_lock);
1005
1006 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1007 "xfsino", ip->i_ino);
1008 }
1009
1010 STATIC void
1011 xfs_fs_evict_inode(
1012 struct inode *inode)
1013 {
1014 xfs_inode_t *ip = XFS_I(inode);
1015
1016 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
1017
1018 trace_xfs_evict_inode(ip);
1019
1020 truncate_inode_pages(&inode->i_data, 0);
1021 clear_inode(inode);
1022 XFS_STATS_INC(vn_rele);
1023 XFS_STATS_INC(vn_remove);
1024 XFS_STATS_DEC(vn_active);
1025
1026 xfs_inactive(ip);
1027 }
1028
1029 /*
1030 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1031 * serialised against cache hits here via the inode->i_lock and igrab() in
1032 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1033 * racing with us, and it avoids needing to grab a spinlock here for every inode
1034 * we drop the final reference on.
1035 */
1036 STATIC int
1037 xfs_fs_drop_inode(
1038 struct inode *inode)
1039 {
1040 struct xfs_inode *ip = XFS_I(inode);
1041
1042 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1043 }
1044
1045 STATIC void
1046 xfs_free_fsname(
1047 struct xfs_mount *mp)
1048 {
1049 kfree(mp->m_fsname);
1050 kfree(mp->m_rtname);
1051 kfree(mp->m_logname);
1052 }
1053
1054 STATIC void
1055 xfs_fs_put_super(
1056 struct super_block *sb)
1057 {
1058 struct xfs_mount *mp = XFS_M(sb);
1059
1060 xfs_filestream_unmount(mp);
1061 xfs_unmountfs(mp);
1062
1063 xfs_freesb(mp);
1064 xfs_icsb_destroy_counters(mp);
1065 xfs_destroy_mount_workqueues(mp);
1066 xfs_close_devices(mp);
1067 xfs_free_fsname(mp);
1068 kfree(mp);
1069 }
1070
1071 STATIC int
1072 xfs_fs_sync_fs(
1073 struct super_block *sb,
1074 int wait)
1075 {
1076 struct xfs_mount *mp = XFS_M(sb);
1077
1078 /*
1079 * Doing anything during the async pass would be counterproductive.
1080 */
1081 if (!wait)
1082 return 0;
1083
1084 xfs_log_force(mp, XFS_LOG_SYNC);
1085 if (laptop_mode) {
1086 /*
1087 * The disk must be active because we're syncing.
1088 * We schedule log work now (now that the disk is
1089 * active) instead of later (when it might not be).
1090 */
1091 flush_delayed_work(&mp->m_log->l_work);
1092 }
1093
1094 return 0;
1095 }
1096
1097 STATIC int
1098 xfs_fs_statfs(
1099 struct dentry *dentry,
1100 struct kstatfs *statp)
1101 {
1102 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1103 xfs_sb_t *sbp = &mp->m_sb;
1104 struct xfs_inode *ip = XFS_I(dentry->d_inode);
1105 __uint64_t fakeinos, id;
1106 xfs_extlen_t lsize;
1107 __int64_t ffree;
1108
1109 statp->f_type = XFS_SB_MAGIC;
1110 statp->f_namelen = MAXNAMELEN - 1;
1111
1112 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1113 statp->f_fsid.val[0] = (u32)id;
1114 statp->f_fsid.val[1] = (u32)(id >> 32);
1115
1116 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1117
1118 spin_lock(&mp->m_sb_lock);
1119 statp->f_bsize = sbp->sb_blocksize;
1120 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1121 statp->f_blocks = sbp->sb_dblocks - lsize;
1122 statp->f_bfree = statp->f_bavail =
1123 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1124 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1125 statp->f_files =
1126 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1127 if (mp->m_maxicount)
1128 statp->f_files = min_t(typeof(statp->f_files),
1129 statp->f_files,
1130 mp->m_maxicount);
1131
1132 /* make sure statp->f_ffree does not underflow */
1133 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1134 statp->f_ffree = max_t(__int64_t, ffree, 0);
1135
1136 spin_unlock(&mp->m_sb_lock);
1137
1138 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1139 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) ==
1140 (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
1141 xfs_qm_statvfs(ip, statp);
1142 return 0;
1143 }
1144
1145 STATIC void
1146 xfs_save_resvblks(struct xfs_mount *mp)
1147 {
1148 __uint64_t resblks = 0;
1149
1150 mp->m_resblks_save = mp->m_resblks;
1151 xfs_reserve_blocks(mp, &resblks, NULL);
1152 }
1153
1154 STATIC void
1155 xfs_restore_resvblks(struct xfs_mount *mp)
1156 {
1157 __uint64_t resblks;
1158
1159 if (mp->m_resblks_save) {
1160 resblks = mp->m_resblks_save;
1161 mp->m_resblks_save = 0;
1162 } else
1163 resblks = xfs_default_resblks(mp);
1164
1165 xfs_reserve_blocks(mp, &resblks, NULL);
1166 }
1167
1168 /*
1169 * Trigger writeback of all the dirty metadata in the file system.
1170 *
1171 * This ensures that the metadata is written to their location on disk rather
1172 * than just existing in transactions in the log. This means after a quiesce
1173 * there is no log replay required to write the inodes to disk - this is the
1174 * primary difference between a sync and a quiesce.
1175 *
1176 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1177 * it is started again when appropriate.
1178 */
1179 void
1180 xfs_quiesce_attr(
1181 struct xfs_mount *mp)
1182 {
1183 int error = 0;
1184
1185 /* wait for all modifications to complete */
1186 while (atomic_read(&mp->m_active_trans) > 0)
1187 delay(100);
1188
1189 /* force the log to unpin objects from the now complete transactions */
1190 xfs_log_force(mp, XFS_LOG_SYNC);
1191
1192 /* reclaim inodes to do any IO before the freeze completes */
1193 xfs_reclaim_inodes(mp, 0);
1194 xfs_reclaim_inodes(mp, SYNC_WAIT);
1195
1196 /* Push the superblock and write an unmount record */
1197 error = xfs_log_sbcount(mp);
1198 if (error)
1199 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1200 "Frozen image may not be consistent.");
1201 /*
1202 * Just warn here till VFS can correctly support
1203 * read-only remount without racing.
1204 */
1205 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1206
1207 xfs_log_quiesce(mp);
1208 }
1209
1210 STATIC int
1211 xfs_fs_remount(
1212 struct super_block *sb,
1213 int *flags,
1214 char *options)
1215 {
1216 struct xfs_mount *mp = XFS_M(sb);
1217 substring_t args[MAX_OPT_ARGS];
1218 char *p;
1219 int error;
1220
1221 while ((p = strsep(&options, ",")) != NULL) {
1222 int token;
1223
1224 if (!*p)
1225 continue;
1226
1227 token = match_token(p, tokens, args);
1228 switch (token) {
1229 case Opt_barrier:
1230 mp->m_flags |= XFS_MOUNT_BARRIER;
1231 break;
1232 case Opt_nobarrier:
1233 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1234 break;
1235 case Opt_inode64:
1236 mp->m_maxagi = xfs_set_inode64(mp);
1237 break;
1238 case Opt_inode32:
1239 mp->m_maxagi = xfs_set_inode32(mp);
1240 break;
1241 default:
1242 /*
1243 * Logically we would return an error here to prevent
1244 * users from believing they might have changed
1245 * mount options using remount which can't be changed.
1246 *
1247 * But unfortunately mount(8) adds all options from
1248 * mtab and fstab to the mount arguments in some cases
1249 * so we can't blindly reject options, but have to
1250 * check for each specified option if it actually
1251 * differs from the currently set option and only
1252 * reject it if that's the case.
1253 *
1254 * Until that is implemented we return success for
1255 * every remount request, and silently ignore all
1256 * options that we can't actually change.
1257 */
1258 #if 0
1259 xfs_info(mp,
1260 "mount option \"%s\" not supported for remount\n", p);
1261 return -EINVAL;
1262 #else
1263 break;
1264 #endif
1265 }
1266 }
1267
1268 /* ro -> rw */
1269 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1270 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1271
1272 /*
1273 * If this is the first remount to writeable state we
1274 * might have some superblock changes to update.
1275 */
1276 if (mp->m_update_flags) {
1277 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1278 if (error) {
1279 xfs_warn(mp, "failed to write sb changes");
1280 return error;
1281 }
1282 mp->m_update_flags = 0;
1283 }
1284
1285 /*
1286 * Fill out the reserve pool if it is empty. Use the stashed
1287 * value if it is non-zero, otherwise go with the default.
1288 */
1289 xfs_restore_resvblks(mp);
1290 xfs_log_work_queue(mp);
1291 }
1292
1293 /* rw -> ro */
1294 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1295 /*
1296 * Before we sync the metadata, we need to free up the reserve
1297 * block pool so that the used block count in the superblock on
1298 * disk is correct at the end of the remount. Stash the current
1299 * reserve pool size so that if we get remounted rw, we can
1300 * return it to the same size.
1301 */
1302 xfs_save_resvblks(mp);
1303 xfs_quiesce_attr(mp);
1304 mp->m_flags |= XFS_MOUNT_RDONLY;
1305 }
1306
1307 return 0;
1308 }
1309
1310 /*
1311 * Second stage of a freeze. The data is already frozen so we only
1312 * need to take care of the metadata. Once that's done write a dummy
1313 * record to dirty the log in case of a crash while frozen.
1314 */
1315 STATIC int
1316 xfs_fs_freeze(
1317 struct super_block *sb)
1318 {
1319 struct xfs_mount *mp = XFS_M(sb);
1320
1321 xfs_save_resvblks(mp);
1322 xfs_quiesce_attr(mp);
1323 return -xfs_fs_log_dummy(mp);
1324 }
1325
1326 STATIC int
1327 xfs_fs_unfreeze(
1328 struct super_block *sb)
1329 {
1330 struct xfs_mount *mp = XFS_M(sb);
1331
1332 xfs_restore_resvblks(mp);
1333 xfs_log_work_queue(mp);
1334 return 0;
1335 }
1336
1337 STATIC int
1338 xfs_fs_show_options(
1339 struct seq_file *m,
1340 struct dentry *root)
1341 {
1342 return -xfs_showargs(XFS_M(root->d_sb), m);
1343 }
1344
1345 /*
1346 * This function fills in xfs_mount_t fields based on mount args.
1347 * Note: the superblock _has_ now been read in.
1348 */
1349 STATIC int
1350 xfs_finish_flags(
1351 struct xfs_mount *mp)
1352 {
1353 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1354
1355 /* Fail a mount where the logbuf is smaller than the log stripe */
1356 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1357 if (mp->m_logbsize <= 0 &&
1358 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1359 mp->m_logbsize = mp->m_sb.sb_logsunit;
1360 } else if (mp->m_logbsize > 0 &&
1361 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1362 xfs_warn(mp,
1363 "logbuf size must be greater than or equal to log stripe size");
1364 return XFS_ERROR(EINVAL);
1365 }
1366 } else {
1367 /* Fail a mount if the logbuf is larger than 32K */
1368 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1369 xfs_warn(mp,
1370 "logbuf size for version 1 logs must be 16K or 32K");
1371 return XFS_ERROR(EINVAL);
1372 }
1373 }
1374
1375 /*
1376 * V5 filesystems always use attr2 format for attributes.
1377 */
1378 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1379 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1380 xfs_warn(mp,
1381 "Cannot mount a V5 filesystem as %s. %s is always enabled for V5 filesystems.",
1382 MNTOPT_NOATTR2, MNTOPT_ATTR2);
1383 return XFS_ERROR(EINVAL);
1384 }
1385
1386 /*
1387 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1388 * told by noattr2 to turn it off
1389 */
1390 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1391 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1392 mp->m_flags |= XFS_MOUNT_ATTR2;
1393
1394 /*
1395 * prohibit r/w mounts of read-only filesystems
1396 */
1397 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1398 xfs_warn(mp,
1399 "cannot mount a read-only filesystem as read-write");
1400 return XFS_ERROR(EROFS);
1401 }
1402
1403 return 0;
1404 }
1405
1406 STATIC int
1407 xfs_fs_fill_super(
1408 struct super_block *sb,
1409 void *data,
1410 int silent)
1411 {
1412 struct inode *root;
1413 struct xfs_mount *mp = NULL;
1414 int flags = 0, error = ENOMEM;
1415
1416 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1417 if (!mp)
1418 goto out;
1419
1420 spin_lock_init(&mp->m_sb_lock);
1421 mutex_init(&mp->m_growlock);
1422 atomic_set(&mp->m_active_trans, 0);
1423 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1424 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1425
1426 mp->m_super = sb;
1427 sb->s_fs_info = mp;
1428
1429 error = xfs_parseargs(mp, (char *)data);
1430 if (error)
1431 goto out_free_fsname;
1432
1433 sb_min_blocksize(sb, BBSIZE);
1434 sb->s_xattr = xfs_xattr_handlers;
1435 sb->s_export_op = &xfs_export_operations;
1436 #ifdef CONFIG_XFS_QUOTA
1437 sb->s_qcop = &xfs_quotactl_operations;
1438 #endif
1439 sb->s_op = &xfs_super_operations;
1440
1441 if (silent)
1442 flags |= XFS_MFSI_QUIET;
1443
1444 error = xfs_open_devices(mp);
1445 if (error)
1446 goto out_free_fsname;
1447
1448 error = xfs_init_mount_workqueues(mp);
1449 if (error)
1450 goto out_close_devices;
1451
1452 error = xfs_icsb_init_counters(mp);
1453 if (error)
1454 goto out_destroy_workqueues;
1455
1456 error = xfs_readsb(mp, flags);
1457 if (error)
1458 goto out_destroy_counters;
1459
1460 error = xfs_finish_flags(mp);
1461 if (error)
1462 goto out_free_sb;
1463
1464 error = xfs_setup_devices(mp);
1465 if (error)
1466 goto out_free_sb;
1467
1468 error = xfs_filestream_mount(mp);
1469 if (error)
1470 goto out_free_sb;
1471
1472 /*
1473 * we must configure the block size in the superblock before we run the
1474 * full mount process as the mount process can lookup and cache inodes.
1475 */
1476 sb->s_magic = XFS_SB_MAGIC;
1477 sb->s_blocksize = mp->m_sb.sb_blocksize;
1478 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1479 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1480 sb->s_max_links = XFS_MAXLINK;
1481 sb->s_time_gran = 1;
1482 set_posix_acl_flag(sb);
1483
1484 error = xfs_mountfs(mp);
1485 if (error)
1486 goto out_filestream_unmount;
1487
1488 root = igrab(VFS_I(mp->m_rootip));
1489 if (!root) {
1490 error = ENOENT;
1491 goto out_unmount;
1492 }
1493 if (is_bad_inode(root)) {
1494 error = EINVAL;
1495 goto out_unmount;
1496 }
1497 sb->s_root = d_make_root(root);
1498 if (!sb->s_root) {
1499 error = ENOMEM;
1500 goto out_unmount;
1501 }
1502
1503 return 0;
1504
1505 out_filestream_unmount:
1506 xfs_filestream_unmount(mp);
1507 out_free_sb:
1508 xfs_freesb(mp);
1509 out_destroy_counters:
1510 xfs_icsb_destroy_counters(mp);
1511 out_destroy_workqueues:
1512 xfs_destroy_mount_workqueues(mp);
1513 out_close_devices:
1514 xfs_close_devices(mp);
1515 out_free_fsname:
1516 xfs_free_fsname(mp);
1517 kfree(mp);
1518 out:
1519 return -error;
1520
1521 out_unmount:
1522 xfs_filestream_unmount(mp);
1523 xfs_unmountfs(mp);
1524 goto out_free_sb;
1525 }
1526
1527 STATIC struct dentry *
1528 xfs_fs_mount(
1529 struct file_system_type *fs_type,
1530 int flags,
1531 const char *dev_name,
1532 void *data)
1533 {
1534 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1535 }
1536
1537 static int
1538 xfs_fs_nr_cached_objects(
1539 struct super_block *sb)
1540 {
1541 return xfs_reclaim_inodes_count(XFS_M(sb));
1542 }
1543
1544 static void
1545 xfs_fs_free_cached_objects(
1546 struct super_block *sb,
1547 int nr_to_scan)
1548 {
1549 xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1550 }
1551
1552 static const struct super_operations xfs_super_operations = {
1553 .alloc_inode = xfs_fs_alloc_inode,
1554 .destroy_inode = xfs_fs_destroy_inode,
1555 .evict_inode = xfs_fs_evict_inode,
1556 .drop_inode = xfs_fs_drop_inode,
1557 .put_super = xfs_fs_put_super,
1558 .sync_fs = xfs_fs_sync_fs,
1559 .freeze_fs = xfs_fs_freeze,
1560 .unfreeze_fs = xfs_fs_unfreeze,
1561 .statfs = xfs_fs_statfs,
1562 .remount_fs = xfs_fs_remount,
1563 .show_options = xfs_fs_show_options,
1564 .nr_cached_objects = xfs_fs_nr_cached_objects,
1565 .free_cached_objects = xfs_fs_free_cached_objects,
1566 };
1567
1568 static struct file_system_type xfs_fs_type = {
1569 .owner = THIS_MODULE,
1570 .name = "xfs",
1571 .mount = xfs_fs_mount,
1572 .kill_sb = kill_block_super,
1573 .fs_flags = FS_REQUIRES_DEV,
1574 };
1575 MODULE_ALIAS_FS("xfs");
1576
1577 STATIC int __init
1578 xfs_init_zones(void)
1579 {
1580
1581 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1582 if (!xfs_ioend_zone)
1583 goto out;
1584
1585 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1586 xfs_ioend_zone);
1587 if (!xfs_ioend_pool)
1588 goto out_destroy_ioend_zone;
1589
1590 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1591 "xfs_log_ticket");
1592 if (!xfs_log_ticket_zone)
1593 goto out_destroy_ioend_pool;
1594
1595 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1596 "xfs_bmap_free_item");
1597 if (!xfs_bmap_free_item_zone)
1598 goto out_destroy_log_ticket_zone;
1599
1600 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1601 "xfs_btree_cur");
1602 if (!xfs_btree_cur_zone)
1603 goto out_destroy_bmap_free_item_zone;
1604
1605 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1606 "xfs_da_state");
1607 if (!xfs_da_state_zone)
1608 goto out_destroy_btree_cur_zone;
1609
1610 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1611 if (!xfs_ifork_zone)
1612 goto out_destroy_da_state_zone;
1613
1614 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1615 if (!xfs_trans_zone)
1616 goto out_destroy_ifork_zone;
1617
1618 xfs_log_item_desc_zone =
1619 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1620 "xfs_log_item_desc");
1621 if (!xfs_log_item_desc_zone)
1622 goto out_destroy_trans_zone;
1623
1624 /*
1625 * The size of the zone allocated buf log item is the maximum
1626 * size possible under XFS. This wastes a little bit of memory,
1627 * but it is much faster.
1628 */
1629 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1630 "xfs_buf_item");
1631 if (!xfs_buf_item_zone)
1632 goto out_destroy_log_item_desc_zone;
1633
1634 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1635 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1636 sizeof(xfs_extent_t))), "xfs_efd_item");
1637 if (!xfs_efd_zone)
1638 goto out_destroy_buf_item_zone;
1639
1640 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1641 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1642 sizeof(xfs_extent_t))), "xfs_efi_item");
1643 if (!xfs_efi_zone)
1644 goto out_destroy_efd_zone;
1645
1646 xfs_inode_zone =
1647 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1648 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1649 xfs_fs_inode_init_once);
1650 if (!xfs_inode_zone)
1651 goto out_destroy_efi_zone;
1652
1653 xfs_ili_zone =
1654 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1655 KM_ZONE_SPREAD, NULL);
1656 if (!xfs_ili_zone)
1657 goto out_destroy_inode_zone;
1658
1659 return 0;
1660
1661 out_destroy_inode_zone:
1662 kmem_zone_destroy(xfs_inode_zone);
1663 out_destroy_efi_zone:
1664 kmem_zone_destroy(xfs_efi_zone);
1665 out_destroy_efd_zone:
1666 kmem_zone_destroy(xfs_efd_zone);
1667 out_destroy_buf_item_zone:
1668 kmem_zone_destroy(xfs_buf_item_zone);
1669 out_destroy_log_item_desc_zone:
1670 kmem_zone_destroy(xfs_log_item_desc_zone);
1671 out_destroy_trans_zone:
1672 kmem_zone_destroy(xfs_trans_zone);
1673 out_destroy_ifork_zone:
1674 kmem_zone_destroy(xfs_ifork_zone);
1675 out_destroy_da_state_zone:
1676 kmem_zone_destroy(xfs_da_state_zone);
1677 out_destroy_btree_cur_zone:
1678 kmem_zone_destroy(xfs_btree_cur_zone);
1679 out_destroy_bmap_free_item_zone:
1680 kmem_zone_destroy(xfs_bmap_free_item_zone);
1681 out_destroy_log_ticket_zone:
1682 kmem_zone_destroy(xfs_log_ticket_zone);
1683 out_destroy_ioend_pool:
1684 mempool_destroy(xfs_ioend_pool);
1685 out_destroy_ioend_zone:
1686 kmem_zone_destroy(xfs_ioend_zone);
1687 out:
1688 return -ENOMEM;
1689 }
1690
1691 STATIC void
1692 xfs_destroy_zones(void)
1693 {
1694 /*
1695 * Make sure all delayed rcu free are flushed before we
1696 * destroy caches.
1697 */
1698 rcu_barrier();
1699 kmem_zone_destroy(xfs_ili_zone);
1700 kmem_zone_destroy(xfs_inode_zone);
1701 kmem_zone_destroy(xfs_efi_zone);
1702 kmem_zone_destroy(xfs_efd_zone);
1703 kmem_zone_destroy(xfs_buf_item_zone);
1704 kmem_zone_destroy(xfs_log_item_desc_zone);
1705 kmem_zone_destroy(xfs_trans_zone);
1706 kmem_zone_destroy(xfs_ifork_zone);
1707 kmem_zone_destroy(xfs_da_state_zone);
1708 kmem_zone_destroy(xfs_btree_cur_zone);
1709 kmem_zone_destroy(xfs_bmap_free_item_zone);
1710 kmem_zone_destroy(xfs_log_ticket_zone);
1711 mempool_destroy(xfs_ioend_pool);
1712 kmem_zone_destroy(xfs_ioend_zone);
1713
1714 }
1715
1716 STATIC int __init
1717 xfs_init_workqueues(void)
1718 {
1719 /*
1720 * The allocation workqueue can be used in memory reclaim situations
1721 * (writepage path), and parallelism is only limited by the number of
1722 * AGs in all the filesystems mounted. Hence use the default large
1723 * max_active value for this workqueue.
1724 */
1725 xfs_alloc_wq = alloc_workqueue("xfsalloc", WQ_MEM_RECLAIM, 0);
1726 if (!xfs_alloc_wq)
1727 return -ENOMEM;
1728
1729 return 0;
1730 }
1731
1732 STATIC void
1733 xfs_destroy_workqueues(void)
1734 {
1735 destroy_workqueue(xfs_alloc_wq);
1736 }
1737
1738 STATIC int __init
1739 init_xfs_fs(void)
1740 {
1741 int error;
1742
1743 printk(KERN_INFO XFS_VERSION_STRING " with "
1744 XFS_BUILD_OPTIONS " enabled\n");
1745
1746 xfs_dir_startup();
1747
1748 error = xfs_init_zones();
1749 if (error)
1750 goto out;
1751
1752 error = xfs_init_workqueues();
1753 if (error)
1754 goto out_destroy_zones;
1755
1756 error = xfs_mru_cache_init();
1757 if (error)
1758 goto out_destroy_wq;
1759
1760 error = xfs_filestream_init();
1761 if (error)
1762 goto out_mru_cache_uninit;
1763
1764 error = xfs_buf_init();
1765 if (error)
1766 goto out_filestream_uninit;
1767
1768 error = xfs_init_procfs();
1769 if (error)
1770 goto out_buf_terminate;
1771
1772 error = xfs_sysctl_register();
1773 if (error)
1774 goto out_cleanup_procfs;
1775
1776 error = xfs_qm_init();
1777 if (error)
1778 goto out_sysctl_unregister;
1779
1780 error = register_filesystem(&xfs_fs_type);
1781 if (error)
1782 goto out_qm_exit;
1783 return 0;
1784
1785 out_qm_exit:
1786 xfs_qm_exit();
1787 out_sysctl_unregister:
1788 xfs_sysctl_unregister();
1789 out_cleanup_procfs:
1790 xfs_cleanup_procfs();
1791 out_buf_terminate:
1792 xfs_buf_terminate();
1793 out_filestream_uninit:
1794 xfs_filestream_uninit();
1795 out_mru_cache_uninit:
1796 xfs_mru_cache_uninit();
1797 out_destroy_wq:
1798 xfs_destroy_workqueues();
1799 out_destroy_zones:
1800 xfs_destroy_zones();
1801 out:
1802 return error;
1803 }
1804
1805 STATIC void __exit
1806 exit_xfs_fs(void)
1807 {
1808 xfs_qm_exit();
1809 unregister_filesystem(&xfs_fs_type);
1810 xfs_sysctl_unregister();
1811 xfs_cleanup_procfs();
1812 xfs_buf_terminate();
1813 xfs_filestream_uninit();
1814 xfs_mru_cache_uninit();
1815 xfs_destroy_workqueues();
1816 xfs_destroy_zones();
1817 }
1818
1819 module_init(init_xfs_fs);
1820 module_exit(exit_xfs_fs);
1821
1822 MODULE_AUTHOR("Silicon Graphics, Inc.");
1823 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1824 MODULE_LICENSE("GPL");
Linus' kernel tree