Loading...
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_sync.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
65static const struct super_operations xfs_super_operations;
66static kmem_zone_t *xfs_ioend_zone;
67mempool_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_IKEEP "ikeep" /* do not free empty inode clusters */
92#define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
93#define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
94#define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
95 * in stat(). */
96#define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
97#define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
98#define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
99#define MNTOPT_QUOTA "quota" /* disk quotas (user) */
100#define MNTOPT_NOQUOTA "noquota" /* no quotas */
101#define MNTOPT_USRQUOTA "usrquota" /* user quota enabled */
102#define MNTOPT_GRPQUOTA "grpquota" /* group quota enabled */
103#define MNTOPT_PRJQUOTA "prjquota" /* project quota enabled */
104#define MNTOPT_UQUOTA "uquota" /* user quota (IRIX variant) */
105#define MNTOPT_GQUOTA "gquota" /* group quota (IRIX variant) */
106#define MNTOPT_PQUOTA "pquota" /* project quota (IRIX variant) */
107#define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
108#define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
109#define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
110#define MNTOPT_QUOTANOENF "qnoenforce" /* same as uqnoenforce */
111#define MNTOPT_DELAYLOG "delaylog" /* Delayed logging enabled */
112#define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed logging disabled */
113#define MNTOPT_DISCARD "discard" /* Discard unused blocks */
114#define MNTOPT_NODISCARD "nodiscard" /* Do not discard unused blocks */
115
116/*
117 * Table driven mount option parser.
118 *
119 * Currently only used for remount, but it will be used for mount
120 * in the future, too.
121 */
122enum {
123 Opt_barrier, Opt_nobarrier, Opt_err
124};
125
126static const match_table_t tokens = {
127 {Opt_barrier, "barrier"},
128 {Opt_nobarrier, "nobarrier"},
129 {Opt_err, NULL}
130};
131
132
133STATIC unsigned long
134suffix_strtoul(char *s, char **endp, unsigned int base)
135{
136 int last, shift_left_factor = 0;
137 char *value = s;
138
139 last = strlen(value) - 1;
140 if (value[last] == 'K' || value[last] == 'k') {
141 shift_left_factor = 10;
142 value[last] = '\0';
143 }
144 if (value[last] == 'M' || value[last] == 'm') {
145 shift_left_factor = 20;
146 value[last] = '\0';
147 }
148 if (value[last] == 'G' || value[last] == 'g') {
149 shift_left_factor = 30;
150 value[last] = '\0';
151 }
152
153 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
154}
155
156/*
157 * This function fills in xfs_mount_t fields based on mount args.
158 * Note: the superblock has _not_ yet been read in.
159 *
160 * Note that this function leaks the various device name allocations on
161 * failure. The caller takes care of them.
162 */
163STATIC int
164xfs_parseargs(
165 struct xfs_mount *mp,
166 char *options)
167{
168 struct super_block *sb = mp->m_super;
169 char *this_char, *value, *eov;
170 int dsunit = 0;
171 int dswidth = 0;
172 int iosize = 0;
173 __uint8_t iosizelog = 0;
174
175 /*
176 * set up the mount name first so all the errors will refer to the
177 * correct device.
178 */
179 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
180 if (!mp->m_fsname)
181 return ENOMEM;
182 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
183
184 /*
185 * Copy binary VFS mount flags we are interested in.
186 */
187 if (sb->s_flags & MS_RDONLY)
188 mp->m_flags |= XFS_MOUNT_RDONLY;
189 if (sb->s_flags & MS_DIRSYNC)
190 mp->m_flags |= XFS_MOUNT_DIRSYNC;
191 if (sb->s_flags & MS_SYNCHRONOUS)
192 mp->m_flags |= XFS_MOUNT_WSYNC;
193
194 /*
195 * Set some default flags that could be cleared by the mount option
196 * parsing.
197 */
198 mp->m_flags |= XFS_MOUNT_BARRIER;
199 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
200 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
201
202 /*
203 * These can be overridden by the mount option parsing.
204 */
205 mp->m_logbufs = -1;
206 mp->m_logbsize = -1;
207
208 if (!options)
209 goto done;
210
211 while ((this_char = strsep(&options, ",")) != NULL) {
212 if (!*this_char)
213 continue;
214 if ((value = strchr(this_char, '=')) != NULL)
215 *value++ = 0;
216
217 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
218 if (!value || !*value) {
219 xfs_warn(mp, "%s option requires an argument",
220 this_char);
221 return EINVAL;
222 }
223 mp->m_logbufs = simple_strtoul(value, &eov, 10);
224 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
225 if (!value || !*value) {
226 xfs_warn(mp, "%s option requires an argument",
227 this_char);
228 return EINVAL;
229 }
230 mp->m_logbsize = suffix_strtoul(value, &eov, 10);
231 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
232 if (!value || !*value) {
233 xfs_warn(mp, "%s option requires an argument",
234 this_char);
235 return EINVAL;
236 }
237 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
238 if (!mp->m_logname)
239 return ENOMEM;
240 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
241 xfs_warn(mp, "%s option not allowed on this system",
242 this_char);
243 return EINVAL;
244 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
245 if (!value || !*value) {
246 xfs_warn(mp, "%s option requires an argument",
247 this_char);
248 return EINVAL;
249 }
250 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
251 if (!mp->m_rtname)
252 return ENOMEM;
253 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
254 if (!value || !*value) {
255 xfs_warn(mp, "%s option requires an argument",
256 this_char);
257 return EINVAL;
258 }
259 iosize = simple_strtoul(value, &eov, 10);
260 iosizelog = ffs(iosize) - 1;
261 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
262 if (!value || !*value) {
263 xfs_warn(mp, "%s option requires an argument",
264 this_char);
265 return EINVAL;
266 }
267 iosize = suffix_strtoul(value, &eov, 10);
268 iosizelog = ffs(iosize) - 1;
269 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
270 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
271 mp->m_flags |= XFS_MOUNT_GRPID;
272 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
273 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
274 mp->m_flags &= ~XFS_MOUNT_GRPID;
275 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
276 mp->m_flags |= XFS_MOUNT_WSYNC;
277 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
278 mp->m_flags |= XFS_MOUNT_NORECOVERY;
279 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
280 mp->m_flags |= XFS_MOUNT_NOALIGN;
281 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
282 mp->m_flags |= XFS_MOUNT_SWALLOC;
283 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
284 if (!value || !*value) {
285 xfs_warn(mp, "%s option requires an argument",
286 this_char);
287 return EINVAL;
288 }
289 dsunit = simple_strtoul(value, &eov, 10);
290 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
291 if (!value || !*value) {
292 xfs_warn(mp, "%s option requires an argument",
293 this_char);
294 return EINVAL;
295 }
296 dswidth = simple_strtoul(value, &eov, 10);
297 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
298 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
299#if !XFS_BIG_INUMS
300 xfs_warn(mp, "%s option not allowed on this system",
301 this_char);
302 return EINVAL;
303#endif
304 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
305 mp->m_flags |= XFS_MOUNT_NOUUID;
306 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
307 mp->m_flags |= XFS_MOUNT_BARRIER;
308 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
309 mp->m_flags &= ~XFS_MOUNT_BARRIER;
310 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
311 mp->m_flags |= XFS_MOUNT_IKEEP;
312 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
313 mp->m_flags &= ~XFS_MOUNT_IKEEP;
314 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
315 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
316 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
317 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
318 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
319 mp->m_flags |= XFS_MOUNT_ATTR2;
320 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
321 mp->m_flags &= ~XFS_MOUNT_ATTR2;
322 mp->m_flags |= XFS_MOUNT_NOATTR2;
323 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
324 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
325 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
326 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
327 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
328 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
329 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
330 !strcmp(this_char, MNTOPT_UQUOTA) ||
331 !strcmp(this_char, MNTOPT_USRQUOTA)) {
332 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
333 XFS_UQUOTA_ENFD);
334 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
335 !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
336 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
337 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
338 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
339 !strcmp(this_char, MNTOPT_PRJQUOTA)) {
340 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
341 XFS_OQUOTA_ENFD);
342 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
343 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
344 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
345 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
346 !strcmp(this_char, MNTOPT_GRPQUOTA)) {
347 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
348 XFS_OQUOTA_ENFD);
349 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
350 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
351 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
352 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
353 xfs_warn(mp,
354 "delaylog is the default now, option is deprecated.");
355 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
356 xfs_warn(mp,
357 "nodelaylog support has been removed, option is deprecated.");
358 } else if (!strcmp(this_char, MNTOPT_DISCARD)) {
359 mp->m_flags |= XFS_MOUNT_DISCARD;
360 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
361 mp->m_flags &= ~XFS_MOUNT_DISCARD;
362 } else if (!strcmp(this_char, "ihashsize")) {
363 xfs_warn(mp,
364 "ihashsize no longer used, option is deprecated.");
365 } else if (!strcmp(this_char, "osyncisdsync")) {
366 xfs_warn(mp,
367 "osyncisdsync has no effect, option is deprecated.");
368 } else if (!strcmp(this_char, "osyncisosync")) {
369 xfs_warn(mp,
370 "osyncisosync has no effect, option is deprecated.");
371 } else if (!strcmp(this_char, "irixsgid")) {
372 xfs_warn(mp,
373 "irixsgid is now a sysctl(2) variable, option is deprecated.");
374 } else {
375 xfs_warn(mp, "unknown mount option [%s].", this_char);
376 return EINVAL;
377 }
378 }
379
380 /*
381 * no recovery flag requires a read-only mount
382 */
383 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
384 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
385 xfs_warn(mp, "no-recovery mounts must be read-only.");
386 return EINVAL;
387 }
388
389 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
390 xfs_warn(mp,
391 "sunit and swidth options incompatible with the noalign option");
392 return EINVAL;
393 }
394
395#ifndef CONFIG_XFS_QUOTA
396 if (XFS_IS_QUOTA_RUNNING(mp)) {
397 xfs_warn(mp, "quota support not available in this kernel.");
398 return EINVAL;
399 }
400#endif
401
402 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
403 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
404 xfs_warn(mp, "cannot mount with both project and group quota");
405 return EINVAL;
406 }
407
408 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
409 xfs_warn(mp, "sunit and swidth must be specified together");
410 return EINVAL;
411 }
412
413 if (dsunit && (dswidth % dsunit != 0)) {
414 xfs_warn(mp,
415 "stripe width (%d) must be a multiple of the stripe unit (%d)",
416 dswidth, dsunit);
417 return EINVAL;
418 }
419
420done:
421 if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
422 /*
423 * At this point the superblock has not been read
424 * in, therefore we do not know the block size.
425 * Before the mount call ends we will convert
426 * these to FSBs.
427 */
428 if (dsunit) {
429 mp->m_dalign = dsunit;
430 mp->m_flags |= XFS_MOUNT_RETERR;
431 }
432
433 if (dswidth)
434 mp->m_swidth = dswidth;
435 }
436
437 if (mp->m_logbufs != -1 &&
438 mp->m_logbufs != 0 &&
439 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
440 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
441 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
442 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
443 return XFS_ERROR(EINVAL);
444 }
445 if (mp->m_logbsize != -1 &&
446 mp->m_logbsize != 0 &&
447 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
448 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
449 !is_power_of_2(mp->m_logbsize))) {
450 xfs_warn(mp,
451 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
452 mp->m_logbsize);
453 return XFS_ERROR(EINVAL);
454 }
455
456 if (iosizelog) {
457 if (iosizelog > XFS_MAX_IO_LOG ||
458 iosizelog < XFS_MIN_IO_LOG) {
459 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
460 iosizelog, XFS_MIN_IO_LOG,
461 XFS_MAX_IO_LOG);
462 return XFS_ERROR(EINVAL);
463 }
464
465 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
466 mp->m_readio_log = iosizelog;
467 mp->m_writeio_log = iosizelog;
468 }
469
470 return 0;
471}
472
473struct proc_xfs_info {
474 int flag;
475 char *str;
476};
477
478STATIC int
479xfs_showargs(
480 struct xfs_mount *mp,
481 struct seq_file *m)
482{
483 static struct proc_xfs_info xfs_info_set[] = {
484 /* the few simple ones we can get from the mount struct */
485 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP },
486 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
487 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
488 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
489 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
490 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
491 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 },
492 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
493 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
494 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD },
495 { 0, NULL }
496 };
497 static struct proc_xfs_info xfs_info_unset[] = {
498 /* the few simple ones we can get from the mount struct */
499 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO },
500 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER },
501 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE },
502 { 0, NULL }
503 };
504 struct proc_xfs_info *xfs_infop;
505
506 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
507 if (mp->m_flags & xfs_infop->flag)
508 seq_puts(m, xfs_infop->str);
509 }
510 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
511 if (!(mp->m_flags & xfs_infop->flag))
512 seq_puts(m, xfs_infop->str);
513 }
514
515 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
516 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
517 (int)(1 << mp->m_writeio_log) >> 10);
518
519 if (mp->m_logbufs > 0)
520 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
521 if (mp->m_logbsize > 0)
522 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
523
524 if (mp->m_logname)
525 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
526 if (mp->m_rtname)
527 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
528
529 if (mp->m_dalign > 0)
530 seq_printf(m, "," MNTOPT_SUNIT "=%d",
531 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
532 if (mp->m_swidth > 0)
533 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
534 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
535
536 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
537 seq_puts(m, "," MNTOPT_USRQUOTA);
538 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
539 seq_puts(m, "," MNTOPT_UQUOTANOENF);
540
541 /* Either project or group quotas can be active, not both */
542
543 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
544 if (mp->m_qflags & XFS_OQUOTA_ENFD)
545 seq_puts(m, "," MNTOPT_PRJQUOTA);
546 else
547 seq_puts(m, "," MNTOPT_PQUOTANOENF);
548 } else if (mp->m_qflags & XFS_GQUOTA_ACCT) {
549 if (mp->m_qflags & XFS_OQUOTA_ENFD)
550 seq_puts(m, "," MNTOPT_GRPQUOTA);
551 else
552 seq_puts(m, "," MNTOPT_GQUOTANOENF);
553 }
554
555 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
556 seq_puts(m, "," MNTOPT_NOQUOTA);
557
558 return 0;
559}
560__uint64_t
561xfs_max_file_offset(
562 unsigned int blockshift)
563{
564 unsigned int pagefactor = 1;
565 unsigned int bitshift = BITS_PER_LONG - 1;
566
567 /* Figure out maximum filesize, on Linux this can depend on
568 * the filesystem blocksize (on 32 bit platforms).
569 * __block_write_begin does this in an [unsigned] long...
570 * page->index << (PAGE_CACHE_SHIFT - bbits)
571 * So, for page sized blocks (4K on 32 bit platforms),
572 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
573 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
574 * but for smaller blocksizes it is less (bbits = log2 bsize).
575 * Note1: get_block_t takes a long (implicit cast from above)
576 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
577 * can optionally convert the [unsigned] long from above into
578 * an [unsigned] long long.
579 */
580
581#if BITS_PER_LONG == 32
582# if defined(CONFIG_LBDAF)
583 ASSERT(sizeof(sector_t) == 8);
584 pagefactor = PAGE_CACHE_SIZE;
585 bitshift = BITS_PER_LONG;
586# else
587 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
588# endif
589#endif
590
591 return (((__uint64_t)pagefactor) << bitshift) - 1;
592}
593
594STATIC int
595xfs_blkdev_get(
596 xfs_mount_t *mp,
597 const char *name,
598 struct block_device **bdevp)
599{
600 int error = 0;
601
602 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
603 mp);
604 if (IS_ERR(*bdevp)) {
605 error = PTR_ERR(*bdevp);
606 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
607 }
608
609 return -error;
610}
611
612STATIC void
613xfs_blkdev_put(
614 struct block_device *bdev)
615{
616 if (bdev)
617 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
618}
619
620void
621xfs_blkdev_issue_flush(
622 xfs_buftarg_t *buftarg)
623{
624 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
625}
626
627STATIC void
628xfs_close_devices(
629 struct xfs_mount *mp)
630{
631 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
632 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
633 xfs_free_buftarg(mp, mp->m_logdev_targp);
634 xfs_blkdev_put(logdev);
635 }
636 if (mp->m_rtdev_targp) {
637 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
638 xfs_free_buftarg(mp, mp->m_rtdev_targp);
639 xfs_blkdev_put(rtdev);
640 }
641 xfs_free_buftarg(mp, mp->m_ddev_targp);
642}
643
644/*
645 * The file system configurations are:
646 * (1) device (partition) with data and internal log
647 * (2) logical volume with data and log subvolumes.
648 * (3) logical volume with data, log, and realtime subvolumes.
649 *
650 * We only have to handle opening the log and realtime volumes here if
651 * they are present. The data subvolume has already been opened by
652 * get_sb_bdev() and is stored in sb->s_bdev.
653 */
654STATIC int
655xfs_open_devices(
656 struct xfs_mount *mp)
657{
658 struct block_device *ddev = mp->m_super->s_bdev;
659 struct block_device *logdev = NULL, *rtdev = NULL;
660 int error;
661
662 /*
663 * Open real time and log devices - order is important.
664 */
665 if (mp->m_logname) {
666 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
667 if (error)
668 goto out;
669 }
670
671 if (mp->m_rtname) {
672 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
673 if (error)
674 goto out_close_logdev;
675
676 if (rtdev == ddev || rtdev == logdev) {
677 xfs_warn(mp,
678 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
679 error = EINVAL;
680 goto out_close_rtdev;
681 }
682 }
683
684 /*
685 * Setup xfs_mount buffer target pointers
686 */
687 error = ENOMEM;
688 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
689 if (!mp->m_ddev_targp)
690 goto out_close_rtdev;
691
692 if (rtdev) {
693 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
694 mp->m_fsname);
695 if (!mp->m_rtdev_targp)
696 goto out_free_ddev_targ;
697 }
698
699 if (logdev && logdev != ddev) {
700 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
701 mp->m_fsname);
702 if (!mp->m_logdev_targp)
703 goto out_free_rtdev_targ;
704 } else {
705 mp->m_logdev_targp = mp->m_ddev_targp;
706 }
707
708 return 0;
709
710 out_free_rtdev_targ:
711 if (mp->m_rtdev_targp)
712 xfs_free_buftarg(mp, mp->m_rtdev_targp);
713 out_free_ddev_targ:
714 xfs_free_buftarg(mp, mp->m_ddev_targp);
715 out_close_rtdev:
716 if (rtdev)
717 xfs_blkdev_put(rtdev);
718 out_close_logdev:
719 if (logdev && logdev != ddev)
720 xfs_blkdev_put(logdev);
721 out:
722 return error;
723}
724
725/*
726 * Setup xfs_mount buffer target pointers based on superblock
727 */
728STATIC int
729xfs_setup_devices(
730 struct xfs_mount *mp)
731{
732 int error;
733
734 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
735 mp->m_sb.sb_sectsize);
736 if (error)
737 return error;
738
739 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
740 unsigned int log_sector_size = BBSIZE;
741
742 if (xfs_sb_version_hassector(&mp->m_sb))
743 log_sector_size = mp->m_sb.sb_logsectsize;
744 error = xfs_setsize_buftarg(mp->m_logdev_targp,
745 mp->m_sb.sb_blocksize,
746 log_sector_size);
747 if (error)
748 return error;
749 }
750 if (mp->m_rtdev_targp) {
751 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
752 mp->m_sb.sb_blocksize,
753 mp->m_sb.sb_sectsize);
754 if (error)
755 return error;
756 }
757
758 return 0;
759}
760
761STATIC int
762xfs_init_mount_workqueues(
763 struct xfs_mount *mp)
764{
765 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
766 WQ_MEM_RECLAIM, 0, mp->m_fsname);
767 if (!mp->m_data_workqueue)
768 goto out;
769
770 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
771 WQ_MEM_RECLAIM, 0, mp->m_fsname);
772 if (!mp->m_unwritten_workqueue)
773 goto out_destroy_data_iodone_queue;
774
775 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
776 WQ_MEM_RECLAIM, 0, mp->m_fsname);
777 if (!mp->m_cil_workqueue)
778 goto out_destroy_unwritten;
779 return 0;
780
781out_destroy_unwritten:
782 destroy_workqueue(mp->m_unwritten_workqueue);
783out_destroy_data_iodone_queue:
784 destroy_workqueue(mp->m_data_workqueue);
785out:
786 return -ENOMEM;
787}
788
789STATIC void
790xfs_destroy_mount_workqueues(
791 struct xfs_mount *mp)
792{
793 destroy_workqueue(mp->m_cil_workqueue);
794 destroy_workqueue(mp->m_data_workqueue);
795 destroy_workqueue(mp->m_unwritten_workqueue);
796}
797
798/* Catch misguided souls that try to use this interface on XFS */
799STATIC struct inode *
800xfs_fs_alloc_inode(
801 struct super_block *sb)
802{
803 BUG();
804 return NULL;
805}
806
807/*
808 * Now that the generic code is guaranteed not to be accessing
809 * the linux inode, we can reclaim the inode.
810 */
811STATIC void
812xfs_fs_destroy_inode(
813 struct inode *inode)
814{
815 struct xfs_inode *ip = XFS_I(inode);
816
817 trace_xfs_destroy_inode(ip);
818
819 XFS_STATS_INC(vn_reclaim);
820
821 /* bad inode, get out here ASAP */
822 if (is_bad_inode(inode))
823 goto out_reclaim;
824
825 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
826
827 /*
828 * We should never get here with one of the reclaim flags already set.
829 */
830 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
831 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
832
833 /*
834 * We always use background reclaim here because even if the
835 * inode is clean, it still may be under IO and hence we have
836 * to take the flush lock. The background reclaim path handles
837 * this more efficiently than we can here, so simply let background
838 * reclaim tear down all inodes.
839 */
840out_reclaim:
841 xfs_inode_set_reclaim_tag(ip);
842}
843
844/*
845 * Slab object creation initialisation for the XFS inode.
846 * This covers only the idempotent fields in the XFS inode;
847 * all other fields need to be initialised on allocation
848 * from the slab. This avoids the need to repeatedly initialise
849 * fields in the xfs inode that left in the initialise state
850 * when freeing the inode.
851 */
852STATIC void
853xfs_fs_inode_init_once(
854 void *inode)
855{
856 struct xfs_inode *ip = inode;
857
858 memset(ip, 0, sizeof(struct xfs_inode));
859
860 /* vfs inode */
861 inode_init_once(VFS_I(ip));
862
863 /* xfs inode */
864 atomic_set(&ip->i_pincount, 0);
865 spin_lock_init(&ip->i_flags_lock);
866
867 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
868 "xfsino", ip->i_ino);
869}
870
871/*
872 * This is called by the VFS when dirtying inode metadata. This can happen
873 * for a few reasons, but we only care about timestamp updates, given that
874 * we handled the rest ourselves. In theory no other calls should happen,
875 * but for example generic_write_end() keeps dirtying the inode after
876 * updating i_size. Thus we check that the flags are exactly I_DIRTY_SYNC,
877 * and skip this call otherwise.
878 *
879 * We'll hopefull get a different method just for updating timestamps soon,
880 * at which point this hack can go away, and maybe we'll also get real
881 * error handling here.
882 */
883STATIC void
884xfs_fs_dirty_inode(
885 struct inode *inode,
886 int flags)
887{
888 struct xfs_inode *ip = XFS_I(inode);
889 struct xfs_mount *mp = ip->i_mount;
890 struct xfs_trans *tp;
891 int error;
892
893 if (flags != I_DIRTY_SYNC)
894 return;
895
896 trace_xfs_dirty_inode(ip);
897
898 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
899 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
900 if (error) {
901 xfs_trans_cancel(tp, 0);
902 goto trouble;
903 }
904 xfs_ilock(ip, XFS_ILOCK_EXCL);
905 /*
906 * Grab all the latest timestamps from the Linux inode.
907 */
908 ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec;
909 ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec;
910 ip->i_d.di_ctime.t_sec = (__int32_t)inode->i_ctime.tv_sec;
911 ip->i_d.di_ctime.t_nsec = (__int32_t)inode->i_ctime.tv_nsec;
912 ip->i_d.di_mtime.t_sec = (__int32_t)inode->i_mtime.tv_sec;
913 ip->i_d.di_mtime.t_nsec = (__int32_t)inode->i_mtime.tv_nsec;
914
915 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
916 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
917 error = xfs_trans_commit(tp, 0);
918 if (error)
919 goto trouble;
920 return;
921
922trouble:
923 xfs_warn(mp, "failed to update timestamps for inode 0x%llx", ip->i_ino);
924}
925
926STATIC void
927xfs_fs_evict_inode(
928 struct inode *inode)
929{
930 xfs_inode_t *ip = XFS_I(inode);
931
932 trace_xfs_evict_inode(ip);
933
934 truncate_inode_pages(&inode->i_data, 0);
935 clear_inode(inode);
936 XFS_STATS_INC(vn_rele);
937 XFS_STATS_INC(vn_remove);
938 XFS_STATS_DEC(vn_active);
939
940 /*
941 * The iolock is used by the file system to coordinate reads,
942 * writes, and block truncates. Up to this point the lock
943 * protected concurrent accesses by users of the inode. But
944 * from here forward we're doing some final processing of the
945 * inode because we're done with it, and although we reuse the
946 * iolock for protection it is really a distinct lock class
947 * (in the lockdep sense) from before. To keep lockdep happy
948 * (and basically indicate what we are doing), we explicitly
949 * re-init the iolock here.
950 */
951 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
952 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
953 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
954 &xfs_iolock_reclaimable, "xfs_iolock_reclaimable");
955
956 xfs_inactive(ip);
957}
958
959/*
960 * We do an unlocked check for XFS_IDONTCACHE here because we are already
961 * serialised against cache hits here via the inode->i_lock and igrab() in
962 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
963 * racing with us, and it avoids needing to grab a spinlock here for every inode
964 * we drop the final reference on.
965 */
966STATIC int
967xfs_fs_drop_inode(
968 struct inode *inode)
969{
970 struct xfs_inode *ip = XFS_I(inode);
971
972 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
973}
974
975STATIC void
976xfs_free_fsname(
977 struct xfs_mount *mp)
978{
979 kfree(mp->m_fsname);
980 kfree(mp->m_rtname);
981 kfree(mp->m_logname);
982}
983
984STATIC void
985xfs_fs_put_super(
986 struct super_block *sb)
987{
988 struct xfs_mount *mp = XFS_M(sb);
989
990 xfs_filestream_unmount(mp);
991 xfs_unmountfs(mp);
992 xfs_syncd_stop(mp);
993 xfs_freesb(mp);
994 xfs_icsb_destroy_counters(mp);
995 xfs_destroy_mount_workqueues(mp);
996 xfs_close_devices(mp);
997 xfs_free_fsname(mp);
998 kfree(mp);
999}
1000
1001STATIC int
1002xfs_fs_sync_fs(
1003 struct super_block *sb,
1004 int wait)
1005{
1006 struct xfs_mount *mp = XFS_M(sb);
1007 int error;
1008
1009 /*
1010 * Doing anything during the async pass would be counterproductive.
1011 */
1012 if (!wait)
1013 return 0;
1014
1015 error = xfs_quiesce_data(mp);
1016 if (error)
1017 return -error;
1018
1019 if (laptop_mode) {
1020 /*
1021 * The disk must be active because we're syncing.
1022 * We schedule xfssyncd now (now that the disk is
1023 * active) instead of later (when it might not be).
1024 */
1025 flush_delayed_work_sync(&mp->m_sync_work);
1026 }
1027
1028 return 0;
1029}
1030
1031STATIC int
1032xfs_fs_statfs(
1033 struct dentry *dentry,
1034 struct kstatfs *statp)
1035{
1036 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1037 xfs_sb_t *sbp = &mp->m_sb;
1038 struct xfs_inode *ip = XFS_I(dentry->d_inode);
1039 __uint64_t fakeinos, id;
1040 xfs_extlen_t lsize;
1041 __int64_t ffree;
1042
1043 statp->f_type = XFS_SB_MAGIC;
1044 statp->f_namelen = MAXNAMELEN - 1;
1045
1046 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1047 statp->f_fsid.val[0] = (u32)id;
1048 statp->f_fsid.val[1] = (u32)(id >> 32);
1049
1050 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1051
1052 spin_lock(&mp->m_sb_lock);
1053 statp->f_bsize = sbp->sb_blocksize;
1054 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1055 statp->f_blocks = sbp->sb_dblocks - lsize;
1056 statp->f_bfree = statp->f_bavail =
1057 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1058 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1059 statp->f_files =
1060 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1061 if (mp->m_maxicount)
1062 statp->f_files = min_t(typeof(statp->f_files),
1063 statp->f_files,
1064 mp->m_maxicount);
1065
1066 /* make sure statp->f_ffree does not underflow */
1067 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1068 statp->f_ffree = max_t(__int64_t, ffree, 0);
1069
1070 spin_unlock(&mp->m_sb_lock);
1071
1072 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1073 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) ==
1074 (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
1075 xfs_qm_statvfs(ip, statp);
1076 return 0;
1077}
1078
1079STATIC void
1080xfs_save_resvblks(struct xfs_mount *mp)
1081{
1082 __uint64_t resblks = 0;
1083
1084 mp->m_resblks_save = mp->m_resblks;
1085 xfs_reserve_blocks(mp, &resblks, NULL);
1086}
1087
1088STATIC void
1089xfs_restore_resvblks(struct xfs_mount *mp)
1090{
1091 __uint64_t resblks;
1092
1093 if (mp->m_resblks_save) {
1094 resblks = mp->m_resblks_save;
1095 mp->m_resblks_save = 0;
1096 } else
1097 resblks = xfs_default_resblks(mp);
1098
1099 xfs_reserve_blocks(mp, &resblks, NULL);
1100}
1101
1102STATIC int
1103xfs_fs_remount(
1104 struct super_block *sb,
1105 int *flags,
1106 char *options)
1107{
1108 struct xfs_mount *mp = XFS_M(sb);
1109 substring_t args[MAX_OPT_ARGS];
1110 char *p;
1111 int error;
1112
1113 while ((p = strsep(&options, ",")) != NULL) {
1114 int token;
1115
1116 if (!*p)
1117 continue;
1118
1119 token = match_token(p, tokens, args);
1120 switch (token) {
1121 case Opt_barrier:
1122 mp->m_flags |= XFS_MOUNT_BARRIER;
1123 break;
1124 case Opt_nobarrier:
1125 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1126 break;
1127 default:
1128 /*
1129 * Logically we would return an error here to prevent
1130 * users from believing they might have changed
1131 * mount options using remount which can't be changed.
1132 *
1133 * But unfortunately mount(8) adds all options from
1134 * mtab and fstab to the mount arguments in some cases
1135 * so we can't blindly reject options, but have to
1136 * check for each specified option if it actually
1137 * differs from the currently set option and only
1138 * reject it if that's the case.
1139 *
1140 * Until that is implemented we return success for
1141 * every remount request, and silently ignore all
1142 * options that we can't actually change.
1143 */
1144#if 0
1145 xfs_info(mp,
1146 "mount option \"%s\" not supported for remount\n", p);
1147 return -EINVAL;
1148#else
1149 break;
1150#endif
1151 }
1152 }
1153
1154 /* ro -> rw */
1155 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1156 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1157
1158 /*
1159 * If this is the first remount to writeable state we
1160 * might have some superblock changes to update.
1161 */
1162 if (mp->m_update_flags) {
1163 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1164 if (error) {
1165 xfs_warn(mp, "failed to write sb changes");
1166 return error;
1167 }
1168 mp->m_update_flags = 0;
1169 }
1170
1171 /*
1172 * Fill out the reserve pool if it is empty. Use the stashed
1173 * value if it is non-zero, otherwise go with the default.
1174 */
1175 xfs_restore_resvblks(mp);
1176 }
1177
1178 /* rw -> ro */
1179 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1180 /*
1181 * After we have synced the data but before we sync the
1182 * metadata, we need to free up the reserve block pool so that
1183 * the used block count in the superblock on disk is correct at
1184 * the end of the remount. Stash the current reserve pool size
1185 * so that if we get remounted rw, we can return it to the same
1186 * size.
1187 */
1188
1189 xfs_quiesce_data(mp);
1190 xfs_save_resvblks(mp);
1191 xfs_quiesce_attr(mp);
1192 mp->m_flags |= XFS_MOUNT_RDONLY;
1193 }
1194
1195 return 0;
1196}
1197
1198/*
1199 * Second stage of a freeze. The data is already frozen so we only
1200 * need to take care of the metadata. Once that's done write a dummy
1201 * record to dirty the log in case of a crash while frozen.
1202 */
1203STATIC int
1204xfs_fs_freeze(
1205 struct super_block *sb)
1206{
1207 struct xfs_mount *mp = XFS_M(sb);
1208
1209 xfs_save_resvblks(mp);
1210 xfs_quiesce_attr(mp);
1211 return -xfs_fs_log_dummy(mp);
1212}
1213
1214STATIC int
1215xfs_fs_unfreeze(
1216 struct super_block *sb)
1217{
1218 struct xfs_mount *mp = XFS_M(sb);
1219
1220 xfs_restore_resvblks(mp);
1221 return 0;
1222}
1223
1224STATIC int
1225xfs_fs_show_options(
1226 struct seq_file *m,
1227 struct dentry *root)
1228{
1229 return -xfs_showargs(XFS_M(root->d_sb), m);
1230}
1231
1232/*
1233 * This function fills in xfs_mount_t fields based on mount args.
1234 * Note: the superblock _has_ now been read in.
1235 */
1236STATIC int
1237xfs_finish_flags(
1238 struct xfs_mount *mp)
1239{
1240 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1241
1242 /* Fail a mount where the logbuf is smaller than the log stripe */
1243 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1244 if (mp->m_logbsize <= 0 &&
1245 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1246 mp->m_logbsize = mp->m_sb.sb_logsunit;
1247 } else if (mp->m_logbsize > 0 &&
1248 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1249 xfs_warn(mp,
1250 "logbuf size must be greater than or equal to log stripe size");
1251 return XFS_ERROR(EINVAL);
1252 }
1253 } else {
1254 /* Fail a mount if the logbuf is larger than 32K */
1255 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1256 xfs_warn(mp,
1257 "logbuf size for version 1 logs must be 16K or 32K");
1258 return XFS_ERROR(EINVAL);
1259 }
1260 }
1261
1262 /*
1263 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1264 * told by noattr2 to turn it off
1265 */
1266 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1267 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1268 mp->m_flags |= XFS_MOUNT_ATTR2;
1269
1270 /*
1271 * prohibit r/w mounts of read-only filesystems
1272 */
1273 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1274 xfs_warn(mp,
1275 "cannot mount a read-only filesystem as read-write");
1276 return XFS_ERROR(EROFS);
1277 }
1278
1279 return 0;
1280}
1281
1282STATIC int
1283xfs_fs_fill_super(
1284 struct super_block *sb,
1285 void *data,
1286 int silent)
1287{
1288 struct inode *root;
1289 struct xfs_mount *mp = NULL;
1290 int flags = 0, error = ENOMEM;
1291
1292 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1293 if (!mp)
1294 goto out;
1295
1296 spin_lock_init(&mp->m_sb_lock);
1297 mutex_init(&mp->m_growlock);
1298 atomic_set(&mp->m_active_trans, 0);
1299
1300 mp->m_super = sb;
1301 sb->s_fs_info = mp;
1302
1303 error = xfs_parseargs(mp, (char *)data);
1304 if (error)
1305 goto out_free_fsname;
1306
1307 sb_min_blocksize(sb, BBSIZE);
1308 sb->s_xattr = xfs_xattr_handlers;
1309 sb->s_export_op = &xfs_export_operations;
1310#ifdef CONFIG_XFS_QUOTA
1311 sb->s_qcop = &xfs_quotactl_operations;
1312#endif
1313 sb->s_op = &xfs_super_operations;
1314
1315 if (silent)
1316 flags |= XFS_MFSI_QUIET;
1317
1318 error = xfs_open_devices(mp);
1319 if (error)
1320 goto out_free_fsname;
1321
1322 error = xfs_init_mount_workqueues(mp);
1323 if (error)
1324 goto out_close_devices;
1325
1326 error = xfs_icsb_init_counters(mp);
1327 if (error)
1328 goto out_destroy_workqueues;
1329
1330 error = xfs_readsb(mp, flags);
1331 if (error)
1332 goto out_destroy_counters;
1333
1334 error = xfs_finish_flags(mp);
1335 if (error)
1336 goto out_free_sb;
1337
1338 error = xfs_setup_devices(mp);
1339 if (error)
1340 goto out_free_sb;
1341
1342 error = xfs_filestream_mount(mp);
1343 if (error)
1344 goto out_free_sb;
1345
1346 /*
1347 * we must configure the block size in the superblock before we run the
1348 * full mount process as the mount process can lookup and cache inodes.
1349 * For the same reason we must also initialise the syncd and register
1350 * the inode cache shrinker so that inodes can be reclaimed during
1351 * operations like a quotacheck that iterate all inodes in the
1352 * filesystem.
1353 */
1354 sb->s_magic = XFS_SB_MAGIC;
1355 sb->s_blocksize = mp->m_sb.sb_blocksize;
1356 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1357 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1358 sb->s_max_links = XFS_MAXLINK;
1359 sb->s_time_gran = 1;
1360 set_posix_acl_flag(sb);
1361
1362 error = xfs_syncd_init(mp);
1363 if (error)
1364 goto out_filestream_unmount;
1365
1366 error = xfs_mountfs(mp);
1367 if (error)
1368 goto out_syncd_stop;
1369
1370 root = igrab(VFS_I(mp->m_rootip));
1371 if (!root) {
1372 error = ENOENT;
1373 goto out_unmount;
1374 }
1375 if (is_bad_inode(root)) {
1376 error = EINVAL;
1377 goto out_unmount;
1378 }
1379 sb->s_root = d_make_root(root);
1380 if (!sb->s_root) {
1381 error = ENOMEM;
1382 goto out_unmount;
1383 }
1384
1385 return 0;
1386 out_syncd_stop:
1387 xfs_syncd_stop(mp);
1388 out_filestream_unmount:
1389 xfs_filestream_unmount(mp);
1390 out_free_sb:
1391 xfs_freesb(mp);
1392 out_destroy_counters:
1393 xfs_icsb_destroy_counters(mp);
1394out_destroy_workqueues:
1395 xfs_destroy_mount_workqueues(mp);
1396 out_close_devices:
1397 xfs_close_devices(mp);
1398 out_free_fsname:
1399 xfs_free_fsname(mp);
1400 kfree(mp);
1401 out:
1402 return -error;
1403
1404 out_unmount:
1405 xfs_filestream_unmount(mp);
1406 xfs_unmountfs(mp);
1407 xfs_syncd_stop(mp);
1408 goto out_free_sb;
1409}
1410
1411STATIC struct dentry *
1412xfs_fs_mount(
1413 struct file_system_type *fs_type,
1414 int flags,
1415 const char *dev_name,
1416 void *data)
1417{
1418 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1419}
1420
1421static int
1422xfs_fs_nr_cached_objects(
1423 struct super_block *sb)
1424{
1425 return xfs_reclaim_inodes_count(XFS_M(sb));
1426}
1427
1428static void
1429xfs_fs_free_cached_objects(
1430 struct super_block *sb,
1431 int nr_to_scan)
1432{
1433 xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1434}
1435
1436static const struct super_operations xfs_super_operations = {
1437 .alloc_inode = xfs_fs_alloc_inode,
1438 .destroy_inode = xfs_fs_destroy_inode,
1439 .dirty_inode = xfs_fs_dirty_inode,
1440 .evict_inode = xfs_fs_evict_inode,
1441 .drop_inode = xfs_fs_drop_inode,
1442 .put_super = xfs_fs_put_super,
1443 .sync_fs = xfs_fs_sync_fs,
1444 .freeze_fs = xfs_fs_freeze,
1445 .unfreeze_fs = xfs_fs_unfreeze,
1446 .statfs = xfs_fs_statfs,
1447 .remount_fs = xfs_fs_remount,
1448 .show_options = xfs_fs_show_options,
1449 .nr_cached_objects = xfs_fs_nr_cached_objects,
1450 .free_cached_objects = xfs_fs_free_cached_objects,
1451};
1452
1453static struct file_system_type xfs_fs_type = {
1454 .owner = THIS_MODULE,
1455 .name = "xfs",
1456 .mount = xfs_fs_mount,
1457 .kill_sb = kill_block_super,
1458 .fs_flags = FS_REQUIRES_DEV,
1459};
1460
1461STATIC int __init
1462xfs_init_zones(void)
1463{
1464
1465 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1466 if (!xfs_ioend_zone)
1467 goto out;
1468
1469 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1470 xfs_ioend_zone);
1471 if (!xfs_ioend_pool)
1472 goto out_destroy_ioend_zone;
1473
1474 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1475 "xfs_log_ticket");
1476 if (!xfs_log_ticket_zone)
1477 goto out_destroy_ioend_pool;
1478
1479 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1480 "xfs_bmap_free_item");
1481 if (!xfs_bmap_free_item_zone)
1482 goto out_destroy_log_ticket_zone;
1483
1484 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1485 "xfs_btree_cur");
1486 if (!xfs_btree_cur_zone)
1487 goto out_destroy_bmap_free_item_zone;
1488
1489 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1490 "xfs_da_state");
1491 if (!xfs_da_state_zone)
1492 goto out_destroy_btree_cur_zone;
1493
1494 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
1495 if (!xfs_dabuf_zone)
1496 goto out_destroy_da_state_zone;
1497
1498 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1499 if (!xfs_ifork_zone)
1500 goto out_destroy_dabuf_zone;
1501
1502 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1503 if (!xfs_trans_zone)
1504 goto out_destroy_ifork_zone;
1505
1506 xfs_log_item_desc_zone =
1507 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1508 "xfs_log_item_desc");
1509 if (!xfs_log_item_desc_zone)
1510 goto out_destroy_trans_zone;
1511
1512 /*
1513 * The size of the zone allocated buf log item is the maximum
1514 * size possible under XFS. This wastes a little bit of memory,
1515 * but it is much faster.
1516 */
1517 xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) +
1518 (((XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK) /
1519 NBWORD) * sizeof(int))), "xfs_buf_item");
1520 if (!xfs_buf_item_zone)
1521 goto out_destroy_log_item_desc_zone;
1522
1523 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1524 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1525 sizeof(xfs_extent_t))), "xfs_efd_item");
1526 if (!xfs_efd_zone)
1527 goto out_destroy_buf_item_zone;
1528
1529 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1530 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1531 sizeof(xfs_extent_t))), "xfs_efi_item");
1532 if (!xfs_efi_zone)
1533 goto out_destroy_efd_zone;
1534
1535 xfs_inode_zone =
1536 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1537 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1538 xfs_fs_inode_init_once);
1539 if (!xfs_inode_zone)
1540 goto out_destroy_efi_zone;
1541
1542 xfs_ili_zone =
1543 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1544 KM_ZONE_SPREAD, NULL);
1545 if (!xfs_ili_zone)
1546 goto out_destroy_inode_zone;
1547
1548 return 0;
1549
1550 out_destroy_inode_zone:
1551 kmem_zone_destroy(xfs_inode_zone);
1552 out_destroy_efi_zone:
1553 kmem_zone_destroy(xfs_efi_zone);
1554 out_destroy_efd_zone:
1555 kmem_zone_destroy(xfs_efd_zone);
1556 out_destroy_buf_item_zone:
1557 kmem_zone_destroy(xfs_buf_item_zone);
1558 out_destroy_log_item_desc_zone:
1559 kmem_zone_destroy(xfs_log_item_desc_zone);
1560 out_destroy_trans_zone:
1561 kmem_zone_destroy(xfs_trans_zone);
1562 out_destroy_ifork_zone:
1563 kmem_zone_destroy(xfs_ifork_zone);
1564 out_destroy_dabuf_zone:
1565 kmem_zone_destroy(xfs_dabuf_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);
1578 out:
1579 return -ENOMEM;
1580}
1581
1582STATIC void
1583xfs_destroy_zones(void)
1584{
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_dabuf_zone);
1594 kmem_zone_destroy(xfs_da_state_zone);
1595 kmem_zone_destroy(xfs_btree_cur_zone);
1596 kmem_zone_destroy(xfs_bmap_free_item_zone);
1597 kmem_zone_destroy(xfs_log_ticket_zone);
1598 mempool_destroy(xfs_ioend_pool);
1599 kmem_zone_destroy(xfs_ioend_zone);
1600
1601}
1602
1603STATIC int __init
1604xfs_init_workqueues(void)
1605{
1606 /*
1607 * We never want to the same work item to run twice, reclaiming inodes
1608 * or idling the log is not going to get any faster by multiple CPUs
1609 * competing for ressources. Use the default large max_active value
1610 * so that even lots of filesystems can perform these task in parallel.
1611 */
1612 xfs_syncd_wq = alloc_workqueue("xfssyncd", WQ_NON_REENTRANT, 0);
1613 if (!xfs_syncd_wq)
1614 return -ENOMEM;
1615
1616 /*
1617 * The allocation workqueue can be used in memory reclaim situations
1618 * (writepage path), and parallelism is only limited by the number of
1619 * AGs in all the filesystems mounted. Hence use the default large
1620 * max_active value for this workqueue.
1621 */
1622 xfs_alloc_wq = alloc_workqueue("xfsalloc", WQ_MEM_RECLAIM, 0);
1623 if (!xfs_alloc_wq)
1624 goto out_destroy_syncd;
1625
1626 return 0;
1627
1628out_destroy_syncd:
1629 destroy_workqueue(xfs_syncd_wq);
1630 return -ENOMEM;
1631}
1632
1633STATIC void
1634xfs_destroy_workqueues(void)
1635{
1636 destroy_workqueue(xfs_alloc_wq);
1637 destroy_workqueue(xfs_syncd_wq);
1638}
1639
1640STATIC int __init
1641init_xfs_fs(void)
1642{
1643 int error;
1644
1645 printk(KERN_INFO XFS_VERSION_STRING " with "
1646 XFS_BUILD_OPTIONS " enabled\n");
1647
1648 xfs_dir_startup();
1649
1650 error = xfs_init_zones();
1651 if (error)
1652 goto out;
1653
1654 error = xfs_init_workqueues();
1655 if (error)
1656 goto out_destroy_zones;
1657
1658 error = xfs_mru_cache_init();
1659 if (error)
1660 goto out_destroy_wq;
1661
1662 error = xfs_filestream_init();
1663 if (error)
1664 goto out_mru_cache_uninit;
1665
1666 error = xfs_buf_init();
1667 if (error)
1668 goto out_filestream_uninit;
1669
1670 error = xfs_init_procfs();
1671 if (error)
1672 goto out_buf_terminate;
1673
1674 error = xfs_sysctl_register();
1675 if (error)
1676 goto out_cleanup_procfs;
1677
1678 error = xfs_qm_init();
1679 if (error)
1680 goto out_sysctl_unregister;
1681
1682 error = register_filesystem(&xfs_fs_type);
1683 if (error)
1684 goto out_qm_exit;
1685 return 0;
1686
1687 out_qm_exit:
1688 xfs_qm_exit();
1689 out_sysctl_unregister:
1690 xfs_sysctl_unregister();
1691 out_cleanup_procfs:
1692 xfs_cleanup_procfs();
1693 out_buf_terminate:
1694 xfs_buf_terminate();
1695 out_filestream_uninit:
1696 xfs_filestream_uninit();
1697 out_mru_cache_uninit:
1698 xfs_mru_cache_uninit();
1699 out_destroy_wq:
1700 xfs_destroy_workqueues();
1701 out_destroy_zones:
1702 xfs_destroy_zones();
1703 out:
1704 return error;
1705}
1706
1707STATIC void __exit
1708exit_xfs_fs(void)
1709{
1710 xfs_qm_exit();
1711 unregister_filesystem(&xfs_fs_type);
1712 xfs_sysctl_unregister();
1713 xfs_cleanup_procfs();
1714 xfs_buf_terminate();
1715 xfs_filestream_uninit();
1716 xfs_mru_cache_uninit();
1717 xfs_destroy_workqueues();
1718 xfs_destroy_zones();
1719}
1720
1721module_init(init_xfs_fs);
1722module_exit(exit_xfs_fs);
1723
1724MODULE_AUTHOR("Silicon Graphics, Inc.");
1725MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1726MODULE_LICENSE("GPL");