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_bit.h"
21#include "xfs_log.h"
22#include "xfs_inum.h"
23#include "xfs_trans.h"
24#include "xfs_sb.h"
25#include "xfs_ag.h"
26#include "xfs_dir2.h"
27#include "xfs_alloc.h"
28#include "xfs_quota.h"
29#include "xfs_mount.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
33#include "xfs_dinode.h"
34#include "xfs_inode.h"
35#include "xfs_btree.h"
36#include "xfs_ialloc.h"
37#include "xfs_bmap.h"
38#include "xfs_rtalloc.h"
39#include "xfs_error.h"
40#include "xfs_itable.h"
41#include "xfs_fsops.h"
42#include "xfs_attr.h"
43#include "xfs_buf_item.h"
44#include "xfs_utils.h"
45#include "xfs_vnodeops.h"
46#include "xfs_log_priv.h"
47#include "xfs_trans_priv.h"
48#include "xfs_filestream.h"
49#include "xfs_da_btree.h"
50#include "xfs_extfree_item.h"
51#include "xfs_mru_cache.h"
52#include "xfs_inode_item.h"
53#include "xfs_sync.h"
54#include "xfs_trace.h"
55
56#include <linux/namei.h>
57#include <linux/init.h>
58#include <linux/slab.h>
59#include <linux/mount.h>
60#include <linux/mempool.h>
61#include <linux/writeback.h>
62#include <linux/kthread.h>
63#include <linux/freezer.h>
64#include <linux/parser.h>
65
66static const struct super_operations xfs_super_operations;
67static kmem_zone_t *xfs_ioend_zone;
68mempool_t *xfs_ioend_pool;
69
70#define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
71#define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
72#define MNTOPT_LOGDEV "logdev" /* log device */
73#define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
74#define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
75#define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
76#define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
77#define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
78#define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
79#define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
80#define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
81#define MNTOPT_MTPT "mtpt" /* filesystem mount point */
82#define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
83#define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
84#define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
85#define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
86#define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
87#define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
88#define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
89 * unwritten extent conversion */
90#define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
91#define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
92#define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
93#define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
94#define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
95#define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
96 * in stat(). */
97#define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
98#define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
99#define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
100#define MNTOPT_QUOTA "quota" /* disk quotas (user) */
101#define MNTOPT_NOQUOTA "noquota" /* no quotas */
102#define MNTOPT_USRQUOTA "usrquota" /* user quota enabled */
103#define MNTOPT_GRPQUOTA "grpquota" /* group quota enabled */
104#define MNTOPT_PRJQUOTA "prjquota" /* project quota enabled */
105#define MNTOPT_UQUOTA "uquota" /* user quota (IRIX variant) */
106#define MNTOPT_GQUOTA "gquota" /* group quota (IRIX variant) */
107#define MNTOPT_PQUOTA "pquota" /* project quota (IRIX variant) */
108#define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
109#define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
110#define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
111#define MNTOPT_QUOTANOENF "qnoenforce" /* same as uqnoenforce */
112#define MNTOPT_DELAYLOG "delaylog" /* Delayed logging enabled */
113#define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed logging disabled */
114#define MNTOPT_DISCARD "discard" /* Discard unused blocks */
115#define MNTOPT_NODISCARD "nodiscard" /* Do not discard unused blocks */
116
117/*
118 * Table driven mount option parser.
119 *
120 * Currently only used for remount, but it will be used for mount
121 * in the future, too.
122 */
123enum {
124 Opt_barrier, Opt_nobarrier, Opt_err
125};
126
127static const match_table_t tokens = {
128 {Opt_barrier, "barrier"},
129 {Opt_nobarrier, "nobarrier"},
130 {Opt_err, NULL}
131};
132
133
134STATIC unsigned long
135suffix_strtoul(char *s, char **endp, unsigned int base)
136{
137 int last, shift_left_factor = 0;
138 char *value = s;
139
140 last = strlen(value) - 1;
141 if (value[last] == 'K' || value[last] == 'k') {
142 shift_left_factor = 10;
143 value[last] = '\0';
144 }
145 if (value[last] == 'M' || value[last] == 'm') {
146 shift_left_factor = 20;
147 value[last] = '\0';
148 }
149 if (value[last] == 'G' || value[last] == 'g') {
150 shift_left_factor = 30;
151 value[last] = '\0';
152 }
153
154 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
155}
156
157/*
158 * This function fills in xfs_mount_t fields based on mount args.
159 * Note: the superblock has _not_ yet been read in.
160 *
161 * Note that this function leaks the various device name allocations on
162 * failure. The caller takes care of them.
163 */
164STATIC int
165xfs_parseargs(
166 struct xfs_mount *mp,
167 char *options)
168{
169 struct super_block *sb = mp->m_super;
170 char *this_char, *value, *eov;
171 int dsunit = 0;
172 int dswidth = 0;
173 int iosize = 0;
174 __uint8_t iosizelog = 0;
175
176 /*
177 * set up the mount name first so all the errors will refer to the
178 * correct device.
179 */
180 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
181 if (!mp->m_fsname)
182 return ENOMEM;
183 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
184
185 /*
186 * Copy binary VFS mount flags we are interested in.
187 */
188 if (sb->s_flags & MS_RDONLY)
189 mp->m_flags |= XFS_MOUNT_RDONLY;
190 if (sb->s_flags & MS_DIRSYNC)
191 mp->m_flags |= XFS_MOUNT_DIRSYNC;
192 if (sb->s_flags & MS_SYNCHRONOUS)
193 mp->m_flags |= XFS_MOUNT_WSYNC;
194
195 /*
196 * Set some default flags that could be cleared by the mount option
197 * parsing.
198 */
199 mp->m_flags |= XFS_MOUNT_BARRIER;
200 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
201 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
202 mp->m_flags |= XFS_MOUNT_DELAYLOG;
203
204 /*
205 * These can be overridden by the mount option parsing.
206 */
207 mp->m_logbufs = -1;
208 mp->m_logbsize = -1;
209
210 if (!options)
211 goto done;
212
213 while ((this_char = strsep(&options, ",")) != NULL) {
214 if (!*this_char)
215 continue;
216 if ((value = strchr(this_char, '=')) != NULL)
217 *value++ = 0;
218
219 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
220 if (!value || !*value) {
221 xfs_warn(mp, "%s option requires an argument",
222 this_char);
223 return EINVAL;
224 }
225 mp->m_logbufs = simple_strtoul(value, &eov, 10);
226 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
227 if (!value || !*value) {
228 xfs_warn(mp, "%s option requires an argument",
229 this_char);
230 return EINVAL;
231 }
232 mp->m_logbsize = suffix_strtoul(value, &eov, 10);
233 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
234 if (!value || !*value) {
235 xfs_warn(mp, "%s option requires an argument",
236 this_char);
237 return EINVAL;
238 }
239 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
240 if (!mp->m_logname)
241 return ENOMEM;
242 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
243 xfs_warn(mp, "%s option not allowed on this system",
244 this_char);
245 return EINVAL;
246 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
247 if (!value || !*value) {
248 xfs_warn(mp, "%s option requires an argument",
249 this_char);
250 return EINVAL;
251 }
252 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
253 if (!mp->m_rtname)
254 return ENOMEM;
255 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
256 if (!value || !*value) {
257 xfs_warn(mp, "%s option requires an argument",
258 this_char);
259 return EINVAL;
260 }
261 iosize = simple_strtoul(value, &eov, 10);
262 iosizelog = ffs(iosize) - 1;
263 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
264 if (!value || !*value) {
265 xfs_warn(mp, "%s option requires an argument",
266 this_char);
267 return EINVAL;
268 }
269 iosize = suffix_strtoul(value, &eov, 10);
270 iosizelog = ffs(iosize) - 1;
271 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
272 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
273 mp->m_flags |= XFS_MOUNT_GRPID;
274 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
275 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
276 mp->m_flags &= ~XFS_MOUNT_GRPID;
277 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
278 mp->m_flags |= XFS_MOUNT_WSYNC;
279 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
280 mp->m_flags |= XFS_MOUNT_NORECOVERY;
281 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
282 mp->m_flags |= XFS_MOUNT_NOALIGN;
283 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
284 mp->m_flags |= XFS_MOUNT_SWALLOC;
285 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
286 if (!value || !*value) {
287 xfs_warn(mp, "%s option requires an argument",
288 this_char);
289 return EINVAL;
290 }
291 dsunit = simple_strtoul(value, &eov, 10);
292 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
293 if (!value || !*value) {
294 xfs_warn(mp, "%s option requires an argument",
295 this_char);
296 return EINVAL;
297 }
298 dswidth = simple_strtoul(value, &eov, 10);
299 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
300 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
301#if !XFS_BIG_INUMS
302 xfs_warn(mp, "%s option not allowed on this system",
303 this_char);
304 return EINVAL;
305#endif
306 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
307 mp->m_flags |= XFS_MOUNT_NOUUID;
308 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
309 mp->m_flags |= XFS_MOUNT_BARRIER;
310 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
311 mp->m_flags &= ~XFS_MOUNT_BARRIER;
312 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
313 mp->m_flags |= XFS_MOUNT_IKEEP;
314 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
315 mp->m_flags &= ~XFS_MOUNT_IKEEP;
316 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
317 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
318 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
319 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
320 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
321 mp->m_flags |= XFS_MOUNT_ATTR2;
322 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
323 mp->m_flags &= ~XFS_MOUNT_ATTR2;
324 mp->m_flags |= XFS_MOUNT_NOATTR2;
325 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
326 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
327 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
328 mp->m_qflags &= ~(XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
329 XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
330 XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
331 XFS_UQUOTA_ENFD | XFS_OQUOTA_ENFD);
332 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
333 !strcmp(this_char, MNTOPT_UQUOTA) ||
334 !strcmp(this_char, MNTOPT_USRQUOTA)) {
335 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
336 XFS_UQUOTA_ENFD);
337 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
338 !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
339 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
340 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
341 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
342 !strcmp(this_char, MNTOPT_PRJQUOTA)) {
343 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
344 XFS_OQUOTA_ENFD);
345 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
346 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
347 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
348 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
349 !strcmp(this_char, MNTOPT_GRPQUOTA)) {
350 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
351 XFS_OQUOTA_ENFD);
352 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
353 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
354 mp->m_qflags &= ~XFS_OQUOTA_ENFD;
355 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
356 mp->m_flags |= XFS_MOUNT_DELAYLOG;
357 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
358 mp->m_flags &= ~XFS_MOUNT_DELAYLOG;
359 xfs_warn(mp,
360 "nodelaylog is deprecated and will be removed in Linux 3.3");
361 } else if (!strcmp(this_char, MNTOPT_DISCARD)) {
362 mp->m_flags |= XFS_MOUNT_DISCARD;
363 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
364 mp->m_flags &= ~XFS_MOUNT_DISCARD;
365 } else if (!strcmp(this_char, "ihashsize")) {
366 xfs_warn(mp,
367 "ihashsize no longer used, option is deprecated.");
368 } else if (!strcmp(this_char, "osyncisdsync")) {
369 xfs_warn(mp,
370 "osyncisdsync has no effect, option is deprecated.");
371 } else if (!strcmp(this_char, "osyncisosync")) {
372 xfs_warn(mp,
373 "osyncisosync has no effect, option is deprecated.");
374 } else if (!strcmp(this_char, "irixsgid")) {
375 xfs_warn(mp,
376 "irixsgid is now a sysctl(2) variable, option is deprecated.");
377 } else {
378 xfs_warn(mp, "unknown mount option [%s].", this_char);
379 return EINVAL;
380 }
381 }
382
383 /*
384 * no recovery flag requires a read-only mount
385 */
386 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
387 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
388 xfs_warn(mp, "no-recovery mounts must be read-only.");
389 return EINVAL;
390 }
391
392 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
393 xfs_warn(mp,
394 "sunit and swidth options incompatible with the noalign option");
395 return EINVAL;
396 }
397
398 if ((mp->m_flags & XFS_MOUNT_DISCARD) &&
399 !(mp->m_flags & XFS_MOUNT_DELAYLOG)) {
400 xfs_warn(mp,
401 "the discard option is incompatible with the nodelaylog option");
402 return EINVAL;
403 }
404
405#ifndef CONFIG_XFS_QUOTA
406 if (XFS_IS_QUOTA_RUNNING(mp)) {
407 xfs_warn(mp, "quota support not available in this kernel.");
408 return EINVAL;
409 }
410#endif
411
412 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
413 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
414 xfs_warn(mp, "cannot mount with both project and group quota");
415 return EINVAL;
416 }
417
418 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
419 xfs_warn(mp, "sunit and swidth must be specified together");
420 return EINVAL;
421 }
422
423 if (dsunit && (dswidth % dsunit != 0)) {
424 xfs_warn(mp,
425 "stripe width (%d) must be a multiple of the stripe unit (%d)",
426 dswidth, dsunit);
427 return EINVAL;
428 }
429
430done:
431 if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
432 /*
433 * At this point the superblock has not been read
434 * in, therefore we do not know the block size.
435 * Before the mount call ends we will convert
436 * these to FSBs.
437 */
438 if (dsunit) {
439 mp->m_dalign = dsunit;
440 mp->m_flags |= XFS_MOUNT_RETERR;
441 }
442
443 if (dswidth)
444 mp->m_swidth = dswidth;
445 }
446
447 if (mp->m_logbufs != -1 &&
448 mp->m_logbufs != 0 &&
449 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
450 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
451 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
452 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
453 return XFS_ERROR(EINVAL);
454 }
455 if (mp->m_logbsize != -1 &&
456 mp->m_logbsize != 0 &&
457 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
458 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
459 !is_power_of_2(mp->m_logbsize))) {
460 xfs_warn(mp,
461 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
462 mp->m_logbsize);
463 return XFS_ERROR(EINVAL);
464 }
465
466 if (iosizelog) {
467 if (iosizelog > XFS_MAX_IO_LOG ||
468 iosizelog < XFS_MIN_IO_LOG) {
469 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
470 iosizelog, XFS_MIN_IO_LOG,
471 XFS_MAX_IO_LOG);
472 return XFS_ERROR(EINVAL);
473 }
474
475 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
476 mp->m_readio_log = iosizelog;
477 mp->m_writeio_log = iosizelog;
478 }
479
480 return 0;
481}
482
483struct proc_xfs_info {
484 int flag;
485 char *str;
486};
487
488STATIC int
489xfs_showargs(
490 struct xfs_mount *mp,
491 struct seq_file *m)
492{
493 static struct proc_xfs_info xfs_info_set[] = {
494 /* the few simple ones we can get from the mount struct */
495 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP },
496 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
497 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
498 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
499 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
500 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
501 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 },
502 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
503 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
504 { XFS_MOUNT_DELAYLOG, "," MNTOPT_DELAYLOG },
505 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD },
506 { 0, NULL }
507 };
508 static struct proc_xfs_info xfs_info_unset[] = {
509 /* the few simple ones we can get from the mount struct */
510 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO },
511 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER },
512 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE },
513 { 0, NULL }
514 };
515 struct proc_xfs_info *xfs_infop;
516
517 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
518 if (mp->m_flags & xfs_infop->flag)
519 seq_puts(m, xfs_infop->str);
520 }
521 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
522 if (!(mp->m_flags & xfs_infop->flag))
523 seq_puts(m, xfs_infop->str);
524 }
525
526 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
527 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
528 (int)(1 << mp->m_writeio_log) >> 10);
529
530 if (mp->m_logbufs > 0)
531 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
532 if (mp->m_logbsize > 0)
533 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
534
535 if (mp->m_logname)
536 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
537 if (mp->m_rtname)
538 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
539
540 if (mp->m_dalign > 0)
541 seq_printf(m, "," MNTOPT_SUNIT "=%d",
542 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
543 if (mp->m_swidth > 0)
544 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
545 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
546
547 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
548 seq_puts(m, "," MNTOPT_USRQUOTA);
549 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
550 seq_puts(m, "," MNTOPT_UQUOTANOENF);
551
552 /* Either project or group quotas can be active, not both */
553
554 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
555 if (mp->m_qflags & XFS_OQUOTA_ENFD)
556 seq_puts(m, "," MNTOPT_PRJQUOTA);
557 else
558 seq_puts(m, "," MNTOPT_PQUOTANOENF);
559 } else if (mp->m_qflags & XFS_GQUOTA_ACCT) {
560 if (mp->m_qflags & XFS_OQUOTA_ENFD)
561 seq_puts(m, "," MNTOPT_GRPQUOTA);
562 else
563 seq_puts(m, "," MNTOPT_GQUOTANOENF);
564 }
565
566 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
567 seq_puts(m, "," MNTOPT_NOQUOTA);
568
569 return 0;
570}
571__uint64_t
572xfs_max_file_offset(
573 unsigned int blockshift)
574{
575 unsigned int pagefactor = 1;
576 unsigned int bitshift = BITS_PER_LONG - 1;
577
578 /* Figure out maximum filesize, on Linux this can depend on
579 * the filesystem blocksize (on 32 bit platforms).
580 * __block_write_begin does this in an [unsigned] long...
581 * page->index << (PAGE_CACHE_SHIFT - bbits)
582 * So, for page sized blocks (4K on 32 bit platforms),
583 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
584 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
585 * but for smaller blocksizes it is less (bbits = log2 bsize).
586 * Note1: get_block_t takes a long (implicit cast from above)
587 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
588 * can optionally convert the [unsigned] long from above into
589 * an [unsigned] long long.
590 */
591
592#if BITS_PER_LONG == 32
593# if defined(CONFIG_LBDAF)
594 ASSERT(sizeof(sector_t) == 8);
595 pagefactor = PAGE_CACHE_SIZE;
596 bitshift = BITS_PER_LONG;
597# else
598 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
599# endif
600#endif
601
602 return (((__uint64_t)pagefactor) << bitshift) - 1;
603}
604
605STATIC int
606xfs_blkdev_get(
607 xfs_mount_t *mp,
608 const char *name,
609 struct block_device **bdevp)
610{
611 int error = 0;
612
613 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
614 mp);
615 if (IS_ERR(*bdevp)) {
616 error = PTR_ERR(*bdevp);
617 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
618 }
619
620 return -error;
621}
622
623STATIC void
624xfs_blkdev_put(
625 struct block_device *bdev)
626{
627 if (bdev)
628 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
629}
630
631void
632xfs_blkdev_issue_flush(
633 xfs_buftarg_t *buftarg)
634{
635 blkdev_issue_flush(buftarg->bt_bdev, GFP_KERNEL, NULL);
636}
637
638STATIC void
639xfs_close_devices(
640 struct xfs_mount *mp)
641{
642 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
643 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
644 xfs_free_buftarg(mp, mp->m_logdev_targp);
645 xfs_blkdev_put(logdev);
646 }
647 if (mp->m_rtdev_targp) {
648 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
649 xfs_free_buftarg(mp, mp->m_rtdev_targp);
650 xfs_blkdev_put(rtdev);
651 }
652 xfs_free_buftarg(mp, mp->m_ddev_targp);
653}
654
655/*
656 * The file system configurations are:
657 * (1) device (partition) with data and internal log
658 * (2) logical volume with data and log subvolumes.
659 * (3) logical volume with data, log, and realtime subvolumes.
660 *
661 * We only have to handle opening the log and realtime volumes here if
662 * they are present. The data subvolume has already been opened by
663 * get_sb_bdev() and is stored in sb->s_bdev.
664 */
665STATIC int
666xfs_open_devices(
667 struct xfs_mount *mp)
668{
669 struct block_device *ddev = mp->m_super->s_bdev;
670 struct block_device *logdev = NULL, *rtdev = NULL;
671 int error;
672
673 /*
674 * Open real time and log devices - order is important.
675 */
676 if (mp->m_logname) {
677 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
678 if (error)
679 goto out;
680 }
681
682 if (mp->m_rtname) {
683 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
684 if (error)
685 goto out_close_logdev;
686
687 if (rtdev == ddev || rtdev == logdev) {
688 xfs_warn(mp,
689 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
690 error = EINVAL;
691 goto out_close_rtdev;
692 }
693 }
694
695 /*
696 * Setup xfs_mount buffer target pointers
697 */
698 error = ENOMEM;
699 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
700 if (!mp->m_ddev_targp)
701 goto out_close_rtdev;
702
703 if (rtdev) {
704 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
705 mp->m_fsname);
706 if (!mp->m_rtdev_targp)
707 goto out_free_ddev_targ;
708 }
709
710 if (logdev && logdev != ddev) {
711 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
712 mp->m_fsname);
713 if (!mp->m_logdev_targp)
714 goto out_free_rtdev_targ;
715 } else {
716 mp->m_logdev_targp = mp->m_ddev_targp;
717 }
718
719 return 0;
720
721 out_free_rtdev_targ:
722 if (mp->m_rtdev_targp)
723 xfs_free_buftarg(mp, mp->m_rtdev_targp);
724 out_free_ddev_targ:
725 xfs_free_buftarg(mp, mp->m_ddev_targp);
726 out_close_rtdev:
727 if (rtdev)
728 xfs_blkdev_put(rtdev);
729 out_close_logdev:
730 if (logdev && logdev != ddev)
731 xfs_blkdev_put(logdev);
732 out:
733 return error;
734}
735
736/*
737 * Setup xfs_mount buffer target pointers based on superblock
738 */
739STATIC int
740xfs_setup_devices(
741 struct xfs_mount *mp)
742{
743 int error;
744
745 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
746 mp->m_sb.sb_sectsize);
747 if (error)
748 return error;
749
750 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
751 unsigned int log_sector_size = BBSIZE;
752
753 if (xfs_sb_version_hassector(&mp->m_sb))
754 log_sector_size = mp->m_sb.sb_logsectsize;
755 error = xfs_setsize_buftarg(mp->m_logdev_targp,
756 mp->m_sb.sb_blocksize,
757 log_sector_size);
758 if (error)
759 return error;
760 }
761 if (mp->m_rtdev_targp) {
762 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
763 mp->m_sb.sb_blocksize,
764 mp->m_sb.sb_sectsize);
765 if (error)
766 return error;
767 }
768
769 return 0;
770}
771
772/* Catch misguided souls that try to use this interface on XFS */
773STATIC struct inode *
774xfs_fs_alloc_inode(
775 struct super_block *sb)
776{
777 BUG();
778 return NULL;
779}
780
781/*
782 * Now that the generic code is guaranteed not to be accessing
783 * the linux inode, we can reclaim the inode.
784 */
785STATIC void
786xfs_fs_destroy_inode(
787 struct inode *inode)
788{
789 struct xfs_inode *ip = XFS_I(inode);
790
791 trace_xfs_destroy_inode(ip);
792
793 XFS_STATS_INC(vn_reclaim);
794
795 /* bad inode, get out here ASAP */
796 if (is_bad_inode(inode))
797 goto out_reclaim;
798
799 xfs_ioend_wait(ip);
800
801 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
802
803 /*
804 * We should never get here with one of the reclaim flags already set.
805 */
806 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
807 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
808
809 /*
810 * We always use background reclaim here because even if the
811 * inode is clean, it still may be under IO and hence we have
812 * to take the flush lock. The background reclaim path handles
813 * this more efficiently than we can here, so simply let background
814 * reclaim tear down all inodes.
815 */
816out_reclaim:
817 xfs_inode_set_reclaim_tag(ip);
818}
819
820/*
821 * Slab object creation initialisation for the XFS inode.
822 * This covers only the idempotent fields in the XFS inode;
823 * all other fields need to be initialised on allocation
824 * from the slab. This avoids the need to repeatedly initialise
825 * fields in the xfs inode that left in the initialise state
826 * when freeing the inode.
827 */
828STATIC void
829xfs_fs_inode_init_once(
830 void *inode)
831{
832 struct xfs_inode *ip = inode;
833
834 memset(ip, 0, sizeof(struct xfs_inode));
835
836 /* vfs inode */
837 inode_init_once(VFS_I(ip));
838
839 /* xfs inode */
840 atomic_set(&ip->i_iocount, 0);
841 atomic_set(&ip->i_pincount, 0);
842 spin_lock_init(&ip->i_flags_lock);
843 init_waitqueue_head(&ip->i_ipin_wait);
844 /*
845 * Because we want to use a counting completion, complete
846 * the flush completion once to allow a single access to
847 * the flush completion without blocking.
848 */
849 init_completion(&ip->i_flush);
850 complete(&ip->i_flush);
851
852 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
853 "xfsino", ip->i_ino);
854}
855
856/*
857 * Dirty the XFS inode when mark_inode_dirty_sync() is called so that
858 * we catch unlogged VFS level updates to the inode.
859 *
860 * We need the barrier() to maintain correct ordering between unlogged
861 * updates and the transaction commit code that clears the i_update_core
862 * field. This requires all updates to be completed before marking the
863 * inode dirty.
864 */
865STATIC void
866xfs_fs_dirty_inode(
867 struct inode *inode,
868 int flags)
869{
870 barrier();
871 XFS_I(inode)->i_update_core = 1;
872}
873
874STATIC int
875xfs_log_inode(
876 struct xfs_inode *ip)
877{
878 struct xfs_mount *mp = ip->i_mount;
879 struct xfs_trans *tp;
880 int error;
881
882 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
883 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
884 if (error) {
885 xfs_trans_cancel(tp, 0);
886 return error;
887 }
888
889 xfs_ilock(ip, XFS_ILOCK_EXCL);
890 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
891 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
892 return xfs_trans_commit(tp, 0);
893}
894
895STATIC int
896xfs_fs_write_inode(
897 struct inode *inode,
898 struct writeback_control *wbc)
899{
900 struct xfs_inode *ip = XFS_I(inode);
901 struct xfs_mount *mp = ip->i_mount;
902 int error = EAGAIN;
903
904 trace_xfs_write_inode(ip);
905
906 if (XFS_FORCED_SHUTDOWN(mp))
907 return -XFS_ERROR(EIO);
908 if (!ip->i_update_core)
909 return 0;
910
911 if (wbc->sync_mode == WB_SYNC_ALL) {
912 /*
913 * Make sure the inode has made it it into the log. Instead
914 * of forcing it all the way to stable storage using a
915 * synchronous transaction we let the log force inside the
916 * ->sync_fs call do that for thus, which reduces the number
917 * of synchronous log foces dramatically.
918 */
919 xfs_ioend_wait(ip);
920 error = xfs_log_inode(ip);
921 if (error)
922 goto out;
923 return 0;
924 } else {
925 /*
926 * We make this non-blocking if the inode is contended, return
927 * EAGAIN to indicate to the caller that they did not succeed.
928 * This prevents the flush path from blocking on inodes inside
929 * another operation right now, they get caught later by
930 * xfs_sync.
931 */
932 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
933 goto out;
934
935 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip))
936 goto out_unlock;
937
938 /*
939 * Now we have the flush lock and the inode is not pinned, we
940 * can check if the inode is really clean as we know that
941 * there are no pending transaction completions, it is not
942 * waiting on the delayed write queue and there is no IO in
943 * progress.
944 */
945 if (xfs_inode_clean(ip)) {
946 xfs_ifunlock(ip);
947 error = 0;
948 goto out_unlock;
949 }
950 error = xfs_iflush(ip, SYNC_TRYLOCK);
951 }
952
953 out_unlock:
954 xfs_iunlock(ip, XFS_ILOCK_SHARED);
955 out:
956 /*
957 * if we failed to write out the inode then mark
958 * it dirty again so we'll try again later.
959 */
960 if (error)
961 xfs_mark_inode_dirty_sync(ip);
962 return -error;
963}
964
965STATIC void
966xfs_fs_evict_inode(
967 struct inode *inode)
968{
969 xfs_inode_t *ip = XFS_I(inode);
970
971 trace_xfs_evict_inode(ip);
972
973 truncate_inode_pages(&inode->i_data, 0);
974 end_writeback(inode);
975 XFS_STATS_INC(vn_rele);
976 XFS_STATS_INC(vn_remove);
977 XFS_STATS_DEC(vn_active);
978
979 /*
980 * The iolock is used by the file system to coordinate reads,
981 * writes, and block truncates. Up to this point the lock
982 * protected concurrent accesses by users of the inode. But
983 * from here forward we're doing some final processing of the
984 * inode because we're done with it, and although we reuse the
985 * iolock for protection it is really a distinct lock class
986 * (in the lockdep sense) from before. To keep lockdep happy
987 * (and basically indicate what we are doing), we explicitly
988 * re-init the iolock here.
989 */
990 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
991 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
992 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
993 &xfs_iolock_reclaimable, "xfs_iolock_reclaimable");
994
995 xfs_inactive(ip);
996}
997
998STATIC void
999xfs_free_fsname(
1000 struct xfs_mount *mp)
1001{
1002 kfree(mp->m_fsname);
1003 kfree(mp->m_rtname);
1004 kfree(mp->m_logname);
1005}
1006
1007STATIC void
1008xfs_fs_put_super(
1009 struct super_block *sb)
1010{
1011 struct xfs_mount *mp = XFS_M(sb);
1012
1013 xfs_syncd_stop(mp);
1014
1015 /*
1016 * Blow away any referenced inode in the filestreams cache.
1017 * This can and will cause log traffic as inodes go inactive
1018 * here.
1019 */
1020 xfs_filestream_unmount(mp);
1021
1022 XFS_bflush(mp->m_ddev_targp);
1023
1024 xfs_unmountfs(mp);
1025 xfs_freesb(mp);
1026 xfs_icsb_destroy_counters(mp);
1027 xfs_close_devices(mp);
1028 xfs_free_fsname(mp);
1029 kfree(mp);
1030}
1031
1032STATIC int
1033xfs_fs_sync_fs(
1034 struct super_block *sb,
1035 int wait)
1036{
1037 struct xfs_mount *mp = XFS_M(sb);
1038 int error;
1039
1040 /*
1041 * Not much we can do for the first async pass. Writing out the
1042 * superblock would be counter-productive as we are going to redirty
1043 * when writing out other data and metadata (and writing out a single
1044 * block is quite fast anyway).
1045 *
1046 * Try to asynchronously kick off quota syncing at least.
1047 */
1048 if (!wait) {
1049 xfs_qm_sync(mp, SYNC_TRYLOCK);
1050 return 0;
1051 }
1052
1053 error = xfs_quiesce_data(mp);
1054 if (error)
1055 return -error;
1056
1057 if (laptop_mode) {
1058 /*
1059 * The disk must be active because we're syncing.
1060 * We schedule xfssyncd now (now that the disk is
1061 * active) instead of later (when it might not be).
1062 */
1063 flush_delayed_work_sync(&mp->m_sync_work);
1064 }
1065
1066 return 0;
1067}
1068
1069STATIC int
1070xfs_fs_statfs(
1071 struct dentry *dentry,
1072 struct kstatfs *statp)
1073{
1074 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1075 xfs_sb_t *sbp = &mp->m_sb;
1076 struct xfs_inode *ip = XFS_I(dentry->d_inode);
1077 __uint64_t fakeinos, id;
1078 xfs_extlen_t lsize;
1079 __int64_t ffree;
1080
1081 statp->f_type = XFS_SB_MAGIC;
1082 statp->f_namelen = MAXNAMELEN - 1;
1083
1084 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1085 statp->f_fsid.val[0] = (u32)id;
1086 statp->f_fsid.val[1] = (u32)(id >> 32);
1087
1088 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1089
1090 spin_lock(&mp->m_sb_lock);
1091 statp->f_bsize = sbp->sb_blocksize;
1092 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1093 statp->f_blocks = sbp->sb_dblocks - lsize;
1094 statp->f_bfree = statp->f_bavail =
1095 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1096 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1097 statp->f_files =
1098 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1099 if (mp->m_maxicount)
1100 statp->f_files = min_t(typeof(statp->f_files),
1101 statp->f_files,
1102 mp->m_maxicount);
1103
1104 /* make sure statp->f_ffree does not underflow */
1105 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1106 statp->f_ffree = max_t(__int64_t, ffree, 0);
1107
1108 spin_unlock(&mp->m_sb_lock);
1109
1110 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) ||
1111 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) ==
1112 (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
1113 xfs_qm_statvfs(ip, statp);
1114 return 0;
1115}
1116
1117STATIC void
1118xfs_save_resvblks(struct xfs_mount *mp)
1119{
1120 __uint64_t resblks = 0;
1121
1122 mp->m_resblks_save = mp->m_resblks;
1123 xfs_reserve_blocks(mp, &resblks, NULL);
1124}
1125
1126STATIC void
1127xfs_restore_resvblks(struct xfs_mount *mp)
1128{
1129 __uint64_t resblks;
1130
1131 if (mp->m_resblks_save) {
1132 resblks = mp->m_resblks_save;
1133 mp->m_resblks_save = 0;
1134 } else
1135 resblks = xfs_default_resblks(mp);
1136
1137 xfs_reserve_blocks(mp, &resblks, NULL);
1138}
1139
1140STATIC int
1141xfs_fs_remount(
1142 struct super_block *sb,
1143 int *flags,
1144 char *options)
1145{
1146 struct xfs_mount *mp = XFS_M(sb);
1147 substring_t args[MAX_OPT_ARGS];
1148 char *p;
1149 int error;
1150
1151 while ((p = strsep(&options, ",")) != NULL) {
1152 int token;
1153
1154 if (!*p)
1155 continue;
1156
1157 token = match_token(p, tokens, args);
1158 switch (token) {
1159 case Opt_barrier:
1160 mp->m_flags |= XFS_MOUNT_BARRIER;
1161 break;
1162 case Opt_nobarrier:
1163 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1164 break;
1165 default:
1166 /*
1167 * Logically we would return an error here to prevent
1168 * users from believing they might have changed
1169 * mount options using remount which can't be changed.
1170 *
1171 * But unfortunately mount(8) adds all options from
1172 * mtab and fstab to the mount arguments in some cases
1173 * so we can't blindly reject options, but have to
1174 * check for each specified option if it actually
1175 * differs from the currently set option and only
1176 * reject it if that's the case.
1177 *
1178 * Until that is implemented we return success for
1179 * every remount request, and silently ignore all
1180 * options that we can't actually change.
1181 */
1182#if 0
1183 xfs_info(mp,
1184 "mount option \"%s\" not supported for remount\n", p);
1185 return -EINVAL;
1186#else
1187 break;
1188#endif
1189 }
1190 }
1191
1192 /* ro -> rw */
1193 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1194 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1195
1196 /*
1197 * If this is the first remount to writeable state we
1198 * might have some superblock changes to update.
1199 */
1200 if (mp->m_update_flags) {
1201 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1202 if (error) {
1203 xfs_warn(mp, "failed to write sb changes");
1204 return error;
1205 }
1206 mp->m_update_flags = 0;
1207 }
1208
1209 /*
1210 * Fill out the reserve pool if it is empty. Use the stashed
1211 * value if it is non-zero, otherwise go with the default.
1212 */
1213 xfs_restore_resvblks(mp);
1214 }
1215
1216 /* rw -> ro */
1217 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1218 /*
1219 * After we have synced the data but before we sync the
1220 * metadata, we need to free up the reserve block pool so that
1221 * the used block count in the superblock on disk is correct at
1222 * the end of the remount. Stash the current reserve pool size
1223 * so that if we get remounted rw, we can return it to the same
1224 * size.
1225 */
1226
1227 xfs_quiesce_data(mp);
1228 xfs_save_resvblks(mp);
1229 xfs_quiesce_attr(mp);
1230 mp->m_flags |= XFS_MOUNT_RDONLY;
1231 }
1232
1233 return 0;
1234}
1235
1236/*
1237 * Second stage of a freeze. The data is already frozen so we only
1238 * need to take care of the metadata. Once that's done write a dummy
1239 * record to dirty the log in case of a crash while frozen.
1240 */
1241STATIC int
1242xfs_fs_freeze(
1243 struct super_block *sb)
1244{
1245 struct xfs_mount *mp = XFS_M(sb);
1246
1247 xfs_save_resvblks(mp);
1248 xfs_quiesce_attr(mp);
1249 return -xfs_fs_log_dummy(mp);
1250}
1251
1252STATIC int
1253xfs_fs_unfreeze(
1254 struct super_block *sb)
1255{
1256 struct xfs_mount *mp = XFS_M(sb);
1257
1258 xfs_restore_resvblks(mp);
1259 return 0;
1260}
1261
1262STATIC int
1263xfs_fs_show_options(
1264 struct seq_file *m,
1265 struct vfsmount *mnt)
1266{
1267 return -xfs_showargs(XFS_M(mnt->mnt_sb), m);
1268}
1269
1270/*
1271 * This function fills in xfs_mount_t fields based on mount args.
1272 * Note: the superblock _has_ now been read in.
1273 */
1274STATIC int
1275xfs_finish_flags(
1276 struct xfs_mount *mp)
1277{
1278 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1279
1280 /* Fail a mount where the logbuf is smaller than the log stripe */
1281 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1282 if (mp->m_logbsize <= 0 &&
1283 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1284 mp->m_logbsize = mp->m_sb.sb_logsunit;
1285 } else if (mp->m_logbsize > 0 &&
1286 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1287 xfs_warn(mp,
1288 "logbuf size must be greater than or equal to log stripe size");
1289 return XFS_ERROR(EINVAL);
1290 }
1291 } else {
1292 /* Fail a mount if the logbuf is larger than 32K */
1293 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1294 xfs_warn(mp,
1295 "logbuf size for version 1 logs must be 16K or 32K");
1296 return XFS_ERROR(EINVAL);
1297 }
1298 }
1299
1300 /*
1301 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1302 * told by noattr2 to turn it off
1303 */
1304 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1305 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1306 mp->m_flags |= XFS_MOUNT_ATTR2;
1307
1308 /*
1309 * prohibit r/w mounts of read-only filesystems
1310 */
1311 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1312 xfs_warn(mp,
1313 "cannot mount a read-only filesystem as read-write");
1314 return XFS_ERROR(EROFS);
1315 }
1316
1317 return 0;
1318}
1319
1320STATIC int
1321xfs_fs_fill_super(
1322 struct super_block *sb,
1323 void *data,
1324 int silent)
1325{
1326 struct inode *root;
1327 struct xfs_mount *mp = NULL;
1328 int flags = 0, error = ENOMEM;
1329
1330 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1331 if (!mp)
1332 goto out;
1333
1334 spin_lock_init(&mp->m_sb_lock);
1335 mutex_init(&mp->m_growlock);
1336 atomic_set(&mp->m_active_trans, 0);
1337
1338 mp->m_super = sb;
1339 sb->s_fs_info = mp;
1340
1341 error = xfs_parseargs(mp, (char *)data);
1342 if (error)
1343 goto out_free_fsname;
1344
1345 sb_min_blocksize(sb, BBSIZE);
1346 sb->s_xattr = xfs_xattr_handlers;
1347 sb->s_export_op = &xfs_export_operations;
1348#ifdef CONFIG_XFS_QUOTA
1349 sb->s_qcop = &xfs_quotactl_operations;
1350#endif
1351 sb->s_op = &xfs_super_operations;
1352
1353 if (silent)
1354 flags |= XFS_MFSI_QUIET;
1355
1356 error = xfs_open_devices(mp);
1357 if (error)
1358 goto out_free_fsname;
1359
1360 error = xfs_icsb_init_counters(mp);
1361 if (error)
1362 goto out_close_devices;
1363
1364 error = xfs_readsb(mp, flags);
1365 if (error)
1366 goto out_destroy_counters;
1367
1368 error = xfs_finish_flags(mp);
1369 if (error)
1370 goto out_free_sb;
1371
1372 error = xfs_setup_devices(mp);
1373 if (error)
1374 goto out_free_sb;
1375
1376 error = xfs_filestream_mount(mp);
1377 if (error)
1378 goto out_free_sb;
1379
1380 /*
1381 * we must configure the block size in the superblock before we run the
1382 * full mount process as the mount process can lookup and cache inodes.
1383 * For the same reason we must also initialise the syncd and register
1384 * the inode cache shrinker so that inodes can be reclaimed during
1385 * operations like a quotacheck that iterate all inodes in the
1386 * filesystem.
1387 */
1388 sb->s_magic = XFS_SB_MAGIC;
1389 sb->s_blocksize = mp->m_sb.sb_blocksize;
1390 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1391 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1392 sb->s_time_gran = 1;
1393 set_posix_acl_flag(sb);
1394
1395 error = xfs_mountfs(mp);
1396 if (error)
1397 goto out_filestream_unmount;
1398
1399 error = xfs_syncd_init(mp);
1400 if (error)
1401 goto out_unmount;
1402
1403 root = igrab(VFS_I(mp->m_rootip));
1404 if (!root) {
1405 error = ENOENT;
1406 goto out_syncd_stop;
1407 }
1408 if (is_bad_inode(root)) {
1409 error = EINVAL;
1410 goto out_syncd_stop;
1411 }
1412 sb->s_root = d_alloc_root(root);
1413 if (!sb->s_root) {
1414 error = ENOMEM;
1415 goto out_iput;
1416 }
1417
1418 return 0;
1419
1420 out_filestream_unmount:
1421 xfs_filestream_unmount(mp);
1422 out_free_sb:
1423 xfs_freesb(mp);
1424 out_destroy_counters:
1425 xfs_icsb_destroy_counters(mp);
1426 out_close_devices:
1427 xfs_close_devices(mp);
1428 out_free_fsname:
1429 xfs_free_fsname(mp);
1430 kfree(mp);
1431 out:
1432 return -error;
1433
1434 out_iput:
1435 iput(root);
1436 out_syncd_stop:
1437 xfs_syncd_stop(mp);
1438 out_unmount:
1439 /*
1440 * Blow away any referenced inode in the filestreams cache.
1441 * This can and will cause log traffic as inodes go inactive
1442 * here.
1443 */
1444 xfs_filestream_unmount(mp);
1445
1446 XFS_bflush(mp->m_ddev_targp);
1447
1448 xfs_unmountfs(mp);
1449 goto out_free_sb;
1450}
1451
1452STATIC struct dentry *
1453xfs_fs_mount(
1454 struct file_system_type *fs_type,
1455 int flags,
1456 const char *dev_name,
1457 void *data)
1458{
1459 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1460}
1461
1462static int
1463xfs_fs_nr_cached_objects(
1464 struct super_block *sb)
1465{
1466 return xfs_reclaim_inodes_count(XFS_M(sb));
1467}
1468
1469static void
1470xfs_fs_free_cached_objects(
1471 struct super_block *sb,
1472 int nr_to_scan)
1473{
1474 xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1475}
1476
1477static const struct super_operations xfs_super_operations = {
1478 .alloc_inode = xfs_fs_alloc_inode,
1479 .destroy_inode = xfs_fs_destroy_inode,
1480 .dirty_inode = xfs_fs_dirty_inode,
1481 .write_inode = xfs_fs_write_inode,
1482 .evict_inode = xfs_fs_evict_inode,
1483 .put_super = xfs_fs_put_super,
1484 .sync_fs = xfs_fs_sync_fs,
1485 .freeze_fs = xfs_fs_freeze,
1486 .unfreeze_fs = xfs_fs_unfreeze,
1487 .statfs = xfs_fs_statfs,
1488 .remount_fs = xfs_fs_remount,
1489 .show_options = xfs_fs_show_options,
1490 .nr_cached_objects = xfs_fs_nr_cached_objects,
1491 .free_cached_objects = xfs_fs_free_cached_objects,
1492};
1493
1494static struct file_system_type xfs_fs_type = {
1495 .owner = THIS_MODULE,
1496 .name = "xfs",
1497 .mount = xfs_fs_mount,
1498 .kill_sb = kill_block_super,
1499 .fs_flags = FS_REQUIRES_DEV,
1500};
1501
1502STATIC int __init
1503xfs_init_zones(void)
1504{
1505
1506 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1507 if (!xfs_ioend_zone)
1508 goto out;
1509
1510 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1511 xfs_ioend_zone);
1512 if (!xfs_ioend_pool)
1513 goto out_destroy_ioend_zone;
1514
1515 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1516 "xfs_log_ticket");
1517 if (!xfs_log_ticket_zone)
1518 goto out_destroy_ioend_pool;
1519
1520 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1521 "xfs_bmap_free_item");
1522 if (!xfs_bmap_free_item_zone)
1523 goto out_destroy_log_ticket_zone;
1524
1525 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1526 "xfs_btree_cur");
1527 if (!xfs_btree_cur_zone)
1528 goto out_destroy_bmap_free_item_zone;
1529
1530 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1531 "xfs_da_state");
1532 if (!xfs_da_state_zone)
1533 goto out_destroy_btree_cur_zone;
1534
1535 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
1536 if (!xfs_dabuf_zone)
1537 goto out_destroy_da_state_zone;
1538
1539 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1540 if (!xfs_ifork_zone)
1541 goto out_destroy_dabuf_zone;
1542
1543 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1544 if (!xfs_trans_zone)
1545 goto out_destroy_ifork_zone;
1546
1547 xfs_log_item_desc_zone =
1548 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1549 "xfs_log_item_desc");
1550 if (!xfs_log_item_desc_zone)
1551 goto out_destroy_trans_zone;
1552
1553 /*
1554 * The size of the zone allocated buf log item is the maximum
1555 * size possible under XFS. This wastes a little bit of memory,
1556 * but it is much faster.
1557 */
1558 xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) +
1559 (((XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK) /
1560 NBWORD) * sizeof(int))), "xfs_buf_item");
1561 if (!xfs_buf_item_zone)
1562 goto out_destroy_log_item_desc_zone;
1563
1564 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1565 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1566 sizeof(xfs_extent_t))), "xfs_efd_item");
1567 if (!xfs_efd_zone)
1568 goto out_destroy_buf_item_zone;
1569
1570 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1571 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1572 sizeof(xfs_extent_t))), "xfs_efi_item");
1573 if (!xfs_efi_zone)
1574 goto out_destroy_efd_zone;
1575
1576 xfs_inode_zone =
1577 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1578 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1579 xfs_fs_inode_init_once);
1580 if (!xfs_inode_zone)
1581 goto out_destroy_efi_zone;
1582
1583 xfs_ili_zone =
1584 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1585 KM_ZONE_SPREAD, NULL);
1586 if (!xfs_ili_zone)
1587 goto out_destroy_inode_zone;
1588
1589 return 0;
1590
1591 out_destroy_inode_zone:
1592 kmem_zone_destroy(xfs_inode_zone);
1593 out_destroy_efi_zone:
1594 kmem_zone_destroy(xfs_efi_zone);
1595 out_destroy_efd_zone:
1596 kmem_zone_destroy(xfs_efd_zone);
1597 out_destroy_buf_item_zone:
1598 kmem_zone_destroy(xfs_buf_item_zone);
1599 out_destroy_log_item_desc_zone:
1600 kmem_zone_destroy(xfs_log_item_desc_zone);
1601 out_destroy_trans_zone:
1602 kmem_zone_destroy(xfs_trans_zone);
1603 out_destroy_ifork_zone:
1604 kmem_zone_destroy(xfs_ifork_zone);
1605 out_destroy_dabuf_zone:
1606 kmem_zone_destroy(xfs_dabuf_zone);
1607 out_destroy_da_state_zone:
1608 kmem_zone_destroy(xfs_da_state_zone);
1609 out_destroy_btree_cur_zone:
1610 kmem_zone_destroy(xfs_btree_cur_zone);
1611 out_destroy_bmap_free_item_zone:
1612 kmem_zone_destroy(xfs_bmap_free_item_zone);
1613 out_destroy_log_ticket_zone:
1614 kmem_zone_destroy(xfs_log_ticket_zone);
1615 out_destroy_ioend_pool:
1616 mempool_destroy(xfs_ioend_pool);
1617 out_destroy_ioend_zone:
1618 kmem_zone_destroy(xfs_ioend_zone);
1619 out:
1620 return -ENOMEM;
1621}
1622
1623STATIC void
1624xfs_destroy_zones(void)
1625{
1626 kmem_zone_destroy(xfs_ili_zone);
1627 kmem_zone_destroy(xfs_inode_zone);
1628 kmem_zone_destroy(xfs_efi_zone);
1629 kmem_zone_destroy(xfs_efd_zone);
1630 kmem_zone_destroy(xfs_buf_item_zone);
1631 kmem_zone_destroy(xfs_log_item_desc_zone);
1632 kmem_zone_destroy(xfs_trans_zone);
1633 kmem_zone_destroy(xfs_ifork_zone);
1634 kmem_zone_destroy(xfs_dabuf_zone);
1635 kmem_zone_destroy(xfs_da_state_zone);
1636 kmem_zone_destroy(xfs_btree_cur_zone);
1637 kmem_zone_destroy(xfs_bmap_free_item_zone);
1638 kmem_zone_destroy(xfs_log_ticket_zone);
1639 mempool_destroy(xfs_ioend_pool);
1640 kmem_zone_destroy(xfs_ioend_zone);
1641
1642}
1643
1644STATIC int __init
1645xfs_init_workqueues(void)
1646{
1647 /*
1648 * max_active is set to 8 to give enough concurency to allow
1649 * multiple work operations on each CPU to run. This allows multiple
1650 * filesystems to be running sync work concurrently, and scales with
1651 * the number of CPUs in the system.
1652 */
1653 xfs_syncd_wq = alloc_workqueue("xfssyncd", WQ_CPU_INTENSIVE, 8);
1654 if (!xfs_syncd_wq)
1655 return -ENOMEM;
1656 return 0;
1657}
1658
1659STATIC void
1660xfs_destroy_workqueues(void)
1661{
1662 destroy_workqueue(xfs_syncd_wq);
1663}
1664
1665STATIC int __init
1666init_xfs_fs(void)
1667{
1668 int error;
1669
1670 printk(KERN_INFO XFS_VERSION_STRING " with "
1671 XFS_BUILD_OPTIONS " enabled\n");
1672
1673 xfs_ioend_init();
1674 xfs_dir_startup();
1675
1676 error = xfs_init_zones();
1677 if (error)
1678 goto out;
1679
1680 error = xfs_init_workqueues();
1681 if (error)
1682 goto out_destroy_zones;
1683
1684 error = xfs_mru_cache_init();
1685 if (error)
1686 goto out_destroy_wq;
1687
1688 error = xfs_filestream_init();
1689 if (error)
1690 goto out_mru_cache_uninit;
1691
1692 error = xfs_buf_init();
1693 if (error)
1694 goto out_filestream_uninit;
1695
1696 error = xfs_init_procfs();
1697 if (error)
1698 goto out_buf_terminate;
1699
1700 error = xfs_sysctl_register();
1701 if (error)
1702 goto out_cleanup_procfs;
1703
1704 vfs_initquota();
1705
1706 error = register_filesystem(&xfs_fs_type);
1707 if (error)
1708 goto out_sysctl_unregister;
1709 return 0;
1710
1711 out_sysctl_unregister:
1712 xfs_sysctl_unregister();
1713 out_cleanup_procfs:
1714 xfs_cleanup_procfs();
1715 out_buf_terminate:
1716 xfs_buf_terminate();
1717 out_filestream_uninit:
1718 xfs_filestream_uninit();
1719 out_mru_cache_uninit:
1720 xfs_mru_cache_uninit();
1721 out_destroy_wq:
1722 xfs_destroy_workqueues();
1723 out_destroy_zones:
1724 xfs_destroy_zones();
1725 out:
1726 return error;
1727}
1728
1729STATIC void __exit
1730exit_xfs_fs(void)
1731{
1732 vfs_exitquota();
1733 unregister_filesystem(&xfs_fs_type);
1734 xfs_sysctl_unregister();
1735 xfs_cleanup_procfs();
1736 xfs_buf_terminate();
1737 xfs_filestream_uninit();
1738 xfs_mru_cache_uninit();
1739 xfs_destroy_workqueues();
1740 xfs_destroy_zones();
1741}
1742
1743module_init(init_xfs_fs);
1744module_exit(exit_xfs_fs);
1745
1746MODULE_AUTHOR("Silicon Graphics, Inc.");
1747MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1748MODULE_LICENSE("GPL");
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_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_inum.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_mount.h"
28#include "xfs_da_format.h"
29#include "xfs_inode.h"
30#include "xfs_btree.h"
31#include "xfs_bmap.h"
32#include "xfs_alloc.h"
33#include "xfs_error.h"
34#include "xfs_fsops.h"
35#include "xfs_trans.h"
36#include "xfs_buf_item.h"
37#include "xfs_log.h"
38#include "xfs_log_priv.h"
39#include "xfs_da_btree.h"
40#include "xfs_dir2.h"
41#include "xfs_extfree_item.h"
42#include "xfs_mru_cache.h"
43#include "xfs_inode_item.h"
44#include "xfs_icache.h"
45#include "xfs_trace.h"
46#include "xfs_icreate_item.h"
47#include "xfs_dinode.h"
48#include "xfs_filestream.h"
49#include "xfs_quota.h"
50
51#include <linux/namei.h>
52#include <linux/init.h>
53#include <linux/slab.h>
54#include <linux/mount.h>
55#include <linux/mempool.h>
56#include <linux/writeback.h>
57#include <linux/kthread.h>
58#include <linux/freezer.h>
59#include <linux/parser.h>
60
61static const struct super_operations xfs_super_operations;
62static kmem_zone_t *xfs_ioend_zone;
63mempool_t *xfs_ioend_pool;
64
65#define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
66#define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
67#define MNTOPT_LOGDEV "logdev" /* log device */
68#define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
69#define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
70#define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
71#define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
72#define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
73#define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
74#define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
75#define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
76#define MNTOPT_MTPT "mtpt" /* filesystem mount point */
77#define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
78#define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
79#define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
80#define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
81#define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
82#define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
83#define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
84 * unwritten extent conversion */
85#define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
86#define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
87#define MNTOPT_32BITINODE "inode32" /* inode allocation limited to
88 * XFS_MAXINUMBER_32 */
89#define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
90#define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
91#define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
92#define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
93 * in stat(). */
94#define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
95#define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
96#define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
97#define MNTOPT_QUOTA "quota" /* disk quotas (user) */
98#define MNTOPT_NOQUOTA "noquota" /* no quotas */
99#define MNTOPT_USRQUOTA "usrquota" /* user quota enabled */
100#define MNTOPT_GRPQUOTA "grpquota" /* group quota enabled */
101#define MNTOPT_PRJQUOTA "prjquota" /* project quota enabled */
102#define MNTOPT_UQUOTA "uquota" /* user quota (IRIX variant) */
103#define MNTOPT_GQUOTA "gquota" /* group quota (IRIX variant) */
104#define MNTOPT_PQUOTA "pquota" /* project quota (IRIX variant) */
105#define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
106#define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
107#define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
108#define MNTOPT_QUOTANOENF "qnoenforce" /* same as uqnoenforce */
109#define MNTOPT_DELAYLOG "delaylog" /* Delayed logging enabled */
110#define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed logging disabled */
111#define MNTOPT_DISCARD "discard" /* Discard unused blocks */
112#define MNTOPT_NODISCARD "nodiscard" /* Do not discard unused blocks */
113
114/*
115 * Table driven mount option parser.
116 *
117 * Currently only used for remount, but it will be used for mount
118 * in the future, too.
119 */
120enum {
121 Opt_barrier,
122 Opt_nobarrier,
123 Opt_inode64,
124 Opt_inode32,
125 Opt_err
126};
127
128static const match_table_t tokens = {
129 {Opt_barrier, "barrier"},
130 {Opt_nobarrier, "nobarrier"},
131 {Opt_inode64, "inode64"},
132 {Opt_inode32, "inode32"},
133 {Opt_err, NULL}
134};
135
136
137STATIC unsigned long
138suffix_kstrtoint(char *s, unsigned int base, int *res)
139{
140 int last, shift_left_factor = 0, _res;
141 char *value = s;
142
143 last = strlen(value) - 1;
144 if (value[last] == 'K' || value[last] == 'k') {
145 shift_left_factor = 10;
146 value[last] = '\0';
147 }
148 if (value[last] == 'M' || value[last] == 'm') {
149 shift_left_factor = 20;
150 value[last] = '\0';
151 }
152 if (value[last] == 'G' || value[last] == 'g') {
153 shift_left_factor = 30;
154 value[last] = '\0';
155 }
156
157 if (kstrtoint(s, base, &_res))
158 return -EINVAL;
159 *res = _res << shift_left_factor;
160 return 0;
161}
162
163/*
164 * This function fills in xfs_mount_t fields based on mount args.
165 * Note: the superblock has _not_ yet been read in.
166 *
167 * Note that this function leaks the various device name allocations on
168 * failure. The caller takes care of them.
169 */
170STATIC int
171xfs_parseargs(
172 struct xfs_mount *mp,
173 char *options)
174{
175 struct super_block *sb = mp->m_super;
176 char *this_char, *value;
177 int dsunit = 0;
178 int dswidth = 0;
179 int iosize = 0;
180 __uint8_t iosizelog = 0;
181
182 /*
183 * set up the mount name first so all the errors will refer to the
184 * correct device.
185 */
186 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
187 if (!mp->m_fsname)
188 return ENOMEM;
189 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
190
191 /*
192 * Copy binary VFS mount flags we are interested in.
193 */
194 if (sb->s_flags & MS_RDONLY)
195 mp->m_flags |= XFS_MOUNT_RDONLY;
196 if (sb->s_flags & MS_DIRSYNC)
197 mp->m_flags |= XFS_MOUNT_DIRSYNC;
198 if (sb->s_flags & MS_SYNCHRONOUS)
199 mp->m_flags |= XFS_MOUNT_WSYNC;
200
201 /*
202 * Set some default flags that could be cleared by the mount option
203 * parsing.
204 */
205 mp->m_flags |= XFS_MOUNT_BARRIER;
206 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
207#if !XFS_BIG_INUMS
208 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
209#endif
210
211 /*
212 * These can be overridden by the mount option parsing.
213 */
214 mp->m_logbufs = -1;
215 mp->m_logbsize = -1;
216
217 if (!options)
218 goto done;
219
220 while ((this_char = strsep(&options, ",")) != NULL) {
221 if (!*this_char)
222 continue;
223 if ((value = strchr(this_char, '=')) != NULL)
224 *value++ = 0;
225
226 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
227 if (!value || !*value) {
228 xfs_warn(mp, "%s option requires an argument",
229 this_char);
230 return EINVAL;
231 }
232 if (kstrtoint(value, 10, &mp->m_logbufs))
233 return EINVAL;
234 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
235 if (!value || !*value) {
236 xfs_warn(mp, "%s option requires an argument",
237 this_char);
238 return EINVAL;
239 }
240 if (suffix_kstrtoint(value, 10, &mp->m_logbsize))
241 return EINVAL;
242 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
243 if (!value || !*value) {
244 xfs_warn(mp, "%s option requires an argument",
245 this_char);
246 return EINVAL;
247 }
248 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
249 if (!mp->m_logname)
250 return ENOMEM;
251 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
252 xfs_warn(mp, "%s option not allowed on this system",
253 this_char);
254 return EINVAL;
255 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
256 if (!value || !*value) {
257 xfs_warn(mp, "%s option requires an argument",
258 this_char);
259 return EINVAL;
260 }
261 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
262 if (!mp->m_rtname)
263 return ENOMEM;
264 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
265 if (!value || !*value) {
266 xfs_warn(mp, "%s option requires an argument",
267 this_char);
268 return EINVAL;
269 }
270 if (kstrtoint(value, 10, &iosize))
271 return EINVAL;
272 iosizelog = ffs(iosize) - 1;
273 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
274 if (!value || !*value) {
275 xfs_warn(mp, "%s option requires an argument",
276 this_char);
277 return EINVAL;
278 }
279 if (suffix_kstrtoint(value, 10, &iosize))
280 return EINVAL;
281 iosizelog = ffs(iosize) - 1;
282 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
283 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
284 mp->m_flags |= XFS_MOUNT_GRPID;
285 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
286 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
287 mp->m_flags &= ~XFS_MOUNT_GRPID;
288 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
289 mp->m_flags |= XFS_MOUNT_WSYNC;
290 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
291 mp->m_flags |= XFS_MOUNT_NORECOVERY;
292 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
293 mp->m_flags |= XFS_MOUNT_NOALIGN;
294 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
295 mp->m_flags |= XFS_MOUNT_SWALLOC;
296 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
297 if (!value || !*value) {
298 xfs_warn(mp, "%s option requires an argument",
299 this_char);
300 return EINVAL;
301 }
302 if (kstrtoint(value, 10, &dsunit))
303 return EINVAL;
304 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
305 if (!value || !*value) {
306 xfs_warn(mp, "%s option requires an argument",
307 this_char);
308 return EINVAL;
309 }
310 if (kstrtoint(value, 10, &dswidth))
311 return EINVAL;
312 } else if (!strcmp(this_char, MNTOPT_32BITINODE)) {
313 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
314 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
315 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
316#if !XFS_BIG_INUMS
317 xfs_warn(mp, "%s option not allowed on this system",
318 this_char);
319 return EINVAL;
320#endif
321 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
322 mp->m_flags |= XFS_MOUNT_NOUUID;
323 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
324 mp->m_flags |= XFS_MOUNT_BARRIER;
325 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
326 mp->m_flags &= ~XFS_MOUNT_BARRIER;
327 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
328 mp->m_flags |= XFS_MOUNT_IKEEP;
329 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
330 mp->m_flags &= ~XFS_MOUNT_IKEEP;
331 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
332 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
333 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
334 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
335 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
336 mp->m_flags |= XFS_MOUNT_ATTR2;
337 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
338 mp->m_flags &= ~XFS_MOUNT_ATTR2;
339 mp->m_flags |= XFS_MOUNT_NOATTR2;
340 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
341 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
342 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
343 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
344 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
345 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
346 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
347 !strcmp(this_char, MNTOPT_UQUOTA) ||
348 !strcmp(this_char, MNTOPT_USRQUOTA)) {
349 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
350 XFS_UQUOTA_ENFD);
351 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
352 !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
353 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
354 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
355 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
356 !strcmp(this_char, MNTOPT_PRJQUOTA)) {
357 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
358 XFS_PQUOTA_ENFD);
359 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
360 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
361 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
362 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
363 !strcmp(this_char, MNTOPT_GRPQUOTA)) {
364 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
365 XFS_GQUOTA_ENFD);
366 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
367 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
368 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
369 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
370 xfs_warn(mp,
371 "delaylog is the default now, option is deprecated.");
372 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
373 xfs_warn(mp,
374 "nodelaylog support has been removed, option is deprecated.");
375 } else if (!strcmp(this_char, MNTOPT_DISCARD)) {
376 mp->m_flags |= XFS_MOUNT_DISCARD;
377 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
378 mp->m_flags &= ~XFS_MOUNT_DISCARD;
379 } else if (!strcmp(this_char, "ihashsize")) {
380 xfs_warn(mp,
381 "ihashsize no longer used, option is deprecated.");
382 } else if (!strcmp(this_char, "osyncisdsync")) {
383 xfs_warn(mp,
384 "osyncisdsync has no effect, option is deprecated.");
385 } else if (!strcmp(this_char, "osyncisosync")) {
386 xfs_warn(mp,
387 "osyncisosync has no effect, option is deprecated.");
388 } else if (!strcmp(this_char, "irixsgid")) {
389 xfs_warn(mp,
390 "irixsgid is now a sysctl(2) variable, option is deprecated.");
391 } else {
392 xfs_warn(mp, "unknown mount option [%s].", this_char);
393 return EINVAL;
394 }
395 }
396
397 /*
398 * no recovery flag requires a read-only mount
399 */
400 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
401 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
402 xfs_warn(mp, "no-recovery mounts must be read-only.");
403 return EINVAL;
404 }
405
406 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
407 xfs_warn(mp,
408 "sunit and swidth options incompatible with the noalign option");
409 return EINVAL;
410 }
411
412#ifndef CONFIG_XFS_QUOTA
413 if (XFS_IS_QUOTA_RUNNING(mp)) {
414 xfs_warn(mp, "quota support not available in this kernel.");
415 return EINVAL;
416 }
417#endif
418
419 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
420 xfs_warn(mp, "sunit and swidth must be specified together");
421 return EINVAL;
422 }
423
424 if (dsunit && (dswidth % dsunit != 0)) {
425 xfs_warn(mp,
426 "stripe width (%d) must be a multiple of the stripe unit (%d)",
427 dswidth, dsunit);
428 return EINVAL;
429 }
430
431done:
432 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
433 /*
434 * At this point the superblock has not been read
435 * in, therefore we do not know the block size.
436 * Before the mount call ends we will convert
437 * these to FSBs.
438 */
439 mp->m_dalign = dsunit;
440 mp->m_swidth = dswidth;
441 }
442
443 if (mp->m_logbufs != -1 &&
444 mp->m_logbufs != 0 &&
445 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
446 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
447 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
448 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
449 return XFS_ERROR(EINVAL);
450 }
451 if (mp->m_logbsize != -1 &&
452 mp->m_logbsize != 0 &&
453 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
454 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
455 !is_power_of_2(mp->m_logbsize))) {
456 xfs_warn(mp,
457 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
458 mp->m_logbsize);
459 return XFS_ERROR(EINVAL);
460 }
461
462 if (iosizelog) {
463 if (iosizelog > XFS_MAX_IO_LOG ||
464 iosizelog < XFS_MIN_IO_LOG) {
465 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
466 iosizelog, XFS_MIN_IO_LOG,
467 XFS_MAX_IO_LOG);
468 return XFS_ERROR(EINVAL);
469 }
470
471 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
472 mp->m_readio_log = iosizelog;
473 mp->m_writeio_log = iosizelog;
474 }
475
476 return 0;
477}
478
479struct proc_xfs_info {
480 int flag;
481 char *str;
482};
483
484STATIC int
485xfs_showargs(
486 struct xfs_mount *mp,
487 struct seq_file *m)
488{
489 static struct proc_xfs_info xfs_info_set[] = {
490 /* the few simple ones we can get from the mount struct */
491 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP },
492 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
493 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
494 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
495 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
496 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
497 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 },
498 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
499 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
500 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD },
501 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_32BITINODE },
502 { 0, NULL }
503 };
504 static struct proc_xfs_info xfs_info_unset[] = {
505 /* the few simple ones we can get from the mount struct */
506 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO },
507 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER },
508 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE },
509 { 0, NULL }
510 };
511 struct proc_xfs_info *xfs_infop;
512
513 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
514 if (mp->m_flags & xfs_infop->flag)
515 seq_puts(m, xfs_infop->str);
516 }
517 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
518 if (!(mp->m_flags & xfs_infop->flag))
519 seq_puts(m, xfs_infop->str);
520 }
521
522 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
523 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
524 (int)(1 << mp->m_writeio_log) >> 10);
525
526 if (mp->m_logbufs > 0)
527 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
528 if (mp->m_logbsize > 0)
529 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
530
531 if (mp->m_logname)
532 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
533 if (mp->m_rtname)
534 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
535
536 if (mp->m_dalign > 0)
537 seq_printf(m, "," MNTOPT_SUNIT "=%d",
538 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
539 if (mp->m_swidth > 0)
540 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
541 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
542
543 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
544 seq_puts(m, "," MNTOPT_USRQUOTA);
545 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
546 seq_puts(m, "," MNTOPT_UQUOTANOENF);
547
548 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
549 if (mp->m_qflags & XFS_PQUOTA_ENFD)
550 seq_puts(m, "," MNTOPT_PRJQUOTA);
551 else
552 seq_puts(m, "," MNTOPT_PQUOTANOENF);
553 }
554 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
555 if (mp->m_qflags & XFS_GQUOTA_ENFD)
556 seq_puts(m, "," MNTOPT_GRPQUOTA);
557 else
558 seq_puts(m, "," MNTOPT_GQUOTANOENF);
559 }
560
561 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
562 seq_puts(m, "," MNTOPT_NOQUOTA);
563
564 return 0;
565}
566__uint64_t
567xfs_max_file_offset(
568 unsigned int blockshift)
569{
570 unsigned int pagefactor = 1;
571 unsigned int bitshift = BITS_PER_LONG - 1;
572
573 /* Figure out maximum filesize, on Linux this can depend on
574 * the filesystem blocksize (on 32 bit platforms).
575 * __block_write_begin does this in an [unsigned] long...
576 * page->index << (PAGE_CACHE_SHIFT - bbits)
577 * So, for page sized blocks (4K on 32 bit platforms),
578 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
579 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
580 * but for smaller blocksizes it is less (bbits = log2 bsize).
581 * Note1: get_block_t takes a long (implicit cast from above)
582 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
583 * can optionally convert the [unsigned] long from above into
584 * an [unsigned] long long.
585 */
586
587#if BITS_PER_LONG == 32
588# if defined(CONFIG_LBDAF)
589 ASSERT(sizeof(sector_t) == 8);
590 pagefactor = PAGE_CACHE_SIZE;
591 bitshift = BITS_PER_LONG;
592# else
593 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
594# endif
595#endif
596
597 return (((__uint64_t)pagefactor) << bitshift) - 1;
598}
599
600xfs_agnumber_t
601xfs_set_inode32(struct xfs_mount *mp)
602{
603 xfs_agnumber_t index = 0;
604 xfs_agnumber_t maxagi = 0;
605 xfs_sb_t *sbp = &mp->m_sb;
606 xfs_agnumber_t max_metadata;
607 xfs_agino_t agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks -1, 0);
608 xfs_ino_t ino = XFS_AGINO_TO_INO(mp, sbp->sb_agcount -1, agino);
609 xfs_perag_t *pag;
610
611 /* Calculate how much should be reserved for inodes to meet
612 * the max inode percentage.
613 */
614 if (mp->m_maxicount) {
615 __uint64_t icount;
616
617 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
618 do_div(icount, 100);
619 icount += sbp->sb_agblocks - 1;
620 do_div(icount, sbp->sb_agblocks);
621 max_metadata = icount;
622 } else {
623 max_metadata = sbp->sb_agcount;
624 }
625
626 for (index = 0; index < sbp->sb_agcount; index++) {
627 ino = XFS_AGINO_TO_INO(mp, index, agino);
628
629 if (ino > XFS_MAXINUMBER_32) {
630 pag = xfs_perag_get(mp, index);
631 pag->pagi_inodeok = 0;
632 pag->pagf_metadata = 0;
633 xfs_perag_put(pag);
634 continue;
635 }
636
637 pag = xfs_perag_get(mp, index);
638 pag->pagi_inodeok = 1;
639 maxagi++;
640 if (index < max_metadata)
641 pag->pagf_metadata = 1;
642 xfs_perag_put(pag);
643 }
644 mp->m_flags |= (XFS_MOUNT_32BITINODES |
645 XFS_MOUNT_SMALL_INUMS);
646
647 return maxagi;
648}
649
650xfs_agnumber_t
651xfs_set_inode64(struct xfs_mount *mp)
652{
653 xfs_agnumber_t index = 0;
654
655 for (index = 0; index < mp->m_sb.sb_agcount; index++) {
656 struct xfs_perag *pag;
657
658 pag = xfs_perag_get(mp, index);
659 pag->pagi_inodeok = 1;
660 pag->pagf_metadata = 0;
661 xfs_perag_put(pag);
662 }
663
664 /* There is no need for lock protection on m_flags,
665 * the rw_semaphore of the VFS superblock is locked
666 * during mount/umount/remount operations, so this is
667 * enough to avoid concurency on the m_flags field
668 */
669 mp->m_flags &= ~(XFS_MOUNT_32BITINODES |
670 XFS_MOUNT_SMALL_INUMS);
671 return index;
672}
673
674STATIC int
675xfs_blkdev_get(
676 xfs_mount_t *mp,
677 const char *name,
678 struct block_device **bdevp)
679{
680 int error = 0;
681
682 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
683 mp);
684 if (IS_ERR(*bdevp)) {
685 error = PTR_ERR(*bdevp);
686 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
687 }
688
689 return -error;
690}
691
692STATIC void
693xfs_blkdev_put(
694 struct block_device *bdev)
695{
696 if (bdev)
697 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
698}
699
700void
701xfs_blkdev_issue_flush(
702 xfs_buftarg_t *buftarg)
703{
704 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
705}
706
707STATIC void
708xfs_close_devices(
709 struct xfs_mount *mp)
710{
711 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
712 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
713 xfs_free_buftarg(mp, mp->m_logdev_targp);
714 xfs_blkdev_put(logdev);
715 }
716 if (mp->m_rtdev_targp) {
717 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
718 xfs_free_buftarg(mp, mp->m_rtdev_targp);
719 xfs_blkdev_put(rtdev);
720 }
721 xfs_free_buftarg(mp, mp->m_ddev_targp);
722}
723
724/*
725 * The file system configurations are:
726 * (1) device (partition) with data and internal log
727 * (2) logical volume with data and log subvolumes.
728 * (3) logical volume with data, log, and realtime subvolumes.
729 *
730 * We only have to handle opening the log and realtime volumes here if
731 * they are present. The data subvolume has already been opened by
732 * get_sb_bdev() and is stored in sb->s_bdev.
733 */
734STATIC int
735xfs_open_devices(
736 struct xfs_mount *mp)
737{
738 struct block_device *ddev = mp->m_super->s_bdev;
739 struct block_device *logdev = NULL, *rtdev = NULL;
740 int error;
741
742 /*
743 * Open real time and log devices - order is important.
744 */
745 if (mp->m_logname) {
746 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
747 if (error)
748 goto out;
749 }
750
751 if (mp->m_rtname) {
752 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
753 if (error)
754 goto out_close_logdev;
755
756 if (rtdev == ddev || rtdev == logdev) {
757 xfs_warn(mp,
758 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
759 error = EINVAL;
760 goto out_close_rtdev;
761 }
762 }
763
764 /*
765 * Setup xfs_mount buffer target pointers
766 */
767 error = ENOMEM;
768 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
769 if (!mp->m_ddev_targp)
770 goto out_close_rtdev;
771
772 if (rtdev) {
773 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
774 mp->m_fsname);
775 if (!mp->m_rtdev_targp)
776 goto out_free_ddev_targ;
777 }
778
779 if (logdev && logdev != ddev) {
780 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
781 mp->m_fsname);
782 if (!mp->m_logdev_targp)
783 goto out_free_rtdev_targ;
784 } else {
785 mp->m_logdev_targp = mp->m_ddev_targp;
786 }
787
788 return 0;
789
790 out_free_rtdev_targ:
791 if (mp->m_rtdev_targp)
792 xfs_free_buftarg(mp, mp->m_rtdev_targp);
793 out_free_ddev_targ:
794 xfs_free_buftarg(mp, mp->m_ddev_targp);
795 out_close_rtdev:
796 if (rtdev)
797 xfs_blkdev_put(rtdev);
798 out_close_logdev:
799 if (logdev && logdev != ddev)
800 xfs_blkdev_put(logdev);
801 out:
802 return error;
803}
804
805/*
806 * Setup xfs_mount buffer target pointers based on superblock
807 */
808STATIC int
809xfs_setup_devices(
810 struct xfs_mount *mp)
811{
812 int error;
813
814 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
815 mp->m_sb.sb_sectsize);
816 if (error)
817 return error;
818
819 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
820 unsigned int log_sector_size = BBSIZE;
821
822 if (xfs_sb_version_hassector(&mp->m_sb))
823 log_sector_size = mp->m_sb.sb_logsectsize;
824 error = xfs_setsize_buftarg(mp->m_logdev_targp,
825 mp->m_sb.sb_blocksize,
826 log_sector_size);
827 if (error)
828 return error;
829 }
830 if (mp->m_rtdev_targp) {
831 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
832 mp->m_sb.sb_blocksize,
833 mp->m_sb.sb_sectsize);
834 if (error)
835 return error;
836 }
837
838 return 0;
839}
840
841STATIC int
842xfs_init_mount_workqueues(
843 struct xfs_mount *mp)
844{
845 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
846 WQ_MEM_RECLAIM, 0, mp->m_fsname);
847 if (!mp->m_data_workqueue)
848 goto out;
849
850 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
851 WQ_MEM_RECLAIM, 0, mp->m_fsname);
852 if (!mp->m_unwritten_workqueue)
853 goto out_destroy_data_iodone_queue;
854
855 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
856 WQ_MEM_RECLAIM, 0, mp->m_fsname);
857 if (!mp->m_cil_workqueue)
858 goto out_destroy_unwritten;
859
860 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
861 0, 0, mp->m_fsname);
862 if (!mp->m_reclaim_workqueue)
863 goto out_destroy_cil;
864
865 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
866 0, 0, mp->m_fsname);
867 if (!mp->m_log_workqueue)
868 goto out_destroy_reclaim;
869
870 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
871 0, 0, mp->m_fsname);
872 if (!mp->m_eofblocks_workqueue)
873 goto out_destroy_log;
874
875 return 0;
876
877out_destroy_log:
878 destroy_workqueue(mp->m_log_workqueue);
879out_destroy_reclaim:
880 destroy_workqueue(mp->m_reclaim_workqueue);
881out_destroy_cil:
882 destroy_workqueue(mp->m_cil_workqueue);
883out_destroy_unwritten:
884 destroy_workqueue(mp->m_unwritten_workqueue);
885out_destroy_data_iodone_queue:
886 destroy_workqueue(mp->m_data_workqueue);
887out:
888 return -ENOMEM;
889}
890
891STATIC void
892xfs_destroy_mount_workqueues(
893 struct xfs_mount *mp)
894{
895 destroy_workqueue(mp->m_eofblocks_workqueue);
896 destroy_workqueue(mp->m_log_workqueue);
897 destroy_workqueue(mp->m_reclaim_workqueue);
898 destroy_workqueue(mp->m_cil_workqueue);
899 destroy_workqueue(mp->m_data_workqueue);
900 destroy_workqueue(mp->m_unwritten_workqueue);
901}
902
903/*
904 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
905 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
906 * for IO to complete so that we effectively throttle multiple callers to the
907 * rate at which IO is completing.
908 */
909void
910xfs_flush_inodes(
911 struct xfs_mount *mp)
912{
913 struct super_block *sb = mp->m_super;
914
915 if (down_read_trylock(&sb->s_umount)) {
916 sync_inodes_sb(sb);
917 up_read(&sb->s_umount);
918 }
919}
920
921/* Catch misguided souls that try to use this interface on XFS */
922STATIC struct inode *
923xfs_fs_alloc_inode(
924 struct super_block *sb)
925{
926 BUG();
927 return NULL;
928}
929
930/*
931 * Now that the generic code is guaranteed not to be accessing
932 * the linux inode, we can reclaim the inode.
933 */
934STATIC void
935xfs_fs_destroy_inode(
936 struct inode *inode)
937{
938 struct xfs_inode *ip = XFS_I(inode);
939
940 trace_xfs_destroy_inode(ip);
941
942 XFS_STATS_INC(vn_reclaim);
943
944 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
945
946 /*
947 * We should never get here with one of the reclaim flags already set.
948 */
949 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
950 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
951
952 /*
953 * We always use background reclaim here because even if the
954 * inode is clean, it still may be under IO and hence we have
955 * to take the flush lock. The background reclaim path handles
956 * this more efficiently than we can here, so simply let background
957 * reclaim tear down all inodes.
958 */
959 xfs_inode_set_reclaim_tag(ip);
960}
961
962/*
963 * Slab object creation initialisation for the XFS inode.
964 * This covers only the idempotent fields in the XFS inode;
965 * all other fields need to be initialised on allocation
966 * from the slab. This avoids the need to repeatedly initialise
967 * fields in the xfs inode that left in the initialise state
968 * when freeing the inode.
969 */
970STATIC void
971xfs_fs_inode_init_once(
972 void *inode)
973{
974 struct xfs_inode *ip = inode;
975
976 memset(ip, 0, sizeof(struct xfs_inode));
977
978 /* vfs inode */
979 inode_init_once(VFS_I(ip));
980
981 /* xfs inode */
982 atomic_set(&ip->i_pincount, 0);
983 spin_lock_init(&ip->i_flags_lock);
984
985 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
986 "xfsino", ip->i_ino);
987}
988
989STATIC void
990xfs_fs_evict_inode(
991 struct inode *inode)
992{
993 xfs_inode_t *ip = XFS_I(inode);
994
995 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
996
997 trace_xfs_evict_inode(ip);
998
999 truncate_inode_pages_final(&inode->i_data);
1000 clear_inode(inode);
1001 XFS_STATS_INC(vn_rele);
1002 XFS_STATS_INC(vn_remove);
1003 XFS_STATS_DEC(vn_active);
1004
1005 xfs_inactive(ip);
1006}
1007
1008/*
1009 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1010 * serialised against cache hits here via the inode->i_lock and igrab() in
1011 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1012 * racing with us, and it avoids needing to grab a spinlock here for every inode
1013 * we drop the final reference on.
1014 */
1015STATIC int
1016xfs_fs_drop_inode(
1017 struct inode *inode)
1018{
1019 struct xfs_inode *ip = XFS_I(inode);
1020
1021 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1022}
1023
1024STATIC void
1025xfs_free_fsname(
1026 struct xfs_mount *mp)
1027{
1028 kfree(mp->m_fsname);
1029 kfree(mp->m_rtname);
1030 kfree(mp->m_logname);
1031}
1032
1033STATIC void
1034xfs_fs_put_super(
1035 struct super_block *sb)
1036{
1037 struct xfs_mount *mp = XFS_M(sb);
1038
1039 xfs_filestream_unmount(mp);
1040 xfs_unmountfs(mp);
1041
1042 xfs_freesb(mp);
1043 xfs_icsb_destroy_counters(mp);
1044 xfs_destroy_mount_workqueues(mp);
1045 xfs_close_devices(mp);
1046 xfs_free_fsname(mp);
1047 kfree(mp);
1048}
1049
1050STATIC int
1051xfs_fs_sync_fs(
1052 struct super_block *sb,
1053 int wait)
1054{
1055 struct xfs_mount *mp = XFS_M(sb);
1056
1057 /*
1058 * Doing anything during the async pass would be counterproductive.
1059 */
1060 if (!wait)
1061 return 0;
1062
1063 xfs_log_force(mp, XFS_LOG_SYNC);
1064 if (laptop_mode) {
1065 /*
1066 * The disk must be active because we're syncing.
1067 * We schedule log work now (now that the disk is
1068 * active) instead of later (when it might not be).
1069 */
1070 flush_delayed_work(&mp->m_log->l_work);
1071 }
1072
1073 return 0;
1074}
1075
1076STATIC int
1077xfs_fs_statfs(
1078 struct dentry *dentry,
1079 struct kstatfs *statp)
1080{
1081 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1082 xfs_sb_t *sbp = &mp->m_sb;
1083 struct xfs_inode *ip = XFS_I(dentry->d_inode);
1084 __uint64_t fakeinos, id;
1085 xfs_extlen_t lsize;
1086 __int64_t ffree;
1087
1088 statp->f_type = XFS_SB_MAGIC;
1089 statp->f_namelen = MAXNAMELEN - 1;
1090
1091 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1092 statp->f_fsid.val[0] = (u32)id;
1093 statp->f_fsid.val[1] = (u32)(id >> 32);
1094
1095 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1096
1097 spin_lock(&mp->m_sb_lock);
1098 statp->f_bsize = sbp->sb_blocksize;
1099 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1100 statp->f_blocks = sbp->sb_dblocks - lsize;
1101 statp->f_bfree = statp->f_bavail =
1102 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1103 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1104 statp->f_files =
1105 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1106 if (mp->m_maxicount)
1107 statp->f_files = min_t(typeof(statp->f_files),
1108 statp->f_files,
1109 mp->m_maxicount);
1110
1111 /* make sure statp->f_ffree does not underflow */
1112 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1113 statp->f_ffree = max_t(__int64_t, ffree, 0);
1114
1115 spin_unlock(&mp->m_sb_lock);
1116
1117 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1118 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1119 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1120 xfs_qm_statvfs(ip, statp);
1121 return 0;
1122}
1123
1124STATIC void
1125xfs_save_resvblks(struct xfs_mount *mp)
1126{
1127 __uint64_t resblks = 0;
1128
1129 mp->m_resblks_save = mp->m_resblks;
1130 xfs_reserve_blocks(mp, &resblks, NULL);
1131}
1132
1133STATIC void
1134xfs_restore_resvblks(struct xfs_mount *mp)
1135{
1136 __uint64_t resblks;
1137
1138 if (mp->m_resblks_save) {
1139 resblks = mp->m_resblks_save;
1140 mp->m_resblks_save = 0;
1141 } else
1142 resblks = xfs_default_resblks(mp);
1143
1144 xfs_reserve_blocks(mp, &resblks, NULL);
1145}
1146
1147/*
1148 * Trigger writeback of all the dirty metadata in the file system.
1149 *
1150 * This ensures that the metadata is written to their location on disk rather
1151 * than just existing in transactions in the log. This means after a quiesce
1152 * there is no log replay required to write the inodes to disk - this is the
1153 * primary difference between a sync and a quiesce.
1154 *
1155 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1156 * it is started again when appropriate.
1157 */
1158static void
1159xfs_quiesce_attr(
1160 struct xfs_mount *mp)
1161{
1162 int error = 0;
1163
1164 /* wait for all modifications to complete */
1165 while (atomic_read(&mp->m_active_trans) > 0)
1166 delay(100);
1167
1168 /* force the log to unpin objects from the now complete transactions */
1169 xfs_log_force(mp, XFS_LOG_SYNC);
1170
1171 /* reclaim inodes to do any IO before the freeze completes */
1172 xfs_reclaim_inodes(mp, 0);
1173 xfs_reclaim_inodes(mp, SYNC_WAIT);
1174
1175 /* Push the superblock and write an unmount record */
1176 error = xfs_log_sbcount(mp);
1177 if (error)
1178 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1179 "Frozen image may not be consistent.");
1180 /*
1181 * Just warn here till VFS can correctly support
1182 * read-only remount without racing.
1183 */
1184 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1185
1186 xfs_log_quiesce(mp);
1187}
1188
1189STATIC int
1190xfs_fs_remount(
1191 struct super_block *sb,
1192 int *flags,
1193 char *options)
1194{
1195 struct xfs_mount *mp = XFS_M(sb);
1196 substring_t args[MAX_OPT_ARGS];
1197 char *p;
1198 int error;
1199
1200 sync_filesystem(sb);
1201 while ((p = strsep(&options, ",")) != NULL) {
1202 int token;
1203
1204 if (!*p)
1205 continue;
1206
1207 token = match_token(p, tokens, args);
1208 switch (token) {
1209 case Opt_barrier:
1210 mp->m_flags |= XFS_MOUNT_BARRIER;
1211 break;
1212 case Opt_nobarrier:
1213 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1214 break;
1215 case Opt_inode64:
1216 mp->m_maxagi = xfs_set_inode64(mp);
1217 break;
1218 case Opt_inode32:
1219 mp->m_maxagi = xfs_set_inode32(mp);
1220 break;
1221 default:
1222 /*
1223 * Logically we would return an error here to prevent
1224 * users from believing they might have changed
1225 * mount options using remount which can't be changed.
1226 *
1227 * But unfortunately mount(8) adds all options from
1228 * mtab and fstab to the mount arguments in some cases
1229 * so we can't blindly reject options, but have to
1230 * check for each specified option if it actually
1231 * differs from the currently set option and only
1232 * reject it if that's the case.
1233 *
1234 * Until that is implemented we return success for
1235 * every remount request, and silently ignore all
1236 * options that we can't actually change.
1237 */
1238#if 0
1239 xfs_info(mp,
1240 "mount option \"%s\" not supported for remount", p);
1241 return -EINVAL;
1242#else
1243 break;
1244#endif
1245 }
1246 }
1247
1248 /* ro -> rw */
1249 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1250 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1251
1252 /*
1253 * If this is the first remount to writeable state we
1254 * might have some superblock changes to update.
1255 */
1256 if (mp->m_update_flags) {
1257 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1258 if (error) {
1259 xfs_warn(mp, "failed to write sb changes");
1260 return error;
1261 }
1262 mp->m_update_flags = 0;
1263 }
1264
1265 /*
1266 * Fill out the reserve pool if it is empty. Use the stashed
1267 * value if it is non-zero, otherwise go with the default.
1268 */
1269 xfs_restore_resvblks(mp);
1270 xfs_log_work_queue(mp);
1271 }
1272
1273 /* rw -> ro */
1274 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1275 /*
1276 * Before we sync the metadata, we need to free up the reserve
1277 * block pool so that the used block count in the superblock on
1278 * disk is correct at the end of the remount. Stash the current
1279 * reserve pool size so that if we get remounted rw, we can
1280 * return it to the same size.
1281 */
1282 xfs_save_resvblks(mp);
1283 xfs_quiesce_attr(mp);
1284 mp->m_flags |= XFS_MOUNT_RDONLY;
1285 }
1286
1287 return 0;
1288}
1289
1290/*
1291 * Second stage of a freeze. The data is already frozen so we only
1292 * need to take care of the metadata. Once that's done write a dummy
1293 * record to dirty the log in case of a crash while frozen.
1294 */
1295STATIC int
1296xfs_fs_freeze(
1297 struct super_block *sb)
1298{
1299 struct xfs_mount *mp = XFS_M(sb);
1300
1301 xfs_save_resvblks(mp);
1302 xfs_quiesce_attr(mp);
1303 return -xfs_fs_log_dummy(mp);
1304}
1305
1306STATIC int
1307xfs_fs_unfreeze(
1308 struct super_block *sb)
1309{
1310 struct xfs_mount *mp = XFS_M(sb);
1311
1312 xfs_restore_resvblks(mp);
1313 xfs_log_work_queue(mp);
1314 return 0;
1315}
1316
1317STATIC int
1318xfs_fs_show_options(
1319 struct seq_file *m,
1320 struct dentry *root)
1321{
1322 return -xfs_showargs(XFS_M(root->d_sb), m);
1323}
1324
1325/*
1326 * This function fills in xfs_mount_t fields based on mount args.
1327 * Note: the superblock _has_ now been read in.
1328 */
1329STATIC int
1330xfs_finish_flags(
1331 struct xfs_mount *mp)
1332{
1333 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1334
1335 /* Fail a mount where the logbuf is smaller than the log stripe */
1336 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1337 if (mp->m_logbsize <= 0 &&
1338 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1339 mp->m_logbsize = mp->m_sb.sb_logsunit;
1340 } else if (mp->m_logbsize > 0 &&
1341 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1342 xfs_warn(mp,
1343 "logbuf size must be greater than or equal to log stripe size");
1344 return XFS_ERROR(EINVAL);
1345 }
1346 } else {
1347 /* Fail a mount if the logbuf is larger than 32K */
1348 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1349 xfs_warn(mp,
1350 "logbuf size for version 1 logs must be 16K or 32K");
1351 return XFS_ERROR(EINVAL);
1352 }
1353 }
1354
1355 /*
1356 * V5 filesystems always use attr2 format for attributes.
1357 */
1358 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1359 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1360 xfs_warn(mp,
1361"Cannot mount a V5 filesystem as %s. %s is always enabled for V5 filesystems.",
1362 MNTOPT_NOATTR2, MNTOPT_ATTR2);
1363 return XFS_ERROR(EINVAL);
1364 }
1365
1366 /*
1367 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1368 * told by noattr2 to turn it off
1369 */
1370 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1371 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1372 mp->m_flags |= XFS_MOUNT_ATTR2;
1373
1374 /*
1375 * prohibit r/w mounts of read-only filesystems
1376 */
1377 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1378 xfs_warn(mp,
1379 "cannot mount a read-only filesystem as read-write");
1380 return XFS_ERROR(EROFS);
1381 }
1382
1383 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1384 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1385 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1386 xfs_warn(mp,
1387 "Super block does not support project and group quota together");
1388 return XFS_ERROR(EINVAL);
1389 }
1390
1391 return 0;
1392}
1393
1394STATIC int
1395xfs_fs_fill_super(
1396 struct super_block *sb,
1397 void *data,
1398 int silent)
1399{
1400 struct inode *root;
1401 struct xfs_mount *mp = NULL;
1402 int flags = 0, error = ENOMEM;
1403
1404 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1405 if (!mp)
1406 goto out;
1407
1408 spin_lock_init(&mp->m_sb_lock);
1409 mutex_init(&mp->m_growlock);
1410 atomic_set(&mp->m_active_trans, 0);
1411 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1412 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1413
1414 mp->m_super = sb;
1415 sb->s_fs_info = mp;
1416
1417 error = xfs_parseargs(mp, (char *)data);
1418 if (error)
1419 goto out_free_fsname;
1420
1421 sb_min_blocksize(sb, BBSIZE);
1422 sb->s_xattr = xfs_xattr_handlers;
1423 sb->s_export_op = &xfs_export_operations;
1424#ifdef CONFIG_XFS_QUOTA
1425 sb->s_qcop = &xfs_quotactl_operations;
1426#endif
1427 sb->s_op = &xfs_super_operations;
1428
1429 if (silent)
1430 flags |= XFS_MFSI_QUIET;
1431
1432 error = xfs_open_devices(mp);
1433 if (error)
1434 goto out_free_fsname;
1435
1436 error = -xfs_init_mount_workqueues(mp);
1437 if (error)
1438 goto out_close_devices;
1439
1440 error = -xfs_icsb_init_counters(mp);
1441 if (error)
1442 goto out_destroy_workqueues;
1443
1444 error = xfs_readsb(mp, flags);
1445 if (error)
1446 goto out_destroy_counters;
1447
1448 error = xfs_finish_flags(mp);
1449 if (error)
1450 goto out_free_sb;
1451
1452 error = xfs_setup_devices(mp);
1453 if (error)
1454 goto out_free_sb;
1455
1456 error = xfs_filestream_mount(mp);
1457 if (error)
1458 goto out_free_sb;
1459
1460 /*
1461 * we must configure the block size in the superblock before we run the
1462 * full mount process as the mount process can lookup and cache inodes.
1463 */
1464 sb->s_magic = XFS_SB_MAGIC;
1465 sb->s_blocksize = mp->m_sb.sb_blocksize;
1466 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1467 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1468 sb->s_max_links = XFS_MAXLINK;
1469 sb->s_time_gran = 1;
1470 set_posix_acl_flag(sb);
1471
1472 /* version 5 superblocks support inode version counters. */
1473 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1474 sb->s_flags |= MS_I_VERSION;
1475
1476 error = xfs_mountfs(mp);
1477 if (error)
1478 goto out_filestream_unmount;
1479
1480 root = igrab(VFS_I(mp->m_rootip));
1481 if (!root) {
1482 error = ENOENT;
1483 goto out_unmount;
1484 }
1485 sb->s_root = d_make_root(root);
1486 if (!sb->s_root) {
1487 error = ENOMEM;
1488 goto out_unmount;
1489 }
1490
1491 return 0;
1492
1493 out_filestream_unmount:
1494 xfs_filestream_unmount(mp);
1495 out_free_sb:
1496 xfs_freesb(mp);
1497 out_destroy_counters:
1498 xfs_icsb_destroy_counters(mp);
1499out_destroy_workqueues:
1500 xfs_destroy_mount_workqueues(mp);
1501 out_close_devices:
1502 xfs_close_devices(mp);
1503 out_free_fsname:
1504 xfs_free_fsname(mp);
1505 kfree(mp);
1506 out:
1507 return -error;
1508
1509 out_unmount:
1510 xfs_filestream_unmount(mp);
1511 xfs_unmountfs(mp);
1512 goto out_free_sb;
1513}
1514
1515STATIC struct dentry *
1516xfs_fs_mount(
1517 struct file_system_type *fs_type,
1518 int flags,
1519 const char *dev_name,
1520 void *data)
1521{
1522 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1523}
1524
1525static long
1526xfs_fs_nr_cached_objects(
1527 struct super_block *sb,
1528 int nid)
1529{
1530 return xfs_reclaim_inodes_count(XFS_M(sb));
1531}
1532
1533static long
1534xfs_fs_free_cached_objects(
1535 struct super_block *sb,
1536 long nr_to_scan,
1537 int nid)
1538{
1539 return xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1540}
1541
1542static const struct super_operations xfs_super_operations = {
1543 .alloc_inode = xfs_fs_alloc_inode,
1544 .destroy_inode = xfs_fs_destroy_inode,
1545 .evict_inode = xfs_fs_evict_inode,
1546 .drop_inode = xfs_fs_drop_inode,
1547 .put_super = xfs_fs_put_super,
1548 .sync_fs = xfs_fs_sync_fs,
1549 .freeze_fs = xfs_fs_freeze,
1550 .unfreeze_fs = xfs_fs_unfreeze,
1551 .statfs = xfs_fs_statfs,
1552 .remount_fs = xfs_fs_remount,
1553 .show_options = xfs_fs_show_options,
1554 .nr_cached_objects = xfs_fs_nr_cached_objects,
1555 .free_cached_objects = xfs_fs_free_cached_objects,
1556};
1557
1558static struct file_system_type xfs_fs_type = {
1559 .owner = THIS_MODULE,
1560 .name = "xfs",
1561 .mount = xfs_fs_mount,
1562 .kill_sb = kill_block_super,
1563 .fs_flags = FS_REQUIRES_DEV,
1564};
1565MODULE_ALIAS_FS("xfs");
1566
1567STATIC int __init
1568xfs_init_zones(void)
1569{
1570
1571 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1572 if (!xfs_ioend_zone)
1573 goto out;
1574
1575 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1576 xfs_ioend_zone);
1577 if (!xfs_ioend_pool)
1578 goto out_destroy_ioend_zone;
1579
1580 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1581 "xfs_log_ticket");
1582 if (!xfs_log_ticket_zone)
1583 goto out_destroy_ioend_pool;
1584
1585 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1586 "xfs_bmap_free_item");
1587 if (!xfs_bmap_free_item_zone)
1588 goto out_destroy_log_ticket_zone;
1589
1590 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1591 "xfs_btree_cur");
1592 if (!xfs_btree_cur_zone)
1593 goto out_destroy_bmap_free_item_zone;
1594
1595 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1596 "xfs_da_state");
1597 if (!xfs_da_state_zone)
1598 goto out_destroy_btree_cur_zone;
1599
1600 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1601 if (!xfs_ifork_zone)
1602 goto out_destroy_da_state_zone;
1603
1604 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1605 if (!xfs_trans_zone)
1606 goto out_destroy_ifork_zone;
1607
1608 xfs_log_item_desc_zone =
1609 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1610 "xfs_log_item_desc");
1611 if (!xfs_log_item_desc_zone)
1612 goto out_destroy_trans_zone;
1613
1614 /*
1615 * The size of the zone allocated buf log item is the maximum
1616 * size possible under XFS. This wastes a little bit of memory,
1617 * but it is much faster.
1618 */
1619 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1620 "xfs_buf_item");
1621 if (!xfs_buf_item_zone)
1622 goto out_destroy_log_item_desc_zone;
1623
1624 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1625 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1626 sizeof(xfs_extent_t))), "xfs_efd_item");
1627 if (!xfs_efd_zone)
1628 goto out_destroy_buf_item_zone;
1629
1630 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1631 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1632 sizeof(xfs_extent_t))), "xfs_efi_item");
1633 if (!xfs_efi_zone)
1634 goto out_destroy_efd_zone;
1635
1636 xfs_inode_zone =
1637 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1638 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1639 xfs_fs_inode_init_once);
1640 if (!xfs_inode_zone)
1641 goto out_destroy_efi_zone;
1642
1643 xfs_ili_zone =
1644 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1645 KM_ZONE_SPREAD, NULL);
1646 if (!xfs_ili_zone)
1647 goto out_destroy_inode_zone;
1648 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1649 "xfs_icr");
1650 if (!xfs_icreate_zone)
1651 goto out_destroy_ili_zone;
1652
1653 return 0;
1654
1655 out_destroy_ili_zone:
1656 kmem_zone_destroy(xfs_ili_zone);
1657 out_destroy_inode_zone:
1658 kmem_zone_destroy(xfs_inode_zone);
1659 out_destroy_efi_zone:
1660 kmem_zone_destroy(xfs_efi_zone);
1661 out_destroy_efd_zone:
1662 kmem_zone_destroy(xfs_efd_zone);
1663 out_destroy_buf_item_zone:
1664 kmem_zone_destroy(xfs_buf_item_zone);
1665 out_destroy_log_item_desc_zone:
1666 kmem_zone_destroy(xfs_log_item_desc_zone);
1667 out_destroy_trans_zone:
1668 kmem_zone_destroy(xfs_trans_zone);
1669 out_destroy_ifork_zone:
1670 kmem_zone_destroy(xfs_ifork_zone);
1671 out_destroy_da_state_zone:
1672 kmem_zone_destroy(xfs_da_state_zone);
1673 out_destroy_btree_cur_zone:
1674 kmem_zone_destroy(xfs_btree_cur_zone);
1675 out_destroy_bmap_free_item_zone:
1676 kmem_zone_destroy(xfs_bmap_free_item_zone);
1677 out_destroy_log_ticket_zone:
1678 kmem_zone_destroy(xfs_log_ticket_zone);
1679 out_destroy_ioend_pool:
1680 mempool_destroy(xfs_ioend_pool);
1681 out_destroy_ioend_zone:
1682 kmem_zone_destroy(xfs_ioend_zone);
1683 out:
1684 return -ENOMEM;
1685}
1686
1687STATIC void
1688xfs_destroy_zones(void)
1689{
1690 /*
1691 * Make sure all delayed rcu free are flushed before we
1692 * destroy caches.
1693 */
1694 rcu_barrier();
1695 kmem_zone_destroy(xfs_icreate_zone);
1696 kmem_zone_destroy(xfs_ili_zone);
1697 kmem_zone_destroy(xfs_inode_zone);
1698 kmem_zone_destroy(xfs_efi_zone);
1699 kmem_zone_destroy(xfs_efd_zone);
1700 kmem_zone_destroy(xfs_buf_item_zone);
1701 kmem_zone_destroy(xfs_log_item_desc_zone);
1702 kmem_zone_destroy(xfs_trans_zone);
1703 kmem_zone_destroy(xfs_ifork_zone);
1704 kmem_zone_destroy(xfs_da_state_zone);
1705 kmem_zone_destroy(xfs_btree_cur_zone);
1706 kmem_zone_destroy(xfs_bmap_free_item_zone);
1707 kmem_zone_destroy(xfs_log_ticket_zone);
1708 mempool_destroy(xfs_ioend_pool);
1709 kmem_zone_destroy(xfs_ioend_zone);
1710
1711}
1712
1713STATIC int __init
1714xfs_init_workqueues(void)
1715{
1716 /*
1717 * The allocation workqueue can be used in memory reclaim situations
1718 * (writepage path), and parallelism is only limited by the number of
1719 * AGs in all the filesystems mounted. Hence use the default large
1720 * max_active value for this workqueue.
1721 */
1722 xfs_alloc_wq = alloc_workqueue("xfsalloc", WQ_MEM_RECLAIM, 0);
1723 if (!xfs_alloc_wq)
1724 return -ENOMEM;
1725
1726 return 0;
1727}
1728
1729STATIC void
1730xfs_destroy_workqueues(void)
1731{
1732 destroy_workqueue(xfs_alloc_wq);
1733}
1734
1735STATIC int __init
1736init_xfs_fs(void)
1737{
1738 int error;
1739
1740 printk(KERN_INFO XFS_VERSION_STRING " with "
1741 XFS_BUILD_OPTIONS " enabled\n");
1742
1743 xfs_dir_startup();
1744
1745 error = xfs_init_zones();
1746 if (error)
1747 goto out;
1748
1749 error = xfs_init_workqueues();
1750 if (error)
1751 goto out_destroy_zones;
1752
1753 error = xfs_mru_cache_init();
1754 if (error)
1755 goto out_destroy_wq;
1756
1757 error = xfs_filestream_init();
1758 if (error)
1759 goto out_mru_cache_uninit;
1760
1761 error = xfs_buf_init();
1762 if (error)
1763 goto out_filestream_uninit;
1764
1765 error = xfs_init_procfs();
1766 if (error)
1767 goto out_buf_terminate;
1768
1769 error = xfs_sysctl_register();
1770 if (error)
1771 goto out_cleanup_procfs;
1772
1773 error = xfs_qm_init();
1774 if (error)
1775 goto out_sysctl_unregister;
1776
1777 error = register_filesystem(&xfs_fs_type);
1778 if (error)
1779 goto out_qm_exit;
1780 return 0;
1781
1782 out_qm_exit:
1783 xfs_qm_exit();
1784 out_sysctl_unregister:
1785 xfs_sysctl_unregister();
1786 out_cleanup_procfs:
1787 xfs_cleanup_procfs();
1788 out_buf_terminate:
1789 xfs_buf_terminate();
1790 out_filestream_uninit:
1791 xfs_filestream_uninit();
1792 out_mru_cache_uninit:
1793 xfs_mru_cache_uninit();
1794 out_destroy_wq:
1795 xfs_destroy_workqueues();
1796 out_destroy_zones:
1797 xfs_destroy_zones();
1798 out:
1799 return error;
1800}
1801
1802STATIC void __exit
1803exit_xfs_fs(void)
1804{
1805 xfs_qm_exit();
1806 unregister_filesystem(&xfs_fs_type);
1807 xfs_sysctl_unregister();
1808 xfs_cleanup_procfs();
1809 xfs_buf_terminate();
1810 xfs_filestream_uninit();
1811 xfs_mru_cache_uninit();
1812 xfs_destroy_workqueues();
1813 xfs_destroy_zones();
1814}
1815
1816module_init(init_xfs_fs);
1817module_exit(exit_xfs_fs);
1818
1819MODULE_AUTHOR("Silicon Graphics, Inc.");
1820MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1821MODULE_LICENSE("GPL");