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_sb.h"
25#include "xfs_mount.h"
26#include "xfs_da_format.h"
27#include "xfs_inode.h"
28#include "xfs_btree.h"
29#include "xfs_bmap.h"
30#include "xfs_alloc.h"
31#include "xfs_error.h"
32#include "xfs_fsops.h"
33#include "xfs_trans.h"
34#include "xfs_buf_item.h"
35#include "xfs_log.h"
36#include "xfs_log_priv.h"
37#include "xfs_da_btree.h"
38#include "xfs_dir2.h"
39#include "xfs_extfree_item.h"
40#include "xfs_mru_cache.h"
41#include "xfs_inode_item.h"
42#include "xfs_icache.h"
43#include "xfs_trace.h"
44#include "xfs_icreate_item.h"
45#include "xfs_filestream.h"
46#include "xfs_quota.h"
47#include "xfs_sysfs.h"
48#include "xfs_ondisk.h"
49
50#include <linux/namei.h>
51#include <linux/init.h>
52#include <linux/slab.h>
53#include <linux/mount.h>
54#include <linux/mempool.h>
55#include <linux/writeback.h>
56#include <linux/kthread.h>
57#include <linux/freezer.h>
58#include <linux/parser.h>
59
60static const struct super_operations xfs_super_operations;
61static kmem_zone_t *xfs_ioend_zone;
62mempool_t *xfs_ioend_pool;
63
64static struct kset *xfs_kset; /* top-level xfs sysfs dir */
65#ifdef DEBUG
66static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
67#endif
68
69/*
70 * Table driven mount option parser.
71 */
72enum {
73 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
74 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
75 Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
76 Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
77 Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
78 Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
79 Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
80 Opt_uquota, Opt_gquota, Opt_pquota,
81 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
82 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
83};
84
85static const match_table_t tokens = {
86 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */
87 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */
88 {Opt_logdev, "logdev=%s"}, /* log device */
89 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */
90 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */
91 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */
92 {Opt_noalign, "noalign"}, /* turn off stripe alignment */
93 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */
94 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */
95 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */
96 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */
97 {Opt_mtpt, "mtpt"}, /* filesystem mount point */
98 {Opt_grpid, "grpid"}, /* group-ID from parent directory */
99 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */
100 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */
101 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */
102 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
103 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */
104 {Opt_barrier, "barrier"}, /* use writer barriers for log write and
105 * unwritten extent conversion */
106 {Opt_nobarrier, "nobarrier"}, /* .. disable */
107 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */
108 {Opt_inode32, "inode32"}, /* inode allocation limited to
109 * XFS_MAXINUMBER_32 */
110 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */
111 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */
112 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */
113 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes
114 * in stat(). */
115 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */
116 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */
117 {Opt_filestreams,"filestreams"},/* use filestreams allocator */
118 {Opt_quota, "quota"}, /* disk quotas (user) */
119 {Opt_noquota, "noquota"}, /* no quotas */
120 {Opt_usrquota, "usrquota"}, /* user quota enabled */
121 {Opt_grpquota, "grpquota"}, /* group quota enabled */
122 {Opt_prjquota, "prjquota"}, /* project quota enabled */
123 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */
124 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */
125 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */
126 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
127 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
128 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
129 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
130 {Opt_discard, "discard"}, /* Discard unused blocks */
131 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */
132
133 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */
134 {Opt_err, NULL},
135};
136
137
138STATIC int
139suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
140{
141 int last, shift_left_factor = 0, _res;
142 char *value;
143 int ret = 0;
144
145 value = match_strdup(s);
146 if (!value)
147 return -ENOMEM;
148
149 last = strlen(value) - 1;
150 if (value[last] == 'K' || value[last] == 'k') {
151 shift_left_factor = 10;
152 value[last] = '\0';
153 }
154 if (value[last] == 'M' || value[last] == 'm') {
155 shift_left_factor = 20;
156 value[last] = '\0';
157 }
158 if (value[last] == 'G' || value[last] == 'g') {
159 shift_left_factor = 30;
160 value[last] = '\0';
161 }
162
163 if (kstrtoint(value, base, &_res))
164 ret = -EINVAL;
165 kfree(value);
166 *res = _res << shift_left_factor;
167 return ret;
168}
169
170/*
171 * This function fills in xfs_mount_t fields based on mount args.
172 * Note: the superblock has _not_ yet been read in.
173 *
174 * Note that this function leaks the various device name allocations on
175 * failure. The caller takes care of them.
176 *
177 * *sb is const because this is also used to test options on the remount
178 * path, and we don't want this to have any side effects at remount time.
179 * Today this function does not change *sb, but just to future-proof...
180 */
181STATIC int
182xfs_parseargs(
183 struct xfs_mount *mp,
184 char *options)
185{
186 const struct super_block *sb = mp->m_super;
187 char *p;
188 substring_t args[MAX_OPT_ARGS];
189 int dsunit = 0;
190 int dswidth = 0;
191 int iosize = 0;
192 __uint8_t iosizelog = 0;
193
194 /*
195 * set up the mount name first so all the errors will refer to the
196 * correct device.
197 */
198 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
199 if (!mp->m_fsname)
200 return -ENOMEM;
201 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
202
203 /*
204 * Copy binary VFS mount flags we are interested in.
205 */
206 if (sb->s_flags & MS_RDONLY)
207 mp->m_flags |= XFS_MOUNT_RDONLY;
208 if (sb->s_flags & MS_DIRSYNC)
209 mp->m_flags |= XFS_MOUNT_DIRSYNC;
210 if (sb->s_flags & MS_SYNCHRONOUS)
211 mp->m_flags |= XFS_MOUNT_WSYNC;
212
213 /*
214 * Set some default flags that could be cleared by the mount option
215 * parsing.
216 */
217 mp->m_flags |= XFS_MOUNT_BARRIER;
218 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
219
220 /*
221 * These can be overridden by the mount option parsing.
222 */
223 mp->m_logbufs = -1;
224 mp->m_logbsize = -1;
225
226 if (!options)
227 goto done;
228
229 while ((p = strsep(&options, ",")) != NULL) {
230 int token;
231
232 if (!*p)
233 continue;
234
235 token = match_token(p, tokens, args);
236 switch (token) {
237 case Opt_logbufs:
238 if (match_int(args, &mp->m_logbufs))
239 return -EINVAL;
240 break;
241 case Opt_logbsize:
242 if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
243 return -EINVAL;
244 break;
245 case Opt_logdev:
246 mp->m_logname = match_strdup(args);
247 if (!mp->m_logname)
248 return -ENOMEM;
249 break;
250 case Opt_mtpt:
251 xfs_warn(mp, "%s option not allowed on this system", p);
252 return -EINVAL;
253 case Opt_rtdev:
254 mp->m_rtname = match_strdup(args);
255 if (!mp->m_rtname)
256 return -ENOMEM;
257 break;
258 case Opt_allocsize:
259 case Opt_biosize:
260 if (suffix_kstrtoint(args, 10, &iosize))
261 return -EINVAL;
262 iosizelog = ffs(iosize) - 1;
263 break;
264 case Opt_grpid:
265 case Opt_bsdgroups:
266 mp->m_flags |= XFS_MOUNT_GRPID;
267 break;
268 case Opt_nogrpid:
269 case Opt_sysvgroups:
270 mp->m_flags &= ~XFS_MOUNT_GRPID;
271 break;
272 case Opt_wsync:
273 mp->m_flags |= XFS_MOUNT_WSYNC;
274 break;
275 case Opt_norecovery:
276 mp->m_flags |= XFS_MOUNT_NORECOVERY;
277 break;
278 case Opt_noalign:
279 mp->m_flags |= XFS_MOUNT_NOALIGN;
280 break;
281 case Opt_swalloc:
282 mp->m_flags |= XFS_MOUNT_SWALLOC;
283 break;
284 case Opt_sunit:
285 if (match_int(args, &dsunit))
286 return -EINVAL;
287 break;
288 case Opt_swidth:
289 if (match_int(args, &dswidth))
290 return -EINVAL;
291 break;
292 case Opt_inode32:
293 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
294 break;
295 case Opt_inode64:
296 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
297 break;
298 case Opt_nouuid:
299 mp->m_flags |= XFS_MOUNT_NOUUID;
300 break;
301 case Opt_barrier:
302 mp->m_flags |= XFS_MOUNT_BARRIER;
303 break;
304 case Opt_nobarrier:
305 mp->m_flags &= ~XFS_MOUNT_BARRIER;
306 break;
307 case Opt_ikeep:
308 mp->m_flags |= XFS_MOUNT_IKEEP;
309 break;
310 case Opt_noikeep:
311 mp->m_flags &= ~XFS_MOUNT_IKEEP;
312 break;
313 case Opt_largeio:
314 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
315 break;
316 case Opt_nolargeio:
317 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
318 break;
319 case Opt_attr2:
320 mp->m_flags |= XFS_MOUNT_ATTR2;
321 break;
322 case Opt_noattr2:
323 mp->m_flags &= ~XFS_MOUNT_ATTR2;
324 mp->m_flags |= XFS_MOUNT_NOATTR2;
325 break;
326 case Opt_filestreams:
327 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
328 break;
329 case Opt_noquota:
330 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
331 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
332 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
333 break;
334 case Opt_quota:
335 case Opt_uquota:
336 case Opt_usrquota:
337 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
338 XFS_UQUOTA_ENFD);
339 break;
340 case Opt_qnoenforce:
341 case Opt_uqnoenforce:
342 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
343 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
344 break;
345 case Opt_pquota:
346 case Opt_prjquota:
347 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
348 XFS_PQUOTA_ENFD);
349 break;
350 case Opt_pqnoenforce:
351 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
352 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
353 case Opt_gquota:
354 case Opt_grpquota:
355 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
356 XFS_GQUOTA_ENFD);
357 break;
358 case Opt_gqnoenforce:
359 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
360 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
361 break;
362 case Opt_discard:
363 mp->m_flags |= XFS_MOUNT_DISCARD;
364 break;
365 case Opt_nodiscard:
366 mp->m_flags &= ~XFS_MOUNT_DISCARD;
367 break;
368#ifdef CONFIG_FS_DAX
369 case Opt_dax:
370 mp->m_flags |= XFS_MOUNT_DAX;
371 break;
372#endif
373 default:
374 xfs_warn(mp, "unknown mount option [%s].", p);
375 return -EINVAL;
376 }
377 }
378
379 /*
380 * no recovery flag requires a read-only mount
381 */
382 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
383 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
384 xfs_warn(mp, "no-recovery mounts must be read-only.");
385 return -EINVAL;
386 }
387
388 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
389 xfs_warn(mp,
390 "sunit and swidth options incompatible with the noalign option");
391 return -EINVAL;
392 }
393
394#ifndef CONFIG_XFS_QUOTA
395 if (XFS_IS_QUOTA_RUNNING(mp)) {
396 xfs_warn(mp, "quota support not available in this kernel.");
397 return -EINVAL;
398 }
399#endif
400
401 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
402 xfs_warn(mp, "sunit and swidth must be specified together");
403 return -EINVAL;
404 }
405
406 if (dsunit && (dswidth % dsunit != 0)) {
407 xfs_warn(mp,
408 "stripe width (%d) must be a multiple of the stripe unit (%d)",
409 dswidth, dsunit);
410 return -EINVAL;
411 }
412
413done:
414 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
415 /*
416 * At this point the superblock has not been read
417 * in, therefore we do not know the block size.
418 * Before the mount call ends we will convert
419 * these to FSBs.
420 */
421 mp->m_dalign = dsunit;
422 mp->m_swidth = dswidth;
423 }
424
425 if (mp->m_logbufs != -1 &&
426 mp->m_logbufs != 0 &&
427 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
428 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
429 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
430 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
431 return -EINVAL;
432 }
433 if (mp->m_logbsize != -1 &&
434 mp->m_logbsize != 0 &&
435 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
436 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
437 !is_power_of_2(mp->m_logbsize))) {
438 xfs_warn(mp,
439 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
440 mp->m_logbsize);
441 return -EINVAL;
442 }
443
444 if (iosizelog) {
445 if (iosizelog > XFS_MAX_IO_LOG ||
446 iosizelog < XFS_MIN_IO_LOG) {
447 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
448 iosizelog, XFS_MIN_IO_LOG,
449 XFS_MAX_IO_LOG);
450 return -EINVAL;
451 }
452
453 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
454 mp->m_readio_log = iosizelog;
455 mp->m_writeio_log = iosizelog;
456 }
457
458 return 0;
459}
460
461struct proc_xfs_info {
462 uint64_t flag;
463 char *str;
464};
465
466STATIC int
467xfs_showargs(
468 struct xfs_mount *mp,
469 struct seq_file *m)
470{
471 static struct proc_xfs_info xfs_info_set[] = {
472 /* the few simple ones we can get from the mount struct */
473 { XFS_MOUNT_IKEEP, ",ikeep" },
474 { XFS_MOUNT_WSYNC, ",wsync" },
475 { XFS_MOUNT_NOALIGN, ",noalign" },
476 { XFS_MOUNT_SWALLOC, ",swalloc" },
477 { XFS_MOUNT_NOUUID, ",nouuid" },
478 { XFS_MOUNT_NORECOVERY, ",norecovery" },
479 { XFS_MOUNT_ATTR2, ",attr2" },
480 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
481 { XFS_MOUNT_GRPID, ",grpid" },
482 { XFS_MOUNT_DISCARD, ",discard" },
483 { XFS_MOUNT_SMALL_INUMS, ",inode32" },
484 { XFS_MOUNT_DAX, ",dax" },
485 { 0, NULL }
486 };
487 static struct proc_xfs_info xfs_info_unset[] = {
488 /* the few simple ones we can get from the mount struct */
489 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" },
490 { XFS_MOUNT_BARRIER, ",nobarrier" },
491 { XFS_MOUNT_SMALL_INUMS, ",inode64" },
492 { 0, NULL }
493 };
494 struct proc_xfs_info *xfs_infop;
495
496 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
497 if (mp->m_flags & xfs_infop->flag)
498 seq_puts(m, xfs_infop->str);
499 }
500 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
501 if (!(mp->m_flags & xfs_infop->flag))
502 seq_puts(m, xfs_infop->str);
503 }
504
505 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
506 seq_printf(m, ",allocsize=%dk",
507 (int)(1 << mp->m_writeio_log) >> 10);
508
509 if (mp->m_logbufs > 0)
510 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
511 if (mp->m_logbsize > 0)
512 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
513
514 if (mp->m_logname)
515 seq_show_option(m, "logdev", mp->m_logname);
516 if (mp->m_rtname)
517 seq_show_option(m, "rtdev", mp->m_rtname);
518
519 if (mp->m_dalign > 0)
520 seq_printf(m, ",sunit=%d",
521 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
522 if (mp->m_swidth > 0)
523 seq_printf(m, ",swidth=%d",
524 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
525
526 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
527 seq_puts(m, ",usrquota");
528 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
529 seq_puts(m, ",uqnoenforce");
530
531 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
532 if (mp->m_qflags & XFS_PQUOTA_ENFD)
533 seq_puts(m, ",prjquota");
534 else
535 seq_puts(m, ",pqnoenforce");
536 }
537 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
538 if (mp->m_qflags & XFS_GQUOTA_ENFD)
539 seq_puts(m, ",grpquota");
540 else
541 seq_puts(m, ",gqnoenforce");
542 }
543
544 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
545 seq_puts(m, ",noquota");
546
547 return 0;
548}
549__uint64_t
550xfs_max_file_offset(
551 unsigned int blockshift)
552{
553 unsigned int pagefactor = 1;
554 unsigned int bitshift = BITS_PER_LONG - 1;
555
556 /* Figure out maximum filesize, on Linux this can depend on
557 * the filesystem blocksize (on 32 bit platforms).
558 * __block_write_begin does this in an [unsigned] long...
559 * page->index << (PAGE_SHIFT - bbits)
560 * So, for page sized blocks (4K on 32 bit platforms),
561 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
562 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
563 * but for smaller blocksizes it is less (bbits = log2 bsize).
564 * Note1: get_block_t takes a long (implicit cast from above)
565 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
566 * can optionally convert the [unsigned] long from above into
567 * an [unsigned] long long.
568 */
569
570#if BITS_PER_LONG == 32
571# if defined(CONFIG_LBDAF)
572 ASSERT(sizeof(sector_t) == 8);
573 pagefactor = PAGE_SIZE;
574 bitshift = BITS_PER_LONG;
575# else
576 pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
577# endif
578#endif
579
580 return (((__uint64_t)pagefactor) << bitshift) - 1;
581}
582
583/*
584 * Set parameters for inode allocation heuristics, taking into account
585 * filesystem size and inode32/inode64 mount options; i.e. specifically
586 * whether or not XFS_MOUNT_SMALL_INUMS is set.
587 *
588 * Inode allocation patterns are altered only if inode32 is requested
589 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
590 * If altered, XFS_MOUNT_32BITINODES is set as well.
591 *
592 * An agcount independent of that in the mount structure is provided
593 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
594 * to the potentially higher ag count.
595 *
596 * Returns the maximum AG index which may contain inodes.
597 */
598xfs_agnumber_t
599xfs_set_inode_alloc(
600 struct xfs_mount *mp,
601 xfs_agnumber_t agcount)
602{
603 xfs_agnumber_t index;
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;
608 xfs_ino_t ino;
609
610 /*
611 * Calculate how much should be reserved for inodes to meet
612 * the max inode percentage. Used only for inode32.
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 = agcount;
624 }
625
626 /* Get the last possible inode in the filesystem */
627 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
628 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
629
630 /*
631 * If user asked for no more than 32-bit inodes, and the fs is
632 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
633 * the allocator to accommodate the request.
634 */
635 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
636 mp->m_flags |= XFS_MOUNT_32BITINODES;
637 else
638 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
639
640 for (index = 0; index < agcount; index++) {
641 struct xfs_perag *pag;
642
643 ino = XFS_AGINO_TO_INO(mp, index, agino);
644
645 pag = xfs_perag_get(mp, index);
646
647 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
648 if (ino > XFS_MAXINUMBER_32) {
649 pag->pagi_inodeok = 0;
650 pag->pagf_metadata = 0;
651 } else {
652 pag->pagi_inodeok = 1;
653 maxagi++;
654 if (index < max_metadata)
655 pag->pagf_metadata = 1;
656 else
657 pag->pagf_metadata = 0;
658 }
659 } else {
660 pag->pagi_inodeok = 1;
661 pag->pagf_metadata = 0;
662 }
663
664 xfs_perag_put(pag);
665 }
666
667 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
668}
669
670STATIC int
671xfs_blkdev_get(
672 xfs_mount_t *mp,
673 const char *name,
674 struct block_device **bdevp)
675{
676 int error = 0;
677
678 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
679 mp);
680 if (IS_ERR(*bdevp)) {
681 error = PTR_ERR(*bdevp);
682 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
683 }
684
685 return error;
686}
687
688STATIC void
689xfs_blkdev_put(
690 struct block_device *bdev)
691{
692 if (bdev)
693 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
694}
695
696void
697xfs_blkdev_issue_flush(
698 xfs_buftarg_t *buftarg)
699{
700 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
701}
702
703STATIC void
704xfs_close_devices(
705 struct xfs_mount *mp)
706{
707 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
708 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
709 xfs_free_buftarg(mp, mp->m_logdev_targp);
710 xfs_blkdev_put(logdev);
711 }
712 if (mp->m_rtdev_targp) {
713 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
714 xfs_free_buftarg(mp, mp->m_rtdev_targp);
715 xfs_blkdev_put(rtdev);
716 }
717 xfs_free_buftarg(mp, mp->m_ddev_targp);
718}
719
720/*
721 * The file system configurations are:
722 * (1) device (partition) with data and internal log
723 * (2) logical volume with data and log subvolumes.
724 * (3) logical volume with data, log, and realtime subvolumes.
725 *
726 * We only have to handle opening the log and realtime volumes here if
727 * they are present. The data subvolume has already been opened by
728 * get_sb_bdev() and is stored in sb->s_bdev.
729 */
730STATIC int
731xfs_open_devices(
732 struct xfs_mount *mp)
733{
734 struct block_device *ddev = mp->m_super->s_bdev;
735 struct block_device *logdev = NULL, *rtdev = NULL;
736 int error;
737
738 /*
739 * Open real time and log devices - order is important.
740 */
741 if (mp->m_logname) {
742 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
743 if (error)
744 goto out;
745 }
746
747 if (mp->m_rtname) {
748 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
749 if (error)
750 goto out_close_logdev;
751
752 if (rtdev == ddev || rtdev == logdev) {
753 xfs_warn(mp,
754 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
755 error = -EINVAL;
756 goto out_close_rtdev;
757 }
758 }
759
760 /*
761 * Setup xfs_mount buffer target pointers
762 */
763 error = -ENOMEM;
764 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
765 if (!mp->m_ddev_targp)
766 goto out_close_rtdev;
767
768 if (rtdev) {
769 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
770 if (!mp->m_rtdev_targp)
771 goto out_free_ddev_targ;
772 }
773
774 if (logdev && logdev != ddev) {
775 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
776 if (!mp->m_logdev_targp)
777 goto out_free_rtdev_targ;
778 } else {
779 mp->m_logdev_targp = mp->m_ddev_targp;
780 }
781
782 return 0;
783
784 out_free_rtdev_targ:
785 if (mp->m_rtdev_targp)
786 xfs_free_buftarg(mp, mp->m_rtdev_targp);
787 out_free_ddev_targ:
788 xfs_free_buftarg(mp, mp->m_ddev_targp);
789 out_close_rtdev:
790 xfs_blkdev_put(rtdev);
791 out_close_logdev:
792 if (logdev && logdev != ddev)
793 xfs_blkdev_put(logdev);
794 out:
795 return error;
796}
797
798/*
799 * Setup xfs_mount buffer target pointers based on superblock
800 */
801STATIC int
802xfs_setup_devices(
803 struct xfs_mount *mp)
804{
805 int error;
806
807 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
808 if (error)
809 return error;
810
811 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
812 unsigned int log_sector_size = BBSIZE;
813
814 if (xfs_sb_version_hassector(&mp->m_sb))
815 log_sector_size = mp->m_sb.sb_logsectsize;
816 error = xfs_setsize_buftarg(mp->m_logdev_targp,
817 log_sector_size);
818 if (error)
819 return error;
820 }
821 if (mp->m_rtdev_targp) {
822 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
823 mp->m_sb.sb_sectsize);
824 if (error)
825 return error;
826 }
827
828 return 0;
829}
830
831STATIC int
832xfs_init_mount_workqueues(
833 struct xfs_mount *mp)
834{
835 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
836 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
837 if (!mp->m_buf_workqueue)
838 goto out;
839
840 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
841 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
842 if (!mp->m_data_workqueue)
843 goto out_destroy_buf;
844
845 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
846 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
847 if (!mp->m_unwritten_workqueue)
848 goto out_destroy_data_iodone_queue;
849
850 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
851 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
852 if (!mp->m_cil_workqueue)
853 goto out_destroy_unwritten;
854
855 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
856 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
857 if (!mp->m_reclaim_workqueue)
858 goto out_destroy_cil;
859
860 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
861 WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
862 mp->m_fsname);
863 if (!mp->m_log_workqueue)
864 goto out_destroy_reclaim;
865
866 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
867 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
868 if (!mp->m_eofblocks_workqueue)
869 goto out_destroy_log;
870
871 return 0;
872
873out_destroy_log:
874 destroy_workqueue(mp->m_log_workqueue);
875out_destroy_reclaim:
876 destroy_workqueue(mp->m_reclaim_workqueue);
877out_destroy_cil:
878 destroy_workqueue(mp->m_cil_workqueue);
879out_destroy_unwritten:
880 destroy_workqueue(mp->m_unwritten_workqueue);
881out_destroy_data_iodone_queue:
882 destroy_workqueue(mp->m_data_workqueue);
883out_destroy_buf:
884 destroy_workqueue(mp->m_buf_workqueue);
885out:
886 return -ENOMEM;
887}
888
889STATIC void
890xfs_destroy_mount_workqueues(
891 struct xfs_mount *mp)
892{
893 destroy_workqueue(mp->m_eofblocks_workqueue);
894 destroy_workqueue(mp->m_log_workqueue);
895 destroy_workqueue(mp->m_reclaim_workqueue);
896 destroy_workqueue(mp->m_cil_workqueue);
897 destroy_workqueue(mp->m_data_workqueue);
898 destroy_workqueue(mp->m_unwritten_workqueue);
899 destroy_workqueue(mp->m_buf_workqueue);
900}
901
902/*
903 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
904 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
905 * for IO to complete so that we effectively throttle multiple callers to the
906 * rate at which IO is completing.
907 */
908void
909xfs_flush_inodes(
910 struct xfs_mount *mp)
911{
912 struct super_block *sb = mp->m_super;
913
914 if (down_read_trylock(&sb->s_umount)) {
915 sync_inodes_sb(sb);
916 up_read(&sb->s_umount);
917 }
918}
919
920/* Catch misguided souls that try to use this interface on XFS */
921STATIC struct inode *
922xfs_fs_alloc_inode(
923 struct super_block *sb)
924{
925 BUG();
926 return NULL;
927}
928
929/*
930 * Now that the generic code is guaranteed not to be accessing
931 * the linux inode, we can reclaim the inode.
932 */
933STATIC void
934xfs_fs_destroy_inode(
935 struct inode *inode)
936{
937 struct xfs_inode *ip = XFS_I(inode);
938
939 trace_xfs_destroy_inode(ip);
940
941 XFS_STATS_INC(ip->i_mount, vn_reclaim);
942
943 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
944
945 /*
946 * We should never get here with one of the reclaim flags already set.
947 */
948 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
949 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
950
951 /*
952 * We always use background reclaim here because even if the
953 * inode is clean, it still may be under IO and hence we have
954 * to take the flush lock. The background reclaim path handles
955 * this more efficiently than we can here, so simply let background
956 * reclaim tear down all inodes.
957 */
958 xfs_inode_set_reclaim_tag(ip);
959}
960
961/*
962 * Slab object creation initialisation for the XFS inode.
963 * This covers only the idempotent fields in the XFS inode;
964 * all other fields need to be initialised on allocation
965 * from the slab. This avoids the need to repeatedly initialise
966 * fields in the xfs inode that left in the initialise state
967 * when freeing the inode.
968 */
969STATIC void
970xfs_fs_inode_init_once(
971 void *inode)
972{
973 struct xfs_inode *ip = inode;
974
975 memset(ip, 0, sizeof(struct xfs_inode));
976
977 /* vfs inode */
978 inode_init_once(VFS_I(ip));
979
980 /* xfs inode */
981 atomic_set(&ip->i_pincount, 0);
982 spin_lock_init(&ip->i_flags_lock);
983
984 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
985 "xfsino", ip->i_ino);
986 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
987 "xfsino", ip->i_ino);
988}
989
990STATIC void
991xfs_fs_evict_inode(
992 struct inode *inode)
993{
994 xfs_inode_t *ip = XFS_I(inode);
995
996 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
997
998 trace_xfs_evict_inode(ip);
999
1000 truncate_inode_pages_final(&inode->i_data);
1001 clear_inode(inode);
1002 XFS_STATS_INC(ip->i_mount, vn_rele);
1003 XFS_STATS_INC(ip->i_mount, vn_remove);
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 int
1034xfs_fs_sync_fs(
1035 struct super_block *sb,
1036 int wait)
1037{
1038 struct xfs_mount *mp = XFS_M(sb);
1039
1040 /*
1041 * Doing anything during the async pass would be counterproductive.
1042 */
1043 if (!wait)
1044 return 0;
1045
1046 xfs_log_force(mp, XFS_LOG_SYNC);
1047 if (laptop_mode) {
1048 /*
1049 * The disk must be active because we're syncing.
1050 * We schedule log work now (now that the disk is
1051 * active) instead of later (when it might not be).
1052 */
1053 flush_delayed_work(&mp->m_log->l_work);
1054 }
1055
1056 return 0;
1057}
1058
1059STATIC int
1060xfs_fs_statfs(
1061 struct dentry *dentry,
1062 struct kstatfs *statp)
1063{
1064 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1065 xfs_sb_t *sbp = &mp->m_sb;
1066 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1067 __uint64_t fakeinos, id;
1068 __uint64_t icount;
1069 __uint64_t ifree;
1070 __uint64_t fdblocks;
1071 xfs_extlen_t lsize;
1072 __int64_t ffree;
1073
1074 statp->f_type = XFS_SB_MAGIC;
1075 statp->f_namelen = MAXNAMELEN - 1;
1076
1077 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1078 statp->f_fsid.val[0] = (u32)id;
1079 statp->f_fsid.val[1] = (u32)(id >> 32);
1080
1081 icount = percpu_counter_sum(&mp->m_icount);
1082 ifree = percpu_counter_sum(&mp->m_ifree);
1083 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1084
1085 spin_lock(&mp->m_sb_lock);
1086 statp->f_bsize = sbp->sb_blocksize;
1087 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1088 statp->f_blocks = sbp->sb_dblocks - lsize;
1089 spin_unlock(&mp->m_sb_lock);
1090
1091 statp->f_bfree = fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1092 statp->f_bavail = statp->f_bfree;
1093
1094 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1095 statp->f_files = MIN(icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1096 if (mp->m_maxicount)
1097 statp->f_files = min_t(typeof(statp->f_files),
1098 statp->f_files,
1099 mp->m_maxicount);
1100
1101 /* If sb_icount overshot maxicount, report actual allocation */
1102 statp->f_files = max_t(typeof(statp->f_files),
1103 statp->f_files,
1104 sbp->sb_icount);
1105
1106 /* make sure statp->f_ffree does not underflow */
1107 ffree = statp->f_files - (icount - ifree);
1108 statp->f_ffree = max_t(__int64_t, ffree, 0);
1109
1110
1111 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1112 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1113 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1114 xfs_qm_statvfs(ip, statp);
1115 return 0;
1116}
1117
1118STATIC void
1119xfs_save_resvblks(struct xfs_mount *mp)
1120{
1121 __uint64_t resblks = 0;
1122
1123 mp->m_resblks_save = mp->m_resblks;
1124 xfs_reserve_blocks(mp, &resblks, NULL);
1125}
1126
1127STATIC void
1128xfs_restore_resvblks(struct xfs_mount *mp)
1129{
1130 __uint64_t resblks;
1131
1132 if (mp->m_resblks_save) {
1133 resblks = mp->m_resblks_save;
1134 mp->m_resblks_save = 0;
1135 } else
1136 resblks = xfs_default_resblks(mp);
1137
1138 xfs_reserve_blocks(mp, &resblks, NULL);
1139}
1140
1141/*
1142 * Trigger writeback of all the dirty metadata in the file system.
1143 *
1144 * This ensures that the metadata is written to their location on disk rather
1145 * than just existing in transactions in the log. This means after a quiesce
1146 * there is no log replay required to write the inodes to disk - this is the
1147 * primary difference between a sync and a quiesce.
1148 *
1149 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1150 * it is started again when appropriate.
1151 */
1152static void
1153xfs_quiesce_attr(
1154 struct xfs_mount *mp)
1155{
1156 int error = 0;
1157
1158 /* wait for all modifications to complete */
1159 while (atomic_read(&mp->m_active_trans) > 0)
1160 delay(100);
1161
1162 /* force the log to unpin objects from the now complete transactions */
1163 xfs_log_force(mp, XFS_LOG_SYNC);
1164
1165 /* reclaim inodes to do any IO before the freeze completes */
1166 xfs_reclaim_inodes(mp, 0);
1167 xfs_reclaim_inodes(mp, SYNC_WAIT);
1168
1169 /* Push the superblock and write an unmount record */
1170 error = xfs_log_sbcount(mp);
1171 if (error)
1172 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1173 "Frozen image may not be consistent.");
1174 /*
1175 * Just warn here till VFS can correctly support
1176 * read-only remount without racing.
1177 */
1178 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1179
1180 xfs_log_quiesce(mp);
1181}
1182
1183STATIC int
1184xfs_test_remount_options(
1185 struct super_block *sb,
1186 struct xfs_mount *mp,
1187 char *options)
1188{
1189 int error = 0;
1190 struct xfs_mount *tmp_mp;
1191
1192 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1193 if (!tmp_mp)
1194 return -ENOMEM;
1195
1196 tmp_mp->m_super = sb;
1197 error = xfs_parseargs(tmp_mp, options);
1198 xfs_free_fsname(tmp_mp);
1199 kfree(tmp_mp);
1200
1201 return error;
1202}
1203
1204STATIC int
1205xfs_fs_remount(
1206 struct super_block *sb,
1207 int *flags,
1208 char *options)
1209{
1210 struct xfs_mount *mp = XFS_M(sb);
1211 xfs_sb_t *sbp = &mp->m_sb;
1212 substring_t args[MAX_OPT_ARGS];
1213 char *p;
1214 int error;
1215
1216 /* First, check for complete junk; i.e. invalid options */
1217 error = xfs_test_remount_options(sb, mp, options);
1218 if (error)
1219 return error;
1220
1221 sync_filesystem(sb);
1222 while ((p = strsep(&options, ",")) != NULL) {
1223 int token;
1224
1225 if (!*p)
1226 continue;
1227
1228 token = match_token(p, tokens, args);
1229 switch (token) {
1230 case Opt_barrier:
1231 mp->m_flags |= XFS_MOUNT_BARRIER;
1232 break;
1233 case Opt_nobarrier:
1234 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1235 break;
1236 case Opt_inode64:
1237 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1238 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1239 break;
1240 case Opt_inode32:
1241 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1242 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1243 break;
1244 default:
1245 /*
1246 * Logically we would return an error here to prevent
1247 * users from believing they might have changed
1248 * mount options using remount which can't be changed.
1249 *
1250 * But unfortunately mount(8) adds all options from
1251 * mtab and fstab to the mount arguments in some cases
1252 * so we can't blindly reject options, but have to
1253 * check for each specified option if it actually
1254 * differs from the currently set option and only
1255 * reject it if that's the case.
1256 *
1257 * Until that is implemented we return success for
1258 * every remount request, and silently ignore all
1259 * options that we can't actually change.
1260 */
1261#if 0
1262 xfs_info(mp,
1263 "mount option \"%s\" not supported for remount", p);
1264 return -EINVAL;
1265#else
1266 break;
1267#endif
1268 }
1269 }
1270
1271 /* ro -> rw */
1272 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1273 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1274 xfs_warn(mp,
1275 "ro->rw transition prohibited on norecovery mount");
1276 return -EINVAL;
1277 }
1278
1279 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1280
1281 /*
1282 * If this is the first remount to writeable state we
1283 * might have some superblock changes to update.
1284 */
1285 if (mp->m_update_sb) {
1286 error = xfs_sync_sb(mp, false);
1287 if (error) {
1288 xfs_warn(mp, "failed to write sb changes");
1289 return error;
1290 }
1291 mp->m_update_sb = false;
1292 }
1293
1294 /*
1295 * Fill out the reserve pool if it is empty. Use the stashed
1296 * value if it is non-zero, otherwise go with the default.
1297 */
1298 xfs_restore_resvblks(mp);
1299 xfs_log_work_queue(mp);
1300 }
1301
1302 /* rw -> ro */
1303 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1304 /*
1305 * Before we sync the metadata, we need to free up the reserve
1306 * block pool so that the used block count in the superblock on
1307 * disk is correct at the end of the remount. Stash the current
1308 * reserve pool size so that if we get remounted rw, we can
1309 * return it to the same size.
1310 */
1311 xfs_save_resvblks(mp);
1312 xfs_quiesce_attr(mp);
1313 mp->m_flags |= XFS_MOUNT_RDONLY;
1314 }
1315
1316 return 0;
1317}
1318
1319/*
1320 * Second stage of a freeze. The data is already frozen so we only
1321 * need to take care of the metadata. Once that's done sync the superblock
1322 * to the log to dirty it in case of a crash while frozen. This ensures that we
1323 * will recover the unlinked inode lists on the next mount.
1324 */
1325STATIC int
1326xfs_fs_freeze(
1327 struct super_block *sb)
1328{
1329 struct xfs_mount *mp = XFS_M(sb);
1330
1331 xfs_save_resvblks(mp);
1332 xfs_quiesce_attr(mp);
1333 return xfs_sync_sb(mp, true);
1334}
1335
1336STATIC int
1337xfs_fs_unfreeze(
1338 struct super_block *sb)
1339{
1340 struct xfs_mount *mp = XFS_M(sb);
1341
1342 xfs_restore_resvblks(mp);
1343 xfs_log_work_queue(mp);
1344 return 0;
1345}
1346
1347STATIC int
1348xfs_fs_show_options(
1349 struct seq_file *m,
1350 struct dentry *root)
1351{
1352 return xfs_showargs(XFS_M(root->d_sb), m);
1353}
1354
1355/*
1356 * This function fills in xfs_mount_t fields based on mount args.
1357 * Note: the superblock _has_ now been read in.
1358 */
1359STATIC int
1360xfs_finish_flags(
1361 struct xfs_mount *mp)
1362{
1363 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1364
1365 /* Fail a mount where the logbuf is smaller than the log stripe */
1366 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1367 if (mp->m_logbsize <= 0 &&
1368 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1369 mp->m_logbsize = mp->m_sb.sb_logsunit;
1370 } else if (mp->m_logbsize > 0 &&
1371 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1372 xfs_warn(mp,
1373 "logbuf size must be greater than or equal to log stripe size");
1374 return -EINVAL;
1375 }
1376 } else {
1377 /* Fail a mount if the logbuf is larger than 32K */
1378 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1379 xfs_warn(mp,
1380 "logbuf size for version 1 logs must be 16K or 32K");
1381 return -EINVAL;
1382 }
1383 }
1384
1385 /*
1386 * V5 filesystems always use attr2 format for attributes.
1387 */
1388 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1389 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1390 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1391 "attr2 is always enabled for V5 filesystems.");
1392 return -EINVAL;
1393 }
1394
1395 /*
1396 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1397 * told by noattr2 to turn it off
1398 */
1399 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1400 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1401 mp->m_flags |= XFS_MOUNT_ATTR2;
1402
1403 /*
1404 * prohibit r/w mounts of read-only filesystems
1405 */
1406 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1407 xfs_warn(mp,
1408 "cannot mount a read-only filesystem as read-write");
1409 return -EROFS;
1410 }
1411
1412 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1413 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1414 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1415 xfs_warn(mp,
1416 "Super block does not support project and group quota together");
1417 return -EINVAL;
1418 }
1419
1420 return 0;
1421}
1422
1423static int
1424xfs_init_percpu_counters(
1425 struct xfs_mount *mp)
1426{
1427 int error;
1428
1429 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1430 if (error)
1431 return -ENOMEM;
1432
1433 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1434 if (error)
1435 goto free_icount;
1436
1437 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1438 if (error)
1439 goto free_ifree;
1440
1441 return 0;
1442
1443free_ifree:
1444 percpu_counter_destroy(&mp->m_ifree);
1445free_icount:
1446 percpu_counter_destroy(&mp->m_icount);
1447 return -ENOMEM;
1448}
1449
1450void
1451xfs_reinit_percpu_counters(
1452 struct xfs_mount *mp)
1453{
1454 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1455 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1456 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1457}
1458
1459static void
1460xfs_destroy_percpu_counters(
1461 struct xfs_mount *mp)
1462{
1463 percpu_counter_destroy(&mp->m_icount);
1464 percpu_counter_destroy(&mp->m_ifree);
1465 percpu_counter_destroy(&mp->m_fdblocks);
1466}
1467
1468STATIC int
1469xfs_fs_fill_super(
1470 struct super_block *sb,
1471 void *data,
1472 int silent)
1473{
1474 struct inode *root;
1475 struct xfs_mount *mp = NULL;
1476 int flags = 0, error = -ENOMEM;
1477
1478 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1479 if (!mp)
1480 goto out;
1481
1482 spin_lock_init(&mp->m_sb_lock);
1483 mutex_init(&mp->m_growlock);
1484 atomic_set(&mp->m_active_trans, 0);
1485 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1486 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1487 mp->m_kobj.kobject.kset = xfs_kset;
1488
1489 mp->m_super = sb;
1490 sb->s_fs_info = mp;
1491
1492 error = xfs_parseargs(mp, (char *)data);
1493 if (error)
1494 goto out_free_fsname;
1495
1496 sb_min_blocksize(sb, BBSIZE);
1497 sb->s_xattr = xfs_xattr_handlers;
1498 sb->s_export_op = &xfs_export_operations;
1499#ifdef CONFIG_XFS_QUOTA
1500 sb->s_qcop = &xfs_quotactl_operations;
1501 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1502#endif
1503 sb->s_op = &xfs_super_operations;
1504
1505 if (silent)
1506 flags |= XFS_MFSI_QUIET;
1507
1508 error = xfs_open_devices(mp);
1509 if (error)
1510 goto out_free_fsname;
1511
1512 error = xfs_init_mount_workqueues(mp);
1513 if (error)
1514 goto out_close_devices;
1515
1516 error = xfs_init_percpu_counters(mp);
1517 if (error)
1518 goto out_destroy_workqueues;
1519
1520 /* Allocate stats memory before we do operations that might use it */
1521 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1522 if (!mp->m_stats.xs_stats) {
1523 error = -ENOMEM;
1524 goto out_destroy_counters;
1525 }
1526
1527 error = xfs_readsb(mp, flags);
1528 if (error)
1529 goto out_free_stats;
1530
1531 error = xfs_finish_flags(mp);
1532 if (error)
1533 goto out_free_sb;
1534
1535 error = xfs_setup_devices(mp);
1536 if (error)
1537 goto out_free_sb;
1538
1539 error = xfs_filestream_mount(mp);
1540 if (error)
1541 goto out_free_sb;
1542
1543 /*
1544 * we must configure the block size in the superblock before we run the
1545 * full mount process as the mount process can lookup and cache inodes.
1546 */
1547 sb->s_magic = XFS_SB_MAGIC;
1548 sb->s_blocksize = mp->m_sb.sb_blocksize;
1549 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1550 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1551 sb->s_max_links = XFS_MAXLINK;
1552 sb->s_time_gran = 1;
1553 set_posix_acl_flag(sb);
1554
1555 /* version 5 superblocks support inode version counters. */
1556 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1557 sb->s_flags |= MS_I_VERSION;
1558
1559 if (mp->m_flags & XFS_MOUNT_DAX) {
1560 xfs_warn(mp,
1561 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1562 if (sb->s_blocksize != PAGE_SIZE) {
1563 xfs_alert(mp,
1564 "Filesystem block size invalid for DAX Turning DAX off.");
1565 mp->m_flags &= ~XFS_MOUNT_DAX;
1566 } else if (!sb->s_bdev->bd_disk->fops->direct_access) {
1567 xfs_alert(mp,
1568 "Block device does not support DAX Turning DAX off.");
1569 mp->m_flags &= ~XFS_MOUNT_DAX;
1570 }
1571 }
1572
1573 if (xfs_sb_version_hassparseinodes(&mp->m_sb))
1574 xfs_alert(mp,
1575 "EXPERIMENTAL sparse inode feature enabled. Use at your own risk!");
1576
1577 error = xfs_mountfs(mp);
1578 if (error)
1579 goto out_filestream_unmount;
1580
1581 root = igrab(VFS_I(mp->m_rootip));
1582 if (!root) {
1583 error = -ENOENT;
1584 goto out_unmount;
1585 }
1586 sb->s_root = d_make_root(root);
1587 if (!sb->s_root) {
1588 error = -ENOMEM;
1589 goto out_unmount;
1590 }
1591
1592 return 0;
1593
1594 out_filestream_unmount:
1595 xfs_filestream_unmount(mp);
1596 out_free_sb:
1597 xfs_freesb(mp);
1598 out_free_stats:
1599 free_percpu(mp->m_stats.xs_stats);
1600 out_destroy_counters:
1601 xfs_destroy_percpu_counters(mp);
1602 out_destroy_workqueues:
1603 xfs_destroy_mount_workqueues(mp);
1604 out_close_devices:
1605 xfs_close_devices(mp);
1606 out_free_fsname:
1607 xfs_free_fsname(mp);
1608 kfree(mp);
1609 out:
1610 return error;
1611
1612 out_unmount:
1613 xfs_filestream_unmount(mp);
1614 xfs_unmountfs(mp);
1615 goto out_free_sb;
1616}
1617
1618STATIC void
1619xfs_fs_put_super(
1620 struct super_block *sb)
1621{
1622 struct xfs_mount *mp = XFS_M(sb);
1623
1624 xfs_notice(mp, "Unmounting Filesystem");
1625 xfs_filestream_unmount(mp);
1626 xfs_unmountfs(mp);
1627
1628 xfs_freesb(mp);
1629 free_percpu(mp->m_stats.xs_stats);
1630 xfs_destroy_percpu_counters(mp);
1631 xfs_destroy_mount_workqueues(mp);
1632 xfs_close_devices(mp);
1633 xfs_free_fsname(mp);
1634 kfree(mp);
1635}
1636
1637STATIC struct dentry *
1638xfs_fs_mount(
1639 struct file_system_type *fs_type,
1640 int flags,
1641 const char *dev_name,
1642 void *data)
1643{
1644 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1645}
1646
1647static long
1648xfs_fs_nr_cached_objects(
1649 struct super_block *sb,
1650 struct shrink_control *sc)
1651{
1652 return xfs_reclaim_inodes_count(XFS_M(sb));
1653}
1654
1655static long
1656xfs_fs_free_cached_objects(
1657 struct super_block *sb,
1658 struct shrink_control *sc)
1659{
1660 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1661}
1662
1663static const struct super_operations xfs_super_operations = {
1664 .alloc_inode = xfs_fs_alloc_inode,
1665 .destroy_inode = xfs_fs_destroy_inode,
1666 .evict_inode = xfs_fs_evict_inode,
1667 .drop_inode = xfs_fs_drop_inode,
1668 .put_super = xfs_fs_put_super,
1669 .sync_fs = xfs_fs_sync_fs,
1670 .freeze_fs = xfs_fs_freeze,
1671 .unfreeze_fs = xfs_fs_unfreeze,
1672 .statfs = xfs_fs_statfs,
1673 .remount_fs = xfs_fs_remount,
1674 .show_options = xfs_fs_show_options,
1675 .nr_cached_objects = xfs_fs_nr_cached_objects,
1676 .free_cached_objects = xfs_fs_free_cached_objects,
1677};
1678
1679static struct file_system_type xfs_fs_type = {
1680 .owner = THIS_MODULE,
1681 .name = "xfs",
1682 .mount = xfs_fs_mount,
1683 .kill_sb = kill_block_super,
1684 .fs_flags = FS_REQUIRES_DEV,
1685};
1686MODULE_ALIAS_FS("xfs");
1687
1688STATIC int __init
1689xfs_init_zones(void)
1690{
1691
1692 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1693 if (!xfs_ioend_zone)
1694 goto out;
1695
1696 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1697 xfs_ioend_zone);
1698 if (!xfs_ioend_pool)
1699 goto out_destroy_ioend_zone;
1700
1701 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1702 "xfs_log_ticket");
1703 if (!xfs_log_ticket_zone)
1704 goto out_destroy_ioend_pool;
1705
1706 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1707 "xfs_bmap_free_item");
1708 if (!xfs_bmap_free_item_zone)
1709 goto out_destroy_log_ticket_zone;
1710
1711 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1712 "xfs_btree_cur");
1713 if (!xfs_btree_cur_zone)
1714 goto out_destroy_bmap_free_item_zone;
1715
1716 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1717 "xfs_da_state");
1718 if (!xfs_da_state_zone)
1719 goto out_destroy_btree_cur_zone;
1720
1721 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1722 if (!xfs_ifork_zone)
1723 goto out_destroy_da_state_zone;
1724
1725 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1726 if (!xfs_trans_zone)
1727 goto out_destroy_ifork_zone;
1728
1729 xfs_log_item_desc_zone =
1730 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1731 "xfs_log_item_desc");
1732 if (!xfs_log_item_desc_zone)
1733 goto out_destroy_trans_zone;
1734
1735 /*
1736 * The size of the zone allocated buf log item is the maximum
1737 * size possible under XFS. This wastes a little bit of memory,
1738 * but it is much faster.
1739 */
1740 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1741 "xfs_buf_item");
1742 if (!xfs_buf_item_zone)
1743 goto out_destroy_log_item_desc_zone;
1744
1745 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1746 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1747 sizeof(xfs_extent_t))), "xfs_efd_item");
1748 if (!xfs_efd_zone)
1749 goto out_destroy_buf_item_zone;
1750
1751 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1752 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1753 sizeof(xfs_extent_t))), "xfs_efi_item");
1754 if (!xfs_efi_zone)
1755 goto out_destroy_efd_zone;
1756
1757 xfs_inode_zone =
1758 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1759 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1760 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1761 if (!xfs_inode_zone)
1762 goto out_destroy_efi_zone;
1763
1764 xfs_ili_zone =
1765 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1766 KM_ZONE_SPREAD, NULL);
1767 if (!xfs_ili_zone)
1768 goto out_destroy_inode_zone;
1769 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1770 "xfs_icr");
1771 if (!xfs_icreate_zone)
1772 goto out_destroy_ili_zone;
1773
1774 return 0;
1775
1776 out_destroy_ili_zone:
1777 kmem_zone_destroy(xfs_ili_zone);
1778 out_destroy_inode_zone:
1779 kmem_zone_destroy(xfs_inode_zone);
1780 out_destroy_efi_zone:
1781 kmem_zone_destroy(xfs_efi_zone);
1782 out_destroy_efd_zone:
1783 kmem_zone_destroy(xfs_efd_zone);
1784 out_destroy_buf_item_zone:
1785 kmem_zone_destroy(xfs_buf_item_zone);
1786 out_destroy_log_item_desc_zone:
1787 kmem_zone_destroy(xfs_log_item_desc_zone);
1788 out_destroy_trans_zone:
1789 kmem_zone_destroy(xfs_trans_zone);
1790 out_destroy_ifork_zone:
1791 kmem_zone_destroy(xfs_ifork_zone);
1792 out_destroy_da_state_zone:
1793 kmem_zone_destroy(xfs_da_state_zone);
1794 out_destroy_btree_cur_zone:
1795 kmem_zone_destroy(xfs_btree_cur_zone);
1796 out_destroy_bmap_free_item_zone:
1797 kmem_zone_destroy(xfs_bmap_free_item_zone);
1798 out_destroy_log_ticket_zone:
1799 kmem_zone_destroy(xfs_log_ticket_zone);
1800 out_destroy_ioend_pool:
1801 mempool_destroy(xfs_ioend_pool);
1802 out_destroy_ioend_zone:
1803 kmem_zone_destroy(xfs_ioend_zone);
1804 out:
1805 return -ENOMEM;
1806}
1807
1808STATIC void
1809xfs_destroy_zones(void)
1810{
1811 /*
1812 * Make sure all delayed rcu free are flushed before we
1813 * destroy caches.
1814 */
1815 rcu_barrier();
1816 kmem_zone_destroy(xfs_icreate_zone);
1817 kmem_zone_destroy(xfs_ili_zone);
1818 kmem_zone_destroy(xfs_inode_zone);
1819 kmem_zone_destroy(xfs_efi_zone);
1820 kmem_zone_destroy(xfs_efd_zone);
1821 kmem_zone_destroy(xfs_buf_item_zone);
1822 kmem_zone_destroy(xfs_log_item_desc_zone);
1823 kmem_zone_destroy(xfs_trans_zone);
1824 kmem_zone_destroy(xfs_ifork_zone);
1825 kmem_zone_destroy(xfs_da_state_zone);
1826 kmem_zone_destroy(xfs_btree_cur_zone);
1827 kmem_zone_destroy(xfs_bmap_free_item_zone);
1828 kmem_zone_destroy(xfs_log_ticket_zone);
1829 mempool_destroy(xfs_ioend_pool);
1830 kmem_zone_destroy(xfs_ioend_zone);
1831
1832}
1833
1834STATIC int __init
1835xfs_init_workqueues(void)
1836{
1837 /*
1838 * The allocation workqueue can be used in memory reclaim situations
1839 * (writepage path), and parallelism is only limited by the number of
1840 * AGs in all the filesystems mounted. Hence use the default large
1841 * max_active value for this workqueue.
1842 */
1843 xfs_alloc_wq = alloc_workqueue("xfsalloc",
1844 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1845 if (!xfs_alloc_wq)
1846 return -ENOMEM;
1847
1848 return 0;
1849}
1850
1851STATIC void
1852xfs_destroy_workqueues(void)
1853{
1854 destroy_workqueue(xfs_alloc_wq);
1855}
1856
1857STATIC int __init
1858init_xfs_fs(void)
1859{
1860 int error;
1861
1862 xfs_check_ondisk_structs();
1863
1864 printk(KERN_INFO XFS_VERSION_STRING " with "
1865 XFS_BUILD_OPTIONS " enabled\n");
1866
1867 xfs_dir_startup();
1868
1869 error = xfs_init_zones();
1870 if (error)
1871 goto out;
1872
1873 error = xfs_init_workqueues();
1874 if (error)
1875 goto out_destroy_zones;
1876
1877 error = xfs_mru_cache_init();
1878 if (error)
1879 goto out_destroy_wq;
1880
1881 error = xfs_buf_init();
1882 if (error)
1883 goto out_mru_cache_uninit;
1884
1885 error = xfs_init_procfs();
1886 if (error)
1887 goto out_buf_terminate;
1888
1889 error = xfs_sysctl_register();
1890 if (error)
1891 goto out_cleanup_procfs;
1892
1893 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
1894 if (!xfs_kset) {
1895 error = -ENOMEM;
1896 goto out_sysctl_unregister;
1897 }
1898
1899 xfsstats.xs_kobj.kobject.kset = xfs_kset;
1900
1901 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
1902 if (!xfsstats.xs_stats) {
1903 error = -ENOMEM;
1904 goto out_kset_unregister;
1905 }
1906
1907 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
1908 "stats");
1909 if (error)
1910 goto out_free_stats;
1911
1912#ifdef DEBUG
1913 xfs_dbg_kobj.kobject.kset = xfs_kset;
1914 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
1915 if (error)
1916 goto out_remove_stats_kobj;
1917#endif
1918
1919 error = xfs_qm_init();
1920 if (error)
1921 goto out_remove_dbg_kobj;
1922
1923 error = register_filesystem(&xfs_fs_type);
1924 if (error)
1925 goto out_qm_exit;
1926 return 0;
1927
1928 out_qm_exit:
1929 xfs_qm_exit();
1930 out_remove_dbg_kobj:
1931#ifdef DEBUG
1932 xfs_sysfs_del(&xfs_dbg_kobj);
1933 out_remove_stats_kobj:
1934#endif
1935 xfs_sysfs_del(&xfsstats.xs_kobj);
1936 out_free_stats:
1937 free_percpu(xfsstats.xs_stats);
1938 out_kset_unregister:
1939 kset_unregister(xfs_kset);
1940 out_sysctl_unregister:
1941 xfs_sysctl_unregister();
1942 out_cleanup_procfs:
1943 xfs_cleanup_procfs();
1944 out_buf_terminate:
1945 xfs_buf_terminate();
1946 out_mru_cache_uninit:
1947 xfs_mru_cache_uninit();
1948 out_destroy_wq:
1949 xfs_destroy_workqueues();
1950 out_destroy_zones:
1951 xfs_destroy_zones();
1952 out:
1953 return error;
1954}
1955
1956STATIC void __exit
1957exit_xfs_fs(void)
1958{
1959 xfs_qm_exit();
1960 unregister_filesystem(&xfs_fs_type);
1961#ifdef DEBUG
1962 xfs_sysfs_del(&xfs_dbg_kobj);
1963#endif
1964 xfs_sysfs_del(&xfsstats.xs_kobj);
1965 free_percpu(xfsstats.xs_stats);
1966 kset_unregister(xfs_kset);
1967 xfs_sysctl_unregister();
1968 xfs_cleanup_procfs();
1969 xfs_buf_terminate();
1970 xfs_mru_cache_uninit();
1971 xfs_destroy_workqueues();
1972 xfs_destroy_zones();
1973 xfs_uuid_table_free();
1974}
1975
1976module_init(init_xfs_fs);
1977module_exit(exit_xfs_fs);
1978
1979MODULE_AUTHOR("Silicon Graphics, Inc.");
1980MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1981MODULE_LICENSE("GPL");