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// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6
7#include "xfs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_sb.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_btree.h"
16#include "xfs_bmap.h"
17#include "xfs_alloc.h"
18#include "xfs_fsops.h"
19#include "xfs_trans.h"
20#include "xfs_buf_item.h"
21#include "xfs_log.h"
22#include "xfs_log_priv.h"
23#include "xfs_dir2.h"
24#include "xfs_extfree_item.h"
25#include "xfs_mru_cache.h"
26#include "xfs_inode_item.h"
27#include "xfs_icache.h"
28#include "xfs_trace.h"
29#include "xfs_icreate_item.h"
30#include "xfs_filestream.h"
31#include "xfs_quota.h"
32#include "xfs_sysfs.h"
33#include "xfs_ondisk.h"
34#include "xfs_rmap_item.h"
35#include "xfs_refcount_item.h"
36#include "xfs_bmap_item.h"
37#include "xfs_reflink.h"
38
39#include <linux/magic.h>
40#include <linux/parser.h>
41
42static const struct super_operations xfs_super_operations;
43struct bio_set xfs_ioend_bioset;
44
45static struct kset *xfs_kset; /* top-level xfs sysfs dir */
46#ifdef DEBUG
47static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
48#endif
49
50/*
51 * Table driven mount option parser.
52 */
53enum {
54 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
55 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
56 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
57 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
58 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
59 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
60 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
61 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
62 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
63};
64
65static const match_table_t tokens = {
66 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */
67 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */
68 {Opt_logdev, "logdev=%s"}, /* log device */
69 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */
70 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */
71 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */
72 {Opt_noalign, "noalign"}, /* turn off stripe alignment */
73 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */
74 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */
75 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */
76 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */
77 {Opt_grpid, "grpid"}, /* group-ID from parent directory */
78 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */
79 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */
80 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */
81 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
82 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */
83 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */
84 {Opt_inode32, "inode32"}, /* inode allocation limited to
85 * XFS_MAXINUMBER_32 */
86 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */
87 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */
88 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */
89 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes
90 * in stat(). */
91 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */
92 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */
93 {Opt_filestreams,"filestreams"},/* use filestreams allocator */
94 {Opt_quota, "quota"}, /* disk quotas (user) */
95 {Opt_noquota, "noquota"}, /* no quotas */
96 {Opt_usrquota, "usrquota"}, /* user quota enabled */
97 {Opt_grpquota, "grpquota"}, /* group quota enabled */
98 {Opt_prjquota, "prjquota"}, /* project quota enabled */
99 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */
100 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */
101 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */
102 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
103 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
104 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
105 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
106 {Opt_discard, "discard"}, /* Discard unused blocks */
107 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */
108 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */
109 {Opt_err, NULL},
110};
111
112
113STATIC int
114suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
115{
116 int last, shift_left_factor = 0, _res;
117 char *value;
118 int ret = 0;
119
120 value = match_strdup(s);
121 if (!value)
122 return -ENOMEM;
123
124 last = strlen(value) - 1;
125 if (value[last] == 'K' || value[last] == 'k') {
126 shift_left_factor = 10;
127 value[last] = '\0';
128 }
129 if (value[last] == 'M' || value[last] == 'm') {
130 shift_left_factor = 20;
131 value[last] = '\0';
132 }
133 if (value[last] == 'G' || value[last] == 'g') {
134 shift_left_factor = 30;
135 value[last] = '\0';
136 }
137
138 if (kstrtoint(value, base, &_res))
139 ret = -EINVAL;
140 kfree(value);
141 *res = _res << shift_left_factor;
142 return ret;
143}
144
145/*
146 * This function fills in xfs_mount_t fields based on mount args.
147 * Note: the superblock has _not_ yet been read in.
148 *
149 * Note that this function leaks the various device name allocations on
150 * failure. The caller takes care of them.
151 *
152 * *sb is const because this is also used to test options on the remount
153 * path, and we don't want this to have any side effects at remount time.
154 * Today this function does not change *sb, but just to future-proof...
155 */
156STATIC int
157xfs_parseargs(
158 struct xfs_mount *mp,
159 char *options)
160{
161 const struct super_block *sb = mp->m_super;
162 char *p;
163 substring_t args[MAX_OPT_ARGS];
164 int dsunit = 0;
165 int dswidth = 0;
166 int iosize = 0;
167 uint8_t iosizelog = 0;
168
169 /*
170 * set up the mount name first so all the errors will refer to the
171 * correct device.
172 */
173 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
174 if (!mp->m_fsname)
175 return -ENOMEM;
176 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
177
178 /*
179 * Copy binary VFS mount flags we are interested in.
180 */
181 if (sb_rdonly(sb))
182 mp->m_flags |= XFS_MOUNT_RDONLY;
183 if (sb->s_flags & SB_DIRSYNC)
184 mp->m_flags |= XFS_MOUNT_DIRSYNC;
185 if (sb->s_flags & SB_SYNCHRONOUS)
186 mp->m_flags |= XFS_MOUNT_WSYNC;
187
188 /*
189 * Set some default flags that could be cleared by the mount option
190 * parsing.
191 */
192 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
193
194 /*
195 * These can be overridden by the mount option parsing.
196 */
197 mp->m_logbufs = -1;
198 mp->m_logbsize = -1;
199
200 if (!options)
201 goto done;
202
203 while ((p = strsep(&options, ",")) != NULL) {
204 int token;
205
206 if (!*p)
207 continue;
208
209 token = match_token(p, tokens, args);
210 switch (token) {
211 case Opt_logbufs:
212 if (match_int(args, &mp->m_logbufs))
213 return -EINVAL;
214 break;
215 case Opt_logbsize:
216 if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
217 return -EINVAL;
218 break;
219 case Opt_logdev:
220 kfree(mp->m_logname);
221 mp->m_logname = match_strdup(args);
222 if (!mp->m_logname)
223 return -ENOMEM;
224 break;
225 case Opt_rtdev:
226 kfree(mp->m_rtname);
227 mp->m_rtname = match_strdup(args);
228 if (!mp->m_rtname)
229 return -ENOMEM;
230 break;
231 case Opt_allocsize:
232 case Opt_biosize:
233 if (suffix_kstrtoint(args, 10, &iosize))
234 return -EINVAL;
235 iosizelog = ffs(iosize) - 1;
236 break;
237 case Opt_grpid:
238 case Opt_bsdgroups:
239 mp->m_flags |= XFS_MOUNT_GRPID;
240 break;
241 case Opt_nogrpid:
242 case Opt_sysvgroups:
243 mp->m_flags &= ~XFS_MOUNT_GRPID;
244 break;
245 case Opt_wsync:
246 mp->m_flags |= XFS_MOUNT_WSYNC;
247 break;
248 case Opt_norecovery:
249 mp->m_flags |= XFS_MOUNT_NORECOVERY;
250 break;
251 case Opt_noalign:
252 mp->m_flags |= XFS_MOUNT_NOALIGN;
253 break;
254 case Opt_swalloc:
255 mp->m_flags |= XFS_MOUNT_SWALLOC;
256 break;
257 case Opt_sunit:
258 if (match_int(args, &dsunit))
259 return -EINVAL;
260 break;
261 case Opt_swidth:
262 if (match_int(args, &dswidth))
263 return -EINVAL;
264 break;
265 case Opt_inode32:
266 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
267 break;
268 case Opt_inode64:
269 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
270 break;
271 case Opt_nouuid:
272 mp->m_flags |= XFS_MOUNT_NOUUID;
273 break;
274 case Opt_ikeep:
275 mp->m_flags |= XFS_MOUNT_IKEEP;
276 break;
277 case Opt_noikeep:
278 mp->m_flags &= ~XFS_MOUNT_IKEEP;
279 break;
280 case Opt_largeio:
281 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
282 break;
283 case Opt_nolargeio:
284 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
285 break;
286 case Opt_attr2:
287 mp->m_flags |= XFS_MOUNT_ATTR2;
288 break;
289 case Opt_noattr2:
290 mp->m_flags &= ~XFS_MOUNT_ATTR2;
291 mp->m_flags |= XFS_MOUNT_NOATTR2;
292 break;
293 case Opt_filestreams:
294 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
295 break;
296 case Opt_noquota:
297 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
298 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
299 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
300 break;
301 case Opt_quota:
302 case Opt_uquota:
303 case Opt_usrquota:
304 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
305 XFS_UQUOTA_ENFD);
306 break;
307 case Opt_qnoenforce:
308 case Opt_uqnoenforce:
309 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
310 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
311 break;
312 case Opt_pquota:
313 case Opt_prjquota:
314 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
315 XFS_PQUOTA_ENFD);
316 break;
317 case Opt_pqnoenforce:
318 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
319 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
320 break;
321 case Opt_gquota:
322 case Opt_grpquota:
323 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
324 XFS_GQUOTA_ENFD);
325 break;
326 case Opt_gqnoenforce:
327 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
328 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
329 break;
330 case Opt_discard:
331 mp->m_flags |= XFS_MOUNT_DISCARD;
332 break;
333 case Opt_nodiscard:
334 mp->m_flags &= ~XFS_MOUNT_DISCARD;
335 break;
336#ifdef CONFIG_FS_DAX
337 case Opt_dax:
338 mp->m_flags |= XFS_MOUNT_DAX;
339 break;
340#endif
341 default:
342 xfs_warn(mp, "unknown mount option [%s].", p);
343 return -EINVAL;
344 }
345 }
346
347 /*
348 * no recovery flag requires a read-only mount
349 */
350 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
351 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
352 xfs_warn(mp, "no-recovery mounts must be read-only.");
353 return -EINVAL;
354 }
355
356 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
357 xfs_warn(mp,
358 "sunit and swidth options incompatible with the noalign option");
359 return -EINVAL;
360 }
361
362#ifndef CONFIG_XFS_QUOTA
363 if (XFS_IS_QUOTA_RUNNING(mp)) {
364 xfs_warn(mp, "quota support not available in this kernel.");
365 return -EINVAL;
366 }
367#endif
368
369 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
370 xfs_warn(mp, "sunit and swidth must be specified together");
371 return -EINVAL;
372 }
373
374 if (dsunit && (dswidth % dsunit != 0)) {
375 xfs_warn(mp,
376 "stripe width (%d) must be a multiple of the stripe unit (%d)",
377 dswidth, dsunit);
378 return -EINVAL;
379 }
380
381done:
382 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
383 /*
384 * At this point the superblock has not been read
385 * in, therefore we do not know the block size.
386 * Before the mount call ends we will convert
387 * these to FSBs.
388 */
389 mp->m_dalign = dsunit;
390 mp->m_swidth = dswidth;
391 }
392
393 if (mp->m_logbufs != -1 &&
394 mp->m_logbufs != 0 &&
395 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
396 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
397 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
398 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
399 return -EINVAL;
400 }
401 if (mp->m_logbsize != -1 &&
402 mp->m_logbsize != 0 &&
403 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
404 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
405 !is_power_of_2(mp->m_logbsize))) {
406 xfs_warn(mp,
407 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
408 mp->m_logbsize);
409 return -EINVAL;
410 }
411
412 if (iosizelog) {
413 if (iosizelog > XFS_MAX_IO_LOG ||
414 iosizelog < XFS_MIN_IO_LOG) {
415 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
416 iosizelog, XFS_MIN_IO_LOG,
417 XFS_MAX_IO_LOG);
418 return -EINVAL;
419 }
420
421 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
422 mp->m_readio_log = iosizelog;
423 mp->m_writeio_log = iosizelog;
424 }
425
426 return 0;
427}
428
429struct proc_xfs_info {
430 uint64_t flag;
431 char *str;
432};
433
434STATIC void
435xfs_showargs(
436 struct xfs_mount *mp,
437 struct seq_file *m)
438{
439 static struct proc_xfs_info xfs_info_set[] = {
440 /* the few simple ones we can get from the mount struct */
441 { XFS_MOUNT_IKEEP, ",ikeep" },
442 { XFS_MOUNT_WSYNC, ",wsync" },
443 { XFS_MOUNT_NOALIGN, ",noalign" },
444 { XFS_MOUNT_SWALLOC, ",swalloc" },
445 { XFS_MOUNT_NOUUID, ",nouuid" },
446 { XFS_MOUNT_NORECOVERY, ",norecovery" },
447 { XFS_MOUNT_ATTR2, ",attr2" },
448 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
449 { XFS_MOUNT_GRPID, ",grpid" },
450 { XFS_MOUNT_DISCARD, ",discard" },
451 { XFS_MOUNT_SMALL_INUMS, ",inode32" },
452 { XFS_MOUNT_DAX, ",dax" },
453 { 0, NULL }
454 };
455 static struct proc_xfs_info xfs_info_unset[] = {
456 /* the few simple ones we can get from the mount struct */
457 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" },
458 { XFS_MOUNT_SMALL_INUMS, ",inode64" },
459 { 0, NULL }
460 };
461 struct proc_xfs_info *xfs_infop;
462
463 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
464 if (mp->m_flags & xfs_infop->flag)
465 seq_puts(m, xfs_infop->str);
466 }
467 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
468 if (!(mp->m_flags & xfs_infop->flag))
469 seq_puts(m, xfs_infop->str);
470 }
471
472 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
473 seq_printf(m, ",allocsize=%dk",
474 (int)(1 << mp->m_writeio_log) >> 10);
475
476 if (mp->m_logbufs > 0)
477 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
478 if (mp->m_logbsize > 0)
479 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
480
481 if (mp->m_logname)
482 seq_show_option(m, "logdev", mp->m_logname);
483 if (mp->m_rtname)
484 seq_show_option(m, "rtdev", mp->m_rtname);
485
486 if (mp->m_dalign > 0)
487 seq_printf(m, ",sunit=%d",
488 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
489 if (mp->m_swidth > 0)
490 seq_printf(m, ",swidth=%d",
491 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
492
493 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
494 seq_puts(m, ",usrquota");
495 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
496 seq_puts(m, ",uqnoenforce");
497
498 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
499 if (mp->m_qflags & XFS_PQUOTA_ENFD)
500 seq_puts(m, ",prjquota");
501 else
502 seq_puts(m, ",pqnoenforce");
503 }
504 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
505 if (mp->m_qflags & XFS_GQUOTA_ENFD)
506 seq_puts(m, ",grpquota");
507 else
508 seq_puts(m, ",gqnoenforce");
509 }
510
511 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
512 seq_puts(m, ",noquota");
513}
514
515static uint64_t
516xfs_max_file_offset(
517 unsigned int blockshift)
518{
519 unsigned int pagefactor = 1;
520 unsigned int bitshift = BITS_PER_LONG - 1;
521
522 /* Figure out maximum filesize, on Linux this can depend on
523 * the filesystem blocksize (on 32 bit platforms).
524 * __block_write_begin does this in an [unsigned] long long...
525 * page->index << (PAGE_SHIFT - bbits)
526 * So, for page sized blocks (4K on 32 bit platforms),
527 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
528 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
529 * but for smaller blocksizes it is less (bbits = log2 bsize).
530 */
531
532#if BITS_PER_LONG == 32
533 ASSERT(sizeof(sector_t) == 8);
534 pagefactor = PAGE_SIZE;
535 bitshift = BITS_PER_LONG;
536#endif
537
538 return (((uint64_t)pagefactor) << bitshift) - 1;
539}
540
541/*
542 * Set parameters for inode allocation heuristics, taking into account
543 * filesystem size and inode32/inode64 mount options; i.e. specifically
544 * whether or not XFS_MOUNT_SMALL_INUMS is set.
545 *
546 * Inode allocation patterns are altered only if inode32 is requested
547 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
548 * If altered, XFS_MOUNT_32BITINODES is set as well.
549 *
550 * An agcount independent of that in the mount structure is provided
551 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
552 * to the potentially higher ag count.
553 *
554 * Returns the maximum AG index which may contain inodes.
555 */
556xfs_agnumber_t
557xfs_set_inode_alloc(
558 struct xfs_mount *mp,
559 xfs_agnumber_t agcount)
560{
561 xfs_agnumber_t index;
562 xfs_agnumber_t maxagi = 0;
563 xfs_sb_t *sbp = &mp->m_sb;
564 xfs_agnumber_t max_metadata;
565 xfs_agino_t agino;
566 xfs_ino_t ino;
567
568 /*
569 * Calculate how much should be reserved for inodes to meet
570 * the max inode percentage. Used only for inode32.
571 */
572 if (M_IGEO(mp)->maxicount) {
573 uint64_t icount;
574
575 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
576 do_div(icount, 100);
577 icount += sbp->sb_agblocks - 1;
578 do_div(icount, sbp->sb_agblocks);
579 max_metadata = icount;
580 } else {
581 max_metadata = agcount;
582 }
583
584 /* Get the last possible inode in the filesystem */
585 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
586 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
587
588 /*
589 * If user asked for no more than 32-bit inodes, and the fs is
590 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
591 * the allocator to accommodate the request.
592 */
593 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
594 mp->m_flags |= XFS_MOUNT_32BITINODES;
595 else
596 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
597
598 for (index = 0; index < agcount; index++) {
599 struct xfs_perag *pag;
600
601 ino = XFS_AGINO_TO_INO(mp, index, agino);
602
603 pag = xfs_perag_get(mp, index);
604
605 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
606 if (ino > XFS_MAXINUMBER_32) {
607 pag->pagi_inodeok = 0;
608 pag->pagf_metadata = 0;
609 } else {
610 pag->pagi_inodeok = 1;
611 maxagi++;
612 if (index < max_metadata)
613 pag->pagf_metadata = 1;
614 else
615 pag->pagf_metadata = 0;
616 }
617 } else {
618 pag->pagi_inodeok = 1;
619 pag->pagf_metadata = 0;
620 }
621
622 xfs_perag_put(pag);
623 }
624
625 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
626}
627
628STATIC int
629xfs_blkdev_get(
630 xfs_mount_t *mp,
631 const char *name,
632 struct block_device **bdevp)
633{
634 int error = 0;
635
636 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
637 mp);
638 if (IS_ERR(*bdevp)) {
639 error = PTR_ERR(*bdevp);
640 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
641 }
642
643 return error;
644}
645
646STATIC void
647xfs_blkdev_put(
648 struct block_device *bdev)
649{
650 if (bdev)
651 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
652}
653
654void
655xfs_blkdev_issue_flush(
656 xfs_buftarg_t *buftarg)
657{
658 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
659}
660
661STATIC void
662xfs_close_devices(
663 struct xfs_mount *mp)
664{
665 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
666
667 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
668 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
669 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
670
671 xfs_free_buftarg(mp->m_logdev_targp);
672 xfs_blkdev_put(logdev);
673 fs_put_dax(dax_logdev);
674 }
675 if (mp->m_rtdev_targp) {
676 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
677 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
678
679 xfs_free_buftarg(mp->m_rtdev_targp);
680 xfs_blkdev_put(rtdev);
681 fs_put_dax(dax_rtdev);
682 }
683 xfs_free_buftarg(mp->m_ddev_targp);
684 fs_put_dax(dax_ddev);
685}
686
687/*
688 * The file system configurations are:
689 * (1) device (partition) with data and internal log
690 * (2) logical volume with data and log subvolumes.
691 * (3) logical volume with data, log, and realtime subvolumes.
692 *
693 * We only have to handle opening the log and realtime volumes here if
694 * they are present. The data subvolume has already been opened by
695 * get_sb_bdev() and is stored in sb->s_bdev.
696 */
697STATIC int
698xfs_open_devices(
699 struct xfs_mount *mp)
700{
701 struct block_device *ddev = mp->m_super->s_bdev;
702 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
703 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
704 struct block_device *logdev = NULL, *rtdev = NULL;
705 int error;
706
707 /*
708 * Open real time and log devices - order is important.
709 */
710 if (mp->m_logname) {
711 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
712 if (error)
713 goto out;
714 dax_logdev = fs_dax_get_by_bdev(logdev);
715 }
716
717 if (mp->m_rtname) {
718 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
719 if (error)
720 goto out_close_logdev;
721
722 if (rtdev == ddev || rtdev == logdev) {
723 xfs_warn(mp,
724 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
725 error = -EINVAL;
726 goto out_close_rtdev;
727 }
728 dax_rtdev = fs_dax_get_by_bdev(rtdev);
729 }
730
731 /*
732 * Setup xfs_mount buffer target pointers
733 */
734 error = -ENOMEM;
735 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
736 if (!mp->m_ddev_targp)
737 goto out_close_rtdev;
738
739 if (rtdev) {
740 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
741 if (!mp->m_rtdev_targp)
742 goto out_free_ddev_targ;
743 }
744
745 if (logdev && logdev != ddev) {
746 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
747 if (!mp->m_logdev_targp)
748 goto out_free_rtdev_targ;
749 } else {
750 mp->m_logdev_targp = mp->m_ddev_targp;
751 }
752
753 return 0;
754
755 out_free_rtdev_targ:
756 if (mp->m_rtdev_targp)
757 xfs_free_buftarg(mp->m_rtdev_targp);
758 out_free_ddev_targ:
759 xfs_free_buftarg(mp->m_ddev_targp);
760 out_close_rtdev:
761 xfs_blkdev_put(rtdev);
762 fs_put_dax(dax_rtdev);
763 out_close_logdev:
764 if (logdev && logdev != ddev) {
765 xfs_blkdev_put(logdev);
766 fs_put_dax(dax_logdev);
767 }
768 out:
769 fs_put_dax(dax_ddev);
770 return error;
771}
772
773/*
774 * Setup xfs_mount buffer target pointers based on superblock
775 */
776STATIC int
777xfs_setup_devices(
778 struct xfs_mount *mp)
779{
780 int error;
781
782 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
783 if (error)
784 return error;
785
786 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
787 unsigned int log_sector_size = BBSIZE;
788
789 if (xfs_sb_version_hassector(&mp->m_sb))
790 log_sector_size = mp->m_sb.sb_logsectsize;
791 error = xfs_setsize_buftarg(mp->m_logdev_targp,
792 log_sector_size);
793 if (error)
794 return error;
795 }
796 if (mp->m_rtdev_targp) {
797 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
798 mp->m_sb.sb_sectsize);
799 if (error)
800 return error;
801 }
802
803 return 0;
804}
805
806STATIC int
807xfs_init_mount_workqueues(
808 struct xfs_mount *mp)
809{
810 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
811 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
812 if (!mp->m_buf_workqueue)
813 goto out;
814
815 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
816 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
817 if (!mp->m_unwritten_workqueue)
818 goto out_destroy_buf;
819
820 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
821 WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
822 0, mp->m_fsname);
823 if (!mp->m_cil_workqueue)
824 goto out_destroy_unwritten;
825
826 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
827 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
828 if (!mp->m_reclaim_workqueue)
829 goto out_destroy_cil;
830
831 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
832 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
833 if (!mp->m_eofblocks_workqueue)
834 goto out_destroy_reclaim;
835
836 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
837 mp->m_fsname);
838 if (!mp->m_sync_workqueue)
839 goto out_destroy_eofb;
840
841 return 0;
842
843out_destroy_eofb:
844 destroy_workqueue(mp->m_eofblocks_workqueue);
845out_destroy_reclaim:
846 destroy_workqueue(mp->m_reclaim_workqueue);
847out_destroy_cil:
848 destroy_workqueue(mp->m_cil_workqueue);
849out_destroy_unwritten:
850 destroy_workqueue(mp->m_unwritten_workqueue);
851out_destroy_buf:
852 destroy_workqueue(mp->m_buf_workqueue);
853out:
854 return -ENOMEM;
855}
856
857STATIC void
858xfs_destroy_mount_workqueues(
859 struct xfs_mount *mp)
860{
861 destroy_workqueue(mp->m_sync_workqueue);
862 destroy_workqueue(mp->m_eofblocks_workqueue);
863 destroy_workqueue(mp->m_reclaim_workqueue);
864 destroy_workqueue(mp->m_cil_workqueue);
865 destroy_workqueue(mp->m_unwritten_workqueue);
866 destroy_workqueue(mp->m_buf_workqueue);
867}
868
869/*
870 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
871 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
872 * for IO to complete so that we effectively throttle multiple callers to the
873 * rate at which IO is completing.
874 */
875void
876xfs_flush_inodes(
877 struct xfs_mount *mp)
878{
879 struct super_block *sb = mp->m_super;
880
881 if (down_read_trylock(&sb->s_umount)) {
882 sync_inodes_sb(sb);
883 up_read(&sb->s_umount);
884 }
885}
886
887/* Catch misguided souls that try to use this interface on XFS */
888STATIC struct inode *
889xfs_fs_alloc_inode(
890 struct super_block *sb)
891{
892 BUG();
893 return NULL;
894}
895
896#ifdef DEBUG
897static void
898xfs_check_delalloc(
899 struct xfs_inode *ip,
900 int whichfork)
901{
902 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
903 struct xfs_bmbt_irec got;
904 struct xfs_iext_cursor icur;
905
906 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
907 return;
908 do {
909 if (isnullstartblock(got.br_startblock)) {
910 xfs_warn(ip->i_mount,
911 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
912 ip->i_ino,
913 whichfork == XFS_DATA_FORK ? "data" : "cow",
914 got.br_startoff, got.br_blockcount);
915 }
916 } while (xfs_iext_next_extent(ifp, &icur, &got));
917}
918#else
919#define xfs_check_delalloc(ip, whichfork) do { } while (0)
920#endif
921
922/*
923 * Now that the generic code is guaranteed not to be accessing
924 * the linux inode, we can inactivate and reclaim the inode.
925 */
926STATIC void
927xfs_fs_destroy_inode(
928 struct inode *inode)
929{
930 struct xfs_inode *ip = XFS_I(inode);
931
932 trace_xfs_destroy_inode(ip);
933
934 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
935 XFS_STATS_INC(ip->i_mount, vn_rele);
936 XFS_STATS_INC(ip->i_mount, vn_remove);
937
938 xfs_inactive(ip);
939
940 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
941 xfs_check_delalloc(ip, XFS_DATA_FORK);
942 xfs_check_delalloc(ip, XFS_COW_FORK);
943 ASSERT(0);
944 }
945
946 XFS_STATS_INC(ip->i_mount, vn_reclaim);
947
948 /*
949 * We should never get here with one of the reclaim flags already set.
950 */
951 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
952 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
953
954 /*
955 * We always use background reclaim here because even if the
956 * inode is clean, it still may be under IO and hence we have
957 * to take the flush lock. The background reclaim path handles
958 * this more efficiently than we can here, so simply let background
959 * reclaim tear down all inodes.
960 */
961 xfs_inode_set_reclaim_tag(ip);
962}
963
964static void
965xfs_fs_dirty_inode(
966 struct inode *inode,
967 int flag)
968{
969 struct xfs_inode *ip = XFS_I(inode);
970 struct xfs_mount *mp = ip->i_mount;
971 struct xfs_trans *tp;
972
973 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
974 return;
975 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
976 return;
977
978 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
979 return;
980 xfs_ilock(ip, XFS_ILOCK_EXCL);
981 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
982 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
983 xfs_trans_commit(tp);
984}
985
986/*
987 * Slab object creation initialisation for the XFS inode.
988 * This covers only the idempotent fields in the XFS inode;
989 * all other fields need to be initialised on allocation
990 * from the slab. This avoids the need to repeatedly initialise
991 * fields in the xfs inode that left in the initialise state
992 * when freeing the inode.
993 */
994STATIC void
995xfs_fs_inode_init_once(
996 void *inode)
997{
998 struct xfs_inode *ip = inode;
999
1000 memset(ip, 0, sizeof(struct xfs_inode));
1001
1002 /* vfs inode */
1003 inode_init_once(VFS_I(ip));
1004
1005 /* xfs inode */
1006 atomic_set(&ip->i_pincount, 0);
1007 spin_lock_init(&ip->i_flags_lock);
1008
1009 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1010 "xfsino", ip->i_ino);
1011 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1012 "xfsino", ip->i_ino);
1013}
1014
1015/*
1016 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1017 * serialised against cache hits here via the inode->i_lock and igrab() in
1018 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1019 * racing with us, and it avoids needing to grab a spinlock here for every inode
1020 * we drop the final reference on.
1021 */
1022STATIC int
1023xfs_fs_drop_inode(
1024 struct inode *inode)
1025{
1026 struct xfs_inode *ip = XFS_I(inode);
1027
1028 /*
1029 * If this unlinked inode is in the middle of recovery, don't
1030 * drop the inode just yet; log recovery will take care of
1031 * that. See the comment for this inode flag.
1032 */
1033 if (ip->i_flags & XFS_IRECOVERY) {
1034 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1035 return 0;
1036 }
1037
1038 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1039}
1040
1041STATIC void
1042xfs_free_fsname(
1043 struct xfs_mount *mp)
1044{
1045 kfree(mp->m_fsname);
1046 kfree(mp->m_rtname);
1047 kfree(mp->m_logname);
1048}
1049
1050STATIC int
1051xfs_fs_sync_fs(
1052 struct super_block *sb,
1053 int wait)
1054{
1055 struct xfs_mount *mp = XFS_M(sb);
1056
1057 /*
1058 * Doing anything during the async pass would be counterproductive.
1059 */
1060 if (!wait)
1061 return 0;
1062
1063 xfs_log_force(mp, XFS_LOG_SYNC);
1064 if (laptop_mode) {
1065 /*
1066 * The disk must be active because we're syncing.
1067 * We schedule log work now (now that the disk is
1068 * active) instead of later (when it might not be).
1069 */
1070 flush_delayed_work(&mp->m_log->l_work);
1071 }
1072
1073 return 0;
1074}
1075
1076STATIC int
1077xfs_fs_statfs(
1078 struct dentry *dentry,
1079 struct kstatfs *statp)
1080{
1081 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1082 xfs_sb_t *sbp = &mp->m_sb;
1083 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1084 uint64_t fakeinos, id;
1085 uint64_t icount;
1086 uint64_t ifree;
1087 uint64_t fdblocks;
1088 xfs_extlen_t lsize;
1089 int64_t ffree;
1090
1091 statp->f_type = XFS_SUPER_MAGIC;
1092 statp->f_namelen = MAXNAMELEN - 1;
1093
1094 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1095 statp->f_fsid.val[0] = (u32)id;
1096 statp->f_fsid.val[1] = (u32)(id >> 32);
1097
1098 icount = percpu_counter_sum(&mp->m_icount);
1099 ifree = percpu_counter_sum(&mp->m_ifree);
1100 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1101
1102 spin_lock(&mp->m_sb_lock);
1103 statp->f_bsize = sbp->sb_blocksize;
1104 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1105 statp->f_blocks = sbp->sb_dblocks - lsize;
1106 spin_unlock(&mp->m_sb_lock);
1107
1108 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1109 statp->f_bavail = statp->f_bfree;
1110
1111 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
1112 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1113 if (M_IGEO(mp)->maxicount)
1114 statp->f_files = min_t(typeof(statp->f_files),
1115 statp->f_files,
1116 M_IGEO(mp)->maxicount);
1117
1118 /* If sb_icount overshot maxicount, report actual allocation */
1119 statp->f_files = max_t(typeof(statp->f_files),
1120 statp->f_files,
1121 sbp->sb_icount);
1122
1123 /* make sure statp->f_ffree does not underflow */
1124 ffree = statp->f_files - (icount - ifree);
1125 statp->f_ffree = max_t(int64_t, ffree, 0);
1126
1127
1128 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1129 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1130 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1131 xfs_qm_statvfs(ip, statp);
1132
1133 if (XFS_IS_REALTIME_MOUNT(mp) &&
1134 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1135 statp->f_blocks = sbp->sb_rblocks;
1136 statp->f_bavail = statp->f_bfree =
1137 sbp->sb_frextents * sbp->sb_rextsize;
1138 }
1139
1140 return 0;
1141}
1142
1143STATIC void
1144xfs_save_resvblks(struct xfs_mount *mp)
1145{
1146 uint64_t resblks = 0;
1147
1148 mp->m_resblks_save = mp->m_resblks;
1149 xfs_reserve_blocks(mp, &resblks, NULL);
1150}
1151
1152STATIC void
1153xfs_restore_resvblks(struct xfs_mount *mp)
1154{
1155 uint64_t resblks;
1156
1157 if (mp->m_resblks_save) {
1158 resblks = mp->m_resblks_save;
1159 mp->m_resblks_save = 0;
1160 } else
1161 resblks = xfs_default_resblks(mp);
1162
1163 xfs_reserve_blocks(mp, &resblks, NULL);
1164}
1165
1166/*
1167 * Trigger writeback of all the dirty metadata in the file system.
1168 *
1169 * This ensures that the metadata is written to their location on disk rather
1170 * than just existing in transactions in the log. This means after a quiesce
1171 * there is no log replay required to write the inodes to disk - this is the
1172 * primary difference between a sync and a quiesce.
1173 *
1174 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1175 * it is started again when appropriate.
1176 */
1177void
1178xfs_quiesce_attr(
1179 struct xfs_mount *mp)
1180{
1181 int error = 0;
1182
1183 /* wait for all modifications to complete */
1184 while (atomic_read(&mp->m_active_trans) > 0)
1185 delay(100);
1186
1187 /* force the log to unpin objects from the now complete transactions */
1188 xfs_log_force(mp, XFS_LOG_SYNC);
1189
1190 /* reclaim inodes to do any IO before the freeze completes */
1191 xfs_reclaim_inodes(mp, 0);
1192 xfs_reclaim_inodes(mp, SYNC_WAIT);
1193
1194 /* Push the superblock and write an unmount record */
1195 error = xfs_log_sbcount(mp);
1196 if (error)
1197 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1198 "Frozen image may not be consistent.");
1199 /*
1200 * Just warn here till VFS can correctly support
1201 * read-only remount without racing.
1202 */
1203 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1204
1205 xfs_log_quiesce(mp);
1206}
1207
1208STATIC int
1209xfs_test_remount_options(
1210 struct super_block *sb,
1211 char *options)
1212{
1213 int error = 0;
1214 struct xfs_mount *tmp_mp;
1215
1216 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1217 if (!tmp_mp)
1218 return -ENOMEM;
1219
1220 tmp_mp->m_super = sb;
1221 error = xfs_parseargs(tmp_mp, options);
1222 xfs_free_fsname(tmp_mp);
1223 kmem_free(tmp_mp);
1224
1225 return error;
1226}
1227
1228STATIC int
1229xfs_fs_remount(
1230 struct super_block *sb,
1231 int *flags,
1232 char *options)
1233{
1234 struct xfs_mount *mp = XFS_M(sb);
1235 xfs_sb_t *sbp = &mp->m_sb;
1236 substring_t args[MAX_OPT_ARGS];
1237 char *p;
1238 int error;
1239
1240 /* First, check for complete junk; i.e. invalid options */
1241 error = xfs_test_remount_options(sb, options);
1242 if (error)
1243 return error;
1244
1245 sync_filesystem(sb);
1246 while ((p = strsep(&options, ",")) != NULL) {
1247 int token;
1248
1249 if (!*p)
1250 continue;
1251
1252 token = match_token(p, tokens, args);
1253 switch (token) {
1254 case Opt_inode64:
1255 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1256 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1257 break;
1258 case Opt_inode32:
1259 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1260 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1261 break;
1262 default:
1263 /*
1264 * Logically we would return an error here to prevent
1265 * users from believing they might have changed
1266 * mount options using remount which can't be changed.
1267 *
1268 * But unfortunately mount(8) adds all options from
1269 * mtab and fstab to the mount arguments in some cases
1270 * so we can't blindly reject options, but have to
1271 * check for each specified option if it actually
1272 * differs from the currently set option and only
1273 * reject it if that's the case.
1274 *
1275 * Until that is implemented we return success for
1276 * every remount request, and silently ignore all
1277 * options that we can't actually change.
1278 */
1279#if 0
1280 xfs_info(mp,
1281 "mount option \"%s\" not supported for remount", p);
1282 return -EINVAL;
1283#else
1284 break;
1285#endif
1286 }
1287 }
1288
1289 /* ro -> rw */
1290 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1291 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1292 xfs_warn(mp,
1293 "ro->rw transition prohibited on norecovery mount");
1294 return -EINVAL;
1295 }
1296
1297 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1298 xfs_sb_has_ro_compat_feature(sbp,
1299 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1300 xfs_warn(mp,
1301"ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1302 (sbp->sb_features_ro_compat &
1303 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1304 return -EINVAL;
1305 }
1306
1307 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1308
1309 /*
1310 * If this is the first remount to writeable state we
1311 * might have some superblock changes to update.
1312 */
1313 if (mp->m_update_sb) {
1314 error = xfs_sync_sb(mp, false);
1315 if (error) {
1316 xfs_warn(mp, "failed to write sb changes");
1317 return error;
1318 }
1319 mp->m_update_sb = false;
1320 }
1321
1322 /*
1323 * Fill out the reserve pool if it is empty. Use the stashed
1324 * value if it is non-zero, otherwise go with the default.
1325 */
1326 xfs_restore_resvblks(mp);
1327 xfs_log_work_queue(mp);
1328
1329 /* Recover any CoW blocks that never got remapped. */
1330 error = xfs_reflink_recover_cow(mp);
1331 if (error) {
1332 xfs_err(mp,
1333 "Error %d recovering leftover CoW allocations.", error);
1334 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1335 return error;
1336 }
1337 xfs_start_block_reaping(mp);
1338
1339 /* Create the per-AG metadata reservation pool .*/
1340 error = xfs_fs_reserve_ag_blocks(mp);
1341 if (error && error != -ENOSPC)
1342 return error;
1343 }
1344
1345 /* rw -> ro */
1346 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1347 /*
1348 * Cancel background eofb scanning so it cannot race with the
1349 * final log force+buftarg wait and deadlock the remount.
1350 */
1351 xfs_stop_block_reaping(mp);
1352
1353 /* Get rid of any leftover CoW reservations... */
1354 error = xfs_icache_free_cowblocks(mp, NULL);
1355 if (error) {
1356 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1357 return error;
1358 }
1359
1360 /* Free the per-AG metadata reservation pool. */
1361 error = xfs_fs_unreserve_ag_blocks(mp);
1362 if (error) {
1363 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1364 return error;
1365 }
1366
1367 /*
1368 * Before we sync the metadata, we need to free up the reserve
1369 * block pool so that the used block count in the superblock on
1370 * disk is correct at the end of the remount. Stash the current
1371 * reserve pool size so that if we get remounted rw, we can
1372 * return it to the same size.
1373 */
1374 xfs_save_resvblks(mp);
1375
1376 xfs_quiesce_attr(mp);
1377 mp->m_flags |= XFS_MOUNT_RDONLY;
1378 }
1379
1380 return 0;
1381}
1382
1383/*
1384 * Second stage of a freeze. The data is already frozen so we only
1385 * need to take care of the metadata. Once that's done sync the superblock
1386 * to the log to dirty it in case of a crash while frozen. This ensures that we
1387 * will recover the unlinked inode lists on the next mount.
1388 */
1389STATIC int
1390xfs_fs_freeze(
1391 struct super_block *sb)
1392{
1393 struct xfs_mount *mp = XFS_M(sb);
1394
1395 xfs_stop_block_reaping(mp);
1396 xfs_save_resvblks(mp);
1397 xfs_quiesce_attr(mp);
1398 return xfs_sync_sb(mp, true);
1399}
1400
1401STATIC int
1402xfs_fs_unfreeze(
1403 struct super_block *sb)
1404{
1405 struct xfs_mount *mp = XFS_M(sb);
1406
1407 xfs_restore_resvblks(mp);
1408 xfs_log_work_queue(mp);
1409 xfs_start_block_reaping(mp);
1410 return 0;
1411}
1412
1413STATIC int
1414xfs_fs_show_options(
1415 struct seq_file *m,
1416 struct dentry *root)
1417{
1418 xfs_showargs(XFS_M(root->d_sb), m);
1419 return 0;
1420}
1421
1422/*
1423 * This function fills in xfs_mount_t fields based on mount args.
1424 * Note: the superblock _has_ now been read in.
1425 */
1426STATIC int
1427xfs_finish_flags(
1428 struct xfs_mount *mp)
1429{
1430 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1431
1432 /* Fail a mount where the logbuf is smaller than the log stripe */
1433 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1434 if (mp->m_logbsize <= 0 &&
1435 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1436 mp->m_logbsize = mp->m_sb.sb_logsunit;
1437 } else if (mp->m_logbsize > 0 &&
1438 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1439 xfs_warn(mp,
1440 "logbuf size must be greater than or equal to log stripe size");
1441 return -EINVAL;
1442 }
1443 } else {
1444 /* Fail a mount if the logbuf is larger than 32K */
1445 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1446 xfs_warn(mp,
1447 "logbuf size for version 1 logs must be 16K or 32K");
1448 return -EINVAL;
1449 }
1450 }
1451
1452 /*
1453 * V5 filesystems always use attr2 format for attributes.
1454 */
1455 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1456 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1457 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1458 "attr2 is always enabled for V5 filesystems.");
1459 return -EINVAL;
1460 }
1461
1462 /*
1463 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1464 * told by noattr2 to turn it off
1465 */
1466 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1467 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1468 mp->m_flags |= XFS_MOUNT_ATTR2;
1469
1470 /*
1471 * prohibit r/w mounts of read-only filesystems
1472 */
1473 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1474 xfs_warn(mp,
1475 "cannot mount a read-only filesystem as read-write");
1476 return -EROFS;
1477 }
1478
1479 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1480 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1481 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1482 xfs_warn(mp,
1483 "Super block does not support project and group quota together");
1484 return -EINVAL;
1485 }
1486
1487 return 0;
1488}
1489
1490static int
1491xfs_init_percpu_counters(
1492 struct xfs_mount *mp)
1493{
1494 int error;
1495
1496 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1497 if (error)
1498 return -ENOMEM;
1499
1500 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1501 if (error)
1502 goto free_icount;
1503
1504 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1505 if (error)
1506 goto free_ifree;
1507
1508 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1509 if (error)
1510 goto free_fdblocks;
1511
1512 return 0;
1513
1514free_fdblocks:
1515 percpu_counter_destroy(&mp->m_fdblocks);
1516free_ifree:
1517 percpu_counter_destroy(&mp->m_ifree);
1518free_icount:
1519 percpu_counter_destroy(&mp->m_icount);
1520 return -ENOMEM;
1521}
1522
1523void
1524xfs_reinit_percpu_counters(
1525 struct xfs_mount *mp)
1526{
1527 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1528 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1529 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1530}
1531
1532static void
1533xfs_destroy_percpu_counters(
1534 struct xfs_mount *mp)
1535{
1536 percpu_counter_destroy(&mp->m_icount);
1537 percpu_counter_destroy(&mp->m_ifree);
1538 percpu_counter_destroy(&mp->m_fdblocks);
1539 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1540 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1541 percpu_counter_destroy(&mp->m_delalloc_blks);
1542}
1543
1544static struct xfs_mount *
1545xfs_mount_alloc(
1546 struct super_block *sb)
1547{
1548 struct xfs_mount *mp;
1549
1550 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1551 if (!mp)
1552 return NULL;
1553
1554 mp->m_super = sb;
1555 spin_lock_init(&mp->m_sb_lock);
1556 spin_lock_init(&mp->m_agirotor_lock);
1557 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1558 spin_lock_init(&mp->m_perag_lock);
1559 mutex_init(&mp->m_growlock);
1560 atomic_set(&mp->m_active_trans, 0);
1561 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1562 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1563 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1564 mp->m_kobj.kobject.kset = xfs_kset;
1565 /*
1566 * We don't create the finobt per-ag space reservation until after log
1567 * recovery, so we must set this to true so that an ifree transaction
1568 * started during log recovery will not depend on space reservations
1569 * for finobt expansion.
1570 */
1571 mp->m_finobt_nores = true;
1572 return mp;
1573}
1574
1575
1576STATIC int
1577xfs_fs_fill_super(
1578 struct super_block *sb,
1579 void *data,
1580 int silent)
1581{
1582 struct inode *root;
1583 struct xfs_mount *mp = NULL;
1584 int flags = 0, error = -ENOMEM;
1585
1586 /*
1587 * allocate mp and do all low-level struct initializations before we
1588 * attach it to the super
1589 */
1590 mp = xfs_mount_alloc(sb);
1591 if (!mp)
1592 goto out;
1593 sb->s_fs_info = mp;
1594
1595 error = xfs_parseargs(mp, (char *)data);
1596 if (error)
1597 goto out_free_fsname;
1598
1599 sb_min_blocksize(sb, BBSIZE);
1600 sb->s_xattr = xfs_xattr_handlers;
1601 sb->s_export_op = &xfs_export_operations;
1602#ifdef CONFIG_XFS_QUOTA
1603 sb->s_qcop = &xfs_quotactl_operations;
1604 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1605#endif
1606 sb->s_op = &xfs_super_operations;
1607
1608 /*
1609 * Delay mount work if the debug hook is set. This is debug
1610 * instrumention to coordinate simulation of xfs mount failures with
1611 * VFS superblock operations
1612 */
1613 if (xfs_globals.mount_delay) {
1614 xfs_notice(mp, "Delaying mount for %d seconds.",
1615 xfs_globals.mount_delay);
1616 msleep(xfs_globals.mount_delay * 1000);
1617 }
1618
1619 if (silent)
1620 flags |= XFS_MFSI_QUIET;
1621
1622 error = xfs_open_devices(mp);
1623 if (error)
1624 goto out_free_fsname;
1625
1626 error = xfs_init_mount_workqueues(mp);
1627 if (error)
1628 goto out_close_devices;
1629
1630 error = xfs_init_percpu_counters(mp);
1631 if (error)
1632 goto out_destroy_workqueues;
1633
1634 /* Allocate stats memory before we do operations that might use it */
1635 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1636 if (!mp->m_stats.xs_stats) {
1637 error = -ENOMEM;
1638 goto out_destroy_counters;
1639 }
1640
1641 error = xfs_readsb(mp, flags);
1642 if (error)
1643 goto out_free_stats;
1644
1645 error = xfs_finish_flags(mp);
1646 if (error)
1647 goto out_free_sb;
1648
1649 error = xfs_setup_devices(mp);
1650 if (error)
1651 goto out_free_sb;
1652
1653 error = xfs_filestream_mount(mp);
1654 if (error)
1655 goto out_free_sb;
1656
1657 /*
1658 * we must configure the block size in the superblock before we run the
1659 * full mount process as the mount process can lookup and cache inodes.
1660 */
1661 sb->s_magic = XFS_SUPER_MAGIC;
1662 sb->s_blocksize = mp->m_sb.sb_blocksize;
1663 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1664 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1665 sb->s_max_links = XFS_MAXLINK;
1666 sb->s_time_gran = 1;
1667 sb->s_time_min = S32_MIN;
1668 sb->s_time_max = S32_MAX;
1669 sb->s_iflags |= SB_I_CGROUPWB;
1670
1671 set_posix_acl_flag(sb);
1672
1673 /* version 5 superblocks support inode version counters. */
1674 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1675 sb->s_flags |= SB_I_VERSION;
1676
1677 if (mp->m_flags & XFS_MOUNT_DAX) {
1678 bool rtdev_is_dax = false, datadev_is_dax;
1679
1680 xfs_warn(mp,
1681 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1682
1683 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1684 sb->s_blocksize);
1685 if (mp->m_rtdev_targp)
1686 rtdev_is_dax = bdev_dax_supported(
1687 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1688 if (!rtdev_is_dax && !datadev_is_dax) {
1689 xfs_alert(mp,
1690 "DAX unsupported by block device. Turning off DAX.");
1691 mp->m_flags &= ~XFS_MOUNT_DAX;
1692 }
1693 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1694 xfs_alert(mp,
1695 "DAX and reflink cannot be used together!");
1696 error = -EINVAL;
1697 goto out_filestream_unmount;
1698 }
1699 }
1700
1701 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1702 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1703
1704 if (!blk_queue_discard(q)) {
1705 xfs_warn(mp, "mounting with \"discard\" option, but "
1706 "the device does not support discard");
1707 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1708 }
1709 }
1710
1711 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1712 if (mp->m_sb.sb_rblocks) {
1713 xfs_alert(mp,
1714 "reflink not compatible with realtime device!");
1715 error = -EINVAL;
1716 goto out_filestream_unmount;
1717 }
1718
1719 if (xfs_globals.always_cow) {
1720 xfs_info(mp, "using DEBUG-only always_cow mode.");
1721 mp->m_always_cow = true;
1722 }
1723 }
1724
1725 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1726 xfs_alert(mp,
1727 "reverse mapping btree not compatible with realtime device!");
1728 error = -EINVAL;
1729 goto out_filestream_unmount;
1730 }
1731
1732 error = xfs_mountfs(mp);
1733 if (error)
1734 goto out_filestream_unmount;
1735
1736 root = igrab(VFS_I(mp->m_rootip));
1737 if (!root) {
1738 error = -ENOENT;
1739 goto out_unmount;
1740 }
1741 sb->s_root = d_make_root(root);
1742 if (!sb->s_root) {
1743 error = -ENOMEM;
1744 goto out_unmount;
1745 }
1746
1747 return 0;
1748
1749 out_filestream_unmount:
1750 xfs_filestream_unmount(mp);
1751 out_free_sb:
1752 xfs_freesb(mp);
1753 out_free_stats:
1754 free_percpu(mp->m_stats.xs_stats);
1755 out_destroy_counters:
1756 xfs_destroy_percpu_counters(mp);
1757 out_destroy_workqueues:
1758 xfs_destroy_mount_workqueues(mp);
1759 out_close_devices:
1760 xfs_close_devices(mp);
1761 out_free_fsname:
1762 sb->s_fs_info = NULL;
1763 xfs_free_fsname(mp);
1764 kfree(mp);
1765 out:
1766 return error;
1767
1768 out_unmount:
1769 xfs_filestream_unmount(mp);
1770 xfs_unmountfs(mp);
1771 goto out_free_sb;
1772}
1773
1774STATIC void
1775xfs_fs_put_super(
1776 struct super_block *sb)
1777{
1778 struct xfs_mount *mp = XFS_M(sb);
1779
1780 /* if ->fill_super failed, we have no mount to tear down */
1781 if (!sb->s_fs_info)
1782 return;
1783
1784 xfs_notice(mp, "Unmounting Filesystem");
1785 xfs_filestream_unmount(mp);
1786 xfs_unmountfs(mp);
1787
1788 xfs_freesb(mp);
1789 free_percpu(mp->m_stats.xs_stats);
1790 xfs_destroy_percpu_counters(mp);
1791 xfs_destroy_mount_workqueues(mp);
1792 xfs_close_devices(mp);
1793
1794 sb->s_fs_info = NULL;
1795 xfs_free_fsname(mp);
1796 kfree(mp);
1797}
1798
1799STATIC struct dentry *
1800xfs_fs_mount(
1801 struct file_system_type *fs_type,
1802 int flags,
1803 const char *dev_name,
1804 void *data)
1805{
1806 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1807}
1808
1809static long
1810xfs_fs_nr_cached_objects(
1811 struct super_block *sb,
1812 struct shrink_control *sc)
1813{
1814 /* Paranoia: catch incorrect calls during mount setup or teardown */
1815 if (WARN_ON_ONCE(!sb->s_fs_info))
1816 return 0;
1817 return xfs_reclaim_inodes_count(XFS_M(sb));
1818}
1819
1820static long
1821xfs_fs_free_cached_objects(
1822 struct super_block *sb,
1823 struct shrink_control *sc)
1824{
1825 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1826}
1827
1828static const struct super_operations xfs_super_operations = {
1829 .alloc_inode = xfs_fs_alloc_inode,
1830 .destroy_inode = xfs_fs_destroy_inode,
1831 .dirty_inode = xfs_fs_dirty_inode,
1832 .drop_inode = xfs_fs_drop_inode,
1833 .put_super = xfs_fs_put_super,
1834 .sync_fs = xfs_fs_sync_fs,
1835 .freeze_fs = xfs_fs_freeze,
1836 .unfreeze_fs = xfs_fs_unfreeze,
1837 .statfs = xfs_fs_statfs,
1838 .remount_fs = xfs_fs_remount,
1839 .show_options = xfs_fs_show_options,
1840 .nr_cached_objects = xfs_fs_nr_cached_objects,
1841 .free_cached_objects = xfs_fs_free_cached_objects,
1842};
1843
1844static struct file_system_type xfs_fs_type = {
1845 .owner = THIS_MODULE,
1846 .name = "xfs",
1847 .mount = xfs_fs_mount,
1848 .kill_sb = kill_block_super,
1849 .fs_flags = FS_REQUIRES_DEV,
1850};
1851MODULE_ALIAS_FS("xfs");
1852
1853STATIC int __init
1854xfs_init_zones(void)
1855{
1856 if (bioset_init(&xfs_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
1857 offsetof(struct xfs_ioend, io_inline_bio),
1858 BIOSET_NEED_BVECS))
1859 goto out;
1860
1861 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1862 "xfs_log_ticket");
1863 if (!xfs_log_ticket_zone)
1864 goto out_free_ioend_bioset;
1865
1866 xfs_bmap_free_item_zone = kmem_zone_init(
1867 sizeof(struct xfs_extent_free_item),
1868 "xfs_bmap_free_item");
1869 if (!xfs_bmap_free_item_zone)
1870 goto out_destroy_log_ticket_zone;
1871
1872 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1873 "xfs_btree_cur");
1874 if (!xfs_btree_cur_zone)
1875 goto out_destroy_bmap_free_item_zone;
1876
1877 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1878 "xfs_da_state");
1879 if (!xfs_da_state_zone)
1880 goto out_destroy_btree_cur_zone;
1881
1882 xfs_ifork_zone = kmem_zone_init(sizeof(struct xfs_ifork), "xfs_ifork");
1883 if (!xfs_ifork_zone)
1884 goto out_destroy_da_state_zone;
1885
1886 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1887 if (!xfs_trans_zone)
1888 goto out_destroy_ifork_zone;
1889
1890
1891 /*
1892 * The size of the zone allocated buf log item is the maximum
1893 * size possible under XFS. This wastes a little bit of memory,
1894 * but it is much faster.
1895 */
1896 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1897 "xfs_buf_item");
1898 if (!xfs_buf_item_zone)
1899 goto out_destroy_trans_zone;
1900
1901 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1902 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1903 sizeof(xfs_extent_t))), "xfs_efd_item");
1904 if (!xfs_efd_zone)
1905 goto out_destroy_buf_item_zone;
1906
1907 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1908 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1909 sizeof(xfs_extent_t))), "xfs_efi_item");
1910 if (!xfs_efi_zone)
1911 goto out_destroy_efd_zone;
1912
1913 xfs_inode_zone =
1914 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1915 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1916 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1917 if (!xfs_inode_zone)
1918 goto out_destroy_efi_zone;
1919
1920 xfs_ili_zone =
1921 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1922 KM_ZONE_SPREAD, NULL);
1923 if (!xfs_ili_zone)
1924 goto out_destroy_inode_zone;
1925 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1926 "xfs_icr");
1927 if (!xfs_icreate_zone)
1928 goto out_destroy_ili_zone;
1929
1930 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1931 "xfs_rud_item");
1932 if (!xfs_rud_zone)
1933 goto out_destroy_icreate_zone;
1934
1935 xfs_rui_zone = kmem_zone_init(
1936 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1937 "xfs_rui_item");
1938 if (!xfs_rui_zone)
1939 goto out_destroy_rud_zone;
1940
1941 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1942 "xfs_cud_item");
1943 if (!xfs_cud_zone)
1944 goto out_destroy_rui_zone;
1945
1946 xfs_cui_zone = kmem_zone_init(
1947 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1948 "xfs_cui_item");
1949 if (!xfs_cui_zone)
1950 goto out_destroy_cud_zone;
1951
1952 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1953 "xfs_bud_item");
1954 if (!xfs_bud_zone)
1955 goto out_destroy_cui_zone;
1956
1957 xfs_bui_zone = kmem_zone_init(
1958 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1959 "xfs_bui_item");
1960 if (!xfs_bui_zone)
1961 goto out_destroy_bud_zone;
1962
1963 return 0;
1964
1965 out_destroy_bud_zone:
1966 kmem_zone_destroy(xfs_bud_zone);
1967 out_destroy_cui_zone:
1968 kmem_zone_destroy(xfs_cui_zone);
1969 out_destroy_cud_zone:
1970 kmem_zone_destroy(xfs_cud_zone);
1971 out_destroy_rui_zone:
1972 kmem_zone_destroy(xfs_rui_zone);
1973 out_destroy_rud_zone:
1974 kmem_zone_destroy(xfs_rud_zone);
1975 out_destroy_icreate_zone:
1976 kmem_zone_destroy(xfs_icreate_zone);
1977 out_destroy_ili_zone:
1978 kmem_zone_destroy(xfs_ili_zone);
1979 out_destroy_inode_zone:
1980 kmem_zone_destroy(xfs_inode_zone);
1981 out_destroy_efi_zone:
1982 kmem_zone_destroy(xfs_efi_zone);
1983 out_destroy_efd_zone:
1984 kmem_zone_destroy(xfs_efd_zone);
1985 out_destroy_buf_item_zone:
1986 kmem_zone_destroy(xfs_buf_item_zone);
1987 out_destroy_trans_zone:
1988 kmem_zone_destroy(xfs_trans_zone);
1989 out_destroy_ifork_zone:
1990 kmem_zone_destroy(xfs_ifork_zone);
1991 out_destroy_da_state_zone:
1992 kmem_zone_destroy(xfs_da_state_zone);
1993 out_destroy_btree_cur_zone:
1994 kmem_zone_destroy(xfs_btree_cur_zone);
1995 out_destroy_bmap_free_item_zone:
1996 kmem_zone_destroy(xfs_bmap_free_item_zone);
1997 out_destroy_log_ticket_zone:
1998 kmem_zone_destroy(xfs_log_ticket_zone);
1999 out_free_ioend_bioset:
2000 bioset_exit(&xfs_ioend_bioset);
2001 out:
2002 return -ENOMEM;
2003}
2004
2005STATIC void
2006xfs_destroy_zones(void)
2007{
2008 /*
2009 * Make sure all delayed rcu free are flushed before we
2010 * destroy caches.
2011 */
2012 rcu_barrier();
2013 kmem_zone_destroy(xfs_bui_zone);
2014 kmem_zone_destroy(xfs_bud_zone);
2015 kmem_zone_destroy(xfs_cui_zone);
2016 kmem_zone_destroy(xfs_cud_zone);
2017 kmem_zone_destroy(xfs_rui_zone);
2018 kmem_zone_destroy(xfs_rud_zone);
2019 kmem_zone_destroy(xfs_icreate_zone);
2020 kmem_zone_destroy(xfs_ili_zone);
2021 kmem_zone_destroy(xfs_inode_zone);
2022 kmem_zone_destroy(xfs_efi_zone);
2023 kmem_zone_destroy(xfs_efd_zone);
2024 kmem_zone_destroy(xfs_buf_item_zone);
2025 kmem_zone_destroy(xfs_trans_zone);
2026 kmem_zone_destroy(xfs_ifork_zone);
2027 kmem_zone_destroy(xfs_da_state_zone);
2028 kmem_zone_destroy(xfs_btree_cur_zone);
2029 kmem_zone_destroy(xfs_bmap_free_item_zone);
2030 kmem_zone_destroy(xfs_log_ticket_zone);
2031 bioset_exit(&xfs_ioend_bioset);
2032}
2033
2034STATIC int __init
2035xfs_init_workqueues(void)
2036{
2037 /*
2038 * The allocation workqueue can be used in memory reclaim situations
2039 * (writepage path), and parallelism is only limited by the number of
2040 * AGs in all the filesystems mounted. Hence use the default large
2041 * max_active value for this workqueue.
2042 */
2043 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2044 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2045 if (!xfs_alloc_wq)
2046 return -ENOMEM;
2047
2048 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2049 if (!xfs_discard_wq)
2050 goto out_free_alloc_wq;
2051
2052 return 0;
2053out_free_alloc_wq:
2054 destroy_workqueue(xfs_alloc_wq);
2055 return -ENOMEM;
2056}
2057
2058STATIC void
2059xfs_destroy_workqueues(void)
2060{
2061 destroy_workqueue(xfs_discard_wq);
2062 destroy_workqueue(xfs_alloc_wq);
2063}
2064
2065STATIC int __init
2066init_xfs_fs(void)
2067{
2068 int error;
2069
2070 xfs_check_ondisk_structs();
2071
2072 printk(KERN_INFO XFS_VERSION_STRING " with "
2073 XFS_BUILD_OPTIONS " enabled\n");
2074
2075 xfs_dir_startup();
2076
2077 error = xfs_init_zones();
2078 if (error)
2079 goto out;
2080
2081 error = xfs_init_workqueues();
2082 if (error)
2083 goto out_destroy_zones;
2084
2085 error = xfs_mru_cache_init();
2086 if (error)
2087 goto out_destroy_wq;
2088
2089 error = xfs_buf_init();
2090 if (error)
2091 goto out_mru_cache_uninit;
2092
2093 error = xfs_init_procfs();
2094 if (error)
2095 goto out_buf_terminate;
2096
2097 error = xfs_sysctl_register();
2098 if (error)
2099 goto out_cleanup_procfs;
2100
2101 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2102 if (!xfs_kset) {
2103 error = -ENOMEM;
2104 goto out_sysctl_unregister;
2105 }
2106
2107 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2108
2109 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2110 if (!xfsstats.xs_stats) {
2111 error = -ENOMEM;
2112 goto out_kset_unregister;
2113 }
2114
2115 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2116 "stats");
2117 if (error)
2118 goto out_free_stats;
2119
2120#ifdef DEBUG
2121 xfs_dbg_kobj.kobject.kset = xfs_kset;
2122 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2123 if (error)
2124 goto out_remove_stats_kobj;
2125#endif
2126
2127 error = xfs_qm_init();
2128 if (error)
2129 goto out_remove_dbg_kobj;
2130
2131 error = register_filesystem(&xfs_fs_type);
2132 if (error)
2133 goto out_qm_exit;
2134 return 0;
2135
2136 out_qm_exit:
2137 xfs_qm_exit();
2138 out_remove_dbg_kobj:
2139#ifdef DEBUG
2140 xfs_sysfs_del(&xfs_dbg_kobj);
2141 out_remove_stats_kobj:
2142#endif
2143 xfs_sysfs_del(&xfsstats.xs_kobj);
2144 out_free_stats:
2145 free_percpu(xfsstats.xs_stats);
2146 out_kset_unregister:
2147 kset_unregister(xfs_kset);
2148 out_sysctl_unregister:
2149 xfs_sysctl_unregister();
2150 out_cleanup_procfs:
2151 xfs_cleanup_procfs();
2152 out_buf_terminate:
2153 xfs_buf_terminate();
2154 out_mru_cache_uninit:
2155 xfs_mru_cache_uninit();
2156 out_destroy_wq:
2157 xfs_destroy_workqueues();
2158 out_destroy_zones:
2159 xfs_destroy_zones();
2160 out:
2161 return error;
2162}
2163
2164STATIC void __exit
2165exit_xfs_fs(void)
2166{
2167 xfs_qm_exit();
2168 unregister_filesystem(&xfs_fs_type);
2169#ifdef DEBUG
2170 xfs_sysfs_del(&xfs_dbg_kobj);
2171#endif
2172 xfs_sysfs_del(&xfsstats.xs_kobj);
2173 free_percpu(xfsstats.xs_stats);
2174 kset_unregister(xfs_kset);
2175 xfs_sysctl_unregister();
2176 xfs_cleanup_procfs();
2177 xfs_buf_terminate();
2178 xfs_mru_cache_uninit();
2179 xfs_destroy_workqueues();
2180 xfs_destroy_zones();
2181 xfs_uuid_table_free();
2182}
2183
2184module_init(init_xfs_fs);
2185module_exit(exit_xfs_fs);
2186
2187MODULE_AUTHOR("Silicon Graphics, Inc.");
2188MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2189MODULE_LICENSE("GPL");