?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'xtensa-20180225' of git://github.com/jcmvbkbc/linux-xtensa
[cris-mirror.git] / fs / xfs / xfs_super.c
blob7aba628dc5279858a7608a78839b0e32753bbee6
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18
19 #include "xfs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_bmap.h"
30 #include "xfs_alloc.h"
31 #include "xfs_error.h"
32 #include "xfs_fsops.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_log.h"
36 #include "xfs_log_priv.h"
37 #include "xfs_da_btree.h"
38 #include "xfs_dir2.h"
39 #include "xfs_extfree_item.h"
40 #include "xfs_mru_cache.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_icache.h"
43 #include "xfs_trace.h"
44 #include "xfs_icreate_item.h"
45 #include "xfs_filestream.h"
46 #include "xfs_quota.h"
47 #include "xfs_sysfs.h"
48 #include "xfs_ondisk.h"
49 #include "xfs_rmap_item.h"
50 #include "xfs_refcount_item.h"
51 #include "xfs_bmap_item.h"
52 #include "xfs_reflink.h"
53
54 #include <linux/namei.h>
55 #include <linux/dax.h>
56 #include <linux/init.h>
57 #include <linux/slab.h>
58 #include <linux/mount.h>
59 #include <linux/mempool.h>
60 #include <linux/writeback.h>
61 #include <linux/kthread.h>
62 #include <linux/freezer.h>
63 #include <linux/parser.h>
64
65 static const struct super_operations xfs_super_operations;
66 struct bio_set *xfs_ioend_bioset;
67
68 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
69 #ifdef DEBUG
70 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
71 #endif
72
73 /*
74 * Table driven mount option parser.
75 */
76 enum {
77 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
78 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
79 Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
80 Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
81 Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
82 Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
83 Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
84 Opt_uquota, Opt_gquota, Opt_pquota,
85 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
86 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
87 };
88
89 static const match_table_t tokens = {
90 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */
91 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */
92 {Opt_logdev, "logdev=%s"}, /* log device */
93 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */
94 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */
95 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */
96 {Opt_noalign, "noalign"}, /* turn off stripe alignment */
97 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */
98 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */
99 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */
100 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */
101 {Opt_mtpt, "mtpt"}, /* filesystem mount point */
102 {Opt_grpid, "grpid"}, /* group-ID from parent directory */
103 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */
104 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */
105 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */
106 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
107 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */
108 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */
109 {Opt_inode32, "inode32"}, /* inode allocation limited to
110 * XFS_MAXINUMBER_32 */
111 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */
112 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */
113 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */
114 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes
115 * in stat(). */
116 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */
117 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */
118 {Opt_filestreams,"filestreams"},/* use filestreams allocator */
119 {Opt_quota, "quota"}, /* disk quotas (user) */
120 {Opt_noquota, "noquota"}, /* no quotas */
121 {Opt_usrquota, "usrquota"}, /* user quota enabled */
122 {Opt_grpquota, "grpquota"}, /* group quota enabled */
123 {Opt_prjquota, "prjquota"}, /* project quota enabled */
124 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */
125 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */
126 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */
127 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
128 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
129 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
130 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
131 {Opt_discard, "discard"}, /* Discard unused blocks */
132 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */
133
134 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */
135
136 /* Deprecated mount options scheduled for removal */
137 {Opt_barrier, "barrier"}, /* use writer barriers for log write and
138 * unwritten extent conversion */
139 {Opt_nobarrier, "nobarrier"}, /* .. disable */
140
141 {Opt_err, NULL},
142 };
143
144
145 STATIC int
146 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
147 {
148 int last, shift_left_factor = 0, _res;
149 char *value;
150 int ret = 0;
151
152 value = match_strdup(s);
153 if (!value)
154 return -ENOMEM;
155
156 last = strlen(value) - 1;
157 if (value[last] == 'K' || value[last] == 'k') {
158 shift_left_factor = 10;
159 value[last] = '\0';
160 }
161 if (value[last] == 'M' || value[last] == 'm') {
162 shift_left_factor = 20;
163 value[last] = '\0';
164 }
165 if (value[last] == 'G' || value[last] == 'g') {
166 shift_left_factor = 30;
167 value[last] = '\0';
168 }
169
170 if (kstrtoint(value, base, &_res))
171 ret = -EINVAL;
172 kfree(value);
173 *res = _res << shift_left_factor;
174 return ret;
175 }
176
177 /*
178 * This function fills in xfs_mount_t fields based on mount args.
179 * Note: the superblock has _not_ yet been read in.
180 *
181 * Note that this function leaks the various device name allocations on
182 * failure. The caller takes care of them.
183 *
184 * *sb is const because this is also used to test options on the remount
185 * path, and we don't want this to have any side effects at remount time.
186 * Today this function does not change *sb, but just to future-proof...
187 */
188 STATIC int
189 xfs_parseargs(
190 struct xfs_mount *mp,
191 char *options)
192 {
193 const struct super_block *sb = mp->m_super;
194 char *p;
195 substring_t args[MAX_OPT_ARGS];
196 int dsunit = 0;
197 int dswidth = 0;
198 int iosize = 0;
199 uint8_t iosizelog = 0;
200
201 /*
202 * set up the mount name first so all the errors will refer to the
203 * correct device.
204 */
205 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
206 if (!mp->m_fsname)
207 return -ENOMEM;
208 mp->m_fsname_len = strlen(mp->m_fsname) + 1;
209
210 /*
211 * Copy binary VFS mount flags we are interested in.
212 */
213 if (sb_rdonly(sb))
214 mp->m_flags |= XFS_MOUNT_RDONLY;
215 if (sb->s_flags & SB_DIRSYNC)
216 mp->m_flags |= XFS_MOUNT_DIRSYNC;
217 if (sb->s_flags & SB_SYNCHRONOUS)
218 mp->m_flags |= XFS_MOUNT_WSYNC;
219
220 /*
221 * Set some default flags that could be cleared by the mount option
222 * parsing.
223 */
224 mp->m_flags |= XFS_MOUNT_BARRIER;
225 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
226
227 /*
228 * These can be overridden by the mount option parsing.
229 */
230 mp->m_logbufs = -1;
231 mp->m_logbsize = -1;
232
233 if (!options)
234 goto done;
235
236 while ((p = strsep(&options, ",")) != NULL) {
237 int token;
238
239 if (!*p)
240 continue;
241
242 token = match_token(p, tokens, args);
243 switch (token) {
244 case Opt_logbufs:
245 if (match_int(args, &mp->m_logbufs))
246 return -EINVAL;
247 break;
248 case Opt_logbsize:
249 if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
250 return -EINVAL;
251 break;
252 case Opt_logdev:
253 mp->m_logname = match_strdup(args);
254 if (!mp->m_logname)
255 return -ENOMEM;
256 break;
257 case Opt_mtpt:
258 xfs_warn(mp, "%s option not allowed on this system", p);
259 return -EINVAL;
260 case Opt_rtdev:
261 mp->m_rtname = match_strdup(args);
262 if (!mp->m_rtname)
263 return -ENOMEM;
264 break;
265 case Opt_allocsize:
266 case Opt_biosize:
267 if (suffix_kstrtoint(args, 10, &iosize))
268 return -EINVAL;
269 iosizelog = ffs(iosize) - 1;
270 break;
271 case Opt_grpid:
272 case Opt_bsdgroups:
273 mp->m_flags |= XFS_MOUNT_GRPID;
274 break;
275 case Opt_nogrpid:
276 case Opt_sysvgroups:
277 mp->m_flags &= ~XFS_MOUNT_GRPID;
278 break;
279 case Opt_wsync:
280 mp->m_flags |= XFS_MOUNT_WSYNC;
281 break;
282 case Opt_norecovery:
283 mp->m_flags |= XFS_MOUNT_NORECOVERY;
284 break;
285 case Opt_noalign:
286 mp->m_flags |= XFS_MOUNT_NOALIGN;
287 break;
288 case Opt_swalloc:
289 mp->m_flags |= XFS_MOUNT_SWALLOC;
290 break;
291 case Opt_sunit:
292 if (match_int(args, &dsunit))
293 return -EINVAL;
294 break;
295 case Opt_swidth:
296 if (match_int(args, &dswidth))
297 return -EINVAL;
298 break;
299 case Opt_inode32:
300 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
301 break;
302 case Opt_inode64:
303 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
304 break;
305 case Opt_nouuid:
306 mp->m_flags |= XFS_MOUNT_NOUUID;
307 break;
308 case Opt_ikeep:
309 mp->m_flags |= XFS_MOUNT_IKEEP;
310 break;
311 case Opt_noikeep:
312 mp->m_flags &= ~XFS_MOUNT_IKEEP;
313 break;
314 case Opt_largeio:
315 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
316 break;
317 case Opt_nolargeio:
318 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
319 break;
320 case Opt_attr2:
321 mp->m_flags |= XFS_MOUNT_ATTR2;
322 break;
323 case Opt_noattr2:
324 mp->m_flags &= ~XFS_MOUNT_ATTR2;
325 mp->m_flags |= XFS_MOUNT_NOATTR2;
326 break;
327 case Opt_filestreams:
328 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
329 break;
330 case Opt_noquota:
331 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
332 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
333 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
334 break;
335 case Opt_quota:
336 case Opt_uquota:
337 case Opt_usrquota:
338 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
339 XFS_UQUOTA_ENFD);
340 break;
341 case Opt_qnoenforce:
342 case Opt_uqnoenforce:
343 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
344 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
345 break;
346 case Opt_pquota:
347 case Opt_prjquota:
348 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
349 XFS_PQUOTA_ENFD);
350 break;
351 case Opt_pqnoenforce:
352 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
353 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
354 break;
355 case Opt_gquota:
356 case Opt_grpquota:
357 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
358 XFS_GQUOTA_ENFD);
359 break;
360 case Opt_gqnoenforce:
361 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
362 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
363 break;
364 case Opt_discard:
365 mp->m_flags |= XFS_MOUNT_DISCARD;
366 break;
367 case Opt_nodiscard:
368 mp->m_flags &= ~XFS_MOUNT_DISCARD;
369 break;
370 #ifdef CONFIG_FS_DAX
371 case Opt_dax:
372 mp->m_flags |= XFS_MOUNT_DAX;
373 break;
374 #endif
375 case Opt_barrier:
376 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
377 mp->m_flags |= XFS_MOUNT_BARRIER;
378 break;
379 case Opt_nobarrier:
380 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
381 mp->m_flags &= ~XFS_MOUNT_BARRIER;
382 break;
383 default:
384 xfs_warn(mp, "unknown mount option [%s].", p);
385 return -EINVAL;
386 }
387 }
388
389 /*
390 * no recovery flag requires a read-only mount
391 */
392 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
393 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
394 xfs_warn(mp, "no-recovery mounts must be read-only.");
395 return -EINVAL;
396 }
397
398 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
399 xfs_warn(mp,
400 "sunit and swidth options incompatible with the noalign option");
401 return -EINVAL;
402 }
403
404 #ifndef CONFIG_XFS_QUOTA
405 if (XFS_IS_QUOTA_RUNNING(mp)) {
406 xfs_warn(mp, "quota support not available in this kernel.");
407 return -EINVAL;
408 }
409 #endif
410
411 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
412 xfs_warn(mp, "sunit and swidth must be specified together");
413 return -EINVAL;
414 }
415
416 if (dsunit && (dswidth % dsunit != 0)) {
417 xfs_warn(mp,
418 "stripe width (%d) must be a multiple of the stripe unit (%d)",
419 dswidth, dsunit);
420 return -EINVAL;
421 }
422
423 done:
424 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
425 /*
426 * At this point the superblock has not been read
427 * in, therefore we do not know the block size.
428 * Before the mount call ends we will convert
429 * these to FSBs.
430 */
431 mp->m_dalign = dsunit;
432 mp->m_swidth = dswidth;
433 }
434
435 if (mp->m_logbufs != -1 &&
436 mp->m_logbufs != 0 &&
437 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
438 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
439 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
440 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
441 return -EINVAL;
442 }
443 if (mp->m_logbsize != -1 &&
444 mp->m_logbsize != 0 &&
445 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
446 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
447 !is_power_of_2(mp->m_logbsize))) {
448 xfs_warn(mp,
449 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
450 mp->m_logbsize);
451 return -EINVAL;
452 }
453
454 if (iosizelog) {
455 if (iosizelog > XFS_MAX_IO_LOG ||
456 iosizelog < XFS_MIN_IO_LOG) {
457 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
458 iosizelog, XFS_MIN_IO_LOG,
459 XFS_MAX_IO_LOG);
460 return -EINVAL;
461 }
462
463 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
464 mp->m_readio_log = iosizelog;
465 mp->m_writeio_log = iosizelog;
466 }
467
468 return 0;
469 }
470
471 struct proc_xfs_info {
472 uint64_t flag;
473 char *str;
474 };
475
476 STATIC int
477 xfs_showargs(
478 struct xfs_mount *mp,
479 struct seq_file *m)
480 {
481 static struct proc_xfs_info xfs_info_set[] = {
482 /* the few simple ones we can get from the mount struct */
483 { XFS_MOUNT_IKEEP, ",ikeep" },
484 { XFS_MOUNT_WSYNC, ",wsync" },
485 { XFS_MOUNT_NOALIGN, ",noalign" },
486 { XFS_MOUNT_SWALLOC, ",swalloc" },
487 { XFS_MOUNT_NOUUID, ",nouuid" },
488 { XFS_MOUNT_NORECOVERY, ",norecovery" },
489 { XFS_MOUNT_ATTR2, ",attr2" },
490 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
491 { XFS_MOUNT_GRPID, ",grpid" },
492 { XFS_MOUNT_DISCARD, ",discard" },
493 { XFS_MOUNT_SMALL_INUMS, ",inode32" },
494 { XFS_MOUNT_DAX, ",dax" },
495 { 0, NULL }
496 };
497 static struct proc_xfs_info xfs_info_unset[] = {
498 /* the few simple ones we can get from the mount struct */
499 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" },
500 { XFS_MOUNT_BARRIER, ",nobarrier" },
501 { XFS_MOUNT_SMALL_INUMS, ",inode64" },
502 { 0, NULL }
503 };
504 struct proc_xfs_info *xfs_infop;
505
506 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
507 if (mp->m_flags & xfs_infop->flag)
508 seq_puts(m, xfs_infop->str);
509 }
510 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
511 if (!(mp->m_flags & xfs_infop->flag))
512 seq_puts(m, xfs_infop->str);
513 }
514
515 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
516 seq_printf(m, ",allocsize=%dk",
517 (int)(1 << mp->m_writeio_log) >> 10);
518
519 if (mp->m_logbufs > 0)
520 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
521 if (mp->m_logbsize > 0)
522 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
523
524 if (mp->m_logname)
525 seq_show_option(m, "logdev", mp->m_logname);
526 if (mp->m_rtname)
527 seq_show_option(m, "rtdev", mp->m_rtname);
528
529 if (mp->m_dalign > 0)
530 seq_printf(m, ",sunit=%d",
531 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
532 if (mp->m_swidth > 0)
533 seq_printf(m, ",swidth=%d",
534 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
535
536 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
537 seq_puts(m, ",usrquota");
538 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
539 seq_puts(m, ",uqnoenforce");
540
541 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
542 if (mp->m_qflags & XFS_PQUOTA_ENFD)
543 seq_puts(m, ",prjquota");
544 else
545 seq_puts(m, ",pqnoenforce");
546 }
547 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
548 if (mp->m_qflags & XFS_GQUOTA_ENFD)
549 seq_puts(m, ",grpquota");
550 else
551 seq_puts(m, ",gqnoenforce");
552 }
553
554 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
555 seq_puts(m, ",noquota");
556
557 return 0;
558 }
559 static uint64_t
560 xfs_max_file_offset(
561 unsigned int blockshift)
562 {
563 unsigned int pagefactor = 1;
564 unsigned int bitshift = BITS_PER_LONG - 1;
565
566 /* Figure out maximum filesize, on Linux this can depend on
567 * the filesystem blocksize (on 32 bit platforms).
568 * __block_write_begin does this in an [unsigned] long...
569 * page->index << (PAGE_SHIFT - bbits)
570 * So, for page sized blocks (4K on 32 bit platforms),
571 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
572 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
573 * but for smaller blocksizes it is less (bbits = log2 bsize).
574 * Note1: get_block_t takes a long (implicit cast from above)
575 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
576 * can optionally convert the [unsigned] long from above into
577 * an [unsigned] long long.
578 */
579
580 #if BITS_PER_LONG == 32
581 # if defined(CONFIG_LBDAF)
582 ASSERT(sizeof(sector_t) == 8);
583 pagefactor = PAGE_SIZE;
584 bitshift = BITS_PER_LONG;
585 # else
586 pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
587 # endif
588 #endif
589
590 return (((uint64_t)pagefactor) << bitshift) - 1;
591 }
592
593 /*
594 * Set parameters for inode allocation heuristics, taking into account
595 * filesystem size and inode32/inode64 mount options; i.e. specifically
596 * whether or not XFS_MOUNT_SMALL_INUMS is set.
597 *
598 * Inode allocation patterns are altered only if inode32 is requested
599 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
600 * If altered, XFS_MOUNT_32BITINODES is set as well.
601 *
602 * An agcount independent of that in the mount structure is provided
603 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
604 * to the potentially higher ag count.
605 *
606 * Returns the maximum AG index which may contain inodes.
607 */
608 xfs_agnumber_t
609 xfs_set_inode_alloc(
610 struct xfs_mount *mp,
611 xfs_agnumber_t agcount)
612 {
613 xfs_agnumber_t index;
614 xfs_agnumber_t maxagi = 0;
615 xfs_sb_t *sbp = &mp->m_sb;
616 xfs_agnumber_t max_metadata;
617 xfs_agino_t agino;
618 xfs_ino_t ino;
619
620 /*
621 * Calculate how much should be reserved for inodes to meet
622 * the max inode percentage. Used only for inode32.
623 */
624 if (mp->m_maxicount) {
625 uint64_t icount;
626
627 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
628 do_div(icount, 100);
629 icount += sbp->sb_agblocks - 1;
630 do_div(icount, sbp->sb_agblocks);
631 max_metadata = icount;
632 } else {
633 max_metadata = agcount;
634 }
635
636 /* Get the last possible inode in the filesystem */
637 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
638 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
639
640 /*
641 * If user asked for no more than 32-bit inodes, and the fs is
642 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
643 * the allocator to accommodate the request.
644 */
645 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
646 mp->m_flags |= XFS_MOUNT_32BITINODES;
647 else
648 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
649
650 for (index = 0; index < agcount; index++) {
651 struct xfs_perag *pag;
652
653 ino = XFS_AGINO_TO_INO(mp, index, agino);
654
655 pag = xfs_perag_get(mp, index);
656
657 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
658 if (ino > XFS_MAXINUMBER_32) {
659 pag->pagi_inodeok = 0;
660 pag->pagf_metadata = 0;
661 } else {
662 pag->pagi_inodeok = 1;
663 maxagi++;
664 if (index < max_metadata)
665 pag->pagf_metadata = 1;
666 else
667 pag->pagf_metadata = 0;
668 }
669 } else {
670 pag->pagi_inodeok = 1;
671 pag->pagf_metadata = 0;
672 }
673
674 xfs_perag_put(pag);
675 }
676
677 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
678 }
679
680 STATIC int
681 xfs_blkdev_get(
682 xfs_mount_t *mp,
683 const char *name,
684 struct block_device **bdevp)
685 {
686 int error = 0;
687
688 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
689 mp);
690 if (IS_ERR(*bdevp)) {
691 error = PTR_ERR(*bdevp);
692 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
693 }
694
695 return error;
696 }
697
698 STATIC void
699 xfs_blkdev_put(
700 struct block_device *bdev)
701 {
702 if (bdev)
703 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
704 }
705
706 void
707 xfs_blkdev_issue_flush(
708 xfs_buftarg_t *buftarg)
709 {
710 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
711 }
712
713 STATIC void
714 xfs_close_devices(
715 struct xfs_mount *mp)
716 {
717 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
718
719 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
720 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
721 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
722
723 xfs_free_buftarg(mp, mp->m_logdev_targp);
724 xfs_blkdev_put(logdev);
725 fs_put_dax(dax_logdev);
726 }
727 if (mp->m_rtdev_targp) {
728 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
729 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
730
731 xfs_free_buftarg(mp, mp->m_rtdev_targp);
732 xfs_blkdev_put(rtdev);
733 fs_put_dax(dax_rtdev);
734 }
735 xfs_free_buftarg(mp, mp->m_ddev_targp);
736 fs_put_dax(dax_ddev);
737 }
738
739 /*
740 * The file system configurations are:
741 * (1) device (partition) with data and internal log
742 * (2) logical volume with data and log subvolumes.
743 * (3) logical volume with data, log, and realtime subvolumes.
744 *
745 * We only have to handle opening the log and realtime volumes here if
746 * they are present. The data subvolume has already been opened by
747 * get_sb_bdev() and is stored in sb->s_bdev.
748 */
749 STATIC int
750 xfs_open_devices(
751 struct xfs_mount *mp)
752 {
753 struct block_device *ddev = mp->m_super->s_bdev;
754 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
755 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
756 struct block_device *logdev = NULL, *rtdev = NULL;
757 int error;
758
759 /*
760 * Open real time and log devices - order is important.
761 */
762 if (mp->m_logname) {
763 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
764 if (error)
765 goto out;
766 dax_logdev = fs_dax_get_by_bdev(logdev);
767 }
768
769 if (mp->m_rtname) {
770 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
771 if (error)
772 goto out_close_logdev;
773
774 if (rtdev == ddev || rtdev == logdev) {
775 xfs_warn(mp,
776 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
777 error = -EINVAL;
778 goto out_close_rtdev;
779 }
780 dax_rtdev = fs_dax_get_by_bdev(rtdev);
781 }
782
783 /*
784 * Setup xfs_mount buffer target pointers
785 */
786 error = -ENOMEM;
787 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
788 if (!mp->m_ddev_targp)
789 goto out_close_rtdev;
790
791 if (rtdev) {
792 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
793 if (!mp->m_rtdev_targp)
794 goto out_free_ddev_targ;
795 }
796
797 if (logdev && logdev != ddev) {
798 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
799 if (!mp->m_logdev_targp)
800 goto out_free_rtdev_targ;
801 } else {
802 mp->m_logdev_targp = mp->m_ddev_targp;
803 }
804
805 return 0;
806
807 out_free_rtdev_targ:
808 if (mp->m_rtdev_targp)
809 xfs_free_buftarg(mp, mp->m_rtdev_targp);
810 out_free_ddev_targ:
811 xfs_free_buftarg(mp, mp->m_ddev_targp);
812 out_close_rtdev:
813 xfs_blkdev_put(rtdev);
814 fs_put_dax(dax_rtdev);
815 out_close_logdev:
816 if (logdev && logdev != ddev) {
817 xfs_blkdev_put(logdev);
818 fs_put_dax(dax_logdev);
819 }
820 out:
821 fs_put_dax(dax_ddev);
822 return error;
823 }
824
825 /*
826 * Setup xfs_mount buffer target pointers based on superblock
827 */
828 STATIC int
829 xfs_setup_devices(
830 struct xfs_mount *mp)
831 {
832 int error;
833
834 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
835 if (error)
836 return error;
837
838 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
839 unsigned int log_sector_size = BBSIZE;
840
841 if (xfs_sb_version_hassector(&mp->m_sb))
842 log_sector_size = mp->m_sb.sb_logsectsize;
843 error = xfs_setsize_buftarg(mp->m_logdev_targp,
844 log_sector_size);
845 if (error)
846 return error;
847 }
848 if (mp->m_rtdev_targp) {
849 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
850 mp->m_sb.sb_sectsize);
851 if (error)
852 return error;
853 }
854
855 return 0;
856 }
857
858 STATIC int
859 xfs_init_mount_workqueues(
860 struct xfs_mount *mp)
861 {
862 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
863 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
864 if (!mp->m_buf_workqueue)
865 goto out;
866
867 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
868 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
869 if (!mp->m_data_workqueue)
870 goto out_destroy_buf;
871
872 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
873 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
874 if (!mp->m_unwritten_workqueue)
875 goto out_destroy_data_iodone_queue;
876
877 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
878 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
879 if (!mp->m_cil_workqueue)
880 goto out_destroy_unwritten;
881
882 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
883 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
884 if (!mp->m_reclaim_workqueue)
885 goto out_destroy_cil;
886
887 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
888 WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
889 mp->m_fsname);
890 if (!mp->m_log_workqueue)
891 goto out_destroy_reclaim;
892
893 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
894 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
895 if (!mp->m_eofblocks_workqueue)
896 goto out_destroy_log;
897
898 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
899 mp->m_fsname);
900 if (!mp->m_sync_workqueue)
901 goto out_destroy_eofb;
902
903 return 0;
904
905 out_destroy_eofb:
906 destroy_workqueue(mp->m_eofblocks_workqueue);
907 out_destroy_log:
908 destroy_workqueue(mp->m_log_workqueue);
909 out_destroy_reclaim:
910 destroy_workqueue(mp->m_reclaim_workqueue);
911 out_destroy_cil:
912 destroy_workqueue(mp->m_cil_workqueue);
913 out_destroy_unwritten:
914 destroy_workqueue(mp->m_unwritten_workqueue);
915 out_destroy_data_iodone_queue:
916 destroy_workqueue(mp->m_data_workqueue);
917 out_destroy_buf:
918 destroy_workqueue(mp->m_buf_workqueue);
919 out:
920 return -ENOMEM;
921 }
922
923 STATIC void
924 xfs_destroy_mount_workqueues(
925 struct xfs_mount *mp)
926 {
927 destroy_workqueue(mp->m_sync_workqueue);
928 destroy_workqueue(mp->m_eofblocks_workqueue);
929 destroy_workqueue(mp->m_log_workqueue);
930 destroy_workqueue(mp->m_reclaim_workqueue);
931 destroy_workqueue(mp->m_cil_workqueue);
932 destroy_workqueue(mp->m_data_workqueue);
933 destroy_workqueue(mp->m_unwritten_workqueue);
934 destroy_workqueue(mp->m_buf_workqueue);
935 }
936
937 /*
938 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
939 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
940 * for IO to complete so that we effectively throttle multiple callers to the
941 * rate at which IO is completing.
942 */
943 void
944 xfs_flush_inodes(
945 struct xfs_mount *mp)
946 {
947 struct super_block *sb = mp->m_super;
948
949 if (down_read_trylock(&sb->s_umount)) {
950 sync_inodes_sb(sb);
951 up_read(&sb->s_umount);
952 }
953 }
954
955 /* Catch misguided souls that try to use this interface on XFS */
956 STATIC struct inode *
957 xfs_fs_alloc_inode(
958 struct super_block *sb)
959 {
960 BUG();
961 return NULL;
962 }
963
964 /*
965 * Now that the generic code is guaranteed not to be accessing
966 * the linux inode, we can inactivate and reclaim the inode.
967 */
968 STATIC void
969 xfs_fs_destroy_inode(
970 struct inode *inode)
971 {
972 struct xfs_inode *ip = XFS_I(inode);
973 int error;
974
975 trace_xfs_destroy_inode(ip);
976
977 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
978 XFS_STATS_INC(ip->i_mount, vn_rele);
979 XFS_STATS_INC(ip->i_mount, vn_remove);
980
981 if (xfs_is_reflink_inode(ip)) {
982 error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
983 if (error && !XFS_FORCED_SHUTDOWN(ip->i_mount))
984 xfs_warn(ip->i_mount,
985 "Error %d while evicting CoW blocks for inode %llu.",
986 error, ip->i_ino);
987 }
988
989 xfs_inactive(ip);
990
991 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
992 XFS_STATS_INC(ip->i_mount, vn_reclaim);
993
994 /*
995 * We should never get here with one of the reclaim flags already set.
996 */
997 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
998 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
999
1000 /*
1001 * We always use background reclaim here because even if the
1002 * inode is clean, it still may be under IO and hence we have
1003 * to take the flush lock. The background reclaim path handles
1004 * this more efficiently than we can here, so simply let background
1005 * reclaim tear down all inodes.
1006 */
1007 xfs_inode_set_reclaim_tag(ip);
1008 }
1009
1010 /*
1011 * Slab object creation initialisation for the XFS inode.
1012 * This covers only the idempotent fields in the XFS inode;
1013 * all other fields need to be initialised on allocation
1014 * from the slab. This avoids the need to repeatedly initialise
1015 * fields in the xfs inode that left in the initialise state
1016 * when freeing the inode.
1017 */
1018 STATIC void
1019 xfs_fs_inode_init_once(
1020 void *inode)
1021 {
1022 struct xfs_inode *ip = inode;
1023
1024 memset(ip, 0, sizeof(struct xfs_inode));
1025
1026 /* vfs inode */
1027 inode_init_once(VFS_I(ip));
1028
1029 /* xfs inode */
1030 atomic_set(&ip->i_pincount, 0);
1031 spin_lock_init(&ip->i_flags_lock);
1032
1033 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1034 "xfsino", ip->i_ino);
1035 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1036 "xfsino", ip->i_ino);
1037 }
1038
1039 /*
1040 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1041 * serialised against cache hits here via the inode->i_lock and igrab() in
1042 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1043 * racing with us, and it avoids needing to grab a spinlock here for every inode
1044 * we drop the final reference on.
1045 */
1046 STATIC int
1047 xfs_fs_drop_inode(
1048 struct inode *inode)
1049 {
1050 struct xfs_inode *ip = XFS_I(inode);
1051
1052 /*
1053 * If this unlinked inode is in the middle of recovery, don't
1054 * drop the inode just yet; log recovery will take care of
1055 * that. See the comment for this inode flag.
1056 */
1057 if (ip->i_flags & XFS_IRECOVERY) {
1058 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1059 return 0;
1060 }
1061
1062 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1063 }
1064
1065 STATIC void
1066 xfs_free_fsname(
1067 struct xfs_mount *mp)
1068 {
1069 kfree(mp->m_fsname);
1070 kfree(mp->m_rtname);
1071 kfree(mp->m_logname);
1072 }
1073
1074 STATIC int
1075 xfs_fs_sync_fs(
1076 struct super_block *sb,
1077 int wait)
1078 {
1079 struct xfs_mount *mp = XFS_M(sb);
1080
1081 /*
1082 * Doing anything during the async pass would be counterproductive.
1083 */
1084 if (!wait)
1085 return 0;
1086
1087 xfs_log_force(mp, XFS_LOG_SYNC);
1088 if (laptop_mode) {
1089 /*
1090 * The disk must be active because we're syncing.
1091 * We schedule log work now (now that the disk is
1092 * active) instead of later (when it might not be).
1093 */
1094 flush_delayed_work(&mp->m_log->l_work);
1095 }
1096
1097 return 0;
1098 }
1099
1100 STATIC int
1101 xfs_fs_statfs(
1102 struct dentry *dentry,
1103 struct kstatfs *statp)
1104 {
1105 struct xfs_mount *mp = XFS_M(dentry->d_sb);
1106 xfs_sb_t *sbp = &mp->m_sb;
1107 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1108 uint64_t fakeinos, id;
1109 uint64_t icount;
1110 uint64_t ifree;
1111 uint64_t fdblocks;
1112 xfs_extlen_t lsize;
1113 int64_t ffree;
1114
1115 statp->f_type = XFS_SB_MAGIC;
1116 statp->f_namelen = MAXNAMELEN - 1;
1117
1118 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1119 statp->f_fsid.val[0] = (u32)id;
1120 statp->f_fsid.val[1] = (u32)(id >> 32);
1121
1122 icount = percpu_counter_sum(&mp->m_icount);
1123 ifree = percpu_counter_sum(&mp->m_ifree);
1124 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1125
1126 spin_lock(&mp->m_sb_lock);
1127 statp->f_bsize = sbp->sb_blocksize;
1128 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1129 statp->f_blocks = sbp->sb_dblocks - lsize;
1130 spin_unlock(&mp->m_sb_lock);
1131
1132 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1133 statp->f_bavail = statp->f_bfree;
1134
1135 fakeinos = statp->f_bfree << sbp->sb_inopblog;
1136 statp->f_files = MIN(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1137 if (mp->m_maxicount)
1138 statp->f_files = min_t(typeof(statp->f_files),
1139 statp->f_files,
1140 mp->m_maxicount);
1141
1142 /* If sb_icount overshot maxicount, report actual allocation */
1143 statp->f_files = max_t(typeof(statp->f_files),
1144 statp->f_files,
1145 sbp->sb_icount);
1146
1147 /* make sure statp->f_ffree does not underflow */
1148 ffree = statp->f_files - (icount - ifree);
1149 statp->f_ffree = max_t(int64_t, ffree, 0);
1150
1151
1152 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1153 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1154 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1155 xfs_qm_statvfs(ip, statp);
1156
1157 if (XFS_IS_REALTIME_MOUNT(mp) &&
1158 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1159 statp->f_blocks = sbp->sb_rblocks;
1160 statp->f_bavail = statp->f_bfree =
1161 sbp->sb_frextents * sbp->sb_rextsize;
1162 }
1163
1164 return 0;
1165 }
1166
1167 STATIC void
1168 xfs_save_resvblks(struct xfs_mount *mp)
1169 {
1170 uint64_t resblks = 0;
1171
1172 mp->m_resblks_save = mp->m_resblks;
1173 xfs_reserve_blocks(mp, &resblks, NULL);
1174 }
1175
1176 STATIC void
1177 xfs_restore_resvblks(struct xfs_mount *mp)
1178 {
1179 uint64_t resblks;
1180
1181 if (mp->m_resblks_save) {
1182 resblks = mp->m_resblks_save;
1183 mp->m_resblks_save = 0;
1184 } else
1185 resblks = xfs_default_resblks(mp);
1186
1187 xfs_reserve_blocks(mp, &resblks, NULL);
1188 }
1189
1190 /*
1191 * Trigger writeback of all the dirty metadata in the file system.
1192 *
1193 * This ensures that the metadata is written to their location on disk rather
1194 * than just existing in transactions in the log. This means after a quiesce
1195 * there is no log replay required to write the inodes to disk - this is the
1196 * primary difference between a sync and a quiesce.
1197 *
1198 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1199 * it is started again when appropriate.
1200 */
1201 void
1202 xfs_quiesce_attr(
1203 struct xfs_mount *mp)
1204 {
1205 int error = 0;
1206
1207 /* wait for all modifications to complete */
1208 while (atomic_read(&mp->m_active_trans) > 0)
1209 delay(100);
1210
1211 /* force the log to unpin objects from the now complete transactions */
1212 xfs_log_force(mp, XFS_LOG_SYNC);
1213
1214 /* reclaim inodes to do any IO before the freeze completes */
1215 xfs_reclaim_inodes(mp, 0);
1216 xfs_reclaim_inodes(mp, SYNC_WAIT);
1217
1218 /* Push the superblock and write an unmount record */
1219 error = xfs_log_sbcount(mp);
1220 if (error)
1221 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1222 "Frozen image may not be consistent.");
1223 /*
1224 * Just warn here till VFS can correctly support
1225 * read-only remount without racing.
1226 */
1227 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1228
1229 xfs_log_quiesce(mp);
1230 }
1231
1232 STATIC int
1233 xfs_test_remount_options(
1234 struct super_block *sb,
1235 struct xfs_mount *mp,
1236 char *options)
1237 {
1238 int error = 0;
1239 struct xfs_mount *tmp_mp;
1240
1241 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1242 if (!tmp_mp)
1243 return -ENOMEM;
1244
1245 tmp_mp->m_super = sb;
1246 error = xfs_parseargs(tmp_mp, options);
1247 xfs_free_fsname(tmp_mp);
1248 kmem_free(tmp_mp);
1249
1250 return error;
1251 }
1252
1253 STATIC int
1254 xfs_fs_remount(
1255 struct super_block *sb,
1256 int *flags,
1257 char *options)
1258 {
1259 struct xfs_mount *mp = XFS_M(sb);
1260 xfs_sb_t *sbp = &mp->m_sb;
1261 substring_t args[MAX_OPT_ARGS];
1262 char *p;
1263 int error;
1264
1265 /* First, check for complete junk; i.e. invalid options */
1266 error = xfs_test_remount_options(sb, mp, options);
1267 if (error)
1268 return error;
1269
1270 sync_filesystem(sb);
1271 while ((p = strsep(&options, ",")) != NULL) {
1272 int token;
1273
1274 if (!*p)
1275 continue;
1276
1277 token = match_token(p, tokens, args);
1278 switch (token) {
1279 case Opt_barrier:
1280 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1281 mp->m_flags |= XFS_MOUNT_BARRIER;
1282 break;
1283 case Opt_nobarrier:
1284 xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1285 mp->m_flags &= ~XFS_MOUNT_BARRIER;
1286 break;
1287 case Opt_inode64:
1288 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1289 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1290 break;
1291 case Opt_inode32:
1292 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1293 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1294 break;
1295 default:
1296 /*
1297 * Logically we would return an error here to prevent
1298 * users from believing they might have changed
1299 * mount options using remount which can't be changed.
1300 *
1301 * But unfortunately mount(8) adds all options from
1302 * mtab and fstab to the mount arguments in some cases
1303 * so we can't blindly reject options, but have to
1304 * check for each specified option if it actually
1305 * differs from the currently set option and only
1306 * reject it if that's the case.
1307 *
1308 * Until that is implemented we return success for
1309 * every remount request, and silently ignore all
1310 * options that we can't actually change.
1311 */
1312 #if 0
1313 xfs_info(mp,
1314 "mount option \"%s\" not supported for remount", p);
1315 return -EINVAL;
1316 #else
1317 break;
1318 #endif
1319 }
1320 }
1321
1322 /* ro -> rw */
1323 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1324 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1325 xfs_warn(mp,
1326 "ro->rw transition prohibited on norecovery mount");
1327 return -EINVAL;
1328 }
1329
1330 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1331 xfs_sb_has_ro_compat_feature(sbp,
1332 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1333 xfs_warn(mp,
1334 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1335 (sbp->sb_features_ro_compat &
1336 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1337 return -EINVAL;
1338 }
1339
1340 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1341
1342 /*
1343 * If this is the first remount to writeable state we
1344 * might have some superblock changes to update.
1345 */
1346 if (mp->m_update_sb) {
1347 error = xfs_sync_sb(mp, false);
1348 if (error) {
1349 xfs_warn(mp, "failed to write sb changes");
1350 return error;
1351 }
1352 mp->m_update_sb = false;
1353 }
1354
1355 /*
1356 * Fill out the reserve pool if it is empty. Use the stashed
1357 * value if it is non-zero, otherwise go with the default.
1358 */
1359 xfs_restore_resvblks(mp);
1360 xfs_log_work_queue(mp);
1361 xfs_queue_eofblocks(mp);
1362
1363 /* Recover any CoW blocks that never got remapped. */
1364 error = xfs_reflink_recover_cow(mp);
1365 if (error) {
1366 xfs_err(mp,
1367 "Error %d recovering leftover CoW allocations.", error);
1368 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1369 return error;
1370 }
1371 xfs_queue_cowblocks(mp);
1372
1373 /* Create the per-AG metadata reservation pool .*/
1374 error = xfs_fs_reserve_ag_blocks(mp);
1375 if (error && error != -ENOSPC)
1376 return error;
1377 }
1378
1379 /* rw -> ro */
1380 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1381 /* Get rid of any leftover CoW reservations... */
1382 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1383 error = xfs_icache_free_cowblocks(mp, NULL);
1384 if (error) {
1385 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1386 return error;
1387 }
1388
1389 /* Free the per-AG metadata reservation pool. */
1390 error = xfs_fs_unreserve_ag_blocks(mp);
1391 if (error) {
1392 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1393 return error;
1394 }
1395
1396 /*
1397 * Before we sync the metadata, we need to free up the reserve
1398 * block pool so that the used block count in the superblock on
1399 * disk is correct at the end of the remount. Stash the current
1400 * reserve pool size so that if we get remounted rw, we can
1401 * return it to the same size.
1402 */
1403 xfs_save_resvblks(mp);
1404
1405 /*
1406 * Cancel background eofb scanning so it cannot race with the
1407 * final log force+buftarg wait and deadlock the remount.
1408 */
1409 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1410
1411 xfs_quiesce_attr(mp);
1412 mp->m_flags |= XFS_MOUNT_RDONLY;
1413 }
1414
1415 return 0;
1416 }
1417
1418 /*
1419 * Second stage of a freeze. The data is already frozen so we only
1420 * need to take care of the metadata. Once that's done sync the superblock
1421 * to the log to dirty it in case of a crash while frozen. This ensures that we
1422 * will recover the unlinked inode lists on the next mount.
1423 */
1424 STATIC int
1425 xfs_fs_freeze(
1426 struct super_block *sb)
1427 {
1428 struct xfs_mount *mp = XFS_M(sb);
1429
1430 xfs_save_resvblks(mp);
1431 xfs_quiesce_attr(mp);
1432 return xfs_sync_sb(mp, true);
1433 }
1434
1435 STATIC int
1436 xfs_fs_unfreeze(
1437 struct super_block *sb)
1438 {
1439 struct xfs_mount *mp = XFS_M(sb);
1440
1441 xfs_restore_resvblks(mp);
1442 xfs_log_work_queue(mp);
1443 return 0;
1444 }
1445
1446 STATIC int
1447 xfs_fs_show_options(
1448 struct seq_file *m,
1449 struct dentry *root)
1450 {
1451 return xfs_showargs(XFS_M(root->d_sb), m);
1452 }
1453
1454 /*
1455 * This function fills in xfs_mount_t fields based on mount args.
1456 * Note: the superblock _has_ now been read in.
1457 */
1458 STATIC int
1459 xfs_finish_flags(
1460 struct xfs_mount *mp)
1461 {
1462 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1463
1464 /* Fail a mount where the logbuf is smaller than the log stripe */
1465 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1466 if (mp->m_logbsize <= 0 &&
1467 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1468 mp->m_logbsize = mp->m_sb.sb_logsunit;
1469 } else if (mp->m_logbsize > 0 &&
1470 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1471 xfs_warn(mp,
1472 "logbuf size must be greater than or equal to log stripe size");
1473 return -EINVAL;
1474 }
1475 } else {
1476 /* Fail a mount if the logbuf is larger than 32K */
1477 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1478 xfs_warn(mp,
1479 "logbuf size for version 1 logs must be 16K or 32K");
1480 return -EINVAL;
1481 }
1482 }
1483
1484 /*
1485 * V5 filesystems always use attr2 format for attributes.
1486 */
1487 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1488 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1489 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1490 "attr2 is always enabled for V5 filesystems.");
1491 return -EINVAL;
1492 }
1493
1494 /*
1495 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1496 * told by noattr2 to turn it off
1497 */
1498 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1499 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1500 mp->m_flags |= XFS_MOUNT_ATTR2;
1501
1502 /*
1503 * prohibit r/w mounts of read-only filesystems
1504 */
1505 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1506 xfs_warn(mp,
1507 "cannot mount a read-only filesystem as read-write");
1508 return -EROFS;
1509 }
1510
1511 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1512 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1513 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1514 xfs_warn(mp,
1515 "Super block does not support project and group quota together");
1516 return -EINVAL;
1517 }
1518
1519 return 0;
1520 }
1521
1522 static int
1523 xfs_init_percpu_counters(
1524 struct xfs_mount *mp)
1525 {
1526 int error;
1527
1528 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1529 if (error)
1530 return -ENOMEM;
1531
1532 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1533 if (error)
1534 goto free_icount;
1535
1536 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1537 if (error)
1538 goto free_ifree;
1539
1540 return 0;
1541
1542 free_ifree:
1543 percpu_counter_destroy(&mp->m_ifree);
1544 free_icount:
1545 percpu_counter_destroy(&mp->m_icount);
1546 return -ENOMEM;
1547 }
1548
1549 void
1550 xfs_reinit_percpu_counters(
1551 struct xfs_mount *mp)
1552 {
1553 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1554 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1555 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1556 }
1557
1558 static void
1559 xfs_destroy_percpu_counters(
1560 struct xfs_mount *mp)
1561 {
1562 percpu_counter_destroy(&mp->m_icount);
1563 percpu_counter_destroy(&mp->m_ifree);
1564 percpu_counter_destroy(&mp->m_fdblocks);
1565 }
1566
1567 STATIC int
1568 xfs_fs_fill_super(
1569 struct super_block *sb,
1570 void *data,
1571 int silent)
1572 {
1573 struct inode *root;
1574 struct xfs_mount *mp = NULL;
1575 int flags = 0, error = -ENOMEM;
1576
1577 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1578 if (!mp)
1579 goto out;
1580
1581 spin_lock_init(&mp->m_sb_lock);
1582 mutex_init(&mp->m_growlock);
1583 atomic_set(&mp->m_active_trans, 0);
1584 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1585 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1586 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1587 mp->m_kobj.kobject.kset = xfs_kset;
1588
1589 mp->m_super = sb;
1590 sb->s_fs_info = mp;
1591
1592 error = xfs_parseargs(mp, (char *)data);
1593 if (error)
1594 goto out_free_fsname;
1595
1596 sb_min_blocksize(sb, BBSIZE);
1597 sb->s_xattr = xfs_xattr_handlers;
1598 sb->s_export_op = &xfs_export_operations;
1599 #ifdef CONFIG_XFS_QUOTA
1600 sb->s_qcop = &xfs_quotactl_operations;
1601 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1602 #endif
1603 sb->s_op = &xfs_super_operations;
1604
1605 if (silent)
1606 flags |= XFS_MFSI_QUIET;
1607
1608 error = xfs_open_devices(mp);
1609 if (error)
1610 goto out_free_fsname;
1611
1612 error = xfs_init_mount_workqueues(mp);
1613 if (error)
1614 goto out_close_devices;
1615
1616 error = xfs_init_percpu_counters(mp);
1617 if (error)
1618 goto out_destroy_workqueues;
1619
1620 /* Allocate stats memory before we do operations that might use it */
1621 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1622 if (!mp->m_stats.xs_stats) {
1623 error = -ENOMEM;
1624 goto out_destroy_counters;
1625 }
1626
1627 error = xfs_readsb(mp, flags);
1628 if (error)
1629 goto out_free_stats;
1630
1631 error = xfs_finish_flags(mp);
1632 if (error)
1633 goto out_free_sb;
1634
1635 error = xfs_setup_devices(mp);
1636 if (error)
1637 goto out_free_sb;
1638
1639 error = xfs_filestream_mount(mp);
1640 if (error)
1641 goto out_free_sb;
1642
1643 /*
1644 * we must configure the block size in the superblock before we run the
1645 * full mount process as the mount process can lookup and cache inodes.
1646 */
1647 sb->s_magic = XFS_SB_MAGIC;
1648 sb->s_blocksize = mp->m_sb.sb_blocksize;
1649 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1650 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1651 sb->s_max_links = XFS_MAXLINK;
1652 sb->s_time_gran = 1;
1653 set_posix_acl_flag(sb);
1654
1655 /* version 5 superblocks support inode version counters. */
1656 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1657 sb->s_flags |= SB_I_VERSION;
1658
1659 if (mp->m_flags & XFS_MOUNT_DAX) {
1660 xfs_warn(mp,
1661 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1662
1663 error = bdev_dax_supported(sb, sb->s_blocksize);
1664 if (error) {
1665 xfs_alert(mp,
1666 "DAX unsupported by block device. Turning off DAX.");
1667 mp->m_flags &= ~XFS_MOUNT_DAX;
1668 }
1669 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1670 xfs_alert(mp,
1671 "DAX and reflink cannot be used together!");
1672 error = -EINVAL;
1673 goto out_filestream_unmount;
1674 }
1675 }
1676
1677 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1678 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1679
1680 if (!blk_queue_discard(q)) {
1681 xfs_warn(mp, "mounting with \"discard\" option, but "
1682 "the device does not support discard");
1683 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1684 }
1685 }
1686
1687 if (xfs_sb_version_hasreflink(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1688 xfs_alert(mp,
1689 "reflink not compatible with realtime device!");
1690 error = -EINVAL;
1691 goto out_filestream_unmount;
1692 }
1693
1694 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1695 xfs_alert(mp,
1696 "reverse mapping btree not compatible with realtime device!");
1697 error = -EINVAL;
1698 goto out_filestream_unmount;
1699 }
1700
1701 error = xfs_mountfs(mp);
1702 if (error)
1703 goto out_filestream_unmount;
1704
1705 root = igrab(VFS_I(mp->m_rootip));
1706 if (!root) {
1707 error = -ENOENT;
1708 goto out_unmount;
1709 }
1710 sb->s_root = d_make_root(root);
1711 if (!sb->s_root) {
1712 error = -ENOMEM;
1713 goto out_unmount;
1714 }
1715
1716 return 0;
1717
1718 out_filestream_unmount:
1719 xfs_filestream_unmount(mp);
1720 out_free_sb:
1721 xfs_freesb(mp);
1722 out_free_stats:
1723 free_percpu(mp->m_stats.xs_stats);
1724 out_destroy_counters:
1725 xfs_destroy_percpu_counters(mp);
1726 out_destroy_workqueues:
1727 xfs_destroy_mount_workqueues(mp);
1728 out_close_devices:
1729 xfs_close_devices(mp);
1730 out_free_fsname:
1731 xfs_free_fsname(mp);
1732 kfree(mp);
1733 out:
1734 return error;
1735
1736 out_unmount:
1737 xfs_filestream_unmount(mp);
1738 xfs_unmountfs(mp);
1739 goto out_free_sb;
1740 }
1741
1742 STATIC void
1743 xfs_fs_put_super(
1744 struct super_block *sb)
1745 {
1746 struct xfs_mount *mp = XFS_M(sb);
1747
1748 xfs_notice(mp, "Unmounting Filesystem");
1749 xfs_filestream_unmount(mp);
1750 xfs_unmountfs(mp);
1751
1752 xfs_freesb(mp);
1753 free_percpu(mp->m_stats.xs_stats);
1754 xfs_destroy_percpu_counters(mp);
1755 xfs_destroy_mount_workqueues(mp);
1756 xfs_close_devices(mp);
1757 xfs_free_fsname(mp);
1758 kfree(mp);
1759 }
1760
1761 STATIC struct dentry *
1762 xfs_fs_mount(
1763 struct file_system_type *fs_type,
1764 int flags,
1765 const char *dev_name,
1766 void *data)
1767 {
1768 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1769 }
1770
1771 static long
1772 xfs_fs_nr_cached_objects(
1773 struct super_block *sb,
1774 struct shrink_control *sc)
1775 {
1776 return xfs_reclaim_inodes_count(XFS_M(sb));
1777 }
1778
1779 static long
1780 xfs_fs_free_cached_objects(
1781 struct super_block *sb,
1782 struct shrink_control *sc)
1783 {
1784 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1785 }
1786
1787 static const struct super_operations xfs_super_operations = {
1788 .alloc_inode = xfs_fs_alloc_inode,
1789 .destroy_inode = xfs_fs_destroy_inode,
1790 .drop_inode = xfs_fs_drop_inode,
1791 .put_super = xfs_fs_put_super,
1792 .sync_fs = xfs_fs_sync_fs,
1793 .freeze_fs = xfs_fs_freeze,
1794 .unfreeze_fs = xfs_fs_unfreeze,
1795 .statfs = xfs_fs_statfs,
1796 .remount_fs = xfs_fs_remount,
1797 .show_options = xfs_fs_show_options,
1798 .nr_cached_objects = xfs_fs_nr_cached_objects,
1799 .free_cached_objects = xfs_fs_free_cached_objects,
1800 };
1801
1802 static struct file_system_type xfs_fs_type = {
1803 .owner = THIS_MODULE,
1804 .name = "xfs",
1805 .mount = xfs_fs_mount,
1806 .kill_sb = kill_block_super,
1807 .fs_flags = FS_REQUIRES_DEV,
1808 };
1809 MODULE_ALIAS_FS("xfs");
1810
1811 STATIC int __init
1812 xfs_init_zones(void)
1813 {
1814 xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1815 offsetof(struct xfs_ioend, io_inline_bio),
1816 BIOSET_NEED_BVECS);
1817 if (!xfs_ioend_bioset)
1818 goto out;
1819
1820 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1821 "xfs_log_ticket");
1822 if (!xfs_log_ticket_zone)
1823 goto out_free_ioend_bioset;
1824
1825 xfs_bmap_free_item_zone = kmem_zone_init(
1826 sizeof(struct xfs_extent_free_item),
1827 "xfs_bmap_free_item");
1828 if (!xfs_bmap_free_item_zone)
1829 goto out_destroy_log_ticket_zone;
1830
1831 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1832 "xfs_btree_cur");
1833 if (!xfs_btree_cur_zone)
1834 goto out_destroy_bmap_free_item_zone;
1835
1836 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1837 "xfs_da_state");
1838 if (!xfs_da_state_zone)
1839 goto out_destroy_btree_cur_zone;
1840
1841 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1842 if (!xfs_ifork_zone)
1843 goto out_destroy_da_state_zone;
1844
1845 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1846 if (!xfs_trans_zone)
1847 goto out_destroy_ifork_zone;
1848
1849 xfs_log_item_desc_zone =
1850 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1851 "xfs_log_item_desc");
1852 if (!xfs_log_item_desc_zone)
1853 goto out_destroy_trans_zone;
1854
1855 /*
1856 * The size of the zone allocated buf log item is the maximum
1857 * size possible under XFS. This wastes a little bit of memory,
1858 * but it is much faster.
1859 */
1860 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1861 "xfs_buf_item");
1862 if (!xfs_buf_item_zone)
1863 goto out_destroy_log_item_desc_zone;
1864
1865 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1866 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1867 sizeof(xfs_extent_t))), "xfs_efd_item");
1868 if (!xfs_efd_zone)
1869 goto out_destroy_buf_item_zone;
1870
1871 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1872 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1873 sizeof(xfs_extent_t))), "xfs_efi_item");
1874 if (!xfs_efi_zone)
1875 goto out_destroy_efd_zone;
1876
1877 xfs_inode_zone =
1878 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1879 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1880 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1881 if (!xfs_inode_zone)
1882 goto out_destroy_efi_zone;
1883
1884 xfs_ili_zone =
1885 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1886 KM_ZONE_SPREAD, NULL);
1887 if (!xfs_ili_zone)
1888 goto out_destroy_inode_zone;
1889 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1890 "xfs_icr");
1891 if (!xfs_icreate_zone)
1892 goto out_destroy_ili_zone;
1893
1894 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1895 "xfs_rud_item");
1896 if (!xfs_rud_zone)
1897 goto out_destroy_icreate_zone;
1898
1899 xfs_rui_zone = kmem_zone_init(
1900 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1901 "xfs_rui_item");
1902 if (!xfs_rui_zone)
1903 goto out_destroy_rud_zone;
1904
1905 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1906 "xfs_cud_item");
1907 if (!xfs_cud_zone)
1908 goto out_destroy_rui_zone;
1909
1910 xfs_cui_zone = kmem_zone_init(
1911 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1912 "xfs_cui_item");
1913 if (!xfs_cui_zone)
1914 goto out_destroy_cud_zone;
1915
1916 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1917 "xfs_bud_item");
1918 if (!xfs_bud_zone)
1919 goto out_destroy_cui_zone;
1920
1921 xfs_bui_zone = kmem_zone_init(
1922 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1923 "xfs_bui_item");
1924 if (!xfs_bui_zone)
1925 goto out_destroy_bud_zone;
1926
1927 return 0;
1928
1929 out_destroy_bud_zone:
1930 kmem_zone_destroy(xfs_bud_zone);
1931 out_destroy_cui_zone:
1932 kmem_zone_destroy(xfs_cui_zone);
1933 out_destroy_cud_zone:
1934 kmem_zone_destroy(xfs_cud_zone);
1935 out_destroy_rui_zone:
1936 kmem_zone_destroy(xfs_rui_zone);
1937 out_destroy_rud_zone:
1938 kmem_zone_destroy(xfs_rud_zone);
1939 out_destroy_icreate_zone:
1940 kmem_zone_destroy(xfs_icreate_zone);
1941 out_destroy_ili_zone:
1942 kmem_zone_destroy(xfs_ili_zone);
1943 out_destroy_inode_zone:
1944 kmem_zone_destroy(xfs_inode_zone);
1945 out_destroy_efi_zone:
1946 kmem_zone_destroy(xfs_efi_zone);
1947 out_destroy_efd_zone:
1948 kmem_zone_destroy(xfs_efd_zone);
1949 out_destroy_buf_item_zone:
1950 kmem_zone_destroy(xfs_buf_item_zone);
1951 out_destroy_log_item_desc_zone:
1952 kmem_zone_destroy(xfs_log_item_desc_zone);
1953 out_destroy_trans_zone:
1954 kmem_zone_destroy(xfs_trans_zone);
1955 out_destroy_ifork_zone:
1956 kmem_zone_destroy(xfs_ifork_zone);
1957 out_destroy_da_state_zone:
1958 kmem_zone_destroy(xfs_da_state_zone);
1959 out_destroy_btree_cur_zone:
1960 kmem_zone_destroy(xfs_btree_cur_zone);
1961 out_destroy_bmap_free_item_zone:
1962 kmem_zone_destroy(xfs_bmap_free_item_zone);
1963 out_destroy_log_ticket_zone:
1964 kmem_zone_destroy(xfs_log_ticket_zone);
1965 out_free_ioend_bioset:
1966 bioset_free(xfs_ioend_bioset);
1967 out:
1968 return -ENOMEM;
1969 }
1970
1971 STATIC void
1972 xfs_destroy_zones(void)
1973 {
1974 /*
1975 * Make sure all delayed rcu free are flushed before we
1976 * destroy caches.
1977 */
1978 rcu_barrier();
1979 kmem_zone_destroy(xfs_bui_zone);
1980 kmem_zone_destroy(xfs_bud_zone);
1981 kmem_zone_destroy(xfs_cui_zone);
1982 kmem_zone_destroy(xfs_cud_zone);
1983 kmem_zone_destroy(xfs_rui_zone);
1984 kmem_zone_destroy(xfs_rud_zone);
1985 kmem_zone_destroy(xfs_icreate_zone);
1986 kmem_zone_destroy(xfs_ili_zone);
1987 kmem_zone_destroy(xfs_inode_zone);
1988 kmem_zone_destroy(xfs_efi_zone);
1989 kmem_zone_destroy(xfs_efd_zone);
1990 kmem_zone_destroy(xfs_buf_item_zone);
1991 kmem_zone_destroy(xfs_log_item_desc_zone);
1992 kmem_zone_destroy(xfs_trans_zone);
1993 kmem_zone_destroy(xfs_ifork_zone);
1994 kmem_zone_destroy(xfs_da_state_zone);
1995 kmem_zone_destroy(xfs_btree_cur_zone);
1996 kmem_zone_destroy(xfs_bmap_free_item_zone);
1997 kmem_zone_destroy(xfs_log_ticket_zone);
1998 bioset_free(xfs_ioend_bioset);
1999 }
2000
2001 STATIC int __init
2002 xfs_init_workqueues(void)
2003 {
2004 /*
2005 * The allocation workqueue can be used in memory reclaim situations
2006 * (writepage path), and parallelism is only limited by the number of
2007 * AGs in all the filesystems mounted. Hence use the default large
2008 * max_active value for this workqueue.
2009 */
2010 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2011 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2012 if (!xfs_alloc_wq)
2013 return -ENOMEM;
2014
2015 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2016 if (!xfs_discard_wq)
2017 goto out_free_alloc_wq;
2018
2019 return 0;
2020 out_free_alloc_wq:
2021 destroy_workqueue(xfs_alloc_wq);
2022 return -ENOMEM;
2023 }
2024
2025 STATIC void
2026 xfs_destroy_workqueues(void)
2027 {
2028 destroy_workqueue(xfs_discard_wq);
2029 destroy_workqueue(xfs_alloc_wq);
2030 }
2031
2032 STATIC int __init
2033 init_xfs_fs(void)
2034 {
2035 int error;
2036
2037 xfs_check_ondisk_structs();
2038
2039 printk(KERN_INFO XFS_VERSION_STRING " with "
2040 XFS_BUILD_OPTIONS " enabled\n");
2041
2042 xfs_extent_free_init_defer_op();
2043 xfs_rmap_update_init_defer_op();
2044 xfs_refcount_update_init_defer_op();
2045 xfs_bmap_update_init_defer_op();
2046
2047 xfs_dir_startup();
2048
2049 error = xfs_init_zones();
2050 if (error)
2051 goto out;
2052
2053 error = xfs_init_workqueues();
2054 if (error)
2055 goto out_destroy_zones;
2056
2057 error = xfs_mru_cache_init();
2058 if (error)
2059 goto out_destroy_wq;
2060
2061 error = xfs_buf_init();
2062 if (error)
2063 goto out_mru_cache_uninit;
2064
2065 error = xfs_init_procfs();
2066 if (error)
2067 goto out_buf_terminate;
2068
2069 error = xfs_sysctl_register();
2070 if (error)
2071 goto out_cleanup_procfs;
2072
2073 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2074 if (!xfs_kset) {
2075 error = -ENOMEM;
2076 goto out_sysctl_unregister;
2077 }
2078
2079 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2080
2081 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2082 if (!xfsstats.xs_stats) {
2083 error = -ENOMEM;
2084 goto out_kset_unregister;
2085 }
2086
2087 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2088 "stats");
2089 if (error)
2090 goto out_free_stats;
2091
2092 #ifdef DEBUG
2093 xfs_dbg_kobj.kobject.kset = xfs_kset;
2094 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2095 if (error)
2096 goto out_remove_stats_kobj;
2097 #endif
2098
2099 error = xfs_qm_init();
2100 if (error)
2101 goto out_remove_dbg_kobj;
2102
2103 error = register_filesystem(&xfs_fs_type);
2104 if (error)
2105 goto out_qm_exit;
2106 return 0;
2107
2108 out_qm_exit:
2109 xfs_qm_exit();
2110 out_remove_dbg_kobj:
2111 #ifdef DEBUG
2112 xfs_sysfs_del(&xfs_dbg_kobj);
2113 out_remove_stats_kobj:
2114 #endif
2115 xfs_sysfs_del(&xfsstats.xs_kobj);
2116 out_free_stats:
2117 free_percpu(xfsstats.xs_stats);
2118 out_kset_unregister:
2119 kset_unregister(xfs_kset);
2120 out_sysctl_unregister:
2121 xfs_sysctl_unregister();
2122 out_cleanup_procfs:
2123 xfs_cleanup_procfs();
2124 out_buf_terminate:
2125 xfs_buf_terminate();
2126 out_mru_cache_uninit:
2127 xfs_mru_cache_uninit();
2128 out_destroy_wq:
2129 xfs_destroy_workqueues();
2130 out_destroy_zones:
2131 xfs_destroy_zones();
2132 out:
2133 return error;
2134 }
2135
2136 STATIC void __exit
2137 exit_xfs_fs(void)
2138 {
2139 xfs_qm_exit();
2140 unregister_filesystem(&xfs_fs_type);
2141 #ifdef DEBUG
2142 xfs_sysfs_del(&xfs_dbg_kobj);
2143 #endif
2144 xfs_sysfs_del(&xfsstats.xs_kobj);
2145 free_percpu(xfsstats.xs_stats);
2146 kset_unregister(xfs_kset);
2147 xfs_sysctl_unregister();
2148 xfs_cleanup_procfs();
2149 xfs_buf_terminate();
2150 xfs_mru_cache_uninit();
2151 xfs_destroy_workqueues();
2152 xfs_destroy_zones();
2153 xfs_uuid_table_free();
2154 }
2155
2156 module_init(init_xfs_fs);
2157 module_exit(exit_xfs_fs);
2158
2159 MODULE_AUTHOR("Silicon Graphics, Inc.");
2160 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2161 MODULE_LICENSE("GPL");
CRIS linux kernel tree