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
42 static const struct super_operations xfs_super_operations;
43 struct bio_set xfs_ioend_bioset;
44
45 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
46 #ifdef DEBUG
47 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
48 #endif
49
50 /*
51 * Table driven mount option parser.
52 */
53 enum {
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
65 static 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
113 STATIC int
suffix_kstrtoint(const substring_t * s,unsigned int base,int * res)114 suffix_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 */
156 STATIC int
xfs_parseargs(struct xfs_mount * mp,char * options)157 xfs_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
381 done:
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
429 struct proc_xfs_info {
430 uint64_t flag;
431 char *str;
432 };
433
434 STATIC void
xfs_showargs(struct xfs_mount * mp,struct seq_file * m)435 xfs_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) {
494 if (mp->m_qflags & XFS_UQUOTA_ENFD)
495 seq_puts(m, ",usrquota");
496 else
497 seq_puts(m, ",uqnoenforce");
498 }
499
500 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
501 if (mp->m_qflags & XFS_PQUOTA_ENFD)
502 seq_puts(m, ",prjquota");
503 else
504 seq_puts(m, ",pqnoenforce");
505 }
506 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
507 if (mp->m_qflags & XFS_GQUOTA_ENFD)
508 seq_puts(m, ",grpquota");
509 else
510 seq_puts(m, ",gqnoenforce");
511 }
512
513 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
514 seq_puts(m, ",noquota");
515 }
516
517 /*
518 * Set parameters for inode allocation heuristics, taking into account
519 * filesystem size and inode32/inode64 mount options; i.e. specifically
520 * whether or not XFS_MOUNT_SMALL_INUMS is set.
521 *
522 * Inode allocation patterns are altered only if inode32 is requested
523 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
524 * If altered, XFS_MOUNT_32BITINODES is set as well.
525 *
526 * An agcount independent of that in the mount structure is provided
527 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
528 * to the potentially higher ag count.
529 *
530 * Returns the maximum AG index which may contain inodes.
531 */
532 xfs_agnumber_t
xfs_set_inode_alloc(struct xfs_mount * mp,xfs_agnumber_t agcount)533 xfs_set_inode_alloc(
534 struct xfs_mount *mp,
535 xfs_agnumber_t agcount)
536 {
537 xfs_agnumber_t index;
538 xfs_agnumber_t maxagi = 0;
539 xfs_sb_t *sbp = &mp->m_sb;
540 xfs_agnumber_t max_metadata;
541 xfs_agino_t agino;
542 xfs_ino_t ino;
543
544 /*
545 * Calculate how much should be reserved for inodes to meet
546 * the max inode percentage. Used only for inode32.
547 */
548 if (M_IGEO(mp)->maxicount) {
549 uint64_t icount;
550
551 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
552 do_div(icount, 100);
553 icount += sbp->sb_agblocks - 1;
554 do_div(icount, sbp->sb_agblocks);
555 max_metadata = icount;
556 } else {
557 max_metadata = agcount;
558 }
559
560 /* Get the last possible inode in the filesystem */
561 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
562 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
563
564 /*
565 * If user asked for no more than 32-bit inodes, and the fs is
566 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
567 * the allocator to accommodate the request.
568 */
569 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
570 mp->m_flags |= XFS_MOUNT_32BITINODES;
571 else
572 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
573
574 for (index = 0; index < agcount; index++) {
575 struct xfs_perag *pag;
576
577 ino = XFS_AGINO_TO_INO(mp, index, agino);
578
579 pag = xfs_perag_get(mp, index);
580
581 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
582 if (ino > XFS_MAXINUMBER_32) {
583 pag->pagi_inodeok = 0;
584 pag->pagf_metadata = 0;
585 } else {
586 pag->pagi_inodeok = 1;
587 maxagi++;
588 if (index < max_metadata)
589 pag->pagf_metadata = 1;
590 else
591 pag->pagf_metadata = 0;
592 }
593 } else {
594 pag->pagi_inodeok = 1;
595 pag->pagf_metadata = 0;
596 }
597
598 xfs_perag_put(pag);
599 }
600
601 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
602 }
603
604 STATIC int
xfs_blkdev_get(xfs_mount_t * mp,const char * name,struct block_device ** bdevp)605 xfs_blkdev_get(
606 xfs_mount_t *mp,
607 const char *name,
608 struct block_device **bdevp)
609 {
610 int error = 0;
611
612 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
613 mp);
614 if (IS_ERR(*bdevp)) {
615 error = PTR_ERR(*bdevp);
616 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
617 }
618
619 return error;
620 }
621
622 STATIC void
xfs_blkdev_put(struct block_device * bdev)623 xfs_blkdev_put(
624 struct block_device *bdev)
625 {
626 if (bdev)
627 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
628 }
629
630 void
xfs_blkdev_issue_flush(xfs_buftarg_t * buftarg)631 xfs_blkdev_issue_flush(
632 xfs_buftarg_t *buftarg)
633 {
634 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
635 }
636
637 STATIC void
xfs_close_devices(struct xfs_mount * mp)638 xfs_close_devices(
639 struct xfs_mount *mp)
640 {
641 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
642
643 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
644 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
645 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
646
647 xfs_free_buftarg(mp->m_logdev_targp);
648 xfs_blkdev_put(logdev);
649 fs_put_dax(dax_logdev);
650 }
651 if (mp->m_rtdev_targp) {
652 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
653 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
654
655 xfs_free_buftarg(mp->m_rtdev_targp);
656 xfs_blkdev_put(rtdev);
657 fs_put_dax(dax_rtdev);
658 }
659 xfs_free_buftarg(mp->m_ddev_targp);
660 fs_put_dax(dax_ddev);
661 }
662
663 /*
664 * The file system configurations are:
665 * (1) device (partition) with data and internal log
666 * (2) logical volume with data and log subvolumes.
667 * (3) logical volume with data, log, and realtime subvolumes.
668 *
669 * We only have to handle opening the log and realtime volumes here if
670 * they are present. The data subvolume has already been opened by
671 * get_sb_bdev() and is stored in sb->s_bdev.
672 */
673 STATIC int
xfs_open_devices(struct xfs_mount * mp)674 xfs_open_devices(
675 struct xfs_mount *mp)
676 {
677 struct block_device *ddev = mp->m_super->s_bdev;
678 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
679 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
680 struct block_device *logdev = NULL, *rtdev = NULL;
681 int error;
682
683 /*
684 * Open real time and log devices - order is important.
685 */
686 if (mp->m_logname) {
687 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
688 if (error)
689 goto out;
690 dax_logdev = fs_dax_get_by_bdev(logdev);
691 }
692
693 if (mp->m_rtname) {
694 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
695 if (error)
696 goto out_close_logdev;
697
698 if (rtdev == ddev || rtdev == logdev) {
699 xfs_warn(mp,
700 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
701 error = -EINVAL;
702 goto out_close_rtdev;
703 }
704 dax_rtdev = fs_dax_get_by_bdev(rtdev);
705 }
706
707 /*
708 * Setup xfs_mount buffer target pointers
709 */
710 error = -ENOMEM;
711 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
712 if (!mp->m_ddev_targp)
713 goto out_close_rtdev;
714
715 if (rtdev) {
716 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
717 if (!mp->m_rtdev_targp)
718 goto out_free_ddev_targ;
719 }
720
721 if (logdev && logdev != ddev) {
722 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
723 if (!mp->m_logdev_targp)
724 goto out_free_rtdev_targ;
725 } else {
726 mp->m_logdev_targp = mp->m_ddev_targp;
727 }
728
729 return 0;
730
731 out_free_rtdev_targ:
732 if (mp->m_rtdev_targp)
733 xfs_free_buftarg(mp->m_rtdev_targp);
734 out_free_ddev_targ:
735 xfs_free_buftarg(mp->m_ddev_targp);
736 out_close_rtdev:
737 xfs_blkdev_put(rtdev);
738 fs_put_dax(dax_rtdev);
739 out_close_logdev:
740 if (logdev && logdev != ddev) {
741 xfs_blkdev_put(logdev);
742 fs_put_dax(dax_logdev);
743 }
744 out:
745 fs_put_dax(dax_ddev);
746 return error;
747 }
748
749 /*
750 * Setup xfs_mount buffer target pointers based on superblock
751 */
752 STATIC int
xfs_setup_devices(struct xfs_mount * mp)753 xfs_setup_devices(
754 struct xfs_mount *mp)
755 {
756 int error;
757
758 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
759 if (error)
760 return error;
761
762 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
763 unsigned int log_sector_size = BBSIZE;
764
765 if (xfs_sb_version_hassector(&mp->m_sb))
766 log_sector_size = mp->m_sb.sb_logsectsize;
767 error = xfs_setsize_buftarg(mp->m_logdev_targp,
768 log_sector_size);
769 if (error)
770 return error;
771 }
772 if (mp->m_rtdev_targp) {
773 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
774 mp->m_sb.sb_sectsize);
775 if (error)
776 return error;
777 }
778
779 return 0;
780 }
781
782 STATIC int
xfs_init_mount_workqueues(struct xfs_mount * mp)783 xfs_init_mount_workqueues(
784 struct xfs_mount *mp)
785 {
786 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
787 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
788 if (!mp->m_buf_workqueue)
789 goto out;
790
791 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
792 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
793 if (!mp->m_unwritten_workqueue)
794 goto out_destroy_buf;
795
796 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
797 WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
798 0, mp->m_fsname);
799 if (!mp->m_cil_workqueue)
800 goto out_destroy_unwritten;
801
802 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
803 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
804 if (!mp->m_reclaim_workqueue)
805 goto out_destroy_cil;
806
807 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
808 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
809 if (!mp->m_eofblocks_workqueue)
810 goto out_destroy_reclaim;
811
812 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
813 mp->m_fsname);
814 if (!mp->m_sync_workqueue)
815 goto out_destroy_eofb;
816
817 return 0;
818
819 out_destroy_eofb:
820 destroy_workqueue(mp->m_eofblocks_workqueue);
821 out_destroy_reclaim:
822 destroy_workqueue(mp->m_reclaim_workqueue);
823 out_destroy_cil:
824 destroy_workqueue(mp->m_cil_workqueue);
825 out_destroy_unwritten:
826 destroy_workqueue(mp->m_unwritten_workqueue);
827 out_destroy_buf:
828 destroy_workqueue(mp->m_buf_workqueue);
829 out:
830 return -ENOMEM;
831 }
832
833 STATIC void
xfs_destroy_mount_workqueues(struct xfs_mount * mp)834 xfs_destroy_mount_workqueues(
835 struct xfs_mount *mp)
836 {
837 destroy_workqueue(mp->m_sync_workqueue);
838 destroy_workqueue(mp->m_eofblocks_workqueue);
839 destroy_workqueue(mp->m_reclaim_workqueue);
840 destroy_workqueue(mp->m_cil_workqueue);
841 destroy_workqueue(mp->m_unwritten_workqueue);
842 destroy_workqueue(mp->m_buf_workqueue);
843 }
844
845 static void
xfs_flush_inodes_worker(struct work_struct * work)846 xfs_flush_inodes_worker(
847 struct work_struct *work)
848 {
849 struct xfs_mount *mp = container_of(work, struct xfs_mount,
850 m_flush_inodes_work);
851 struct super_block *sb = mp->m_super;
852
853 if (down_read_trylock(&sb->s_umount)) {
854 sync_inodes_sb(sb);
855 up_read(&sb->s_umount);
856 }
857 }
858
859 /*
860 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
861 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
862 * for IO to complete so that we effectively throttle multiple callers to the
863 * rate at which IO is completing.
864 */
865 void
xfs_flush_inodes(struct xfs_mount * mp)866 xfs_flush_inodes(
867 struct xfs_mount *mp)
868 {
869 /*
870 * If flush_work() returns true then that means we waited for a flush
871 * which was already in progress. Don't bother running another scan.
872 */
873 if (flush_work(&mp->m_flush_inodes_work))
874 return;
875
876 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
877 flush_work(&mp->m_flush_inodes_work);
878 }
879
880 /* Catch misguided souls that try to use this interface on XFS */
881 STATIC struct inode *
xfs_fs_alloc_inode(struct super_block * sb)882 xfs_fs_alloc_inode(
883 struct super_block *sb)
884 {
885 BUG();
886 return NULL;
887 }
888
889 #ifdef DEBUG
890 static void
xfs_check_delalloc(struct xfs_inode * ip,int whichfork)891 xfs_check_delalloc(
892 struct xfs_inode *ip,
893 int whichfork)
894 {
895 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
896 struct xfs_bmbt_irec got;
897 struct xfs_iext_cursor icur;
898
899 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
900 return;
901 do {
902 if (isnullstartblock(got.br_startblock)) {
903 xfs_warn(ip->i_mount,
904 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
905 ip->i_ino,
906 whichfork == XFS_DATA_FORK ? "data" : "cow",
907 got.br_startoff, got.br_blockcount);
908 }
909 } while (xfs_iext_next_extent(ifp, &icur, &got));
910 }
911 #else
912 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
913 #endif
914
915 /*
916 * Now that the generic code is guaranteed not to be accessing
917 * the linux inode, we can inactivate and reclaim the inode.
918 */
919 STATIC void
xfs_fs_destroy_inode(struct inode * inode)920 xfs_fs_destroy_inode(
921 struct inode *inode)
922 {
923 struct xfs_inode *ip = XFS_I(inode);
924
925 trace_xfs_destroy_inode(ip);
926
927 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
928 XFS_STATS_INC(ip->i_mount, vn_rele);
929 XFS_STATS_INC(ip->i_mount, vn_remove);
930
931 xfs_inactive(ip);
932
933 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
934 xfs_check_delalloc(ip, XFS_DATA_FORK);
935 xfs_check_delalloc(ip, XFS_COW_FORK);
936 ASSERT(0);
937 }
938
939 XFS_STATS_INC(ip->i_mount, vn_reclaim);
940
941 /*
942 * We should never get here with one of the reclaim flags already set.
943 */
944 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
945 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
946
947 /*
948 * We always use background reclaim here because even if the
949 * inode is clean, it still may be under IO and hence we have
950 * to take the flush lock. The background reclaim path handles
951 * this more efficiently than we can here, so simply let background
952 * reclaim tear down all inodes.
953 */
954 xfs_inode_set_reclaim_tag(ip);
955 }
956
957 static void
xfs_fs_dirty_inode(struct inode * inode,int flag)958 xfs_fs_dirty_inode(
959 struct inode *inode,
960 int flag)
961 {
962 struct xfs_inode *ip = XFS_I(inode);
963 struct xfs_mount *mp = ip->i_mount;
964 struct xfs_trans *tp;
965
966 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
967 return;
968 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
969 return;
970
971 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
972 return;
973 xfs_ilock(ip, XFS_ILOCK_EXCL);
974 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
975 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
976 xfs_trans_commit(tp);
977 }
978
979 /*
980 * Slab object creation initialisation for the XFS inode.
981 * This covers only the idempotent fields in the XFS inode;
982 * all other fields need to be initialised on allocation
983 * from the slab. This avoids the need to repeatedly initialise
984 * fields in the xfs inode that left in the initialise state
985 * when freeing the inode.
986 */
987 STATIC void
xfs_fs_inode_init_once(void * inode)988 xfs_fs_inode_init_once(
989 void *inode)
990 {
991 struct xfs_inode *ip = inode;
992
993 memset(ip, 0, sizeof(struct xfs_inode));
994
995 /* vfs inode */
996 inode_init_once(VFS_I(ip));
997
998 /* xfs inode */
999 atomic_set(&ip->i_pincount, 0);
1000 spin_lock_init(&ip->i_flags_lock);
1001
1002 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1003 "xfsino", ip->i_ino);
1004 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1005 "xfsino", ip->i_ino);
1006 }
1007
1008 /*
1009 * We do an unlocked check for XFS_IDONTCACHE here because we are already
1010 * serialised against cache hits here via the inode->i_lock and igrab() in
1011 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1012 * racing with us, and it avoids needing to grab a spinlock here for every inode
1013 * we drop the final reference on.
1014 */
1015 STATIC int
xfs_fs_drop_inode(struct inode * inode)1016 xfs_fs_drop_inode(
1017 struct inode *inode)
1018 {
1019 struct xfs_inode *ip = XFS_I(inode);
1020
1021 /*
1022 * If this unlinked inode is in the middle of recovery, don't
1023 * drop the inode just yet; log recovery will take care of
1024 * that. See the comment for this inode flag.
1025 */
1026 if (ip->i_flags & XFS_IRECOVERY) {
1027 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1028 return 0;
1029 }
1030
1031 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1032 }
1033
1034 STATIC void
xfs_free_fsname(struct xfs_mount * mp)1035 xfs_free_fsname(
1036 struct xfs_mount *mp)
1037 {
1038 kfree(mp->m_fsname);
1039 kfree(mp->m_rtname);
1040 kfree(mp->m_logname);
1041 }
1042
1043 STATIC int
xfs_fs_sync_fs(struct super_block * sb,int wait)1044 xfs_fs_sync_fs(
1045 struct super_block *sb,
1046 int wait)
1047 {
1048 struct xfs_mount *mp = XFS_M(sb);
1049
1050 /*
1051 * Doing anything during the async pass would be counterproductive.
1052 */
1053 if (!wait)
1054 return 0;
1055
1056 xfs_log_force(mp, XFS_LOG_SYNC);
1057 if (laptop_mode) {
1058 /*
1059 * The disk must be active because we're syncing.
1060 * We schedule log work now (now that the disk is
1061 * active) instead of later (when it might not be).
1062 */
1063 flush_delayed_work(&mp->m_log->l_work);
1064 }
1065
1066 return 0;
1067 }
1068
1069 STATIC int
xfs_fs_statfs(struct dentry * dentry,struct kstatfs * statp)1070 xfs_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(d_inode(dentry));
1077 uint64_t fakeinos, id;
1078 uint64_t icount;
1079 uint64_t ifree;
1080 uint64_t fdblocks;
1081 xfs_extlen_t lsize;
1082 int64_t ffree;
1083
1084 statp->f_type = XFS_SUPER_MAGIC;
1085 statp->f_namelen = MAXNAMELEN - 1;
1086
1087 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1088 statp->f_fsid.val[0] = (u32)id;
1089 statp->f_fsid.val[1] = (u32)(id >> 32);
1090
1091 icount = percpu_counter_sum(&mp->m_icount);
1092 ifree = percpu_counter_sum(&mp->m_ifree);
1093 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1094
1095 spin_lock(&mp->m_sb_lock);
1096 statp->f_bsize = sbp->sb_blocksize;
1097 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1098 statp->f_blocks = sbp->sb_dblocks - lsize;
1099 spin_unlock(&mp->m_sb_lock);
1100
1101 statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1102 statp->f_bavail = statp->f_bfree;
1103
1104 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
1105 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1106 if (M_IGEO(mp)->maxicount)
1107 statp->f_files = min_t(typeof(statp->f_files),
1108 statp->f_files,
1109 M_IGEO(mp)->maxicount);
1110
1111 /* If sb_icount overshot maxicount, report actual allocation */
1112 statp->f_files = max_t(typeof(statp->f_files),
1113 statp->f_files,
1114 sbp->sb_icount);
1115
1116 /* make sure statp->f_ffree does not underflow */
1117 ffree = statp->f_files - (icount - ifree);
1118 statp->f_ffree = max_t(int64_t, ffree, 0);
1119
1120
1121 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1122 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1123 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1124 xfs_qm_statvfs(ip, statp);
1125
1126 if (XFS_IS_REALTIME_MOUNT(mp) &&
1127 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1128 statp->f_blocks = sbp->sb_rblocks;
1129 statp->f_bavail = statp->f_bfree =
1130 sbp->sb_frextents * sbp->sb_rextsize;
1131 }
1132
1133 return 0;
1134 }
1135
1136 STATIC void
xfs_save_resvblks(struct xfs_mount * mp)1137 xfs_save_resvblks(struct xfs_mount *mp)
1138 {
1139 uint64_t resblks = 0;
1140
1141 mp->m_resblks_save = mp->m_resblks;
1142 xfs_reserve_blocks(mp, &resblks, NULL);
1143 }
1144
1145 STATIC void
xfs_restore_resvblks(struct xfs_mount * mp)1146 xfs_restore_resvblks(struct xfs_mount *mp)
1147 {
1148 uint64_t resblks;
1149
1150 if (mp->m_resblks_save) {
1151 resblks = mp->m_resblks_save;
1152 mp->m_resblks_save = 0;
1153 } else
1154 resblks = xfs_default_resblks(mp);
1155
1156 xfs_reserve_blocks(mp, &resblks, NULL);
1157 }
1158
1159 /*
1160 * Trigger writeback of all the dirty metadata in the file system.
1161 *
1162 * This ensures that the metadata is written to their location on disk rather
1163 * than just existing in transactions in the log. This means after a quiesce
1164 * there is no log replay required to write the inodes to disk - this is the
1165 * primary difference between a sync and a quiesce.
1166 *
1167 * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1168 * it is started again when appropriate.
1169 */
1170 void
xfs_quiesce_attr(struct xfs_mount * mp)1171 xfs_quiesce_attr(
1172 struct xfs_mount *mp)
1173 {
1174 int error = 0;
1175
1176 /* wait for all modifications to complete */
1177 while (atomic_read(&mp->m_active_trans) > 0)
1178 delay(100);
1179
1180 /* force the log to unpin objects from the now complete transactions */
1181 xfs_log_force(mp, XFS_LOG_SYNC);
1182
1183 /* reclaim inodes to do any IO before the freeze completes */
1184 xfs_reclaim_inodes(mp, 0);
1185 xfs_reclaim_inodes(mp, SYNC_WAIT);
1186
1187 /* Push the superblock and write an unmount record */
1188 error = xfs_log_sbcount(mp);
1189 if (error)
1190 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1191 "Frozen image may not be consistent.");
1192 /*
1193 * Just warn here till VFS can correctly support
1194 * read-only remount without racing.
1195 */
1196 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1197
1198 xfs_log_quiesce(mp);
1199 }
1200
1201 STATIC int
xfs_test_remount_options(struct super_block * sb,char * options)1202 xfs_test_remount_options(
1203 struct super_block *sb,
1204 char *options)
1205 {
1206 int error = 0;
1207 struct xfs_mount *tmp_mp;
1208
1209 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1210 if (!tmp_mp)
1211 return -ENOMEM;
1212
1213 tmp_mp->m_super = sb;
1214 error = xfs_parseargs(tmp_mp, options);
1215 xfs_free_fsname(tmp_mp);
1216 kmem_free(tmp_mp);
1217
1218 return error;
1219 }
1220
1221 STATIC int
xfs_fs_remount(struct super_block * sb,int * flags,char * options)1222 xfs_fs_remount(
1223 struct super_block *sb,
1224 int *flags,
1225 char *options)
1226 {
1227 struct xfs_mount *mp = XFS_M(sb);
1228 xfs_sb_t *sbp = &mp->m_sb;
1229 substring_t args[MAX_OPT_ARGS];
1230 char *p;
1231 int error;
1232
1233 /* version 5 superblocks always support version counters. */
1234 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1235 *flags |= SB_I_VERSION;
1236
1237 /* First, check for complete junk; i.e. invalid options */
1238 error = xfs_test_remount_options(sb, options);
1239 if (error)
1240 return error;
1241
1242 sync_filesystem(sb);
1243 while ((p = strsep(&options, ",")) != NULL) {
1244 int token;
1245
1246 if (!*p)
1247 continue;
1248
1249 token = match_token(p, tokens, args);
1250 switch (token) {
1251 case Opt_inode64:
1252 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1253 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1254 break;
1255 case Opt_inode32:
1256 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1257 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1258 break;
1259 default:
1260 /*
1261 * Logically we would return an error here to prevent
1262 * users from believing they might have changed
1263 * mount options using remount which can't be changed.
1264 *
1265 * But unfortunately mount(8) adds all options from
1266 * mtab and fstab to the mount arguments in some cases
1267 * so we can't blindly reject options, but have to
1268 * check for each specified option if it actually
1269 * differs from the currently set option and only
1270 * reject it if that's the case.
1271 *
1272 * Until that is implemented we return success for
1273 * every remount request, and silently ignore all
1274 * options that we can't actually change.
1275 */
1276 #if 0
1277 xfs_info(mp,
1278 "mount option \"%s\" not supported for remount", p);
1279 return -EINVAL;
1280 #else
1281 break;
1282 #endif
1283 }
1284 }
1285
1286 /* ro -> rw */
1287 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1288 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1289 xfs_warn(mp,
1290 "ro->rw transition prohibited on norecovery mount");
1291 return -EINVAL;
1292 }
1293
1294 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1295 xfs_sb_has_ro_compat_feature(sbp,
1296 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1297 xfs_warn(mp,
1298 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1299 (sbp->sb_features_ro_compat &
1300 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1301 return -EINVAL;
1302 }
1303
1304 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1305
1306 /*
1307 * If this is the first remount to writeable state we
1308 * might have some superblock changes to update.
1309 */
1310 if (mp->m_update_sb) {
1311 error = xfs_sync_sb(mp, false);
1312 if (error) {
1313 xfs_warn(mp, "failed to write sb changes");
1314 return error;
1315 }
1316 mp->m_update_sb = false;
1317 }
1318
1319 /*
1320 * Fill out the reserve pool if it is empty. Use the stashed
1321 * value if it is non-zero, otherwise go with the default.
1322 */
1323 xfs_restore_resvblks(mp);
1324 xfs_log_work_queue(mp);
1325
1326 /* Recover any CoW blocks that never got remapped. */
1327 error = xfs_reflink_recover_cow(mp);
1328 if (error) {
1329 xfs_err(mp,
1330 "Error %d recovering leftover CoW allocations.", error);
1331 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1332 return error;
1333 }
1334 xfs_start_block_reaping(mp);
1335
1336 /* Create the per-AG metadata reservation pool .*/
1337 error = xfs_fs_reserve_ag_blocks(mp);
1338 if (error && error != -ENOSPC)
1339 return error;
1340 }
1341
1342 /* rw -> ro */
1343 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1344 /*
1345 * Cancel background eofb scanning so it cannot race with the
1346 * final log force+buftarg wait and deadlock the remount.
1347 */
1348 xfs_stop_block_reaping(mp);
1349
1350 /* Get rid of any leftover CoW reservations... */
1351 error = xfs_icache_free_cowblocks(mp, NULL);
1352 if (error) {
1353 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1354 return error;
1355 }
1356
1357 /* Free the per-AG metadata reservation pool. */
1358 error = xfs_fs_unreserve_ag_blocks(mp);
1359 if (error) {
1360 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1361 return error;
1362 }
1363
1364 /*
1365 * Before we sync the metadata, we need to free up the reserve
1366 * block pool so that the used block count in the superblock on
1367 * disk is correct at the end of the remount. Stash the current
1368 * reserve pool size so that if we get remounted rw, we can
1369 * return it to the same size.
1370 */
1371 xfs_save_resvblks(mp);
1372
1373 xfs_quiesce_attr(mp);
1374 mp->m_flags |= XFS_MOUNT_RDONLY;
1375 }
1376
1377 return 0;
1378 }
1379
1380 /*
1381 * Second stage of a freeze. The data is already frozen so we only
1382 * need to take care of the metadata. Once that's done sync the superblock
1383 * to the log to dirty it in case of a crash while frozen. This ensures that we
1384 * will recover the unlinked inode lists on the next mount.
1385 */
1386 STATIC int
xfs_fs_freeze(struct super_block * sb)1387 xfs_fs_freeze(
1388 struct super_block *sb)
1389 {
1390 struct xfs_mount *mp = XFS_M(sb);
1391
1392 xfs_stop_block_reaping(mp);
1393 xfs_save_resvblks(mp);
1394 xfs_quiesce_attr(mp);
1395 return xfs_sync_sb(mp, true);
1396 }
1397
1398 STATIC int
xfs_fs_unfreeze(struct super_block * sb)1399 xfs_fs_unfreeze(
1400 struct super_block *sb)
1401 {
1402 struct xfs_mount *mp = XFS_M(sb);
1403
1404 xfs_restore_resvblks(mp);
1405 xfs_log_work_queue(mp);
1406 xfs_start_block_reaping(mp);
1407 return 0;
1408 }
1409
1410 STATIC int
xfs_fs_show_options(struct seq_file * m,struct dentry * root)1411 xfs_fs_show_options(
1412 struct seq_file *m,
1413 struct dentry *root)
1414 {
1415 xfs_showargs(XFS_M(root->d_sb), m);
1416 return 0;
1417 }
1418
1419 /*
1420 * This function fills in xfs_mount_t fields based on mount args.
1421 * Note: the superblock _has_ now been read in.
1422 */
1423 STATIC int
xfs_finish_flags(struct xfs_mount * mp)1424 xfs_finish_flags(
1425 struct xfs_mount *mp)
1426 {
1427 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1428
1429 /* Fail a mount where the logbuf is smaller than the log stripe */
1430 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1431 if (mp->m_logbsize <= 0 &&
1432 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1433 mp->m_logbsize = mp->m_sb.sb_logsunit;
1434 } else if (mp->m_logbsize > 0 &&
1435 mp->m_logbsize < mp->m_sb.sb_logsunit) {
1436 xfs_warn(mp,
1437 "logbuf size must be greater than or equal to log stripe size");
1438 return -EINVAL;
1439 }
1440 } else {
1441 /* Fail a mount if the logbuf is larger than 32K */
1442 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1443 xfs_warn(mp,
1444 "logbuf size for version 1 logs must be 16K or 32K");
1445 return -EINVAL;
1446 }
1447 }
1448
1449 /*
1450 * V5 filesystems always use attr2 format for attributes.
1451 */
1452 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1453 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1454 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1455 "attr2 is always enabled for V5 filesystems.");
1456 return -EINVAL;
1457 }
1458
1459 /*
1460 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1461 * told by noattr2 to turn it off
1462 */
1463 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1464 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1465 mp->m_flags |= XFS_MOUNT_ATTR2;
1466
1467 /*
1468 * prohibit r/w mounts of read-only filesystems
1469 */
1470 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1471 xfs_warn(mp,
1472 "cannot mount a read-only filesystem as read-write");
1473 return -EROFS;
1474 }
1475
1476 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1477 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1478 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1479 xfs_warn(mp,
1480 "Super block does not support project and group quota together");
1481 return -EINVAL;
1482 }
1483
1484 return 0;
1485 }
1486
1487 static int
xfs_init_percpu_counters(struct xfs_mount * mp)1488 xfs_init_percpu_counters(
1489 struct xfs_mount *mp)
1490 {
1491 int error;
1492
1493 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1494 if (error)
1495 return -ENOMEM;
1496
1497 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1498 if (error)
1499 goto free_icount;
1500
1501 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1502 if (error)
1503 goto free_ifree;
1504
1505 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1506 if (error)
1507 goto free_fdblocks;
1508
1509 return 0;
1510
1511 free_fdblocks:
1512 percpu_counter_destroy(&mp->m_fdblocks);
1513 free_ifree:
1514 percpu_counter_destroy(&mp->m_ifree);
1515 free_icount:
1516 percpu_counter_destroy(&mp->m_icount);
1517 return -ENOMEM;
1518 }
1519
1520 void
xfs_reinit_percpu_counters(struct xfs_mount * mp)1521 xfs_reinit_percpu_counters(
1522 struct xfs_mount *mp)
1523 {
1524 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1525 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1526 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1527 }
1528
1529 static void
xfs_destroy_percpu_counters(struct xfs_mount * mp)1530 xfs_destroy_percpu_counters(
1531 struct xfs_mount *mp)
1532 {
1533 percpu_counter_destroy(&mp->m_icount);
1534 percpu_counter_destroy(&mp->m_ifree);
1535 percpu_counter_destroy(&mp->m_fdblocks);
1536 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1537 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1538 percpu_counter_destroy(&mp->m_delalloc_blks);
1539 }
1540
1541 static struct xfs_mount *
xfs_mount_alloc(struct super_block * sb)1542 xfs_mount_alloc(
1543 struct super_block *sb)
1544 {
1545 struct xfs_mount *mp;
1546
1547 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1548 if (!mp)
1549 return NULL;
1550
1551 mp->m_super = sb;
1552 spin_lock_init(&mp->m_sb_lock);
1553 spin_lock_init(&mp->m_agirotor_lock);
1554 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1555 spin_lock_init(&mp->m_perag_lock);
1556 mutex_init(&mp->m_growlock);
1557 atomic_set(&mp->m_active_trans, 0);
1558 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1559 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1560 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1561 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1562 mp->m_kobj.kobject.kset = xfs_kset;
1563 /*
1564 * We don't create the finobt per-ag space reservation until after log
1565 * recovery, so we must set this to true so that an ifree transaction
1566 * started during log recovery will not depend on space reservations
1567 * for finobt expansion.
1568 */
1569 mp->m_finobt_nores = true;
1570 return mp;
1571 }
1572
1573
1574 STATIC int
xfs_fs_fill_super(struct super_block * sb,void * data,int silent)1575 xfs_fs_fill_super(
1576 struct super_block *sb,
1577 void *data,
1578 int silent)
1579 {
1580 struct inode *root;
1581 struct xfs_mount *mp = NULL;
1582 int flags = 0, error = -ENOMEM;
1583
1584 /*
1585 * allocate mp and do all low-level struct initializations before we
1586 * attach it to the super
1587 */
1588 mp = xfs_mount_alloc(sb);
1589 if (!mp)
1590 goto out;
1591 sb->s_fs_info = mp;
1592
1593 error = xfs_parseargs(mp, (char *)data);
1594 if (error)
1595 goto out_free_fsname;
1596
1597 sb_min_blocksize(sb, BBSIZE);
1598 sb->s_xattr = xfs_xattr_handlers;
1599 sb->s_export_op = &xfs_export_operations;
1600 #ifdef CONFIG_XFS_QUOTA
1601 sb->s_qcop = &xfs_quotactl_operations;
1602 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1603 #endif
1604 sb->s_op = &xfs_super_operations;
1605
1606 /*
1607 * Delay mount work if the debug hook is set. This is debug
1608 * instrumention to coordinate simulation of xfs mount failures with
1609 * VFS superblock operations
1610 */
1611 if (xfs_globals.mount_delay) {
1612 xfs_notice(mp, "Delaying mount for %d seconds.",
1613 xfs_globals.mount_delay);
1614 msleep(xfs_globals.mount_delay * 1000);
1615 }
1616
1617 if (silent)
1618 flags |= XFS_MFSI_QUIET;
1619
1620 error = xfs_open_devices(mp);
1621 if (error)
1622 goto out_free_fsname;
1623
1624 error = xfs_init_mount_workqueues(mp);
1625 if (error)
1626 goto out_close_devices;
1627
1628 error = xfs_init_percpu_counters(mp);
1629 if (error)
1630 goto out_destroy_workqueues;
1631
1632 /* Allocate stats memory before we do operations that might use it */
1633 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1634 if (!mp->m_stats.xs_stats) {
1635 error = -ENOMEM;
1636 goto out_destroy_counters;
1637 }
1638
1639 error = xfs_readsb(mp, flags);
1640 if (error)
1641 goto out_free_stats;
1642
1643 error = xfs_finish_flags(mp);
1644 if (error)
1645 goto out_free_sb;
1646
1647 error = xfs_setup_devices(mp);
1648 if (error)
1649 goto out_free_sb;
1650
1651 /*
1652 * XFS block mappings use 54 bits to store the logical block offset.
1653 * This should suffice to handle the maximum file size that the VFS
1654 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1655 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1656 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1657 * to check this assertion.
1658 *
1659 * Avoid integer overflow by comparing the maximum bmbt offset to the
1660 * maximum pagecache offset in units of fs blocks.
1661 */
1662 if (XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE) > XFS_MAX_FILEOFF) {
1663 xfs_warn(mp,
1664 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1665 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1666 XFS_MAX_FILEOFF);
1667 error = -EINVAL;
1668 goto out_free_sb;
1669 }
1670
1671 error = xfs_filestream_mount(mp);
1672 if (error)
1673 goto out_free_sb;
1674
1675 /*
1676 * we must configure the block size in the superblock before we run the
1677 * full mount process as the mount process can lookup and cache inodes.
1678 */
1679 sb->s_magic = XFS_SUPER_MAGIC;
1680 sb->s_blocksize = mp->m_sb.sb_blocksize;
1681 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1682 sb->s_maxbytes = MAX_LFS_FILESIZE;
1683 sb->s_max_links = XFS_MAXLINK;
1684 sb->s_time_gran = 1;
1685 sb->s_time_min = S32_MIN;
1686 sb->s_time_max = S32_MAX;
1687 sb->s_iflags |= SB_I_CGROUPWB;
1688
1689 set_posix_acl_flag(sb);
1690
1691 /* version 5 superblocks support inode version counters. */
1692 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1693 sb->s_flags |= SB_I_VERSION;
1694
1695 if (mp->m_flags & XFS_MOUNT_DAX) {
1696 bool rtdev_is_dax = false, datadev_is_dax;
1697
1698 xfs_warn(mp,
1699 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1700
1701 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1702 sb->s_blocksize);
1703 if (mp->m_rtdev_targp)
1704 rtdev_is_dax = bdev_dax_supported(
1705 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1706 if (!rtdev_is_dax && !datadev_is_dax) {
1707 xfs_alert(mp,
1708 "DAX unsupported by block device. Turning off DAX.");
1709 mp->m_flags &= ~XFS_MOUNT_DAX;
1710 }
1711 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1712 xfs_alert(mp,
1713 "DAX and reflink cannot be used together!");
1714 error = -EINVAL;
1715 goto out_filestream_unmount;
1716 }
1717 }
1718
1719 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1720 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1721
1722 if (!blk_queue_discard(q)) {
1723 xfs_warn(mp, "mounting with \"discard\" option, but "
1724 "the device does not support discard");
1725 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1726 }
1727 }
1728
1729 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1730 if (mp->m_sb.sb_rblocks) {
1731 xfs_alert(mp,
1732 "reflink not compatible with realtime device!");
1733 error = -EINVAL;
1734 goto out_filestream_unmount;
1735 }
1736
1737 if (xfs_globals.always_cow) {
1738 xfs_info(mp, "using DEBUG-only always_cow mode.");
1739 mp->m_always_cow = true;
1740 }
1741 }
1742
1743 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1744 xfs_alert(mp,
1745 "reverse mapping btree not compatible with realtime device!");
1746 error = -EINVAL;
1747 goto out_filestream_unmount;
1748 }
1749
1750 error = xfs_mountfs(mp);
1751 if (error)
1752 goto out_filestream_unmount;
1753
1754 root = igrab(VFS_I(mp->m_rootip));
1755 if (!root) {
1756 error = -ENOENT;
1757 goto out_unmount;
1758 }
1759 sb->s_root = d_make_root(root);
1760 if (!sb->s_root) {
1761 error = -ENOMEM;
1762 goto out_unmount;
1763 }
1764
1765 return 0;
1766
1767 out_filestream_unmount:
1768 xfs_filestream_unmount(mp);
1769 out_free_sb:
1770 xfs_freesb(mp);
1771 out_free_stats:
1772 free_percpu(mp->m_stats.xs_stats);
1773 out_destroy_counters:
1774 xfs_destroy_percpu_counters(mp);
1775 out_destroy_workqueues:
1776 xfs_destroy_mount_workqueues(mp);
1777 out_close_devices:
1778 xfs_close_devices(mp);
1779 out_free_fsname:
1780 sb->s_fs_info = NULL;
1781 xfs_free_fsname(mp);
1782 kfree(mp);
1783 out:
1784 return error;
1785
1786 out_unmount:
1787 xfs_filestream_unmount(mp);
1788 xfs_unmountfs(mp);
1789 goto out_free_sb;
1790 }
1791
1792 STATIC void
xfs_fs_put_super(struct super_block * sb)1793 xfs_fs_put_super(
1794 struct super_block *sb)
1795 {
1796 struct xfs_mount *mp = XFS_M(sb);
1797
1798 /* if ->fill_super failed, we have no mount to tear down */
1799 if (!sb->s_fs_info)
1800 return;
1801
1802 xfs_notice(mp, "Unmounting Filesystem");
1803 xfs_filestream_unmount(mp);
1804 xfs_unmountfs(mp);
1805
1806 xfs_freesb(mp);
1807 free_percpu(mp->m_stats.xs_stats);
1808 xfs_destroy_percpu_counters(mp);
1809 xfs_destroy_mount_workqueues(mp);
1810 xfs_close_devices(mp);
1811
1812 sb->s_fs_info = NULL;
1813 xfs_free_fsname(mp);
1814 kfree(mp);
1815 }
1816
1817 STATIC struct dentry *
xfs_fs_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1818 xfs_fs_mount(
1819 struct file_system_type *fs_type,
1820 int flags,
1821 const char *dev_name,
1822 void *data)
1823 {
1824 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1825 }
1826
1827 static long
xfs_fs_nr_cached_objects(struct super_block * sb,struct shrink_control * sc)1828 xfs_fs_nr_cached_objects(
1829 struct super_block *sb,
1830 struct shrink_control *sc)
1831 {
1832 /* Paranoia: catch incorrect calls during mount setup or teardown */
1833 if (WARN_ON_ONCE(!sb->s_fs_info))
1834 return 0;
1835 return xfs_reclaim_inodes_count(XFS_M(sb));
1836 }
1837
1838 static long
xfs_fs_free_cached_objects(struct super_block * sb,struct shrink_control * sc)1839 xfs_fs_free_cached_objects(
1840 struct super_block *sb,
1841 struct shrink_control *sc)
1842 {
1843 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1844 }
1845
1846 static const struct super_operations xfs_super_operations = {
1847 .alloc_inode = xfs_fs_alloc_inode,
1848 .destroy_inode = xfs_fs_destroy_inode,
1849 .dirty_inode = xfs_fs_dirty_inode,
1850 .drop_inode = xfs_fs_drop_inode,
1851 .put_super = xfs_fs_put_super,
1852 .sync_fs = xfs_fs_sync_fs,
1853 .freeze_fs = xfs_fs_freeze,
1854 .unfreeze_fs = xfs_fs_unfreeze,
1855 .statfs = xfs_fs_statfs,
1856 .remount_fs = xfs_fs_remount,
1857 .show_options = xfs_fs_show_options,
1858 .nr_cached_objects = xfs_fs_nr_cached_objects,
1859 .free_cached_objects = xfs_fs_free_cached_objects,
1860 };
1861
1862 static struct file_system_type xfs_fs_type = {
1863 .owner = THIS_MODULE,
1864 .name = "xfs",
1865 .mount = xfs_fs_mount,
1866 .kill_sb = kill_block_super,
1867 .fs_flags = FS_REQUIRES_DEV,
1868 };
1869 MODULE_ALIAS_FS("xfs");
1870
1871 STATIC int __init
xfs_init_zones(void)1872 xfs_init_zones(void)
1873 {
1874 if (bioset_init(&xfs_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
1875 offsetof(struct xfs_ioend, io_inline_bio),
1876 BIOSET_NEED_BVECS))
1877 goto out;
1878
1879 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1880 "xfs_log_ticket");
1881 if (!xfs_log_ticket_zone)
1882 goto out_free_ioend_bioset;
1883
1884 xfs_bmap_free_item_zone = kmem_zone_init(
1885 sizeof(struct xfs_extent_free_item),
1886 "xfs_bmap_free_item");
1887 if (!xfs_bmap_free_item_zone)
1888 goto out_destroy_log_ticket_zone;
1889
1890 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1891 "xfs_btree_cur");
1892 if (!xfs_btree_cur_zone)
1893 goto out_destroy_bmap_free_item_zone;
1894
1895 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1896 "xfs_da_state");
1897 if (!xfs_da_state_zone)
1898 goto out_destroy_btree_cur_zone;
1899
1900 xfs_ifork_zone = kmem_zone_init(sizeof(struct xfs_ifork), "xfs_ifork");
1901 if (!xfs_ifork_zone)
1902 goto out_destroy_da_state_zone;
1903
1904 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1905 if (!xfs_trans_zone)
1906 goto out_destroy_ifork_zone;
1907
1908
1909 /*
1910 * The size of the zone allocated buf log item is the maximum
1911 * size possible under XFS. This wastes a little bit of memory,
1912 * but it is much faster.
1913 */
1914 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1915 "xfs_buf_item");
1916 if (!xfs_buf_item_zone)
1917 goto out_destroy_trans_zone;
1918
1919 xfs_efd_zone = kmem_zone_init((sizeof(struct xfs_efd_log_item) +
1920 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1921 sizeof(xfs_extent_t))), "xfs_efd_item");
1922 if (!xfs_efd_zone)
1923 goto out_destroy_buf_item_zone;
1924
1925 xfs_efi_zone = kmem_zone_init((sizeof(struct xfs_efi_log_item) +
1926 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1927 sizeof(xfs_extent_t))), "xfs_efi_item");
1928 if (!xfs_efi_zone)
1929 goto out_destroy_efd_zone;
1930
1931 xfs_inode_zone =
1932 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1933 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1934 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1935 if (!xfs_inode_zone)
1936 goto out_destroy_efi_zone;
1937
1938 xfs_ili_zone =
1939 kmem_zone_init_flags(sizeof(struct xfs_inode_log_item),
1940 "xfs_ili", KM_ZONE_SPREAD, NULL);
1941 if (!xfs_ili_zone)
1942 goto out_destroy_inode_zone;
1943 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1944 "xfs_icr");
1945 if (!xfs_icreate_zone)
1946 goto out_destroy_ili_zone;
1947
1948 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1949 "xfs_rud_item");
1950 if (!xfs_rud_zone)
1951 goto out_destroy_icreate_zone;
1952
1953 xfs_rui_zone = kmem_zone_init(
1954 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1955 "xfs_rui_item");
1956 if (!xfs_rui_zone)
1957 goto out_destroy_rud_zone;
1958
1959 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1960 "xfs_cud_item");
1961 if (!xfs_cud_zone)
1962 goto out_destroy_rui_zone;
1963
1964 xfs_cui_zone = kmem_zone_init(
1965 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1966 "xfs_cui_item");
1967 if (!xfs_cui_zone)
1968 goto out_destroy_cud_zone;
1969
1970 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1971 "xfs_bud_item");
1972 if (!xfs_bud_zone)
1973 goto out_destroy_cui_zone;
1974
1975 xfs_bui_zone = kmem_zone_init(
1976 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1977 "xfs_bui_item");
1978 if (!xfs_bui_zone)
1979 goto out_destroy_bud_zone;
1980
1981 return 0;
1982
1983 out_destroy_bud_zone:
1984 kmem_zone_destroy(xfs_bud_zone);
1985 out_destroy_cui_zone:
1986 kmem_zone_destroy(xfs_cui_zone);
1987 out_destroy_cud_zone:
1988 kmem_zone_destroy(xfs_cud_zone);
1989 out_destroy_rui_zone:
1990 kmem_zone_destroy(xfs_rui_zone);
1991 out_destroy_rud_zone:
1992 kmem_zone_destroy(xfs_rud_zone);
1993 out_destroy_icreate_zone:
1994 kmem_zone_destroy(xfs_icreate_zone);
1995 out_destroy_ili_zone:
1996 kmem_zone_destroy(xfs_ili_zone);
1997 out_destroy_inode_zone:
1998 kmem_zone_destroy(xfs_inode_zone);
1999 out_destroy_efi_zone:
2000 kmem_zone_destroy(xfs_efi_zone);
2001 out_destroy_efd_zone:
2002 kmem_zone_destroy(xfs_efd_zone);
2003 out_destroy_buf_item_zone:
2004 kmem_zone_destroy(xfs_buf_item_zone);
2005 out_destroy_trans_zone:
2006 kmem_zone_destroy(xfs_trans_zone);
2007 out_destroy_ifork_zone:
2008 kmem_zone_destroy(xfs_ifork_zone);
2009 out_destroy_da_state_zone:
2010 kmem_zone_destroy(xfs_da_state_zone);
2011 out_destroy_btree_cur_zone:
2012 kmem_zone_destroy(xfs_btree_cur_zone);
2013 out_destroy_bmap_free_item_zone:
2014 kmem_zone_destroy(xfs_bmap_free_item_zone);
2015 out_destroy_log_ticket_zone:
2016 kmem_zone_destroy(xfs_log_ticket_zone);
2017 out_free_ioend_bioset:
2018 bioset_exit(&xfs_ioend_bioset);
2019 out:
2020 return -ENOMEM;
2021 }
2022
2023 STATIC void
xfs_destroy_zones(void)2024 xfs_destroy_zones(void)
2025 {
2026 /*
2027 * Make sure all delayed rcu free are flushed before we
2028 * destroy caches.
2029 */
2030 rcu_barrier();
2031 kmem_zone_destroy(xfs_bui_zone);
2032 kmem_zone_destroy(xfs_bud_zone);
2033 kmem_zone_destroy(xfs_cui_zone);
2034 kmem_zone_destroy(xfs_cud_zone);
2035 kmem_zone_destroy(xfs_rui_zone);
2036 kmem_zone_destroy(xfs_rud_zone);
2037 kmem_zone_destroy(xfs_icreate_zone);
2038 kmem_zone_destroy(xfs_ili_zone);
2039 kmem_zone_destroy(xfs_inode_zone);
2040 kmem_zone_destroy(xfs_efi_zone);
2041 kmem_zone_destroy(xfs_efd_zone);
2042 kmem_zone_destroy(xfs_buf_item_zone);
2043 kmem_zone_destroy(xfs_trans_zone);
2044 kmem_zone_destroy(xfs_ifork_zone);
2045 kmem_zone_destroy(xfs_da_state_zone);
2046 kmem_zone_destroy(xfs_btree_cur_zone);
2047 kmem_zone_destroy(xfs_bmap_free_item_zone);
2048 kmem_zone_destroy(xfs_log_ticket_zone);
2049 bioset_exit(&xfs_ioend_bioset);
2050 }
2051
2052 STATIC int __init
xfs_init_workqueues(void)2053 xfs_init_workqueues(void)
2054 {
2055 /*
2056 * The allocation workqueue can be used in memory reclaim situations
2057 * (writepage path), and parallelism is only limited by the number of
2058 * AGs in all the filesystems mounted. Hence use the default large
2059 * max_active value for this workqueue.
2060 */
2061 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2062 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2063 if (!xfs_alloc_wq)
2064 return -ENOMEM;
2065
2066 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2067 if (!xfs_discard_wq)
2068 goto out_free_alloc_wq;
2069
2070 return 0;
2071 out_free_alloc_wq:
2072 destroy_workqueue(xfs_alloc_wq);
2073 return -ENOMEM;
2074 }
2075
2076 STATIC void
xfs_destroy_workqueues(void)2077 xfs_destroy_workqueues(void)
2078 {
2079 destroy_workqueue(xfs_discard_wq);
2080 destroy_workqueue(xfs_alloc_wq);
2081 }
2082
2083 STATIC int __init
init_xfs_fs(void)2084 init_xfs_fs(void)
2085 {
2086 int error;
2087
2088 xfs_check_ondisk_structs();
2089
2090 printk(KERN_INFO XFS_VERSION_STRING " with "
2091 XFS_BUILD_OPTIONS " enabled\n");
2092
2093 xfs_dir_startup();
2094
2095 error = xfs_init_zones();
2096 if (error)
2097 goto out;
2098
2099 error = xfs_init_workqueues();
2100 if (error)
2101 goto out_destroy_zones;
2102
2103 error = xfs_mru_cache_init();
2104 if (error)
2105 goto out_destroy_wq;
2106
2107 error = xfs_buf_init();
2108 if (error)
2109 goto out_mru_cache_uninit;
2110
2111 error = xfs_init_procfs();
2112 if (error)
2113 goto out_buf_terminate;
2114
2115 error = xfs_sysctl_register();
2116 if (error)
2117 goto out_cleanup_procfs;
2118
2119 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2120 if (!xfs_kset) {
2121 error = -ENOMEM;
2122 goto out_sysctl_unregister;
2123 }
2124
2125 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2126
2127 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2128 if (!xfsstats.xs_stats) {
2129 error = -ENOMEM;
2130 goto out_kset_unregister;
2131 }
2132
2133 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2134 "stats");
2135 if (error)
2136 goto out_free_stats;
2137
2138 #ifdef DEBUG
2139 xfs_dbg_kobj.kobject.kset = xfs_kset;
2140 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2141 if (error)
2142 goto out_remove_stats_kobj;
2143 #endif
2144
2145 error = xfs_qm_init();
2146 if (error)
2147 goto out_remove_dbg_kobj;
2148
2149 error = register_filesystem(&xfs_fs_type);
2150 if (error)
2151 goto out_qm_exit;
2152 return 0;
2153
2154 out_qm_exit:
2155 xfs_qm_exit();
2156 out_remove_dbg_kobj:
2157 #ifdef DEBUG
2158 xfs_sysfs_del(&xfs_dbg_kobj);
2159 out_remove_stats_kobj:
2160 #endif
2161 xfs_sysfs_del(&xfsstats.xs_kobj);
2162 out_free_stats:
2163 free_percpu(xfsstats.xs_stats);
2164 out_kset_unregister:
2165 kset_unregister(xfs_kset);
2166 out_sysctl_unregister:
2167 xfs_sysctl_unregister();
2168 out_cleanup_procfs:
2169 xfs_cleanup_procfs();
2170 out_buf_terminate:
2171 xfs_buf_terminate();
2172 out_mru_cache_uninit:
2173 xfs_mru_cache_uninit();
2174 out_destroy_wq:
2175 xfs_destroy_workqueues();
2176 out_destroy_zones:
2177 xfs_destroy_zones();
2178 out:
2179 return error;
2180 }
2181
2182 STATIC void __exit
exit_xfs_fs(void)2183 exit_xfs_fs(void)
2184 {
2185 xfs_qm_exit();
2186 unregister_filesystem(&xfs_fs_type);
2187 #ifdef DEBUG
2188 xfs_sysfs_del(&xfs_dbg_kobj);
2189 #endif
2190 xfs_sysfs_del(&xfsstats.xs_kobj);
2191 free_percpu(xfsstats.xs_stats);
2192 kset_unregister(xfs_kset);
2193 xfs_sysctl_unregister();
2194 xfs_cleanup_procfs();
2195 xfs_buf_terminate();
2196 xfs_mru_cache_uninit();
2197 xfs_destroy_workqueues();
2198 xfs_destroy_zones();
2199 xfs_uuid_table_free();
2200 }
2201
2202 module_init(init_xfs_fs);
2203 module_exit(exit_xfs_fs);
2204
2205 MODULE_AUTHOR("Silicon Graphics, Inc.");
2206 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2207 MODULE_LICENSE("GPL");
2208