Loading...
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/blkdev.h>
20#include <linux/module.h>
21#include <linux/buffer_head.h>
22#include <linux/fs.h>
23#include <linux/pagemap.h>
24#include <linux/highmem.h>
25#include <linux/time.h>
26#include <linux/init.h>
27#include <linux/seq_file.h>
28#include <linux/string.h>
29#include <linux/backing-dev.h>
30#include <linux/mount.h>
31#include <linux/mpage.h>
32#include <linux/swap.h>
33#include <linux/writeback.h>
34#include <linux/statfs.h>
35#include <linux/compat.h>
36#include <linux/parser.h>
37#include <linux/ctype.h>
38#include <linux/namei.h>
39#include <linux/miscdevice.h>
40#include <linux/magic.h>
41#include <linux/slab.h>
42#include <linux/cleancache.h>
43#include <linux/ratelimit.h>
44#include <linux/btrfs.h>
45#include "delayed-inode.h"
46#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
50#include "print-tree.h"
51#include "hash.h"
52#include "props.h"
53#include "xattr.h"
54#include "volumes.h"
55#include "export.h"
56#include "compression.h"
57#include "rcu-string.h"
58#include "dev-replace.h"
59#include "free-space-cache.h"
60#include "backref.h"
61#include "tests/btrfs-tests.h"
62
63#include "qgroup.h"
64#define CREATE_TRACE_POINTS
65#include <trace/events/btrfs.h>
66
67static const struct super_operations btrfs_super_ops;
68static struct file_system_type btrfs_fs_type;
69
70static int btrfs_remount(struct super_block *sb, int *flags, char *data);
71
72const char *btrfs_decode_error(int errno)
73{
74 char *errstr = "unknown";
75
76 switch (errno) {
77 case -EIO:
78 errstr = "IO failure";
79 break;
80 case -ENOMEM:
81 errstr = "Out of memory";
82 break;
83 case -EROFS:
84 errstr = "Readonly filesystem";
85 break;
86 case -EEXIST:
87 errstr = "Object already exists";
88 break;
89 case -ENOSPC:
90 errstr = "No space left";
91 break;
92 case -ENOENT:
93 errstr = "No such entry";
94 break;
95 }
96
97 return errstr;
98}
99
100/* btrfs handle error by forcing the filesystem readonly */
101static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
102{
103 struct super_block *sb = fs_info->sb;
104
105 if (sb->s_flags & MS_RDONLY)
106 return;
107
108 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
109 sb->s_flags |= MS_RDONLY;
110 btrfs_info(fs_info, "forced readonly");
111 /*
112 * Note that a running device replace operation is not
113 * canceled here although there is no way to update
114 * the progress. It would add the risk of a deadlock,
115 * therefore the canceling is omitted. The only penalty
116 * is that some I/O remains active until the procedure
117 * completes. The next time when the filesystem is
118 * mounted writeable again, the device replace
119 * operation continues.
120 */
121 }
122}
123
124/*
125 * __btrfs_handle_fs_error decodes expected errors from the caller and
126 * invokes the approciate error response.
127 */
128__cold
129void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
130 unsigned int line, int errno, const char *fmt, ...)
131{
132 struct super_block *sb = fs_info->sb;
133#ifdef CONFIG_PRINTK
134 const char *errstr;
135#endif
136
137 /*
138 * Special case: if the error is EROFS, and we're already
139 * under MS_RDONLY, then it is safe here.
140 */
141 if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
142 return;
143
144#ifdef CONFIG_PRINTK
145 errstr = btrfs_decode_error(errno);
146 if (fmt) {
147 struct va_format vaf;
148 va_list args;
149
150 va_start(args, fmt);
151 vaf.fmt = fmt;
152 vaf.va = &args;
153
154 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
155 sb->s_id, function, line, errno, errstr, &vaf);
156 va_end(args);
157 } else {
158 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
159 sb->s_id, function, line, errno, errstr);
160 }
161#endif
162
163 /*
164 * Today we only save the error info to memory. Long term we'll
165 * also send it down to the disk
166 */
167 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
168
169 /* Don't go through full error handling during mount */
170 if (sb->s_flags & MS_BORN)
171 btrfs_handle_error(fs_info);
172}
173
174#ifdef CONFIG_PRINTK
175static const char * const logtypes[] = {
176 "emergency",
177 "alert",
178 "critical",
179 "error",
180 "warning",
181 "notice",
182 "info",
183 "debug",
184};
185
186
187/*
188 * Use one ratelimit state per log level so that a flood of less important
189 * messages doesn't cause more important ones to be dropped.
190 */
191static struct ratelimit_state printk_limits[] = {
192 RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
193 RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
194 RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
195 RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
196 RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
197 RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
198 RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
199 RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
200};
201
202void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
203{
204 struct super_block *sb = fs_info->sb;
205 char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
206 struct va_format vaf;
207 va_list args;
208 int kern_level;
209 const char *type = logtypes[4];
210 struct ratelimit_state *ratelimit = &printk_limits[4];
211
212 va_start(args, fmt);
213
214 while ((kern_level = printk_get_level(fmt)) != 0) {
215 size_t size = printk_skip_level(fmt) - fmt;
216
217 if (kern_level >= '0' && kern_level <= '7') {
218 memcpy(lvl, fmt, size);
219 lvl[size] = '\0';
220 type = logtypes[kern_level - '0'];
221 ratelimit = &printk_limits[kern_level - '0'];
222 }
223 fmt += size;
224 }
225
226 vaf.fmt = fmt;
227 vaf.va = &args;
228
229 if (__ratelimit(ratelimit))
230 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
231
232 va_end(args);
233}
234#endif
235
236/*
237 * We only mark the transaction aborted and then set the file system read-only.
238 * This will prevent new transactions from starting or trying to join this
239 * one.
240 *
241 * This means that error recovery at the call site is limited to freeing
242 * any local memory allocations and passing the error code up without
243 * further cleanup. The transaction should complete as it normally would
244 * in the call path but will return -EIO.
245 *
246 * We'll complete the cleanup in btrfs_end_transaction and
247 * btrfs_commit_transaction.
248 */
249__cold
250void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
251 const char *function,
252 unsigned int line, int errno)
253{
254 struct btrfs_fs_info *fs_info = trans->fs_info;
255
256 trans->aborted = errno;
257 /* Nothing used. The other threads that have joined this
258 * transaction may be able to continue. */
259 if (!trans->dirty && list_empty(&trans->new_bgs)) {
260 const char *errstr;
261
262 errstr = btrfs_decode_error(errno);
263 btrfs_warn(fs_info,
264 "%s:%d: Aborting unused transaction(%s).",
265 function, line, errstr);
266 return;
267 }
268 ACCESS_ONCE(trans->transaction->aborted) = errno;
269 /* Wake up anybody who may be waiting on this transaction */
270 wake_up(&fs_info->transaction_wait);
271 wake_up(&fs_info->transaction_blocked_wait);
272 __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
273}
274/*
275 * __btrfs_panic decodes unexpected, fatal errors from the caller,
276 * issues an alert, and either panics or BUGs, depending on mount options.
277 */
278__cold
279void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
280 unsigned int line, int errno, const char *fmt, ...)
281{
282 char *s_id = "<unknown>";
283 const char *errstr;
284 struct va_format vaf = { .fmt = fmt };
285 va_list args;
286
287 if (fs_info)
288 s_id = fs_info->sb->s_id;
289
290 va_start(args, fmt);
291 vaf.va = &args;
292
293 errstr = btrfs_decode_error(errno);
294 if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
295 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
296 s_id, function, line, &vaf, errno, errstr);
297
298 btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
299 function, line, &vaf, errno, errstr);
300 va_end(args);
301 /* Caller calls BUG() */
302}
303
304static void btrfs_put_super(struct super_block *sb)
305{
306 close_ctree(btrfs_sb(sb));
307}
308
309enum {
310 Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
311 Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
312 Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
313 Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
314 Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
315 Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
316 Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
317 Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
318 Opt_skip_balance, Opt_check_integrity,
319 Opt_check_integrity_including_extent_data,
320 Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
321 Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
322 Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
323 Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
324 Opt_nologreplay, Opt_norecovery,
325#ifdef CONFIG_BTRFS_DEBUG
326 Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
327#endif
328 Opt_err,
329};
330
331static const match_table_t tokens = {
332 {Opt_degraded, "degraded"},
333 {Opt_subvol, "subvol=%s"},
334 {Opt_subvolid, "subvolid=%s"},
335 {Opt_device, "device=%s"},
336 {Opt_nodatasum, "nodatasum"},
337 {Opt_datasum, "datasum"},
338 {Opt_nodatacow, "nodatacow"},
339 {Opt_datacow, "datacow"},
340 {Opt_nobarrier, "nobarrier"},
341 {Opt_barrier, "barrier"},
342 {Opt_max_inline, "max_inline=%s"},
343 {Opt_alloc_start, "alloc_start=%s"},
344 {Opt_thread_pool, "thread_pool=%d"},
345 {Opt_compress, "compress"},
346 {Opt_compress_type, "compress=%s"},
347 {Opt_compress_force, "compress-force"},
348 {Opt_compress_force_type, "compress-force=%s"},
349 {Opt_ssd, "ssd"},
350 {Opt_ssd_spread, "ssd_spread"},
351 {Opt_nossd, "nossd"},
352 {Opt_acl, "acl"},
353 {Opt_noacl, "noacl"},
354 {Opt_notreelog, "notreelog"},
355 {Opt_treelog, "treelog"},
356 {Opt_nologreplay, "nologreplay"},
357 {Opt_norecovery, "norecovery"},
358 {Opt_flushoncommit, "flushoncommit"},
359 {Opt_noflushoncommit, "noflushoncommit"},
360 {Opt_ratio, "metadata_ratio=%d"},
361 {Opt_discard, "discard"},
362 {Opt_nodiscard, "nodiscard"},
363 {Opt_space_cache, "space_cache"},
364 {Opt_space_cache_version, "space_cache=%s"},
365 {Opt_clear_cache, "clear_cache"},
366 {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
367 {Opt_enospc_debug, "enospc_debug"},
368 {Opt_noenospc_debug, "noenospc_debug"},
369 {Opt_subvolrootid, "subvolrootid=%d"},
370 {Opt_defrag, "autodefrag"},
371 {Opt_nodefrag, "noautodefrag"},
372 {Opt_inode_cache, "inode_cache"},
373 {Opt_noinode_cache, "noinode_cache"},
374 {Opt_no_space_cache, "nospace_cache"},
375 {Opt_recovery, "recovery"}, /* deprecated */
376 {Opt_usebackuproot, "usebackuproot"},
377 {Opt_skip_balance, "skip_balance"},
378 {Opt_check_integrity, "check_int"},
379 {Opt_check_integrity_including_extent_data, "check_int_data"},
380 {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
381 {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
382 {Opt_fatal_errors, "fatal_errors=%s"},
383 {Opt_commit_interval, "commit=%d"},
384#ifdef CONFIG_BTRFS_DEBUG
385 {Opt_fragment_data, "fragment=data"},
386 {Opt_fragment_metadata, "fragment=metadata"},
387 {Opt_fragment_all, "fragment=all"},
388#endif
389 {Opt_err, NULL},
390};
391
392/*
393 * Regular mount options parser. Everything that is needed only when
394 * reading in a new superblock is parsed here.
395 * XXX JDM: This needs to be cleaned up for remount.
396 */
397int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
398 unsigned long new_flags)
399{
400 substring_t args[MAX_OPT_ARGS];
401 char *p, *num, *orig = NULL;
402 u64 cache_gen;
403 int intarg;
404 int ret = 0;
405 char *compress_type;
406 bool compress_force = false;
407 enum btrfs_compression_type saved_compress_type;
408 bool saved_compress_force;
409 int no_compress = 0;
410
411 cache_gen = btrfs_super_cache_generation(info->super_copy);
412 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
413 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
414 else if (cache_gen)
415 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
416
417 /*
418 * Even the options are empty, we still need to do extra check
419 * against new flags
420 */
421 if (!options)
422 goto check;
423
424 /*
425 * strsep changes the string, duplicate it because parse_options
426 * gets called twice
427 */
428 options = kstrdup(options, GFP_NOFS);
429 if (!options)
430 return -ENOMEM;
431
432 orig = options;
433
434 while ((p = strsep(&options, ",")) != NULL) {
435 int token;
436 if (!*p)
437 continue;
438
439 token = match_token(p, tokens, args);
440 switch (token) {
441 case Opt_degraded:
442 btrfs_info(info, "allowing degraded mounts");
443 btrfs_set_opt(info->mount_opt, DEGRADED);
444 break;
445 case Opt_subvol:
446 case Opt_subvolid:
447 case Opt_subvolrootid:
448 case Opt_device:
449 /*
450 * These are parsed by btrfs_parse_early_options
451 * and can be happily ignored here.
452 */
453 break;
454 case Opt_nodatasum:
455 btrfs_set_and_info(info, NODATASUM,
456 "setting nodatasum");
457 break;
458 case Opt_datasum:
459 if (btrfs_test_opt(info, NODATASUM)) {
460 if (btrfs_test_opt(info, NODATACOW))
461 btrfs_info(info,
462 "setting datasum, datacow enabled");
463 else
464 btrfs_info(info, "setting datasum");
465 }
466 btrfs_clear_opt(info->mount_opt, NODATACOW);
467 btrfs_clear_opt(info->mount_opt, NODATASUM);
468 break;
469 case Opt_nodatacow:
470 if (!btrfs_test_opt(info, NODATACOW)) {
471 if (!btrfs_test_opt(info, COMPRESS) ||
472 !btrfs_test_opt(info, FORCE_COMPRESS)) {
473 btrfs_info(info,
474 "setting nodatacow, compression disabled");
475 } else {
476 btrfs_info(info, "setting nodatacow");
477 }
478 }
479 btrfs_clear_opt(info->mount_opt, COMPRESS);
480 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
481 btrfs_set_opt(info->mount_opt, NODATACOW);
482 btrfs_set_opt(info->mount_opt, NODATASUM);
483 break;
484 case Opt_datacow:
485 btrfs_clear_and_info(info, NODATACOW,
486 "setting datacow");
487 break;
488 case Opt_compress_force:
489 case Opt_compress_force_type:
490 compress_force = true;
491 /* Fallthrough */
492 case Opt_compress:
493 case Opt_compress_type:
494 saved_compress_type = btrfs_test_opt(info,
495 COMPRESS) ?
496 info->compress_type : BTRFS_COMPRESS_NONE;
497 saved_compress_force =
498 btrfs_test_opt(info, FORCE_COMPRESS);
499 if (token == Opt_compress ||
500 token == Opt_compress_force ||
501 strcmp(args[0].from, "zlib") == 0) {
502 compress_type = "zlib";
503 info->compress_type = BTRFS_COMPRESS_ZLIB;
504 btrfs_set_opt(info->mount_opt, COMPRESS);
505 btrfs_clear_opt(info->mount_opt, NODATACOW);
506 btrfs_clear_opt(info->mount_opt, NODATASUM);
507 no_compress = 0;
508 } else if (strcmp(args[0].from, "lzo") == 0) {
509 compress_type = "lzo";
510 info->compress_type = BTRFS_COMPRESS_LZO;
511 btrfs_set_opt(info->mount_opt, COMPRESS);
512 btrfs_clear_opt(info->mount_opt, NODATACOW);
513 btrfs_clear_opt(info->mount_opt, NODATASUM);
514 btrfs_set_fs_incompat(info, COMPRESS_LZO);
515 no_compress = 0;
516 } else if (strncmp(args[0].from, "no", 2) == 0) {
517 compress_type = "no";
518 btrfs_clear_opt(info->mount_opt, COMPRESS);
519 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
520 compress_force = false;
521 no_compress++;
522 } else {
523 ret = -EINVAL;
524 goto out;
525 }
526
527 if (compress_force) {
528 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
529 } else {
530 /*
531 * If we remount from compress-force=xxx to
532 * compress=xxx, we need clear FORCE_COMPRESS
533 * flag, otherwise, there is no way for users
534 * to disable forcible compression separately.
535 */
536 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
537 }
538 if ((btrfs_test_opt(info, COMPRESS) &&
539 (info->compress_type != saved_compress_type ||
540 compress_force != saved_compress_force)) ||
541 (!btrfs_test_opt(info, COMPRESS) &&
542 no_compress == 1)) {
543 btrfs_info(info, "%s %s compression",
544 (compress_force) ? "force" : "use",
545 compress_type);
546 }
547 compress_force = false;
548 break;
549 case Opt_ssd:
550 btrfs_set_and_info(info, SSD,
551 "use ssd allocation scheme");
552 break;
553 case Opt_ssd_spread:
554 btrfs_set_and_info(info, SSD_SPREAD,
555 "use spread ssd allocation scheme");
556 btrfs_set_opt(info->mount_opt, SSD);
557 break;
558 case Opt_nossd:
559 btrfs_set_and_info(info, NOSSD,
560 "not using ssd allocation scheme");
561 btrfs_clear_opt(info->mount_opt, SSD);
562 break;
563 case Opt_barrier:
564 btrfs_clear_and_info(info, NOBARRIER,
565 "turning on barriers");
566 break;
567 case Opt_nobarrier:
568 btrfs_set_and_info(info, NOBARRIER,
569 "turning off barriers");
570 break;
571 case Opt_thread_pool:
572 ret = match_int(&args[0], &intarg);
573 if (ret) {
574 goto out;
575 } else if (intarg > 0) {
576 info->thread_pool_size = intarg;
577 } else {
578 ret = -EINVAL;
579 goto out;
580 }
581 break;
582 case Opt_max_inline:
583 num = match_strdup(&args[0]);
584 if (num) {
585 info->max_inline = memparse(num, NULL);
586 kfree(num);
587
588 if (info->max_inline) {
589 info->max_inline = min_t(u64,
590 info->max_inline,
591 info->sectorsize);
592 }
593 btrfs_info(info, "max_inline at %llu",
594 info->max_inline);
595 } else {
596 ret = -ENOMEM;
597 goto out;
598 }
599 break;
600 case Opt_alloc_start:
601 num = match_strdup(&args[0]);
602 if (num) {
603 mutex_lock(&info->chunk_mutex);
604 info->alloc_start = memparse(num, NULL);
605 mutex_unlock(&info->chunk_mutex);
606 kfree(num);
607 btrfs_info(info, "allocations start at %llu",
608 info->alloc_start);
609 } else {
610 ret = -ENOMEM;
611 goto out;
612 }
613 break;
614 case Opt_acl:
615#ifdef CONFIG_BTRFS_FS_POSIX_ACL
616 info->sb->s_flags |= MS_POSIXACL;
617 break;
618#else
619 btrfs_err(info, "support for ACL not compiled in!");
620 ret = -EINVAL;
621 goto out;
622#endif
623 case Opt_noacl:
624 info->sb->s_flags &= ~MS_POSIXACL;
625 break;
626 case Opt_notreelog:
627 btrfs_set_and_info(info, NOTREELOG,
628 "disabling tree log");
629 break;
630 case Opt_treelog:
631 btrfs_clear_and_info(info, NOTREELOG,
632 "enabling tree log");
633 break;
634 case Opt_norecovery:
635 case Opt_nologreplay:
636 btrfs_set_and_info(info, NOLOGREPLAY,
637 "disabling log replay at mount time");
638 break;
639 case Opt_flushoncommit:
640 btrfs_set_and_info(info, FLUSHONCOMMIT,
641 "turning on flush-on-commit");
642 break;
643 case Opt_noflushoncommit:
644 btrfs_clear_and_info(info, FLUSHONCOMMIT,
645 "turning off flush-on-commit");
646 break;
647 case Opt_ratio:
648 ret = match_int(&args[0], &intarg);
649 if (ret) {
650 goto out;
651 } else if (intarg >= 0) {
652 info->metadata_ratio = intarg;
653 btrfs_info(info, "metadata ratio %d",
654 info->metadata_ratio);
655 } else {
656 ret = -EINVAL;
657 goto out;
658 }
659 break;
660 case Opt_discard:
661 btrfs_set_and_info(info, DISCARD,
662 "turning on discard");
663 break;
664 case Opt_nodiscard:
665 btrfs_clear_and_info(info, DISCARD,
666 "turning off discard");
667 break;
668 case Opt_space_cache:
669 case Opt_space_cache_version:
670 if (token == Opt_space_cache ||
671 strcmp(args[0].from, "v1") == 0) {
672 btrfs_clear_opt(info->mount_opt,
673 FREE_SPACE_TREE);
674 btrfs_set_and_info(info, SPACE_CACHE,
675 "enabling disk space caching");
676 } else if (strcmp(args[0].from, "v2") == 0) {
677 btrfs_clear_opt(info->mount_opt,
678 SPACE_CACHE);
679 btrfs_set_and_info(info, FREE_SPACE_TREE,
680 "enabling free space tree");
681 } else {
682 ret = -EINVAL;
683 goto out;
684 }
685 break;
686 case Opt_rescan_uuid_tree:
687 btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
688 break;
689 case Opt_no_space_cache:
690 if (btrfs_test_opt(info, SPACE_CACHE)) {
691 btrfs_clear_and_info(info, SPACE_CACHE,
692 "disabling disk space caching");
693 }
694 if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
695 btrfs_clear_and_info(info, FREE_SPACE_TREE,
696 "disabling free space tree");
697 }
698 break;
699 case Opt_inode_cache:
700 btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
701 "enabling inode map caching");
702 break;
703 case Opt_noinode_cache:
704 btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
705 "disabling inode map caching");
706 break;
707 case Opt_clear_cache:
708 btrfs_set_and_info(info, CLEAR_CACHE,
709 "force clearing of disk cache");
710 break;
711 case Opt_user_subvol_rm_allowed:
712 btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
713 break;
714 case Opt_enospc_debug:
715 btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
716 break;
717 case Opt_noenospc_debug:
718 btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
719 break;
720 case Opt_defrag:
721 btrfs_set_and_info(info, AUTO_DEFRAG,
722 "enabling auto defrag");
723 break;
724 case Opt_nodefrag:
725 btrfs_clear_and_info(info, AUTO_DEFRAG,
726 "disabling auto defrag");
727 break;
728 case Opt_recovery:
729 btrfs_warn(info,
730 "'recovery' is deprecated, use 'usebackuproot' instead");
731 case Opt_usebackuproot:
732 btrfs_info(info,
733 "trying to use backup root at mount time");
734 btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
735 break;
736 case Opt_skip_balance:
737 btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
738 break;
739#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
740 case Opt_check_integrity_including_extent_data:
741 btrfs_info(info,
742 "enabling check integrity including extent data");
743 btrfs_set_opt(info->mount_opt,
744 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
745 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
746 break;
747 case Opt_check_integrity:
748 btrfs_info(info, "enabling check integrity");
749 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
750 break;
751 case Opt_check_integrity_print_mask:
752 ret = match_int(&args[0], &intarg);
753 if (ret) {
754 goto out;
755 } else if (intarg >= 0) {
756 info->check_integrity_print_mask = intarg;
757 btrfs_info(info,
758 "check_integrity_print_mask 0x%x",
759 info->check_integrity_print_mask);
760 } else {
761 ret = -EINVAL;
762 goto out;
763 }
764 break;
765#else
766 case Opt_check_integrity_including_extent_data:
767 case Opt_check_integrity:
768 case Opt_check_integrity_print_mask:
769 btrfs_err(info,
770 "support for check_integrity* not compiled in!");
771 ret = -EINVAL;
772 goto out;
773#endif
774 case Opt_fatal_errors:
775 if (strcmp(args[0].from, "panic") == 0)
776 btrfs_set_opt(info->mount_opt,
777 PANIC_ON_FATAL_ERROR);
778 else if (strcmp(args[0].from, "bug") == 0)
779 btrfs_clear_opt(info->mount_opt,
780 PANIC_ON_FATAL_ERROR);
781 else {
782 ret = -EINVAL;
783 goto out;
784 }
785 break;
786 case Opt_commit_interval:
787 intarg = 0;
788 ret = match_int(&args[0], &intarg);
789 if (ret < 0) {
790 btrfs_err(info, "invalid commit interval");
791 ret = -EINVAL;
792 goto out;
793 }
794 if (intarg > 0) {
795 if (intarg > 300) {
796 btrfs_warn(info,
797 "excessive commit interval %d",
798 intarg);
799 }
800 info->commit_interval = intarg;
801 } else {
802 btrfs_info(info,
803 "using default commit interval %ds",
804 BTRFS_DEFAULT_COMMIT_INTERVAL);
805 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
806 }
807 break;
808#ifdef CONFIG_BTRFS_DEBUG
809 case Opt_fragment_all:
810 btrfs_info(info, "fragmenting all space");
811 btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
812 btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
813 break;
814 case Opt_fragment_metadata:
815 btrfs_info(info, "fragmenting metadata");
816 btrfs_set_opt(info->mount_opt,
817 FRAGMENT_METADATA);
818 break;
819 case Opt_fragment_data:
820 btrfs_info(info, "fragmenting data");
821 btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
822 break;
823#endif
824 case Opt_err:
825 btrfs_info(info, "unrecognized mount option '%s'", p);
826 ret = -EINVAL;
827 goto out;
828 default:
829 break;
830 }
831 }
832check:
833 /*
834 * Extra check for current option against current flag
835 */
836 if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
837 btrfs_err(info,
838 "nologreplay must be used with ro mount option");
839 ret = -EINVAL;
840 }
841out:
842 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
843 !btrfs_test_opt(info, FREE_SPACE_TREE) &&
844 !btrfs_test_opt(info, CLEAR_CACHE)) {
845 btrfs_err(info, "cannot disable free space tree");
846 ret = -EINVAL;
847
848 }
849 if (!ret && btrfs_test_opt(info, SPACE_CACHE))
850 btrfs_info(info, "disk space caching is enabled");
851 if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
852 btrfs_info(info, "using free space tree");
853 kfree(orig);
854 return ret;
855}
856
857/*
858 * Parse mount options that are required early in the mount process.
859 *
860 * All other options will be parsed on much later in the mount process and
861 * only when we need to allocate a new super block.
862 */
863static int btrfs_parse_early_options(const char *options, fmode_t flags,
864 void *holder, char **subvol_name, u64 *subvol_objectid,
865 struct btrfs_fs_devices **fs_devices)
866{
867 substring_t args[MAX_OPT_ARGS];
868 char *device_name, *opts, *orig, *p;
869 char *num = NULL;
870 int error = 0;
871
872 if (!options)
873 return 0;
874
875 /*
876 * strsep changes the string, duplicate it because parse_options
877 * gets called twice
878 */
879 opts = kstrdup(options, GFP_KERNEL);
880 if (!opts)
881 return -ENOMEM;
882 orig = opts;
883
884 while ((p = strsep(&opts, ",")) != NULL) {
885 int token;
886 if (!*p)
887 continue;
888
889 token = match_token(p, tokens, args);
890 switch (token) {
891 case Opt_subvol:
892 kfree(*subvol_name);
893 *subvol_name = match_strdup(&args[0]);
894 if (!*subvol_name) {
895 error = -ENOMEM;
896 goto out;
897 }
898 break;
899 case Opt_subvolid:
900 num = match_strdup(&args[0]);
901 if (num) {
902 *subvol_objectid = memparse(num, NULL);
903 kfree(num);
904 /* we want the original fs_tree */
905 if (!*subvol_objectid)
906 *subvol_objectid =
907 BTRFS_FS_TREE_OBJECTID;
908 } else {
909 error = -EINVAL;
910 goto out;
911 }
912 break;
913 case Opt_subvolrootid:
914 pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
915 break;
916 case Opt_device:
917 device_name = match_strdup(&args[0]);
918 if (!device_name) {
919 error = -ENOMEM;
920 goto out;
921 }
922 error = btrfs_scan_one_device(device_name,
923 flags, holder, fs_devices);
924 kfree(device_name);
925 if (error)
926 goto out;
927 break;
928 default:
929 break;
930 }
931 }
932
933out:
934 kfree(orig);
935 return error;
936}
937
938static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
939 u64 subvol_objectid)
940{
941 struct btrfs_root *root = fs_info->tree_root;
942 struct btrfs_root *fs_root;
943 struct btrfs_root_ref *root_ref;
944 struct btrfs_inode_ref *inode_ref;
945 struct btrfs_key key;
946 struct btrfs_path *path = NULL;
947 char *name = NULL, *ptr;
948 u64 dirid;
949 int len;
950 int ret;
951
952 path = btrfs_alloc_path();
953 if (!path) {
954 ret = -ENOMEM;
955 goto err;
956 }
957 path->leave_spinning = 1;
958
959 name = kmalloc(PATH_MAX, GFP_NOFS);
960 if (!name) {
961 ret = -ENOMEM;
962 goto err;
963 }
964 ptr = name + PATH_MAX - 1;
965 ptr[0] = '\0';
966
967 /*
968 * Walk up the subvolume trees in the tree of tree roots by root
969 * backrefs until we hit the top-level subvolume.
970 */
971 while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
972 key.objectid = subvol_objectid;
973 key.type = BTRFS_ROOT_BACKREF_KEY;
974 key.offset = (u64)-1;
975
976 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
977 if (ret < 0) {
978 goto err;
979 } else if (ret > 0) {
980 ret = btrfs_previous_item(root, path, subvol_objectid,
981 BTRFS_ROOT_BACKREF_KEY);
982 if (ret < 0) {
983 goto err;
984 } else if (ret > 0) {
985 ret = -ENOENT;
986 goto err;
987 }
988 }
989
990 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
991 subvol_objectid = key.offset;
992
993 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
994 struct btrfs_root_ref);
995 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
996 ptr -= len + 1;
997 if (ptr < name) {
998 ret = -ENAMETOOLONG;
999 goto err;
1000 }
1001 read_extent_buffer(path->nodes[0], ptr + 1,
1002 (unsigned long)(root_ref + 1), len);
1003 ptr[0] = '/';
1004 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1005 btrfs_release_path(path);
1006
1007 key.objectid = subvol_objectid;
1008 key.type = BTRFS_ROOT_ITEM_KEY;
1009 key.offset = (u64)-1;
1010 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1011 if (IS_ERR(fs_root)) {
1012 ret = PTR_ERR(fs_root);
1013 goto err;
1014 }
1015
1016 /*
1017 * Walk up the filesystem tree by inode refs until we hit the
1018 * root directory.
1019 */
1020 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1021 key.objectid = dirid;
1022 key.type = BTRFS_INODE_REF_KEY;
1023 key.offset = (u64)-1;
1024
1025 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1026 if (ret < 0) {
1027 goto err;
1028 } else if (ret > 0) {
1029 ret = btrfs_previous_item(fs_root, path, dirid,
1030 BTRFS_INODE_REF_KEY);
1031 if (ret < 0) {
1032 goto err;
1033 } else if (ret > 0) {
1034 ret = -ENOENT;
1035 goto err;
1036 }
1037 }
1038
1039 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1040 dirid = key.offset;
1041
1042 inode_ref = btrfs_item_ptr(path->nodes[0],
1043 path->slots[0],
1044 struct btrfs_inode_ref);
1045 len = btrfs_inode_ref_name_len(path->nodes[0],
1046 inode_ref);
1047 ptr -= len + 1;
1048 if (ptr < name) {
1049 ret = -ENAMETOOLONG;
1050 goto err;
1051 }
1052 read_extent_buffer(path->nodes[0], ptr + 1,
1053 (unsigned long)(inode_ref + 1), len);
1054 ptr[0] = '/';
1055 btrfs_release_path(path);
1056 }
1057 }
1058
1059 btrfs_free_path(path);
1060 if (ptr == name + PATH_MAX - 1) {
1061 name[0] = '/';
1062 name[1] = '\0';
1063 } else {
1064 memmove(name, ptr, name + PATH_MAX - ptr);
1065 }
1066 return name;
1067
1068err:
1069 btrfs_free_path(path);
1070 kfree(name);
1071 return ERR_PTR(ret);
1072}
1073
1074static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1075{
1076 struct btrfs_root *root = fs_info->tree_root;
1077 struct btrfs_dir_item *di;
1078 struct btrfs_path *path;
1079 struct btrfs_key location;
1080 u64 dir_id;
1081
1082 path = btrfs_alloc_path();
1083 if (!path)
1084 return -ENOMEM;
1085 path->leave_spinning = 1;
1086
1087 /*
1088 * Find the "default" dir item which points to the root item that we
1089 * will mount by default if we haven't been given a specific subvolume
1090 * to mount.
1091 */
1092 dir_id = btrfs_super_root_dir(fs_info->super_copy);
1093 di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1094 if (IS_ERR(di)) {
1095 btrfs_free_path(path);
1096 return PTR_ERR(di);
1097 }
1098 if (!di) {
1099 /*
1100 * Ok the default dir item isn't there. This is weird since
1101 * it's always been there, but don't freak out, just try and
1102 * mount the top-level subvolume.
1103 */
1104 btrfs_free_path(path);
1105 *objectid = BTRFS_FS_TREE_OBJECTID;
1106 return 0;
1107 }
1108
1109 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1110 btrfs_free_path(path);
1111 *objectid = location.objectid;
1112 return 0;
1113}
1114
1115static int btrfs_fill_super(struct super_block *sb,
1116 struct btrfs_fs_devices *fs_devices,
1117 void *data, int silent)
1118{
1119 struct inode *inode;
1120 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1121 struct btrfs_key key;
1122 int err;
1123
1124 sb->s_maxbytes = MAX_LFS_FILESIZE;
1125 sb->s_magic = BTRFS_SUPER_MAGIC;
1126 sb->s_op = &btrfs_super_ops;
1127 sb->s_d_op = &btrfs_dentry_operations;
1128 sb->s_export_op = &btrfs_export_ops;
1129 sb->s_xattr = btrfs_xattr_handlers;
1130 sb->s_time_gran = 1;
1131#ifdef CONFIG_BTRFS_FS_POSIX_ACL
1132 sb->s_flags |= MS_POSIXACL;
1133#endif
1134 sb->s_flags |= MS_I_VERSION;
1135 sb->s_iflags |= SB_I_CGROUPWB;
1136 err = open_ctree(sb, fs_devices, (char *)data);
1137 if (err) {
1138 btrfs_err(fs_info, "open_ctree failed");
1139 return err;
1140 }
1141
1142 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1143 key.type = BTRFS_INODE_ITEM_KEY;
1144 key.offset = 0;
1145 inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1146 if (IS_ERR(inode)) {
1147 err = PTR_ERR(inode);
1148 goto fail_close;
1149 }
1150
1151 sb->s_root = d_make_root(inode);
1152 if (!sb->s_root) {
1153 err = -ENOMEM;
1154 goto fail_close;
1155 }
1156
1157 save_mount_options(sb, data);
1158 cleancache_init_fs(sb);
1159 sb->s_flags |= MS_ACTIVE;
1160 return 0;
1161
1162fail_close:
1163 close_ctree(fs_info);
1164 return err;
1165}
1166
1167int btrfs_sync_fs(struct super_block *sb, int wait)
1168{
1169 struct btrfs_trans_handle *trans;
1170 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1171 struct btrfs_root *root = fs_info->tree_root;
1172
1173 trace_btrfs_sync_fs(fs_info, wait);
1174
1175 if (!wait) {
1176 filemap_flush(fs_info->btree_inode->i_mapping);
1177 return 0;
1178 }
1179
1180 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
1181
1182 trans = btrfs_attach_transaction_barrier(root);
1183 if (IS_ERR(trans)) {
1184 /* no transaction, don't bother */
1185 if (PTR_ERR(trans) == -ENOENT) {
1186 /*
1187 * Exit unless we have some pending changes
1188 * that need to go through commit
1189 */
1190 if (fs_info->pending_changes == 0)
1191 return 0;
1192 /*
1193 * A non-blocking test if the fs is frozen. We must not
1194 * start a new transaction here otherwise a deadlock
1195 * happens. The pending operations are delayed to the
1196 * next commit after thawing.
1197 */
1198 if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1199 __sb_end_write(sb, SB_FREEZE_WRITE);
1200 else
1201 return 0;
1202 trans = btrfs_start_transaction(root, 0);
1203 }
1204 if (IS_ERR(trans))
1205 return PTR_ERR(trans);
1206 }
1207 return btrfs_commit_transaction(trans);
1208}
1209
1210static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1211{
1212 struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1213 char *compress_type;
1214
1215 if (btrfs_test_opt(info, DEGRADED))
1216 seq_puts(seq, ",degraded");
1217 if (btrfs_test_opt(info, NODATASUM))
1218 seq_puts(seq, ",nodatasum");
1219 if (btrfs_test_opt(info, NODATACOW))
1220 seq_puts(seq, ",nodatacow");
1221 if (btrfs_test_opt(info, NOBARRIER))
1222 seq_puts(seq, ",nobarrier");
1223 if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1224 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1225 if (info->alloc_start != 0)
1226 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1227 if (info->thread_pool_size != min_t(unsigned long,
1228 num_online_cpus() + 2, 8))
1229 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1230 if (btrfs_test_opt(info, COMPRESS)) {
1231 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1232 compress_type = "zlib";
1233 else
1234 compress_type = "lzo";
1235 if (btrfs_test_opt(info, FORCE_COMPRESS))
1236 seq_printf(seq, ",compress-force=%s", compress_type);
1237 else
1238 seq_printf(seq, ",compress=%s", compress_type);
1239 }
1240 if (btrfs_test_opt(info, NOSSD))
1241 seq_puts(seq, ",nossd");
1242 if (btrfs_test_opt(info, SSD_SPREAD))
1243 seq_puts(seq, ",ssd_spread");
1244 else if (btrfs_test_opt(info, SSD))
1245 seq_puts(seq, ",ssd");
1246 if (btrfs_test_opt(info, NOTREELOG))
1247 seq_puts(seq, ",notreelog");
1248 if (btrfs_test_opt(info, NOLOGREPLAY))
1249 seq_puts(seq, ",nologreplay");
1250 if (btrfs_test_opt(info, FLUSHONCOMMIT))
1251 seq_puts(seq, ",flushoncommit");
1252 if (btrfs_test_opt(info, DISCARD))
1253 seq_puts(seq, ",discard");
1254 if (!(info->sb->s_flags & MS_POSIXACL))
1255 seq_puts(seq, ",noacl");
1256 if (btrfs_test_opt(info, SPACE_CACHE))
1257 seq_puts(seq, ",space_cache");
1258 else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1259 seq_puts(seq, ",space_cache=v2");
1260 else
1261 seq_puts(seq, ",nospace_cache");
1262 if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1263 seq_puts(seq, ",rescan_uuid_tree");
1264 if (btrfs_test_opt(info, CLEAR_CACHE))
1265 seq_puts(seq, ",clear_cache");
1266 if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1267 seq_puts(seq, ",user_subvol_rm_allowed");
1268 if (btrfs_test_opt(info, ENOSPC_DEBUG))
1269 seq_puts(seq, ",enospc_debug");
1270 if (btrfs_test_opt(info, AUTO_DEFRAG))
1271 seq_puts(seq, ",autodefrag");
1272 if (btrfs_test_opt(info, INODE_MAP_CACHE))
1273 seq_puts(seq, ",inode_cache");
1274 if (btrfs_test_opt(info, SKIP_BALANCE))
1275 seq_puts(seq, ",skip_balance");
1276#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1277 if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1278 seq_puts(seq, ",check_int_data");
1279 else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1280 seq_puts(seq, ",check_int");
1281 if (info->check_integrity_print_mask)
1282 seq_printf(seq, ",check_int_print_mask=%d",
1283 info->check_integrity_print_mask);
1284#endif
1285 if (info->metadata_ratio)
1286 seq_printf(seq, ",metadata_ratio=%d",
1287 info->metadata_ratio);
1288 if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1289 seq_puts(seq, ",fatal_errors=panic");
1290 if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1291 seq_printf(seq, ",commit=%d", info->commit_interval);
1292#ifdef CONFIG_BTRFS_DEBUG
1293 if (btrfs_test_opt(info, FRAGMENT_DATA))
1294 seq_puts(seq, ",fragment=data");
1295 if (btrfs_test_opt(info, FRAGMENT_METADATA))
1296 seq_puts(seq, ",fragment=metadata");
1297#endif
1298 seq_printf(seq, ",subvolid=%llu",
1299 BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1300 seq_puts(seq, ",subvol=");
1301 seq_dentry(seq, dentry, " \t\n\\");
1302 return 0;
1303}
1304
1305static int btrfs_test_super(struct super_block *s, void *data)
1306{
1307 struct btrfs_fs_info *p = data;
1308 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1309
1310 return fs_info->fs_devices == p->fs_devices;
1311}
1312
1313static int btrfs_set_super(struct super_block *s, void *data)
1314{
1315 int err = set_anon_super(s, data);
1316 if (!err)
1317 s->s_fs_info = data;
1318 return err;
1319}
1320
1321/*
1322 * subvolumes are identified by ino 256
1323 */
1324static inline int is_subvolume_inode(struct inode *inode)
1325{
1326 if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1327 return 1;
1328 return 0;
1329}
1330
1331/*
1332 * This will add subvolid=0 to the argument string while removing any subvol=
1333 * and subvolid= arguments to make sure we get the top-level root for path
1334 * walking to the subvol we want.
1335 */
1336static char *setup_root_args(char *args)
1337{
1338 char *buf, *dst, *sep;
1339
1340 if (!args)
1341 return kstrdup("subvolid=0", GFP_NOFS);
1342
1343 /* The worst case is that we add ",subvolid=0" to the end. */
1344 buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1345 if (!buf)
1346 return NULL;
1347
1348 while (1) {
1349 sep = strchrnul(args, ',');
1350 if (!strstarts(args, "subvol=") &&
1351 !strstarts(args, "subvolid=")) {
1352 memcpy(dst, args, sep - args);
1353 dst += sep - args;
1354 *dst++ = ',';
1355 }
1356 if (*sep)
1357 args = sep + 1;
1358 else
1359 break;
1360 }
1361 strcpy(dst, "subvolid=0");
1362
1363 return buf;
1364}
1365
1366static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1367 int flags, const char *device_name,
1368 char *data)
1369{
1370 struct dentry *root;
1371 struct vfsmount *mnt = NULL;
1372 char *newargs;
1373 int ret;
1374
1375 newargs = setup_root_args(data);
1376 if (!newargs) {
1377 root = ERR_PTR(-ENOMEM);
1378 goto out;
1379 }
1380
1381 mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1382 if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1383 if (flags & MS_RDONLY) {
1384 mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1385 device_name, newargs);
1386 } else {
1387 mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1388 device_name, newargs);
1389 if (IS_ERR(mnt)) {
1390 root = ERR_CAST(mnt);
1391 mnt = NULL;
1392 goto out;
1393 }
1394
1395 down_write(&mnt->mnt_sb->s_umount);
1396 ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1397 up_write(&mnt->mnt_sb->s_umount);
1398 if (ret < 0) {
1399 root = ERR_PTR(ret);
1400 goto out;
1401 }
1402 }
1403 }
1404 if (IS_ERR(mnt)) {
1405 root = ERR_CAST(mnt);
1406 mnt = NULL;
1407 goto out;
1408 }
1409
1410 if (!subvol_name) {
1411 if (!subvol_objectid) {
1412 ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1413 &subvol_objectid);
1414 if (ret) {
1415 root = ERR_PTR(ret);
1416 goto out;
1417 }
1418 }
1419 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1420 subvol_objectid);
1421 if (IS_ERR(subvol_name)) {
1422 root = ERR_CAST(subvol_name);
1423 subvol_name = NULL;
1424 goto out;
1425 }
1426
1427 }
1428
1429 root = mount_subtree(mnt, subvol_name);
1430 /* mount_subtree() drops our reference on the vfsmount. */
1431 mnt = NULL;
1432
1433 if (!IS_ERR(root)) {
1434 struct super_block *s = root->d_sb;
1435 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1436 struct inode *root_inode = d_inode(root);
1437 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1438
1439 ret = 0;
1440 if (!is_subvolume_inode(root_inode)) {
1441 btrfs_err(fs_info, "'%s' is not a valid subvolume",
1442 subvol_name);
1443 ret = -EINVAL;
1444 }
1445 if (subvol_objectid && root_objectid != subvol_objectid) {
1446 /*
1447 * This will also catch a race condition where a
1448 * subvolume which was passed by ID is renamed and
1449 * another subvolume is renamed over the old location.
1450 */
1451 btrfs_err(fs_info,
1452 "subvol '%s' does not match subvolid %llu",
1453 subvol_name, subvol_objectid);
1454 ret = -EINVAL;
1455 }
1456 if (ret) {
1457 dput(root);
1458 root = ERR_PTR(ret);
1459 deactivate_locked_super(s);
1460 }
1461 }
1462
1463out:
1464 mntput(mnt);
1465 kfree(newargs);
1466 kfree(subvol_name);
1467 return root;
1468}
1469
1470static int parse_security_options(char *orig_opts,
1471 struct security_mnt_opts *sec_opts)
1472{
1473 char *secdata = NULL;
1474 int ret = 0;
1475
1476 secdata = alloc_secdata();
1477 if (!secdata)
1478 return -ENOMEM;
1479 ret = security_sb_copy_data(orig_opts, secdata);
1480 if (ret) {
1481 free_secdata(secdata);
1482 return ret;
1483 }
1484 ret = security_sb_parse_opts_str(secdata, sec_opts);
1485 free_secdata(secdata);
1486 return ret;
1487}
1488
1489static int setup_security_options(struct btrfs_fs_info *fs_info,
1490 struct super_block *sb,
1491 struct security_mnt_opts *sec_opts)
1492{
1493 int ret = 0;
1494
1495 /*
1496 * Call security_sb_set_mnt_opts() to check whether new sec_opts
1497 * is valid.
1498 */
1499 ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1500 if (ret)
1501 return ret;
1502
1503#ifdef CONFIG_SECURITY
1504 if (!fs_info->security_opts.num_mnt_opts) {
1505 /* first time security setup, copy sec_opts to fs_info */
1506 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1507 } else {
1508 /*
1509 * Since SELinux (the only one supporting security_mnt_opts)
1510 * does NOT support changing context during remount/mount of
1511 * the same sb, this must be the same or part of the same
1512 * security options, just free it.
1513 */
1514 security_free_mnt_opts(sec_opts);
1515 }
1516#endif
1517 return ret;
1518}
1519
1520/*
1521 * Find a superblock for the given device / mount point.
1522 *
1523 * Note: This is based on get_sb_bdev from fs/super.c with a few additions
1524 * for multiple device setup. Make sure to keep it in sync.
1525 */
1526static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1527 const char *device_name, void *data)
1528{
1529 struct block_device *bdev = NULL;
1530 struct super_block *s;
1531 struct btrfs_fs_devices *fs_devices = NULL;
1532 struct btrfs_fs_info *fs_info = NULL;
1533 struct security_mnt_opts new_sec_opts;
1534 fmode_t mode = FMODE_READ;
1535 char *subvol_name = NULL;
1536 u64 subvol_objectid = 0;
1537 int error = 0;
1538
1539 if (!(flags & MS_RDONLY))
1540 mode |= FMODE_WRITE;
1541
1542 error = btrfs_parse_early_options(data, mode, fs_type,
1543 &subvol_name, &subvol_objectid,
1544 &fs_devices);
1545 if (error) {
1546 kfree(subvol_name);
1547 return ERR_PTR(error);
1548 }
1549
1550 if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1551 /* mount_subvol() will free subvol_name. */
1552 return mount_subvol(subvol_name, subvol_objectid, flags,
1553 device_name, data);
1554 }
1555
1556 security_init_mnt_opts(&new_sec_opts);
1557 if (data) {
1558 error = parse_security_options(data, &new_sec_opts);
1559 if (error)
1560 return ERR_PTR(error);
1561 }
1562
1563 error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1564 if (error)
1565 goto error_sec_opts;
1566
1567 /*
1568 * Setup a dummy root and fs_info for test/set super. This is because
1569 * we don't actually fill this stuff out until open_ctree, but we need
1570 * it for searching for existing supers, so this lets us do that and
1571 * then open_ctree will properly initialize everything later.
1572 */
1573 fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1574 if (!fs_info) {
1575 error = -ENOMEM;
1576 goto error_sec_opts;
1577 }
1578
1579 fs_info->fs_devices = fs_devices;
1580
1581 fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1582 fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1583 security_init_mnt_opts(&fs_info->security_opts);
1584 if (!fs_info->super_copy || !fs_info->super_for_commit) {
1585 error = -ENOMEM;
1586 goto error_fs_info;
1587 }
1588
1589 error = btrfs_open_devices(fs_devices, mode, fs_type);
1590 if (error)
1591 goto error_fs_info;
1592
1593 if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1594 error = -EACCES;
1595 goto error_close_devices;
1596 }
1597
1598 bdev = fs_devices->latest_bdev;
1599 s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1600 fs_info);
1601 if (IS_ERR(s)) {
1602 error = PTR_ERR(s);
1603 goto error_close_devices;
1604 }
1605
1606 if (s->s_root) {
1607 btrfs_close_devices(fs_devices);
1608 free_fs_info(fs_info);
1609 if ((flags ^ s->s_flags) & MS_RDONLY)
1610 error = -EBUSY;
1611 } else {
1612 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1613 btrfs_sb(s)->bdev_holder = fs_type;
1614 error = btrfs_fill_super(s, fs_devices, data,
1615 flags & MS_SILENT ? 1 : 0);
1616 }
1617 if (error) {
1618 deactivate_locked_super(s);
1619 goto error_sec_opts;
1620 }
1621
1622 fs_info = btrfs_sb(s);
1623 error = setup_security_options(fs_info, s, &new_sec_opts);
1624 if (error) {
1625 deactivate_locked_super(s);
1626 goto error_sec_opts;
1627 }
1628
1629 return dget(s->s_root);
1630
1631error_close_devices:
1632 btrfs_close_devices(fs_devices);
1633error_fs_info:
1634 free_fs_info(fs_info);
1635error_sec_opts:
1636 security_free_mnt_opts(&new_sec_opts);
1637 return ERR_PTR(error);
1638}
1639
1640static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1641 int new_pool_size, int old_pool_size)
1642{
1643 if (new_pool_size == old_pool_size)
1644 return;
1645
1646 fs_info->thread_pool_size = new_pool_size;
1647
1648 btrfs_info(fs_info, "resize thread pool %d -> %d",
1649 old_pool_size, new_pool_size);
1650
1651 btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1652 btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1653 btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1654 btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1655 btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1656 btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1657 btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1658 new_pool_size);
1659 btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1660 btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1661 btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1662 btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1663 btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1664 new_pool_size);
1665}
1666
1667static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1668{
1669 set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1670}
1671
1672static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1673 unsigned long old_opts, int flags)
1674{
1675 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1676 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1677 (flags & MS_RDONLY))) {
1678 /* wait for any defraggers to finish */
1679 wait_event(fs_info->transaction_wait,
1680 (atomic_read(&fs_info->defrag_running) == 0));
1681 if (flags & MS_RDONLY)
1682 sync_filesystem(fs_info->sb);
1683 }
1684}
1685
1686static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1687 unsigned long old_opts)
1688{
1689 /*
1690 * We need to cleanup all defragable inodes if the autodefragment is
1691 * close or the filesystem is read only.
1692 */
1693 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1694 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1695 (fs_info->sb->s_flags & MS_RDONLY))) {
1696 btrfs_cleanup_defrag_inodes(fs_info);
1697 }
1698
1699 clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1700}
1701
1702static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1703{
1704 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1705 struct btrfs_root *root = fs_info->tree_root;
1706 unsigned old_flags = sb->s_flags;
1707 unsigned long old_opts = fs_info->mount_opt;
1708 unsigned long old_compress_type = fs_info->compress_type;
1709 u64 old_max_inline = fs_info->max_inline;
1710 u64 old_alloc_start = fs_info->alloc_start;
1711 int old_thread_pool_size = fs_info->thread_pool_size;
1712 unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1713 int ret;
1714
1715 sync_filesystem(sb);
1716 btrfs_remount_prepare(fs_info);
1717
1718 if (data) {
1719 struct security_mnt_opts new_sec_opts;
1720
1721 security_init_mnt_opts(&new_sec_opts);
1722 ret = parse_security_options(data, &new_sec_opts);
1723 if (ret)
1724 goto restore;
1725 ret = setup_security_options(fs_info, sb,
1726 &new_sec_opts);
1727 if (ret) {
1728 security_free_mnt_opts(&new_sec_opts);
1729 goto restore;
1730 }
1731 }
1732
1733 ret = btrfs_parse_options(fs_info, data, *flags);
1734 if (ret) {
1735 ret = -EINVAL;
1736 goto restore;
1737 }
1738
1739 btrfs_remount_begin(fs_info, old_opts, *flags);
1740 btrfs_resize_thread_pool(fs_info,
1741 fs_info->thread_pool_size, old_thread_pool_size);
1742
1743 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1744 goto out;
1745
1746 if (*flags & MS_RDONLY) {
1747 /*
1748 * this also happens on 'umount -rf' or on shutdown, when
1749 * the filesystem is busy.
1750 */
1751 cancel_work_sync(&fs_info->async_reclaim_work);
1752
1753 /* wait for the uuid_scan task to finish */
1754 down(&fs_info->uuid_tree_rescan_sem);
1755 /* avoid complains from lockdep et al. */
1756 up(&fs_info->uuid_tree_rescan_sem);
1757
1758 sb->s_flags |= MS_RDONLY;
1759
1760 /*
1761 * Setting MS_RDONLY will put the cleaner thread to
1762 * sleep at the next loop if it's already active.
1763 * If it's already asleep, we'll leave unused block
1764 * groups on disk until we're mounted read-write again
1765 * unless we clean them up here.
1766 */
1767 btrfs_delete_unused_bgs(fs_info);
1768
1769 btrfs_dev_replace_suspend_for_unmount(fs_info);
1770 btrfs_scrub_cancel(fs_info);
1771 btrfs_pause_balance(fs_info);
1772
1773 ret = btrfs_commit_super(fs_info);
1774 if (ret)
1775 goto restore;
1776 } else {
1777 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1778 btrfs_err(fs_info,
1779 "Remounting read-write after error is not allowed");
1780 ret = -EINVAL;
1781 goto restore;
1782 }
1783 if (fs_info->fs_devices->rw_devices == 0) {
1784 ret = -EACCES;
1785 goto restore;
1786 }
1787
1788 if (fs_info->fs_devices->missing_devices >
1789 fs_info->num_tolerated_disk_barrier_failures &&
1790 !(*flags & MS_RDONLY)) {
1791 btrfs_warn(fs_info,
1792 "too many missing devices, writeable remount is not allowed");
1793 ret = -EACCES;
1794 goto restore;
1795 }
1796
1797 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1798 ret = -EINVAL;
1799 goto restore;
1800 }
1801
1802 ret = btrfs_cleanup_fs_roots(fs_info);
1803 if (ret)
1804 goto restore;
1805
1806 /* recover relocation */
1807 mutex_lock(&fs_info->cleaner_mutex);
1808 ret = btrfs_recover_relocation(root);
1809 mutex_unlock(&fs_info->cleaner_mutex);
1810 if (ret)
1811 goto restore;
1812
1813 ret = btrfs_resume_balance_async(fs_info);
1814 if (ret)
1815 goto restore;
1816
1817 ret = btrfs_resume_dev_replace_async(fs_info);
1818 if (ret) {
1819 btrfs_warn(fs_info, "failed to resume dev_replace");
1820 goto restore;
1821 }
1822
1823 if (!fs_info->uuid_root) {
1824 btrfs_info(fs_info, "creating UUID tree");
1825 ret = btrfs_create_uuid_tree(fs_info);
1826 if (ret) {
1827 btrfs_warn(fs_info,
1828 "failed to create the UUID tree %d",
1829 ret);
1830 goto restore;
1831 }
1832 }
1833 sb->s_flags &= ~MS_RDONLY;
1834
1835 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1836 }
1837out:
1838 wake_up_process(fs_info->transaction_kthread);
1839 btrfs_remount_cleanup(fs_info, old_opts);
1840 return 0;
1841
1842restore:
1843 /* We've hit an error - don't reset MS_RDONLY */
1844 if (sb->s_flags & MS_RDONLY)
1845 old_flags |= MS_RDONLY;
1846 sb->s_flags = old_flags;
1847 fs_info->mount_opt = old_opts;
1848 fs_info->compress_type = old_compress_type;
1849 fs_info->max_inline = old_max_inline;
1850 mutex_lock(&fs_info->chunk_mutex);
1851 fs_info->alloc_start = old_alloc_start;
1852 mutex_unlock(&fs_info->chunk_mutex);
1853 btrfs_resize_thread_pool(fs_info,
1854 old_thread_pool_size, fs_info->thread_pool_size);
1855 fs_info->metadata_ratio = old_metadata_ratio;
1856 btrfs_remount_cleanup(fs_info, old_opts);
1857 return ret;
1858}
1859
1860/* Used to sort the devices by max_avail(descending sort) */
1861static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1862 const void *dev_info2)
1863{
1864 if (((struct btrfs_device_info *)dev_info1)->max_avail >
1865 ((struct btrfs_device_info *)dev_info2)->max_avail)
1866 return -1;
1867 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1868 ((struct btrfs_device_info *)dev_info2)->max_avail)
1869 return 1;
1870 else
1871 return 0;
1872}
1873
1874/*
1875 * sort the devices by max_avail, in which max free extent size of each device
1876 * is stored.(Descending Sort)
1877 */
1878static inline void btrfs_descending_sort_devices(
1879 struct btrfs_device_info *devices,
1880 size_t nr_devices)
1881{
1882 sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1883 btrfs_cmp_device_free_bytes, NULL);
1884}
1885
1886/*
1887 * The helper to calc the free space on the devices that can be used to store
1888 * file data.
1889 */
1890static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1891 u64 *free_bytes)
1892{
1893 struct btrfs_root *root = fs_info->tree_root;
1894 struct btrfs_device_info *devices_info;
1895 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1896 struct btrfs_device *device;
1897 u64 skip_space;
1898 u64 type;
1899 u64 avail_space;
1900 u64 used_space;
1901 u64 min_stripe_size;
1902 int min_stripes = 1, num_stripes = 1;
1903 int i = 0, nr_devices;
1904 int ret;
1905
1906 /*
1907 * We aren't under the device list lock, so this is racy-ish, but good
1908 * enough for our purposes.
1909 */
1910 nr_devices = fs_info->fs_devices->open_devices;
1911 if (!nr_devices) {
1912 smp_mb();
1913 nr_devices = fs_info->fs_devices->open_devices;
1914 ASSERT(nr_devices);
1915 if (!nr_devices) {
1916 *free_bytes = 0;
1917 return 0;
1918 }
1919 }
1920
1921 devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1922 GFP_NOFS);
1923 if (!devices_info)
1924 return -ENOMEM;
1925
1926 /* calc min stripe number for data space allocation */
1927 type = btrfs_get_alloc_profile(root, 1);
1928 if (type & BTRFS_BLOCK_GROUP_RAID0) {
1929 min_stripes = 2;
1930 num_stripes = nr_devices;
1931 } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1932 min_stripes = 2;
1933 num_stripes = 2;
1934 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1935 min_stripes = 4;
1936 num_stripes = 4;
1937 }
1938
1939 if (type & BTRFS_BLOCK_GROUP_DUP)
1940 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1941 else
1942 min_stripe_size = BTRFS_STRIPE_LEN;
1943
1944 if (fs_info->alloc_start)
1945 mutex_lock(&fs_devices->device_list_mutex);
1946 rcu_read_lock();
1947 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1948 if (!device->in_fs_metadata || !device->bdev ||
1949 device->is_tgtdev_for_dev_replace)
1950 continue;
1951
1952 if (i >= nr_devices)
1953 break;
1954
1955 avail_space = device->total_bytes - device->bytes_used;
1956
1957 /* align with stripe_len */
1958 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1959 avail_space *= BTRFS_STRIPE_LEN;
1960
1961 /*
1962 * In order to avoid overwriting the superblock on the drive,
1963 * btrfs starts at an offset of at least 1MB when doing chunk
1964 * allocation.
1965 */
1966 skip_space = SZ_1M;
1967
1968 /* user can set the offset in fs_info->alloc_start. */
1969 if (fs_info->alloc_start &&
1970 fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1971 device->total_bytes) {
1972 rcu_read_unlock();
1973 skip_space = max(fs_info->alloc_start, skip_space);
1974
1975 /*
1976 * btrfs can not use the free space in
1977 * [0, skip_space - 1], we must subtract it from the
1978 * total. In order to implement it, we account the used
1979 * space in this range first.
1980 */
1981 ret = btrfs_account_dev_extents_size(device, 0,
1982 skip_space - 1,
1983 &used_space);
1984 if (ret) {
1985 kfree(devices_info);
1986 mutex_unlock(&fs_devices->device_list_mutex);
1987 return ret;
1988 }
1989
1990 rcu_read_lock();
1991
1992 /* calc the free space in [0, skip_space - 1] */
1993 skip_space -= used_space;
1994 }
1995
1996 /*
1997 * we can use the free space in [0, skip_space - 1], subtract
1998 * it from the total.
1999 */
2000 if (avail_space && avail_space >= skip_space)
2001 avail_space -= skip_space;
2002 else
2003 avail_space = 0;
2004
2005 if (avail_space < min_stripe_size)
2006 continue;
2007
2008 devices_info[i].dev = device;
2009 devices_info[i].max_avail = avail_space;
2010
2011 i++;
2012 }
2013 rcu_read_unlock();
2014 if (fs_info->alloc_start)
2015 mutex_unlock(&fs_devices->device_list_mutex);
2016
2017 nr_devices = i;
2018
2019 btrfs_descending_sort_devices(devices_info, nr_devices);
2020
2021 i = nr_devices - 1;
2022 avail_space = 0;
2023 while (nr_devices >= min_stripes) {
2024 if (num_stripes > nr_devices)
2025 num_stripes = nr_devices;
2026
2027 if (devices_info[i].max_avail >= min_stripe_size) {
2028 int j;
2029 u64 alloc_size;
2030
2031 avail_space += devices_info[i].max_avail * num_stripes;
2032 alloc_size = devices_info[i].max_avail;
2033 for (j = i + 1 - num_stripes; j <= i; j++)
2034 devices_info[j].max_avail -= alloc_size;
2035 }
2036 i--;
2037 nr_devices--;
2038 }
2039
2040 kfree(devices_info);
2041 *free_bytes = avail_space;
2042 return 0;
2043}
2044
2045/*
2046 * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2047 *
2048 * If there's a redundant raid level at DATA block groups, use the respective
2049 * multiplier to scale the sizes.
2050 *
2051 * Unused device space usage is based on simulating the chunk allocator
2052 * algorithm that respects the device sizes, order of allocations and the
2053 * 'alloc_start' value, this is a close approximation of the actual use but
2054 * there are other factors that may change the result (like a new metadata
2055 * chunk).
2056 *
2057 * If metadata is exhausted, f_bavail will be 0.
2058 */
2059static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2060{
2061 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2062 struct btrfs_super_block *disk_super = fs_info->super_copy;
2063 struct list_head *head = &fs_info->space_info;
2064 struct btrfs_space_info *found;
2065 u64 total_used = 0;
2066 u64 total_free_data = 0;
2067 u64 total_free_meta = 0;
2068 int bits = dentry->d_sb->s_blocksize_bits;
2069 __be32 *fsid = (__be32 *)fs_info->fsid;
2070 unsigned factor = 1;
2071 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2072 int ret;
2073 u64 thresh = 0;
2074 int mixed = 0;
2075
2076 rcu_read_lock();
2077 list_for_each_entry_rcu(found, head, list) {
2078 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2079 int i;
2080
2081 total_free_data += found->disk_total - found->disk_used;
2082 total_free_data -=
2083 btrfs_account_ro_block_groups_free_space(found);
2084
2085 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2086 if (!list_empty(&found->block_groups[i])) {
2087 switch (i) {
2088 case BTRFS_RAID_DUP:
2089 case BTRFS_RAID_RAID1:
2090 case BTRFS_RAID_RAID10:
2091 factor = 2;
2092 }
2093 }
2094 }
2095 }
2096
2097 /*
2098 * Metadata in mixed block goup profiles are accounted in data
2099 */
2100 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2101 if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2102 mixed = 1;
2103 else
2104 total_free_meta += found->disk_total -
2105 found->disk_used;
2106 }
2107
2108 total_used += found->disk_used;
2109 }
2110
2111 rcu_read_unlock();
2112
2113 buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2114 buf->f_blocks >>= bits;
2115 buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2116
2117 /* Account global block reserve as used, it's in logical size already */
2118 spin_lock(&block_rsv->lock);
2119 /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2120 if (buf->f_bfree >= block_rsv->size >> bits)
2121 buf->f_bfree -= block_rsv->size >> bits;
2122 else
2123 buf->f_bfree = 0;
2124 spin_unlock(&block_rsv->lock);
2125
2126 buf->f_bavail = div_u64(total_free_data, factor);
2127 ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2128 if (ret)
2129 return ret;
2130 buf->f_bavail += div_u64(total_free_data, factor);
2131 buf->f_bavail = buf->f_bavail >> bits;
2132
2133 /*
2134 * We calculate the remaining metadata space minus global reserve. If
2135 * this is (supposedly) smaller than zero, there's no space. But this
2136 * does not hold in practice, the exhausted state happens where's still
2137 * some positive delta. So we apply some guesswork and compare the
2138 * delta to a 4M threshold. (Practically observed delta was ~2M.)
2139 *
2140 * We probably cannot calculate the exact threshold value because this
2141 * depends on the internal reservations requested by various
2142 * operations, so some operations that consume a few metadata will
2143 * succeed even if the Avail is zero. But this is better than the other
2144 * way around.
2145 */
2146 thresh = 4 * 1024 * 1024;
2147
2148 if (!mixed && total_free_meta - thresh < block_rsv->size)
2149 buf->f_bavail = 0;
2150
2151 buf->f_type = BTRFS_SUPER_MAGIC;
2152 buf->f_bsize = dentry->d_sb->s_blocksize;
2153 buf->f_namelen = BTRFS_NAME_LEN;
2154
2155 /* We treat it as constant endianness (it doesn't matter _which_)
2156 because we want the fsid to come out the same whether mounted
2157 on a big-endian or little-endian host */
2158 buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2159 buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2160 /* Mask in the root object ID too, to disambiguate subvols */
2161 buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2162 buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2163
2164 return 0;
2165}
2166
2167static void btrfs_kill_super(struct super_block *sb)
2168{
2169 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2170 kill_anon_super(sb);
2171 free_fs_info(fs_info);
2172}
2173
2174static struct file_system_type btrfs_fs_type = {
2175 .owner = THIS_MODULE,
2176 .name = "btrfs",
2177 .mount = btrfs_mount,
2178 .kill_sb = btrfs_kill_super,
2179 .fs_flags = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2180};
2181MODULE_ALIAS_FS("btrfs");
2182
2183static int btrfs_control_open(struct inode *inode, struct file *file)
2184{
2185 /*
2186 * The control file's private_data is used to hold the
2187 * transaction when it is started and is used to keep
2188 * track of whether a transaction is already in progress.
2189 */
2190 file->private_data = NULL;
2191 return 0;
2192}
2193
2194/*
2195 * used by btrfsctl to scan devices when no FS is mounted
2196 */
2197static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2198 unsigned long arg)
2199{
2200 struct btrfs_ioctl_vol_args *vol;
2201 struct btrfs_fs_devices *fs_devices;
2202 int ret = -ENOTTY;
2203
2204 if (!capable(CAP_SYS_ADMIN))
2205 return -EPERM;
2206
2207 vol = memdup_user((void __user *)arg, sizeof(*vol));
2208 if (IS_ERR(vol))
2209 return PTR_ERR(vol);
2210
2211 switch (cmd) {
2212 case BTRFS_IOC_SCAN_DEV:
2213 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2214 &btrfs_fs_type, &fs_devices);
2215 break;
2216 case BTRFS_IOC_DEVICES_READY:
2217 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2218 &btrfs_fs_type, &fs_devices);
2219 if (ret)
2220 break;
2221 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2222 break;
2223 case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2224 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2225 break;
2226 }
2227
2228 kfree(vol);
2229 return ret;
2230}
2231
2232static int btrfs_freeze(struct super_block *sb)
2233{
2234 struct btrfs_trans_handle *trans;
2235 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2236 struct btrfs_root *root = fs_info->tree_root;
2237
2238 fs_info->fs_frozen = 1;
2239 /*
2240 * We don't need a barrier here, we'll wait for any transaction that
2241 * could be in progress on other threads (and do delayed iputs that
2242 * we want to avoid on a frozen filesystem), or do the commit
2243 * ourselves.
2244 */
2245 trans = btrfs_attach_transaction_barrier(root);
2246 if (IS_ERR(trans)) {
2247 /* no transaction, don't bother */
2248 if (PTR_ERR(trans) == -ENOENT)
2249 return 0;
2250 return PTR_ERR(trans);
2251 }
2252 return btrfs_commit_transaction(trans);
2253}
2254
2255static int btrfs_unfreeze(struct super_block *sb)
2256{
2257 btrfs_sb(sb)->fs_frozen = 0;
2258 return 0;
2259}
2260
2261static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2262{
2263 struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2264 struct btrfs_fs_devices *cur_devices;
2265 struct btrfs_device *dev, *first_dev = NULL;
2266 struct list_head *head;
2267 struct rcu_string *name;
2268
2269 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2270 cur_devices = fs_info->fs_devices;
2271 while (cur_devices) {
2272 head = &cur_devices->devices;
2273 list_for_each_entry(dev, head, dev_list) {
2274 if (dev->missing)
2275 continue;
2276 if (!dev->name)
2277 continue;
2278 if (!first_dev || dev->devid < first_dev->devid)
2279 first_dev = dev;
2280 }
2281 cur_devices = cur_devices->seed;
2282 }
2283
2284 if (first_dev) {
2285 rcu_read_lock();
2286 name = rcu_dereference(first_dev->name);
2287 seq_escape(m, name->str, " \t\n\\");
2288 rcu_read_unlock();
2289 } else {
2290 WARN_ON(1);
2291 }
2292 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2293 return 0;
2294}
2295
2296static const struct super_operations btrfs_super_ops = {
2297 .drop_inode = btrfs_drop_inode,
2298 .evict_inode = btrfs_evict_inode,
2299 .put_super = btrfs_put_super,
2300 .sync_fs = btrfs_sync_fs,
2301 .show_options = btrfs_show_options,
2302 .show_devname = btrfs_show_devname,
2303 .write_inode = btrfs_write_inode,
2304 .alloc_inode = btrfs_alloc_inode,
2305 .destroy_inode = btrfs_destroy_inode,
2306 .statfs = btrfs_statfs,
2307 .remount_fs = btrfs_remount,
2308 .freeze_fs = btrfs_freeze,
2309 .unfreeze_fs = btrfs_unfreeze,
2310};
2311
2312static const struct file_operations btrfs_ctl_fops = {
2313 .open = btrfs_control_open,
2314 .unlocked_ioctl = btrfs_control_ioctl,
2315 .compat_ioctl = btrfs_control_ioctl,
2316 .owner = THIS_MODULE,
2317 .llseek = noop_llseek,
2318};
2319
2320static struct miscdevice btrfs_misc = {
2321 .minor = BTRFS_MINOR,
2322 .name = "btrfs-control",
2323 .fops = &btrfs_ctl_fops
2324};
2325
2326MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2327MODULE_ALIAS("devname:btrfs-control");
2328
2329static int btrfs_interface_init(void)
2330{
2331 return misc_register(&btrfs_misc);
2332}
2333
2334static void btrfs_interface_exit(void)
2335{
2336 misc_deregister(&btrfs_misc);
2337}
2338
2339static void btrfs_print_mod_info(void)
2340{
2341 pr_info("Btrfs loaded, crc32c=%s"
2342#ifdef CONFIG_BTRFS_DEBUG
2343 ", debug=on"
2344#endif
2345#ifdef CONFIG_BTRFS_ASSERT
2346 ", assert=on"
2347#endif
2348#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2349 ", integrity-checker=on"
2350#endif
2351 "\n",
2352 btrfs_crc32c_impl());
2353}
2354
2355static int __init init_btrfs_fs(void)
2356{
2357 int err;
2358
2359 err = btrfs_hash_init();
2360 if (err)
2361 return err;
2362
2363 btrfs_props_init();
2364
2365 err = btrfs_init_sysfs();
2366 if (err)
2367 goto free_hash;
2368
2369 btrfs_init_compress();
2370
2371 err = btrfs_init_cachep();
2372 if (err)
2373 goto free_compress;
2374
2375 err = extent_io_init();
2376 if (err)
2377 goto free_cachep;
2378
2379 err = extent_map_init();
2380 if (err)
2381 goto free_extent_io;
2382
2383 err = ordered_data_init();
2384 if (err)
2385 goto free_extent_map;
2386
2387 err = btrfs_delayed_inode_init();
2388 if (err)
2389 goto free_ordered_data;
2390
2391 err = btrfs_auto_defrag_init();
2392 if (err)
2393 goto free_delayed_inode;
2394
2395 err = btrfs_delayed_ref_init();
2396 if (err)
2397 goto free_auto_defrag;
2398
2399 err = btrfs_prelim_ref_init();
2400 if (err)
2401 goto free_delayed_ref;
2402
2403 err = btrfs_end_io_wq_init();
2404 if (err)
2405 goto free_prelim_ref;
2406
2407 err = btrfs_interface_init();
2408 if (err)
2409 goto free_end_io_wq;
2410
2411 btrfs_init_lockdep();
2412
2413 btrfs_print_mod_info();
2414
2415 err = btrfs_run_sanity_tests();
2416 if (err)
2417 goto unregister_ioctl;
2418
2419 err = register_filesystem(&btrfs_fs_type);
2420 if (err)
2421 goto unregister_ioctl;
2422
2423 return 0;
2424
2425unregister_ioctl:
2426 btrfs_interface_exit();
2427free_end_io_wq:
2428 btrfs_end_io_wq_exit();
2429free_prelim_ref:
2430 btrfs_prelim_ref_exit();
2431free_delayed_ref:
2432 btrfs_delayed_ref_exit();
2433free_auto_defrag:
2434 btrfs_auto_defrag_exit();
2435free_delayed_inode:
2436 btrfs_delayed_inode_exit();
2437free_ordered_data:
2438 ordered_data_exit();
2439free_extent_map:
2440 extent_map_exit();
2441free_extent_io:
2442 extent_io_exit();
2443free_cachep:
2444 btrfs_destroy_cachep();
2445free_compress:
2446 btrfs_exit_compress();
2447 btrfs_exit_sysfs();
2448free_hash:
2449 btrfs_hash_exit();
2450 return err;
2451}
2452
2453static void __exit exit_btrfs_fs(void)
2454{
2455 btrfs_destroy_cachep();
2456 btrfs_delayed_ref_exit();
2457 btrfs_auto_defrag_exit();
2458 btrfs_delayed_inode_exit();
2459 btrfs_prelim_ref_exit();
2460 ordered_data_exit();
2461 extent_map_exit();
2462 extent_io_exit();
2463 btrfs_interface_exit();
2464 btrfs_end_io_wq_exit();
2465 unregister_filesystem(&btrfs_fs_type);
2466 btrfs_exit_sysfs();
2467 btrfs_cleanup_fs_uuids();
2468 btrfs_exit_compress();
2469 btrfs_hash_exit();
2470}
2471
2472late_initcall(init_btrfs_fs);
2473module_exit(exit_btrfs_fs)
2474
2475MODULE_LICENSE("GPL");
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/blkdev.h>
20#include <linux/module.h>
21#include <linux/buffer_head.h>
22#include <linux/fs.h>
23#include <linux/pagemap.h>
24#include <linux/highmem.h>
25#include <linux/time.h>
26#include <linux/init.h>
27#include <linux/seq_file.h>
28#include <linux/string.h>
29#include <linux/backing-dev.h>
30#include <linux/mount.h>
31#include <linux/mpage.h>
32#include <linux/swap.h>
33#include <linux/writeback.h>
34#include <linux/statfs.h>
35#include <linux/compat.h>
36#include <linux/parser.h>
37#include <linux/ctype.h>
38#include <linux/namei.h>
39#include <linux/miscdevice.h>
40#include <linux/magic.h>
41#include <linux/slab.h>
42#include <linux/cleancache.h>
43#include <linux/ratelimit.h>
44#include <linux/btrfs.h>
45#include "delayed-inode.h"
46#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
50#include "print-tree.h"
51#include "hash.h"
52#include "props.h"
53#include "xattr.h"
54#include "volumes.h"
55#include "export.h"
56#include "compression.h"
57#include "rcu-string.h"
58#include "dev-replace.h"
59#include "free-space-cache.h"
60#include "backref.h"
61#include "tests/btrfs-tests.h"
62
63#define CREATE_TRACE_POINTS
64#include <trace/events/btrfs.h>
65
66static const struct super_operations btrfs_super_ops;
67static struct file_system_type btrfs_fs_type;
68
69static int btrfs_remount(struct super_block *sb, int *flags, char *data);
70
71static const char *btrfs_decode_error(int errno)
72{
73 char *errstr = "unknown";
74
75 switch (errno) {
76 case -EIO:
77 errstr = "IO failure";
78 break;
79 case -ENOMEM:
80 errstr = "Out of memory";
81 break;
82 case -EROFS:
83 errstr = "Readonly filesystem";
84 break;
85 case -EEXIST:
86 errstr = "Object already exists";
87 break;
88 case -ENOSPC:
89 errstr = "No space left";
90 break;
91 case -ENOENT:
92 errstr = "No such entry";
93 break;
94 }
95
96 return errstr;
97}
98
99static void save_error_info(struct btrfs_fs_info *fs_info)
100{
101 /*
102 * today we only save the error info into ram. Long term we'll
103 * also send it down to the disk
104 */
105 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
106}
107
108/* btrfs handle error by forcing the filesystem readonly */
109static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
110{
111 struct super_block *sb = fs_info->sb;
112
113 if (sb->s_flags & MS_RDONLY)
114 return;
115
116 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
117 sb->s_flags |= MS_RDONLY;
118 btrfs_info(fs_info, "forced readonly");
119 /*
120 * Note that a running device replace operation is not
121 * canceled here although there is no way to update
122 * the progress. It would add the risk of a deadlock,
123 * therefore the canceling is ommited. The only penalty
124 * is that some I/O remains active until the procedure
125 * completes. The next time when the filesystem is
126 * mounted writeable again, the device replace
127 * operation continues.
128 */
129 }
130}
131
132#ifdef CONFIG_PRINTK
133/*
134 * __btrfs_std_error decodes expected errors from the caller and
135 * invokes the approciate error response.
136 */
137void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
138 unsigned int line, int errno, const char *fmt, ...)
139{
140 struct super_block *sb = fs_info->sb;
141 const char *errstr;
142
143 /*
144 * Special case: if the error is EROFS, and we're already
145 * under MS_RDONLY, then it is safe here.
146 */
147 if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
148 return;
149
150 errstr = btrfs_decode_error(errno);
151 if (fmt) {
152 struct va_format vaf;
153 va_list args;
154
155 va_start(args, fmt);
156 vaf.fmt = fmt;
157 vaf.va = &args;
158
159 printk(KERN_CRIT
160 "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
161 sb->s_id, function, line, errno, errstr, &vaf);
162 va_end(args);
163 } else {
164 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
165 sb->s_id, function, line, errno, errstr);
166 }
167
168 /* Don't go through full error handling during mount */
169 save_error_info(fs_info);
170 if (sb->s_flags & MS_BORN)
171 btrfs_handle_error(fs_info);
172}
173
174static const char * const logtypes[] = {
175 "emergency",
176 "alert",
177 "critical",
178 "error",
179 "warning",
180 "notice",
181 "info",
182 "debug",
183};
184
185void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
186{
187 struct super_block *sb = fs_info->sb;
188 char lvl[4];
189 struct va_format vaf;
190 va_list args;
191 const char *type = logtypes[4];
192 int kern_level;
193
194 va_start(args, fmt);
195
196 kern_level = printk_get_level(fmt);
197 if (kern_level) {
198 size_t size = printk_skip_level(fmt) - fmt;
199 memcpy(lvl, fmt, size);
200 lvl[size] = '\0';
201 fmt += size;
202 type = logtypes[kern_level - '0'];
203 } else
204 *lvl = '\0';
205
206 vaf.fmt = fmt;
207 vaf.va = &args;
208
209 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
210
211 va_end(args);
212}
213
214#else
215
216void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
217 unsigned int line, int errno, const char *fmt, ...)
218{
219 struct super_block *sb = fs_info->sb;
220
221 /*
222 * Special case: if the error is EROFS, and we're already
223 * under MS_RDONLY, then it is safe here.
224 */
225 if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
226 return;
227
228 /* Don't go through full error handling during mount */
229 if (sb->s_flags & MS_BORN) {
230 save_error_info(fs_info);
231 btrfs_handle_error(fs_info);
232 }
233}
234#endif
235
236/*
237 * We only mark the transaction aborted and then set the file system read-only.
238 * This will prevent new transactions from starting or trying to join this
239 * one.
240 *
241 * This means that error recovery at the call site is limited to freeing
242 * any local memory allocations and passing the error code up without
243 * further cleanup. The transaction should complete as it normally would
244 * in the call path but will return -EIO.
245 *
246 * We'll complete the cleanup in btrfs_end_transaction and
247 * btrfs_commit_transaction.
248 */
249void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
250 struct btrfs_root *root, const char *function,
251 unsigned int line, int errno)
252{
253 /*
254 * Report first abort since mount
255 */
256 if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,
257 &root->fs_info->fs_state)) {
258 WARN(1, KERN_DEBUG "BTRFS: Transaction aborted (error %d)\n",
259 errno);
260 }
261 trans->aborted = errno;
262 /* Nothing used. The other threads that have joined this
263 * transaction may be able to continue. */
264 if (!trans->blocks_used) {
265 const char *errstr;
266
267 errstr = btrfs_decode_error(errno);
268 btrfs_warn(root->fs_info,
269 "%s:%d: Aborting unused transaction(%s).",
270 function, line, errstr);
271 return;
272 }
273 ACCESS_ONCE(trans->transaction->aborted) = errno;
274 /* Wake up anybody who may be waiting on this transaction */
275 wake_up(&root->fs_info->transaction_wait);
276 wake_up(&root->fs_info->transaction_blocked_wait);
277 __btrfs_std_error(root->fs_info, function, line, errno, NULL);
278}
279/*
280 * __btrfs_panic decodes unexpected, fatal errors from the caller,
281 * issues an alert, and either panics or BUGs, depending on mount options.
282 */
283void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
284 unsigned int line, int errno, const char *fmt, ...)
285{
286 char *s_id = "<unknown>";
287 const char *errstr;
288 struct va_format vaf = { .fmt = fmt };
289 va_list args;
290
291 if (fs_info)
292 s_id = fs_info->sb->s_id;
293
294 va_start(args, fmt);
295 vaf.va = &args;
296
297 errstr = btrfs_decode_error(errno);
298 if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
299 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
300 s_id, function, line, &vaf, errno, errstr);
301
302 btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
303 function, line, &vaf, errno, errstr);
304 va_end(args);
305 /* Caller calls BUG() */
306}
307
308static void btrfs_put_super(struct super_block *sb)
309{
310 (void)close_ctree(btrfs_sb(sb)->tree_root);
311 /* FIXME: need to fix VFS to return error? */
312 /* AV: return it _where_? ->put_super() can be triggered by any number
313 * of async events, up to and including delivery of SIGKILL to the
314 * last process that kept it busy. Or segfault in the aforementioned
315 * process... Whom would you report that to?
316 */
317}
318
319enum {
320 Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
321 Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
322 Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
323 Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
324 Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
325 Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
326 Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
327 Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
328 Opt_check_integrity, Opt_check_integrity_including_extent_data,
329 Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
330 Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
331 Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
332 Opt_datasum, Opt_treelog, Opt_noinode_cache,
333 Opt_err,
334};
335
336static match_table_t tokens = {
337 {Opt_degraded, "degraded"},
338 {Opt_subvol, "subvol=%s"},
339 {Opt_subvolid, "subvolid=%s"},
340 {Opt_device, "device=%s"},
341 {Opt_nodatasum, "nodatasum"},
342 {Opt_datasum, "datasum"},
343 {Opt_nodatacow, "nodatacow"},
344 {Opt_datacow, "datacow"},
345 {Opt_nobarrier, "nobarrier"},
346 {Opt_barrier, "barrier"},
347 {Opt_max_inline, "max_inline=%s"},
348 {Opt_alloc_start, "alloc_start=%s"},
349 {Opt_thread_pool, "thread_pool=%d"},
350 {Opt_compress, "compress"},
351 {Opt_compress_type, "compress=%s"},
352 {Opt_compress_force, "compress-force"},
353 {Opt_compress_force_type, "compress-force=%s"},
354 {Opt_ssd, "ssd"},
355 {Opt_ssd_spread, "ssd_spread"},
356 {Opt_nossd, "nossd"},
357 {Opt_acl, "acl"},
358 {Opt_noacl, "noacl"},
359 {Opt_notreelog, "notreelog"},
360 {Opt_treelog, "treelog"},
361 {Opt_flushoncommit, "flushoncommit"},
362 {Opt_noflushoncommit, "noflushoncommit"},
363 {Opt_ratio, "metadata_ratio=%d"},
364 {Opt_discard, "discard"},
365 {Opt_nodiscard, "nodiscard"},
366 {Opt_space_cache, "space_cache"},
367 {Opt_clear_cache, "clear_cache"},
368 {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
369 {Opt_enospc_debug, "enospc_debug"},
370 {Opt_noenospc_debug, "noenospc_debug"},
371 {Opt_subvolrootid, "subvolrootid=%d"},
372 {Opt_defrag, "autodefrag"},
373 {Opt_nodefrag, "noautodefrag"},
374 {Opt_inode_cache, "inode_cache"},
375 {Opt_noinode_cache, "noinode_cache"},
376 {Opt_no_space_cache, "nospace_cache"},
377 {Opt_recovery, "recovery"},
378 {Opt_skip_balance, "skip_balance"},
379 {Opt_check_integrity, "check_int"},
380 {Opt_check_integrity_including_extent_data, "check_int_data"},
381 {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
382 {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
383 {Opt_fatal_errors, "fatal_errors=%s"},
384 {Opt_commit_interval, "commit=%d"},
385 {Opt_err, NULL},
386};
387
388/*
389 * Regular mount options parser. Everything that is needed only when
390 * reading in a new superblock is parsed here.
391 * XXX JDM: This needs to be cleaned up for remount.
392 */
393int btrfs_parse_options(struct btrfs_root *root, char *options)
394{
395 struct btrfs_fs_info *info = root->fs_info;
396 substring_t args[MAX_OPT_ARGS];
397 char *p, *num, *orig = NULL;
398 u64 cache_gen;
399 int intarg;
400 int ret = 0;
401 char *compress_type;
402 bool compress_force = false;
403 bool compress = false;
404
405 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
406 if (cache_gen)
407 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
408
409 if (!options)
410 goto out;
411
412 /*
413 * strsep changes the string, duplicate it because parse_options
414 * gets called twice
415 */
416 options = kstrdup(options, GFP_NOFS);
417 if (!options)
418 return -ENOMEM;
419
420 orig = options;
421
422 while ((p = strsep(&options, ",")) != NULL) {
423 int token;
424 if (!*p)
425 continue;
426
427 token = match_token(p, tokens, args);
428 switch (token) {
429 case Opt_degraded:
430 btrfs_info(root->fs_info, "allowing degraded mounts");
431 btrfs_set_opt(info->mount_opt, DEGRADED);
432 break;
433 case Opt_subvol:
434 case Opt_subvolid:
435 case Opt_subvolrootid:
436 case Opt_device:
437 /*
438 * These are parsed by btrfs_parse_early_options
439 * and can be happily ignored here.
440 */
441 break;
442 case Opt_nodatasum:
443 btrfs_set_and_info(root, NODATASUM,
444 "setting nodatasum");
445 break;
446 case Opt_datasum:
447 if (btrfs_test_opt(root, NODATASUM)) {
448 if (btrfs_test_opt(root, NODATACOW))
449 btrfs_info(root->fs_info, "setting datasum, datacow enabled");
450 else
451 btrfs_info(root->fs_info, "setting datasum");
452 }
453 btrfs_clear_opt(info->mount_opt, NODATACOW);
454 btrfs_clear_opt(info->mount_opt, NODATASUM);
455 break;
456 case Opt_nodatacow:
457 if (!btrfs_test_opt(root, NODATACOW)) {
458 if (!btrfs_test_opt(root, COMPRESS) ||
459 !btrfs_test_opt(root, FORCE_COMPRESS)) {
460 btrfs_info(root->fs_info,
461 "setting nodatacow, compression disabled");
462 } else {
463 btrfs_info(root->fs_info, "setting nodatacow");
464 }
465 }
466 btrfs_clear_opt(info->mount_opt, COMPRESS);
467 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
468 btrfs_set_opt(info->mount_opt, NODATACOW);
469 btrfs_set_opt(info->mount_opt, NODATASUM);
470 break;
471 case Opt_datacow:
472 btrfs_clear_and_info(root, NODATACOW,
473 "setting datacow");
474 break;
475 case Opt_compress_force:
476 case Opt_compress_force_type:
477 compress_force = true;
478 /* Fallthrough */
479 case Opt_compress:
480 case Opt_compress_type:
481 compress = true;
482 if (token == Opt_compress ||
483 token == Opt_compress_force ||
484 strcmp(args[0].from, "zlib") == 0) {
485 compress_type = "zlib";
486 info->compress_type = BTRFS_COMPRESS_ZLIB;
487 btrfs_set_opt(info->mount_opt, COMPRESS);
488 btrfs_clear_opt(info->mount_opt, NODATACOW);
489 btrfs_clear_opt(info->mount_opt, NODATASUM);
490 } else if (strcmp(args[0].from, "lzo") == 0) {
491 compress_type = "lzo";
492 info->compress_type = BTRFS_COMPRESS_LZO;
493 btrfs_set_opt(info->mount_opt, COMPRESS);
494 btrfs_clear_opt(info->mount_opt, NODATACOW);
495 btrfs_clear_opt(info->mount_opt, NODATASUM);
496 btrfs_set_fs_incompat(info, COMPRESS_LZO);
497 } else if (strncmp(args[0].from, "no", 2) == 0) {
498 compress_type = "no";
499 btrfs_clear_opt(info->mount_opt, COMPRESS);
500 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
501 compress_force = false;
502 } else {
503 ret = -EINVAL;
504 goto out;
505 }
506
507 if (compress_force) {
508 btrfs_set_and_info(root, FORCE_COMPRESS,
509 "force %s compression",
510 compress_type);
511 } else if (compress) {
512 if (!btrfs_test_opt(root, COMPRESS))
513 btrfs_info(root->fs_info,
514 "btrfs: use %s compression\n",
515 compress_type);
516 }
517 break;
518 case Opt_ssd:
519 btrfs_set_and_info(root, SSD,
520 "use ssd allocation scheme");
521 break;
522 case Opt_ssd_spread:
523 btrfs_set_and_info(root, SSD_SPREAD,
524 "use spread ssd allocation scheme");
525 break;
526 case Opt_nossd:
527 btrfs_clear_and_info(root, NOSSD,
528 "not using ssd allocation scheme");
529 btrfs_clear_opt(info->mount_opt, SSD);
530 break;
531 case Opt_barrier:
532 btrfs_clear_and_info(root, NOBARRIER,
533 "turning on barriers");
534 break;
535 case Opt_nobarrier:
536 btrfs_set_and_info(root, NOBARRIER,
537 "turning off barriers");
538 break;
539 case Opt_thread_pool:
540 ret = match_int(&args[0], &intarg);
541 if (ret) {
542 goto out;
543 } else if (intarg > 0) {
544 info->thread_pool_size = intarg;
545 } else {
546 ret = -EINVAL;
547 goto out;
548 }
549 break;
550 case Opt_max_inline:
551 num = match_strdup(&args[0]);
552 if (num) {
553 info->max_inline = memparse(num, NULL);
554 kfree(num);
555
556 if (info->max_inline) {
557 info->max_inline = min_t(u64,
558 info->max_inline,
559 root->sectorsize);
560 }
561 btrfs_info(root->fs_info, "max_inline at %llu",
562 info->max_inline);
563 } else {
564 ret = -ENOMEM;
565 goto out;
566 }
567 break;
568 case Opt_alloc_start:
569 num = match_strdup(&args[0]);
570 if (num) {
571 mutex_lock(&info->chunk_mutex);
572 info->alloc_start = memparse(num, NULL);
573 mutex_unlock(&info->chunk_mutex);
574 kfree(num);
575 btrfs_info(root->fs_info, "allocations start at %llu",
576 info->alloc_start);
577 } else {
578 ret = -ENOMEM;
579 goto out;
580 }
581 break;
582 case Opt_acl:
583 root->fs_info->sb->s_flags |= MS_POSIXACL;
584 break;
585 case Opt_noacl:
586 root->fs_info->sb->s_flags &= ~MS_POSIXACL;
587 break;
588 case Opt_notreelog:
589 btrfs_set_and_info(root, NOTREELOG,
590 "disabling tree log");
591 break;
592 case Opt_treelog:
593 btrfs_clear_and_info(root, NOTREELOG,
594 "enabling tree log");
595 break;
596 case Opt_flushoncommit:
597 btrfs_set_and_info(root, FLUSHONCOMMIT,
598 "turning on flush-on-commit");
599 break;
600 case Opt_noflushoncommit:
601 btrfs_clear_and_info(root, FLUSHONCOMMIT,
602 "turning off flush-on-commit");
603 break;
604 case Opt_ratio:
605 ret = match_int(&args[0], &intarg);
606 if (ret) {
607 goto out;
608 } else if (intarg >= 0) {
609 info->metadata_ratio = intarg;
610 btrfs_info(root->fs_info, "metadata ratio %d",
611 info->metadata_ratio);
612 } else {
613 ret = -EINVAL;
614 goto out;
615 }
616 break;
617 case Opt_discard:
618 btrfs_set_and_info(root, DISCARD,
619 "turning on discard");
620 break;
621 case Opt_nodiscard:
622 btrfs_clear_and_info(root, DISCARD,
623 "turning off discard");
624 break;
625 case Opt_space_cache:
626 btrfs_set_and_info(root, SPACE_CACHE,
627 "enabling disk space caching");
628 break;
629 case Opt_rescan_uuid_tree:
630 btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
631 break;
632 case Opt_no_space_cache:
633 btrfs_clear_and_info(root, SPACE_CACHE,
634 "disabling disk space caching");
635 break;
636 case Opt_inode_cache:
637 btrfs_set_and_info(root, CHANGE_INODE_CACHE,
638 "enabling inode map caching");
639 break;
640 case Opt_noinode_cache:
641 btrfs_clear_and_info(root, CHANGE_INODE_CACHE,
642 "disabling inode map caching");
643 break;
644 case Opt_clear_cache:
645 btrfs_set_and_info(root, CLEAR_CACHE,
646 "force clearing of disk cache");
647 break;
648 case Opt_user_subvol_rm_allowed:
649 btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
650 break;
651 case Opt_enospc_debug:
652 btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
653 break;
654 case Opt_noenospc_debug:
655 btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
656 break;
657 case Opt_defrag:
658 btrfs_set_and_info(root, AUTO_DEFRAG,
659 "enabling auto defrag");
660 break;
661 case Opt_nodefrag:
662 btrfs_clear_and_info(root, AUTO_DEFRAG,
663 "disabling auto defrag");
664 break;
665 case Opt_recovery:
666 btrfs_info(root->fs_info, "enabling auto recovery");
667 btrfs_set_opt(info->mount_opt, RECOVERY);
668 break;
669 case Opt_skip_balance:
670 btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
671 break;
672#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
673 case Opt_check_integrity_including_extent_data:
674 btrfs_info(root->fs_info,
675 "enabling check integrity including extent data");
676 btrfs_set_opt(info->mount_opt,
677 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
678 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
679 break;
680 case Opt_check_integrity:
681 btrfs_info(root->fs_info, "enabling check integrity");
682 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
683 break;
684 case Opt_check_integrity_print_mask:
685 ret = match_int(&args[0], &intarg);
686 if (ret) {
687 goto out;
688 } else if (intarg >= 0) {
689 info->check_integrity_print_mask = intarg;
690 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
691 info->check_integrity_print_mask);
692 } else {
693 ret = -EINVAL;
694 goto out;
695 }
696 break;
697#else
698 case Opt_check_integrity_including_extent_data:
699 case Opt_check_integrity:
700 case Opt_check_integrity_print_mask:
701 btrfs_err(root->fs_info,
702 "support for check_integrity* not compiled in!");
703 ret = -EINVAL;
704 goto out;
705#endif
706 case Opt_fatal_errors:
707 if (strcmp(args[0].from, "panic") == 0)
708 btrfs_set_opt(info->mount_opt,
709 PANIC_ON_FATAL_ERROR);
710 else if (strcmp(args[0].from, "bug") == 0)
711 btrfs_clear_opt(info->mount_opt,
712 PANIC_ON_FATAL_ERROR);
713 else {
714 ret = -EINVAL;
715 goto out;
716 }
717 break;
718 case Opt_commit_interval:
719 intarg = 0;
720 ret = match_int(&args[0], &intarg);
721 if (ret < 0) {
722 btrfs_err(root->fs_info, "invalid commit interval");
723 ret = -EINVAL;
724 goto out;
725 }
726 if (intarg > 0) {
727 if (intarg > 300) {
728 btrfs_warn(root->fs_info, "excessive commit interval %d",
729 intarg);
730 }
731 info->commit_interval = intarg;
732 } else {
733 btrfs_info(root->fs_info, "using default commit interval %ds",
734 BTRFS_DEFAULT_COMMIT_INTERVAL);
735 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
736 }
737 break;
738 case Opt_err:
739 btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
740 ret = -EINVAL;
741 goto out;
742 default:
743 break;
744 }
745 }
746out:
747 if (!ret && btrfs_test_opt(root, SPACE_CACHE))
748 btrfs_info(root->fs_info, "disk space caching is enabled");
749 kfree(orig);
750 return ret;
751}
752
753/*
754 * Parse mount options that are required early in the mount process.
755 *
756 * All other options will be parsed on much later in the mount process and
757 * only when we need to allocate a new super block.
758 */
759static int btrfs_parse_early_options(const char *options, fmode_t flags,
760 void *holder, char **subvol_name, u64 *subvol_objectid,
761 struct btrfs_fs_devices **fs_devices)
762{
763 substring_t args[MAX_OPT_ARGS];
764 char *device_name, *opts, *orig, *p;
765 char *num = NULL;
766 int error = 0;
767
768 if (!options)
769 return 0;
770
771 /*
772 * strsep changes the string, duplicate it because parse_options
773 * gets called twice
774 */
775 opts = kstrdup(options, GFP_KERNEL);
776 if (!opts)
777 return -ENOMEM;
778 orig = opts;
779
780 while ((p = strsep(&opts, ",")) != NULL) {
781 int token;
782 if (!*p)
783 continue;
784
785 token = match_token(p, tokens, args);
786 switch (token) {
787 case Opt_subvol:
788 kfree(*subvol_name);
789 *subvol_name = match_strdup(&args[0]);
790 if (!*subvol_name) {
791 error = -ENOMEM;
792 goto out;
793 }
794 break;
795 case Opt_subvolid:
796 num = match_strdup(&args[0]);
797 if (num) {
798 *subvol_objectid = memparse(num, NULL);
799 kfree(num);
800 /* we want the original fs_tree */
801 if (!*subvol_objectid)
802 *subvol_objectid =
803 BTRFS_FS_TREE_OBJECTID;
804 } else {
805 error = -EINVAL;
806 goto out;
807 }
808 break;
809 case Opt_subvolrootid:
810 printk(KERN_WARNING
811 "BTRFS: 'subvolrootid' mount option is deprecated and has "
812 "no effect\n");
813 break;
814 case Opt_device:
815 device_name = match_strdup(&args[0]);
816 if (!device_name) {
817 error = -ENOMEM;
818 goto out;
819 }
820 error = btrfs_scan_one_device(device_name,
821 flags, holder, fs_devices);
822 kfree(device_name);
823 if (error)
824 goto out;
825 break;
826 default:
827 break;
828 }
829 }
830
831out:
832 kfree(orig);
833 return error;
834}
835
836static struct dentry *get_default_root(struct super_block *sb,
837 u64 subvol_objectid)
838{
839 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
840 struct btrfs_root *root = fs_info->tree_root;
841 struct btrfs_root *new_root;
842 struct btrfs_dir_item *di;
843 struct btrfs_path *path;
844 struct btrfs_key location;
845 struct inode *inode;
846 struct dentry *dentry;
847 u64 dir_id;
848 int new = 0;
849
850 /*
851 * We have a specific subvol we want to mount, just setup location and
852 * go look up the root.
853 */
854 if (subvol_objectid) {
855 location.objectid = subvol_objectid;
856 location.type = BTRFS_ROOT_ITEM_KEY;
857 location.offset = (u64)-1;
858 goto find_root;
859 }
860
861 path = btrfs_alloc_path();
862 if (!path)
863 return ERR_PTR(-ENOMEM);
864 path->leave_spinning = 1;
865
866 /*
867 * Find the "default" dir item which points to the root item that we
868 * will mount by default if we haven't been given a specific subvolume
869 * to mount.
870 */
871 dir_id = btrfs_super_root_dir(fs_info->super_copy);
872 di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
873 if (IS_ERR(di)) {
874 btrfs_free_path(path);
875 return ERR_CAST(di);
876 }
877 if (!di) {
878 /*
879 * Ok the default dir item isn't there. This is weird since
880 * it's always been there, but don't freak out, just try and
881 * mount to root most subvolume.
882 */
883 btrfs_free_path(path);
884 dir_id = BTRFS_FIRST_FREE_OBJECTID;
885 new_root = fs_info->fs_root;
886 goto setup_root;
887 }
888
889 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
890 btrfs_free_path(path);
891
892find_root:
893 new_root = btrfs_read_fs_root_no_name(fs_info, &location);
894 if (IS_ERR(new_root))
895 return ERR_CAST(new_root);
896
897 dir_id = btrfs_root_dirid(&new_root->root_item);
898setup_root:
899 location.objectid = dir_id;
900 location.type = BTRFS_INODE_ITEM_KEY;
901 location.offset = 0;
902
903 inode = btrfs_iget(sb, &location, new_root, &new);
904 if (IS_ERR(inode))
905 return ERR_CAST(inode);
906
907 /*
908 * If we're just mounting the root most subvol put the inode and return
909 * a reference to the dentry. We will have already gotten a reference
910 * to the inode in btrfs_fill_super so we're good to go.
911 */
912 if (!new && sb->s_root->d_inode == inode) {
913 iput(inode);
914 return dget(sb->s_root);
915 }
916
917 dentry = d_obtain_alias(inode);
918 if (!IS_ERR(dentry)) {
919 spin_lock(&dentry->d_lock);
920 dentry->d_flags &= ~DCACHE_DISCONNECTED;
921 spin_unlock(&dentry->d_lock);
922 }
923 return dentry;
924}
925
926static int btrfs_fill_super(struct super_block *sb,
927 struct btrfs_fs_devices *fs_devices,
928 void *data, int silent)
929{
930 struct inode *inode;
931 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
932 struct btrfs_key key;
933 int err;
934
935 sb->s_maxbytes = MAX_LFS_FILESIZE;
936 sb->s_magic = BTRFS_SUPER_MAGIC;
937 sb->s_op = &btrfs_super_ops;
938 sb->s_d_op = &btrfs_dentry_operations;
939 sb->s_export_op = &btrfs_export_ops;
940 sb->s_xattr = btrfs_xattr_handlers;
941 sb->s_time_gran = 1;
942#ifdef CONFIG_BTRFS_FS_POSIX_ACL
943 sb->s_flags |= MS_POSIXACL;
944#endif
945 sb->s_flags |= MS_I_VERSION;
946 err = open_ctree(sb, fs_devices, (char *)data);
947 if (err) {
948 printk(KERN_ERR "BTRFS: open_ctree failed\n");
949 return err;
950 }
951
952 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
953 key.type = BTRFS_INODE_ITEM_KEY;
954 key.offset = 0;
955 inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
956 if (IS_ERR(inode)) {
957 err = PTR_ERR(inode);
958 goto fail_close;
959 }
960
961 sb->s_root = d_make_root(inode);
962 if (!sb->s_root) {
963 err = -ENOMEM;
964 goto fail_close;
965 }
966
967 save_mount_options(sb, data);
968 cleancache_init_fs(sb);
969 sb->s_flags |= MS_ACTIVE;
970 return 0;
971
972fail_close:
973 close_ctree(fs_info->tree_root);
974 return err;
975}
976
977int btrfs_sync_fs(struct super_block *sb, int wait)
978{
979 struct btrfs_trans_handle *trans;
980 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
981 struct btrfs_root *root = fs_info->tree_root;
982
983 trace_btrfs_sync_fs(wait);
984
985 if (!wait) {
986 filemap_flush(fs_info->btree_inode->i_mapping);
987 return 0;
988 }
989
990 btrfs_wait_ordered_roots(fs_info, -1);
991
992 trans = btrfs_attach_transaction_barrier(root);
993 if (IS_ERR(trans)) {
994 /* no transaction, don't bother */
995 if (PTR_ERR(trans) == -ENOENT)
996 return 0;
997 return PTR_ERR(trans);
998 }
999 return btrfs_commit_transaction(trans, root);
1000}
1001
1002static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1003{
1004 struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1005 struct btrfs_root *root = info->tree_root;
1006 char *compress_type;
1007
1008 if (btrfs_test_opt(root, DEGRADED))
1009 seq_puts(seq, ",degraded");
1010 if (btrfs_test_opt(root, NODATASUM))
1011 seq_puts(seq, ",nodatasum");
1012 if (btrfs_test_opt(root, NODATACOW))
1013 seq_puts(seq, ",nodatacow");
1014 if (btrfs_test_opt(root, NOBARRIER))
1015 seq_puts(seq, ",nobarrier");
1016 if (info->max_inline != 8192 * 1024)
1017 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1018 if (info->alloc_start != 0)
1019 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1020 if (info->thread_pool_size != min_t(unsigned long,
1021 num_online_cpus() + 2, 8))
1022 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1023 if (btrfs_test_opt(root, COMPRESS)) {
1024 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1025 compress_type = "zlib";
1026 else
1027 compress_type = "lzo";
1028 if (btrfs_test_opt(root, FORCE_COMPRESS))
1029 seq_printf(seq, ",compress-force=%s", compress_type);
1030 else
1031 seq_printf(seq, ",compress=%s", compress_type);
1032 }
1033 if (btrfs_test_opt(root, NOSSD))
1034 seq_puts(seq, ",nossd");
1035 if (btrfs_test_opt(root, SSD_SPREAD))
1036 seq_puts(seq, ",ssd_spread");
1037 else if (btrfs_test_opt(root, SSD))
1038 seq_puts(seq, ",ssd");
1039 if (btrfs_test_opt(root, NOTREELOG))
1040 seq_puts(seq, ",notreelog");
1041 if (btrfs_test_opt(root, FLUSHONCOMMIT))
1042 seq_puts(seq, ",flushoncommit");
1043 if (btrfs_test_opt(root, DISCARD))
1044 seq_puts(seq, ",discard");
1045 if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1046 seq_puts(seq, ",noacl");
1047 if (btrfs_test_opt(root, SPACE_CACHE))
1048 seq_puts(seq, ",space_cache");
1049 else
1050 seq_puts(seq, ",nospace_cache");
1051 if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1052 seq_puts(seq, ",rescan_uuid_tree");
1053 if (btrfs_test_opt(root, CLEAR_CACHE))
1054 seq_puts(seq, ",clear_cache");
1055 if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1056 seq_puts(seq, ",user_subvol_rm_allowed");
1057 if (btrfs_test_opt(root, ENOSPC_DEBUG))
1058 seq_puts(seq, ",enospc_debug");
1059 if (btrfs_test_opt(root, AUTO_DEFRAG))
1060 seq_puts(seq, ",autodefrag");
1061 if (btrfs_test_opt(root, INODE_MAP_CACHE))
1062 seq_puts(seq, ",inode_cache");
1063 if (btrfs_test_opt(root, SKIP_BALANCE))
1064 seq_puts(seq, ",skip_balance");
1065 if (btrfs_test_opt(root, RECOVERY))
1066 seq_puts(seq, ",recovery");
1067#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1068 if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1069 seq_puts(seq, ",check_int_data");
1070 else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1071 seq_puts(seq, ",check_int");
1072 if (info->check_integrity_print_mask)
1073 seq_printf(seq, ",check_int_print_mask=%d",
1074 info->check_integrity_print_mask);
1075#endif
1076 if (info->metadata_ratio)
1077 seq_printf(seq, ",metadata_ratio=%d",
1078 info->metadata_ratio);
1079 if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1080 seq_puts(seq, ",fatal_errors=panic");
1081 if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1082 seq_printf(seq, ",commit=%d", info->commit_interval);
1083 return 0;
1084}
1085
1086static int btrfs_test_super(struct super_block *s, void *data)
1087{
1088 struct btrfs_fs_info *p = data;
1089 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1090
1091 return fs_info->fs_devices == p->fs_devices;
1092}
1093
1094static int btrfs_set_super(struct super_block *s, void *data)
1095{
1096 int err = set_anon_super(s, data);
1097 if (!err)
1098 s->s_fs_info = data;
1099 return err;
1100}
1101
1102/*
1103 * subvolumes are identified by ino 256
1104 */
1105static inline int is_subvolume_inode(struct inode *inode)
1106{
1107 if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1108 return 1;
1109 return 0;
1110}
1111
1112/*
1113 * This will strip out the subvol=%s argument for an argument string and add
1114 * subvolid=0 to make sure we get the actual tree root for path walking to the
1115 * subvol we want.
1116 */
1117static char *setup_root_args(char *args)
1118{
1119 unsigned len = strlen(args) + 2 + 1;
1120 char *src, *dst, *buf;
1121
1122 /*
1123 * We need the same args as before, but with this substitution:
1124 * s!subvol=[^,]+!subvolid=0!
1125 *
1126 * Since the replacement string is up to 2 bytes longer than the
1127 * original, allocate strlen(args) + 2 + 1 bytes.
1128 */
1129
1130 src = strstr(args, "subvol=");
1131 /* This shouldn't happen, but just in case.. */
1132 if (!src)
1133 return NULL;
1134
1135 buf = dst = kmalloc(len, GFP_NOFS);
1136 if (!buf)
1137 return NULL;
1138
1139 /*
1140 * If the subvol= arg is not at the start of the string,
1141 * copy whatever precedes it into buf.
1142 */
1143 if (src != args) {
1144 *src++ = '\0';
1145 strcpy(buf, args);
1146 dst += strlen(args);
1147 }
1148
1149 strcpy(dst, "subvolid=0");
1150 dst += strlen("subvolid=0");
1151
1152 /*
1153 * If there is a "," after the original subvol=... string,
1154 * copy that suffix into our buffer. Otherwise, we're done.
1155 */
1156 src = strchr(src, ',');
1157 if (src)
1158 strcpy(dst, src);
1159
1160 return buf;
1161}
1162
1163static struct dentry *mount_subvol(const char *subvol_name, int flags,
1164 const char *device_name, char *data)
1165{
1166 struct dentry *root;
1167 struct vfsmount *mnt;
1168 char *newargs;
1169
1170 newargs = setup_root_args(data);
1171 if (!newargs)
1172 return ERR_PTR(-ENOMEM);
1173 mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1174 newargs);
1175
1176 if (PTR_RET(mnt) == -EBUSY) {
1177 if (flags & MS_RDONLY) {
1178 mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY, device_name,
1179 newargs);
1180 } else {
1181 int r;
1182 mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY, device_name,
1183 newargs);
1184 if (IS_ERR(mnt)) {
1185 kfree(newargs);
1186 return ERR_CAST(mnt);
1187 }
1188
1189 r = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1190 if (r < 0) {
1191 /* FIXME: release vfsmount mnt ??*/
1192 kfree(newargs);
1193 return ERR_PTR(r);
1194 }
1195 }
1196 }
1197
1198 kfree(newargs);
1199
1200 if (IS_ERR(mnt))
1201 return ERR_CAST(mnt);
1202
1203 root = mount_subtree(mnt, subvol_name);
1204
1205 if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1206 struct super_block *s = root->d_sb;
1207 dput(root);
1208 root = ERR_PTR(-EINVAL);
1209 deactivate_locked_super(s);
1210 printk(KERN_ERR "BTRFS: '%s' is not a valid subvolume\n",
1211 subvol_name);
1212 }
1213
1214 return root;
1215}
1216
1217/*
1218 * Find a superblock for the given device / mount point.
1219 *
1220 * Note: This is based on get_sb_bdev from fs/super.c with a few additions
1221 * for multiple device setup. Make sure to keep it in sync.
1222 */
1223static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1224 const char *device_name, void *data)
1225{
1226 struct block_device *bdev = NULL;
1227 struct super_block *s;
1228 struct dentry *root;
1229 struct btrfs_fs_devices *fs_devices = NULL;
1230 struct btrfs_fs_info *fs_info = NULL;
1231 fmode_t mode = FMODE_READ;
1232 char *subvol_name = NULL;
1233 u64 subvol_objectid = 0;
1234 int error = 0;
1235
1236 if (!(flags & MS_RDONLY))
1237 mode |= FMODE_WRITE;
1238
1239 error = btrfs_parse_early_options(data, mode, fs_type,
1240 &subvol_name, &subvol_objectid,
1241 &fs_devices);
1242 if (error) {
1243 kfree(subvol_name);
1244 return ERR_PTR(error);
1245 }
1246
1247 if (subvol_name) {
1248 root = mount_subvol(subvol_name, flags, device_name, data);
1249 kfree(subvol_name);
1250 return root;
1251 }
1252
1253 error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1254 if (error)
1255 return ERR_PTR(error);
1256
1257 /*
1258 * Setup a dummy root and fs_info for test/set super. This is because
1259 * we don't actually fill this stuff out until open_ctree, but we need
1260 * it for searching for existing supers, so this lets us do that and
1261 * then open_ctree will properly initialize everything later.
1262 */
1263 fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1264 if (!fs_info)
1265 return ERR_PTR(-ENOMEM);
1266
1267 fs_info->fs_devices = fs_devices;
1268
1269 fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1270 fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1271 if (!fs_info->super_copy || !fs_info->super_for_commit) {
1272 error = -ENOMEM;
1273 goto error_fs_info;
1274 }
1275
1276 error = btrfs_open_devices(fs_devices, mode, fs_type);
1277 if (error)
1278 goto error_fs_info;
1279
1280 if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1281 error = -EACCES;
1282 goto error_close_devices;
1283 }
1284
1285 bdev = fs_devices->latest_bdev;
1286 s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1287 fs_info);
1288 if (IS_ERR(s)) {
1289 error = PTR_ERR(s);
1290 goto error_close_devices;
1291 }
1292
1293 if (s->s_root) {
1294 btrfs_close_devices(fs_devices);
1295 free_fs_info(fs_info);
1296 if ((flags ^ s->s_flags) & MS_RDONLY)
1297 error = -EBUSY;
1298 } else {
1299 char b[BDEVNAME_SIZE];
1300
1301 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1302 btrfs_sb(s)->bdev_holder = fs_type;
1303 error = btrfs_fill_super(s, fs_devices, data,
1304 flags & MS_SILENT ? 1 : 0);
1305 }
1306
1307 root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1308 if (IS_ERR(root))
1309 deactivate_locked_super(s);
1310
1311 return root;
1312
1313error_close_devices:
1314 btrfs_close_devices(fs_devices);
1315error_fs_info:
1316 free_fs_info(fs_info);
1317 return ERR_PTR(error);
1318}
1319
1320static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1321 int new_pool_size, int old_pool_size)
1322{
1323 if (new_pool_size == old_pool_size)
1324 return;
1325
1326 fs_info->thread_pool_size = new_pool_size;
1327
1328 btrfs_info(fs_info, "resize thread pool %d -> %d",
1329 old_pool_size, new_pool_size);
1330
1331 btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1332 btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1333 btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1334 btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1335 btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1336 btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1337 btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1338 new_pool_size);
1339 btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1340 btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1341 btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1342 btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1343 btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1344 new_pool_size);
1345}
1346
1347static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1348{
1349 set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1350}
1351
1352static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1353 unsigned long old_opts, int flags)
1354{
1355 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1356 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1357 (flags & MS_RDONLY))) {
1358 /* wait for any defraggers to finish */
1359 wait_event(fs_info->transaction_wait,
1360 (atomic_read(&fs_info->defrag_running) == 0));
1361 if (flags & MS_RDONLY)
1362 sync_filesystem(fs_info->sb);
1363 }
1364}
1365
1366static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1367 unsigned long old_opts)
1368{
1369 /*
1370 * We need cleanup all defragable inodes if the autodefragment is
1371 * close or the fs is R/O.
1372 */
1373 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1374 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1375 (fs_info->sb->s_flags & MS_RDONLY))) {
1376 btrfs_cleanup_defrag_inodes(fs_info);
1377 }
1378
1379 clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1380}
1381
1382static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1383{
1384 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1385 struct btrfs_root *root = fs_info->tree_root;
1386 unsigned old_flags = sb->s_flags;
1387 unsigned long old_opts = fs_info->mount_opt;
1388 unsigned long old_compress_type = fs_info->compress_type;
1389 u64 old_max_inline = fs_info->max_inline;
1390 u64 old_alloc_start = fs_info->alloc_start;
1391 int old_thread_pool_size = fs_info->thread_pool_size;
1392 unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1393 int ret;
1394
1395 sync_filesystem(sb);
1396 btrfs_remount_prepare(fs_info);
1397
1398 ret = btrfs_parse_options(root, data);
1399 if (ret) {
1400 ret = -EINVAL;
1401 goto restore;
1402 }
1403
1404 btrfs_remount_begin(fs_info, old_opts, *flags);
1405 btrfs_resize_thread_pool(fs_info,
1406 fs_info->thread_pool_size, old_thread_pool_size);
1407
1408 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1409 goto out;
1410
1411 if (*flags & MS_RDONLY) {
1412 /*
1413 * this also happens on 'umount -rf' or on shutdown, when
1414 * the filesystem is busy.
1415 */
1416
1417 /* wait for the uuid_scan task to finish */
1418 down(&fs_info->uuid_tree_rescan_sem);
1419 /* avoid complains from lockdep et al. */
1420 up(&fs_info->uuid_tree_rescan_sem);
1421
1422 sb->s_flags |= MS_RDONLY;
1423
1424 btrfs_dev_replace_suspend_for_unmount(fs_info);
1425 btrfs_scrub_cancel(fs_info);
1426 btrfs_pause_balance(fs_info);
1427
1428 ret = btrfs_commit_super(root);
1429 if (ret)
1430 goto restore;
1431 } else {
1432 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1433 btrfs_err(fs_info,
1434 "Remounting read-write after error is not allowed");
1435 ret = -EINVAL;
1436 goto restore;
1437 }
1438 if (fs_info->fs_devices->rw_devices == 0) {
1439 ret = -EACCES;
1440 goto restore;
1441 }
1442
1443 if (fs_info->fs_devices->missing_devices >
1444 fs_info->num_tolerated_disk_barrier_failures &&
1445 !(*flags & MS_RDONLY)) {
1446 btrfs_warn(fs_info,
1447 "too many missing devices, writeable remount is not allowed");
1448 ret = -EACCES;
1449 goto restore;
1450 }
1451
1452 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1453 ret = -EINVAL;
1454 goto restore;
1455 }
1456
1457 ret = btrfs_cleanup_fs_roots(fs_info);
1458 if (ret)
1459 goto restore;
1460
1461 /* recover relocation */
1462 ret = btrfs_recover_relocation(root);
1463 if (ret)
1464 goto restore;
1465
1466 ret = btrfs_resume_balance_async(fs_info);
1467 if (ret)
1468 goto restore;
1469
1470 ret = btrfs_resume_dev_replace_async(fs_info);
1471 if (ret) {
1472 btrfs_warn(fs_info, "failed to resume dev_replace");
1473 goto restore;
1474 }
1475
1476 if (!fs_info->uuid_root) {
1477 btrfs_info(fs_info, "creating UUID tree");
1478 ret = btrfs_create_uuid_tree(fs_info);
1479 if (ret) {
1480 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1481 goto restore;
1482 }
1483 }
1484 sb->s_flags &= ~MS_RDONLY;
1485 }
1486out:
1487 wake_up_process(fs_info->transaction_kthread);
1488 btrfs_remount_cleanup(fs_info, old_opts);
1489 return 0;
1490
1491restore:
1492 /* We've hit an error - don't reset MS_RDONLY */
1493 if (sb->s_flags & MS_RDONLY)
1494 old_flags |= MS_RDONLY;
1495 sb->s_flags = old_flags;
1496 fs_info->mount_opt = old_opts;
1497 fs_info->compress_type = old_compress_type;
1498 fs_info->max_inline = old_max_inline;
1499 mutex_lock(&fs_info->chunk_mutex);
1500 fs_info->alloc_start = old_alloc_start;
1501 mutex_unlock(&fs_info->chunk_mutex);
1502 btrfs_resize_thread_pool(fs_info,
1503 old_thread_pool_size, fs_info->thread_pool_size);
1504 fs_info->metadata_ratio = old_metadata_ratio;
1505 btrfs_remount_cleanup(fs_info, old_opts);
1506 return ret;
1507}
1508
1509/* Used to sort the devices by max_avail(descending sort) */
1510static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1511 const void *dev_info2)
1512{
1513 if (((struct btrfs_device_info *)dev_info1)->max_avail >
1514 ((struct btrfs_device_info *)dev_info2)->max_avail)
1515 return -1;
1516 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1517 ((struct btrfs_device_info *)dev_info2)->max_avail)
1518 return 1;
1519 else
1520 return 0;
1521}
1522
1523/*
1524 * sort the devices by max_avail, in which max free extent size of each device
1525 * is stored.(Descending Sort)
1526 */
1527static inline void btrfs_descending_sort_devices(
1528 struct btrfs_device_info *devices,
1529 size_t nr_devices)
1530{
1531 sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1532 btrfs_cmp_device_free_bytes, NULL);
1533}
1534
1535/*
1536 * The helper to calc the free space on the devices that can be used to store
1537 * file data.
1538 */
1539static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1540{
1541 struct btrfs_fs_info *fs_info = root->fs_info;
1542 struct btrfs_device_info *devices_info;
1543 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1544 struct btrfs_device *device;
1545 u64 skip_space;
1546 u64 type;
1547 u64 avail_space;
1548 u64 used_space;
1549 u64 min_stripe_size;
1550 int min_stripes = 1, num_stripes = 1;
1551 int i = 0, nr_devices;
1552 int ret;
1553
1554 nr_devices = fs_info->fs_devices->open_devices;
1555 BUG_ON(!nr_devices);
1556
1557 devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1558 GFP_NOFS);
1559 if (!devices_info)
1560 return -ENOMEM;
1561
1562 /* calc min stripe number for data space alloction */
1563 type = btrfs_get_alloc_profile(root, 1);
1564 if (type & BTRFS_BLOCK_GROUP_RAID0) {
1565 min_stripes = 2;
1566 num_stripes = nr_devices;
1567 } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1568 min_stripes = 2;
1569 num_stripes = 2;
1570 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1571 min_stripes = 4;
1572 num_stripes = 4;
1573 }
1574
1575 if (type & BTRFS_BLOCK_GROUP_DUP)
1576 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1577 else
1578 min_stripe_size = BTRFS_STRIPE_LEN;
1579
1580 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1581 if (!device->in_fs_metadata || !device->bdev ||
1582 device->is_tgtdev_for_dev_replace)
1583 continue;
1584
1585 avail_space = device->total_bytes - device->bytes_used;
1586
1587 /* align with stripe_len */
1588 do_div(avail_space, BTRFS_STRIPE_LEN);
1589 avail_space *= BTRFS_STRIPE_LEN;
1590
1591 /*
1592 * In order to avoid overwritting the superblock on the drive,
1593 * btrfs starts at an offset of at least 1MB when doing chunk
1594 * allocation.
1595 */
1596 skip_space = 1024 * 1024;
1597
1598 /* user can set the offset in fs_info->alloc_start. */
1599 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1600 device->total_bytes)
1601 skip_space = max(fs_info->alloc_start, skip_space);
1602
1603 /*
1604 * btrfs can not use the free space in [0, skip_space - 1],
1605 * we must subtract it from the total. In order to implement
1606 * it, we account the used space in this range first.
1607 */
1608 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1609 &used_space);
1610 if (ret) {
1611 kfree(devices_info);
1612 return ret;
1613 }
1614
1615 /* calc the free space in [0, skip_space - 1] */
1616 skip_space -= used_space;
1617
1618 /*
1619 * we can use the free space in [0, skip_space - 1], subtract
1620 * it from the total.
1621 */
1622 if (avail_space && avail_space >= skip_space)
1623 avail_space -= skip_space;
1624 else
1625 avail_space = 0;
1626
1627 if (avail_space < min_stripe_size)
1628 continue;
1629
1630 devices_info[i].dev = device;
1631 devices_info[i].max_avail = avail_space;
1632
1633 i++;
1634 }
1635
1636 nr_devices = i;
1637
1638 btrfs_descending_sort_devices(devices_info, nr_devices);
1639
1640 i = nr_devices - 1;
1641 avail_space = 0;
1642 while (nr_devices >= min_stripes) {
1643 if (num_stripes > nr_devices)
1644 num_stripes = nr_devices;
1645
1646 if (devices_info[i].max_avail >= min_stripe_size) {
1647 int j;
1648 u64 alloc_size;
1649
1650 avail_space += devices_info[i].max_avail * num_stripes;
1651 alloc_size = devices_info[i].max_avail;
1652 for (j = i + 1 - num_stripes; j <= i; j++)
1653 devices_info[j].max_avail -= alloc_size;
1654 }
1655 i--;
1656 nr_devices--;
1657 }
1658
1659 kfree(devices_info);
1660 *free_bytes = avail_space;
1661 return 0;
1662}
1663
1664static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1665{
1666 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1667 struct btrfs_super_block *disk_super = fs_info->super_copy;
1668 struct list_head *head = &fs_info->space_info;
1669 struct btrfs_space_info *found;
1670 u64 total_used = 0;
1671 u64 total_free_data = 0;
1672 int bits = dentry->d_sb->s_blocksize_bits;
1673 __be32 *fsid = (__be32 *)fs_info->fsid;
1674 int ret;
1675
1676 /* holding chunk_muext to avoid allocating new chunks */
1677 mutex_lock(&fs_info->chunk_mutex);
1678 rcu_read_lock();
1679 list_for_each_entry_rcu(found, head, list) {
1680 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1681 total_free_data += found->disk_total - found->disk_used;
1682 total_free_data -=
1683 btrfs_account_ro_block_groups_free_space(found);
1684 }
1685
1686 total_used += found->disk_used;
1687 }
1688 rcu_read_unlock();
1689
1690 buf->f_namelen = BTRFS_NAME_LEN;
1691 buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1692 buf->f_bfree = buf->f_blocks - (total_used >> bits);
1693 buf->f_bsize = dentry->d_sb->s_blocksize;
1694 buf->f_type = BTRFS_SUPER_MAGIC;
1695 buf->f_bavail = total_free_data;
1696 ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1697 if (ret) {
1698 mutex_unlock(&fs_info->chunk_mutex);
1699 return ret;
1700 }
1701 buf->f_bavail += total_free_data;
1702 buf->f_bavail = buf->f_bavail >> bits;
1703 mutex_unlock(&fs_info->chunk_mutex);
1704
1705 /* We treat it as constant endianness (it doesn't matter _which_)
1706 because we want the fsid to come out the same whether mounted
1707 on a big-endian or little-endian host */
1708 buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1709 buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1710 /* Mask in the root object ID too, to disambiguate subvols */
1711 buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1712 buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1713
1714 return 0;
1715}
1716
1717static void btrfs_kill_super(struct super_block *sb)
1718{
1719 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1720 kill_anon_super(sb);
1721 free_fs_info(fs_info);
1722}
1723
1724static struct file_system_type btrfs_fs_type = {
1725 .owner = THIS_MODULE,
1726 .name = "btrfs",
1727 .mount = btrfs_mount,
1728 .kill_sb = btrfs_kill_super,
1729 .fs_flags = FS_REQUIRES_DEV,
1730};
1731MODULE_ALIAS_FS("btrfs");
1732
1733/*
1734 * used by btrfsctl to scan devices when no FS is mounted
1735 */
1736static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1737 unsigned long arg)
1738{
1739 struct btrfs_ioctl_vol_args *vol;
1740 struct btrfs_fs_devices *fs_devices;
1741 int ret = -ENOTTY;
1742
1743 if (!capable(CAP_SYS_ADMIN))
1744 return -EPERM;
1745
1746 vol = memdup_user((void __user *)arg, sizeof(*vol));
1747 if (IS_ERR(vol))
1748 return PTR_ERR(vol);
1749
1750 switch (cmd) {
1751 case BTRFS_IOC_SCAN_DEV:
1752 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1753 &btrfs_fs_type, &fs_devices);
1754 break;
1755 case BTRFS_IOC_DEVICES_READY:
1756 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1757 &btrfs_fs_type, &fs_devices);
1758 if (ret)
1759 break;
1760 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1761 break;
1762 }
1763
1764 kfree(vol);
1765 return ret;
1766}
1767
1768static int btrfs_freeze(struct super_block *sb)
1769{
1770 struct btrfs_trans_handle *trans;
1771 struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1772
1773 trans = btrfs_attach_transaction_barrier(root);
1774 if (IS_ERR(trans)) {
1775 /* no transaction, don't bother */
1776 if (PTR_ERR(trans) == -ENOENT)
1777 return 0;
1778 return PTR_ERR(trans);
1779 }
1780 return btrfs_commit_transaction(trans, root);
1781}
1782
1783static int btrfs_unfreeze(struct super_block *sb)
1784{
1785 return 0;
1786}
1787
1788static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1789{
1790 struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1791 struct btrfs_fs_devices *cur_devices;
1792 struct btrfs_device *dev, *first_dev = NULL;
1793 struct list_head *head;
1794 struct rcu_string *name;
1795
1796 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1797 cur_devices = fs_info->fs_devices;
1798 while (cur_devices) {
1799 head = &cur_devices->devices;
1800 list_for_each_entry(dev, head, dev_list) {
1801 if (dev->missing)
1802 continue;
1803 if (!first_dev || dev->devid < first_dev->devid)
1804 first_dev = dev;
1805 }
1806 cur_devices = cur_devices->seed;
1807 }
1808
1809 if (first_dev) {
1810 rcu_read_lock();
1811 name = rcu_dereference(first_dev->name);
1812 seq_escape(m, name->str, " \t\n\\");
1813 rcu_read_unlock();
1814 } else {
1815 WARN_ON(1);
1816 }
1817 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1818 return 0;
1819}
1820
1821static const struct super_operations btrfs_super_ops = {
1822 .drop_inode = btrfs_drop_inode,
1823 .evict_inode = btrfs_evict_inode,
1824 .put_super = btrfs_put_super,
1825 .sync_fs = btrfs_sync_fs,
1826 .show_options = btrfs_show_options,
1827 .show_devname = btrfs_show_devname,
1828 .write_inode = btrfs_write_inode,
1829 .alloc_inode = btrfs_alloc_inode,
1830 .destroy_inode = btrfs_destroy_inode,
1831 .statfs = btrfs_statfs,
1832 .remount_fs = btrfs_remount,
1833 .freeze_fs = btrfs_freeze,
1834 .unfreeze_fs = btrfs_unfreeze,
1835};
1836
1837static const struct file_operations btrfs_ctl_fops = {
1838 .unlocked_ioctl = btrfs_control_ioctl,
1839 .compat_ioctl = btrfs_control_ioctl,
1840 .owner = THIS_MODULE,
1841 .llseek = noop_llseek,
1842};
1843
1844static struct miscdevice btrfs_misc = {
1845 .minor = BTRFS_MINOR,
1846 .name = "btrfs-control",
1847 .fops = &btrfs_ctl_fops
1848};
1849
1850MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1851MODULE_ALIAS("devname:btrfs-control");
1852
1853static int btrfs_interface_init(void)
1854{
1855 return misc_register(&btrfs_misc);
1856}
1857
1858static void btrfs_interface_exit(void)
1859{
1860 if (misc_deregister(&btrfs_misc) < 0)
1861 printk(KERN_INFO "BTRFS: misc_deregister failed for control device\n");
1862}
1863
1864static void btrfs_print_info(void)
1865{
1866 printk(KERN_INFO "Btrfs loaded"
1867#ifdef CONFIG_BTRFS_DEBUG
1868 ", debug=on"
1869#endif
1870#ifdef CONFIG_BTRFS_ASSERT
1871 ", assert=on"
1872#endif
1873#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1874 ", integrity-checker=on"
1875#endif
1876 "\n");
1877}
1878
1879static int btrfs_run_sanity_tests(void)
1880{
1881 int ret;
1882
1883 ret = btrfs_init_test_fs();
1884 if (ret)
1885 return ret;
1886
1887 ret = btrfs_test_free_space_cache();
1888 if (ret)
1889 goto out;
1890 ret = btrfs_test_extent_buffer_operations();
1891 if (ret)
1892 goto out;
1893 ret = btrfs_test_extent_io();
1894 if (ret)
1895 goto out;
1896 ret = btrfs_test_inodes();
1897out:
1898 btrfs_destroy_test_fs();
1899 return ret;
1900}
1901
1902static int __init init_btrfs_fs(void)
1903{
1904 int err;
1905
1906 err = btrfs_hash_init();
1907 if (err)
1908 return err;
1909
1910 btrfs_props_init();
1911
1912 err = btrfs_init_sysfs();
1913 if (err)
1914 goto free_hash;
1915
1916 btrfs_init_compress();
1917
1918 err = btrfs_init_cachep();
1919 if (err)
1920 goto free_compress;
1921
1922 err = extent_io_init();
1923 if (err)
1924 goto free_cachep;
1925
1926 err = extent_map_init();
1927 if (err)
1928 goto free_extent_io;
1929
1930 err = ordered_data_init();
1931 if (err)
1932 goto free_extent_map;
1933
1934 err = btrfs_delayed_inode_init();
1935 if (err)
1936 goto free_ordered_data;
1937
1938 err = btrfs_auto_defrag_init();
1939 if (err)
1940 goto free_delayed_inode;
1941
1942 err = btrfs_delayed_ref_init();
1943 if (err)
1944 goto free_auto_defrag;
1945
1946 err = btrfs_prelim_ref_init();
1947 if (err)
1948 goto free_prelim_ref;
1949
1950 err = btrfs_interface_init();
1951 if (err)
1952 goto free_delayed_ref;
1953
1954 btrfs_init_lockdep();
1955
1956 btrfs_print_info();
1957
1958 err = btrfs_run_sanity_tests();
1959 if (err)
1960 goto unregister_ioctl;
1961
1962 err = register_filesystem(&btrfs_fs_type);
1963 if (err)
1964 goto unregister_ioctl;
1965
1966 return 0;
1967
1968unregister_ioctl:
1969 btrfs_interface_exit();
1970free_prelim_ref:
1971 btrfs_prelim_ref_exit();
1972free_delayed_ref:
1973 btrfs_delayed_ref_exit();
1974free_auto_defrag:
1975 btrfs_auto_defrag_exit();
1976free_delayed_inode:
1977 btrfs_delayed_inode_exit();
1978free_ordered_data:
1979 ordered_data_exit();
1980free_extent_map:
1981 extent_map_exit();
1982free_extent_io:
1983 extent_io_exit();
1984free_cachep:
1985 btrfs_destroy_cachep();
1986free_compress:
1987 btrfs_exit_compress();
1988 btrfs_exit_sysfs();
1989free_hash:
1990 btrfs_hash_exit();
1991 return err;
1992}
1993
1994static void __exit exit_btrfs_fs(void)
1995{
1996 btrfs_destroy_cachep();
1997 btrfs_delayed_ref_exit();
1998 btrfs_auto_defrag_exit();
1999 btrfs_delayed_inode_exit();
2000 btrfs_prelim_ref_exit();
2001 ordered_data_exit();
2002 extent_map_exit();
2003 extent_io_exit();
2004 btrfs_interface_exit();
2005 unregister_filesystem(&btrfs_fs_type);
2006 btrfs_exit_sysfs();
2007 btrfs_cleanup_fs_uuids();
2008 btrfs_exit_compress();
2009 btrfs_hash_exit();
2010}
2011
2012late_initcall(init_btrfs_fs);
2013module_exit(exit_btrfs_fs)
2014
2015MODULE_LICENSE("GPL");