Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/blkdev.h>
7#include <linux/module.h>
8#include <linux/fs.h>
9#include <linux/pagemap.h>
10#include <linux/highmem.h>
11#include <linux/time.h>
12#include <linux/init.h>
13#include <linux/seq_file.h>
14#include <linux/string.h>
15#include <linux/backing-dev.h>
16#include <linux/mount.h>
17#include <linux/writeback.h>
18#include <linux/statfs.h>
19#include <linux/compat.h>
20#include <linux/parser.h>
21#include <linux/ctype.h>
22#include <linux/namei.h>
23#include <linux/miscdevice.h>
24#include <linux/magic.h>
25#include <linux/slab.h>
26#include <linux/cleancache.h>
27#include <linux/ratelimit.h>
28#include <linux/crc32c.h>
29#include <linux/btrfs.h>
30#include "delayed-inode.h"
31#include "ctree.h"
32#include "disk-io.h"
33#include "transaction.h"
34#include "btrfs_inode.h"
35#include "print-tree.h"
36#include "props.h"
37#include "xattr.h"
38#include "volumes.h"
39#include "export.h"
40#include "compression.h"
41#include "rcu-string.h"
42#include "dev-replace.h"
43#include "free-space-cache.h"
44#include "backref.h"
45#include "space-info.h"
46#include "sysfs.h"
47#include "tests/btrfs-tests.h"
48#include "block-group.h"
49#include "discard.h"
50
51#include "qgroup.h"
52#define CREATE_TRACE_POINTS
53#include <trace/events/btrfs.h>
54
55static const struct super_operations btrfs_super_ops;
56
57/*
58 * Types for mounting the default subvolume and a subvolume explicitly
59 * requested by subvol=/path. That way the callchain is straightforward and we
60 * don't have to play tricks with the mount options and recursive calls to
61 * btrfs_mount.
62 *
63 * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
64 */
65static struct file_system_type btrfs_fs_type;
66static struct file_system_type btrfs_root_fs_type;
67
68static int btrfs_remount(struct super_block *sb, int *flags, char *data);
69
70/*
71 * Generally the error codes correspond to their respective errors, but there
72 * are a few special cases.
73 *
74 * EUCLEAN: Any sort of corruption that we encounter. The tree-checker for
75 * instance will return EUCLEAN if any of the blocks are corrupted in
76 * a way that is problematic. We want to reserve EUCLEAN for these
77 * sort of corruptions.
78 *
79 * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we
80 * need to use EROFS for this case. We will have no idea of the
81 * original failure, that will have been reported at the time we tripped
82 * over the error. Each subsequent error that doesn't have any context
83 * of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR.
84 */
85const char * __attribute_const__ btrfs_decode_error(int errno)
86{
87 char *errstr = "unknown";
88
89 switch (errno) {
90 case -ENOENT: /* -2 */
91 errstr = "No such entry";
92 break;
93 case -EIO: /* -5 */
94 errstr = "IO failure";
95 break;
96 case -ENOMEM: /* -12*/
97 errstr = "Out of memory";
98 break;
99 case -EEXIST: /* -17 */
100 errstr = "Object already exists";
101 break;
102 case -ENOSPC: /* -28 */
103 errstr = "No space left";
104 break;
105 case -EROFS: /* -30 */
106 errstr = "Readonly filesystem";
107 break;
108 case -EOPNOTSUPP: /* -95 */
109 errstr = "Operation not supported";
110 break;
111 case -EUCLEAN: /* -117 */
112 errstr = "Filesystem corrupted";
113 break;
114 case -EDQUOT: /* -122 */
115 errstr = "Quota exceeded";
116 break;
117 }
118
119 return errstr;
120}
121
122/*
123 * __btrfs_handle_fs_error decodes expected errors from the caller and
124 * invokes the appropriate error response.
125 */
126__cold
127void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
128 unsigned int line, int errno, const char *fmt, ...)
129{
130 struct super_block *sb = fs_info->sb;
131#ifdef CONFIG_PRINTK
132 const char *errstr;
133#endif
134
135 /*
136 * Special case: if the error is EROFS, and we're already
137 * under SB_RDONLY, then it is safe here.
138 */
139 if (errno == -EROFS && sb_rdonly(sb))
140 return;
141
142#ifdef CONFIG_PRINTK
143 errstr = btrfs_decode_error(errno);
144 if (fmt) {
145 struct va_format vaf;
146 va_list args;
147
148 va_start(args, fmt);
149 vaf.fmt = fmt;
150 vaf.va = &args;
151
152 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
153 sb->s_id, function, line, errno, errstr, &vaf);
154 va_end(args);
155 } else {
156 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
157 sb->s_id, function, line, errno, errstr);
158 }
159#endif
160
161 /*
162 * Today we only save the error info to memory. Long term we'll
163 * also send it down to the disk
164 */
165 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
166
167 /* Don't go through full error handling during mount */
168 if (!(sb->s_flags & SB_BORN))
169 return;
170
171 if (sb_rdonly(sb))
172 return;
173
174 btrfs_discard_stop(fs_info);
175
176 /* btrfs handle error by forcing the filesystem readonly */
177 sb->s_flags |= SB_RDONLY;
178 btrfs_info(fs_info, "forced readonly");
179 /*
180 * Note that a running device replace operation is not canceled here
181 * although there is no way to update the progress. It would add the
182 * risk of a deadlock, therefore the canceling is omitted. The only
183 * penalty is that some I/O remains active until the procedure
184 * completes. The next time when the filesystem is mounted writable
185 * again, the device replace operation continues.
186 */
187}
188
189#ifdef CONFIG_PRINTK
190static const char * const logtypes[] = {
191 "emergency",
192 "alert",
193 "critical",
194 "error",
195 "warning",
196 "notice",
197 "info",
198 "debug",
199};
200
201
202/*
203 * Use one ratelimit state per log level so that a flood of less important
204 * messages doesn't cause more important ones to be dropped.
205 */
206static struct ratelimit_state printk_limits[] = {
207 RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
208 RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
209 RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
210 RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
211 RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
212 RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
213 RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
214 RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
215};
216
217void __cold btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
218{
219 char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
220 struct va_format vaf;
221 va_list args;
222 int kern_level;
223 const char *type = logtypes[4];
224 struct ratelimit_state *ratelimit = &printk_limits[4];
225
226 va_start(args, fmt);
227
228 while ((kern_level = printk_get_level(fmt)) != 0) {
229 size_t size = printk_skip_level(fmt) - fmt;
230
231 if (kern_level >= '0' && kern_level <= '7') {
232 memcpy(lvl, fmt, size);
233 lvl[size] = '\0';
234 type = logtypes[kern_level - '0'];
235 ratelimit = &printk_limits[kern_level - '0'];
236 }
237 fmt += size;
238 }
239
240 vaf.fmt = fmt;
241 vaf.va = &args;
242
243 if (__ratelimit(ratelimit))
244 printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
245 fs_info ? fs_info->sb->s_id : "<unknown>", &vaf);
246
247 va_end(args);
248}
249#endif
250
251/*
252 * We only mark the transaction aborted and then set the file system read-only.
253 * This will prevent new transactions from starting or trying to join this
254 * one.
255 *
256 * This means that error recovery at the call site is limited to freeing
257 * any local memory allocations and passing the error code up without
258 * further cleanup. The transaction should complete as it normally would
259 * in the call path but will return -EIO.
260 *
261 * We'll complete the cleanup in btrfs_end_transaction and
262 * btrfs_commit_transaction.
263 */
264__cold
265void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
266 const char *function,
267 unsigned int line, int errno)
268{
269 struct btrfs_fs_info *fs_info = trans->fs_info;
270
271 WRITE_ONCE(trans->aborted, errno);
272 /* Nothing used. The other threads that have joined this
273 * transaction may be able to continue. */
274 if (!trans->dirty && list_empty(&trans->new_bgs)) {
275 const char *errstr;
276
277 errstr = btrfs_decode_error(errno);
278 btrfs_warn(fs_info,
279 "%s:%d: Aborting unused transaction(%s).",
280 function, line, errstr);
281 return;
282 }
283 WRITE_ONCE(trans->transaction->aborted, errno);
284 /* Wake up anybody who may be waiting on this transaction */
285 wake_up(&fs_info->transaction_wait);
286 wake_up(&fs_info->transaction_blocked_wait);
287 __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
288}
289/*
290 * __btrfs_panic decodes unexpected, fatal errors from the caller,
291 * issues an alert, and either panics or BUGs, depending on mount options.
292 */
293__cold
294void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
295 unsigned int line, int errno, const char *fmt, ...)
296{
297 char *s_id = "<unknown>";
298 const char *errstr;
299 struct va_format vaf = { .fmt = fmt };
300 va_list args;
301
302 if (fs_info)
303 s_id = fs_info->sb->s_id;
304
305 va_start(args, fmt);
306 vaf.va = &args;
307
308 errstr = btrfs_decode_error(errno);
309 if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
310 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
311 s_id, function, line, &vaf, errno, errstr);
312
313 btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
314 function, line, &vaf, errno, errstr);
315 va_end(args);
316 /* Caller calls BUG() */
317}
318
319static void btrfs_put_super(struct super_block *sb)
320{
321 close_ctree(btrfs_sb(sb));
322}
323
324enum {
325 Opt_acl, Opt_noacl,
326 Opt_clear_cache,
327 Opt_commit_interval,
328 Opt_compress,
329 Opt_compress_force,
330 Opt_compress_force_type,
331 Opt_compress_type,
332 Opt_degraded,
333 Opt_device,
334 Opt_fatal_errors,
335 Opt_flushoncommit, Opt_noflushoncommit,
336 Opt_inode_cache, Opt_noinode_cache,
337 Opt_max_inline,
338 Opt_barrier, Opt_nobarrier,
339 Opt_datacow, Opt_nodatacow,
340 Opt_datasum, Opt_nodatasum,
341 Opt_defrag, Opt_nodefrag,
342 Opt_discard, Opt_nodiscard,
343 Opt_discard_mode,
344 Opt_norecovery,
345 Opt_ratio,
346 Opt_rescan_uuid_tree,
347 Opt_skip_balance,
348 Opt_space_cache, Opt_no_space_cache,
349 Opt_space_cache_version,
350 Opt_ssd, Opt_nossd,
351 Opt_ssd_spread, Opt_nossd_spread,
352 Opt_subvol,
353 Opt_subvol_empty,
354 Opt_subvolid,
355 Opt_thread_pool,
356 Opt_treelog, Opt_notreelog,
357 Opt_user_subvol_rm_allowed,
358
359 /* Rescue options */
360 Opt_rescue,
361 Opt_usebackuproot,
362 Opt_nologreplay,
363
364 /* Deprecated options */
365 Opt_recovery,
366
367 /* Debugging options */
368 Opt_check_integrity,
369 Opt_check_integrity_including_extent_data,
370 Opt_check_integrity_print_mask,
371 Opt_enospc_debug, Opt_noenospc_debug,
372#ifdef CONFIG_BTRFS_DEBUG
373 Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
374#endif
375#ifdef CONFIG_BTRFS_FS_REF_VERIFY
376 Opt_ref_verify,
377#endif
378 Opt_err,
379};
380
381static const match_table_t tokens = {
382 {Opt_acl, "acl"},
383 {Opt_noacl, "noacl"},
384 {Opt_clear_cache, "clear_cache"},
385 {Opt_commit_interval, "commit=%u"},
386 {Opt_compress, "compress"},
387 {Opt_compress_type, "compress=%s"},
388 {Opt_compress_force, "compress-force"},
389 {Opt_compress_force_type, "compress-force=%s"},
390 {Opt_degraded, "degraded"},
391 {Opt_device, "device=%s"},
392 {Opt_fatal_errors, "fatal_errors=%s"},
393 {Opt_flushoncommit, "flushoncommit"},
394 {Opt_noflushoncommit, "noflushoncommit"},
395 {Opt_inode_cache, "inode_cache"},
396 {Opt_noinode_cache, "noinode_cache"},
397 {Opt_max_inline, "max_inline=%s"},
398 {Opt_barrier, "barrier"},
399 {Opt_nobarrier, "nobarrier"},
400 {Opt_datacow, "datacow"},
401 {Opt_nodatacow, "nodatacow"},
402 {Opt_datasum, "datasum"},
403 {Opt_nodatasum, "nodatasum"},
404 {Opt_defrag, "autodefrag"},
405 {Opt_nodefrag, "noautodefrag"},
406 {Opt_discard, "discard"},
407 {Opt_discard_mode, "discard=%s"},
408 {Opt_nodiscard, "nodiscard"},
409 {Opt_norecovery, "norecovery"},
410 {Opt_ratio, "metadata_ratio=%u"},
411 {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
412 {Opt_skip_balance, "skip_balance"},
413 {Opt_space_cache, "space_cache"},
414 {Opt_no_space_cache, "nospace_cache"},
415 {Opt_space_cache_version, "space_cache=%s"},
416 {Opt_ssd, "ssd"},
417 {Opt_nossd, "nossd"},
418 {Opt_ssd_spread, "ssd_spread"},
419 {Opt_nossd_spread, "nossd_spread"},
420 {Opt_subvol, "subvol=%s"},
421 {Opt_subvol_empty, "subvol="},
422 {Opt_subvolid, "subvolid=%s"},
423 {Opt_thread_pool, "thread_pool=%u"},
424 {Opt_treelog, "treelog"},
425 {Opt_notreelog, "notreelog"},
426 {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
427
428 /* Rescue options */
429 {Opt_rescue, "rescue=%s"},
430 /* Deprecated, with alias rescue=nologreplay */
431 {Opt_nologreplay, "nologreplay"},
432 /* Deprecated, with alias rescue=usebackuproot */
433 {Opt_usebackuproot, "usebackuproot"},
434
435 /* Deprecated options */
436 {Opt_recovery, "recovery"},
437
438 /* Debugging options */
439 {Opt_check_integrity, "check_int"},
440 {Opt_check_integrity_including_extent_data, "check_int_data"},
441 {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
442 {Opt_enospc_debug, "enospc_debug"},
443 {Opt_noenospc_debug, "noenospc_debug"},
444#ifdef CONFIG_BTRFS_DEBUG
445 {Opt_fragment_data, "fragment=data"},
446 {Opt_fragment_metadata, "fragment=metadata"},
447 {Opt_fragment_all, "fragment=all"},
448#endif
449#ifdef CONFIG_BTRFS_FS_REF_VERIFY
450 {Opt_ref_verify, "ref_verify"},
451#endif
452 {Opt_err, NULL},
453};
454
455static const match_table_t rescue_tokens = {
456 {Opt_usebackuproot, "usebackuproot"},
457 {Opt_nologreplay, "nologreplay"},
458 {Opt_err, NULL},
459};
460
461static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
462{
463 char *opts;
464 char *orig;
465 char *p;
466 substring_t args[MAX_OPT_ARGS];
467 int ret = 0;
468
469 opts = kstrdup(options, GFP_KERNEL);
470 if (!opts)
471 return -ENOMEM;
472 orig = opts;
473
474 while ((p = strsep(&opts, ":")) != NULL) {
475 int token;
476
477 if (!*p)
478 continue;
479 token = match_token(p, rescue_tokens, args);
480 switch (token){
481 case Opt_usebackuproot:
482 btrfs_info(info,
483 "trying to use backup root at mount time");
484 btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
485 break;
486 case Opt_nologreplay:
487 btrfs_set_and_info(info, NOLOGREPLAY,
488 "disabling log replay at mount time");
489 break;
490 case Opt_err:
491 btrfs_info(info, "unrecognized rescue option '%s'", p);
492 ret = -EINVAL;
493 goto out;
494 default:
495 break;
496 }
497
498 }
499out:
500 kfree(orig);
501 return ret;
502}
503
504/*
505 * Regular mount options parser. Everything that is needed only when
506 * reading in a new superblock is parsed here.
507 * XXX JDM: This needs to be cleaned up for remount.
508 */
509int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
510 unsigned long new_flags)
511{
512 substring_t args[MAX_OPT_ARGS];
513 char *p, *num;
514 u64 cache_gen;
515 int intarg;
516 int ret = 0;
517 char *compress_type;
518 bool compress_force = false;
519 enum btrfs_compression_type saved_compress_type;
520 int saved_compress_level;
521 bool saved_compress_force;
522 int no_compress = 0;
523
524 cache_gen = btrfs_super_cache_generation(info->super_copy);
525 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
526 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
527 else if (cache_gen)
528 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
529
530 /*
531 * Even the options are empty, we still need to do extra check
532 * against new flags
533 */
534 if (!options)
535 goto check;
536
537 while ((p = strsep(&options, ",")) != NULL) {
538 int token;
539 if (!*p)
540 continue;
541
542 token = match_token(p, tokens, args);
543 switch (token) {
544 case Opt_degraded:
545 btrfs_info(info, "allowing degraded mounts");
546 btrfs_set_opt(info->mount_opt, DEGRADED);
547 break;
548 case Opt_subvol:
549 case Opt_subvol_empty:
550 case Opt_subvolid:
551 case Opt_device:
552 /*
553 * These are parsed by btrfs_parse_subvol_options or
554 * btrfs_parse_device_options and can be ignored here.
555 */
556 break;
557 case Opt_nodatasum:
558 btrfs_set_and_info(info, NODATASUM,
559 "setting nodatasum");
560 break;
561 case Opt_datasum:
562 if (btrfs_test_opt(info, NODATASUM)) {
563 if (btrfs_test_opt(info, NODATACOW))
564 btrfs_info(info,
565 "setting datasum, datacow enabled");
566 else
567 btrfs_info(info, "setting datasum");
568 }
569 btrfs_clear_opt(info->mount_opt, NODATACOW);
570 btrfs_clear_opt(info->mount_opt, NODATASUM);
571 break;
572 case Opt_nodatacow:
573 if (!btrfs_test_opt(info, NODATACOW)) {
574 if (!btrfs_test_opt(info, COMPRESS) ||
575 !btrfs_test_opt(info, FORCE_COMPRESS)) {
576 btrfs_info(info,
577 "setting nodatacow, compression disabled");
578 } else {
579 btrfs_info(info, "setting nodatacow");
580 }
581 }
582 btrfs_clear_opt(info->mount_opt, COMPRESS);
583 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
584 btrfs_set_opt(info->mount_opt, NODATACOW);
585 btrfs_set_opt(info->mount_opt, NODATASUM);
586 break;
587 case Opt_datacow:
588 btrfs_clear_and_info(info, NODATACOW,
589 "setting datacow");
590 break;
591 case Opt_compress_force:
592 case Opt_compress_force_type:
593 compress_force = true;
594 fallthrough;
595 case Opt_compress:
596 case Opt_compress_type:
597 saved_compress_type = btrfs_test_opt(info,
598 COMPRESS) ?
599 info->compress_type : BTRFS_COMPRESS_NONE;
600 saved_compress_force =
601 btrfs_test_opt(info, FORCE_COMPRESS);
602 saved_compress_level = info->compress_level;
603 if (token == Opt_compress ||
604 token == Opt_compress_force ||
605 strncmp(args[0].from, "zlib", 4) == 0) {
606 compress_type = "zlib";
607
608 info->compress_type = BTRFS_COMPRESS_ZLIB;
609 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
610 /*
611 * args[0] contains uninitialized data since
612 * for these tokens we don't expect any
613 * parameter.
614 */
615 if (token != Opt_compress &&
616 token != Opt_compress_force)
617 info->compress_level =
618 btrfs_compress_str2level(
619 BTRFS_COMPRESS_ZLIB,
620 args[0].from + 4);
621 btrfs_set_opt(info->mount_opt, COMPRESS);
622 btrfs_clear_opt(info->mount_opt, NODATACOW);
623 btrfs_clear_opt(info->mount_opt, NODATASUM);
624 no_compress = 0;
625 } else if (strncmp(args[0].from, "lzo", 3) == 0) {
626 compress_type = "lzo";
627 info->compress_type = BTRFS_COMPRESS_LZO;
628 info->compress_level = 0;
629 btrfs_set_opt(info->mount_opt, COMPRESS);
630 btrfs_clear_opt(info->mount_opt, NODATACOW);
631 btrfs_clear_opt(info->mount_opt, NODATASUM);
632 btrfs_set_fs_incompat(info, COMPRESS_LZO);
633 no_compress = 0;
634 } else if (strncmp(args[0].from, "zstd", 4) == 0) {
635 compress_type = "zstd";
636 info->compress_type = BTRFS_COMPRESS_ZSTD;
637 info->compress_level =
638 btrfs_compress_str2level(
639 BTRFS_COMPRESS_ZSTD,
640 args[0].from + 4);
641 btrfs_set_opt(info->mount_opt, COMPRESS);
642 btrfs_clear_opt(info->mount_opt, NODATACOW);
643 btrfs_clear_opt(info->mount_opt, NODATASUM);
644 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
645 no_compress = 0;
646 } else if (strncmp(args[0].from, "no", 2) == 0) {
647 compress_type = "no";
648 info->compress_level = 0;
649 info->compress_type = 0;
650 btrfs_clear_opt(info->mount_opt, COMPRESS);
651 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
652 compress_force = false;
653 no_compress++;
654 } else {
655 ret = -EINVAL;
656 goto out;
657 }
658
659 if (compress_force) {
660 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
661 } else {
662 /*
663 * If we remount from compress-force=xxx to
664 * compress=xxx, we need clear FORCE_COMPRESS
665 * flag, otherwise, there is no way for users
666 * to disable forcible compression separately.
667 */
668 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
669 }
670 if (no_compress == 1) {
671 btrfs_info(info, "use no compression");
672 } else if ((info->compress_type != saved_compress_type) ||
673 (compress_force != saved_compress_force) ||
674 (info->compress_level != saved_compress_level)) {
675 btrfs_info(info, "%s %s compression, level %d",
676 (compress_force) ? "force" : "use",
677 compress_type, info->compress_level);
678 }
679 compress_force = false;
680 break;
681 case Opt_ssd:
682 btrfs_set_and_info(info, SSD,
683 "enabling ssd optimizations");
684 btrfs_clear_opt(info->mount_opt, NOSSD);
685 break;
686 case Opt_ssd_spread:
687 btrfs_set_and_info(info, SSD,
688 "enabling ssd optimizations");
689 btrfs_set_and_info(info, SSD_SPREAD,
690 "using spread ssd allocation scheme");
691 btrfs_clear_opt(info->mount_opt, NOSSD);
692 break;
693 case Opt_nossd:
694 btrfs_set_opt(info->mount_opt, NOSSD);
695 btrfs_clear_and_info(info, SSD,
696 "not using ssd optimizations");
697 fallthrough;
698 case Opt_nossd_spread:
699 btrfs_clear_and_info(info, SSD_SPREAD,
700 "not using spread ssd allocation scheme");
701 break;
702 case Opt_barrier:
703 btrfs_clear_and_info(info, NOBARRIER,
704 "turning on barriers");
705 break;
706 case Opt_nobarrier:
707 btrfs_set_and_info(info, NOBARRIER,
708 "turning off barriers");
709 break;
710 case Opt_thread_pool:
711 ret = match_int(&args[0], &intarg);
712 if (ret) {
713 goto out;
714 } else if (intarg == 0) {
715 ret = -EINVAL;
716 goto out;
717 }
718 info->thread_pool_size = intarg;
719 break;
720 case Opt_max_inline:
721 num = match_strdup(&args[0]);
722 if (num) {
723 info->max_inline = memparse(num, NULL);
724 kfree(num);
725
726 if (info->max_inline) {
727 info->max_inline = min_t(u64,
728 info->max_inline,
729 info->sectorsize);
730 }
731 btrfs_info(info, "max_inline at %llu",
732 info->max_inline);
733 } else {
734 ret = -ENOMEM;
735 goto out;
736 }
737 break;
738 case Opt_acl:
739#ifdef CONFIG_BTRFS_FS_POSIX_ACL
740 info->sb->s_flags |= SB_POSIXACL;
741 break;
742#else
743 btrfs_err(info, "support for ACL not compiled in!");
744 ret = -EINVAL;
745 goto out;
746#endif
747 case Opt_noacl:
748 info->sb->s_flags &= ~SB_POSIXACL;
749 break;
750 case Opt_notreelog:
751 btrfs_set_and_info(info, NOTREELOG,
752 "disabling tree log");
753 break;
754 case Opt_treelog:
755 btrfs_clear_and_info(info, NOTREELOG,
756 "enabling tree log");
757 break;
758 case Opt_norecovery:
759 case Opt_nologreplay:
760 btrfs_warn(info,
761 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
762 btrfs_set_and_info(info, NOLOGREPLAY,
763 "disabling log replay at mount time");
764 break;
765 case Opt_flushoncommit:
766 btrfs_set_and_info(info, FLUSHONCOMMIT,
767 "turning on flush-on-commit");
768 break;
769 case Opt_noflushoncommit:
770 btrfs_clear_and_info(info, FLUSHONCOMMIT,
771 "turning off flush-on-commit");
772 break;
773 case Opt_ratio:
774 ret = match_int(&args[0], &intarg);
775 if (ret)
776 goto out;
777 info->metadata_ratio = intarg;
778 btrfs_info(info, "metadata ratio %u",
779 info->metadata_ratio);
780 break;
781 case Opt_discard:
782 case Opt_discard_mode:
783 if (token == Opt_discard ||
784 strcmp(args[0].from, "sync") == 0) {
785 btrfs_clear_opt(info->mount_opt, DISCARD_ASYNC);
786 btrfs_set_and_info(info, DISCARD_SYNC,
787 "turning on sync discard");
788 } else if (strcmp(args[0].from, "async") == 0) {
789 btrfs_clear_opt(info->mount_opt, DISCARD_SYNC);
790 btrfs_set_and_info(info, DISCARD_ASYNC,
791 "turning on async discard");
792 } else {
793 ret = -EINVAL;
794 goto out;
795 }
796 break;
797 case Opt_nodiscard:
798 btrfs_clear_and_info(info, DISCARD_SYNC,
799 "turning off discard");
800 btrfs_clear_and_info(info, DISCARD_ASYNC,
801 "turning off async discard");
802 break;
803 case Opt_space_cache:
804 case Opt_space_cache_version:
805 if (token == Opt_space_cache ||
806 strcmp(args[0].from, "v1") == 0) {
807 btrfs_clear_opt(info->mount_opt,
808 FREE_SPACE_TREE);
809 btrfs_set_and_info(info, SPACE_CACHE,
810 "enabling disk space caching");
811 } else if (strcmp(args[0].from, "v2") == 0) {
812 btrfs_clear_opt(info->mount_opt,
813 SPACE_CACHE);
814 btrfs_set_and_info(info, FREE_SPACE_TREE,
815 "enabling free space tree");
816 } else {
817 ret = -EINVAL;
818 goto out;
819 }
820 break;
821 case Opt_rescan_uuid_tree:
822 btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
823 break;
824 case Opt_no_space_cache:
825 if (btrfs_test_opt(info, SPACE_CACHE)) {
826 btrfs_clear_and_info(info, SPACE_CACHE,
827 "disabling disk space caching");
828 }
829 if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
830 btrfs_clear_and_info(info, FREE_SPACE_TREE,
831 "disabling free space tree");
832 }
833 break;
834 case Opt_inode_cache:
835 btrfs_warn(info,
836 "the 'inode_cache' option is deprecated and will have no effect from 5.11");
837 btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
838 "enabling inode map caching");
839 break;
840 case Opt_noinode_cache:
841 btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
842 "disabling inode map caching");
843 break;
844 case Opt_clear_cache:
845 btrfs_set_and_info(info, CLEAR_CACHE,
846 "force clearing of disk cache");
847 break;
848 case Opt_user_subvol_rm_allowed:
849 btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
850 break;
851 case Opt_enospc_debug:
852 btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
853 break;
854 case Opt_noenospc_debug:
855 btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
856 break;
857 case Opt_defrag:
858 btrfs_set_and_info(info, AUTO_DEFRAG,
859 "enabling auto defrag");
860 break;
861 case Opt_nodefrag:
862 btrfs_clear_and_info(info, AUTO_DEFRAG,
863 "disabling auto defrag");
864 break;
865 case Opt_recovery:
866 case Opt_usebackuproot:
867 btrfs_warn(info,
868 "'%s' is deprecated, use 'rescue=usebackuproot' instead",
869 token == Opt_recovery ? "recovery" :
870 "usebackuproot");
871 btrfs_info(info,
872 "trying to use backup root at mount time");
873 btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
874 break;
875 case Opt_skip_balance:
876 btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
877 break;
878#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
879 case Opt_check_integrity_including_extent_data:
880 btrfs_info(info,
881 "enabling check integrity including extent data");
882 btrfs_set_opt(info->mount_opt,
883 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
884 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
885 break;
886 case Opt_check_integrity:
887 btrfs_info(info, "enabling check integrity");
888 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
889 break;
890 case Opt_check_integrity_print_mask:
891 ret = match_int(&args[0], &intarg);
892 if (ret)
893 goto out;
894 info->check_integrity_print_mask = intarg;
895 btrfs_info(info, "check_integrity_print_mask 0x%x",
896 info->check_integrity_print_mask);
897 break;
898#else
899 case Opt_check_integrity_including_extent_data:
900 case Opt_check_integrity:
901 case Opt_check_integrity_print_mask:
902 btrfs_err(info,
903 "support for check_integrity* not compiled in!");
904 ret = -EINVAL;
905 goto out;
906#endif
907 case Opt_fatal_errors:
908 if (strcmp(args[0].from, "panic") == 0)
909 btrfs_set_opt(info->mount_opt,
910 PANIC_ON_FATAL_ERROR);
911 else if (strcmp(args[0].from, "bug") == 0)
912 btrfs_clear_opt(info->mount_opt,
913 PANIC_ON_FATAL_ERROR);
914 else {
915 ret = -EINVAL;
916 goto out;
917 }
918 break;
919 case Opt_commit_interval:
920 intarg = 0;
921 ret = match_int(&args[0], &intarg);
922 if (ret)
923 goto out;
924 if (intarg == 0) {
925 btrfs_info(info,
926 "using default commit interval %us",
927 BTRFS_DEFAULT_COMMIT_INTERVAL);
928 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
929 } else if (intarg > 300) {
930 btrfs_warn(info, "excessive commit interval %d",
931 intarg);
932 }
933 info->commit_interval = intarg;
934 break;
935 case Opt_rescue:
936 ret = parse_rescue_options(info, args[0].from);
937 if (ret < 0)
938 goto out;
939 break;
940#ifdef CONFIG_BTRFS_DEBUG
941 case Opt_fragment_all:
942 btrfs_info(info, "fragmenting all space");
943 btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
944 btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
945 break;
946 case Opt_fragment_metadata:
947 btrfs_info(info, "fragmenting metadata");
948 btrfs_set_opt(info->mount_opt,
949 FRAGMENT_METADATA);
950 break;
951 case Opt_fragment_data:
952 btrfs_info(info, "fragmenting data");
953 btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
954 break;
955#endif
956#ifdef CONFIG_BTRFS_FS_REF_VERIFY
957 case Opt_ref_verify:
958 btrfs_info(info, "doing ref verification");
959 btrfs_set_opt(info->mount_opt, REF_VERIFY);
960 break;
961#endif
962 case Opt_err:
963 btrfs_err(info, "unrecognized mount option '%s'", p);
964 ret = -EINVAL;
965 goto out;
966 default:
967 break;
968 }
969 }
970check:
971 /*
972 * Extra check for current option against current flag
973 */
974 if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & SB_RDONLY)) {
975 btrfs_err(info,
976 "nologreplay must be used with ro mount option");
977 ret = -EINVAL;
978 }
979out:
980 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
981 !btrfs_test_opt(info, FREE_SPACE_TREE) &&
982 !btrfs_test_opt(info, CLEAR_CACHE)) {
983 btrfs_err(info, "cannot disable free space tree");
984 ret = -EINVAL;
985
986 }
987 if (!ret && btrfs_test_opt(info, SPACE_CACHE))
988 btrfs_info(info, "disk space caching is enabled");
989 if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
990 btrfs_info(info, "using free space tree");
991 return ret;
992}
993
994/*
995 * Parse mount options that are required early in the mount process.
996 *
997 * All other options will be parsed on much later in the mount process and
998 * only when we need to allocate a new super block.
999 */
1000static int btrfs_parse_device_options(const char *options, fmode_t flags,
1001 void *holder)
1002{
1003 substring_t args[MAX_OPT_ARGS];
1004 char *device_name, *opts, *orig, *p;
1005 struct btrfs_device *device = NULL;
1006 int error = 0;
1007
1008 lockdep_assert_held(&uuid_mutex);
1009
1010 if (!options)
1011 return 0;
1012
1013 /*
1014 * strsep changes the string, duplicate it because btrfs_parse_options
1015 * gets called later
1016 */
1017 opts = kstrdup(options, GFP_KERNEL);
1018 if (!opts)
1019 return -ENOMEM;
1020 orig = opts;
1021
1022 while ((p = strsep(&opts, ",")) != NULL) {
1023 int token;
1024
1025 if (!*p)
1026 continue;
1027
1028 token = match_token(p, tokens, args);
1029 if (token == Opt_device) {
1030 device_name = match_strdup(&args[0]);
1031 if (!device_name) {
1032 error = -ENOMEM;
1033 goto out;
1034 }
1035 device = btrfs_scan_one_device(device_name, flags,
1036 holder);
1037 kfree(device_name);
1038 if (IS_ERR(device)) {
1039 error = PTR_ERR(device);
1040 goto out;
1041 }
1042 }
1043 }
1044
1045out:
1046 kfree(orig);
1047 return error;
1048}
1049
1050/*
1051 * Parse mount options that are related to subvolume id
1052 *
1053 * The value is later passed to mount_subvol()
1054 */
1055static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1056 u64 *subvol_objectid)
1057{
1058 substring_t args[MAX_OPT_ARGS];
1059 char *opts, *orig, *p;
1060 int error = 0;
1061 u64 subvolid;
1062
1063 if (!options)
1064 return 0;
1065
1066 /*
1067 * strsep changes the string, duplicate it because
1068 * btrfs_parse_device_options gets called later
1069 */
1070 opts = kstrdup(options, GFP_KERNEL);
1071 if (!opts)
1072 return -ENOMEM;
1073 orig = opts;
1074
1075 while ((p = strsep(&opts, ",")) != NULL) {
1076 int token;
1077 if (!*p)
1078 continue;
1079
1080 token = match_token(p, tokens, args);
1081 switch (token) {
1082 case Opt_subvol:
1083 kfree(*subvol_name);
1084 *subvol_name = match_strdup(&args[0]);
1085 if (!*subvol_name) {
1086 error = -ENOMEM;
1087 goto out;
1088 }
1089 break;
1090 case Opt_subvolid:
1091 error = match_u64(&args[0], &subvolid);
1092 if (error)
1093 goto out;
1094
1095 /* we want the original fs_tree */
1096 if (subvolid == 0)
1097 subvolid = BTRFS_FS_TREE_OBJECTID;
1098
1099 *subvol_objectid = subvolid;
1100 break;
1101 default:
1102 break;
1103 }
1104 }
1105
1106out:
1107 kfree(orig);
1108 return error;
1109}
1110
1111char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1112 u64 subvol_objectid)
1113{
1114 struct btrfs_root *root = fs_info->tree_root;
1115 struct btrfs_root *fs_root = NULL;
1116 struct btrfs_root_ref *root_ref;
1117 struct btrfs_inode_ref *inode_ref;
1118 struct btrfs_key key;
1119 struct btrfs_path *path = NULL;
1120 char *name = NULL, *ptr;
1121 u64 dirid;
1122 int len;
1123 int ret;
1124
1125 path = btrfs_alloc_path();
1126 if (!path) {
1127 ret = -ENOMEM;
1128 goto err;
1129 }
1130 path->leave_spinning = 1;
1131
1132 name = kmalloc(PATH_MAX, GFP_KERNEL);
1133 if (!name) {
1134 ret = -ENOMEM;
1135 goto err;
1136 }
1137 ptr = name + PATH_MAX - 1;
1138 ptr[0] = '\0';
1139
1140 /*
1141 * Walk up the subvolume trees in the tree of tree roots by root
1142 * backrefs until we hit the top-level subvolume.
1143 */
1144 while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1145 key.objectid = subvol_objectid;
1146 key.type = BTRFS_ROOT_BACKREF_KEY;
1147 key.offset = (u64)-1;
1148
1149 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1150 if (ret < 0) {
1151 goto err;
1152 } else if (ret > 0) {
1153 ret = btrfs_previous_item(root, path, subvol_objectid,
1154 BTRFS_ROOT_BACKREF_KEY);
1155 if (ret < 0) {
1156 goto err;
1157 } else if (ret > 0) {
1158 ret = -ENOENT;
1159 goto err;
1160 }
1161 }
1162
1163 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1164 subvol_objectid = key.offset;
1165
1166 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1167 struct btrfs_root_ref);
1168 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1169 ptr -= len + 1;
1170 if (ptr < name) {
1171 ret = -ENAMETOOLONG;
1172 goto err;
1173 }
1174 read_extent_buffer(path->nodes[0], ptr + 1,
1175 (unsigned long)(root_ref + 1), len);
1176 ptr[0] = '/';
1177 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1178 btrfs_release_path(path);
1179
1180 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1181 if (IS_ERR(fs_root)) {
1182 ret = PTR_ERR(fs_root);
1183 fs_root = NULL;
1184 goto err;
1185 }
1186
1187 /*
1188 * Walk up the filesystem tree by inode refs until we hit the
1189 * root directory.
1190 */
1191 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1192 key.objectid = dirid;
1193 key.type = BTRFS_INODE_REF_KEY;
1194 key.offset = (u64)-1;
1195
1196 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1197 if (ret < 0) {
1198 goto err;
1199 } else if (ret > 0) {
1200 ret = btrfs_previous_item(fs_root, path, dirid,
1201 BTRFS_INODE_REF_KEY);
1202 if (ret < 0) {
1203 goto err;
1204 } else if (ret > 0) {
1205 ret = -ENOENT;
1206 goto err;
1207 }
1208 }
1209
1210 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1211 dirid = key.offset;
1212
1213 inode_ref = btrfs_item_ptr(path->nodes[0],
1214 path->slots[0],
1215 struct btrfs_inode_ref);
1216 len = btrfs_inode_ref_name_len(path->nodes[0],
1217 inode_ref);
1218 ptr -= len + 1;
1219 if (ptr < name) {
1220 ret = -ENAMETOOLONG;
1221 goto err;
1222 }
1223 read_extent_buffer(path->nodes[0], ptr + 1,
1224 (unsigned long)(inode_ref + 1), len);
1225 ptr[0] = '/';
1226 btrfs_release_path(path);
1227 }
1228 btrfs_put_root(fs_root);
1229 fs_root = NULL;
1230 }
1231
1232 btrfs_free_path(path);
1233 if (ptr == name + PATH_MAX - 1) {
1234 name[0] = '/';
1235 name[1] = '\0';
1236 } else {
1237 memmove(name, ptr, name + PATH_MAX - ptr);
1238 }
1239 return name;
1240
1241err:
1242 btrfs_put_root(fs_root);
1243 btrfs_free_path(path);
1244 kfree(name);
1245 return ERR_PTR(ret);
1246}
1247
1248static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1249{
1250 struct btrfs_root *root = fs_info->tree_root;
1251 struct btrfs_dir_item *di;
1252 struct btrfs_path *path;
1253 struct btrfs_key location;
1254 u64 dir_id;
1255
1256 path = btrfs_alloc_path();
1257 if (!path)
1258 return -ENOMEM;
1259 path->leave_spinning = 1;
1260
1261 /*
1262 * Find the "default" dir item which points to the root item that we
1263 * will mount by default if we haven't been given a specific subvolume
1264 * to mount.
1265 */
1266 dir_id = btrfs_super_root_dir(fs_info->super_copy);
1267 di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1268 if (IS_ERR(di)) {
1269 btrfs_free_path(path);
1270 return PTR_ERR(di);
1271 }
1272 if (!di) {
1273 /*
1274 * Ok the default dir item isn't there. This is weird since
1275 * it's always been there, but don't freak out, just try and
1276 * mount the top-level subvolume.
1277 */
1278 btrfs_free_path(path);
1279 *objectid = BTRFS_FS_TREE_OBJECTID;
1280 return 0;
1281 }
1282
1283 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1284 btrfs_free_path(path);
1285 *objectid = location.objectid;
1286 return 0;
1287}
1288
1289static int btrfs_fill_super(struct super_block *sb,
1290 struct btrfs_fs_devices *fs_devices,
1291 void *data)
1292{
1293 struct inode *inode;
1294 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1295 int err;
1296
1297 sb->s_maxbytes = MAX_LFS_FILESIZE;
1298 sb->s_magic = BTRFS_SUPER_MAGIC;
1299 sb->s_op = &btrfs_super_ops;
1300 sb->s_d_op = &btrfs_dentry_operations;
1301 sb->s_export_op = &btrfs_export_ops;
1302 sb->s_xattr = btrfs_xattr_handlers;
1303 sb->s_time_gran = 1;
1304#ifdef CONFIG_BTRFS_FS_POSIX_ACL
1305 sb->s_flags |= SB_POSIXACL;
1306#endif
1307 sb->s_flags |= SB_I_VERSION;
1308 sb->s_iflags |= SB_I_CGROUPWB;
1309
1310 err = super_setup_bdi(sb);
1311 if (err) {
1312 btrfs_err(fs_info, "super_setup_bdi failed");
1313 return err;
1314 }
1315
1316 err = open_ctree(sb, fs_devices, (char *)data);
1317 if (err) {
1318 btrfs_err(fs_info, "open_ctree failed");
1319 return err;
1320 }
1321
1322 inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1323 if (IS_ERR(inode)) {
1324 err = PTR_ERR(inode);
1325 goto fail_close;
1326 }
1327
1328 sb->s_root = d_make_root(inode);
1329 if (!sb->s_root) {
1330 err = -ENOMEM;
1331 goto fail_close;
1332 }
1333
1334 cleancache_init_fs(sb);
1335 sb->s_flags |= SB_ACTIVE;
1336 return 0;
1337
1338fail_close:
1339 close_ctree(fs_info);
1340 return err;
1341}
1342
1343int btrfs_sync_fs(struct super_block *sb, int wait)
1344{
1345 struct btrfs_trans_handle *trans;
1346 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1347 struct btrfs_root *root = fs_info->tree_root;
1348
1349 trace_btrfs_sync_fs(fs_info, wait);
1350
1351 if (!wait) {
1352 filemap_flush(fs_info->btree_inode->i_mapping);
1353 return 0;
1354 }
1355
1356 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1357
1358 trans = btrfs_attach_transaction_barrier(root);
1359 if (IS_ERR(trans)) {
1360 /* no transaction, don't bother */
1361 if (PTR_ERR(trans) == -ENOENT) {
1362 /*
1363 * Exit unless we have some pending changes
1364 * that need to go through commit
1365 */
1366 if (fs_info->pending_changes == 0)
1367 return 0;
1368 /*
1369 * A non-blocking test if the fs is frozen. We must not
1370 * start a new transaction here otherwise a deadlock
1371 * happens. The pending operations are delayed to the
1372 * next commit after thawing.
1373 */
1374 if (sb_start_write_trylock(sb))
1375 sb_end_write(sb);
1376 else
1377 return 0;
1378 trans = btrfs_start_transaction(root, 0);
1379 }
1380 if (IS_ERR(trans))
1381 return PTR_ERR(trans);
1382 }
1383 return btrfs_commit_transaction(trans);
1384}
1385
1386static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1387{
1388 struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1389 const char *compress_type;
1390 const char *subvol_name;
1391
1392 if (btrfs_test_opt(info, DEGRADED))
1393 seq_puts(seq, ",degraded");
1394 if (btrfs_test_opt(info, NODATASUM))
1395 seq_puts(seq, ",nodatasum");
1396 if (btrfs_test_opt(info, NODATACOW))
1397 seq_puts(seq, ",nodatacow");
1398 if (btrfs_test_opt(info, NOBARRIER))
1399 seq_puts(seq, ",nobarrier");
1400 if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1401 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1402 if (info->thread_pool_size != min_t(unsigned long,
1403 num_online_cpus() + 2, 8))
1404 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1405 if (btrfs_test_opt(info, COMPRESS)) {
1406 compress_type = btrfs_compress_type2str(info->compress_type);
1407 if (btrfs_test_opt(info, FORCE_COMPRESS))
1408 seq_printf(seq, ",compress-force=%s", compress_type);
1409 else
1410 seq_printf(seq, ",compress=%s", compress_type);
1411 if (info->compress_level)
1412 seq_printf(seq, ":%d", info->compress_level);
1413 }
1414 if (btrfs_test_opt(info, NOSSD))
1415 seq_puts(seq, ",nossd");
1416 if (btrfs_test_opt(info, SSD_SPREAD))
1417 seq_puts(seq, ",ssd_spread");
1418 else if (btrfs_test_opt(info, SSD))
1419 seq_puts(seq, ",ssd");
1420 if (btrfs_test_opt(info, NOTREELOG))
1421 seq_puts(seq, ",notreelog");
1422 if (btrfs_test_opt(info, NOLOGREPLAY))
1423 seq_puts(seq, ",rescue=nologreplay");
1424 if (btrfs_test_opt(info, FLUSHONCOMMIT))
1425 seq_puts(seq, ",flushoncommit");
1426 if (btrfs_test_opt(info, DISCARD_SYNC))
1427 seq_puts(seq, ",discard");
1428 if (btrfs_test_opt(info, DISCARD_ASYNC))
1429 seq_puts(seq, ",discard=async");
1430 if (!(info->sb->s_flags & SB_POSIXACL))
1431 seq_puts(seq, ",noacl");
1432 if (btrfs_test_opt(info, SPACE_CACHE))
1433 seq_puts(seq, ",space_cache");
1434 else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1435 seq_puts(seq, ",space_cache=v2");
1436 else
1437 seq_puts(seq, ",nospace_cache");
1438 if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1439 seq_puts(seq, ",rescan_uuid_tree");
1440 if (btrfs_test_opt(info, CLEAR_CACHE))
1441 seq_puts(seq, ",clear_cache");
1442 if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1443 seq_puts(seq, ",user_subvol_rm_allowed");
1444 if (btrfs_test_opt(info, ENOSPC_DEBUG))
1445 seq_puts(seq, ",enospc_debug");
1446 if (btrfs_test_opt(info, AUTO_DEFRAG))
1447 seq_puts(seq, ",autodefrag");
1448 if (btrfs_test_opt(info, INODE_MAP_CACHE))
1449 seq_puts(seq, ",inode_cache");
1450 if (btrfs_test_opt(info, SKIP_BALANCE))
1451 seq_puts(seq, ",skip_balance");
1452#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1453 if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1454 seq_puts(seq, ",check_int_data");
1455 else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1456 seq_puts(seq, ",check_int");
1457 if (info->check_integrity_print_mask)
1458 seq_printf(seq, ",check_int_print_mask=%d",
1459 info->check_integrity_print_mask);
1460#endif
1461 if (info->metadata_ratio)
1462 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1463 if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1464 seq_puts(seq, ",fatal_errors=panic");
1465 if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1466 seq_printf(seq, ",commit=%u", info->commit_interval);
1467#ifdef CONFIG_BTRFS_DEBUG
1468 if (btrfs_test_opt(info, FRAGMENT_DATA))
1469 seq_puts(seq, ",fragment=data");
1470 if (btrfs_test_opt(info, FRAGMENT_METADATA))
1471 seq_puts(seq, ",fragment=metadata");
1472#endif
1473 if (btrfs_test_opt(info, REF_VERIFY))
1474 seq_puts(seq, ",ref_verify");
1475 seq_printf(seq, ",subvolid=%llu",
1476 BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1477 subvol_name = btrfs_get_subvol_name_from_objectid(info,
1478 BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1479 if (!IS_ERR(subvol_name)) {
1480 seq_puts(seq, ",subvol=");
1481 seq_escape(seq, subvol_name, " \t\n\\");
1482 kfree(subvol_name);
1483 }
1484 return 0;
1485}
1486
1487static int btrfs_test_super(struct super_block *s, void *data)
1488{
1489 struct btrfs_fs_info *p = data;
1490 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1491
1492 return fs_info->fs_devices == p->fs_devices;
1493}
1494
1495static int btrfs_set_super(struct super_block *s, void *data)
1496{
1497 int err = set_anon_super(s, data);
1498 if (!err)
1499 s->s_fs_info = data;
1500 return err;
1501}
1502
1503/*
1504 * subvolumes are identified by ino 256
1505 */
1506static inline int is_subvolume_inode(struct inode *inode)
1507{
1508 if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1509 return 1;
1510 return 0;
1511}
1512
1513static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1514 struct vfsmount *mnt)
1515{
1516 struct dentry *root;
1517 int ret;
1518
1519 if (!subvol_name) {
1520 if (!subvol_objectid) {
1521 ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1522 &subvol_objectid);
1523 if (ret) {
1524 root = ERR_PTR(ret);
1525 goto out;
1526 }
1527 }
1528 subvol_name = btrfs_get_subvol_name_from_objectid(
1529 btrfs_sb(mnt->mnt_sb), subvol_objectid);
1530 if (IS_ERR(subvol_name)) {
1531 root = ERR_CAST(subvol_name);
1532 subvol_name = NULL;
1533 goto out;
1534 }
1535
1536 }
1537
1538 root = mount_subtree(mnt, subvol_name);
1539 /* mount_subtree() drops our reference on the vfsmount. */
1540 mnt = NULL;
1541
1542 if (!IS_ERR(root)) {
1543 struct super_block *s = root->d_sb;
1544 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1545 struct inode *root_inode = d_inode(root);
1546 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1547
1548 ret = 0;
1549 if (!is_subvolume_inode(root_inode)) {
1550 btrfs_err(fs_info, "'%s' is not a valid subvolume",
1551 subvol_name);
1552 ret = -EINVAL;
1553 }
1554 if (subvol_objectid && root_objectid != subvol_objectid) {
1555 /*
1556 * This will also catch a race condition where a
1557 * subvolume which was passed by ID is renamed and
1558 * another subvolume is renamed over the old location.
1559 */
1560 btrfs_err(fs_info,
1561 "subvol '%s' does not match subvolid %llu",
1562 subvol_name, subvol_objectid);
1563 ret = -EINVAL;
1564 }
1565 if (ret) {
1566 dput(root);
1567 root = ERR_PTR(ret);
1568 deactivate_locked_super(s);
1569 }
1570 }
1571
1572out:
1573 mntput(mnt);
1574 kfree(subvol_name);
1575 return root;
1576}
1577
1578/*
1579 * Find a superblock for the given device / mount point.
1580 *
1581 * Note: This is based on mount_bdev from fs/super.c with a few additions
1582 * for multiple device setup. Make sure to keep it in sync.
1583 */
1584static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1585 int flags, const char *device_name, void *data)
1586{
1587 struct block_device *bdev = NULL;
1588 struct super_block *s;
1589 struct btrfs_device *device = NULL;
1590 struct btrfs_fs_devices *fs_devices = NULL;
1591 struct btrfs_fs_info *fs_info = NULL;
1592 void *new_sec_opts = NULL;
1593 fmode_t mode = FMODE_READ;
1594 int error = 0;
1595
1596 if (!(flags & SB_RDONLY))
1597 mode |= FMODE_WRITE;
1598
1599 if (data) {
1600 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1601 if (error)
1602 return ERR_PTR(error);
1603 }
1604
1605 /*
1606 * Setup a dummy root and fs_info for test/set super. This is because
1607 * we don't actually fill this stuff out until open_ctree, but we need
1608 * then open_ctree will properly initialize the file system specific
1609 * settings later. btrfs_init_fs_info initializes the static elements
1610 * of the fs_info (locks and such) to make cleanup easier if we find a
1611 * superblock with our given fs_devices later on at sget() time.
1612 */
1613 fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1614 if (!fs_info) {
1615 error = -ENOMEM;
1616 goto error_sec_opts;
1617 }
1618 btrfs_init_fs_info(fs_info);
1619
1620 fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1621 fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1622 if (!fs_info->super_copy || !fs_info->super_for_commit) {
1623 error = -ENOMEM;
1624 goto error_fs_info;
1625 }
1626
1627 mutex_lock(&uuid_mutex);
1628 error = btrfs_parse_device_options(data, mode, fs_type);
1629 if (error) {
1630 mutex_unlock(&uuid_mutex);
1631 goto error_fs_info;
1632 }
1633
1634 device = btrfs_scan_one_device(device_name, mode, fs_type);
1635 if (IS_ERR(device)) {
1636 mutex_unlock(&uuid_mutex);
1637 error = PTR_ERR(device);
1638 goto error_fs_info;
1639 }
1640
1641 fs_devices = device->fs_devices;
1642 fs_info->fs_devices = fs_devices;
1643
1644 error = btrfs_open_devices(fs_devices, mode, fs_type);
1645 mutex_unlock(&uuid_mutex);
1646 if (error)
1647 goto error_fs_info;
1648
1649 if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1650 error = -EACCES;
1651 goto error_close_devices;
1652 }
1653
1654 bdev = fs_devices->latest_bdev;
1655 s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1656 fs_info);
1657 if (IS_ERR(s)) {
1658 error = PTR_ERR(s);
1659 goto error_close_devices;
1660 }
1661
1662 if (s->s_root) {
1663 btrfs_close_devices(fs_devices);
1664 btrfs_free_fs_info(fs_info);
1665 if ((flags ^ s->s_flags) & SB_RDONLY)
1666 error = -EBUSY;
1667 } else {
1668 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1669 btrfs_sb(s)->bdev_holder = fs_type;
1670 if (!strstr(crc32c_impl(), "generic"))
1671 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1672 error = btrfs_fill_super(s, fs_devices, data);
1673 }
1674 if (!error)
1675 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1676 security_free_mnt_opts(&new_sec_opts);
1677 if (error) {
1678 deactivate_locked_super(s);
1679 return ERR_PTR(error);
1680 }
1681
1682 return dget(s->s_root);
1683
1684error_close_devices:
1685 btrfs_close_devices(fs_devices);
1686error_fs_info:
1687 btrfs_free_fs_info(fs_info);
1688error_sec_opts:
1689 security_free_mnt_opts(&new_sec_opts);
1690 return ERR_PTR(error);
1691}
1692
1693/*
1694 * Mount function which is called by VFS layer.
1695 *
1696 * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1697 * which needs vfsmount* of device's root (/). This means device's root has to
1698 * be mounted internally in any case.
1699 *
1700 * Operation flow:
1701 * 1. Parse subvol id related options for later use in mount_subvol().
1702 *
1703 * 2. Mount device's root (/) by calling vfs_kern_mount().
1704 *
1705 * NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1706 * first place. In order to avoid calling btrfs_mount() again, we use
1707 * different file_system_type which is not registered to VFS by
1708 * register_filesystem() (btrfs_root_fs_type). As a result,
1709 * btrfs_mount_root() is called. The return value will be used by
1710 * mount_subtree() in mount_subvol().
1711 *
1712 * 3. Call mount_subvol() to get the dentry of subvolume. Since there is
1713 * "btrfs subvolume set-default", mount_subvol() is called always.
1714 */
1715static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1716 const char *device_name, void *data)
1717{
1718 struct vfsmount *mnt_root;
1719 struct dentry *root;
1720 char *subvol_name = NULL;
1721 u64 subvol_objectid = 0;
1722 int error = 0;
1723
1724 error = btrfs_parse_subvol_options(data, &subvol_name,
1725 &subvol_objectid);
1726 if (error) {
1727 kfree(subvol_name);
1728 return ERR_PTR(error);
1729 }
1730
1731 /* mount device's root (/) */
1732 mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1733 if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1734 if (flags & SB_RDONLY) {
1735 mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1736 flags & ~SB_RDONLY, device_name, data);
1737 } else {
1738 mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1739 flags | SB_RDONLY, device_name, data);
1740 if (IS_ERR(mnt_root)) {
1741 root = ERR_CAST(mnt_root);
1742 kfree(subvol_name);
1743 goto out;
1744 }
1745
1746 down_write(&mnt_root->mnt_sb->s_umount);
1747 error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1748 up_write(&mnt_root->mnt_sb->s_umount);
1749 if (error < 0) {
1750 root = ERR_PTR(error);
1751 mntput(mnt_root);
1752 kfree(subvol_name);
1753 goto out;
1754 }
1755 }
1756 }
1757 if (IS_ERR(mnt_root)) {
1758 root = ERR_CAST(mnt_root);
1759 kfree(subvol_name);
1760 goto out;
1761 }
1762
1763 /* mount_subvol() will free subvol_name and mnt_root */
1764 root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1765
1766out:
1767 return root;
1768}
1769
1770static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1771 u32 new_pool_size, u32 old_pool_size)
1772{
1773 if (new_pool_size == old_pool_size)
1774 return;
1775
1776 fs_info->thread_pool_size = new_pool_size;
1777
1778 btrfs_info(fs_info, "resize thread pool %d -> %d",
1779 old_pool_size, new_pool_size);
1780
1781 btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1782 btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1783 btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1784 btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1785 btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1786 btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1787 new_pool_size);
1788 btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1789 btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1790 btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1791 btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1792 btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1793 new_pool_size);
1794}
1795
1796static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1797 unsigned long old_opts, int flags)
1798{
1799 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1800 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1801 (flags & SB_RDONLY))) {
1802 /* wait for any defraggers to finish */
1803 wait_event(fs_info->transaction_wait,
1804 (atomic_read(&fs_info->defrag_running) == 0));
1805 if (flags & SB_RDONLY)
1806 sync_filesystem(fs_info->sb);
1807 }
1808}
1809
1810static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1811 unsigned long old_opts)
1812{
1813 /*
1814 * We need to cleanup all defragable inodes if the autodefragment is
1815 * close or the filesystem is read only.
1816 */
1817 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1818 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1819 btrfs_cleanup_defrag_inodes(fs_info);
1820 }
1821
1822 /* If we toggled discard async */
1823 if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1824 btrfs_test_opt(fs_info, DISCARD_ASYNC))
1825 btrfs_discard_resume(fs_info);
1826 else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1827 !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1828 btrfs_discard_cleanup(fs_info);
1829}
1830
1831static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1832{
1833 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1834 struct btrfs_root *root = fs_info->tree_root;
1835 unsigned old_flags = sb->s_flags;
1836 unsigned long old_opts = fs_info->mount_opt;
1837 unsigned long old_compress_type = fs_info->compress_type;
1838 u64 old_max_inline = fs_info->max_inline;
1839 u32 old_thread_pool_size = fs_info->thread_pool_size;
1840 u32 old_metadata_ratio = fs_info->metadata_ratio;
1841 int ret;
1842
1843 sync_filesystem(sb);
1844 set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1845
1846 if (data) {
1847 void *new_sec_opts = NULL;
1848
1849 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
1850 if (!ret)
1851 ret = security_sb_remount(sb, new_sec_opts);
1852 security_free_mnt_opts(&new_sec_opts);
1853 if (ret)
1854 goto restore;
1855 }
1856
1857 ret = btrfs_parse_options(fs_info, data, *flags);
1858 if (ret)
1859 goto restore;
1860
1861 btrfs_remount_begin(fs_info, old_opts, *flags);
1862 btrfs_resize_thread_pool(fs_info,
1863 fs_info->thread_pool_size, old_thread_pool_size);
1864
1865 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1866 goto out;
1867
1868 if (*flags & SB_RDONLY) {
1869 /*
1870 * this also happens on 'umount -rf' or on shutdown, when
1871 * the filesystem is busy.
1872 */
1873 cancel_work_sync(&fs_info->async_reclaim_work);
1874
1875 btrfs_discard_cleanup(fs_info);
1876
1877 /* wait for the uuid_scan task to finish */
1878 down(&fs_info->uuid_tree_rescan_sem);
1879 /* avoid complains from lockdep et al. */
1880 up(&fs_info->uuid_tree_rescan_sem);
1881
1882 sb->s_flags |= SB_RDONLY;
1883
1884 /*
1885 * Setting SB_RDONLY will put the cleaner thread to
1886 * sleep at the next loop if it's already active.
1887 * If it's already asleep, we'll leave unused block
1888 * groups on disk until we're mounted read-write again
1889 * unless we clean them up here.
1890 */
1891 btrfs_delete_unused_bgs(fs_info);
1892
1893 btrfs_dev_replace_suspend_for_unmount(fs_info);
1894 btrfs_scrub_cancel(fs_info);
1895 btrfs_pause_balance(fs_info);
1896
1897 ret = btrfs_commit_super(fs_info);
1898 if (ret)
1899 goto restore;
1900 } else {
1901 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1902 btrfs_err(fs_info,
1903 "Remounting read-write after error is not allowed");
1904 ret = -EINVAL;
1905 goto restore;
1906 }
1907 if (fs_info->fs_devices->rw_devices == 0) {
1908 ret = -EACCES;
1909 goto restore;
1910 }
1911
1912 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1913 btrfs_warn(fs_info,
1914 "too many missing devices, writable remount is not allowed");
1915 ret = -EACCES;
1916 goto restore;
1917 }
1918
1919 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1920 btrfs_warn(fs_info,
1921 "mount required to replay tree-log, cannot remount read-write");
1922 ret = -EINVAL;
1923 goto restore;
1924 }
1925
1926 ret = btrfs_cleanup_fs_roots(fs_info);
1927 if (ret)
1928 goto restore;
1929
1930 /* recover relocation */
1931 mutex_lock(&fs_info->cleaner_mutex);
1932 ret = btrfs_recover_relocation(root);
1933 mutex_unlock(&fs_info->cleaner_mutex);
1934 if (ret)
1935 goto restore;
1936
1937 ret = btrfs_resume_balance_async(fs_info);
1938 if (ret)
1939 goto restore;
1940
1941 ret = btrfs_resume_dev_replace_async(fs_info);
1942 if (ret) {
1943 btrfs_warn(fs_info, "failed to resume dev_replace");
1944 goto restore;
1945 }
1946
1947 btrfs_qgroup_rescan_resume(fs_info);
1948
1949 if (!fs_info->uuid_root) {
1950 btrfs_info(fs_info, "creating UUID tree");
1951 ret = btrfs_create_uuid_tree(fs_info);
1952 if (ret) {
1953 btrfs_warn(fs_info,
1954 "failed to create the UUID tree %d",
1955 ret);
1956 goto restore;
1957 }
1958 }
1959 sb->s_flags &= ~SB_RDONLY;
1960
1961 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1962 }
1963out:
1964 /*
1965 * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
1966 * since the absence of the flag means it can be toggled off by remount.
1967 */
1968 *flags |= SB_I_VERSION;
1969
1970 wake_up_process(fs_info->transaction_kthread);
1971 btrfs_remount_cleanup(fs_info, old_opts);
1972 clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1973
1974 return 0;
1975
1976restore:
1977 /* We've hit an error - don't reset SB_RDONLY */
1978 if (sb_rdonly(sb))
1979 old_flags |= SB_RDONLY;
1980 sb->s_flags = old_flags;
1981 fs_info->mount_opt = old_opts;
1982 fs_info->compress_type = old_compress_type;
1983 fs_info->max_inline = old_max_inline;
1984 btrfs_resize_thread_pool(fs_info,
1985 old_thread_pool_size, fs_info->thread_pool_size);
1986 fs_info->metadata_ratio = old_metadata_ratio;
1987 btrfs_remount_cleanup(fs_info, old_opts);
1988 clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1989
1990 return ret;
1991}
1992
1993/* Used to sort the devices by max_avail(descending sort) */
1994static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
1995 const void *dev_info2)
1996{
1997 if (((struct btrfs_device_info *)dev_info1)->max_avail >
1998 ((struct btrfs_device_info *)dev_info2)->max_avail)
1999 return -1;
2000 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2001 ((struct btrfs_device_info *)dev_info2)->max_avail)
2002 return 1;
2003 else
2004 return 0;
2005}
2006
2007/*
2008 * sort the devices by max_avail, in which max free extent size of each device
2009 * is stored.(Descending Sort)
2010 */
2011static inline void btrfs_descending_sort_devices(
2012 struct btrfs_device_info *devices,
2013 size_t nr_devices)
2014{
2015 sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2016 btrfs_cmp_device_free_bytes, NULL);
2017}
2018
2019/*
2020 * The helper to calc the free space on the devices that can be used to store
2021 * file data.
2022 */
2023static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2024 u64 *free_bytes)
2025{
2026 struct btrfs_device_info *devices_info;
2027 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2028 struct btrfs_device *device;
2029 u64 type;
2030 u64 avail_space;
2031 u64 min_stripe_size;
2032 int num_stripes = 1;
2033 int i = 0, nr_devices;
2034 const struct btrfs_raid_attr *rattr;
2035
2036 /*
2037 * We aren't under the device list lock, so this is racy-ish, but good
2038 * enough for our purposes.
2039 */
2040 nr_devices = fs_info->fs_devices->open_devices;
2041 if (!nr_devices) {
2042 smp_mb();
2043 nr_devices = fs_info->fs_devices->open_devices;
2044 ASSERT(nr_devices);
2045 if (!nr_devices) {
2046 *free_bytes = 0;
2047 return 0;
2048 }
2049 }
2050
2051 devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2052 GFP_KERNEL);
2053 if (!devices_info)
2054 return -ENOMEM;
2055
2056 /* calc min stripe number for data space allocation */
2057 type = btrfs_data_alloc_profile(fs_info);
2058 rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2059
2060 if (type & BTRFS_BLOCK_GROUP_RAID0)
2061 num_stripes = nr_devices;
2062 else if (type & BTRFS_BLOCK_GROUP_RAID1)
2063 num_stripes = 2;
2064 else if (type & BTRFS_BLOCK_GROUP_RAID1C3)
2065 num_stripes = 3;
2066 else if (type & BTRFS_BLOCK_GROUP_RAID1C4)
2067 num_stripes = 4;
2068 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2069 num_stripes = 4;
2070
2071 /* Adjust for more than 1 stripe per device */
2072 min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2073
2074 rcu_read_lock();
2075 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2076 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2077 &device->dev_state) ||
2078 !device->bdev ||
2079 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2080 continue;
2081
2082 if (i >= nr_devices)
2083 break;
2084
2085 avail_space = device->total_bytes - device->bytes_used;
2086
2087 /* align with stripe_len */
2088 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2089
2090 /*
2091 * In order to avoid overwriting the superblock on the drive,
2092 * btrfs starts at an offset of at least 1MB when doing chunk
2093 * allocation.
2094 *
2095 * This ensures we have at least min_stripe_size free space
2096 * after excluding 1MB.
2097 */
2098 if (avail_space <= SZ_1M + min_stripe_size)
2099 continue;
2100
2101 avail_space -= SZ_1M;
2102
2103 devices_info[i].dev = device;
2104 devices_info[i].max_avail = avail_space;
2105
2106 i++;
2107 }
2108 rcu_read_unlock();
2109
2110 nr_devices = i;
2111
2112 btrfs_descending_sort_devices(devices_info, nr_devices);
2113
2114 i = nr_devices - 1;
2115 avail_space = 0;
2116 while (nr_devices >= rattr->devs_min) {
2117 num_stripes = min(num_stripes, nr_devices);
2118
2119 if (devices_info[i].max_avail >= min_stripe_size) {
2120 int j;
2121 u64 alloc_size;
2122
2123 avail_space += devices_info[i].max_avail * num_stripes;
2124 alloc_size = devices_info[i].max_avail;
2125 for (j = i + 1 - num_stripes; j <= i; j++)
2126 devices_info[j].max_avail -= alloc_size;
2127 }
2128 i--;
2129 nr_devices--;
2130 }
2131
2132 kfree(devices_info);
2133 *free_bytes = avail_space;
2134 return 0;
2135}
2136
2137/*
2138 * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2139 *
2140 * If there's a redundant raid level at DATA block groups, use the respective
2141 * multiplier to scale the sizes.
2142 *
2143 * Unused device space usage is based on simulating the chunk allocator
2144 * algorithm that respects the device sizes and order of allocations. This is
2145 * a close approximation of the actual use but there are other factors that may
2146 * change the result (like a new metadata chunk).
2147 *
2148 * If metadata is exhausted, f_bavail will be 0.
2149 */
2150static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2151{
2152 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2153 struct btrfs_super_block *disk_super = fs_info->super_copy;
2154 struct btrfs_space_info *found;
2155 u64 total_used = 0;
2156 u64 total_free_data = 0;
2157 u64 total_free_meta = 0;
2158 int bits = dentry->d_sb->s_blocksize_bits;
2159 __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2160 unsigned factor = 1;
2161 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2162 int ret;
2163 u64 thresh = 0;
2164 int mixed = 0;
2165
2166 rcu_read_lock();
2167 list_for_each_entry_rcu(found, &fs_info->space_info, list) {
2168 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2169 int i;
2170
2171 total_free_data += found->disk_total - found->disk_used;
2172 total_free_data -=
2173 btrfs_account_ro_block_groups_free_space(found);
2174
2175 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2176 if (!list_empty(&found->block_groups[i]))
2177 factor = btrfs_bg_type_to_factor(
2178 btrfs_raid_array[i].bg_flag);
2179 }
2180 }
2181
2182 /*
2183 * Metadata in mixed block goup profiles are accounted in data
2184 */
2185 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2186 if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2187 mixed = 1;
2188 else
2189 total_free_meta += found->disk_total -
2190 found->disk_used;
2191 }
2192
2193 total_used += found->disk_used;
2194 }
2195
2196 rcu_read_unlock();
2197
2198 buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2199 buf->f_blocks >>= bits;
2200 buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2201
2202 /* Account global block reserve as used, it's in logical size already */
2203 spin_lock(&block_rsv->lock);
2204 /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2205 if (buf->f_bfree >= block_rsv->size >> bits)
2206 buf->f_bfree -= block_rsv->size >> bits;
2207 else
2208 buf->f_bfree = 0;
2209 spin_unlock(&block_rsv->lock);
2210
2211 buf->f_bavail = div_u64(total_free_data, factor);
2212 ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2213 if (ret)
2214 return ret;
2215 buf->f_bavail += div_u64(total_free_data, factor);
2216 buf->f_bavail = buf->f_bavail >> bits;
2217
2218 /*
2219 * We calculate the remaining metadata space minus global reserve. If
2220 * this is (supposedly) smaller than zero, there's no space. But this
2221 * does not hold in practice, the exhausted state happens where's still
2222 * some positive delta. So we apply some guesswork and compare the
2223 * delta to a 4M threshold. (Practically observed delta was ~2M.)
2224 *
2225 * We probably cannot calculate the exact threshold value because this
2226 * depends on the internal reservations requested by various
2227 * operations, so some operations that consume a few metadata will
2228 * succeed even if the Avail is zero. But this is better than the other
2229 * way around.
2230 */
2231 thresh = SZ_4M;
2232
2233 /*
2234 * We only want to claim there's no available space if we can no longer
2235 * allocate chunks for our metadata profile and our global reserve will
2236 * not fit in the free metadata space. If we aren't ->full then we
2237 * still can allocate chunks and thus are fine using the currently
2238 * calculated f_bavail.
2239 */
2240 if (!mixed && block_rsv->space_info->full &&
2241 total_free_meta - thresh < block_rsv->size)
2242 buf->f_bavail = 0;
2243
2244 buf->f_type = BTRFS_SUPER_MAGIC;
2245 buf->f_bsize = dentry->d_sb->s_blocksize;
2246 buf->f_namelen = BTRFS_NAME_LEN;
2247
2248 /* We treat it as constant endianness (it doesn't matter _which_)
2249 because we want the fsid to come out the same whether mounted
2250 on a big-endian or little-endian host */
2251 buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2252 buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2253 /* Mask in the root object ID too, to disambiguate subvols */
2254 buf->f_fsid.val[0] ^=
2255 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2256 buf->f_fsid.val[1] ^=
2257 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2258
2259 return 0;
2260}
2261
2262static void btrfs_kill_super(struct super_block *sb)
2263{
2264 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2265 kill_anon_super(sb);
2266 btrfs_free_fs_info(fs_info);
2267}
2268
2269static struct file_system_type btrfs_fs_type = {
2270 .owner = THIS_MODULE,
2271 .name = "btrfs",
2272 .mount = btrfs_mount,
2273 .kill_sb = btrfs_kill_super,
2274 .fs_flags = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2275};
2276
2277static struct file_system_type btrfs_root_fs_type = {
2278 .owner = THIS_MODULE,
2279 .name = "btrfs",
2280 .mount = btrfs_mount_root,
2281 .kill_sb = btrfs_kill_super,
2282 .fs_flags = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2283};
2284
2285MODULE_ALIAS_FS("btrfs");
2286
2287static int btrfs_control_open(struct inode *inode, struct file *file)
2288{
2289 /*
2290 * The control file's private_data is used to hold the
2291 * transaction when it is started and is used to keep
2292 * track of whether a transaction is already in progress.
2293 */
2294 file->private_data = NULL;
2295 return 0;
2296}
2297
2298/*
2299 * Used by /dev/btrfs-control for devices ioctls.
2300 */
2301static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2302 unsigned long arg)
2303{
2304 struct btrfs_ioctl_vol_args *vol;
2305 struct btrfs_device *device = NULL;
2306 int ret = -ENOTTY;
2307
2308 if (!capable(CAP_SYS_ADMIN))
2309 return -EPERM;
2310
2311 vol = memdup_user((void __user *)arg, sizeof(*vol));
2312 if (IS_ERR(vol))
2313 return PTR_ERR(vol);
2314 vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2315
2316 switch (cmd) {
2317 case BTRFS_IOC_SCAN_DEV:
2318 mutex_lock(&uuid_mutex);
2319 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2320 &btrfs_root_fs_type);
2321 ret = PTR_ERR_OR_ZERO(device);
2322 mutex_unlock(&uuid_mutex);
2323 break;
2324 case BTRFS_IOC_FORGET_DEV:
2325 ret = btrfs_forget_devices(vol->name);
2326 break;
2327 case BTRFS_IOC_DEVICES_READY:
2328 mutex_lock(&uuid_mutex);
2329 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2330 &btrfs_root_fs_type);
2331 if (IS_ERR(device)) {
2332 mutex_unlock(&uuid_mutex);
2333 ret = PTR_ERR(device);
2334 break;
2335 }
2336 ret = !(device->fs_devices->num_devices ==
2337 device->fs_devices->total_devices);
2338 mutex_unlock(&uuid_mutex);
2339 break;
2340 case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2341 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2342 break;
2343 }
2344
2345 kfree(vol);
2346 return ret;
2347}
2348
2349static int btrfs_freeze(struct super_block *sb)
2350{
2351 struct btrfs_trans_handle *trans;
2352 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2353 struct btrfs_root *root = fs_info->tree_root;
2354
2355 set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2356 /*
2357 * We don't need a barrier here, we'll wait for any transaction that
2358 * could be in progress on other threads (and do delayed iputs that
2359 * we want to avoid on a frozen filesystem), or do the commit
2360 * ourselves.
2361 */
2362 trans = btrfs_attach_transaction_barrier(root);
2363 if (IS_ERR(trans)) {
2364 /* no transaction, don't bother */
2365 if (PTR_ERR(trans) == -ENOENT)
2366 return 0;
2367 return PTR_ERR(trans);
2368 }
2369 return btrfs_commit_transaction(trans);
2370}
2371
2372static int btrfs_unfreeze(struct super_block *sb)
2373{
2374 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2375
2376 clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2377 return 0;
2378}
2379
2380static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2381{
2382 struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2383 struct btrfs_device *dev, *first_dev = NULL;
2384
2385 /*
2386 * Lightweight locking of the devices. We should not need
2387 * device_list_mutex here as we only read the device data and the list
2388 * is protected by RCU. Even if a device is deleted during the list
2389 * traversals, we'll get valid data, the freeing callback will wait at
2390 * least until the rcu_read_unlock.
2391 */
2392 rcu_read_lock();
2393 list_for_each_entry_rcu(dev, &fs_info->fs_devices->devices, dev_list) {
2394 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2395 continue;
2396 if (!dev->name)
2397 continue;
2398 if (!first_dev || dev->devid < first_dev->devid)
2399 first_dev = dev;
2400 }
2401
2402 if (first_dev)
2403 seq_escape(m, rcu_str_deref(first_dev->name), " \t\n\\");
2404 else
2405 WARN_ON(1);
2406 rcu_read_unlock();
2407 return 0;
2408}
2409
2410static const struct super_operations btrfs_super_ops = {
2411 .drop_inode = btrfs_drop_inode,
2412 .evict_inode = btrfs_evict_inode,
2413 .put_super = btrfs_put_super,
2414 .sync_fs = btrfs_sync_fs,
2415 .show_options = btrfs_show_options,
2416 .show_devname = btrfs_show_devname,
2417 .alloc_inode = btrfs_alloc_inode,
2418 .destroy_inode = btrfs_destroy_inode,
2419 .free_inode = btrfs_free_inode,
2420 .statfs = btrfs_statfs,
2421 .remount_fs = btrfs_remount,
2422 .freeze_fs = btrfs_freeze,
2423 .unfreeze_fs = btrfs_unfreeze,
2424};
2425
2426static const struct file_operations btrfs_ctl_fops = {
2427 .open = btrfs_control_open,
2428 .unlocked_ioctl = btrfs_control_ioctl,
2429 .compat_ioctl = compat_ptr_ioctl,
2430 .owner = THIS_MODULE,
2431 .llseek = noop_llseek,
2432};
2433
2434static struct miscdevice btrfs_misc = {
2435 .minor = BTRFS_MINOR,
2436 .name = "btrfs-control",
2437 .fops = &btrfs_ctl_fops
2438};
2439
2440MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2441MODULE_ALIAS("devname:btrfs-control");
2442
2443static int __init btrfs_interface_init(void)
2444{
2445 return misc_register(&btrfs_misc);
2446}
2447
2448static __cold void btrfs_interface_exit(void)
2449{
2450 misc_deregister(&btrfs_misc);
2451}
2452
2453static void __init btrfs_print_mod_info(void)
2454{
2455 static const char options[] = ""
2456#ifdef CONFIG_BTRFS_DEBUG
2457 ", debug=on"
2458#endif
2459#ifdef CONFIG_BTRFS_ASSERT
2460 ", assert=on"
2461#endif
2462#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2463 ", integrity-checker=on"
2464#endif
2465#ifdef CONFIG_BTRFS_FS_REF_VERIFY
2466 ", ref-verify=on"
2467#endif
2468 ;
2469 pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2470}
2471
2472static int __init init_btrfs_fs(void)
2473{
2474 int err;
2475
2476 btrfs_props_init();
2477
2478 err = btrfs_init_sysfs();
2479 if (err)
2480 return err;
2481
2482 btrfs_init_compress();
2483
2484 err = btrfs_init_cachep();
2485 if (err)
2486 goto free_compress;
2487
2488 err = extent_io_init();
2489 if (err)
2490 goto free_cachep;
2491
2492 err = extent_state_cache_init();
2493 if (err)
2494 goto free_extent_io;
2495
2496 err = extent_map_init();
2497 if (err)
2498 goto free_extent_state_cache;
2499
2500 err = ordered_data_init();
2501 if (err)
2502 goto free_extent_map;
2503
2504 err = btrfs_delayed_inode_init();
2505 if (err)
2506 goto free_ordered_data;
2507
2508 err = btrfs_auto_defrag_init();
2509 if (err)
2510 goto free_delayed_inode;
2511
2512 err = btrfs_delayed_ref_init();
2513 if (err)
2514 goto free_auto_defrag;
2515
2516 err = btrfs_prelim_ref_init();
2517 if (err)
2518 goto free_delayed_ref;
2519
2520 err = btrfs_end_io_wq_init();
2521 if (err)
2522 goto free_prelim_ref;
2523
2524 err = btrfs_interface_init();
2525 if (err)
2526 goto free_end_io_wq;
2527
2528 btrfs_init_lockdep();
2529
2530 btrfs_print_mod_info();
2531
2532 err = btrfs_run_sanity_tests();
2533 if (err)
2534 goto unregister_ioctl;
2535
2536 err = register_filesystem(&btrfs_fs_type);
2537 if (err)
2538 goto unregister_ioctl;
2539
2540 return 0;
2541
2542unregister_ioctl:
2543 btrfs_interface_exit();
2544free_end_io_wq:
2545 btrfs_end_io_wq_exit();
2546free_prelim_ref:
2547 btrfs_prelim_ref_exit();
2548free_delayed_ref:
2549 btrfs_delayed_ref_exit();
2550free_auto_defrag:
2551 btrfs_auto_defrag_exit();
2552free_delayed_inode:
2553 btrfs_delayed_inode_exit();
2554free_ordered_data:
2555 ordered_data_exit();
2556free_extent_map:
2557 extent_map_exit();
2558free_extent_state_cache:
2559 extent_state_cache_exit();
2560free_extent_io:
2561 extent_io_exit();
2562free_cachep:
2563 btrfs_destroy_cachep();
2564free_compress:
2565 btrfs_exit_compress();
2566 btrfs_exit_sysfs();
2567
2568 return err;
2569}
2570
2571static void __exit exit_btrfs_fs(void)
2572{
2573 btrfs_destroy_cachep();
2574 btrfs_delayed_ref_exit();
2575 btrfs_auto_defrag_exit();
2576 btrfs_delayed_inode_exit();
2577 btrfs_prelim_ref_exit();
2578 ordered_data_exit();
2579 extent_map_exit();
2580 extent_state_cache_exit();
2581 extent_io_exit();
2582 btrfs_interface_exit();
2583 btrfs_end_io_wq_exit();
2584 unregister_filesystem(&btrfs_fs_type);
2585 btrfs_exit_sysfs();
2586 btrfs_cleanup_fs_uuids();
2587 btrfs_exit_compress();
2588}
2589
2590late_initcall(init_btrfs_fs);
2591module_exit(exit_btrfs_fs)
2592
2593MODULE_LICENSE("GPL");
2594MODULE_SOFTDEP("pre: crc32c");
2595MODULE_SOFTDEP("pre: xxhash64");
2596MODULE_SOFTDEP("pre: sha256");
2597MODULE_SOFTDEP("pre: blake2b-256");
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
100static void save_error_info(struct btrfs_fs_info *fs_info)
101{
102 /*
103 * today we only save the error info into ram. Long term we'll
104 * also send it down to the disk
105 */
106 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
107}
108
109/* btrfs handle error by forcing the filesystem readonly */
110static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
111{
112 struct super_block *sb = fs_info->sb;
113
114 if (sb->s_flags & MS_RDONLY)
115 return;
116
117 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
118 sb->s_flags |= MS_RDONLY;
119 btrfs_info(fs_info, "forced readonly");
120 /*
121 * Note that a running device replace operation is not
122 * canceled here although there is no way to update
123 * the progress. It would add the risk of a deadlock,
124 * therefore the canceling is ommited. The only penalty
125 * is that some I/O remains active until the procedure
126 * completes. The next time when the filesystem is
127 * mounted writeable again, the device replace
128 * operation continues.
129 */
130 }
131}
132
133/*
134 * __btrfs_std_error decodes expected errors from the caller and
135 * invokes the approciate error response.
136 */
137__cold
138void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
139 unsigned int line, int errno, const char *fmt, ...)
140{
141 struct super_block *sb = fs_info->sb;
142#ifdef CONFIG_PRINTK
143 const char *errstr;
144#endif
145
146 /*
147 * Special case: if the error is EROFS, and we're already
148 * under MS_RDONLY, then it is safe here.
149 */
150 if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
151 return;
152
153#ifdef CONFIG_PRINTK
154 errstr = btrfs_decode_error(errno);
155 if (fmt) {
156 struct va_format vaf;
157 va_list args;
158
159 va_start(args, fmt);
160 vaf.fmt = fmt;
161 vaf.va = &args;
162
163 printk(KERN_CRIT
164 "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
165 sb->s_id, function, line, errno, errstr, &vaf);
166 va_end(args);
167 } else {
168 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
169 sb->s_id, function, line, errno, errstr);
170 }
171#endif
172
173 /* Don't go through full error handling during mount */
174 save_error_info(fs_info);
175 if (sb->s_flags & MS_BORN)
176 btrfs_handle_error(fs_info);
177}
178
179#ifdef CONFIG_PRINTK
180static const char * const logtypes[] = {
181 "emergency",
182 "alert",
183 "critical",
184 "error",
185 "warning",
186 "notice",
187 "info",
188 "debug",
189};
190
191void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
192{
193 struct super_block *sb = fs_info->sb;
194 char lvl[4];
195 struct va_format vaf;
196 va_list args;
197 const char *type = logtypes[4];
198 int kern_level;
199
200 va_start(args, fmt);
201
202 kern_level = printk_get_level(fmt);
203 if (kern_level) {
204 size_t size = printk_skip_level(fmt) - fmt;
205 memcpy(lvl, fmt, size);
206 lvl[size] = '\0';
207 fmt += size;
208 type = logtypes[kern_level - '0'];
209 } else
210 *lvl = '\0';
211
212 vaf.fmt = fmt;
213 vaf.va = &args;
214
215 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
216
217 va_end(args);
218}
219#endif
220
221/*
222 * We only mark the transaction aborted and then set the file system read-only.
223 * This will prevent new transactions from starting or trying to join this
224 * one.
225 *
226 * This means that error recovery at the call site is limited to freeing
227 * any local memory allocations and passing the error code up without
228 * further cleanup. The transaction should complete as it normally would
229 * in the call path but will return -EIO.
230 *
231 * We'll complete the cleanup in btrfs_end_transaction and
232 * btrfs_commit_transaction.
233 */
234__cold
235void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
236 struct btrfs_root *root, const char *function,
237 unsigned int line, int errno)
238{
239 trans->aborted = errno;
240 /* Nothing used. The other threads that have joined this
241 * transaction may be able to continue. */
242 if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
243 const char *errstr;
244
245 errstr = btrfs_decode_error(errno);
246 btrfs_warn(root->fs_info,
247 "%s:%d: Aborting unused transaction(%s).",
248 function, line, errstr);
249 return;
250 }
251 ACCESS_ONCE(trans->transaction->aborted) = errno;
252 /* Wake up anybody who may be waiting on this transaction */
253 wake_up(&root->fs_info->transaction_wait);
254 wake_up(&root->fs_info->transaction_blocked_wait);
255 __btrfs_std_error(root->fs_info, function, line, errno, NULL);
256}
257/*
258 * __btrfs_panic decodes unexpected, fatal errors from the caller,
259 * issues an alert, and either panics or BUGs, depending on mount options.
260 */
261__cold
262void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
263 unsigned int line, int errno, const char *fmt, ...)
264{
265 char *s_id = "<unknown>";
266 const char *errstr;
267 struct va_format vaf = { .fmt = fmt };
268 va_list args;
269
270 if (fs_info)
271 s_id = fs_info->sb->s_id;
272
273 va_start(args, fmt);
274 vaf.va = &args;
275
276 errstr = btrfs_decode_error(errno);
277 if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
278 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
279 s_id, function, line, &vaf, errno, errstr);
280
281 btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
282 function, line, &vaf, errno, errstr);
283 va_end(args);
284 /* Caller calls BUG() */
285}
286
287static void btrfs_put_super(struct super_block *sb)
288{
289 close_ctree(btrfs_sb(sb)->tree_root);
290}
291
292enum {
293 Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
294 Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
295 Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
296 Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
297 Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
298 Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
299 Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
300 Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
301 Opt_skip_balance, Opt_check_integrity,
302 Opt_check_integrity_including_extent_data,
303 Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
304 Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
305 Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
306 Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
307 Opt_nologreplay, Opt_norecovery,
308#ifdef CONFIG_BTRFS_DEBUG
309 Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
310#endif
311 Opt_err,
312};
313
314static const match_table_t tokens = {
315 {Opt_degraded, "degraded"},
316 {Opt_subvol, "subvol=%s"},
317 {Opt_subvolid, "subvolid=%s"},
318 {Opt_device, "device=%s"},
319 {Opt_nodatasum, "nodatasum"},
320 {Opt_datasum, "datasum"},
321 {Opt_nodatacow, "nodatacow"},
322 {Opt_datacow, "datacow"},
323 {Opt_nobarrier, "nobarrier"},
324 {Opt_barrier, "barrier"},
325 {Opt_max_inline, "max_inline=%s"},
326 {Opt_alloc_start, "alloc_start=%s"},
327 {Opt_thread_pool, "thread_pool=%d"},
328 {Opt_compress, "compress"},
329 {Opt_compress_type, "compress=%s"},
330 {Opt_compress_force, "compress-force"},
331 {Opt_compress_force_type, "compress-force=%s"},
332 {Opt_ssd, "ssd"},
333 {Opt_ssd_spread, "ssd_spread"},
334 {Opt_nossd, "nossd"},
335 {Opt_acl, "acl"},
336 {Opt_noacl, "noacl"},
337 {Opt_notreelog, "notreelog"},
338 {Opt_treelog, "treelog"},
339 {Opt_nologreplay, "nologreplay"},
340 {Opt_norecovery, "norecovery"},
341 {Opt_flushoncommit, "flushoncommit"},
342 {Opt_noflushoncommit, "noflushoncommit"},
343 {Opt_ratio, "metadata_ratio=%d"},
344 {Opt_discard, "discard"},
345 {Opt_nodiscard, "nodiscard"},
346 {Opt_space_cache, "space_cache"},
347 {Opt_space_cache_version, "space_cache=%s"},
348 {Opt_clear_cache, "clear_cache"},
349 {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
350 {Opt_enospc_debug, "enospc_debug"},
351 {Opt_noenospc_debug, "noenospc_debug"},
352 {Opt_subvolrootid, "subvolrootid=%d"},
353 {Opt_defrag, "autodefrag"},
354 {Opt_nodefrag, "noautodefrag"},
355 {Opt_inode_cache, "inode_cache"},
356 {Opt_noinode_cache, "noinode_cache"},
357 {Opt_no_space_cache, "nospace_cache"},
358 {Opt_recovery, "recovery"}, /* deprecated */
359 {Opt_usebackuproot, "usebackuproot"},
360 {Opt_skip_balance, "skip_balance"},
361 {Opt_check_integrity, "check_int"},
362 {Opt_check_integrity_including_extent_data, "check_int_data"},
363 {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
364 {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
365 {Opt_fatal_errors, "fatal_errors=%s"},
366 {Opt_commit_interval, "commit=%d"},
367#ifdef CONFIG_BTRFS_DEBUG
368 {Opt_fragment_data, "fragment=data"},
369 {Opt_fragment_metadata, "fragment=metadata"},
370 {Opt_fragment_all, "fragment=all"},
371#endif
372 {Opt_err, NULL},
373};
374
375/*
376 * Regular mount options parser. Everything that is needed only when
377 * reading in a new superblock is parsed here.
378 * XXX JDM: This needs to be cleaned up for remount.
379 */
380int btrfs_parse_options(struct btrfs_root *root, char *options,
381 unsigned long new_flags)
382{
383 struct btrfs_fs_info *info = root->fs_info;
384 substring_t args[MAX_OPT_ARGS];
385 char *p, *num, *orig = NULL;
386 u64 cache_gen;
387 int intarg;
388 int ret = 0;
389 char *compress_type;
390 bool compress_force = false;
391 enum btrfs_compression_type saved_compress_type;
392 bool saved_compress_force;
393 int no_compress = 0;
394
395 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
396 if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
397 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
398 else if (cache_gen)
399 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
400
401 /*
402 * Even the options are empty, we still need to do extra check
403 * against new flags
404 */
405 if (!options)
406 goto check;
407
408 /*
409 * strsep changes the string, duplicate it because parse_options
410 * gets called twice
411 */
412 options = kstrdup(options, GFP_NOFS);
413 if (!options)
414 return -ENOMEM;
415
416 orig = options;
417
418 while ((p = strsep(&options, ",")) != NULL) {
419 int token;
420 if (!*p)
421 continue;
422
423 token = match_token(p, tokens, args);
424 switch (token) {
425 case Opt_degraded:
426 btrfs_info(root->fs_info, "allowing degraded mounts");
427 btrfs_set_opt(info->mount_opt, DEGRADED);
428 break;
429 case Opt_subvol:
430 case Opt_subvolid:
431 case Opt_subvolrootid:
432 case Opt_device:
433 /*
434 * These are parsed by btrfs_parse_early_options
435 * and can be happily ignored here.
436 */
437 break;
438 case Opt_nodatasum:
439 btrfs_set_and_info(root, NODATASUM,
440 "setting nodatasum");
441 break;
442 case Opt_datasum:
443 if (btrfs_test_opt(root, NODATASUM)) {
444 if (btrfs_test_opt(root, NODATACOW))
445 btrfs_info(root->fs_info, "setting datasum, datacow enabled");
446 else
447 btrfs_info(root->fs_info, "setting datasum");
448 }
449 btrfs_clear_opt(info->mount_opt, NODATACOW);
450 btrfs_clear_opt(info->mount_opt, NODATASUM);
451 break;
452 case Opt_nodatacow:
453 if (!btrfs_test_opt(root, NODATACOW)) {
454 if (!btrfs_test_opt(root, COMPRESS) ||
455 !btrfs_test_opt(root, FORCE_COMPRESS)) {
456 btrfs_info(root->fs_info,
457 "setting nodatacow, compression disabled");
458 } else {
459 btrfs_info(root->fs_info, "setting nodatacow");
460 }
461 }
462 btrfs_clear_opt(info->mount_opt, COMPRESS);
463 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
464 btrfs_set_opt(info->mount_opt, NODATACOW);
465 btrfs_set_opt(info->mount_opt, NODATASUM);
466 break;
467 case Opt_datacow:
468 btrfs_clear_and_info(root, NODATACOW,
469 "setting datacow");
470 break;
471 case Opt_compress_force:
472 case Opt_compress_force_type:
473 compress_force = true;
474 /* Fallthrough */
475 case Opt_compress:
476 case Opt_compress_type:
477 saved_compress_type = btrfs_test_opt(root, COMPRESS) ?
478 info->compress_type : BTRFS_COMPRESS_NONE;
479 saved_compress_force =
480 btrfs_test_opt(root, FORCE_COMPRESS);
481 if (token == Opt_compress ||
482 token == Opt_compress_force ||
483 strcmp(args[0].from, "zlib") == 0) {
484 compress_type = "zlib";
485 info->compress_type = BTRFS_COMPRESS_ZLIB;
486 btrfs_set_opt(info->mount_opt, COMPRESS);
487 btrfs_clear_opt(info->mount_opt, NODATACOW);
488 btrfs_clear_opt(info->mount_opt, NODATASUM);
489 no_compress = 0;
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 no_compress = 0;
498 } else if (strncmp(args[0].from, "no", 2) == 0) {
499 compress_type = "no";
500 btrfs_clear_opt(info->mount_opt, COMPRESS);
501 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
502 compress_force = false;
503 no_compress++;
504 } else {
505 ret = -EINVAL;
506 goto out;
507 }
508
509 if (compress_force) {
510 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
511 } else {
512 /*
513 * If we remount from compress-force=xxx to
514 * compress=xxx, we need clear FORCE_COMPRESS
515 * flag, otherwise, there is no way for users
516 * to disable forcible compression separately.
517 */
518 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
519 }
520 if ((btrfs_test_opt(root, COMPRESS) &&
521 (info->compress_type != saved_compress_type ||
522 compress_force != saved_compress_force)) ||
523 (!btrfs_test_opt(root, COMPRESS) &&
524 no_compress == 1)) {
525 btrfs_info(root->fs_info,
526 "%s %s compression",
527 (compress_force) ? "force" : "use",
528 compress_type);
529 }
530 compress_force = false;
531 break;
532 case Opt_ssd:
533 btrfs_set_and_info(root, SSD,
534 "use ssd allocation scheme");
535 break;
536 case Opt_ssd_spread:
537 btrfs_set_and_info(root, SSD_SPREAD,
538 "use spread ssd allocation scheme");
539 btrfs_set_opt(info->mount_opt, SSD);
540 break;
541 case Opt_nossd:
542 btrfs_set_and_info(root, NOSSD,
543 "not using ssd allocation scheme");
544 btrfs_clear_opt(info->mount_opt, SSD);
545 break;
546 case Opt_barrier:
547 btrfs_clear_and_info(root, NOBARRIER,
548 "turning on barriers");
549 break;
550 case Opt_nobarrier:
551 btrfs_set_and_info(root, NOBARRIER,
552 "turning off barriers");
553 break;
554 case Opt_thread_pool:
555 ret = match_int(&args[0], &intarg);
556 if (ret) {
557 goto out;
558 } else if (intarg > 0) {
559 info->thread_pool_size = intarg;
560 } else {
561 ret = -EINVAL;
562 goto out;
563 }
564 break;
565 case Opt_max_inline:
566 num = match_strdup(&args[0]);
567 if (num) {
568 info->max_inline = memparse(num, NULL);
569 kfree(num);
570
571 if (info->max_inline) {
572 info->max_inline = min_t(u64,
573 info->max_inline,
574 root->sectorsize);
575 }
576 btrfs_info(root->fs_info, "max_inline at %llu",
577 info->max_inline);
578 } else {
579 ret = -ENOMEM;
580 goto out;
581 }
582 break;
583 case Opt_alloc_start:
584 num = match_strdup(&args[0]);
585 if (num) {
586 mutex_lock(&info->chunk_mutex);
587 info->alloc_start = memparse(num, NULL);
588 mutex_unlock(&info->chunk_mutex);
589 kfree(num);
590 btrfs_info(root->fs_info, "allocations start at %llu",
591 info->alloc_start);
592 } else {
593 ret = -ENOMEM;
594 goto out;
595 }
596 break;
597 case Opt_acl:
598#ifdef CONFIG_BTRFS_FS_POSIX_ACL
599 root->fs_info->sb->s_flags |= MS_POSIXACL;
600 break;
601#else
602 btrfs_err(root->fs_info,
603 "support for ACL not compiled in!");
604 ret = -EINVAL;
605 goto out;
606#endif
607 case Opt_noacl:
608 root->fs_info->sb->s_flags &= ~MS_POSIXACL;
609 break;
610 case Opt_notreelog:
611 btrfs_set_and_info(root, NOTREELOG,
612 "disabling tree log");
613 break;
614 case Opt_treelog:
615 btrfs_clear_and_info(root, NOTREELOG,
616 "enabling tree log");
617 break;
618 case Opt_norecovery:
619 case Opt_nologreplay:
620 btrfs_set_and_info(root, NOLOGREPLAY,
621 "disabling log replay at mount time");
622 break;
623 case Opt_flushoncommit:
624 btrfs_set_and_info(root, FLUSHONCOMMIT,
625 "turning on flush-on-commit");
626 break;
627 case Opt_noflushoncommit:
628 btrfs_clear_and_info(root, FLUSHONCOMMIT,
629 "turning off flush-on-commit");
630 break;
631 case Opt_ratio:
632 ret = match_int(&args[0], &intarg);
633 if (ret) {
634 goto out;
635 } else if (intarg >= 0) {
636 info->metadata_ratio = intarg;
637 btrfs_info(root->fs_info, "metadata ratio %d",
638 info->metadata_ratio);
639 } else {
640 ret = -EINVAL;
641 goto out;
642 }
643 break;
644 case Opt_discard:
645 btrfs_set_and_info(root, DISCARD,
646 "turning on discard");
647 break;
648 case Opt_nodiscard:
649 btrfs_clear_and_info(root, DISCARD,
650 "turning off discard");
651 break;
652 case Opt_space_cache:
653 case Opt_space_cache_version:
654 if (token == Opt_space_cache ||
655 strcmp(args[0].from, "v1") == 0) {
656 btrfs_clear_opt(root->fs_info->mount_opt,
657 FREE_SPACE_TREE);
658 btrfs_set_and_info(root, SPACE_CACHE,
659 "enabling disk space caching");
660 } else if (strcmp(args[0].from, "v2") == 0) {
661 btrfs_clear_opt(root->fs_info->mount_opt,
662 SPACE_CACHE);
663 btrfs_set_and_info(root, FREE_SPACE_TREE,
664 "enabling free space tree");
665 } else {
666 ret = -EINVAL;
667 goto out;
668 }
669 break;
670 case Opt_rescan_uuid_tree:
671 btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
672 break;
673 case Opt_no_space_cache:
674 if (btrfs_test_opt(root, SPACE_CACHE)) {
675 btrfs_clear_and_info(root, SPACE_CACHE,
676 "disabling disk space caching");
677 }
678 if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
679 btrfs_clear_and_info(root, FREE_SPACE_TREE,
680 "disabling free space tree");
681 }
682 break;
683 case Opt_inode_cache:
684 btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
685 "enabling inode map caching");
686 break;
687 case Opt_noinode_cache:
688 btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
689 "disabling inode map caching");
690 break;
691 case Opt_clear_cache:
692 btrfs_set_and_info(root, CLEAR_CACHE,
693 "force clearing of disk cache");
694 break;
695 case Opt_user_subvol_rm_allowed:
696 btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
697 break;
698 case Opt_enospc_debug:
699 btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
700 break;
701 case Opt_noenospc_debug:
702 btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
703 break;
704 case Opt_defrag:
705 btrfs_set_and_info(root, AUTO_DEFRAG,
706 "enabling auto defrag");
707 break;
708 case Opt_nodefrag:
709 btrfs_clear_and_info(root, AUTO_DEFRAG,
710 "disabling auto defrag");
711 break;
712 case Opt_recovery:
713 btrfs_warn(root->fs_info,
714 "'recovery' is deprecated, use 'usebackuproot' instead");
715 case Opt_usebackuproot:
716 btrfs_info(root->fs_info,
717 "trying to use backup root at mount time");
718 btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
719 break;
720 case Opt_skip_balance:
721 btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
722 break;
723#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
724 case Opt_check_integrity_including_extent_data:
725 btrfs_info(root->fs_info,
726 "enabling check integrity including extent data");
727 btrfs_set_opt(info->mount_opt,
728 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
729 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
730 break;
731 case Opt_check_integrity:
732 btrfs_info(root->fs_info, "enabling check integrity");
733 btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
734 break;
735 case Opt_check_integrity_print_mask:
736 ret = match_int(&args[0], &intarg);
737 if (ret) {
738 goto out;
739 } else if (intarg >= 0) {
740 info->check_integrity_print_mask = intarg;
741 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
742 info->check_integrity_print_mask);
743 } else {
744 ret = -EINVAL;
745 goto out;
746 }
747 break;
748#else
749 case Opt_check_integrity_including_extent_data:
750 case Opt_check_integrity:
751 case Opt_check_integrity_print_mask:
752 btrfs_err(root->fs_info,
753 "support for check_integrity* not compiled in!");
754 ret = -EINVAL;
755 goto out;
756#endif
757 case Opt_fatal_errors:
758 if (strcmp(args[0].from, "panic") == 0)
759 btrfs_set_opt(info->mount_opt,
760 PANIC_ON_FATAL_ERROR);
761 else if (strcmp(args[0].from, "bug") == 0)
762 btrfs_clear_opt(info->mount_opt,
763 PANIC_ON_FATAL_ERROR);
764 else {
765 ret = -EINVAL;
766 goto out;
767 }
768 break;
769 case Opt_commit_interval:
770 intarg = 0;
771 ret = match_int(&args[0], &intarg);
772 if (ret < 0) {
773 btrfs_err(root->fs_info, "invalid commit interval");
774 ret = -EINVAL;
775 goto out;
776 }
777 if (intarg > 0) {
778 if (intarg > 300) {
779 btrfs_warn(root->fs_info, "excessive commit interval %d",
780 intarg);
781 }
782 info->commit_interval = intarg;
783 } else {
784 btrfs_info(root->fs_info, "using default commit interval %ds",
785 BTRFS_DEFAULT_COMMIT_INTERVAL);
786 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
787 }
788 break;
789#ifdef CONFIG_BTRFS_DEBUG
790 case Opt_fragment_all:
791 btrfs_info(root->fs_info, "fragmenting all space");
792 btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
793 btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
794 break;
795 case Opt_fragment_metadata:
796 btrfs_info(root->fs_info, "fragmenting metadata");
797 btrfs_set_opt(info->mount_opt,
798 FRAGMENT_METADATA);
799 break;
800 case Opt_fragment_data:
801 btrfs_info(root->fs_info, "fragmenting data");
802 btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
803 break;
804#endif
805 case Opt_err:
806 btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
807 ret = -EINVAL;
808 goto out;
809 default:
810 break;
811 }
812 }
813check:
814 /*
815 * Extra check for current option against current flag
816 */
817 if (btrfs_test_opt(root, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
818 btrfs_err(root->fs_info,
819 "nologreplay must be used with ro mount option");
820 ret = -EINVAL;
821 }
822out:
823 if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
824 !btrfs_test_opt(root, FREE_SPACE_TREE) &&
825 !btrfs_test_opt(root, CLEAR_CACHE)) {
826 btrfs_err(root->fs_info, "cannot disable free space tree");
827 ret = -EINVAL;
828
829 }
830 if (!ret && btrfs_test_opt(root, SPACE_CACHE))
831 btrfs_info(root->fs_info, "disk space caching is enabled");
832 if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
833 btrfs_info(root->fs_info, "using free space tree");
834 kfree(orig);
835 return ret;
836}
837
838/*
839 * Parse mount options that are required early in the mount process.
840 *
841 * All other options will be parsed on much later in the mount process and
842 * only when we need to allocate a new super block.
843 */
844static int btrfs_parse_early_options(const char *options, fmode_t flags,
845 void *holder, char **subvol_name, u64 *subvol_objectid,
846 struct btrfs_fs_devices **fs_devices)
847{
848 substring_t args[MAX_OPT_ARGS];
849 char *device_name, *opts, *orig, *p;
850 char *num = NULL;
851 int error = 0;
852
853 if (!options)
854 return 0;
855
856 /*
857 * strsep changes the string, duplicate it because parse_options
858 * gets called twice
859 */
860 opts = kstrdup(options, GFP_KERNEL);
861 if (!opts)
862 return -ENOMEM;
863 orig = opts;
864
865 while ((p = strsep(&opts, ",")) != NULL) {
866 int token;
867 if (!*p)
868 continue;
869
870 token = match_token(p, tokens, args);
871 switch (token) {
872 case Opt_subvol:
873 kfree(*subvol_name);
874 *subvol_name = match_strdup(&args[0]);
875 if (!*subvol_name) {
876 error = -ENOMEM;
877 goto out;
878 }
879 break;
880 case Opt_subvolid:
881 num = match_strdup(&args[0]);
882 if (num) {
883 *subvol_objectid = memparse(num, NULL);
884 kfree(num);
885 /* we want the original fs_tree */
886 if (!*subvol_objectid)
887 *subvol_objectid =
888 BTRFS_FS_TREE_OBJECTID;
889 } else {
890 error = -EINVAL;
891 goto out;
892 }
893 break;
894 case Opt_subvolrootid:
895 printk(KERN_WARNING
896 "BTRFS: 'subvolrootid' mount option is deprecated and has "
897 "no effect\n");
898 break;
899 case Opt_device:
900 device_name = match_strdup(&args[0]);
901 if (!device_name) {
902 error = -ENOMEM;
903 goto out;
904 }
905 error = btrfs_scan_one_device(device_name,
906 flags, holder, fs_devices);
907 kfree(device_name);
908 if (error)
909 goto out;
910 break;
911 default:
912 break;
913 }
914 }
915
916out:
917 kfree(orig);
918 return error;
919}
920
921static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
922 u64 subvol_objectid)
923{
924 struct btrfs_root *root = fs_info->tree_root;
925 struct btrfs_root *fs_root;
926 struct btrfs_root_ref *root_ref;
927 struct btrfs_inode_ref *inode_ref;
928 struct btrfs_key key;
929 struct btrfs_path *path = NULL;
930 char *name = NULL, *ptr;
931 u64 dirid;
932 int len;
933 int ret;
934
935 path = btrfs_alloc_path();
936 if (!path) {
937 ret = -ENOMEM;
938 goto err;
939 }
940 path->leave_spinning = 1;
941
942 name = kmalloc(PATH_MAX, GFP_NOFS);
943 if (!name) {
944 ret = -ENOMEM;
945 goto err;
946 }
947 ptr = name + PATH_MAX - 1;
948 ptr[0] = '\0';
949
950 /*
951 * Walk up the subvolume trees in the tree of tree roots by root
952 * backrefs until we hit the top-level subvolume.
953 */
954 while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
955 key.objectid = subvol_objectid;
956 key.type = BTRFS_ROOT_BACKREF_KEY;
957 key.offset = (u64)-1;
958
959 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
960 if (ret < 0) {
961 goto err;
962 } else if (ret > 0) {
963 ret = btrfs_previous_item(root, path, subvol_objectid,
964 BTRFS_ROOT_BACKREF_KEY);
965 if (ret < 0) {
966 goto err;
967 } else if (ret > 0) {
968 ret = -ENOENT;
969 goto err;
970 }
971 }
972
973 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
974 subvol_objectid = key.offset;
975
976 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
977 struct btrfs_root_ref);
978 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
979 ptr -= len + 1;
980 if (ptr < name) {
981 ret = -ENAMETOOLONG;
982 goto err;
983 }
984 read_extent_buffer(path->nodes[0], ptr + 1,
985 (unsigned long)(root_ref + 1), len);
986 ptr[0] = '/';
987 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
988 btrfs_release_path(path);
989
990 key.objectid = subvol_objectid;
991 key.type = BTRFS_ROOT_ITEM_KEY;
992 key.offset = (u64)-1;
993 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
994 if (IS_ERR(fs_root)) {
995 ret = PTR_ERR(fs_root);
996 goto err;
997 }
998
999 /*
1000 * Walk up the filesystem tree by inode refs until we hit the
1001 * root directory.
1002 */
1003 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1004 key.objectid = dirid;
1005 key.type = BTRFS_INODE_REF_KEY;
1006 key.offset = (u64)-1;
1007
1008 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1009 if (ret < 0) {
1010 goto err;
1011 } else if (ret > 0) {
1012 ret = btrfs_previous_item(fs_root, path, dirid,
1013 BTRFS_INODE_REF_KEY);
1014 if (ret < 0) {
1015 goto err;
1016 } else if (ret > 0) {
1017 ret = -ENOENT;
1018 goto err;
1019 }
1020 }
1021
1022 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1023 dirid = key.offset;
1024
1025 inode_ref = btrfs_item_ptr(path->nodes[0],
1026 path->slots[0],
1027 struct btrfs_inode_ref);
1028 len = btrfs_inode_ref_name_len(path->nodes[0],
1029 inode_ref);
1030 ptr -= len + 1;
1031 if (ptr < name) {
1032 ret = -ENAMETOOLONG;
1033 goto err;
1034 }
1035 read_extent_buffer(path->nodes[0], ptr + 1,
1036 (unsigned long)(inode_ref + 1), len);
1037 ptr[0] = '/';
1038 btrfs_release_path(path);
1039 }
1040 }
1041
1042 btrfs_free_path(path);
1043 if (ptr == name + PATH_MAX - 1) {
1044 name[0] = '/';
1045 name[1] = '\0';
1046 } else {
1047 memmove(name, ptr, name + PATH_MAX - ptr);
1048 }
1049 return name;
1050
1051err:
1052 btrfs_free_path(path);
1053 kfree(name);
1054 return ERR_PTR(ret);
1055}
1056
1057static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1058{
1059 struct btrfs_root *root = fs_info->tree_root;
1060 struct btrfs_dir_item *di;
1061 struct btrfs_path *path;
1062 struct btrfs_key location;
1063 u64 dir_id;
1064
1065 path = btrfs_alloc_path();
1066 if (!path)
1067 return -ENOMEM;
1068 path->leave_spinning = 1;
1069
1070 /*
1071 * Find the "default" dir item which points to the root item that we
1072 * will mount by default if we haven't been given a specific subvolume
1073 * to mount.
1074 */
1075 dir_id = btrfs_super_root_dir(fs_info->super_copy);
1076 di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1077 if (IS_ERR(di)) {
1078 btrfs_free_path(path);
1079 return PTR_ERR(di);
1080 }
1081 if (!di) {
1082 /*
1083 * Ok the default dir item isn't there. This is weird since
1084 * it's always been there, but don't freak out, just try and
1085 * mount the top-level subvolume.
1086 */
1087 btrfs_free_path(path);
1088 *objectid = BTRFS_FS_TREE_OBJECTID;
1089 return 0;
1090 }
1091
1092 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1093 btrfs_free_path(path);
1094 *objectid = location.objectid;
1095 return 0;
1096}
1097
1098static int btrfs_fill_super(struct super_block *sb,
1099 struct btrfs_fs_devices *fs_devices,
1100 void *data, int silent)
1101{
1102 struct inode *inode;
1103 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1104 struct btrfs_key key;
1105 int err;
1106
1107 sb->s_maxbytes = MAX_LFS_FILESIZE;
1108 sb->s_magic = BTRFS_SUPER_MAGIC;
1109 sb->s_op = &btrfs_super_ops;
1110 sb->s_d_op = &btrfs_dentry_operations;
1111 sb->s_export_op = &btrfs_export_ops;
1112 sb->s_xattr = btrfs_xattr_handlers;
1113 sb->s_time_gran = 1;
1114#ifdef CONFIG_BTRFS_FS_POSIX_ACL
1115 sb->s_flags |= MS_POSIXACL;
1116#endif
1117 sb->s_flags |= MS_I_VERSION;
1118 sb->s_iflags |= SB_I_CGROUPWB;
1119 err = open_ctree(sb, fs_devices, (char *)data);
1120 if (err) {
1121 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1122 return err;
1123 }
1124
1125 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1126 key.type = BTRFS_INODE_ITEM_KEY;
1127 key.offset = 0;
1128 inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1129 if (IS_ERR(inode)) {
1130 err = PTR_ERR(inode);
1131 goto fail_close;
1132 }
1133
1134 sb->s_root = d_make_root(inode);
1135 if (!sb->s_root) {
1136 err = -ENOMEM;
1137 goto fail_close;
1138 }
1139
1140 save_mount_options(sb, data);
1141 cleancache_init_fs(sb);
1142 sb->s_flags |= MS_ACTIVE;
1143 return 0;
1144
1145fail_close:
1146 close_ctree(fs_info->tree_root);
1147 return err;
1148}
1149
1150int btrfs_sync_fs(struct super_block *sb, int wait)
1151{
1152 struct btrfs_trans_handle *trans;
1153 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1154 struct btrfs_root *root = fs_info->tree_root;
1155
1156 trace_btrfs_sync_fs(wait);
1157
1158 if (!wait) {
1159 filemap_flush(fs_info->btree_inode->i_mapping);
1160 return 0;
1161 }
1162
1163 btrfs_wait_ordered_roots(fs_info, -1);
1164
1165 trans = btrfs_attach_transaction_barrier(root);
1166 if (IS_ERR(trans)) {
1167 /* no transaction, don't bother */
1168 if (PTR_ERR(trans) == -ENOENT) {
1169 /*
1170 * Exit unless we have some pending changes
1171 * that need to go through commit
1172 */
1173 if (fs_info->pending_changes == 0)
1174 return 0;
1175 /*
1176 * A non-blocking test if the fs is frozen. We must not
1177 * start a new transaction here otherwise a deadlock
1178 * happens. The pending operations are delayed to the
1179 * next commit after thawing.
1180 */
1181 if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1182 __sb_end_write(sb, SB_FREEZE_WRITE);
1183 else
1184 return 0;
1185 trans = btrfs_start_transaction(root, 0);
1186 }
1187 if (IS_ERR(trans))
1188 return PTR_ERR(trans);
1189 }
1190 return btrfs_commit_transaction(trans, root);
1191}
1192
1193static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1194{
1195 struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1196 struct btrfs_root *root = info->tree_root;
1197 char *compress_type;
1198
1199 if (btrfs_test_opt(root, DEGRADED))
1200 seq_puts(seq, ",degraded");
1201 if (btrfs_test_opt(root, NODATASUM))
1202 seq_puts(seq, ",nodatasum");
1203 if (btrfs_test_opt(root, NODATACOW))
1204 seq_puts(seq, ",nodatacow");
1205 if (btrfs_test_opt(root, NOBARRIER))
1206 seq_puts(seq, ",nobarrier");
1207 if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1208 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1209 if (info->alloc_start != 0)
1210 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1211 if (info->thread_pool_size != min_t(unsigned long,
1212 num_online_cpus() + 2, 8))
1213 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1214 if (btrfs_test_opt(root, COMPRESS)) {
1215 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1216 compress_type = "zlib";
1217 else
1218 compress_type = "lzo";
1219 if (btrfs_test_opt(root, FORCE_COMPRESS))
1220 seq_printf(seq, ",compress-force=%s", compress_type);
1221 else
1222 seq_printf(seq, ",compress=%s", compress_type);
1223 }
1224 if (btrfs_test_opt(root, NOSSD))
1225 seq_puts(seq, ",nossd");
1226 if (btrfs_test_opt(root, SSD_SPREAD))
1227 seq_puts(seq, ",ssd_spread");
1228 else if (btrfs_test_opt(root, SSD))
1229 seq_puts(seq, ",ssd");
1230 if (btrfs_test_opt(root, NOTREELOG))
1231 seq_puts(seq, ",notreelog");
1232 if (btrfs_test_opt(root, NOLOGREPLAY))
1233 seq_puts(seq, ",nologreplay");
1234 if (btrfs_test_opt(root, FLUSHONCOMMIT))
1235 seq_puts(seq, ",flushoncommit");
1236 if (btrfs_test_opt(root, DISCARD))
1237 seq_puts(seq, ",discard");
1238 if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1239 seq_puts(seq, ",noacl");
1240 if (btrfs_test_opt(root, SPACE_CACHE))
1241 seq_puts(seq, ",space_cache");
1242 else if (btrfs_test_opt(root, FREE_SPACE_TREE))
1243 seq_puts(seq, ",space_cache=v2");
1244 else
1245 seq_puts(seq, ",nospace_cache");
1246 if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1247 seq_puts(seq, ",rescan_uuid_tree");
1248 if (btrfs_test_opt(root, CLEAR_CACHE))
1249 seq_puts(seq, ",clear_cache");
1250 if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1251 seq_puts(seq, ",user_subvol_rm_allowed");
1252 if (btrfs_test_opt(root, ENOSPC_DEBUG))
1253 seq_puts(seq, ",enospc_debug");
1254 if (btrfs_test_opt(root, AUTO_DEFRAG))
1255 seq_puts(seq, ",autodefrag");
1256 if (btrfs_test_opt(root, INODE_MAP_CACHE))
1257 seq_puts(seq, ",inode_cache");
1258 if (btrfs_test_opt(root, SKIP_BALANCE))
1259 seq_puts(seq, ",skip_balance");
1260#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1261 if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1262 seq_puts(seq, ",check_int_data");
1263 else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1264 seq_puts(seq, ",check_int");
1265 if (info->check_integrity_print_mask)
1266 seq_printf(seq, ",check_int_print_mask=%d",
1267 info->check_integrity_print_mask);
1268#endif
1269 if (info->metadata_ratio)
1270 seq_printf(seq, ",metadata_ratio=%d",
1271 info->metadata_ratio);
1272 if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1273 seq_puts(seq, ",fatal_errors=panic");
1274 if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1275 seq_printf(seq, ",commit=%d", info->commit_interval);
1276#ifdef CONFIG_BTRFS_DEBUG
1277 if (btrfs_test_opt(root, FRAGMENT_DATA))
1278 seq_puts(seq, ",fragment=data");
1279 if (btrfs_test_opt(root, FRAGMENT_METADATA))
1280 seq_puts(seq, ",fragment=metadata");
1281#endif
1282 seq_printf(seq, ",subvolid=%llu",
1283 BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1284 seq_puts(seq, ",subvol=");
1285 seq_dentry(seq, dentry, " \t\n\\");
1286 return 0;
1287}
1288
1289static int btrfs_test_super(struct super_block *s, void *data)
1290{
1291 struct btrfs_fs_info *p = data;
1292 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1293
1294 return fs_info->fs_devices == p->fs_devices;
1295}
1296
1297static int btrfs_set_super(struct super_block *s, void *data)
1298{
1299 int err = set_anon_super(s, data);
1300 if (!err)
1301 s->s_fs_info = data;
1302 return err;
1303}
1304
1305/*
1306 * subvolumes are identified by ino 256
1307 */
1308static inline int is_subvolume_inode(struct inode *inode)
1309{
1310 if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1311 return 1;
1312 return 0;
1313}
1314
1315/*
1316 * This will add subvolid=0 to the argument string while removing any subvol=
1317 * and subvolid= arguments to make sure we get the top-level root for path
1318 * walking to the subvol we want.
1319 */
1320static char *setup_root_args(char *args)
1321{
1322 char *buf, *dst, *sep;
1323
1324 if (!args)
1325 return kstrdup("subvolid=0", GFP_NOFS);
1326
1327 /* The worst case is that we add ",subvolid=0" to the end. */
1328 buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1329 if (!buf)
1330 return NULL;
1331
1332 while (1) {
1333 sep = strchrnul(args, ',');
1334 if (!strstarts(args, "subvol=") &&
1335 !strstarts(args, "subvolid=")) {
1336 memcpy(dst, args, sep - args);
1337 dst += sep - args;
1338 *dst++ = ',';
1339 }
1340 if (*sep)
1341 args = sep + 1;
1342 else
1343 break;
1344 }
1345 strcpy(dst, "subvolid=0");
1346
1347 return buf;
1348}
1349
1350static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1351 int flags, const char *device_name,
1352 char *data)
1353{
1354 struct dentry *root;
1355 struct vfsmount *mnt = NULL;
1356 char *newargs;
1357 int ret;
1358
1359 newargs = setup_root_args(data);
1360 if (!newargs) {
1361 root = ERR_PTR(-ENOMEM);
1362 goto out;
1363 }
1364
1365 mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1366 if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1367 if (flags & MS_RDONLY) {
1368 mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1369 device_name, newargs);
1370 } else {
1371 mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1372 device_name, newargs);
1373 if (IS_ERR(mnt)) {
1374 root = ERR_CAST(mnt);
1375 mnt = NULL;
1376 goto out;
1377 }
1378
1379 down_write(&mnt->mnt_sb->s_umount);
1380 ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1381 up_write(&mnt->mnt_sb->s_umount);
1382 if (ret < 0) {
1383 root = ERR_PTR(ret);
1384 goto out;
1385 }
1386 }
1387 }
1388 if (IS_ERR(mnt)) {
1389 root = ERR_CAST(mnt);
1390 mnt = NULL;
1391 goto out;
1392 }
1393
1394 if (!subvol_name) {
1395 if (!subvol_objectid) {
1396 ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1397 &subvol_objectid);
1398 if (ret) {
1399 root = ERR_PTR(ret);
1400 goto out;
1401 }
1402 }
1403 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1404 subvol_objectid);
1405 if (IS_ERR(subvol_name)) {
1406 root = ERR_CAST(subvol_name);
1407 subvol_name = NULL;
1408 goto out;
1409 }
1410
1411 }
1412
1413 root = mount_subtree(mnt, subvol_name);
1414 /* mount_subtree() drops our reference on the vfsmount. */
1415 mnt = NULL;
1416
1417 if (!IS_ERR(root)) {
1418 struct super_block *s = root->d_sb;
1419 struct inode *root_inode = d_inode(root);
1420 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1421
1422 ret = 0;
1423 if (!is_subvolume_inode(root_inode)) {
1424 pr_err("BTRFS: '%s' is not a valid subvolume\n",
1425 subvol_name);
1426 ret = -EINVAL;
1427 }
1428 if (subvol_objectid && root_objectid != subvol_objectid) {
1429 /*
1430 * This will also catch a race condition where a
1431 * subvolume which was passed by ID is renamed and
1432 * another subvolume is renamed over the old location.
1433 */
1434 pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1435 subvol_name, subvol_objectid);
1436 ret = -EINVAL;
1437 }
1438 if (ret) {
1439 dput(root);
1440 root = ERR_PTR(ret);
1441 deactivate_locked_super(s);
1442 }
1443 }
1444
1445out:
1446 mntput(mnt);
1447 kfree(newargs);
1448 kfree(subvol_name);
1449 return root;
1450}
1451
1452static int parse_security_options(char *orig_opts,
1453 struct security_mnt_opts *sec_opts)
1454{
1455 char *secdata = NULL;
1456 int ret = 0;
1457
1458 secdata = alloc_secdata();
1459 if (!secdata)
1460 return -ENOMEM;
1461 ret = security_sb_copy_data(orig_opts, secdata);
1462 if (ret) {
1463 free_secdata(secdata);
1464 return ret;
1465 }
1466 ret = security_sb_parse_opts_str(secdata, sec_opts);
1467 free_secdata(secdata);
1468 return ret;
1469}
1470
1471static int setup_security_options(struct btrfs_fs_info *fs_info,
1472 struct super_block *sb,
1473 struct security_mnt_opts *sec_opts)
1474{
1475 int ret = 0;
1476
1477 /*
1478 * Call security_sb_set_mnt_opts() to check whether new sec_opts
1479 * is valid.
1480 */
1481 ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1482 if (ret)
1483 return ret;
1484
1485#ifdef CONFIG_SECURITY
1486 if (!fs_info->security_opts.num_mnt_opts) {
1487 /* first time security setup, copy sec_opts to fs_info */
1488 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1489 } else {
1490 /*
1491 * Since SELinux(the only one supports security_mnt_opts) does
1492 * NOT support changing context during remount/mount same sb,
1493 * This must be the same or part of the same security options,
1494 * just free it.
1495 */
1496 security_free_mnt_opts(sec_opts);
1497 }
1498#endif
1499 return ret;
1500}
1501
1502/*
1503 * Find a superblock for the given device / mount point.
1504 *
1505 * Note: This is based on get_sb_bdev from fs/super.c with a few additions
1506 * for multiple device setup. Make sure to keep it in sync.
1507 */
1508static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1509 const char *device_name, void *data)
1510{
1511 struct block_device *bdev = NULL;
1512 struct super_block *s;
1513 struct btrfs_fs_devices *fs_devices = NULL;
1514 struct btrfs_fs_info *fs_info = NULL;
1515 struct security_mnt_opts new_sec_opts;
1516 fmode_t mode = FMODE_READ;
1517 char *subvol_name = NULL;
1518 u64 subvol_objectid = 0;
1519 int error = 0;
1520
1521 if (!(flags & MS_RDONLY))
1522 mode |= FMODE_WRITE;
1523
1524 error = btrfs_parse_early_options(data, mode, fs_type,
1525 &subvol_name, &subvol_objectid,
1526 &fs_devices);
1527 if (error) {
1528 kfree(subvol_name);
1529 return ERR_PTR(error);
1530 }
1531
1532 if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1533 /* mount_subvol() will free subvol_name. */
1534 return mount_subvol(subvol_name, subvol_objectid, flags,
1535 device_name, data);
1536 }
1537
1538 security_init_mnt_opts(&new_sec_opts);
1539 if (data) {
1540 error = parse_security_options(data, &new_sec_opts);
1541 if (error)
1542 return ERR_PTR(error);
1543 }
1544
1545 error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1546 if (error)
1547 goto error_sec_opts;
1548
1549 /*
1550 * Setup a dummy root and fs_info for test/set super. This is because
1551 * we don't actually fill this stuff out until open_ctree, but we need
1552 * it for searching for existing supers, so this lets us do that and
1553 * then open_ctree will properly initialize everything later.
1554 */
1555 fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1556 if (!fs_info) {
1557 error = -ENOMEM;
1558 goto error_sec_opts;
1559 }
1560
1561 fs_info->fs_devices = fs_devices;
1562
1563 fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1564 fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1565 security_init_mnt_opts(&fs_info->security_opts);
1566 if (!fs_info->super_copy || !fs_info->super_for_commit) {
1567 error = -ENOMEM;
1568 goto error_fs_info;
1569 }
1570
1571 error = btrfs_open_devices(fs_devices, mode, fs_type);
1572 if (error)
1573 goto error_fs_info;
1574
1575 if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1576 error = -EACCES;
1577 goto error_close_devices;
1578 }
1579
1580 bdev = fs_devices->latest_bdev;
1581 s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1582 fs_info);
1583 if (IS_ERR(s)) {
1584 error = PTR_ERR(s);
1585 goto error_close_devices;
1586 }
1587
1588 if (s->s_root) {
1589 btrfs_close_devices(fs_devices);
1590 free_fs_info(fs_info);
1591 if ((flags ^ s->s_flags) & MS_RDONLY)
1592 error = -EBUSY;
1593 } else {
1594 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1595 btrfs_sb(s)->bdev_holder = fs_type;
1596 error = btrfs_fill_super(s, fs_devices, data,
1597 flags & MS_SILENT ? 1 : 0);
1598 }
1599 if (error) {
1600 deactivate_locked_super(s);
1601 goto error_sec_opts;
1602 }
1603
1604 fs_info = btrfs_sb(s);
1605 error = setup_security_options(fs_info, s, &new_sec_opts);
1606 if (error) {
1607 deactivate_locked_super(s);
1608 goto error_sec_opts;
1609 }
1610
1611 return dget(s->s_root);
1612
1613error_close_devices:
1614 btrfs_close_devices(fs_devices);
1615error_fs_info:
1616 free_fs_info(fs_info);
1617error_sec_opts:
1618 security_free_mnt_opts(&new_sec_opts);
1619 return ERR_PTR(error);
1620}
1621
1622static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1623 int new_pool_size, int old_pool_size)
1624{
1625 if (new_pool_size == old_pool_size)
1626 return;
1627
1628 fs_info->thread_pool_size = new_pool_size;
1629
1630 btrfs_info(fs_info, "resize thread pool %d -> %d",
1631 old_pool_size, new_pool_size);
1632
1633 btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1634 btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1635 btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1636 btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1637 btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1638 btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1639 btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1640 new_pool_size);
1641 btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1642 btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1643 btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1644 btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1645 btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1646 new_pool_size);
1647}
1648
1649static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1650{
1651 set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1652}
1653
1654static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1655 unsigned long old_opts, int flags)
1656{
1657 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1658 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1659 (flags & MS_RDONLY))) {
1660 /* wait for any defraggers to finish */
1661 wait_event(fs_info->transaction_wait,
1662 (atomic_read(&fs_info->defrag_running) == 0));
1663 if (flags & MS_RDONLY)
1664 sync_filesystem(fs_info->sb);
1665 }
1666}
1667
1668static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1669 unsigned long old_opts)
1670{
1671 /*
1672 * We need cleanup all defragable inodes if the autodefragment is
1673 * close or the fs is R/O.
1674 */
1675 if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1676 (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1677 (fs_info->sb->s_flags & MS_RDONLY))) {
1678 btrfs_cleanup_defrag_inodes(fs_info);
1679 }
1680
1681 clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1682}
1683
1684static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1685{
1686 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1687 struct btrfs_root *root = fs_info->tree_root;
1688 unsigned old_flags = sb->s_flags;
1689 unsigned long old_opts = fs_info->mount_opt;
1690 unsigned long old_compress_type = fs_info->compress_type;
1691 u64 old_max_inline = fs_info->max_inline;
1692 u64 old_alloc_start = fs_info->alloc_start;
1693 int old_thread_pool_size = fs_info->thread_pool_size;
1694 unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1695 int ret;
1696
1697 sync_filesystem(sb);
1698 btrfs_remount_prepare(fs_info);
1699
1700 if (data) {
1701 struct security_mnt_opts new_sec_opts;
1702
1703 security_init_mnt_opts(&new_sec_opts);
1704 ret = parse_security_options(data, &new_sec_opts);
1705 if (ret)
1706 goto restore;
1707 ret = setup_security_options(fs_info, sb,
1708 &new_sec_opts);
1709 if (ret) {
1710 security_free_mnt_opts(&new_sec_opts);
1711 goto restore;
1712 }
1713 }
1714
1715 ret = btrfs_parse_options(root, data, *flags);
1716 if (ret) {
1717 ret = -EINVAL;
1718 goto restore;
1719 }
1720
1721 btrfs_remount_begin(fs_info, old_opts, *flags);
1722 btrfs_resize_thread_pool(fs_info,
1723 fs_info->thread_pool_size, old_thread_pool_size);
1724
1725 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1726 goto out;
1727
1728 if (*flags & MS_RDONLY) {
1729 /*
1730 * this also happens on 'umount -rf' or on shutdown, when
1731 * the filesystem is busy.
1732 */
1733 cancel_work_sync(&fs_info->async_reclaim_work);
1734
1735 /* wait for the uuid_scan task to finish */
1736 down(&fs_info->uuid_tree_rescan_sem);
1737 /* avoid complains from lockdep et al. */
1738 up(&fs_info->uuid_tree_rescan_sem);
1739
1740 sb->s_flags |= MS_RDONLY;
1741
1742 /*
1743 * Setting MS_RDONLY will put the cleaner thread to
1744 * sleep at the next loop if it's already active.
1745 * If it's already asleep, we'll leave unused block
1746 * groups on disk until we're mounted read-write again
1747 * unless we clean them up here.
1748 */
1749 btrfs_delete_unused_bgs(fs_info);
1750
1751 btrfs_dev_replace_suspend_for_unmount(fs_info);
1752 btrfs_scrub_cancel(fs_info);
1753 btrfs_pause_balance(fs_info);
1754
1755 ret = btrfs_commit_super(root);
1756 if (ret)
1757 goto restore;
1758 } else {
1759 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1760 btrfs_err(fs_info,
1761 "Remounting read-write after error is not allowed");
1762 ret = -EINVAL;
1763 goto restore;
1764 }
1765 if (fs_info->fs_devices->rw_devices == 0) {
1766 ret = -EACCES;
1767 goto restore;
1768 }
1769
1770 if (fs_info->fs_devices->missing_devices >
1771 fs_info->num_tolerated_disk_barrier_failures &&
1772 !(*flags & MS_RDONLY)) {
1773 btrfs_warn(fs_info,
1774 "too many missing devices, writeable remount is not allowed");
1775 ret = -EACCES;
1776 goto restore;
1777 }
1778
1779 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1780 ret = -EINVAL;
1781 goto restore;
1782 }
1783
1784 ret = btrfs_cleanup_fs_roots(fs_info);
1785 if (ret)
1786 goto restore;
1787
1788 /* recover relocation */
1789 mutex_lock(&fs_info->cleaner_mutex);
1790 ret = btrfs_recover_relocation(root);
1791 mutex_unlock(&fs_info->cleaner_mutex);
1792 if (ret)
1793 goto restore;
1794
1795 ret = btrfs_resume_balance_async(fs_info);
1796 if (ret)
1797 goto restore;
1798
1799 ret = btrfs_resume_dev_replace_async(fs_info);
1800 if (ret) {
1801 btrfs_warn(fs_info, "failed to resume dev_replace");
1802 goto restore;
1803 }
1804
1805 if (!fs_info->uuid_root) {
1806 btrfs_info(fs_info, "creating UUID tree");
1807 ret = btrfs_create_uuid_tree(fs_info);
1808 if (ret) {
1809 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1810 goto restore;
1811 }
1812 }
1813 sb->s_flags &= ~MS_RDONLY;
1814 }
1815out:
1816 wake_up_process(fs_info->transaction_kthread);
1817 btrfs_remount_cleanup(fs_info, old_opts);
1818 return 0;
1819
1820restore:
1821 /* We've hit an error - don't reset MS_RDONLY */
1822 if (sb->s_flags & MS_RDONLY)
1823 old_flags |= MS_RDONLY;
1824 sb->s_flags = old_flags;
1825 fs_info->mount_opt = old_opts;
1826 fs_info->compress_type = old_compress_type;
1827 fs_info->max_inline = old_max_inline;
1828 mutex_lock(&fs_info->chunk_mutex);
1829 fs_info->alloc_start = old_alloc_start;
1830 mutex_unlock(&fs_info->chunk_mutex);
1831 btrfs_resize_thread_pool(fs_info,
1832 old_thread_pool_size, fs_info->thread_pool_size);
1833 fs_info->metadata_ratio = old_metadata_ratio;
1834 btrfs_remount_cleanup(fs_info, old_opts);
1835 return ret;
1836}
1837
1838/* Used to sort the devices by max_avail(descending sort) */
1839static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1840 const void *dev_info2)
1841{
1842 if (((struct btrfs_device_info *)dev_info1)->max_avail >
1843 ((struct btrfs_device_info *)dev_info2)->max_avail)
1844 return -1;
1845 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1846 ((struct btrfs_device_info *)dev_info2)->max_avail)
1847 return 1;
1848 else
1849 return 0;
1850}
1851
1852/*
1853 * sort the devices by max_avail, in which max free extent size of each device
1854 * is stored.(Descending Sort)
1855 */
1856static inline void btrfs_descending_sort_devices(
1857 struct btrfs_device_info *devices,
1858 size_t nr_devices)
1859{
1860 sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1861 btrfs_cmp_device_free_bytes, NULL);
1862}
1863
1864/*
1865 * The helper to calc the free space on the devices that can be used to store
1866 * file data.
1867 */
1868static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1869{
1870 struct btrfs_fs_info *fs_info = root->fs_info;
1871 struct btrfs_device_info *devices_info;
1872 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1873 struct btrfs_device *device;
1874 u64 skip_space;
1875 u64 type;
1876 u64 avail_space;
1877 u64 used_space;
1878 u64 min_stripe_size;
1879 int min_stripes = 1, num_stripes = 1;
1880 int i = 0, nr_devices;
1881 int ret;
1882
1883 /*
1884 * We aren't under the device list lock, so this is racey-ish, but good
1885 * enough for our purposes.
1886 */
1887 nr_devices = fs_info->fs_devices->open_devices;
1888 if (!nr_devices) {
1889 smp_mb();
1890 nr_devices = fs_info->fs_devices->open_devices;
1891 ASSERT(nr_devices);
1892 if (!nr_devices) {
1893 *free_bytes = 0;
1894 return 0;
1895 }
1896 }
1897
1898 devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1899 GFP_NOFS);
1900 if (!devices_info)
1901 return -ENOMEM;
1902
1903 /* calc min stripe number for data space alloction */
1904 type = btrfs_get_alloc_profile(root, 1);
1905 if (type & BTRFS_BLOCK_GROUP_RAID0) {
1906 min_stripes = 2;
1907 num_stripes = nr_devices;
1908 } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1909 min_stripes = 2;
1910 num_stripes = 2;
1911 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1912 min_stripes = 4;
1913 num_stripes = 4;
1914 }
1915
1916 if (type & BTRFS_BLOCK_GROUP_DUP)
1917 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1918 else
1919 min_stripe_size = BTRFS_STRIPE_LEN;
1920
1921 if (fs_info->alloc_start)
1922 mutex_lock(&fs_devices->device_list_mutex);
1923 rcu_read_lock();
1924 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1925 if (!device->in_fs_metadata || !device->bdev ||
1926 device->is_tgtdev_for_dev_replace)
1927 continue;
1928
1929 if (i >= nr_devices)
1930 break;
1931
1932 avail_space = device->total_bytes - device->bytes_used;
1933
1934 /* align with stripe_len */
1935 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1936 avail_space *= BTRFS_STRIPE_LEN;
1937
1938 /*
1939 * In order to avoid overwritting the superblock on the drive,
1940 * btrfs starts at an offset of at least 1MB when doing chunk
1941 * allocation.
1942 */
1943 skip_space = SZ_1M;
1944
1945 /* user can set the offset in fs_info->alloc_start. */
1946 if (fs_info->alloc_start &&
1947 fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1948 device->total_bytes) {
1949 rcu_read_unlock();
1950 skip_space = max(fs_info->alloc_start, skip_space);
1951
1952 /*
1953 * btrfs can not use the free space in
1954 * [0, skip_space - 1], we must subtract it from the
1955 * total. In order to implement it, we account the used
1956 * space in this range first.
1957 */
1958 ret = btrfs_account_dev_extents_size(device, 0,
1959 skip_space - 1,
1960 &used_space);
1961 if (ret) {
1962 kfree(devices_info);
1963 mutex_unlock(&fs_devices->device_list_mutex);
1964 return ret;
1965 }
1966
1967 rcu_read_lock();
1968
1969 /* calc the free space in [0, skip_space - 1] */
1970 skip_space -= used_space;
1971 }
1972
1973 /*
1974 * we can use the free space in [0, skip_space - 1], subtract
1975 * it from the total.
1976 */
1977 if (avail_space && avail_space >= skip_space)
1978 avail_space -= skip_space;
1979 else
1980 avail_space = 0;
1981
1982 if (avail_space < min_stripe_size)
1983 continue;
1984
1985 devices_info[i].dev = device;
1986 devices_info[i].max_avail = avail_space;
1987
1988 i++;
1989 }
1990 rcu_read_unlock();
1991 if (fs_info->alloc_start)
1992 mutex_unlock(&fs_devices->device_list_mutex);
1993
1994 nr_devices = i;
1995
1996 btrfs_descending_sort_devices(devices_info, nr_devices);
1997
1998 i = nr_devices - 1;
1999 avail_space = 0;
2000 while (nr_devices >= min_stripes) {
2001 if (num_stripes > nr_devices)
2002 num_stripes = nr_devices;
2003
2004 if (devices_info[i].max_avail >= min_stripe_size) {
2005 int j;
2006 u64 alloc_size;
2007
2008 avail_space += devices_info[i].max_avail * num_stripes;
2009 alloc_size = devices_info[i].max_avail;
2010 for (j = i + 1 - num_stripes; j <= i; j++)
2011 devices_info[j].max_avail -= alloc_size;
2012 }
2013 i--;
2014 nr_devices--;
2015 }
2016
2017 kfree(devices_info);
2018 *free_bytes = avail_space;
2019 return 0;
2020}
2021
2022/*
2023 * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2024 *
2025 * If there's a redundant raid level at DATA block groups, use the respective
2026 * multiplier to scale the sizes.
2027 *
2028 * Unused device space usage is based on simulating the chunk allocator
2029 * algorithm that respects the device sizes, order of allocations and the
2030 * 'alloc_start' value, this is a close approximation of the actual use but
2031 * there are other factors that may change the result (like a new metadata
2032 * chunk).
2033 *
2034 * If metadata is exhausted, f_bavail will be 0.
2035 *
2036 * FIXME: not accurate for mixed block groups, total and free/used are ok,
2037 * available appears slightly larger.
2038 */
2039static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2040{
2041 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2042 struct btrfs_super_block *disk_super = fs_info->super_copy;
2043 struct list_head *head = &fs_info->space_info;
2044 struct btrfs_space_info *found;
2045 u64 total_used = 0;
2046 u64 total_free_data = 0;
2047 u64 total_free_meta = 0;
2048 int bits = dentry->d_sb->s_blocksize_bits;
2049 __be32 *fsid = (__be32 *)fs_info->fsid;
2050 unsigned factor = 1;
2051 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2052 int ret;
2053 u64 thresh = 0;
2054
2055 /*
2056 * holding chunk_muext to avoid allocating new chunks, holding
2057 * device_list_mutex to avoid the device being removed
2058 */
2059 rcu_read_lock();
2060 list_for_each_entry_rcu(found, head, list) {
2061 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2062 int i;
2063
2064 total_free_data += found->disk_total - found->disk_used;
2065 total_free_data -=
2066 btrfs_account_ro_block_groups_free_space(found);
2067
2068 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2069 if (!list_empty(&found->block_groups[i])) {
2070 switch (i) {
2071 case BTRFS_RAID_DUP:
2072 case BTRFS_RAID_RAID1:
2073 case BTRFS_RAID_RAID10:
2074 factor = 2;
2075 }
2076 }
2077 }
2078 }
2079 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2080 total_free_meta += found->disk_total - found->disk_used;
2081
2082 total_used += found->disk_used;
2083 }
2084
2085 rcu_read_unlock();
2086
2087 buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2088 buf->f_blocks >>= bits;
2089 buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2090
2091 /* Account global block reserve as used, it's in logical size already */
2092 spin_lock(&block_rsv->lock);
2093 buf->f_bfree -= block_rsv->size >> bits;
2094 spin_unlock(&block_rsv->lock);
2095
2096 buf->f_bavail = div_u64(total_free_data, factor);
2097 ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2098 if (ret)
2099 return ret;
2100 buf->f_bavail += div_u64(total_free_data, factor);
2101 buf->f_bavail = buf->f_bavail >> bits;
2102
2103 /*
2104 * We calculate the remaining metadata space minus global reserve. If
2105 * this is (supposedly) smaller than zero, there's no space. But this
2106 * does not hold in practice, the exhausted state happens where's still
2107 * some positive delta. So we apply some guesswork and compare the
2108 * delta to a 4M threshold. (Practically observed delta was ~2M.)
2109 *
2110 * We probably cannot calculate the exact threshold value because this
2111 * depends on the internal reservations requested by various
2112 * operations, so some operations that consume a few metadata will
2113 * succeed even if the Avail is zero. But this is better than the other
2114 * way around.
2115 */
2116 thresh = 4 * 1024 * 1024;
2117
2118 if (total_free_meta - thresh < block_rsv->size)
2119 buf->f_bavail = 0;
2120
2121 buf->f_type = BTRFS_SUPER_MAGIC;
2122 buf->f_bsize = dentry->d_sb->s_blocksize;
2123 buf->f_namelen = BTRFS_NAME_LEN;
2124
2125 /* We treat it as constant endianness (it doesn't matter _which_)
2126 because we want the fsid to come out the same whether mounted
2127 on a big-endian or little-endian host */
2128 buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2129 buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2130 /* Mask in the root object ID too, to disambiguate subvols */
2131 buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2132 buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2133
2134 return 0;
2135}
2136
2137static void btrfs_kill_super(struct super_block *sb)
2138{
2139 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2140 kill_anon_super(sb);
2141 free_fs_info(fs_info);
2142}
2143
2144static struct file_system_type btrfs_fs_type = {
2145 .owner = THIS_MODULE,
2146 .name = "btrfs",
2147 .mount = btrfs_mount,
2148 .kill_sb = btrfs_kill_super,
2149 .fs_flags = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2150};
2151MODULE_ALIAS_FS("btrfs");
2152
2153static int btrfs_control_open(struct inode *inode, struct file *file)
2154{
2155 /*
2156 * The control file's private_data is used to hold the
2157 * transaction when it is started and is used to keep
2158 * track of whether a transaction is already in progress.
2159 */
2160 file->private_data = NULL;
2161 return 0;
2162}
2163
2164/*
2165 * used by btrfsctl to scan devices when no FS is mounted
2166 */
2167static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2168 unsigned long arg)
2169{
2170 struct btrfs_ioctl_vol_args *vol;
2171 struct btrfs_fs_devices *fs_devices;
2172 int ret = -ENOTTY;
2173
2174 if (!capable(CAP_SYS_ADMIN))
2175 return -EPERM;
2176
2177 vol = memdup_user((void __user *)arg, sizeof(*vol));
2178 if (IS_ERR(vol))
2179 return PTR_ERR(vol);
2180
2181 switch (cmd) {
2182 case BTRFS_IOC_SCAN_DEV:
2183 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2184 &btrfs_fs_type, &fs_devices);
2185 break;
2186 case BTRFS_IOC_DEVICES_READY:
2187 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2188 &btrfs_fs_type, &fs_devices);
2189 if (ret)
2190 break;
2191 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2192 break;
2193 case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2194 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2195 break;
2196 }
2197
2198 kfree(vol);
2199 return ret;
2200}
2201
2202static int btrfs_freeze(struct super_block *sb)
2203{
2204 struct btrfs_trans_handle *trans;
2205 struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2206
2207 trans = btrfs_attach_transaction_barrier(root);
2208 if (IS_ERR(trans)) {
2209 /* no transaction, don't bother */
2210 if (PTR_ERR(trans) == -ENOENT)
2211 return 0;
2212 return PTR_ERR(trans);
2213 }
2214 return btrfs_commit_transaction(trans, root);
2215}
2216
2217static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2218{
2219 struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2220 struct btrfs_fs_devices *cur_devices;
2221 struct btrfs_device *dev, *first_dev = NULL;
2222 struct list_head *head;
2223 struct rcu_string *name;
2224
2225 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2226 cur_devices = fs_info->fs_devices;
2227 while (cur_devices) {
2228 head = &cur_devices->devices;
2229 list_for_each_entry(dev, head, dev_list) {
2230 if (dev->missing)
2231 continue;
2232 if (!dev->name)
2233 continue;
2234 if (!first_dev || dev->devid < first_dev->devid)
2235 first_dev = dev;
2236 }
2237 cur_devices = cur_devices->seed;
2238 }
2239
2240 if (first_dev) {
2241 rcu_read_lock();
2242 name = rcu_dereference(first_dev->name);
2243 seq_escape(m, name->str, " \t\n\\");
2244 rcu_read_unlock();
2245 } else {
2246 WARN_ON(1);
2247 }
2248 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2249 return 0;
2250}
2251
2252static const struct super_operations btrfs_super_ops = {
2253 .drop_inode = btrfs_drop_inode,
2254 .evict_inode = btrfs_evict_inode,
2255 .put_super = btrfs_put_super,
2256 .sync_fs = btrfs_sync_fs,
2257 .show_options = btrfs_show_options,
2258 .show_devname = btrfs_show_devname,
2259 .write_inode = btrfs_write_inode,
2260 .alloc_inode = btrfs_alloc_inode,
2261 .destroy_inode = btrfs_destroy_inode,
2262 .statfs = btrfs_statfs,
2263 .remount_fs = btrfs_remount,
2264 .freeze_fs = btrfs_freeze,
2265};
2266
2267static const struct file_operations btrfs_ctl_fops = {
2268 .open = btrfs_control_open,
2269 .unlocked_ioctl = btrfs_control_ioctl,
2270 .compat_ioctl = btrfs_control_ioctl,
2271 .owner = THIS_MODULE,
2272 .llseek = noop_llseek,
2273};
2274
2275static struct miscdevice btrfs_misc = {
2276 .minor = BTRFS_MINOR,
2277 .name = "btrfs-control",
2278 .fops = &btrfs_ctl_fops
2279};
2280
2281MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2282MODULE_ALIAS("devname:btrfs-control");
2283
2284static int btrfs_interface_init(void)
2285{
2286 return misc_register(&btrfs_misc);
2287}
2288
2289static void btrfs_interface_exit(void)
2290{
2291 misc_deregister(&btrfs_misc);
2292}
2293
2294static void btrfs_print_mod_info(void)
2295{
2296 printk(KERN_INFO "Btrfs loaded"
2297#ifdef CONFIG_BTRFS_DEBUG
2298 ", debug=on"
2299#endif
2300#ifdef CONFIG_BTRFS_ASSERT
2301 ", assert=on"
2302#endif
2303#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2304 ", integrity-checker=on"
2305#endif
2306 "\n");
2307}
2308
2309static int btrfs_run_sanity_tests(void)
2310{
2311 int ret;
2312
2313 ret = btrfs_init_test_fs();
2314 if (ret)
2315 return ret;
2316
2317 ret = btrfs_test_free_space_cache();
2318 if (ret)
2319 goto out;
2320 ret = btrfs_test_extent_buffer_operations();
2321 if (ret)
2322 goto out;
2323 ret = btrfs_test_extent_io();
2324 if (ret)
2325 goto out;
2326 ret = btrfs_test_inodes();
2327 if (ret)
2328 goto out;
2329 ret = btrfs_test_qgroups();
2330 if (ret)
2331 goto out;
2332 ret = btrfs_test_free_space_tree();
2333out:
2334 btrfs_destroy_test_fs();
2335 return ret;
2336}
2337
2338static int __init init_btrfs_fs(void)
2339{
2340 int err;
2341
2342 err = btrfs_hash_init();
2343 if (err)
2344 return err;
2345
2346 btrfs_props_init();
2347
2348 err = btrfs_init_sysfs();
2349 if (err)
2350 goto free_hash;
2351
2352 btrfs_init_compress();
2353
2354 err = btrfs_init_cachep();
2355 if (err)
2356 goto free_compress;
2357
2358 err = extent_io_init();
2359 if (err)
2360 goto free_cachep;
2361
2362 err = extent_map_init();
2363 if (err)
2364 goto free_extent_io;
2365
2366 err = ordered_data_init();
2367 if (err)
2368 goto free_extent_map;
2369
2370 err = btrfs_delayed_inode_init();
2371 if (err)
2372 goto free_ordered_data;
2373
2374 err = btrfs_auto_defrag_init();
2375 if (err)
2376 goto free_delayed_inode;
2377
2378 err = btrfs_delayed_ref_init();
2379 if (err)
2380 goto free_auto_defrag;
2381
2382 err = btrfs_prelim_ref_init();
2383 if (err)
2384 goto free_delayed_ref;
2385
2386 err = btrfs_end_io_wq_init();
2387 if (err)
2388 goto free_prelim_ref;
2389
2390 err = btrfs_interface_init();
2391 if (err)
2392 goto free_end_io_wq;
2393
2394 btrfs_init_lockdep();
2395
2396 btrfs_print_mod_info();
2397
2398 err = btrfs_run_sanity_tests();
2399 if (err)
2400 goto unregister_ioctl;
2401
2402 err = register_filesystem(&btrfs_fs_type);
2403 if (err)
2404 goto unregister_ioctl;
2405
2406 return 0;
2407
2408unregister_ioctl:
2409 btrfs_interface_exit();
2410free_end_io_wq:
2411 btrfs_end_io_wq_exit();
2412free_prelim_ref:
2413 btrfs_prelim_ref_exit();
2414free_delayed_ref:
2415 btrfs_delayed_ref_exit();
2416free_auto_defrag:
2417 btrfs_auto_defrag_exit();
2418free_delayed_inode:
2419 btrfs_delayed_inode_exit();
2420free_ordered_data:
2421 ordered_data_exit();
2422free_extent_map:
2423 extent_map_exit();
2424free_extent_io:
2425 extent_io_exit();
2426free_cachep:
2427 btrfs_destroy_cachep();
2428free_compress:
2429 btrfs_exit_compress();
2430 btrfs_exit_sysfs();
2431free_hash:
2432 btrfs_hash_exit();
2433 return err;
2434}
2435
2436static void __exit exit_btrfs_fs(void)
2437{
2438 btrfs_destroy_cachep();
2439 btrfs_delayed_ref_exit();
2440 btrfs_auto_defrag_exit();
2441 btrfs_delayed_inode_exit();
2442 btrfs_prelim_ref_exit();
2443 ordered_data_exit();
2444 extent_map_exit();
2445 extent_io_exit();
2446 btrfs_interface_exit();
2447 btrfs_end_io_wq_exit();
2448 unregister_filesystem(&btrfs_fs_type);
2449 btrfs_exit_sysfs();
2450 btrfs_cleanup_fs_uuids();
2451 btrfs_exit_compress();
2452 btrfs_hash_exit();
2453}
2454
2455late_initcall(init_btrfs_fs);
2456module_exit(exit_btrfs_fs)
2457
2458MODULE_LICENSE("GPL");