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