xref: /linux/fs/btrfs/super.c (revision 25cc4eb44b0c840eff0e5a46a85b9ccbde77401b)
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 #include "backref.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/btrfs.h>
67 
68 static const struct super_operations btrfs_super_ops;
69 static struct file_system_type btrfs_fs_type;
70 
71 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
72 
73 const char *btrfs_decode_error(int errno)
74 {
75 	char *errstr = "unknown";
76 
77 	switch (errno) {
78 	case -EIO:
79 		errstr = "IO failure";
80 		break;
81 	case -ENOMEM:
82 		errstr = "Out of memory";
83 		break;
84 	case -EROFS:
85 		errstr = "Readonly filesystem";
86 		break;
87 	case -EEXIST:
88 		errstr = "Object already exists";
89 		break;
90 	case -ENOSPC:
91 		errstr = "No space left";
92 		break;
93 	case -ENOENT:
94 		errstr = "No such entry";
95 		break;
96 	}
97 
98 	return errstr;
99 }
100 
101 /* btrfs handle error by forcing the filesystem readonly */
102 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
103 {
104 	struct super_block *sb = fs_info->sb;
105 
106 	if (sb->s_flags & MS_RDONLY)
107 		return;
108 
109 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
110 		sb->s_flags |= MS_RDONLY;
111 		btrfs_info(fs_info, "forced readonly");
112 		/*
113 		 * Note that a running device replace operation is not
114 		 * canceled here although there is no way to update
115 		 * the progress. It would add the risk of a deadlock,
116 		 * therefore the canceling is omitted. The only penalty
117 		 * is that some I/O remains active until the procedure
118 		 * completes. The next time when the filesystem is
119 		 * mounted writeable again, the device replace
120 		 * operation continues.
121 		 */
122 	}
123 }
124 
125 /*
126  * __btrfs_handle_fs_error decodes expected errors from the caller and
127  * invokes the approciate error response.
128  */
129 __cold
130 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
131 		       unsigned int line, int errno, const char *fmt, ...)
132 {
133 	struct super_block *sb = fs_info->sb;
134 #ifdef CONFIG_PRINTK
135 	const char *errstr;
136 #endif
137 
138 	/*
139 	 * Special case: if the error is EROFS, and we're already
140 	 * under MS_RDONLY, then it is safe here.
141 	 */
142 	if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
143   		return;
144 
145 #ifdef CONFIG_PRINTK
146 	errstr = btrfs_decode_error(errno);
147 	if (fmt) {
148 		struct va_format vaf;
149 		va_list args;
150 
151 		va_start(args, fmt);
152 		vaf.fmt = fmt;
153 		vaf.va = &args;
154 
155 		pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
156 			sb->s_id, function, line, errno, errstr, &vaf);
157 		va_end(args);
158 	} else {
159 		pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
160 			sb->s_id, function, line, errno, errstr);
161 	}
162 #endif
163 
164 	/*
165 	 * Today we only save the error info to memory.  Long term we'll
166 	 * also send it down to the disk
167 	 */
168 	set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
169 
170 	/* Don't go through full error handling during mount */
171 	if (sb->s_flags & MS_BORN)
172 		btrfs_handle_error(fs_info);
173 }
174 
175 #ifdef CONFIG_PRINTK
176 static const char * const logtypes[] = {
177 	"emergency",
178 	"alert",
179 	"critical",
180 	"error",
181 	"warning",
182 	"notice",
183 	"info",
184 	"debug",
185 };
186 
187 
188 /*
189  * Use one ratelimit state per log level so that a flood of less important
190  * messages doesn't cause more important ones to be dropped.
191  */
192 static struct ratelimit_state printk_limits[] = {
193 	RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
194 	RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
195 	RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
196 	RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
197 	RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
198 	RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
199 	RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
200 	RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
201 };
202 
203 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
204 {
205 	struct super_block *sb = fs_info->sb;
206 	char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
207 	struct va_format vaf;
208 	va_list args;
209 	int kern_level;
210 	const char *type = logtypes[4];
211 	struct ratelimit_state *ratelimit = &printk_limits[4];
212 
213 	va_start(args, fmt);
214 
215 	while ((kern_level = printk_get_level(fmt)) != 0) {
216 		size_t size = printk_skip_level(fmt) - fmt;
217 
218 		if (kern_level >= '0' && kern_level <= '7') {
219 			memcpy(lvl, fmt,  size);
220 			lvl[size] = '\0';
221 			type = logtypes[kern_level - '0'];
222 			ratelimit = &printk_limits[kern_level - '0'];
223 		}
224 		fmt += size;
225 	}
226 
227 	vaf.fmt = fmt;
228 	vaf.va = &args;
229 
230 	if (__ratelimit(ratelimit))
231 		printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
232 
233 	va_end(args);
234 }
235 #endif
236 
237 /*
238  * We only mark the transaction aborted and then set the file system read-only.
239  * This will prevent new transactions from starting or trying to join this
240  * one.
241  *
242  * This means that error recovery at the call site is limited to freeing
243  * any local memory allocations and passing the error code up without
244  * further cleanup. The transaction should complete as it normally would
245  * in the call path but will return -EIO.
246  *
247  * We'll complete the cleanup in btrfs_end_transaction and
248  * btrfs_commit_transaction.
249  */
250 __cold
251 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
252 			       const char *function,
253 			       unsigned int line, int errno)
254 {
255 	struct btrfs_fs_info *fs_info = trans->fs_info;
256 
257 	trans->aborted = errno;
258 	/* Nothing used. The other threads that have joined this
259 	 * transaction may be able to continue. */
260 	if (!trans->dirty && list_empty(&trans->new_bgs)) {
261 		const char *errstr;
262 
263 		errstr = btrfs_decode_error(errno);
264 		btrfs_warn(fs_info,
265 		           "%s:%d: Aborting unused transaction(%s).",
266 		           function, line, errstr);
267 		return;
268 	}
269 	WRITE_ONCE(trans->transaction->aborted, errno);
270 	/* Wake up anybody who may be waiting on this transaction */
271 	wake_up(&fs_info->transaction_wait);
272 	wake_up(&fs_info->transaction_blocked_wait);
273 	__btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
274 }
275 /*
276  * __btrfs_panic decodes unexpected, fatal errors from the caller,
277  * issues an alert, and either panics or BUGs, depending on mount options.
278  */
279 __cold
280 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
281 		   unsigned int line, int errno, const char *fmt, ...)
282 {
283 	char *s_id = "<unknown>";
284 	const char *errstr;
285 	struct va_format vaf = { .fmt = fmt };
286 	va_list args;
287 
288 	if (fs_info)
289 		s_id = fs_info->sb->s_id;
290 
291 	va_start(args, fmt);
292 	vaf.va = &args;
293 
294 	errstr = btrfs_decode_error(errno);
295 	if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
296 		panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
297 			s_id, function, line, &vaf, errno, errstr);
298 
299 	btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
300 		   function, line, &vaf, errno, errstr);
301 	va_end(args);
302 	/* Caller calls BUG() */
303 }
304 
305 static void btrfs_put_super(struct super_block *sb)
306 {
307 	close_ctree(btrfs_sb(sb));
308 }
309 
310 enum {
311 	Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
312 	Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
313 	Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
314 	Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
315 	Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
316 	Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
317 	Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
318 	Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
319 	Opt_skip_balance, Opt_check_integrity,
320 	Opt_check_integrity_including_extent_data,
321 	Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
322 	Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
323 	Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
324 	Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
325 	Opt_nologreplay, Opt_norecovery,
326 #ifdef CONFIG_BTRFS_DEBUG
327 	Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
328 #endif
329 	Opt_err,
330 };
331 
332 static const match_table_t tokens = {
333 	{Opt_degraded, "degraded"},
334 	{Opt_subvol, "subvol=%s"},
335 	{Opt_subvolid, "subvolid=%s"},
336 	{Opt_device, "device=%s"},
337 	{Opt_nodatasum, "nodatasum"},
338 	{Opt_datasum, "datasum"},
339 	{Opt_nodatacow, "nodatacow"},
340 	{Opt_datacow, "datacow"},
341 	{Opt_nobarrier, "nobarrier"},
342 	{Opt_barrier, "barrier"},
343 	{Opt_max_inline, "max_inline=%s"},
344 	{Opt_alloc_start, "alloc_start=%s"},
345 	{Opt_thread_pool, "thread_pool=%d"},
346 	{Opt_compress, "compress"},
347 	{Opt_compress_type, "compress=%s"},
348 	{Opt_compress_force, "compress-force"},
349 	{Opt_compress_force_type, "compress-force=%s"},
350 	{Opt_ssd, "ssd"},
351 	{Opt_ssd_spread, "ssd_spread"},
352 	{Opt_nossd, "nossd"},
353 	{Opt_acl, "acl"},
354 	{Opt_noacl, "noacl"},
355 	{Opt_notreelog, "notreelog"},
356 	{Opt_treelog, "treelog"},
357 	{Opt_nologreplay, "nologreplay"},
358 	{Opt_norecovery, "norecovery"},
359 	{Opt_flushoncommit, "flushoncommit"},
360 	{Opt_noflushoncommit, "noflushoncommit"},
361 	{Opt_ratio, "metadata_ratio=%d"},
362 	{Opt_discard, "discard"},
363 	{Opt_nodiscard, "nodiscard"},
364 	{Opt_space_cache, "space_cache"},
365 	{Opt_space_cache_version, "space_cache=%s"},
366 	{Opt_clear_cache, "clear_cache"},
367 	{Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
368 	{Opt_enospc_debug, "enospc_debug"},
369 	{Opt_noenospc_debug, "noenospc_debug"},
370 	{Opt_subvolrootid, "subvolrootid=%d"},
371 	{Opt_defrag, "autodefrag"},
372 	{Opt_nodefrag, "noautodefrag"},
373 	{Opt_inode_cache, "inode_cache"},
374 	{Opt_noinode_cache, "noinode_cache"},
375 	{Opt_no_space_cache, "nospace_cache"},
376 	{Opt_recovery, "recovery"}, /* deprecated */
377 	{Opt_usebackuproot, "usebackuproot"},
378 	{Opt_skip_balance, "skip_balance"},
379 	{Opt_check_integrity, "check_int"},
380 	{Opt_check_integrity_including_extent_data, "check_int_data"},
381 	{Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
382 	{Opt_rescan_uuid_tree, "rescan_uuid_tree"},
383 	{Opt_fatal_errors, "fatal_errors=%s"},
384 	{Opt_commit_interval, "commit=%d"},
385 #ifdef CONFIG_BTRFS_DEBUG
386 	{Opt_fragment_data, "fragment=data"},
387 	{Opt_fragment_metadata, "fragment=metadata"},
388 	{Opt_fragment_all, "fragment=all"},
389 #endif
390 	{Opt_err, NULL},
391 };
392 
393 /*
394  * Regular mount options parser.  Everything that is needed only when
395  * reading in a new superblock is parsed here.
396  * XXX JDM: This needs to be cleaned up for remount.
397  */
398 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
399 			unsigned long new_flags)
400 {
401 	substring_t args[MAX_OPT_ARGS];
402 	char *p, *num, *orig = NULL;
403 	u64 cache_gen;
404 	int intarg;
405 	int ret = 0;
406 	char *compress_type;
407 	bool compress_force = false;
408 	enum btrfs_compression_type saved_compress_type;
409 	bool saved_compress_force;
410 	int no_compress = 0;
411 
412 	cache_gen = btrfs_super_cache_generation(info->super_copy);
413 	if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
414 		btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
415 	else if (cache_gen)
416 		btrfs_set_opt(info->mount_opt, SPACE_CACHE);
417 
418 	/*
419 	 * Even the options are empty, we still need to do extra check
420 	 * against new flags
421 	 */
422 	if (!options)
423 		goto check;
424 
425 	/*
426 	 * strsep changes the string, duplicate it because parse_options
427 	 * gets called twice
428 	 */
429 	options = kstrdup(options, GFP_KERNEL);
430 	if (!options)
431 		return -ENOMEM;
432 
433 	orig = options;
434 
435 	while ((p = strsep(&options, ",")) != NULL) {
436 		int token;
437 		if (!*p)
438 			continue;
439 
440 		token = match_token(p, tokens, args);
441 		switch (token) {
442 		case Opt_degraded:
443 			btrfs_info(info, "allowing degraded mounts");
444 			btrfs_set_opt(info->mount_opt, DEGRADED);
445 			break;
446 		case Opt_subvol:
447 		case Opt_subvolid:
448 		case Opt_subvolrootid:
449 		case Opt_device:
450 			/*
451 			 * These are parsed by btrfs_parse_early_options
452 			 * and can be happily ignored here.
453 			 */
454 			break;
455 		case Opt_nodatasum:
456 			btrfs_set_and_info(info, NODATASUM,
457 					   "setting nodatasum");
458 			break;
459 		case Opt_datasum:
460 			if (btrfs_test_opt(info, NODATASUM)) {
461 				if (btrfs_test_opt(info, NODATACOW))
462 					btrfs_info(info,
463 						   "setting datasum, datacow enabled");
464 				else
465 					btrfs_info(info, "setting datasum");
466 			}
467 			btrfs_clear_opt(info->mount_opt, NODATACOW);
468 			btrfs_clear_opt(info->mount_opt, NODATASUM);
469 			break;
470 		case Opt_nodatacow:
471 			if (!btrfs_test_opt(info, NODATACOW)) {
472 				if (!btrfs_test_opt(info, COMPRESS) ||
473 				    !btrfs_test_opt(info, FORCE_COMPRESS)) {
474 					btrfs_info(info,
475 						   "setting nodatacow, compression disabled");
476 				} else {
477 					btrfs_info(info, "setting nodatacow");
478 				}
479 			}
480 			btrfs_clear_opt(info->mount_opt, COMPRESS);
481 			btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
482 			btrfs_set_opt(info->mount_opt, NODATACOW);
483 			btrfs_set_opt(info->mount_opt, NODATASUM);
484 			break;
485 		case Opt_datacow:
486 			btrfs_clear_and_info(info, NODATACOW,
487 					     "setting datacow");
488 			break;
489 		case Opt_compress_force:
490 		case Opt_compress_force_type:
491 			compress_force = true;
492 			/* Fallthrough */
493 		case Opt_compress:
494 		case Opt_compress_type:
495 			saved_compress_type = btrfs_test_opt(info,
496 							     COMPRESS) ?
497 				info->compress_type : BTRFS_COMPRESS_NONE;
498 			saved_compress_force =
499 				btrfs_test_opt(info, FORCE_COMPRESS);
500 			if (token == Opt_compress ||
501 			    token == Opt_compress_force ||
502 			    strncmp(args[0].from, "zlib", 4) == 0) {
503 				compress_type = "zlib";
504 				info->compress_type = BTRFS_COMPRESS_ZLIB;
505 				btrfs_set_opt(info->mount_opt, COMPRESS);
506 				btrfs_clear_opt(info->mount_opt, NODATACOW);
507 				btrfs_clear_opt(info->mount_opt, NODATASUM);
508 				no_compress = 0;
509 			} else if (strncmp(args[0].from, "lzo", 3) == 0) {
510 				compress_type = "lzo";
511 				info->compress_type = BTRFS_COMPRESS_LZO;
512 				btrfs_set_opt(info->mount_opt, COMPRESS);
513 				btrfs_clear_opt(info->mount_opt, NODATACOW);
514 				btrfs_clear_opt(info->mount_opt, NODATASUM);
515 				btrfs_set_fs_incompat(info, COMPRESS_LZO);
516 				no_compress = 0;
517 			} else if (strncmp(args[0].from, "no", 2) == 0) {
518 				compress_type = "no";
519 				btrfs_clear_opt(info->mount_opt, COMPRESS);
520 				btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
521 				compress_force = false;
522 				no_compress++;
523 			} else {
524 				ret = -EINVAL;
525 				goto out;
526 			}
527 
528 			if (compress_force) {
529 				btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
530 			} else {
531 				/*
532 				 * If we remount from compress-force=xxx to
533 				 * compress=xxx, we need clear FORCE_COMPRESS
534 				 * flag, otherwise, there is no way for users
535 				 * to disable forcible compression separately.
536 				 */
537 				btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
538 			}
539 			if ((btrfs_test_opt(info, COMPRESS) &&
540 			     (info->compress_type != saved_compress_type ||
541 			      compress_force != saved_compress_force)) ||
542 			    (!btrfs_test_opt(info, COMPRESS) &&
543 			     no_compress == 1)) {
544 				btrfs_info(info, "%s %s compression",
545 					   (compress_force) ? "force" : "use",
546 					   compress_type);
547 			}
548 			compress_force = false;
549 			break;
550 		case Opt_ssd:
551 			btrfs_set_and_info(info, SSD,
552 					   "enabling ssd optimizations");
553 			btrfs_clear_opt(info->mount_opt, NOSSD);
554 			break;
555 		case Opt_ssd_spread:
556 			btrfs_set_and_info(info, SSD,
557 					   "enabling ssd optimizations");
558 			btrfs_set_and_info(info, SSD_SPREAD,
559 					   "using spread ssd allocation scheme");
560 			btrfs_clear_opt(info->mount_opt, NOSSD);
561 			break;
562 		case Opt_nossd:
563 			btrfs_set_opt(info->mount_opt, NOSSD);
564 			btrfs_clear_and_info(info, SSD,
565 					     "not using ssd optimizations");
566 			btrfs_clear_and_info(info, SSD_SPREAD,
567 					     "not using spread ssd allocation scheme");
568 			break;
569 		case Opt_barrier:
570 			btrfs_clear_and_info(info, NOBARRIER,
571 					     "turning on barriers");
572 			break;
573 		case Opt_nobarrier:
574 			btrfs_set_and_info(info, NOBARRIER,
575 					   "turning off barriers");
576 			break;
577 		case Opt_thread_pool:
578 			ret = match_int(&args[0], &intarg);
579 			if (ret) {
580 				goto out;
581 			} else if (intarg > 0) {
582 				info->thread_pool_size = intarg;
583 			} else {
584 				ret = -EINVAL;
585 				goto out;
586 			}
587 			break;
588 		case Opt_max_inline:
589 			num = match_strdup(&args[0]);
590 			if (num) {
591 				info->max_inline = memparse(num, NULL);
592 				kfree(num);
593 
594 				if (info->max_inline) {
595 					info->max_inline = min_t(u64,
596 						info->max_inline,
597 						info->sectorsize);
598 				}
599 				btrfs_info(info, "max_inline at %llu",
600 					   info->max_inline);
601 			} else {
602 				ret = -ENOMEM;
603 				goto out;
604 			}
605 			break;
606 		case Opt_alloc_start:
607 			btrfs_info(info,
608 				"option alloc_start is obsolete, ignored");
609 			break;
610 		case Opt_acl:
611 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
612 			info->sb->s_flags |= MS_POSIXACL;
613 			break;
614 #else
615 			btrfs_err(info, "support for ACL not compiled in!");
616 			ret = -EINVAL;
617 			goto out;
618 #endif
619 		case Opt_noacl:
620 			info->sb->s_flags &= ~MS_POSIXACL;
621 			break;
622 		case Opt_notreelog:
623 			btrfs_set_and_info(info, NOTREELOG,
624 					   "disabling tree log");
625 			break;
626 		case Opt_treelog:
627 			btrfs_clear_and_info(info, NOTREELOG,
628 					     "enabling tree log");
629 			break;
630 		case Opt_norecovery:
631 		case Opt_nologreplay:
632 			btrfs_set_and_info(info, NOLOGREPLAY,
633 					   "disabling log replay at mount time");
634 			break;
635 		case Opt_flushoncommit:
636 			btrfs_set_and_info(info, FLUSHONCOMMIT,
637 					   "turning on flush-on-commit");
638 			break;
639 		case Opt_noflushoncommit:
640 			btrfs_clear_and_info(info, FLUSHONCOMMIT,
641 					     "turning off flush-on-commit");
642 			break;
643 		case Opt_ratio:
644 			ret = match_int(&args[0], &intarg);
645 			if (ret) {
646 				goto out;
647 			} else if (intarg >= 0) {
648 				info->metadata_ratio = intarg;
649 				btrfs_info(info, "metadata ratio %d",
650 					   info->metadata_ratio);
651 			} else {
652 				ret = -EINVAL;
653 				goto out;
654 			}
655 			break;
656 		case Opt_discard:
657 			btrfs_set_and_info(info, DISCARD,
658 					   "turning on discard");
659 			break;
660 		case Opt_nodiscard:
661 			btrfs_clear_and_info(info, DISCARD,
662 					     "turning off discard");
663 			break;
664 		case Opt_space_cache:
665 		case Opt_space_cache_version:
666 			if (token == Opt_space_cache ||
667 			    strcmp(args[0].from, "v1") == 0) {
668 				btrfs_clear_opt(info->mount_opt,
669 						FREE_SPACE_TREE);
670 				btrfs_set_and_info(info, SPACE_CACHE,
671 					   "enabling disk space caching");
672 			} else if (strcmp(args[0].from, "v2") == 0) {
673 				btrfs_clear_opt(info->mount_opt,
674 						SPACE_CACHE);
675 				btrfs_set_and_info(info, FREE_SPACE_TREE,
676 						   "enabling free space tree");
677 			} else {
678 				ret = -EINVAL;
679 				goto out;
680 			}
681 			break;
682 		case Opt_rescan_uuid_tree:
683 			btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
684 			break;
685 		case Opt_no_space_cache:
686 			if (btrfs_test_opt(info, SPACE_CACHE)) {
687 				btrfs_clear_and_info(info, SPACE_CACHE,
688 					     "disabling disk space caching");
689 			}
690 			if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
691 				btrfs_clear_and_info(info, FREE_SPACE_TREE,
692 					     "disabling free space tree");
693 			}
694 			break;
695 		case Opt_inode_cache:
696 			btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
697 					   "enabling inode map caching");
698 			break;
699 		case Opt_noinode_cache:
700 			btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
701 					     "disabling inode map caching");
702 			break;
703 		case Opt_clear_cache:
704 			btrfs_set_and_info(info, CLEAR_CACHE,
705 					   "force clearing of disk cache");
706 			break;
707 		case Opt_user_subvol_rm_allowed:
708 			btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
709 			break;
710 		case Opt_enospc_debug:
711 			btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
712 			break;
713 		case Opt_noenospc_debug:
714 			btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
715 			break;
716 		case Opt_defrag:
717 			btrfs_set_and_info(info, AUTO_DEFRAG,
718 					   "enabling auto defrag");
719 			break;
720 		case Opt_nodefrag:
721 			btrfs_clear_and_info(info, AUTO_DEFRAG,
722 					     "disabling auto defrag");
723 			break;
724 		case Opt_recovery:
725 			btrfs_warn(info,
726 				   "'recovery' is deprecated, use 'usebackuproot' instead");
727 		case Opt_usebackuproot:
728 			btrfs_info(info,
729 				   "trying to use backup root at mount time");
730 			btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
731 			break;
732 		case Opt_skip_balance:
733 			btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
734 			break;
735 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
736 		case Opt_check_integrity_including_extent_data:
737 			btrfs_info(info,
738 				   "enabling check integrity including extent data");
739 			btrfs_set_opt(info->mount_opt,
740 				      CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
741 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
742 			break;
743 		case Opt_check_integrity:
744 			btrfs_info(info, "enabling check integrity");
745 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
746 			break;
747 		case Opt_check_integrity_print_mask:
748 			ret = match_int(&args[0], &intarg);
749 			if (ret) {
750 				goto out;
751 			} else if (intarg >= 0) {
752 				info->check_integrity_print_mask = intarg;
753 				btrfs_info(info,
754 					   "check_integrity_print_mask 0x%x",
755 					   info->check_integrity_print_mask);
756 			} else {
757 				ret = -EINVAL;
758 				goto out;
759 			}
760 			break;
761 #else
762 		case Opt_check_integrity_including_extent_data:
763 		case Opt_check_integrity:
764 		case Opt_check_integrity_print_mask:
765 			btrfs_err(info,
766 				  "support for check_integrity* not compiled in!");
767 			ret = -EINVAL;
768 			goto out;
769 #endif
770 		case Opt_fatal_errors:
771 			if (strcmp(args[0].from, "panic") == 0)
772 				btrfs_set_opt(info->mount_opt,
773 					      PANIC_ON_FATAL_ERROR);
774 			else if (strcmp(args[0].from, "bug") == 0)
775 				btrfs_clear_opt(info->mount_opt,
776 					      PANIC_ON_FATAL_ERROR);
777 			else {
778 				ret = -EINVAL;
779 				goto out;
780 			}
781 			break;
782 		case Opt_commit_interval:
783 			intarg = 0;
784 			ret = match_int(&args[0], &intarg);
785 			if (ret < 0) {
786 				btrfs_err(info, "invalid commit interval");
787 				ret = -EINVAL;
788 				goto out;
789 			}
790 			if (intarg > 0) {
791 				if (intarg > 300) {
792 					btrfs_warn(info,
793 						"excessive commit interval %d",
794 						intarg);
795 				}
796 				info->commit_interval = intarg;
797 			} else {
798 				btrfs_info(info,
799 					   "using default commit interval %ds",
800 					   BTRFS_DEFAULT_COMMIT_INTERVAL);
801 				info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
802 			}
803 			break;
804 #ifdef CONFIG_BTRFS_DEBUG
805 		case Opt_fragment_all:
806 			btrfs_info(info, "fragmenting all space");
807 			btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
808 			btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
809 			break;
810 		case Opt_fragment_metadata:
811 			btrfs_info(info, "fragmenting metadata");
812 			btrfs_set_opt(info->mount_opt,
813 				      FRAGMENT_METADATA);
814 			break;
815 		case Opt_fragment_data:
816 			btrfs_info(info, "fragmenting data");
817 			btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
818 			break;
819 #endif
820 		case Opt_err:
821 			btrfs_info(info, "unrecognized mount option '%s'", p);
822 			ret = -EINVAL;
823 			goto out;
824 		default:
825 			break;
826 		}
827 	}
828 check:
829 	/*
830 	 * Extra check for current option against current flag
831 	 */
832 	if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
833 		btrfs_err(info,
834 			  "nologreplay must be used with ro mount option");
835 		ret = -EINVAL;
836 	}
837 out:
838 	if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
839 	    !btrfs_test_opt(info, FREE_SPACE_TREE) &&
840 	    !btrfs_test_opt(info, CLEAR_CACHE)) {
841 		btrfs_err(info, "cannot disable free space tree");
842 		ret = -EINVAL;
843 
844 	}
845 	if (!ret && btrfs_test_opt(info, SPACE_CACHE))
846 		btrfs_info(info, "disk space caching is enabled");
847 	if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
848 		btrfs_info(info, "using free space tree");
849 	kfree(orig);
850 	return ret;
851 }
852 
853 /*
854  * Parse mount options that are required early in the mount process.
855  *
856  * All other options will be parsed on much later in the mount process and
857  * only when we need to allocate a new super block.
858  */
859 static int btrfs_parse_early_options(const char *options, fmode_t flags,
860 		void *holder, char **subvol_name, u64 *subvol_objectid,
861 		struct btrfs_fs_devices **fs_devices)
862 {
863 	substring_t args[MAX_OPT_ARGS];
864 	char *device_name, *opts, *orig, *p;
865 	char *num = NULL;
866 	int error = 0;
867 
868 	if (!options)
869 		return 0;
870 
871 	/*
872 	 * strsep changes the string, duplicate it because parse_options
873 	 * gets called twice
874 	 */
875 	opts = kstrdup(options, GFP_KERNEL);
876 	if (!opts)
877 		return -ENOMEM;
878 	orig = opts;
879 
880 	while ((p = strsep(&opts, ",")) != NULL) {
881 		int token;
882 		if (!*p)
883 			continue;
884 
885 		token = match_token(p, tokens, args);
886 		switch (token) {
887 		case Opt_subvol:
888 			kfree(*subvol_name);
889 			*subvol_name = match_strdup(&args[0]);
890 			if (!*subvol_name) {
891 				error = -ENOMEM;
892 				goto out;
893 			}
894 			break;
895 		case Opt_subvolid:
896 			num = match_strdup(&args[0]);
897 			if (num) {
898 				*subvol_objectid = memparse(num, NULL);
899 				kfree(num);
900 				/* we want the original fs_tree */
901 				if (!*subvol_objectid)
902 					*subvol_objectid =
903 						BTRFS_FS_TREE_OBJECTID;
904 			} else {
905 				error = -EINVAL;
906 				goto out;
907 			}
908 			break;
909 		case Opt_subvolrootid:
910 			pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
911 			break;
912 		case Opt_device:
913 			device_name = match_strdup(&args[0]);
914 			if (!device_name) {
915 				error = -ENOMEM;
916 				goto out;
917 			}
918 			error = btrfs_scan_one_device(device_name,
919 					flags, holder, fs_devices);
920 			kfree(device_name);
921 			if (error)
922 				goto out;
923 			break;
924 		default:
925 			break;
926 		}
927 	}
928 
929 out:
930 	kfree(orig);
931 	return error;
932 }
933 
934 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
935 					   u64 subvol_objectid)
936 {
937 	struct btrfs_root *root = fs_info->tree_root;
938 	struct btrfs_root *fs_root;
939 	struct btrfs_root_ref *root_ref;
940 	struct btrfs_inode_ref *inode_ref;
941 	struct btrfs_key key;
942 	struct btrfs_path *path = NULL;
943 	char *name = NULL, *ptr;
944 	u64 dirid;
945 	int len;
946 	int ret;
947 
948 	path = btrfs_alloc_path();
949 	if (!path) {
950 		ret = -ENOMEM;
951 		goto err;
952 	}
953 	path->leave_spinning = 1;
954 
955 	name = kmalloc(PATH_MAX, GFP_KERNEL);
956 	if (!name) {
957 		ret = -ENOMEM;
958 		goto err;
959 	}
960 	ptr = name + PATH_MAX - 1;
961 	ptr[0] = '\0';
962 
963 	/*
964 	 * Walk up the subvolume trees in the tree of tree roots by root
965 	 * backrefs until we hit the top-level subvolume.
966 	 */
967 	while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
968 		key.objectid = subvol_objectid;
969 		key.type = BTRFS_ROOT_BACKREF_KEY;
970 		key.offset = (u64)-1;
971 
972 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
973 		if (ret < 0) {
974 			goto err;
975 		} else if (ret > 0) {
976 			ret = btrfs_previous_item(root, path, subvol_objectid,
977 						  BTRFS_ROOT_BACKREF_KEY);
978 			if (ret < 0) {
979 				goto err;
980 			} else if (ret > 0) {
981 				ret = -ENOENT;
982 				goto err;
983 			}
984 		}
985 
986 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
987 		subvol_objectid = key.offset;
988 
989 		root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
990 					  struct btrfs_root_ref);
991 		len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
992 		ptr -= len + 1;
993 		if (ptr < name) {
994 			ret = -ENAMETOOLONG;
995 			goto err;
996 		}
997 		read_extent_buffer(path->nodes[0], ptr + 1,
998 				   (unsigned long)(root_ref + 1), len);
999 		ptr[0] = '/';
1000 		dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1001 		btrfs_release_path(path);
1002 
1003 		key.objectid = subvol_objectid;
1004 		key.type = BTRFS_ROOT_ITEM_KEY;
1005 		key.offset = (u64)-1;
1006 		fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1007 		if (IS_ERR(fs_root)) {
1008 			ret = PTR_ERR(fs_root);
1009 			goto err;
1010 		}
1011 
1012 		/*
1013 		 * Walk up the filesystem tree by inode refs until we hit the
1014 		 * root directory.
1015 		 */
1016 		while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1017 			key.objectid = dirid;
1018 			key.type = BTRFS_INODE_REF_KEY;
1019 			key.offset = (u64)-1;
1020 
1021 			ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1022 			if (ret < 0) {
1023 				goto err;
1024 			} else if (ret > 0) {
1025 				ret = btrfs_previous_item(fs_root, path, dirid,
1026 							  BTRFS_INODE_REF_KEY);
1027 				if (ret < 0) {
1028 					goto err;
1029 				} else if (ret > 0) {
1030 					ret = -ENOENT;
1031 					goto err;
1032 				}
1033 			}
1034 
1035 			btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1036 			dirid = key.offset;
1037 
1038 			inode_ref = btrfs_item_ptr(path->nodes[0],
1039 						   path->slots[0],
1040 						   struct btrfs_inode_ref);
1041 			len = btrfs_inode_ref_name_len(path->nodes[0],
1042 						       inode_ref);
1043 			ptr -= len + 1;
1044 			if (ptr < name) {
1045 				ret = -ENAMETOOLONG;
1046 				goto err;
1047 			}
1048 			read_extent_buffer(path->nodes[0], ptr + 1,
1049 					   (unsigned long)(inode_ref + 1), len);
1050 			ptr[0] = '/';
1051 			btrfs_release_path(path);
1052 		}
1053 	}
1054 
1055 	btrfs_free_path(path);
1056 	if (ptr == name + PATH_MAX - 1) {
1057 		name[0] = '/';
1058 		name[1] = '\0';
1059 	} else {
1060 		memmove(name, ptr, name + PATH_MAX - ptr);
1061 	}
1062 	return name;
1063 
1064 err:
1065 	btrfs_free_path(path);
1066 	kfree(name);
1067 	return ERR_PTR(ret);
1068 }
1069 
1070 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1071 {
1072 	struct btrfs_root *root = fs_info->tree_root;
1073 	struct btrfs_dir_item *di;
1074 	struct btrfs_path *path;
1075 	struct btrfs_key location;
1076 	u64 dir_id;
1077 
1078 	path = btrfs_alloc_path();
1079 	if (!path)
1080 		return -ENOMEM;
1081 	path->leave_spinning = 1;
1082 
1083 	/*
1084 	 * Find the "default" dir item which points to the root item that we
1085 	 * will mount by default if we haven't been given a specific subvolume
1086 	 * to mount.
1087 	 */
1088 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
1089 	di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1090 	if (IS_ERR(di)) {
1091 		btrfs_free_path(path);
1092 		return PTR_ERR(di);
1093 	}
1094 	if (!di) {
1095 		/*
1096 		 * Ok the default dir item isn't there.  This is weird since
1097 		 * it's always been there, but don't freak out, just try and
1098 		 * mount the top-level subvolume.
1099 		 */
1100 		btrfs_free_path(path);
1101 		*objectid = BTRFS_FS_TREE_OBJECTID;
1102 		return 0;
1103 	}
1104 
1105 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1106 	btrfs_free_path(path);
1107 	*objectid = location.objectid;
1108 	return 0;
1109 }
1110 
1111 static int btrfs_fill_super(struct super_block *sb,
1112 			    struct btrfs_fs_devices *fs_devices,
1113 			    void *data)
1114 {
1115 	struct inode *inode;
1116 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1117 	struct btrfs_key key;
1118 	int err;
1119 
1120 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1121 	sb->s_magic = BTRFS_SUPER_MAGIC;
1122 	sb->s_op = &btrfs_super_ops;
1123 	sb->s_d_op = &btrfs_dentry_operations;
1124 	sb->s_export_op = &btrfs_export_ops;
1125 	sb->s_xattr = btrfs_xattr_handlers;
1126 	sb->s_time_gran = 1;
1127 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1128 	sb->s_flags |= MS_POSIXACL;
1129 #endif
1130 	sb->s_flags |= MS_I_VERSION;
1131 	sb->s_iflags |= SB_I_CGROUPWB;
1132 
1133 	err = super_setup_bdi(sb);
1134 	if (err) {
1135 		btrfs_err(fs_info, "super_setup_bdi failed");
1136 		return err;
1137 	}
1138 
1139 	err = open_ctree(sb, fs_devices, (char *)data);
1140 	if (err) {
1141 		btrfs_err(fs_info, "open_ctree failed");
1142 		return err;
1143 	}
1144 
1145 	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1146 	key.type = BTRFS_INODE_ITEM_KEY;
1147 	key.offset = 0;
1148 	inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1149 	if (IS_ERR(inode)) {
1150 		err = PTR_ERR(inode);
1151 		goto fail_close;
1152 	}
1153 
1154 	sb->s_root = d_make_root(inode);
1155 	if (!sb->s_root) {
1156 		err = -ENOMEM;
1157 		goto fail_close;
1158 	}
1159 
1160 	cleancache_init_fs(sb);
1161 	sb->s_flags |= MS_ACTIVE;
1162 	return 0;
1163 
1164 fail_close:
1165 	close_ctree(fs_info);
1166 	return err;
1167 }
1168 
1169 int btrfs_sync_fs(struct super_block *sb, int wait)
1170 {
1171 	struct btrfs_trans_handle *trans;
1172 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1173 	struct btrfs_root *root = fs_info->tree_root;
1174 
1175 	trace_btrfs_sync_fs(fs_info, wait);
1176 
1177 	if (!wait) {
1178 		filemap_flush(fs_info->btree_inode->i_mapping);
1179 		return 0;
1180 	}
1181 
1182 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1183 
1184 	trans = btrfs_attach_transaction_barrier(root);
1185 	if (IS_ERR(trans)) {
1186 		/* no transaction, don't bother */
1187 		if (PTR_ERR(trans) == -ENOENT) {
1188 			/*
1189 			 * Exit unless we have some pending changes
1190 			 * that need to go through commit
1191 			 */
1192 			if (fs_info->pending_changes == 0)
1193 				return 0;
1194 			/*
1195 			 * A non-blocking test if the fs is frozen. We must not
1196 			 * start a new transaction here otherwise a deadlock
1197 			 * happens. The pending operations are delayed to the
1198 			 * next commit after thawing.
1199 			 */
1200 			if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1201 				__sb_end_write(sb, SB_FREEZE_WRITE);
1202 			else
1203 				return 0;
1204 			trans = btrfs_start_transaction(root, 0);
1205 		}
1206 		if (IS_ERR(trans))
1207 			return PTR_ERR(trans);
1208 	}
1209 	return btrfs_commit_transaction(trans);
1210 }
1211 
1212 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1213 {
1214 	struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1215 	char *compress_type;
1216 
1217 	if (btrfs_test_opt(info, DEGRADED))
1218 		seq_puts(seq, ",degraded");
1219 	if (btrfs_test_opt(info, NODATASUM))
1220 		seq_puts(seq, ",nodatasum");
1221 	if (btrfs_test_opt(info, NODATACOW))
1222 		seq_puts(seq, ",nodatacow");
1223 	if (btrfs_test_opt(info, NOBARRIER))
1224 		seq_puts(seq, ",nobarrier");
1225 	if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1226 		seq_printf(seq, ",max_inline=%llu", info->max_inline);
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 
1305 static 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 
1313 static 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  */
1324 static 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  */
1336 static char *setup_root_args(char *args)
1337 {
1338 	char *buf, *dst, *sep;
1339 
1340 	if (!args)
1341 		return kstrdup("subvolid=0", GFP_KERNEL);
1342 
1343 	/* The worst case is that we add ",subvolid=0" to the end. */
1344 	buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1,
1345 			GFP_KERNEL);
1346 	if (!buf)
1347 		return NULL;
1348 
1349 	while (1) {
1350 		sep = strchrnul(args, ',');
1351 		if (!strstarts(args, "subvol=") &&
1352 		    !strstarts(args, "subvolid=")) {
1353 			memcpy(dst, args, sep - args);
1354 			dst += sep - args;
1355 			*dst++ = ',';
1356 		}
1357 		if (*sep)
1358 			args = sep + 1;
1359 		else
1360 			break;
1361 	}
1362 	strcpy(dst, "subvolid=0");
1363 
1364 	return buf;
1365 }
1366 
1367 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1368 				   int flags, const char *device_name,
1369 				   char *data)
1370 {
1371 	struct dentry *root;
1372 	struct vfsmount *mnt = NULL;
1373 	char *newargs;
1374 	int ret;
1375 
1376 	newargs = setup_root_args(data);
1377 	if (!newargs) {
1378 		root = ERR_PTR(-ENOMEM);
1379 		goto out;
1380 	}
1381 
1382 	mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1383 	if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1384 		if (flags & MS_RDONLY) {
1385 			mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1386 					     device_name, newargs);
1387 		} else {
1388 			mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1389 					     device_name, newargs);
1390 			if (IS_ERR(mnt)) {
1391 				root = ERR_CAST(mnt);
1392 				mnt = NULL;
1393 				goto out;
1394 			}
1395 
1396 			down_write(&mnt->mnt_sb->s_umount);
1397 			ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1398 			up_write(&mnt->mnt_sb->s_umount);
1399 			if (ret < 0) {
1400 				root = ERR_PTR(ret);
1401 				goto out;
1402 			}
1403 		}
1404 	}
1405 	if (IS_ERR(mnt)) {
1406 		root = ERR_CAST(mnt);
1407 		mnt = NULL;
1408 		goto out;
1409 	}
1410 
1411 	if (!subvol_name) {
1412 		if (!subvol_objectid) {
1413 			ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1414 							  &subvol_objectid);
1415 			if (ret) {
1416 				root = ERR_PTR(ret);
1417 				goto out;
1418 			}
1419 		}
1420 		subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1421 							    subvol_objectid);
1422 		if (IS_ERR(subvol_name)) {
1423 			root = ERR_CAST(subvol_name);
1424 			subvol_name = NULL;
1425 			goto out;
1426 		}
1427 
1428 	}
1429 
1430 	root = mount_subtree(mnt, subvol_name);
1431 	/* mount_subtree() drops our reference on the vfsmount. */
1432 	mnt = NULL;
1433 
1434 	if (!IS_ERR(root)) {
1435 		struct super_block *s = root->d_sb;
1436 		struct btrfs_fs_info *fs_info = btrfs_sb(s);
1437 		struct inode *root_inode = d_inode(root);
1438 		u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1439 
1440 		ret = 0;
1441 		if (!is_subvolume_inode(root_inode)) {
1442 			btrfs_err(fs_info, "'%s' is not a valid subvolume",
1443 			       subvol_name);
1444 			ret = -EINVAL;
1445 		}
1446 		if (subvol_objectid && root_objectid != subvol_objectid) {
1447 			/*
1448 			 * This will also catch a race condition where a
1449 			 * subvolume which was passed by ID is renamed and
1450 			 * another subvolume is renamed over the old location.
1451 			 */
1452 			btrfs_err(fs_info,
1453 				  "subvol '%s' does not match subvolid %llu",
1454 				  subvol_name, subvol_objectid);
1455 			ret = -EINVAL;
1456 		}
1457 		if (ret) {
1458 			dput(root);
1459 			root = ERR_PTR(ret);
1460 			deactivate_locked_super(s);
1461 		}
1462 	}
1463 
1464 out:
1465 	mntput(mnt);
1466 	kfree(newargs);
1467 	kfree(subvol_name);
1468 	return root;
1469 }
1470 
1471 static int parse_security_options(char *orig_opts,
1472 				  struct security_mnt_opts *sec_opts)
1473 {
1474 	char *secdata = NULL;
1475 	int ret = 0;
1476 
1477 	secdata = alloc_secdata();
1478 	if (!secdata)
1479 		return -ENOMEM;
1480 	ret = security_sb_copy_data(orig_opts, secdata);
1481 	if (ret) {
1482 		free_secdata(secdata);
1483 		return ret;
1484 	}
1485 	ret = security_sb_parse_opts_str(secdata, sec_opts);
1486 	free_secdata(secdata);
1487 	return ret;
1488 }
1489 
1490 static int setup_security_options(struct btrfs_fs_info *fs_info,
1491 				  struct super_block *sb,
1492 				  struct security_mnt_opts *sec_opts)
1493 {
1494 	int ret = 0;
1495 
1496 	/*
1497 	 * Call security_sb_set_mnt_opts() to check whether new sec_opts
1498 	 * is valid.
1499 	 */
1500 	ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1501 	if (ret)
1502 		return ret;
1503 
1504 #ifdef CONFIG_SECURITY
1505 	if (!fs_info->security_opts.num_mnt_opts) {
1506 		/* first time security setup, copy sec_opts to fs_info */
1507 		memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1508 	} else {
1509 		/*
1510 		 * Since SELinux (the only one supporting security_mnt_opts)
1511 		 * does NOT support changing context during remount/mount of
1512 		 * the same sb, this must be the same or part of the same
1513 		 * security options, just free it.
1514 		 */
1515 		security_free_mnt_opts(sec_opts);
1516 	}
1517 #endif
1518 	return ret;
1519 }
1520 
1521 /*
1522  * Find a superblock for the given device / mount point.
1523  *
1524  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1525  *	  for multiple device setup.  Make sure to keep it in sync.
1526  */
1527 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1528 		const char *device_name, void *data)
1529 {
1530 	struct block_device *bdev = NULL;
1531 	struct super_block *s;
1532 	struct btrfs_fs_devices *fs_devices = NULL;
1533 	struct btrfs_fs_info *fs_info = NULL;
1534 	struct security_mnt_opts new_sec_opts;
1535 	fmode_t mode = FMODE_READ;
1536 	char *subvol_name = NULL;
1537 	u64 subvol_objectid = 0;
1538 	int error = 0;
1539 
1540 	if (!(flags & MS_RDONLY))
1541 		mode |= FMODE_WRITE;
1542 
1543 	error = btrfs_parse_early_options(data, mode, fs_type,
1544 					  &subvol_name, &subvol_objectid,
1545 					  &fs_devices);
1546 	if (error) {
1547 		kfree(subvol_name);
1548 		return ERR_PTR(error);
1549 	}
1550 
1551 	if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1552 		/* mount_subvol() will free subvol_name. */
1553 		return mount_subvol(subvol_name, subvol_objectid, flags,
1554 				    device_name, data);
1555 	}
1556 
1557 	security_init_mnt_opts(&new_sec_opts);
1558 	if (data) {
1559 		error = parse_security_options(data, &new_sec_opts);
1560 		if (error)
1561 			return ERR_PTR(error);
1562 	}
1563 
1564 	error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1565 	if (error)
1566 		goto error_sec_opts;
1567 
1568 	/*
1569 	 * Setup a dummy root and fs_info for test/set super.  This is because
1570 	 * we don't actually fill this stuff out until open_ctree, but we need
1571 	 * it for searching for existing supers, so this lets us do that and
1572 	 * then open_ctree will properly initialize everything later.
1573 	 */
1574 	fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1575 	if (!fs_info) {
1576 		error = -ENOMEM;
1577 		goto error_sec_opts;
1578 	}
1579 
1580 	fs_info->fs_devices = fs_devices;
1581 
1582 	fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1583 	fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1584 	security_init_mnt_opts(&fs_info->security_opts);
1585 	if (!fs_info->super_copy || !fs_info->super_for_commit) {
1586 		error = -ENOMEM;
1587 		goto error_fs_info;
1588 	}
1589 
1590 	error = btrfs_open_devices(fs_devices, mode, fs_type);
1591 	if (error)
1592 		goto error_fs_info;
1593 
1594 	if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1595 		error = -EACCES;
1596 		goto error_close_devices;
1597 	}
1598 
1599 	bdev = fs_devices->latest_bdev;
1600 	s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1601 		 fs_info);
1602 	if (IS_ERR(s)) {
1603 		error = PTR_ERR(s);
1604 		goto error_close_devices;
1605 	}
1606 
1607 	if (s->s_root) {
1608 		btrfs_close_devices(fs_devices);
1609 		free_fs_info(fs_info);
1610 		if ((flags ^ s->s_flags) & MS_RDONLY)
1611 			error = -EBUSY;
1612 	} else {
1613 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1614 		btrfs_sb(s)->bdev_holder = fs_type;
1615 		error = btrfs_fill_super(s, fs_devices, data);
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 
1631 error_close_devices:
1632 	btrfs_close_devices(fs_devices);
1633 error_fs_info:
1634 	free_fs_info(fs_info);
1635 error_sec_opts:
1636 	security_free_mnt_opts(&new_sec_opts);
1637 	return ERR_PTR(error);
1638 }
1639 
1640 static 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 
1667 static 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 
1672 static 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 
1686 static 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 
1702 static 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 	int old_thread_pool_size = fs_info->thread_pool_size;
1711 	unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1712 	int ret;
1713 
1714 	sync_filesystem(sb);
1715 	btrfs_remount_prepare(fs_info);
1716 
1717 	if (data) {
1718 		struct security_mnt_opts new_sec_opts;
1719 
1720 		security_init_mnt_opts(&new_sec_opts);
1721 		ret = parse_security_options(data, &new_sec_opts);
1722 		if (ret)
1723 			goto restore;
1724 		ret = setup_security_options(fs_info, sb,
1725 					     &new_sec_opts);
1726 		if (ret) {
1727 			security_free_mnt_opts(&new_sec_opts);
1728 			goto restore;
1729 		}
1730 	}
1731 
1732 	ret = btrfs_parse_options(fs_info, data, *flags);
1733 	if (ret) {
1734 		ret = -EINVAL;
1735 		goto restore;
1736 	}
1737 
1738 	btrfs_remount_begin(fs_info, old_opts, *flags);
1739 	btrfs_resize_thread_pool(fs_info,
1740 		fs_info->thread_pool_size, old_thread_pool_size);
1741 
1742 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1743 		goto out;
1744 
1745 	if (*flags & MS_RDONLY) {
1746 		/*
1747 		 * this also happens on 'umount -rf' or on shutdown, when
1748 		 * the filesystem is busy.
1749 		 */
1750 		cancel_work_sync(&fs_info->async_reclaim_work);
1751 
1752 		/* wait for the uuid_scan task to finish */
1753 		down(&fs_info->uuid_tree_rescan_sem);
1754 		/* avoid complains from lockdep et al. */
1755 		up(&fs_info->uuid_tree_rescan_sem);
1756 
1757 		sb->s_flags |= MS_RDONLY;
1758 
1759 		/*
1760 		 * Setting MS_RDONLY will put the cleaner thread to
1761 		 * sleep at the next loop if it's already active.
1762 		 * If it's already asleep, we'll leave unused block
1763 		 * groups on disk until we're mounted read-write again
1764 		 * unless we clean them up here.
1765 		 */
1766 		btrfs_delete_unused_bgs(fs_info);
1767 
1768 		btrfs_dev_replace_suspend_for_unmount(fs_info);
1769 		btrfs_scrub_cancel(fs_info);
1770 		btrfs_pause_balance(fs_info);
1771 
1772 		ret = btrfs_commit_super(fs_info);
1773 		if (ret)
1774 			goto restore;
1775 	} else {
1776 		if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1777 			btrfs_err(fs_info,
1778 				"Remounting read-write after error is not allowed");
1779 			ret = -EINVAL;
1780 			goto restore;
1781 		}
1782 		if (fs_info->fs_devices->rw_devices == 0) {
1783 			ret = -EACCES;
1784 			goto restore;
1785 		}
1786 
1787 		if (!btrfs_check_rw_degradable(fs_info)) {
1788 			btrfs_warn(fs_info,
1789 				"too many missing devices, writeable remount is not allowed");
1790 			ret = -EACCES;
1791 			goto restore;
1792 		}
1793 
1794 		if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1795 			ret = -EINVAL;
1796 			goto restore;
1797 		}
1798 
1799 		ret = btrfs_cleanup_fs_roots(fs_info);
1800 		if (ret)
1801 			goto restore;
1802 
1803 		/* recover relocation */
1804 		mutex_lock(&fs_info->cleaner_mutex);
1805 		ret = btrfs_recover_relocation(root);
1806 		mutex_unlock(&fs_info->cleaner_mutex);
1807 		if (ret)
1808 			goto restore;
1809 
1810 		ret = btrfs_resume_balance_async(fs_info);
1811 		if (ret)
1812 			goto restore;
1813 
1814 		ret = btrfs_resume_dev_replace_async(fs_info);
1815 		if (ret) {
1816 			btrfs_warn(fs_info, "failed to resume dev_replace");
1817 			goto restore;
1818 		}
1819 
1820 		btrfs_qgroup_rescan_resume(fs_info);
1821 
1822 		if (!fs_info->uuid_root) {
1823 			btrfs_info(fs_info, "creating UUID tree");
1824 			ret = btrfs_create_uuid_tree(fs_info);
1825 			if (ret) {
1826 				btrfs_warn(fs_info,
1827 					   "failed to create the UUID tree %d",
1828 					   ret);
1829 				goto restore;
1830 			}
1831 		}
1832 		sb->s_flags &= ~MS_RDONLY;
1833 
1834 		set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1835 	}
1836 out:
1837 	wake_up_process(fs_info->transaction_kthread);
1838 	btrfs_remount_cleanup(fs_info, old_opts);
1839 	return 0;
1840 
1841 restore:
1842 	/* We've hit an error - don't reset MS_RDONLY */
1843 	if (sb->s_flags & MS_RDONLY)
1844 		old_flags |= MS_RDONLY;
1845 	sb->s_flags = old_flags;
1846 	fs_info->mount_opt = old_opts;
1847 	fs_info->compress_type = old_compress_type;
1848 	fs_info->max_inline = old_max_inline;
1849 	btrfs_resize_thread_pool(fs_info,
1850 		old_thread_pool_size, fs_info->thread_pool_size);
1851 	fs_info->metadata_ratio = old_metadata_ratio;
1852 	btrfs_remount_cleanup(fs_info, old_opts);
1853 	return ret;
1854 }
1855 
1856 /* Used to sort the devices by max_avail(descending sort) */
1857 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1858 				       const void *dev_info2)
1859 {
1860 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
1861 	    ((struct btrfs_device_info *)dev_info2)->max_avail)
1862 		return -1;
1863 	else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1864 		 ((struct btrfs_device_info *)dev_info2)->max_avail)
1865 		return 1;
1866 	else
1867 	return 0;
1868 }
1869 
1870 /*
1871  * sort the devices by max_avail, in which max free extent size of each device
1872  * is stored.(Descending Sort)
1873  */
1874 static inline void btrfs_descending_sort_devices(
1875 					struct btrfs_device_info *devices,
1876 					size_t nr_devices)
1877 {
1878 	sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1879 	     btrfs_cmp_device_free_bytes, NULL);
1880 }
1881 
1882 /*
1883  * The helper to calc the free space on the devices that can be used to store
1884  * file data.
1885  */
1886 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1887 				       u64 *free_bytes)
1888 {
1889 	struct btrfs_device_info *devices_info;
1890 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1891 	struct btrfs_device *device;
1892 	u64 skip_space;
1893 	u64 type;
1894 	u64 avail_space;
1895 	u64 min_stripe_size;
1896 	int min_stripes = 1, num_stripes = 1;
1897 	int i = 0, nr_devices;
1898 
1899 	/*
1900 	 * We aren't under the device list lock, so this is racy-ish, but good
1901 	 * enough for our purposes.
1902 	 */
1903 	nr_devices = fs_info->fs_devices->open_devices;
1904 	if (!nr_devices) {
1905 		smp_mb();
1906 		nr_devices = fs_info->fs_devices->open_devices;
1907 		ASSERT(nr_devices);
1908 		if (!nr_devices) {
1909 			*free_bytes = 0;
1910 			return 0;
1911 		}
1912 	}
1913 
1914 	devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1915 			       GFP_KERNEL);
1916 	if (!devices_info)
1917 		return -ENOMEM;
1918 
1919 	/* calc min stripe number for data space allocation */
1920 	type = btrfs_data_alloc_profile(fs_info);
1921 	if (type & BTRFS_BLOCK_GROUP_RAID0) {
1922 		min_stripes = 2;
1923 		num_stripes = nr_devices;
1924 	} else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1925 		min_stripes = 2;
1926 		num_stripes = 2;
1927 	} else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1928 		min_stripes = 4;
1929 		num_stripes = 4;
1930 	}
1931 
1932 	if (type & BTRFS_BLOCK_GROUP_DUP)
1933 		min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1934 	else
1935 		min_stripe_size = BTRFS_STRIPE_LEN;
1936 
1937 	rcu_read_lock();
1938 	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1939 		if (!device->in_fs_metadata || !device->bdev ||
1940 		    device->is_tgtdev_for_dev_replace)
1941 			continue;
1942 
1943 		if (i >= nr_devices)
1944 			break;
1945 
1946 		avail_space = device->total_bytes - device->bytes_used;
1947 
1948 		/* align with stripe_len */
1949 		avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1950 		avail_space *= BTRFS_STRIPE_LEN;
1951 
1952 		/*
1953 		 * In order to avoid overwriting the superblock on the drive,
1954 		 * btrfs starts at an offset of at least 1MB when doing chunk
1955 		 * allocation.
1956 		 */
1957 		skip_space = SZ_1M;
1958 
1959 		/*
1960 		 * we can use the free space in [0, skip_space - 1], subtract
1961 		 * it from the total.
1962 		 */
1963 		if (avail_space && avail_space >= skip_space)
1964 			avail_space -= skip_space;
1965 		else
1966 			avail_space = 0;
1967 
1968 		if (avail_space < min_stripe_size)
1969 			continue;
1970 
1971 		devices_info[i].dev = device;
1972 		devices_info[i].max_avail = avail_space;
1973 
1974 		i++;
1975 	}
1976 	rcu_read_unlock();
1977 
1978 	nr_devices = i;
1979 
1980 	btrfs_descending_sort_devices(devices_info, nr_devices);
1981 
1982 	i = nr_devices - 1;
1983 	avail_space = 0;
1984 	while (nr_devices >= min_stripes) {
1985 		if (num_stripes > nr_devices)
1986 			num_stripes = nr_devices;
1987 
1988 		if (devices_info[i].max_avail >= min_stripe_size) {
1989 			int j;
1990 			u64 alloc_size;
1991 
1992 			avail_space += devices_info[i].max_avail * num_stripes;
1993 			alloc_size = devices_info[i].max_avail;
1994 			for (j = i + 1 - num_stripes; j <= i; j++)
1995 				devices_info[j].max_avail -= alloc_size;
1996 		}
1997 		i--;
1998 		nr_devices--;
1999 	}
2000 
2001 	kfree(devices_info);
2002 	*free_bytes = avail_space;
2003 	return 0;
2004 }
2005 
2006 /*
2007  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2008  *
2009  * If there's a redundant raid level at DATA block groups, use the respective
2010  * multiplier to scale the sizes.
2011  *
2012  * Unused device space usage is based on simulating the chunk allocator
2013  * algorithm that respects the device sizes and order of allocations.  This is
2014  * a close approximation of the actual use but there are other factors that may
2015  * change the result (like a new metadata chunk).
2016  *
2017  * If metadata is exhausted, f_bavail will be 0.
2018  */
2019 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2020 {
2021 	struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2022 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2023 	struct list_head *head = &fs_info->space_info;
2024 	struct btrfs_space_info *found;
2025 	u64 total_used = 0;
2026 	u64 total_free_data = 0;
2027 	u64 total_free_meta = 0;
2028 	int bits = dentry->d_sb->s_blocksize_bits;
2029 	__be32 *fsid = (__be32 *)fs_info->fsid;
2030 	unsigned factor = 1;
2031 	struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2032 	int ret;
2033 	u64 thresh = 0;
2034 	int mixed = 0;
2035 
2036 	rcu_read_lock();
2037 	list_for_each_entry_rcu(found, head, list) {
2038 		if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2039 			int i;
2040 
2041 			total_free_data += found->disk_total - found->disk_used;
2042 			total_free_data -=
2043 				btrfs_account_ro_block_groups_free_space(found);
2044 
2045 			for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2046 				if (!list_empty(&found->block_groups[i])) {
2047 					switch (i) {
2048 					case BTRFS_RAID_DUP:
2049 					case BTRFS_RAID_RAID1:
2050 					case BTRFS_RAID_RAID10:
2051 						factor = 2;
2052 					}
2053 				}
2054 			}
2055 		}
2056 
2057 		/*
2058 		 * Metadata in mixed block goup profiles are accounted in data
2059 		 */
2060 		if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2061 			if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2062 				mixed = 1;
2063 			else
2064 				total_free_meta += found->disk_total -
2065 					found->disk_used;
2066 		}
2067 
2068 		total_used += found->disk_used;
2069 	}
2070 
2071 	rcu_read_unlock();
2072 
2073 	buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2074 	buf->f_blocks >>= bits;
2075 	buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2076 
2077 	/* Account global block reserve as used, it's in logical size already */
2078 	spin_lock(&block_rsv->lock);
2079 	/* Mixed block groups accounting is not byte-accurate, avoid overflow */
2080 	if (buf->f_bfree >= block_rsv->size >> bits)
2081 		buf->f_bfree -= block_rsv->size >> bits;
2082 	else
2083 		buf->f_bfree = 0;
2084 	spin_unlock(&block_rsv->lock);
2085 
2086 	buf->f_bavail = div_u64(total_free_data, factor);
2087 	ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2088 	if (ret)
2089 		return ret;
2090 	buf->f_bavail += div_u64(total_free_data, factor);
2091 	buf->f_bavail = buf->f_bavail >> bits;
2092 
2093 	/*
2094 	 * We calculate the remaining metadata space minus global reserve. If
2095 	 * this is (supposedly) smaller than zero, there's no space. But this
2096 	 * does not hold in practice, the exhausted state happens where's still
2097 	 * some positive delta. So we apply some guesswork and compare the
2098 	 * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2099 	 *
2100 	 * We probably cannot calculate the exact threshold value because this
2101 	 * depends on the internal reservations requested by various
2102 	 * operations, so some operations that consume a few metadata will
2103 	 * succeed even if the Avail is zero. But this is better than the other
2104 	 * way around.
2105 	 */
2106 	thresh = 4 * 1024 * 1024;
2107 
2108 	if (!mixed && total_free_meta - thresh < block_rsv->size)
2109 		buf->f_bavail = 0;
2110 
2111 	buf->f_type = BTRFS_SUPER_MAGIC;
2112 	buf->f_bsize = dentry->d_sb->s_blocksize;
2113 	buf->f_namelen = BTRFS_NAME_LEN;
2114 
2115 	/* We treat it as constant endianness (it doesn't matter _which_)
2116 	   because we want the fsid to come out the same whether mounted
2117 	   on a big-endian or little-endian host */
2118 	buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2119 	buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2120 	/* Mask in the root object ID too, to disambiguate subvols */
2121 	buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2122 	buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2123 
2124 	return 0;
2125 }
2126 
2127 static void btrfs_kill_super(struct super_block *sb)
2128 {
2129 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2130 	kill_anon_super(sb);
2131 	free_fs_info(fs_info);
2132 }
2133 
2134 static struct file_system_type btrfs_fs_type = {
2135 	.owner		= THIS_MODULE,
2136 	.name		= "btrfs",
2137 	.mount		= btrfs_mount,
2138 	.kill_sb	= btrfs_kill_super,
2139 	.fs_flags	= FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2140 };
2141 MODULE_ALIAS_FS("btrfs");
2142 
2143 static int btrfs_control_open(struct inode *inode, struct file *file)
2144 {
2145 	/*
2146 	 * The control file's private_data is used to hold the
2147 	 * transaction when it is started and is used to keep
2148 	 * track of whether a transaction is already in progress.
2149 	 */
2150 	file->private_data = NULL;
2151 	return 0;
2152 }
2153 
2154 /*
2155  * used by btrfsctl to scan devices when no FS is mounted
2156  */
2157 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2158 				unsigned long arg)
2159 {
2160 	struct btrfs_ioctl_vol_args *vol;
2161 	struct btrfs_fs_devices *fs_devices;
2162 	int ret = -ENOTTY;
2163 
2164 	if (!capable(CAP_SYS_ADMIN))
2165 		return -EPERM;
2166 
2167 	vol = memdup_user((void __user *)arg, sizeof(*vol));
2168 	if (IS_ERR(vol))
2169 		return PTR_ERR(vol);
2170 
2171 	switch (cmd) {
2172 	case BTRFS_IOC_SCAN_DEV:
2173 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2174 					    &btrfs_fs_type, &fs_devices);
2175 		break;
2176 	case BTRFS_IOC_DEVICES_READY:
2177 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2178 					    &btrfs_fs_type, &fs_devices);
2179 		if (ret)
2180 			break;
2181 		ret = !(fs_devices->num_devices == fs_devices->total_devices);
2182 		break;
2183 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2184 		ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2185 		break;
2186 	}
2187 
2188 	kfree(vol);
2189 	return ret;
2190 }
2191 
2192 static int btrfs_freeze(struct super_block *sb)
2193 {
2194 	struct btrfs_trans_handle *trans;
2195 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2196 	struct btrfs_root *root = fs_info->tree_root;
2197 
2198 	set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2199 	/*
2200 	 * We don't need a barrier here, we'll wait for any transaction that
2201 	 * could be in progress on other threads (and do delayed iputs that
2202 	 * we want to avoid on a frozen filesystem), or do the commit
2203 	 * ourselves.
2204 	 */
2205 	trans = btrfs_attach_transaction_barrier(root);
2206 	if (IS_ERR(trans)) {
2207 		/* no transaction, don't bother */
2208 		if (PTR_ERR(trans) == -ENOENT)
2209 			return 0;
2210 		return PTR_ERR(trans);
2211 	}
2212 	return btrfs_commit_transaction(trans);
2213 }
2214 
2215 static int btrfs_unfreeze(struct super_block *sb)
2216 {
2217 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2218 
2219 	clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2220 	return 0;
2221 }
2222 
2223 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2224 {
2225 	struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2226 	struct btrfs_fs_devices *cur_devices;
2227 	struct btrfs_device *dev, *first_dev = NULL;
2228 	struct list_head *head;
2229 	struct rcu_string *name;
2230 
2231 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2232 	cur_devices = fs_info->fs_devices;
2233 	while (cur_devices) {
2234 		head = &cur_devices->devices;
2235 		list_for_each_entry(dev, head, dev_list) {
2236 			if (dev->missing)
2237 				continue;
2238 			if (!dev->name)
2239 				continue;
2240 			if (!first_dev || dev->devid < first_dev->devid)
2241 				first_dev = dev;
2242 		}
2243 		cur_devices = cur_devices->seed;
2244 	}
2245 
2246 	if (first_dev) {
2247 		rcu_read_lock();
2248 		name = rcu_dereference(first_dev->name);
2249 		seq_escape(m, name->str, " \t\n\\");
2250 		rcu_read_unlock();
2251 	} else {
2252 		WARN_ON(1);
2253 	}
2254 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2255 	return 0;
2256 }
2257 
2258 static const struct super_operations btrfs_super_ops = {
2259 	.drop_inode	= btrfs_drop_inode,
2260 	.evict_inode	= btrfs_evict_inode,
2261 	.put_super	= btrfs_put_super,
2262 	.sync_fs	= btrfs_sync_fs,
2263 	.show_options	= btrfs_show_options,
2264 	.show_devname	= btrfs_show_devname,
2265 	.write_inode	= btrfs_write_inode,
2266 	.alloc_inode	= btrfs_alloc_inode,
2267 	.destroy_inode	= btrfs_destroy_inode,
2268 	.statfs		= btrfs_statfs,
2269 	.remount_fs	= btrfs_remount,
2270 	.freeze_fs	= btrfs_freeze,
2271 	.unfreeze_fs	= btrfs_unfreeze,
2272 };
2273 
2274 static const struct file_operations btrfs_ctl_fops = {
2275 	.open = btrfs_control_open,
2276 	.unlocked_ioctl	 = btrfs_control_ioctl,
2277 	.compat_ioctl = btrfs_control_ioctl,
2278 	.owner	 = THIS_MODULE,
2279 	.llseek = noop_llseek,
2280 };
2281 
2282 static struct miscdevice btrfs_misc = {
2283 	.minor		= BTRFS_MINOR,
2284 	.name		= "btrfs-control",
2285 	.fops		= &btrfs_ctl_fops
2286 };
2287 
2288 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2289 MODULE_ALIAS("devname:btrfs-control");
2290 
2291 static int btrfs_interface_init(void)
2292 {
2293 	return misc_register(&btrfs_misc);
2294 }
2295 
2296 static void btrfs_interface_exit(void)
2297 {
2298 	misc_deregister(&btrfs_misc);
2299 }
2300 
2301 static void btrfs_print_mod_info(void)
2302 {
2303 	pr_info("Btrfs loaded, crc32c=%s"
2304 #ifdef CONFIG_BTRFS_DEBUG
2305 			", debug=on"
2306 #endif
2307 #ifdef CONFIG_BTRFS_ASSERT
2308 			", assert=on"
2309 #endif
2310 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2311 			", integrity-checker=on"
2312 #endif
2313 			"\n",
2314 			btrfs_crc32c_impl());
2315 }
2316 
2317 static int __init init_btrfs_fs(void)
2318 {
2319 	int err;
2320 
2321 	err = btrfs_hash_init();
2322 	if (err)
2323 		return err;
2324 
2325 	btrfs_props_init();
2326 
2327 	err = btrfs_init_sysfs();
2328 	if (err)
2329 		goto free_hash;
2330 
2331 	btrfs_init_compress();
2332 
2333 	err = btrfs_init_cachep();
2334 	if (err)
2335 		goto free_compress;
2336 
2337 	err = extent_io_init();
2338 	if (err)
2339 		goto free_cachep;
2340 
2341 	err = extent_map_init();
2342 	if (err)
2343 		goto free_extent_io;
2344 
2345 	err = ordered_data_init();
2346 	if (err)
2347 		goto free_extent_map;
2348 
2349 	err = btrfs_delayed_inode_init();
2350 	if (err)
2351 		goto free_ordered_data;
2352 
2353 	err = btrfs_auto_defrag_init();
2354 	if (err)
2355 		goto free_delayed_inode;
2356 
2357 	err = btrfs_delayed_ref_init();
2358 	if (err)
2359 		goto free_auto_defrag;
2360 
2361 	err = btrfs_prelim_ref_init();
2362 	if (err)
2363 		goto free_delayed_ref;
2364 
2365 	err = btrfs_end_io_wq_init();
2366 	if (err)
2367 		goto free_prelim_ref;
2368 
2369 	err = btrfs_interface_init();
2370 	if (err)
2371 		goto free_end_io_wq;
2372 
2373 	btrfs_init_lockdep();
2374 
2375 	btrfs_print_mod_info();
2376 
2377 	err = btrfs_run_sanity_tests();
2378 	if (err)
2379 		goto unregister_ioctl;
2380 
2381 	err = register_filesystem(&btrfs_fs_type);
2382 	if (err)
2383 		goto unregister_ioctl;
2384 
2385 	return 0;
2386 
2387 unregister_ioctl:
2388 	btrfs_interface_exit();
2389 free_end_io_wq:
2390 	btrfs_end_io_wq_exit();
2391 free_prelim_ref:
2392 	btrfs_prelim_ref_exit();
2393 free_delayed_ref:
2394 	btrfs_delayed_ref_exit();
2395 free_auto_defrag:
2396 	btrfs_auto_defrag_exit();
2397 free_delayed_inode:
2398 	btrfs_delayed_inode_exit();
2399 free_ordered_data:
2400 	ordered_data_exit();
2401 free_extent_map:
2402 	extent_map_exit();
2403 free_extent_io:
2404 	extent_io_exit();
2405 free_cachep:
2406 	btrfs_destroy_cachep();
2407 free_compress:
2408 	btrfs_exit_compress();
2409 	btrfs_exit_sysfs();
2410 free_hash:
2411 	btrfs_hash_exit();
2412 	return err;
2413 }
2414 
2415 static void __exit exit_btrfs_fs(void)
2416 {
2417 	btrfs_destroy_cachep();
2418 	btrfs_delayed_ref_exit();
2419 	btrfs_auto_defrag_exit();
2420 	btrfs_delayed_inode_exit();
2421 	btrfs_prelim_ref_exit();
2422 	ordered_data_exit();
2423 	extent_map_exit();
2424 	extent_io_exit();
2425 	btrfs_interface_exit();
2426 	btrfs_end_io_wq_exit();
2427 	unregister_filesystem(&btrfs_fs_type);
2428 	btrfs_exit_sysfs();
2429 	btrfs_cleanup_fs_uuids();
2430 	btrfs_exit_compress();
2431 	btrfs_hash_exit();
2432 }
2433 
2434 late_initcall(init_btrfs_fs);
2435 module_exit(exit_btrfs_fs)
2436 
2437 MODULE_LICENSE("GPL");
2438