xref: /linux/fs/btrfs/super.c (revision 79e24da00b1137031245f3341828e4215b1b5b59)
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 
67 static const struct super_operations btrfs_super_ops;
68 static struct file_system_type btrfs_fs_type;
69 
70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
71 
72 const 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 static void save_error_info(struct btrfs_fs_info *fs_info)
101 {
102 	/*
103 	 * today we only save the error info into ram.  Long term we'll
104 	 * also send it down to the disk
105 	 */
106 	set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
107 }
108 
109 /* btrfs handle error by forcing the filesystem readonly */
110 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
111 {
112 	struct super_block *sb = fs_info->sb;
113 
114 	if (sb->s_flags & MS_RDONLY)
115 		return;
116 
117 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
118 		sb->s_flags |= MS_RDONLY;
119 		btrfs_info(fs_info, "forced readonly");
120 		/*
121 		 * Note that a running device replace operation is not
122 		 * canceled here although there is no way to update
123 		 * the progress. It would add the risk of a deadlock,
124 		 * therefore the canceling is ommited. The only penalty
125 		 * is that some I/O remains active until the procedure
126 		 * completes. The next time when the filesystem is
127 		 * mounted writeable again, the device replace
128 		 * operation continues.
129 		 */
130 	}
131 }
132 
133 /*
134  * __btrfs_std_error decodes expected errors from the caller and
135  * invokes the approciate error response.
136  */
137 __cold
138 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
139 		       unsigned int line, int errno, const char *fmt, ...)
140 {
141 	struct super_block *sb = fs_info->sb;
142 #ifdef CONFIG_PRINTK
143 	const char *errstr;
144 #endif
145 
146 	/*
147 	 * Special case: if the error is EROFS, and we're already
148 	 * under MS_RDONLY, then it is safe here.
149 	 */
150 	if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
151   		return;
152 
153 #ifdef CONFIG_PRINTK
154 	errstr = btrfs_decode_error(errno);
155 	if (fmt) {
156 		struct va_format vaf;
157 		va_list args;
158 
159 		va_start(args, fmt);
160 		vaf.fmt = fmt;
161 		vaf.va = &args;
162 
163 		printk(KERN_CRIT
164 			"BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
165 			sb->s_id, function, line, errno, errstr, &vaf);
166 		va_end(args);
167 	} else {
168 		printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
169 			sb->s_id, function, line, errno, errstr);
170 	}
171 #endif
172 
173 	/* Don't go through full error handling during mount */
174 	save_error_info(fs_info);
175 	if (sb->s_flags & MS_BORN)
176 		btrfs_handle_error(fs_info);
177 }
178 
179 #ifdef CONFIG_PRINTK
180 static const char * const logtypes[] = {
181 	"emergency",
182 	"alert",
183 	"critical",
184 	"error",
185 	"warning",
186 	"notice",
187 	"info",
188 	"debug",
189 };
190 
191 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
192 {
193 	struct super_block *sb = fs_info->sb;
194 	char lvl[4];
195 	struct va_format vaf;
196 	va_list args;
197 	const char *type = logtypes[4];
198 	int kern_level;
199 
200 	va_start(args, fmt);
201 
202 	kern_level = printk_get_level(fmt);
203 	if (kern_level) {
204 		size_t size = printk_skip_level(fmt) - fmt;
205 		memcpy(lvl, fmt,  size);
206 		lvl[size] = '\0';
207 		fmt += size;
208 		type = logtypes[kern_level - '0'];
209 	} else
210 		*lvl = '\0';
211 
212 	vaf.fmt = fmt;
213 	vaf.va = &args;
214 
215 	printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
216 
217 	va_end(args);
218 }
219 #endif
220 
221 /*
222  * We only mark the transaction aborted and then set the file system read-only.
223  * This will prevent new transactions from starting or trying to join this
224  * one.
225  *
226  * This means that error recovery at the call site is limited to freeing
227  * any local memory allocations and passing the error code up without
228  * further cleanup. The transaction should complete as it normally would
229  * in the call path but will return -EIO.
230  *
231  * We'll complete the cleanup in btrfs_end_transaction and
232  * btrfs_commit_transaction.
233  */
234 __cold
235 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
236 			       struct btrfs_root *root, const char *function,
237 			       unsigned int line, int errno)
238 {
239 	trans->aborted = errno;
240 	/* Nothing used. The other threads that have joined this
241 	 * transaction may be able to continue. */
242 	if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
243 		const char *errstr;
244 
245 		errstr = btrfs_decode_error(errno);
246 		btrfs_warn(root->fs_info,
247 		           "%s:%d: Aborting unused transaction(%s).",
248 		           function, line, errstr);
249 		return;
250 	}
251 	ACCESS_ONCE(trans->transaction->aborted) = errno;
252 	/* Wake up anybody who may be waiting on this transaction */
253 	wake_up(&root->fs_info->transaction_wait);
254 	wake_up(&root->fs_info->transaction_blocked_wait);
255 	__btrfs_std_error(root->fs_info, function, line, errno, NULL);
256 }
257 /*
258  * __btrfs_panic decodes unexpected, fatal errors from the caller,
259  * issues an alert, and either panics or BUGs, depending on mount options.
260  */
261 __cold
262 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
263 		   unsigned int line, int errno, const char *fmt, ...)
264 {
265 	char *s_id = "<unknown>";
266 	const char *errstr;
267 	struct va_format vaf = { .fmt = fmt };
268 	va_list args;
269 
270 	if (fs_info)
271 		s_id = fs_info->sb->s_id;
272 
273 	va_start(args, fmt);
274 	vaf.va = &args;
275 
276 	errstr = btrfs_decode_error(errno);
277 	if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
278 		panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
279 			s_id, function, line, &vaf, errno, errstr);
280 
281 	btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
282 		   function, line, &vaf, errno, errstr);
283 	va_end(args);
284 	/* Caller calls BUG() */
285 }
286 
287 static void btrfs_put_super(struct super_block *sb)
288 {
289 	close_ctree(btrfs_sb(sb)->tree_root);
290 }
291 
292 enum {
293 	Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
294 	Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
295 	Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
296 	Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
297 	Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
298 	Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
299 	Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
300 	Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
301 	Opt_skip_balance, Opt_check_integrity,
302 	Opt_check_integrity_including_extent_data,
303 	Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
304 	Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
305 	Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
306 	Opt_datasum, Opt_treelog, Opt_noinode_cache,
307 #ifdef CONFIG_BTRFS_DEBUG
308 	Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
309 #endif
310 	Opt_err,
311 };
312 
313 static const match_table_t tokens = {
314 	{Opt_degraded, "degraded"},
315 	{Opt_subvol, "subvol=%s"},
316 	{Opt_subvolid, "subvolid=%s"},
317 	{Opt_device, "device=%s"},
318 	{Opt_nodatasum, "nodatasum"},
319 	{Opt_datasum, "datasum"},
320 	{Opt_nodatacow, "nodatacow"},
321 	{Opt_datacow, "datacow"},
322 	{Opt_nobarrier, "nobarrier"},
323 	{Opt_barrier, "barrier"},
324 	{Opt_max_inline, "max_inline=%s"},
325 	{Opt_alloc_start, "alloc_start=%s"},
326 	{Opt_thread_pool, "thread_pool=%d"},
327 	{Opt_compress, "compress"},
328 	{Opt_compress_type, "compress=%s"},
329 	{Opt_compress_force, "compress-force"},
330 	{Opt_compress_force_type, "compress-force=%s"},
331 	{Opt_ssd, "ssd"},
332 	{Opt_ssd_spread, "ssd_spread"},
333 	{Opt_nossd, "nossd"},
334 	{Opt_acl, "acl"},
335 	{Opt_noacl, "noacl"},
336 	{Opt_notreelog, "notreelog"},
337 	{Opt_treelog, "treelog"},
338 	{Opt_flushoncommit, "flushoncommit"},
339 	{Opt_noflushoncommit, "noflushoncommit"},
340 	{Opt_ratio, "metadata_ratio=%d"},
341 	{Opt_discard, "discard"},
342 	{Opt_nodiscard, "nodiscard"},
343 	{Opt_space_cache, "space_cache"},
344 	{Opt_space_cache_version, "space_cache=%s"},
345 	{Opt_clear_cache, "clear_cache"},
346 	{Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
347 	{Opt_enospc_debug, "enospc_debug"},
348 	{Opt_noenospc_debug, "noenospc_debug"},
349 	{Opt_subvolrootid, "subvolrootid=%d"},
350 	{Opt_defrag, "autodefrag"},
351 	{Opt_nodefrag, "noautodefrag"},
352 	{Opt_inode_cache, "inode_cache"},
353 	{Opt_noinode_cache, "noinode_cache"},
354 	{Opt_no_space_cache, "nospace_cache"},
355 	{Opt_recovery, "recovery"},
356 	{Opt_skip_balance, "skip_balance"},
357 	{Opt_check_integrity, "check_int"},
358 	{Opt_check_integrity_including_extent_data, "check_int_data"},
359 	{Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
360 	{Opt_rescan_uuid_tree, "rescan_uuid_tree"},
361 	{Opt_fatal_errors, "fatal_errors=%s"},
362 	{Opt_commit_interval, "commit=%d"},
363 #ifdef CONFIG_BTRFS_DEBUG
364 	{Opt_fragment_data, "fragment=data"},
365 	{Opt_fragment_metadata, "fragment=metadata"},
366 	{Opt_fragment_all, "fragment=all"},
367 #endif
368 	{Opt_err, NULL},
369 };
370 
371 /*
372  * Regular mount options parser.  Everything that is needed only when
373  * reading in a new superblock is parsed here.
374  * XXX JDM: This needs to be cleaned up for remount.
375  */
376 int btrfs_parse_options(struct btrfs_root *root, char *options)
377 {
378 	struct btrfs_fs_info *info = root->fs_info;
379 	substring_t args[MAX_OPT_ARGS];
380 	char *p, *num, *orig = NULL;
381 	u64 cache_gen;
382 	int intarg;
383 	int ret = 0;
384 	char *compress_type;
385 	bool compress_force = false;
386 	enum btrfs_compression_type saved_compress_type;
387 	bool saved_compress_force;
388 	int no_compress = 0;
389 
390 	cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
391 	if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
392 		btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
393 	else if (cache_gen)
394 		btrfs_set_opt(info->mount_opt, SPACE_CACHE);
395 
396 	if (!options)
397 		goto out;
398 
399 	/*
400 	 * strsep changes the string, duplicate it because parse_options
401 	 * gets called twice
402 	 */
403 	options = kstrdup(options, GFP_NOFS);
404 	if (!options)
405 		return -ENOMEM;
406 
407 	orig = options;
408 
409 	while ((p = strsep(&options, ",")) != NULL) {
410 		int token;
411 		if (!*p)
412 			continue;
413 
414 		token = match_token(p, tokens, args);
415 		switch (token) {
416 		case Opt_degraded:
417 			btrfs_info(root->fs_info, "allowing degraded mounts");
418 			btrfs_set_opt(info->mount_opt, DEGRADED);
419 			break;
420 		case Opt_subvol:
421 		case Opt_subvolid:
422 		case Opt_subvolrootid:
423 		case Opt_device:
424 			/*
425 			 * These are parsed by btrfs_parse_early_options
426 			 * and can be happily ignored here.
427 			 */
428 			break;
429 		case Opt_nodatasum:
430 			btrfs_set_and_info(root, NODATASUM,
431 					   "setting nodatasum");
432 			break;
433 		case Opt_datasum:
434 			if (btrfs_test_opt(root, NODATASUM)) {
435 				if (btrfs_test_opt(root, NODATACOW))
436 					btrfs_info(root->fs_info, "setting datasum, datacow enabled");
437 				else
438 					btrfs_info(root->fs_info, "setting datasum");
439 			}
440 			btrfs_clear_opt(info->mount_opt, NODATACOW);
441 			btrfs_clear_opt(info->mount_opt, NODATASUM);
442 			break;
443 		case Opt_nodatacow:
444 			if (!btrfs_test_opt(root, NODATACOW)) {
445 				if (!btrfs_test_opt(root, COMPRESS) ||
446 				    !btrfs_test_opt(root, FORCE_COMPRESS)) {
447 					btrfs_info(root->fs_info,
448 						   "setting nodatacow, compression disabled");
449 				} else {
450 					btrfs_info(root->fs_info, "setting nodatacow");
451 				}
452 			}
453 			btrfs_clear_opt(info->mount_opt, COMPRESS);
454 			btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
455 			btrfs_set_opt(info->mount_opt, NODATACOW);
456 			btrfs_set_opt(info->mount_opt, NODATASUM);
457 			break;
458 		case Opt_datacow:
459 			btrfs_clear_and_info(root, NODATACOW,
460 					     "setting datacow");
461 			break;
462 		case Opt_compress_force:
463 		case Opt_compress_force_type:
464 			compress_force = true;
465 			/* Fallthrough */
466 		case Opt_compress:
467 		case Opt_compress_type:
468 			saved_compress_type = btrfs_test_opt(root, COMPRESS) ?
469 				info->compress_type : BTRFS_COMPRESS_NONE;
470 			saved_compress_force =
471 				btrfs_test_opt(root, FORCE_COMPRESS);
472 			if (token == Opt_compress ||
473 			    token == Opt_compress_force ||
474 			    strcmp(args[0].from, "zlib") == 0) {
475 				compress_type = "zlib";
476 				info->compress_type = BTRFS_COMPRESS_ZLIB;
477 				btrfs_set_opt(info->mount_opt, COMPRESS);
478 				btrfs_clear_opt(info->mount_opt, NODATACOW);
479 				btrfs_clear_opt(info->mount_opt, NODATASUM);
480 				no_compress = 0;
481 			} else if (strcmp(args[0].from, "lzo") == 0) {
482 				compress_type = "lzo";
483 				info->compress_type = BTRFS_COMPRESS_LZO;
484 				btrfs_set_opt(info->mount_opt, COMPRESS);
485 				btrfs_clear_opt(info->mount_opt, NODATACOW);
486 				btrfs_clear_opt(info->mount_opt, NODATASUM);
487 				btrfs_set_fs_incompat(info, COMPRESS_LZO);
488 				no_compress = 0;
489 			} else if (strncmp(args[0].from, "no", 2) == 0) {
490 				compress_type = "no";
491 				btrfs_clear_opt(info->mount_opt, COMPRESS);
492 				btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
493 				compress_force = false;
494 				no_compress++;
495 			} else {
496 				ret = -EINVAL;
497 				goto out;
498 			}
499 
500 			if (compress_force) {
501 				btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
502 			} else {
503 				/*
504 				 * If we remount from compress-force=xxx to
505 				 * compress=xxx, we need clear FORCE_COMPRESS
506 				 * flag, otherwise, there is no way for users
507 				 * to disable forcible compression separately.
508 				 */
509 				btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
510 			}
511 			if ((btrfs_test_opt(root, COMPRESS) &&
512 			     (info->compress_type != saved_compress_type ||
513 			      compress_force != saved_compress_force)) ||
514 			    (!btrfs_test_opt(root, COMPRESS) &&
515 			     no_compress == 1)) {
516 				btrfs_info(root->fs_info,
517 					   "%s %s compression",
518 					   (compress_force) ? "force" : "use",
519 					   compress_type);
520 			}
521 			compress_force = false;
522 			break;
523 		case Opt_ssd:
524 			btrfs_set_and_info(root, SSD,
525 					   "use ssd allocation scheme");
526 			break;
527 		case Opt_ssd_spread:
528 			btrfs_set_and_info(root, SSD_SPREAD,
529 					   "use spread ssd allocation scheme");
530 			btrfs_set_opt(info->mount_opt, SSD);
531 			break;
532 		case Opt_nossd:
533 			btrfs_set_and_info(root, NOSSD,
534 					     "not using ssd allocation scheme");
535 			btrfs_clear_opt(info->mount_opt, SSD);
536 			break;
537 		case Opt_barrier:
538 			btrfs_clear_and_info(root, NOBARRIER,
539 					     "turning on barriers");
540 			break;
541 		case Opt_nobarrier:
542 			btrfs_set_and_info(root, NOBARRIER,
543 					   "turning off barriers");
544 			break;
545 		case Opt_thread_pool:
546 			ret = match_int(&args[0], &intarg);
547 			if (ret) {
548 				goto out;
549 			} else if (intarg > 0) {
550 				info->thread_pool_size = intarg;
551 			} else {
552 				ret = -EINVAL;
553 				goto out;
554 			}
555 			break;
556 		case Opt_max_inline:
557 			num = match_strdup(&args[0]);
558 			if (num) {
559 				info->max_inline = memparse(num, NULL);
560 				kfree(num);
561 
562 				if (info->max_inline) {
563 					info->max_inline = min_t(u64,
564 						info->max_inline,
565 						root->sectorsize);
566 				}
567 				btrfs_info(root->fs_info, "max_inline at %llu",
568 					info->max_inline);
569 			} else {
570 				ret = -ENOMEM;
571 				goto out;
572 			}
573 			break;
574 		case Opt_alloc_start:
575 			num = match_strdup(&args[0]);
576 			if (num) {
577 				mutex_lock(&info->chunk_mutex);
578 				info->alloc_start = memparse(num, NULL);
579 				mutex_unlock(&info->chunk_mutex);
580 				kfree(num);
581 				btrfs_info(root->fs_info, "allocations start at %llu",
582 					info->alloc_start);
583 			} else {
584 				ret = -ENOMEM;
585 				goto out;
586 			}
587 			break;
588 		case Opt_acl:
589 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
590 			root->fs_info->sb->s_flags |= MS_POSIXACL;
591 			break;
592 #else
593 			btrfs_err(root->fs_info,
594 				"support for ACL not compiled in!");
595 			ret = -EINVAL;
596 			goto out;
597 #endif
598 		case Opt_noacl:
599 			root->fs_info->sb->s_flags &= ~MS_POSIXACL;
600 			break;
601 		case Opt_notreelog:
602 			btrfs_set_and_info(root, NOTREELOG,
603 					   "disabling tree log");
604 			break;
605 		case Opt_treelog:
606 			btrfs_clear_and_info(root, NOTREELOG,
607 					     "enabling tree log");
608 			break;
609 		case Opt_flushoncommit:
610 			btrfs_set_and_info(root, FLUSHONCOMMIT,
611 					   "turning on flush-on-commit");
612 			break;
613 		case Opt_noflushoncommit:
614 			btrfs_clear_and_info(root, FLUSHONCOMMIT,
615 					     "turning off flush-on-commit");
616 			break;
617 		case Opt_ratio:
618 			ret = match_int(&args[0], &intarg);
619 			if (ret) {
620 				goto out;
621 			} else if (intarg >= 0) {
622 				info->metadata_ratio = intarg;
623 				btrfs_info(root->fs_info, "metadata ratio %d",
624 				       info->metadata_ratio);
625 			} else {
626 				ret = -EINVAL;
627 				goto out;
628 			}
629 			break;
630 		case Opt_discard:
631 			btrfs_set_and_info(root, DISCARD,
632 					   "turning on discard");
633 			break;
634 		case Opt_nodiscard:
635 			btrfs_clear_and_info(root, DISCARD,
636 					     "turning off discard");
637 			break;
638 		case Opt_space_cache:
639 		case Opt_space_cache_version:
640 			if (token == Opt_space_cache ||
641 			    strcmp(args[0].from, "v1") == 0) {
642 				btrfs_clear_opt(root->fs_info->mount_opt,
643 						FREE_SPACE_TREE);
644 				btrfs_set_and_info(root, SPACE_CACHE,
645 						   "enabling disk space caching");
646 			} else if (strcmp(args[0].from, "v2") == 0) {
647 				btrfs_clear_opt(root->fs_info->mount_opt,
648 						SPACE_CACHE);
649 				btrfs_set_and_info(root, FREE_SPACE_TREE,
650 						   "enabling free space tree");
651 			} else {
652 				ret = -EINVAL;
653 				goto out;
654 			}
655 			break;
656 		case Opt_rescan_uuid_tree:
657 			btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
658 			break;
659 		case Opt_no_space_cache:
660 			if (btrfs_test_opt(root, SPACE_CACHE)) {
661 				btrfs_clear_and_info(root, SPACE_CACHE,
662 						     "disabling disk space caching");
663 			}
664 			if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
665 				btrfs_clear_and_info(root, FREE_SPACE_TREE,
666 						     "disabling free space tree");
667 			}
668 			break;
669 		case Opt_inode_cache:
670 			btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
671 					   "enabling inode map caching");
672 			break;
673 		case Opt_noinode_cache:
674 			btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
675 					     "disabling inode map caching");
676 			break;
677 		case Opt_clear_cache:
678 			btrfs_set_and_info(root, CLEAR_CACHE,
679 					   "force clearing of disk cache");
680 			break;
681 		case Opt_user_subvol_rm_allowed:
682 			btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
683 			break;
684 		case Opt_enospc_debug:
685 			btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
686 			break;
687 		case Opt_noenospc_debug:
688 			btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
689 			break;
690 		case Opt_defrag:
691 			btrfs_set_and_info(root, AUTO_DEFRAG,
692 					   "enabling auto defrag");
693 			break;
694 		case Opt_nodefrag:
695 			btrfs_clear_and_info(root, AUTO_DEFRAG,
696 					     "disabling auto defrag");
697 			break;
698 		case Opt_recovery:
699 			btrfs_info(root->fs_info, "enabling auto recovery");
700 			btrfs_set_opt(info->mount_opt, RECOVERY);
701 			break;
702 		case Opt_skip_balance:
703 			btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
704 			break;
705 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
706 		case Opt_check_integrity_including_extent_data:
707 			btrfs_info(root->fs_info,
708 				   "enabling check integrity including extent data");
709 			btrfs_set_opt(info->mount_opt,
710 				      CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
711 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
712 			break;
713 		case Opt_check_integrity:
714 			btrfs_info(root->fs_info, "enabling check integrity");
715 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
716 			break;
717 		case Opt_check_integrity_print_mask:
718 			ret = match_int(&args[0], &intarg);
719 			if (ret) {
720 				goto out;
721 			} else if (intarg >= 0) {
722 				info->check_integrity_print_mask = intarg;
723 				btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
724 				       info->check_integrity_print_mask);
725 			} else {
726 				ret = -EINVAL;
727 				goto out;
728 			}
729 			break;
730 #else
731 		case Opt_check_integrity_including_extent_data:
732 		case Opt_check_integrity:
733 		case Opt_check_integrity_print_mask:
734 			btrfs_err(root->fs_info,
735 				"support for check_integrity* not compiled in!");
736 			ret = -EINVAL;
737 			goto out;
738 #endif
739 		case Opt_fatal_errors:
740 			if (strcmp(args[0].from, "panic") == 0)
741 				btrfs_set_opt(info->mount_opt,
742 					      PANIC_ON_FATAL_ERROR);
743 			else if (strcmp(args[0].from, "bug") == 0)
744 				btrfs_clear_opt(info->mount_opt,
745 					      PANIC_ON_FATAL_ERROR);
746 			else {
747 				ret = -EINVAL;
748 				goto out;
749 			}
750 			break;
751 		case Opt_commit_interval:
752 			intarg = 0;
753 			ret = match_int(&args[0], &intarg);
754 			if (ret < 0) {
755 				btrfs_err(root->fs_info, "invalid commit interval");
756 				ret = -EINVAL;
757 				goto out;
758 			}
759 			if (intarg > 0) {
760 				if (intarg > 300) {
761 					btrfs_warn(root->fs_info, "excessive commit interval %d",
762 							intarg);
763 				}
764 				info->commit_interval = intarg;
765 			} else {
766 				btrfs_info(root->fs_info, "using default commit interval %ds",
767 				    BTRFS_DEFAULT_COMMIT_INTERVAL);
768 				info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
769 			}
770 			break;
771 #ifdef CONFIG_BTRFS_DEBUG
772 		case Opt_fragment_all:
773 			btrfs_info(root->fs_info, "fragmenting all space");
774 			btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
775 			btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
776 			break;
777 		case Opt_fragment_metadata:
778 			btrfs_info(root->fs_info, "fragmenting metadata");
779 			btrfs_set_opt(info->mount_opt,
780 				      FRAGMENT_METADATA);
781 			break;
782 		case Opt_fragment_data:
783 			btrfs_info(root->fs_info, "fragmenting data");
784 			btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
785 			break;
786 #endif
787 		case Opt_err:
788 			btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
789 			ret = -EINVAL;
790 			goto out;
791 		default:
792 			break;
793 		}
794 	}
795 out:
796 	if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
797 	    !btrfs_test_opt(root, FREE_SPACE_TREE) &&
798 	    !btrfs_test_opt(root, CLEAR_CACHE)) {
799 		btrfs_err(root->fs_info, "cannot disable free space tree");
800 		ret = -EINVAL;
801 
802 	}
803 	if (!ret && btrfs_test_opt(root, SPACE_CACHE))
804 		btrfs_info(root->fs_info, "disk space caching is enabled");
805 	if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
806 		btrfs_info(root->fs_info, "using free space tree");
807 	kfree(orig);
808 	return ret;
809 }
810 
811 /*
812  * Parse mount options that are required early in the mount process.
813  *
814  * All other options will be parsed on much later in the mount process and
815  * only when we need to allocate a new super block.
816  */
817 static int btrfs_parse_early_options(const char *options, fmode_t flags,
818 		void *holder, char **subvol_name, u64 *subvol_objectid,
819 		struct btrfs_fs_devices **fs_devices)
820 {
821 	substring_t args[MAX_OPT_ARGS];
822 	char *device_name, *opts, *orig, *p;
823 	char *num = NULL;
824 	int error = 0;
825 
826 	if (!options)
827 		return 0;
828 
829 	/*
830 	 * strsep changes the string, duplicate it because parse_options
831 	 * gets called twice
832 	 */
833 	opts = kstrdup(options, GFP_KERNEL);
834 	if (!opts)
835 		return -ENOMEM;
836 	orig = opts;
837 
838 	while ((p = strsep(&opts, ",")) != NULL) {
839 		int token;
840 		if (!*p)
841 			continue;
842 
843 		token = match_token(p, tokens, args);
844 		switch (token) {
845 		case Opt_subvol:
846 			kfree(*subvol_name);
847 			*subvol_name = match_strdup(&args[0]);
848 			if (!*subvol_name) {
849 				error = -ENOMEM;
850 				goto out;
851 			}
852 			break;
853 		case Opt_subvolid:
854 			num = match_strdup(&args[0]);
855 			if (num) {
856 				*subvol_objectid = memparse(num, NULL);
857 				kfree(num);
858 				/* we want the original fs_tree */
859 				if (!*subvol_objectid)
860 					*subvol_objectid =
861 						BTRFS_FS_TREE_OBJECTID;
862 			} else {
863 				error = -EINVAL;
864 				goto out;
865 			}
866 			break;
867 		case Opt_subvolrootid:
868 			printk(KERN_WARNING
869 				"BTRFS: 'subvolrootid' mount option is deprecated and has "
870 				"no effect\n");
871 			break;
872 		case Opt_device:
873 			device_name = match_strdup(&args[0]);
874 			if (!device_name) {
875 				error = -ENOMEM;
876 				goto out;
877 			}
878 			error = btrfs_scan_one_device(device_name,
879 					flags, holder, fs_devices);
880 			kfree(device_name);
881 			if (error)
882 				goto out;
883 			break;
884 		default:
885 			break;
886 		}
887 	}
888 
889 out:
890 	kfree(orig);
891 	return error;
892 }
893 
894 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
895 					   u64 subvol_objectid)
896 {
897 	struct btrfs_root *root = fs_info->tree_root;
898 	struct btrfs_root *fs_root;
899 	struct btrfs_root_ref *root_ref;
900 	struct btrfs_inode_ref *inode_ref;
901 	struct btrfs_key key;
902 	struct btrfs_path *path = NULL;
903 	char *name = NULL, *ptr;
904 	u64 dirid;
905 	int len;
906 	int ret;
907 
908 	path = btrfs_alloc_path();
909 	if (!path) {
910 		ret = -ENOMEM;
911 		goto err;
912 	}
913 	path->leave_spinning = 1;
914 
915 	name = kmalloc(PATH_MAX, GFP_NOFS);
916 	if (!name) {
917 		ret = -ENOMEM;
918 		goto err;
919 	}
920 	ptr = name + PATH_MAX - 1;
921 	ptr[0] = '\0';
922 
923 	/*
924 	 * Walk up the subvolume trees in the tree of tree roots by root
925 	 * backrefs until we hit the top-level subvolume.
926 	 */
927 	while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
928 		key.objectid = subvol_objectid;
929 		key.type = BTRFS_ROOT_BACKREF_KEY;
930 		key.offset = (u64)-1;
931 
932 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
933 		if (ret < 0) {
934 			goto err;
935 		} else if (ret > 0) {
936 			ret = btrfs_previous_item(root, path, subvol_objectid,
937 						  BTRFS_ROOT_BACKREF_KEY);
938 			if (ret < 0) {
939 				goto err;
940 			} else if (ret > 0) {
941 				ret = -ENOENT;
942 				goto err;
943 			}
944 		}
945 
946 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
947 		subvol_objectid = key.offset;
948 
949 		root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
950 					  struct btrfs_root_ref);
951 		len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
952 		ptr -= len + 1;
953 		if (ptr < name) {
954 			ret = -ENAMETOOLONG;
955 			goto err;
956 		}
957 		read_extent_buffer(path->nodes[0], ptr + 1,
958 				   (unsigned long)(root_ref + 1), len);
959 		ptr[0] = '/';
960 		dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
961 		btrfs_release_path(path);
962 
963 		key.objectid = subvol_objectid;
964 		key.type = BTRFS_ROOT_ITEM_KEY;
965 		key.offset = (u64)-1;
966 		fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
967 		if (IS_ERR(fs_root)) {
968 			ret = PTR_ERR(fs_root);
969 			goto err;
970 		}
971 
972 		/*
973 		 * Walk up the filesystem tree by inode refs until we hit the
974 		 * root directory.
975 		 */
976 		while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
977 			key.objectid = dirid;
978 			key.type = BTRFS_INODE_REF_KEY;
979 			key.offset = (u64)-1;
980 
981 			ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
982 			if (ret < 0) {
983 				goto err;
984 			} else if (ret > 0) {
985 				ret = btrfs_previous_item(fs_root, path, dirid,
986 							  BTRFS_INODE_REF_KEY);
987 				if (ret < 0) {
988 					goto err;
989 				} else if (ret > 0) {
990 					ret = -ENOENT;
991 					goto err;
992 				}
993 			}
994 
995 			btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
996 			dirid = key.offset;
997 
998 			inode_ref = btrfs_item_ptr(path->nodes[0],
999 						   path->slots[0],
1000 						   struct btrfs_inode_ref);
1001 			len = btrfs_inode_ref_name_len(path->nodes[0],
1002 						       inode_ref);
1003 			ptr -= len + 1;
1004 			if (ptr < name) {
1005 				ret = -ENAMETOOLONG;
1006 				goto err;
1007 			}
1008 			read_extent_buffer(path->nodes[0], ptr + 1,
1009 					   (unsigned long)(inode_ref + 1), len);
1010 			ptr[0] = '/';
1011 			btrfs_release_path(path);
1012 		}
1013 	}
1014 
1015 	btrfs_free_path(path);
1016 	if (ptr == name + PATH_MAX - 1) {
1017 		name[0] = '/';
1018 		name[1] = '\0';
1019 	} else {
1020 		memmove(name, ptr, name + PATH_MAX - ptr);
1021 	}
1022 	return name;
1023 
1024 err:
1025 	btrfs_free_path(path);
1026 	kfree(name);
1027 	return ERR_PTR(ret);
1028 }
1029 
1030 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1031 {
1032 	struct btrfs_root *root = fs_info->tree_root;
1033 	struct btrfs_dir_item *di;
1034 	struct btrfs_path *path;
1035 	struct btrfs_key location;
1036 	u64 dir_id;
1037 
1038 	path = btrfs_alloc_path();
1039 	if (!path)
1040 		return -ENOMEM;
1041 	path->leave_spinning = 1;
1042 
1043 	/*
1044 	 * Find the "default" dir item which points to the root item that we
1045 	 * will mount by default if we haven't been given a specific subvolume
1046 	 * to mount.
1047 	 */
1048 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
1049 	di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1050 	if (IS_ERR(di)) {
1051 		btrfs_free_path(path);
1052 		return PTR_ERR(di);
1053 	}
1054 	if (!di) {
1055 		/*
1056 		 * Ok the default dir item isn't there.  This is weird since
1057 		 * it's always been there, but don't freak out, just try and
1058 		 * mount the top-level subvolume.
1059 		 */
1060 		btrfs_free_path(path);
1061 		*objectid = BTRFS_FS_TREE_OBJECTID;
1062 		return 0;
1063 	}
1064 
1065 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1066 	btrfs_free_path(path);
1067 	*objectid = location.objectid;
1068 	return 0;
1069 }
1070 
1071 static int btrfs_fill_super(struct super_block *sb,
1072 			    struct btrfs_fs_devices *fs_devices,
1073 			    void *data, int silent)
1074 {
1075 	struct inode *inode;
1076 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1077 	struct btrfs_key key;
1078 	int err;
1079 
1080 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1081 	sb->s_magic = BTRFS_SUPER_MAGIC;
1082 	sb->s_op = &btrfs_super_ops;
1083 	sb->s_d_op = &btrfs_dentry_operations;
1084 	sb->s_export_op = &btrfs_export_ops;
1085 	sb->s_xattr = btrfs_xattr_handlers;
1086 	sb->s_time_gran = 1;
1087 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1088 	sb->s_flags |= MS_POSIXACL;
1089 #endif
1090 	sb->s_flags |= MS_I_VERSION;
1091 	sb->s_iflags |= SB_I_CGROUPWB;
1092 	err = open_ctree(sb, fs_devices, (char *)data);
1093 	if (err) {
1094 		printk(KERN_ERR "BTRFS: open_ctree failed\n");
1095 		return err;
1096 	}
1097 
1098 	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1099 	key.type = BTRFS_INODE_ITEM_KEY;
1100 	key.offset = 0;
1101 	inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1102 	if (IS_ERR(inode)) {
1103 		err = PTR_ERR(inode);
1104 		goto fail_close;
1105 	}
1106 
1107 	sb->s_root = d_make_root(inode);
1108 	if (!sb->s_root) {
1109 		err = -ENOMEM;
1110 		goto fail_close;
1111 	}
1112 
1113 	save_mount_options(sb, data);
1114 	cleancache_init_fs(sb);
1115 	sb->s_flags |= MS_ACTIVE;
1116 	return 0;
1117 
1118 fail_close:
1119 	close_ctree(fs_info->tree_root);
1120 	return err;
1121 }
1122 
1123 int btrfs_sync_fs(struct super_block *sb, int wait)
1124 {
1125 	struct btrfs_trans_handle *trans;
1126 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1127 	struct btrfs_root *root = fs_info->tree_root;
1128 
1129 	trace_btrfs_sync_fs(wait);
1130 
1131 	if (!wait) {
1132 		filemap_flush(fs_info->btree_inode->i_mapping);
1133 		return 0;
1134 	}
1135 
1136 	btrfs_wait_ordered_roots(fs_info, -1);
1137 
1138 	trans = btrfs_attach_transaction_barrier(root);
1139 	if (IS_ERR(trans)) {
1140 		/* no transaction, don't bother */
1141 		if (PTR_ERR(trans) == -ENOENT) {
1142 			/*
1143 			 * Exit unless we have some pending changes
1144 			 * that need to go through commit
1145 			 */
1146 			if (fs_info->pending_changes == 0)
1147 				return 0;
1148 			/*
1149 			 * A non-blocking test if the fs is frozen. We must not
1150 			 * start a new transaction here otherwise a deadlock
1151 			 * happens. The pending operations are delayed to the
1152 			 * next commit after thawing.
1153 			 */
1154 			if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1155 				__sb_end_write(sb, SB_FREEZE_WRITE);
1156 			else
1157 				return 0;
1158 			trans = btrfs_start_transaction(root, 0);
1159 		}
1160 		if (IS_ERR(trans))
1161 			return PTR_ERR(trans);
1162 	}
1163 	return btrfs_commit_transaction(trans, root);
1164 }
1165 
1166 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1167 {
1168 	struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1169 	struct btrfs_root *root = info->tree_root;
1170 	char *compress_type;
1171 
1172 	if (btrfs_test_opt(root, DEGRADED))
1173 		seq_puts(seq, ",degraded");
1174 	if (btrfs_test_opt(root, NODATASUM))
1175 		seq_puts(seq, ",nodatasum");
1176 	if (btrfs_test_opt(root, NODATACOW))
1177 		seq_puts(seq, ",nodatacow");
1178 	if (btrfs_test_opt(root, NOBARRIER))
1179 		seq_puts(seq, ",nobarrier");
1180 	if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1181 		seq_printf(seq, ",max_inline=%llu", info->max_inline);
1182 	if (info->alloc_start != 0)
1183 		seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1184 	if (info->thread_pool_size !=  min_t(unsigned long,
1185 					     num_online_cpus() + 2, 8))
1186 		seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1187 	if (btrfs_test_opt(root, COMPRESS)) {
1188 		if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1189 			compress_type = "zlib";
1190 		else
1191 			compress_type = "lzo";
1192 		if (btrfs_test_opt(root, FORCE_COMPRESS))
1193 			seq_printf(seq, ",compress-force=%s", compress_type);
1194 		else
1195 			seq_printf(seq, ",compress=%s", compress_type);
1196 	}
1197 	if (btrfs_test_opt(root, NOSSD))
1198 		seq_puts(seq, ",nossd");
1199 	if (btrfs_test_opt(root, SSD_SPREAD))
1200 		seq_puts(seq, ",ssd_spread");
1201 	else if (btrfs_test_opt(root, SSD))
1202 		seq_puts(seq, ",ssd");
1203 	if (btrfs_test_opt(root, NOTREELOG))
1204 		seq_puts(seq, ",notreelog");
1205 	if (btrfs_test_opt(root, FLUSHONCOMMIT))
1206 		seq_puts(seq, ",flushoncommit");
1207 	if (btrfs_test_opt(root, DISCARD))
1208 		seq_puts(seq, ",discard");
1209 	if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1210 		seq_puts(seq, ",noacl");
1211 	if (btrfs_test_opt(root, SPACE_CACHE))
1212 		seq_puts(seq, ",space_cache");
1213 	else if (btrfs_test_opt(root, FREE_SPACE_TREE))
1214 		seq_puts(seq, ",space_cache=v2");
1215 	else
1216 		seq_puts(seq, ",nospace_cache");
1217 	if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1218 		seq_puts(seq, ",rescan_uuid_tree");
1219 	if (btrfs_test_opt(root, CLEAR_CACHE))
1220 		seq_puts(seq, ",clear_cache");
1221 	if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1222 		seq_puts(seq, ",user_subvol_rm_allowed");
1223 	if (btrfs_test_opt(root, ENOSPC_DEBUG))
1224 		seq_puts(seq, ",enospc_debug");
1225 	if (btrfs_test_opt(root, AUTO_DEFRAG))
1226 		seq_puts(seq, ",autodefrag");
1227 	if (btrfs_test_opt(root, INODE_MAP_CACHE))
1228 		seq_puts(seq, ",inode_cache");
1229 	if (btrfs_test_opt(root, SKIP_BALANCE))
1230 		seq_puts(seq, ",skip_balance");
1231 	if (btrfs_test_opt(root, RECOVERY))
1232 		seq_puts(seq, ",recovery");
1233 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1234 	if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1235 		seq_puts(seq, ",check_int_data");
1236 	else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1237 		seq_puts(seq, ",check_int");
1238 	if (info->check_integrity_print_mask)
1239 		seq_printf(seq, ",check_int_print_mask=%d",
1240 				info->check_integrity_print_mask);
1241 #endif
1242 	if (info->metadata_ratio)
1243 		seq_printf(seq, ",metadata_ratio=%d",
1244 				info->metadata_ratio);
1245 	if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1246 		seq_puts(seq, ",fatal_errors=panic");
1247 	if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1248 		seq_printf(seq, ",commit=%d", info->commit_interval);
1249 #ifdef CONFIG_BTRFS_DEBUG
1250 	if (btrfs_test_opt(root, FRAGMENT_DATA))
1251 		seq_puts(seq, ",fragment=data");
1252 	if (btrfs_test_opt(root, FRAGMENT_METADATA))
1253 		seq_puts(seq, ",fragment=metadata");
1254 #endif
1255 	seq_printf(seq, ",subvolid=%llu",
1256 		  BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1257 	seq_puts(seq, ",subvol=");
1258 	seq_dentry(seq, dentry, " \t\n\\");
1259 	return 0;
1260 }
1261 
1262 static int btrfs_test_super(struct super_block *s, void *data)
1263 {
1264 	struct btrfs_fs_info *p = data;
1265 	struct btrfs_fs_info *fs_info = btrfs_sb(s);
1266 
1267 	return fs_info->fs_devices == p->fs_devices;
1268 }
1269 
1270 static int btrfs_set_super(struct super_block *s, void *data)
1271 {
1272 	int err = set_anon_super(s, data);
1273 	if (!err)
1274 		s->s_fs_info = data;
1275 	return err;
1276 }
1277 
1278 /*
1279  * subvolumes are identified by ino 256
1280  */
1281 static inline int is_subvolume_inode(struct inode *inode)
1282 {
1283 	if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1284 		return 1;
1285 	return 0;
1286 }
1287 
1288 /*
1289  * This will add subvolid=0 to the argument string while removing any subvol=
1290  * and subvolid= arguments to make sure we get the top-level root for path
1291  * walking to the subvol we want.
1292  */
1293 static char *setup_root_args(char *args)
1294 {
1295 	char *buf, *dst, *sep;
1296 
1297 	if (!args)
1298 		return kstrdup("subvolid=0", GFP_NOFS);
1299 
1300 	/* The worst case is that we add ",subvolid=0" to the end. */
1301 	buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1302 	if (!buf)
1303 		return NULL;
1304 
1305 	while (1) {
1306 		sep = strchrnul(args, ',');
1307 		if (!strstarts(args, "subvol=") &&
1308 		    !strstarts(args, "subvolid=")) {
1309 			memcpy(dst, args, sep - args);
1310 			dst += sep - args;
1311 			*dst++ = ',';
1312 		}
1313 		if (*sep)
1314 			args = sep + 1;
1315 		else
1316 			break;
1317 	}
1318 	strcpy(dst, "subvolid=0");
1319 
1320 	return buf;
1321 }
1322 
1323 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1324 				   int flags, const char *device_name,
1325 				   char *data)
1326 {
1327 	struct dentry *root;
1328 	struct vfsmount *mnt = NULL;
1329 	char *newargs;
1330 	int ret;
1331 
1332 	newargs = setup_root_args(data);
1333 	if (!newargs) {
1334 		root = ERR_PTR(-ENOMEM);
1335 		goto out;
1336 	}
1337 
1338 	mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1339 	if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1340 		if (flags & MS_RDONLY) {
1341 			mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1342 					     device_name, newargs);
1343 		} else {
1344 			mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1345 					     device_name, newargs);
1346 			if (IS_ERR(mnt)) {
1347 				root = ERR_CAST(mnt);
1348 				mnt = NULL;
1349 				goto out;
1350 			}
1351 
1352 			down_write(&mnt->mnt_sb->s_umount);
1353 			ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1354 			up_write(&mnt->mnt_sb->s_umount);
1355 			if (ret < 0) {
1356 				root = ERR_PTR(ret);
1357 				goto out;
1358 			}
1359 		}
1360 	}
1361 	if (IS_ERR(mnt)) {
1362 		root = ERR_CAST(mnt);
1363 		mnt = NULL;
1364 		goto out;
1365 	}
1366 
1367 	if (!subvol_name) {
1368 		if (!subvol_objectid) {
1369 			ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1370 							  &subvol_objectid);
1371 			if (ret) {
1372 				root = ERR_PTR(ret);
1373 				goto out;
1374 			}
1375 		}
1376 		subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1377 							    subvol_objectid);
1378 		if (IS_ERR(subvol_name)) {
1379 			root = ERR_CAST(subvol_name);
1380 			subvol_name = NULL;
1381 			goto out;
1382 		}
1383 
1384 	}
1385 
1386 	root = mount_subtree(mnt, subvol_name);
1387 	/* mount_subtree() drops our reference on the vfsmount. */
1388 	mnt = NULL;
1389 
1390 	if (!IS_ERR(root)) {
1391 		struct super_block *s = root->d_sb;
1392 		struct inode *root_inode = d_inode(root);
1393 		u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1394 
1395 		ret = 0;
1396 		if (!is_subvolume_inode(root_inode)) {
1397 			pr_err("BTRFS: '%s' is not a valid subvolume\n",
1398 			       subvol_name);
1399 			ret = -EINVAL;
1400 		}
1401 		if (subvol_objectid && root_objectid != subvol_objectid) {
1402 			/*
1403 			 * This will also catch a race condition where a
1404 			 * subvolume which was passed by ID is renamed and
1405 			 * another subvolume is renamed over the old location.
1406 			 */
1407 			pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1408 			       subvol_name, subvol_objectid);
1409 			ret = -EINVAL;
1410 		}
1411 		if (ret) {
1412 			dput(root);
1413 			root = ERR_PTR(ret);
1414 			deactivate_locked_super(s);
1415 		}
1416 	}
1417 
1418 out:
1419 	mntput(mnt);
1420 	kfree(newargs);
1421 	kfree(subvol_name);
1422 	return root;
1423 }
1424 
1425 static int parse_security_options(char *orig_opts,
1426 				  struct security_mnt_opts *sec_opts)
1427 {
1428 	char *secdata = NULL;
1429 	int ret = 0;
1430 
1431 	secdata = alloc_secdata();
1432 	if (!secdata)
1433 		return -ENOMEM;
1434 	ret = security_sb_copy_data(orig_opts, secdata);
1435 	if (ret) {
1436 		free_secdata(secdata);
1437 		return ret;
1438 	}
1439 	ret = security_sb_parse_opts_str(secdata, sec_opts);
1440 	free_secdata(secdata);
1441 	return ret;
1442 }
1443 
1444 static int setup_security_options(struct btrfs_fs_info *fs_info,
1445 				  struct super_block *sb,
1446 				  struct security_mnt_opts *sec_opts)
1447 {
1448 	int ret = 0;
1449 
1450 	/*
1451 	 * Call security_sb_set_mnt_opts() to check whether new sec_opts
1452 	 * is valid.
1453 	 */
1454 	ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1455 	if (ret)
1456 		return ret;
1457 
1458 #ifdef CONFIG_SECURITY
1459 	if (!fs_info->security_opts.num_mnt_opts) {
1460 		/* first time security setup, copy sec_opts to fs_info */
1461 		memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1462 	} else {
1463 		/*
1464 		 * Since SELinux(the only one supports security_mnt_opts) does
1465 		 * NOT support changing context during remount/mount same sb,
1466 		 * This must be the same or part of the same security options,
1467 		 * just free it.
1468 		 */
1469 		security_free_mnt_opts(sec_opts);
1470 	}
1471 #endif
1472 	return ret;
1473 }
1474 
1475 /*
1476  * Find a superblock for the given device / mount point.
1477  *
1478  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1479  *	  for multiple device setup.  Make sure to keep it in sync.
1480  */
1481 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1482 		const char *device_name, void *data)
1483 {
1484 	struct block_device *bdev = NULL;
1485 	struct super_block *s;
1486 	struct btrfs_fs_devices *fs_devices = NULL;
1487 	struct btrfs_fs_info *fs_info = NULL;
1488 	struct security_mnt_opts new_sec_opts;
1489 	fmode_t mode = FMODE_READ;
1490 	char *subvol_name = NULL;
1491 	u64 subvol_objectid = 0;
1492 	int error = 0;
1493 
1494 	if (!(flags & MS_RDONLY))
1495 		mode |= FMODE_WRITE;
1496 
1497 	error = btrfs_parse_early_options(data, mode, fs_type,
1498 					  &subvol_name, &subvol_objectid,
1499 					  &fs_devices);
1500 	if (error) {
1501 		kfree(subvol_name);
1502 		return ERR_PTR(error);
1503 	}
1504 
1505 	if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1506 		/* mount_subvol() will free subvol_name. */
1507 		return mount_subvol(subvol_name, subvol_objectid, flags,
1508 				    device_name, data);
1509 	}
1510 
1511 	security_init_mnt_opts(&new_sec_opts);
1512 	if (data) {
1513 		error = parse_security_options(data, &new_sec_opts);
1514 		if (error)
1515 			return ERR_PTR(error);
1516 	}
1517 
1518 	error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1519 	if (error)
1520 		goto error_sec_opts;
1521 
1522 	/*
1523 	 * Setup a dummy root and fs_info for test/set super.  This is because
1524 	 * we don't actually fill this stuff out until open_ctree, but we need
1525 	 * it for searching for existing supers, so this lets us do that and
1526 	 * then open_ctree will properly initialize everything later.
1527 	 */
1528 	fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1529 	if (!fs_info) {
1530 		error = -ENOMEM;
1531 		goto error_sec_opts;
1532 	}
1533 
1534 	fs_info->fs_devices = fs_devices;
1535 
1536 	fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1537 	fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1538 	security_init_mnt_opts(&fs_info->security_opts);
1539 	if (!fs_info->super_copy || !fs_info->super_for_commit) {
1540 		error = -ENOMEM;
1541 		goto error_fs_info;
1542 	}
1543 
1544 	error = btrfs_open_devices(fs_devices, mode, fs_type);
1545 	if (error)
1546 		goto error_fs_info;
1547 
1548 	if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1549 		error = -EACCES;
1550 		goto error_close_devices;
1551 	}
1552 
1553 	bdev = fs_devices->latest_bdev;
1554 	s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1555 		 fs_info);
1556 	if (IS_ERR(s)) {
1557 		error = PTR_ERR(s);
1558 		goto error_close_devices;
1559 	}
1560 
1561 	if (s->s_root) {
1562 		btrfs_close_devices(fs_devices);
1563 		free_fs_info(fs_info);
1564 		if ((flags ^ s->s_flags) & MS_RDONLY)
1565 			error = -EBUSY;
1566 	} else {
1567 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1568 		btrfs_sb(s)->bdev_holder = fs_type;
1569 		error = btrfs_fill_super(s, fs_devices, data,
1570 					 flags & MS_SILENT ? 1 : 0);
1571 	}
1572 	if (error) {
1573 		deactivate_locked_super(s);
1574 		goto error_sec_opts;
1575 	}
1576 
1577 	fs_info = btrfs_sb(s);
1578 	error = setup_security_options(fs_info, s, &new_sec_opts);
1579 	if (error) {
1580 		deactivate_locked_super(s);
1581 		goto error_sec_opts;
1582 	}
1583 
1584 	return dget(s->s_root);
1585 
1586 error_close_devices:
1587 	btrfs_close_devices(fs_devices);
1588 error_fs_info:
1589 	free_fs_info(fs_info);
1590 error_sec_opts:
1591 	security_free_mnt_opts(&new_sec_opts);
1592 	return ERR_PTR(error);
1593 }
1594 
1595 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1596 				     int new_pool_size, int old_pool_size)
1597 {
1598 	if (new_pool_size == old_pool_size)
1599 		return;
1600 
1601 	fs_info->thread_pool_size = new_pool_size;
1602 
1603 	btrfs_info(fs_info, "resize thread pool %d -> %d",
1604 	       old_pool_size, new_pool_size);
1605 
1606 	btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1607 	btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1608 	btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1609 	btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1610 	btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1611 	btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1612 	btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1613 				new_pool_size);
1614 	btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1615 	btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1616 	btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1617 	btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1618 	btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1619 				new_pool_size);
1620 }
1621 
1622 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1623 {
1624 	set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1625 }
1626 
1627 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1628 				       unsigned long old_opts, int flags)
1629 {
1630 	if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1631 	    (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1632 	     (flags & MS_RDONLY))) {
1633 		/* wait for any defraggers to finish */
1634 		wait_event(fs_info->transaction_wait,
1635 			   (atomic_read(&fs_info->defrag_running) == 0));
1636 		if (flags & MS_RDONLY)
1637 			sync_filesystem(fs_info->sb);
1638 	}
1639 }
1640 
1641 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1642 					 unsigned long old_opts)
1643 {
1644 	/*
1645 	 * We need cleanup all defragable inodes if the autodefragment is
1646 	 * close or the fs is R/O.
1647 	 */
1648 	if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1649 	    (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1650 	     (fs_info->sb->s_flags & MS_RDONLY))) {
1651 		btrfs_cleanup_defrag_inodes(fs_info);
1652 	}
1653 
1654 	clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1655 }
1656 
1657 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1658 {
1659 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1660 	struct btrfs_root *root = fs_info->tree_root;
1661 	unsigned old_flags = sb->s_flags;
1662 	unsigned long old_opts = fs_info->mount_opt;
1663 	unsigned long old_compress_type = fs_info->compress_type;
1664 	u64 old_max_inline = fs_info->max_inline;
1665 	u64 old_alloc_start = fs_info->alloc_start;
1666 	int old_thread_pool_size = fs_info->thread_pool_size;
1667 	unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1668 	int ret;
1669 
1670 	sync_filesystem(sb);
1671 	btrfs_remount_prepare(fs_info);
1672 
1673 	if (data) {
1674 		struct security_mnt_opts new_sec_opts;
1675 
1676 		security_init_mnt_opts(&new_sec_opts);
1677 		ret = parse_security_options(data, &new_sec_opts);
1678 		if (ret)
1679 			goto restore;
1680 		ret = setup_security_options(fs_info, sb,
1681 					     &new_sec_opts);
1682 		if (ret) {
1683 			security_free_mnt_opts(&new_sec_opts);
1684 			goto restore;
1685 		}
1686 	}
1687 
1688 	ret = btrfs_parse_options(root, data);
1689 	if (ret) {
1690 		ret = -EINVAL;
1691 		goto restore;
1692 	}
1693 
1694 	btrfs_remount_begin(fs_info, old_opts, *flags);
1695 	btrfs_resize_thread_pool(fs_info,
1696 		fs_info->thread_pool_size, old_thread_pool_size);
1697 
1698 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1699 		goto out;
1700 
1701 	if (*flags & MS_RDONLY) {
1702 		/*
1703 		 * this also happens on 'umount -rf' or on shutdown, when
1704 		 * the filesystem is busy.
1705 		 */
1706 		cancel_work_sync(&fs_info->async_reclaim_work);
1707 
1708 		/* wait for the uuid_scan task to finish */
1709 		down(&fs_info->uuid_tree_rescan_sem);
1710 		/* avoid complains from lockdep et al. */
1711 		up(&fs_info->uuid_tree_rescan_sem);
1712 
1713 		sb->s_flags |= MS_RDONLY;
1714 
1715 		/*
1716 		 * Setting MS_RDONLY will put the cleaner thread to
1717 		 * sleep at the next loop if it's already active.
1718 		 * If it's already asleep, we'll leave unused block
1719 		 * groups on disk until we're mounted read-write again
1720 		 * unless we clean them up here.
1721 		 */
1722 		btrfs_delete_unused_bgs(fs_info);
1723 
1724 		btrfs_dev_replace_suspend_for_unmount(fs_info);
1725 		btrfs_scrub_cancel(fs_info);
1726 		btrfs_pause_balance(fs_info);
1727 
1728 		ret = btrfs_commit_super(root);
1729 		if (ret)
1730 			goto restore;
1731 	} else {
1732 		if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1733 			btrfs_err(fs_info,
1734 				"Remounting read-write after error is not allowed");
1735 			ret = -EINVAL;
1736 			goto restore;
1737 		}
1738 		if (fs_info->fs_devices->rw_devices == 0) {
1739 			ret = -EACCES;
1740 			goto restore;
1741 		}
1742 
1743 		if (fs_info->fs_devices->missing_devices >
1744 		     fs_info->num_tolerated_disk_barrier_failures &&
1745 		    !(*flags & MS_RDONLY)) {
1746 			btrfs_warn(fs_info,
1747 				"too many missing devices, writeable remount is not allowed");
1748 			ret = -EACCES;
1749 			goto restore;
1750 		}
1751 
1752 		if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1753 			ret = -EINVAL;
1754 			goto restore;
1755 		}
1756 
1757 		ret = btrfs_cleanup_fs_roots(fs_info);
1758 		if (ret)
1759 			goto restore;
1760 
1761 		/* recover relocation */
1762 		mutex_lock(&fs_info->cleaner_mutex);
1763 		ret = btrfs_recover_relocation(root);
1764 		mutex_unlock(&fs_info->cleaner_mutex);
1765 		if (ret)
1766 			goto restore;
1767 
1768 		ret = btrfs_resume_balance_async(fs_info);
1769 		if (ret)
1770 			goto restore;
1771 
1772 		ret = btrfs_resume_dev_replace_async(fs_info);
1773 		if (ret) {
1774 			btrfs_warn(fs_info, "failed to resume dev_replace");
1775 			goto restore;
1776 		}
1777 
1778 		if (!fs_info->uuid_root) {
1779 			btrfs_info(fs_info, "creating UUID tree");
1780 			ret = btrfs_create_uuid_tree(fs_info);
1781 			if (ret) {
1782 				btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1783 				goto restore;
1784 			}
1785 		}
1786 		sb->s_flags &= ~MS_RDONLY;
1787 	}
1788 out:
1789 	wake_up_process(fs_info->transaction_kthread);
1790 	btrfs_remount_cleanup(fs_info, old_opts);
1791 	return 0;
1792 
1793 restore:
1794 	/* We've hit an error - don't reset MS_RDONLY */
1795 	if (sb->s_flags & MS_RDONLY)
1796 		old_flags |= MS_RDONLY;
1797 	sb->s_flags = old_flags;
1798 	fs_info->mount_opt = old_opts;
1799 	fs_info->compress_type = old_compress_type;
1800 	fs_info->max_inline = old_max_inline;
1801 	mutex_lock(&fs_info->chunk_mutex);
1802 	fs_info->alloc_start = old_alloc_start;
1803 	mutex_unlock(&fs_info->chunk_mutex);
1804 	btrfs_resize_thread_pool(fs_info,
1805 		old_thread_pool_size, fs_info->thread_pool_size);
1806 	fs_info->metadata_ratio = old_metadata_ratio;
1807 	btrfs_remount_cleanup(fs_info, old_opts);
1808 	return ret;
1809 }
1810 
1811 /* Used to sort the devices by max_avail(descending sort) */
1812 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1813 				       const void *dev_info2)
1814 {
1815 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
1816 	    ((struct btrfs_device_info *)dev_info2)->max_avail)
1817 		return -1;
1818 	else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1819 		 ((struct btrfs_device_info *)dev_info2)->max_avail)
1820 		return 1;
1821 	else
1822 	return 0;
1823 }
1824 
1825 /*
1826  * sort the devices by max_avail, in which max free extent size of each device
1827  * is stored.(Descending Sort)
1828  */
1829 static inline void btrfs_descending_sort_devices(
1830 					struct btrfs_device_info *devices,
1831 					size_t nr_devices)
1832 {
1833 	sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1834 	     btrfs_cmp_device_free_bytes, NULL);
1835 }
1836 
1837 /*
1838  * The helper to calc the free space on the devices that can be used to store
1839  * file data.
1840  */
1841 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1842 {
1843 	struct btrfs_fs_info *fs_info = root->fs_info;
1844 	struct btrfs_device_info *devices_info;
1845 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1846 	struct btrfs_device *device;
1847 	u64 skip_space;
1848 	u64 type;
1849 	u64 avail_space;
1850 	u64 used_space;
1851 	u64 min_stripe_size;
1852 	int min_stripes = 1, num_stripes = 1;
1853 	int i = 0, nr_devices;
1854 	int ret;
1855 
1856 	/*
1857 	 * We aren't under the device list lock, so this is racey-ish, but good
1858 	 * enough for our purposes.
1859 	 */
1860 	nr_devices = fs_info->fs_devices->open_devices;
1861 	if (!nr_devices) {
1862 		smp_mb();
1863 		nr_devices = fs_info->fs_devices->open_devices;
1864 		ASSERT(nr_devices);
1865 		if (!nr_devices) {
1866 			*free_bytes = 0;
1867 			return 0;
1868 		}
1869 	}
1870 
1871 	devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1872 			       GFP_NOFS);
1873 	if (!devices_info)
1874 		return -ENOMEM;
1875 
1876 	/* calc min stripe number for data space alloction */
1877 	type = btrfs_get_alloc_profile(root, 1);
1878 	if (type & BTRFS_BLOCK_GROUP_RAID0) {
1879 		min_stripes = 2;
1880 		num_stripes = nr_devices;
1881 	} else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1882 		min_stripes = 2;
1883 		num_stripes = 2;
1884 	} else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1885 		min_stripes = 4;
1886 		num_stripes = 4;
1887 	}
1888 
1889 	if (type & BTRFS_BLOCK_GROUP_DUP)
1890 		min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1891 	else
1892 		min_stripe_size = BTRFS_STRIPE_LEN;
1893 
1894 	if (fs_info->alloc_start)
1895 		mutex_lock(&fs_devices->device_list_mutex);
1896 	rcu_read_lock();
1897 	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1898 		if (!device->in_fs_metadata || !device->bdev ||
1899 		    device->is_tgtdev_for_dev_replace)
1900 			continue;
1901 
1902 		if (i >= nr_devices)
1903 			break;
1904 
1905 		avail_space = device->total_bytes - device->bytes_used;
1906 
1907 		/* align with stripe_len */
1908 		avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1909 		avail_space *= BTRFS_STRIPE_LEN;
1910 
1911 		/*
1912 		 * In order to avoid overwritting the superblock on the drive,
1913 		 * btrfs starts at an offset of at least 1MB when doing chunk
1914 		 * allocation.
1915 		 */
1916 		skip_space = SZ_1M;
1917 
1918 		/* user can set the offset in fs_info->alloc_start. */
1919 		if (fs_info->alloc_start &&
1920 		    fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1921 		    device->total_bytes) {
1922 			rcu_read_unlock();
1923 			skip_space = max(fs_info->alloc_start, skip_space);
1924 
1925 			/*
1926 			 * btrfs can not use the free space in
1927 			 * [0, skip_space - 1], we must subtract it from the
1928 			 * total. In order to implement it, we account the used
1929 			 * space in this range first.
1930 			 */
1931 			ret = btrfs_account_dev_extents_size(device, 0,
1932 							     skip_space - 1,
1933 							     &used_space);
1934 			if (ret) {
1935 				kfree(devices_info);
1936 				mutex_unlock(&fs_devices->device_list_mutex);
1937 				return ret;
1938 			}
1939 
1940 			rcu_read_lock();
1941 
1942 			/* calc the free space in [0, skip_space - 1] */
1943 			skip_space -= used_space;
1944 		}
1945 
1946 		/*
1947 		 * we can use the free space in [0, skip_space - 1], subtract
1948 		 * it from the total.
1949 		 */
1950 		if (avail_space && avail_space >= skip_space)
1951 			avail_space -= skip_space;
1952 		else
1953 			avail_space = 0;
1954 
1955 		if (avail_space < min_stripe_size)
1956 			continue;
1957 
1958 		devices_info[i].dev = device;
1959 		devices_info[i].max_avail = avail_space;
1960 
1961 		i++;
1962 	}
1963 	rcu_read_unlock();
1964 	if (fs_info->alloc_start)
1965 		mutex_unlock(&fs_devices->device_list_mutex);
1966 
1967 	nr_devices = i;
1968 
1969 	btrfs_descending_sort_devices(devices_info, nr_devices);
1970 
1971 	i = nr_devices - 1;
1972 	avail_space = 0;
1973 	while (nr_devices >= min_stripes) {
1974 		if (num_stripes > nr_devices)
1975 			num_stripes = nr_devices;
1976 
1977 		if (devices_info[i].max_avail >= min_stripe_size) {
1978 			int j;
1979 			u64 alloc_size;
1980 
1981 			avail_space += devices_info[i].max_avail * num_stripes;
1982 			alloc_size = devices_info[i].max_avail;
1983 			for (j = i + 1 - num_stripes; j <= i; j++)
1984 				devices_info[j].max_avail -= alloc_size;
1985 		}
1986 		i--;
1987 		nr_devices--;
1988 	}
1989 
1990 	kfree(devices_info);
1991 	*free_bytes = avail_space;
1992 	return 0;
1993 }
1994 
1995 /*
1996  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1997  *
1998  * If there's a redundant raid level at DATA block groups, use the respective
1999  * multiplier to scale the sizes.
2000  *
2001  * Unused device space usage is based on simulating the chunk allocator
2002  * algorithm that respects the device sizes, order of allocations and the
2003  * 'alloc_start' value, this is a close approximation of the actual use but
2004  * there are other factors that may change the result (like a new metadata
2005  * chunk).
2006  *
2007  * If metadata is exhausted, f_bavail will be 0.
2008  *
2009  * FIXME: not accurate for mixed block groups, total and free/used are ok,
2010  * available appears slightly larger.
2011  */
2012 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2013 {
2014 	struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2015 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2016 	struct list_head *head = &fs_info->space_info;
2017 	struct btrfs_space_info *found;
2018 	u64 total_used = 0;
2019 	u64 total_free_data = 0;
2020 	u64 total_free_meta = 0;
2021 	int bits = dentry->d_sb->s_blocksize_bits;
2022 	__be32 *fsid = (__be32 *)fs_info->fsid;
2023 	unsigned factor = 1;
2024 	struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2025 	int ret;
2026 	u64 thresh = 0;
2027 
2028 	/*
2029 	 * holding chunk_muext to avoid allocating new chunks, holding
2030 	 * device_list_mutex to avoid the device being removed
2031 	 */
2032 	rcu_read_lock();
2033 	list_for_each_entry_rcu(found, head, list) {
2034 		if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2035 			int i;
2036 
2037 			total_free_data += found->disk_total - found->disk_used;
2038 			total_free_data -=
2039 				btrfs_account_ro_block_groups_free_space(found);
2040 
2041 			for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2042 				if (!list_empty(&found->block_groups[i])) {
2043 					switch (i) {
2044 					case BTRFS_RAID_DUP:
2045 					case BTRFS_RAID_RAID1:
2046 					case BTRFS_RAID_RAID10:
2047 						factor = 2;
2048 					}
2049 				}
2050 			}
2051 		}
2052 		if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2053 			total_free_meta += found->disk_total - found->disk_used;
2054 
2055 		total_used += found->disk_used;
2056 	}
2057 
2058 	rcu_read_unlock();
2059 
2060 	buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2061 	buf->f_blocks >>= bits;
2062 	buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2063 
2064 	/* Account global block reserve as used, it's in logical size already */
2065 	spin_lock(&block_rsv->lock);
2066 	buf->f_bfree -= block_rsv->size >> bits;
2067 	spin_unlock(&block_rsv->lock);
2068 
2069 	buf->f_bavail = div_u64(total_free_data, factor);
2070 	ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2071 	if (ret)
2072 		return ret;
2073 	buf->f_bavail += div_u64(total_free_data, factor);
2074 	buf->f_bavail = buf->f_bavail >> bits;
2075 
2076 	/*
2077 	 * We calculate the remaining metadata space minus global reserve. If
2078 	 * this is (supposedly) smaller than zero, there's no space. But this
2079 	 * does not hold in practice, the exhausted state happens where's still
2080 	 * some positive delta. So we apply some guesswork and compare the
2081 	 * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2082 	 *
2083 	 * We probably cannot calculate the exact threshold value because this
2084 	 * depends on the internal reservations requested by various
2085 	 * operations, so some operations that consume a few metadata will
2086 	 * succeed even if the Avail is zero. But this is better than the other
2087 	 * way around.
2088 	 */
2089 	thresh = 4 * 1024 * 1024;
2090 
2091 	if (total_free_meta - thresh < block_rsv->size)
2092 		buf->f_bavail = 0;
2093 
2094 	buf->f_type = BTRFS_SUPER_MAGIC;
2095 	buf->f_bsize = dentry->d_sb->s_blocksize;
2096 	buf->f_namelen = BTRFS_NAME_LEN;
2097 
2098 	/* We treat it as constant endianness (it doesn't matter _which_)
2099 	   because we want the fsid to come out the same whether mounted
2100 	   on a big-endian or little-endian host */
2101 	buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2102 	buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2103 	/* Mask in the root object ID too, to disambiguate subvols */
2104 	buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2105 	buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2106 
2107 	return 0;
2108 }
2109 
2110 static void btrfs_kill_super(struct super_block *sb)
2111 {
2112 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2113 	kill_anon_super(sb);
2114 	free_fs_info(fs_info);
2115 }
2116 
2117 static struct file_system_type btrfs_fs_type = {
2118 	.owner		= THIS_MODULE,
2119 	.name		= "btrfs",
2120 	.mount		= btrfs_mount,
2121 	.kill_sb	= btrfs_kill_super,
2122 	.fs_flags	= FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2123 };
2124 MODULE_ALIAS_FS("btrfs");
2125 
2126 static int btrfs_control_open(struct inode *inode, struct file *file)
2127 {
2128 	/*
2129 	 * The control file's private_data is used to hold the
2130 	 * transaction when it is started and is used to keep
2131 	 * track of whether a transaction is already in progress.
2132 	 */
2133 	file->private_data = NULL;
2134 	return 0;
2135 }
2136 
2137 /*
2138  * used by btrfsctl to scan devices when no FS is mounted
2139  */
2140 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2141 				unsigned long arg)
2142 {
2143 	struct btrfs_ioctl_vol_args *vol;
2144 	struct btrfs_fs_devices *fs_devices;
2145 	int ret = -ENOTTY;
2146 
2147 	if (!capable(CAP_SYS_ADMIN))
2148 		return -EPERM;
2149 
2150 	vol = memdup_user((void __user *)arg, sizeof(*vol));
2151 	if (IS_ERR(vol))
2152 		return PTR_ERR(vol);
2153 
2154 	switch (cmd) {
2155 	case BTRFS_IOC_SCAN_DEV:
2156 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2157 					    &btrfs_fs_type, &fs_devices);
2158 		break;
2159 	case BTRFS_IOC_DEVICES_READY:
2160 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2161 					    &btrfs_fs_type, &fs_devices);
2162 		if (ret)
2163 			break;
2164 		ret = !(fs_devices->num_devices == fs_devices->total_devices);
2165 		break;
2166 	}
2167 
2168 	kfree(vol);
2169 	return ret;
2170 }
2171 
2172 static int btrfs_freeze(struct super_block *sb)
2173 {
2174 	struct btrfs_trans_handle *trans;
2175 	struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2176 
2177 	trans = btrfs_attach_transaction_barrier(root);
2178 	if (IS_ERR(trans)) {
2179 		/* no transaction, don't bother */
2180 		if (PTR_ERR(trans) == -ENOENT)
2181 			return 0;
2182 		return PTR_ERR(trans);
2183 	}
2184 	return btrfs_commit_transaction(trans, root);
2185 }
2186 
2187 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2188 {
2189 	struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2190 	struct btrfs_fs_devices *cur_devices;
2191 	struct btrfs_device *dev, *first_dev = NULL;
2192 	struct list_head *head;
2193 	struct rcu_string *name;
2194 
2195 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2196 	cur_devices = fs_info->fs_devices;
2197 	while (cur_devices) {
2198 		head = &cur_devices->devices;
2199 		list_for_each_entry(dev, head, dev_list) {
2200 			if (dev->missing)
2201 				continue;
2202 			if (!dev->name)
2203 				continue;
2204 			if (!first_dev || dev->devid < first_dev->devid)
2205 				first_dev = dev;
2206 		}
2207 		cur_devices = cur_devices->seed;
2208 	}
2209 
2210 	if (first_dev) {
2211 		rcu_read_lock();
2212 		name = rcu_dereference(first_dev->name);
2213 		seq_escape(m, name->str, " \t\n\\");
2214 		rcu_read_unlock();
2215 	} else {
2216 		WARN_ON(1);
2217 	}
2218 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2219 	return 0;
2220 }
2221 
2222 static const struct super_operations btrfs_super_ops = {
2223 	.drop_inode	= btrfs_drop_inode,
2224 	.evict_inode	= btrfs_evict_inode,
2225 	.put_super	= btrfs_put_super,
2226 	.sync_fs	= btrfs_sync_fs,
2227 	.show_options	= btrfs_show_options,
2228 	.show_devname	= btrfs_show_devname,
2229 	.write_inode	= btrfs_write_inode,
2230 	.alloc_inode	= btrfs_alloc_inode,
2231 	.destroy_inode	= btrfs_destroy_inode,
2232 	.statfs		= btrfs_statfs,
2233 	.remount_fs	= btrfs_remount,
2234 	.freeze_fs	= btrfs_freeze,
2235 };
2236 
2237 static const struct file_operations btrfs_ctl_fops = {
2238 	.open = btrfs_control_open,
2239 	.unlocked_ioctl	 = btrfs_control_ioctl,
2240 	.compat_ioctl = btrfs_control_ioctl,
2241 	.owner	 = THIS_MODULE,
2242 	.llseek = noop_llseek,
2243 };
2244 
2245 static struct miscdevice btrfs_misc = {
2246 	.minor		= BTRFS_MINOR,
2247 	.name		= "btrfs-control",
2248 	.fops		= &btrfs_ctl_fops
2249 };
2250 
2251 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2252 MODULE_ALIAS("devname:btrfs-control");
2253 
2254 static int btrfs_interface_init(void)
2255 {
2256 	return misc_register(&btrfs_misc);
2257 }
2258 
2259 static void btrfs_interface_exit(void)
2260 {
2261 	misc_deregister(&btrfs_misc);
2262 }
2263 
2264 static void btrfs_print_info(void)
2265 {
2266 	printk(KERN_INFO "Btrfs loaded"
2267 #ifdef CONFIG_BTRFS_DEBUG
2268 			", debug=on"
2269 #endif
2270 #ifdef CONFIG_BTRFS_ASSERT
2271 			", assert=on"
2272 #endif
2273 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2274 			", integrity-checker=on"
2275 #endif
2276 			"\n");
2277 }
2278 
2279 static int btrfs_run_sanity_tests(void)
2280 {
2281 	int ret;
2282 
2283 	ret = btrfs_init_test_fs();
2284 	if (ret)
2285 		return ret;
2286 
2287 	ret = btrfs_test_free_space_cache();
2288 	if (ret)
2289 		goto out;
2290 	ret = btrfs_test_extent_buffer_operations();
2291 	if (ret)
2292 		goto out;
2293 	ret = btrfs_test_extent_io();
2294 	if (ret)
2295 		goto out;
2296 	ret = btrfs_test_inodes();
2297 	if (ret)
2298 		goto out;
2299 	ret = btrfs_test_qgroups();
2300 	if (ret)
2301 		goto out;
2302 	ret = btrfs_test_free_space_tree();
2303 out:
2304 	btrfs_destroy_test_fs();
2305 	return ret;
2306 }
2307 
2308 static int __init init_btrfs_fs(void)
2309 {
2310 	int err;
2311 
2312 	err = btrfs_hash_init();
2313 	if (err)
2314 		return err;
2315 
2316 	btrfs_props_init();
2317 
2318 	err = btrfs_init_sysfs();
2319 	if (err)
2320 		goto free_hash;
2321 
2322 	btrfs_init_compress();
2323 
2324 	err = btrfs_init_cachep();
2325 	if (err)
2326 		goto free_compress;
2327 
2328 	err = extent_io_init();
2329 	if (err)
2330 		goto free_cachep;
2331 
2332 	err = extent_map_init();
2333 	if (err)
2334 		goto free_extent_io;
2335 
2336 	err = ordered_data_init();
2337 	if (err)
2338 		goto free_extent_map;
2339 
2340 	err = btrfs_delayed_inode_init();
2341 	if (err)
2342 		goto free_ordered_data;
2343 
2344 	err = btrfs_auto_defrag_init();
2345 	if (err)
2346 		goto free_delayed_inode;
2347 
2348 	err = btrfs_delayed_ref_init();
2349 	if (err)
2350 		goto free_auto_defrag;
2351 
2352 	err = btrfs_prelim_ref_init();
2353 	if (err)
2354 		goto free_delayed_ref;
2355 
2356 	err = btrfs_end_io_wq_init();
2357 	if (err)
2358 		goto free_prelim_ref;
2359 
2360 	err = btrfs_interface_init();
2361 	if (err)
2362 		goto free_end_io_wq;
2363 
2364 	btrfs_init_lockdep();
2365 
2366 	btrfs_print_info();
2367 
2368 	err = btrfs_run_sanity_tests();
2369 	if (err)
2370 		goto unregister_ioctl;
2371 
2372 	err = register_filesystem(&btrfs_fs_type);
2373 	if (err)
2374 		goto unregister_ioctl;
2375 
2376 	return 0;
2377 
2378 unregister_ioctl:
2379 	btrfs_interface_exit();
2380 free_end_io_wq:
2381 	btrfs_end_io_wq_exit();
2382 free_prelim_ref:
2383 	btrfs_prelim_ref_exit();
2384 free_delayed_ref:
2385 	btrfs_delayed_ref_exit();
2386 free_auto_defrag:
2387 	btrfs_auto_defrag_exit();
2388 free_delayed_inode:
2389 	btrfs_delayed_inode_exit();
2390 free_ordered_data:
2391 	ordered_data_exit();
2392 free_extent_map:
2393 	extent_map_exit();
2394 free_extent_io:
2395 	extent_io_exit();
2396 free_cachep:
2397 	btrfs_destroy_cachep();
2398 free_compress:
2399 	btrfs_exit_compress();
2400 	btrfs_exit_sysfs();
2401 free_hash:
2402 	btrfs_hash_exit();
2403 	return err;
2404 }
2405 
2406 static void __exit exit_btrfs_fs(void)
2407 {
2408 	btrfs_destroy_cachep();
2409 	btrfs_delayed_ref_exit();
2410 	btrfs_auto_defrag_exit();
2411 	btrfs_delayed_inode_exit();
2412 	btrfs_prelim_ref_exit();
2413 	ordered_data_exit();
2414 	extent_map_exit();
2415 	extent_io_exit();
2416 	btrfs_interface_exit();
2417 	btrfs_end_io_wq_exit();
2418 	unregister_filesystem(&btrfs_fs_type);
2419 	btrfs_exit_sysfs();
2420 	btrfs_cleanup_fs_uuids();
2421 	btrfs_exit_compress();
2422 	btrfs_hash_exit();
2423 }
2424 
2425 late_initcall(init_btrfs_fs);
2426 module_exit(exit_btrfs_fs)
2427 
2428 MODULE_LICENSE("GPL");
2429