xref: /linux/fs/btrfs/super.c (revision 65929215d85702dd4a76db0ba9fbfc022ab99e55)
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 "compat.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 "ioctl.h"
51 #include "print-tree.h"
52 #include "xattr.h"
53 #include "volumes.h"
54 #include "version.h"
55 #include "export.h"
56 #include "compression.h"
57 #include "rcu-string.h"
58 
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/btrfs.h>
61 
62 static const struct super_operations btrfs_super_ops;
63 static struct file_system_type btrfs_fs_type;
64 
65 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
66 				      char nbuf[16])
67 {
68 	char *errstr = NULL;
69 
70 	switch (errno) {
71 	case -EIO:
72 		errstr = "IO failure";
73 		break;
74 	case -ENOMEM:
75 		errstr = "Out of memory";
76 		break;
77 	case -EROFS:
78 		errstr = "Readonly filesystem";
79 		break;
80 	case -EEXIST:
81 		errstr = "Object already exists";
82 		break;
83 	default:
84 		if (nbuf) {
85 			if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
86 				errstr = nbuf;
87 		}
88 		break;
89 	}
90 
91 	return errstr;
92 }
93 
94 static void __save_error_info(struct btrfs_fs_info *fs_info)
95 {
96 	/*
97 	 * today we only save the error info into ram.  Long term we'll
98 	 * also send it down to the disk
99 	 */
100 	fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
101 }
102 
103 static void save_error_info(struct btrfs_fs_info *fs_info)
104 {
105 	__save_error_info(fs_info);
106 }
107 
108 /* btrfs handle error by forcing the filesystem readonly */
109 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
110 {
111 	struct super_block *sb = fs_info->sb;
112 
113 	if (sb->s_flags & MS_RDONLY)
114 		return;
115 
116 	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
117 		sb->s_flags |= MS_RDONLY;
118 		printk(KERN_INFO "btrfs is forced readonly\n");
119 		__btrfs_scrub_cancel(fs_info);
120 //		WARN_ON(1);
121 	}
122 }
123 
124 #ifdef CONFIG_PRINTK
125 /*
126  * __btrfs_std_error decodes expected errors from the caller and
127  * invokes the approciate error response.
128  */
129 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
130 		       unsigned int line, int errno, const char *fmt, ...)
131 {
132 	struct super_block *sb = fs_info->sb;
133 	char nbuf[16];
134 	const char *errstr;
135 	va_list args;
136 	va_start(args, fmt);
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   	errstr = btrfs_decode_error(fs_info, errno, nbuf);
146 	if (fmt) {
147 		struct va_format vaf = {
148 			.fmt = fmt,
149 			.va = &args,
150 		};
151 
152 		printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
153 			sb->s_id, function, line, errstr, &vaf);
154 	} else {
155 		printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
156 			sb->s_id, function, line, errstr);
157 	}
158 
159 	/* Don't go through full error handling during mount */
160 	if (sb->s_flags & MS_BORN) {
161 		save_error_info(fs_info);
162 		btrfs_handle_error(fs_info);
163 	}
164 	va_end(args);
165 }
166 
167 static const char * const logtypes[] = {
168 	"emergency",
169 	"alert",
170 	"critical",
171 	"error",
172 	"warning",
173 	"notice",
174 	"info",
175 	"debug",
176 };
177 
178 void btrfs_printk(struct btrfs_fs_info *fs_info, const char *fmt, ...)
179 {
180 	struct super_block *sb = fs_info->sb;
181 	char lvl[4];
182 	struct va_format vaf;
183 	va_list args;
184 	const char *type = logtypes[4];
185 	int kern_level;
186 
187 	va_start(args, fmt);
188 
189 	kern_level = printk_get_level(fmt);
190 	if (kern_level) {
191 		size_t size = printk_skip_level(fmt) - fmt;
192 		memcpy(lvl, fmt,  size);
193 		lvl[size] = '\0';
194 		fmt += size;
195 		type = logtypes[kern_level - '0'];
196 	} else
197 		*lvl = '\0';
198 
199 	vaf.fmt = fmt;
200 	vaf.va = &args;
201 
202 	printk("%sBTRFS %s (device %s): %pV", lvl, type, sb->s_id, &vaf);
203 
204 	va_end(args);
205 }
206 
207 #else
208 
209 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
210 		       unsigned int line, int errno, const char *fmt, ...)
211 {
212 	struct super_block *sb = fs_info->sb;
213 
214 	/*
215 	 * Special case: if the error is EROFS, and we're already
216 	 * under MS_RDONLY, then it is safe here.
217 	 */
218 	if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
219 		return;
220 
221 	/* Don't go through full error handling during mount */
222 	if (sb->s_flags & MS_BORN) {
223 		save_error_info(fs_info);
224 		btrfs_handle_error(fs_info);
225 	}
226 }
227 #endif
228 
229 /*
230  * We only mark the transaction aborted and then set the file system read-only.
231  * This will prevent new transactions from starting or trying to join this
232  * one.
233  *
234  * This means that error recovery at the call site is limited to freeing
235  * any local memory allocations and passing the error code up without
236  * further cleanup. The transaction should complete as it normally would
237  * in the call path but will return -EIO.
238  *
239  * We'll complete the cleanup in btrfs_end_transaction and
240  * btrfs_commit_transaction.
241  */
242 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
243 			       struct btrfs_root *root, const char *function,
244 			       unsigned int line, int errno)
245 {
246 	WARN_ONCE(1, KERN_DEBUG "btrfs: Transaction aborted");
247 	trans->aborted = errno;
248 	/* Nothing used. The other threads that have joined this
249 	 * transaction may be able to continue. */
250 	if (!trans->blocks_used) {
251 		btrfs_printk(root->fs_info, "Aborting unused transaction.\n");
252 		return;
253 	}
254 	trans->transaction->aborted = errno;
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 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
262 		   unsigned int line, int errno, const char *fmt, ...)
263 {
264 	char nbuf[16];
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(fs_info, errno, nbuf);
277 	if (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR)
278 		panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
279 			s_id, function, line, &vaf, errstr);
280 
281 	printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
282 	       s_id, function, line, &vaf, errstr);
283 	va_end(args);
284 	/* Caller calls BUG() */
285 }
286 
287 static void btrfs_put_super(struct super_block *sb)
288 {
289 	(void)close_ctree(btrfs_sb(sb)->tree_root);
290 	/* FIXME: need to fix VFS to return error? */
291 	/* AV: return it _where_?  ->put_super() can be triggered by any number
292 	 * of async events, up to and including delivery of SIGKILL to the
293 	 * last process that kept it busy.  Or segfault in the aforementioned
294 	 * process...  Whom would you report that to?
295 	 */
296 }
297 
298 enum {
299 	Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
300 	Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
301 	Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
302 	Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
303 	Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
304 	Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
305 	Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
306 	Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
307 	Opt_check_integrity, Opt_check_integrity_including_extent_data,
308 	Opt_check_integrity_print_mask, Opt_fatal_errors,
309 	Opt_err,
310 };
311 
312 static match_table_t tokens = {
313 	{Opt_degraded, "degraded"},
314 	{Opt_subvol, "subvol=%s"},
315 	{Opt_subvolid, "subvolid=%d"},
316 	{Opt_device, "device=%s"},
317 	{Opt_nodatasum, "nodatasum"},
318 	{Opt_nodatacow, "nodatacow"},
319 	{Opt_nobarrier, "nobarrier"},
320 	{Opt_max_inline, "max_inline=%s"},
321 	{Opt_alloc_start, "alloc_start=%s"},
322 	{Opt_thread_pool, "thread_pool=%d"},
323 	{Opt_compress, "compress"},
324 	{Opt_compress_type, "compress=%s"},
325 	{Opt_compress_force, "compress-force"},
326 	{Opt_compress_force_type, "compress-force=%s"},
327 	{Opt_ssd, "ssd"},
328 	{Opt_ssd_spread, "ssd_spread"},
329 	{Opt_nossd, "nossd"},
330 	{Opt_noacl, "noacl"},
331 	{Opt_notreelog, "notreelog"},
332 	{Opt_flushoncommit, "flushoncommit"},
333 	{Opt_ratio, "metadata_ratio=%d"},
334 	{Opt_discard, "discard"},
335 	{Opt_space_cache, "space_cache"},
336 	{Opt_clear_cache, "clear_cache"},
337 	{Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
338 	{Opt_enospc_debug, "enospc_debug"},
339 	{Opt_subvolrootid, "subvolrootid=%d"},
340 	{Opt_defrag, "autodefrag"},
341 	{Opt_inode_cache, "inode_cache"},
342 	{Opt_no_space_cache, "nospace_cache"},
343 	{Opt_recovery, "recovery"},
344 	{Opt_skip_balance, "skip_balance"},
345 	{Opt_check_integrity, "check_int"},
346 	{Opt_check_integrity_including_extent_data, "check_int_data"},
347 	{Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
348 	{Opt_fatal_errors, "fatal_errors=%s"},
349 	{Opt_err, NULL},
350 };
351 
352 /*
353  * Regular mount options parser.  Everything that is needed only when
354  * reading in a new superblock is parsed here.
355  * XXX JDM: This needs to be cleaned up for remount.
356  */
357 int btrfs_parse_options(struct btrfs_root *root, char *options)
358 {
359 	struct btrfs_fs_info *info = root->fs_info;
360 	substring_t args[MAX_OPT_ARGS];
361 	char *p, *num, *orig = NULL;
362 	u64 cache_gen;
363 	int intarg;
364 	int ret = 0;
365 	char *compress_type;
366 	bool compress_force = false;
367 
368 	cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
369 	if (cache_gen)
370 		btrfs_set_opt(info->mount_opt, SPACE_CACHE);
371 
372 	if (!options)
373 		goto out;
374 
375 	/*
376 	 * strsep changes the string, duplicate it because parse_options
377 	 * gets called twice
378 	 */
379 	options = kstrdup(options, GFP_NOFS);
380 	if (!options)
381 		return -ENOMEM;
382 
383 	orig = options;
384 
385 	while ((p = strsep(&options, ",")) != NULL) {
386 		int token;
387 		if (!*p)
388 			continue;
389 
390 		token = match_token(p, tokens, args);
391 		switch (token) {
392 		case Opt_degraded:
393 			printk(KERN_INFO "btrfs: allowing degraded mounts\n");
394 			btrfs_set_opt(info->mount_opt, DEGRADED);
395 			break;
396 		case Opt_subvol:
397 		case Opt_subvolid:
398 		case Opt_subvolrootid:
399 		case Opt_device:
400 			/*
401 			 * These are parsed by btrfs_parse_early_options
402 			 * and can be happily ignored here.
403 			 */
404 			break;
405 		case Opt_nodatasum:
406 			printk(KERN_INFO "btrfs: setting nodatasum\n");
407 			btrfs_set_opt(info->mount_opt, NODATASUM);
408 			break;
409 		case Opt_nodatacow:
410 			printk(KERN_INFO "btrfs: setting nodatacow\n");
411 			btrfs_set_opt(info->mount_opt, NODATACOW);
412 			btrfs_set_opt(info->mount_opt, NODATASUM);
413 			break;
414 		case Opt_compress_force:
415 		case Opt_compress_force_type:
416 			compress_force = true;
417 		case Opt_compress:
418 		case Opt_compress_type:
419 			if (token == Opt_compress ||
420 			    token == Opt_compress_force ||
421 			    strcmp(args[0].from, "zlib") == 0) {
422 				compress_type = "zlib";
423 				info->compress_type = BTRFS_COMPRESS_ZLIB;
424 				btrfs_set_opt(info->mount_opt, COMPRESS);
425 			} else if (strcmp(args[0].from, "lzo") == 0) {
426 				compress_type = "lzo";
427 				info->compress_type = BTRFS_COMPRESS_LZO;
428 				btrfs_set_opt(info->mount_opt, COMPRESS);
429 				btrfs_set_fs_incompat(info, COMPRESS_LZO);
430 			} else if (strncmp(args[0].from, "no", 2) == 0) {
431 				compress_type = "no";
432 				info->compress_type = BTRFS_COMPRESS_NONE;
433 				btrfs_clear_opt(info->mount_opt, COMPRESS);
434 				btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
435 				compress_force = false;
436 			} else {
437 				ret = -EINVAL;
438 				goto out;
439 			}
440 
441 			if (compress_force) {
442 				btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
443 				pr_info("btrfs: force %s compression\n",
444 					compress_type);
445 			} else
446 				pr_info("btrfs: use %s compression\n",
447 					compress_type);
448 			break;
449 		case Opt_ssd:
450 			printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
451 			btrfs_set_opt(info->mount_opt, SSD);
452 			break;
453 		case Opt_ssd_spread:
454 			printk(KERN_INFO "btrfs: use spread ssd "
455 			       "allocation scheme\n");
456 			btrfs_set_opt(info->mount_opt, SSD);
457 			btrfs_set_opt(info->mount_opt, SSD_SPREAD);
458 			break;
459 		case Opt_nossd:
460 			printk(KERN_INFO "btrfs: not using ssd allocation "
461 			       "scheme\n");
462 			btrfs_set_opt(info->mount_opt, NOSSD);
463 			btrfs_clear_opt(info->mount_opt, SSD);
464 			btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
465 			break;
466 		case Opt_nobarrier:
467 			printk(KERN_INFO "btrfs: turning off barriers\n");
468 			btrfs_set_opt(info->mount_opt, NOBARRIER);
469 			break;
470 		case Opt_thread_pool:
471 			intarg = 0;
472 			match_int(&args[0], &intarg);
473 			if (intarg)
474 				info->thread_pool_size = intarg;
475 			break;
476 		case Opt_max_inline:
477 			num = match_strdup(&args[0]);
478 			if (num) {
479 				info->max_inline = memparse(num, NULL);
480 				kfree(num);
481 
482 				if (info->max_inline) {
483 					info->max_inline = max_t(u64,
484 						info->max_inline,
485 						root->sectorsize);
486 				}
487 				printk(KERN_INFO "btrfs: max_inline at %llu\n",
488 					(unsigned long long)info->max_inline);
489 			}
490 			break;
491 		case Opt_alloc_start:
492 			num = match_strdup(&args[0]);
493 			if (num) {
494 				info->alloc_start = memparse(num, NULL);
495 				kfree(num);
496 				printk(KERN_INFO
497 					"btrfs: allocations start at %llu\n",
498 					(unsigned long long)info->alloc_start);
499 			}
500 			break;
501 		case Opt_noacl:
502 			root->fs_info->sb->s_flags &= ~MS_POSIXACL;
503 			break;
504 		case Opt_notreelog:
505 			printk(KERN_INFO "btrfs: disabling tree log\n");
506 			btrfs_set_opt(info->mount_opt, NOTREELOG);
507 			break;
508 		case Opt_flushoncommit:
509 			printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
510 			btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
511 			break;
512 		case Opt_ratio:
513 			intarg = 0;
514 			match_int(&args[0], &intarg);
515 			if (intarg) {
516 				info->metadata_ratio = intarg;
517 				printk(KERN_INFO "btrfs: metadata ratio %d\n",
518 				       info->metadata_ratio);
519 			}
520 			break;
521 		case Opt_discard:
522 			btrfs_set_opt(info->mount_opt, DISCARD);
523 			break;
524 		case Opt_space_cache:
525 			btrfs_set_opt(info->mount_opt, SPACE_CACHE);
526 			break;
527 		case Opt_no_space_cache:
528 			printk(KERN_INFO "btrfs: disabling disk space caching\n");
529 			btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
530 			break;
531 		case Opt_inode_cache:
532 			printk(KERN_INFO "btrfs: enabling inode map caching\n");
533 			btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
534 			break;
535 		case Opt_clear_cache:
536 			printk(KERN_INFO "btrfs: force clearing of disk cache\n");
537 			btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
538 			break;
539 		case Opt_user_subvol_rm_allowed:
540 			btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
541 			break;
542 		case Opt_enospc_debug:
543 			btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
544 			break;
545 		case Opt_defrag:
546 			printk(KERN_INFO "btrfs: enabling auto defrag");
547 			btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
548 			break;
549 		case Opt_recovery:
550 			printk(KERN_INFO "btrfs: enabling auto recovery");
551 			btrfs_set_opt(info->mount_opt, RECOVERY);
552 			break;
553 		case Opt_skip_balance:
554 			btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
555 			break;
556 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
557 		case Opt_check_integrity_including_extent_data:
558 			printk(KERN_INFO "btrfs: enabling check integrity"
559 			       " including extent data\n");
560 			btrfs_set_opt(info->mount_opt,
561 				      CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
562 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
563 			break;
564 		case Opt_check_integrity:
565 			printk(KERN_INFO "btrfs: enabling check integrity\n");
566 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
567 			break;
568 		case Opt_check_integrity_print_mask:
569 			intarg = 0;
570 			match_int(&args[0], &intarg);
571 			if (intarg) {
572 				info->check_integrity_print_mask = intarg;
573 				printk(KERN_INFO "btrfs:"
574 				       " check_integrity_print_mask 0x%x\n",
575 				       info->check_integrity_print_mask);
576 			}
577 			break;
578 #else
579 		case Opt_check_integrity_including_extent_data:
580 		case Opt_check_integrity:
581 		case Opt_check_integrity_print_mask:
582 			printk(KERN_ERR "btrfs: support for check_integrity*"
583 			       " not compiled in!\n");
584 			ret = -EINVAL;
585 			goto out;
586 #endif
587 		case Opt_fatal_errors:
588 			if (strcmp(args[0].from, "panic") == 0)
589 				btrfs_set_opt(info->mount_opt,
590 					      PANIC_ON_FATAL_ERROR);
591 			else if (strcmp(args[0].from, "bug") == 0)
592 				btrfs_clear_opt(info->mount_opt,
593 					      PANIC_ON_FATAL_ERROR);
594 			else {
595 				ret = -EINVAL;
596 				goto out;
597 			}
598 			break;
599 		case Opt_err:
600 			printk(KERN_INFO "btrfs: unrecognized mount option "
601 			       "'%s'\n", p);
602 			ret = -EINVAL;
603 			goto out;
604 		default:
605 			break;
606 		}
607 	}
608 out:
609 	if (!ret && btrfs_test_opt(root, SPACE_CACHE))
610 		printk(KERN_INFO "btrfs: disk space caching is enabled\n");
611 	kfree(orig);
612 	return ret;
613 }
614 
615 /*
616  * Parse mount options that are required early in the mount process.
617  *
618  * All other options will be parsed on much later in the mount process and
619  * only when we need to allocate a new super block.
620  */
621 static int btrfs_parse_early_options(const char *options, fmode_t flags,
622 		void *holder, char **subvol_name, u64 *subvol_objectid,
623 		u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
624 {
625 	substring_t args[MAX_OPT_ARGS];
626 	char *device_name, *opts, *orig, *p;
627 	int error = 0;
628 	int intarg;
629 
630 	if (!options)
631 		return 0;
632 
633 	/*
634 	 * strsep changes the string, duplicate it because parse_options
635 	 * gets called twice
636 	 */
637 	opts = kstrdup(options, GFP_KERNEL);
638 	if (!opts)
639 		return -ENOMEM;
640 	orig = opts;
641 
642 	while ((p = strsep(&opts, ",")) != NULL) {
643 		int token;
644 		if (!*p)
645 			continue;
646 
647 		token = match_token(p, tokens, args);
648 		switch (token) {
649 		case Opt_subvol:
650 			kfree(*subvol_name);
651 			*subvol_name = match_strdup(&args[0]);
652 			break;
653 		case Opt_subvolid:
654 			intarg = 0;
655 			error = match_int(&args[0], &intarg);
656 			if (!error) {
657 				/* we want the original fs_tree */
658 				if (!intarg)
659 					*subvol_objectid =
660 						BTRFS_FS_TREE_OBJECTID;
661 				else
662 					*subvol_objectid = intarg;
663 			}
664 			break;
665 		case Opt_subvolrootid:
666 			intarg = 0;
667 			error = match_int(&args[0], &intarg);
668 			if (!error) {
669 				/* we want the original fs_tree */
670 				if (!intarg)
671 					*subvol_rootid =
672 						BTRFS_FS_TREE_OBJECTID;
673 				else
674 					*subvol_rootid = intarg;
675 			}
676 			break;
677 		case Opt_device:
678 			device_name = match_strdup(&args[0]);
679 			if (!device_name) {
680 				error = -ENOMEM;
681 				goto out;
682 			}
683 			error = btrfs_scan_one_device(device_name,
684 					flags, holder, fs_devices);
685 			kfree(device_name);
686 			if (error)
687 				goto out;
688 			break;
689 		default:
690 			break;
691 		}
692 	}
693 
694 out:
695 	kfree(orig);
696 	return error;
697 }
698 
699 static struct dentry *get_default_root(struct super_block *sb,
700 				       u64 subvol_objectid)
701 {
702 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
703 	struct btrfs_root *root = fs_info->tree_root;
704 	struct btrfs_root *new_root;
705 	struct btrfs_dir_item *di;
706 	struct btrfs_path *path;
707 	struct btrfs_key location;
708 	struct inode *inode;
709 	u64 dir_id;
710 	int new = 0;
711 
712 	/*
713 	 * We have a specific subvol we want to mount, just setup location and
714 	 * go look up the root.
715 	 */
716 	if (subvol_objectid) {
717 		location.objectid = subvol_objectid;
718 		location.type = BTRFS_ROOT_ITEM_KEY;
719 		location.offset = (u64)-1;
720 		goto find_root;
721 	}
722 
723 	path = btrfs_alloc_path();
724 	if (!path)
725 		return ERR_PTR(-ENOMEM);
726 	path->leave_spinning = 1;
727 
728 	/*
729 	 * Find the "default" dir item which points to the root item that we
730 	 * will mount by default if we haven't been given a specific subvolume
731 	 * to mount.
732 	 */
733 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
734 	di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
735 	if (IS_ERR(di)) {
736 		btrfs_free_path(path);
737 		return ERR_CAST(di);
738 	}
739 	if (!di) {
740 		/*
741 		 * Ok the default dir item isn't there.  This is weird since
742 		 * it's always been there, but don't freak out, just try and
743 		 * mount to root most subvolume.
744 		 */
745 		btrfs_free_path(path);
746 		dir_id = BTRFS_FIRST_FREE_OBJECTID;
747 		new_root = fs_info->fs_root;
748 		goto setup_root;
749 	}
750 
751 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
752 	btrfs_free_path(path);
753 
754 find_root:
755 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
756 	if (IS_ERR(new_root))
757 		return ERR_CAST(new_root);
758 
759 	if (btrfs_root_refs(&new_root->root_item) == 0)
760 		return ERR_PTR(-ENOENT);
761 
762 	dir_id = btrfs_root_dirid(&new_root->root_item);
763 setup_root:
764 	location.objectid = dir_id;
765 	location.type = BTRFS_INODE_ITEM_KEY;
766 	location.offset = 0;
767 
768 	inode = btrfs_iget(sb, &location, new_root, &new);
769 	if (IS_ERR(inode))
770 		return ERR_CAST(inode);
771 
772 	/*
773 	 * If we're just mounting the root most subvol put the inode and return
774 	 * a reference to the dentry.  We will have already gotten a reference
775 	 * to the inode in btrfs_fill_super so we're good to go.
776 	 */
777 	if (!new && sb->s_root->d_inode == inode) {
778 		iput(inode);
779 		return dget(sb->s_root);
780 	}
781 
782 	return d_obtain_alias(inode);
783 }
784 
785 static int btrfs_fill_super(struct super_block *sb,
786 			    struct btrfs_fs_devices *fs_devices,
787 			    void *data, int silent)
788 {
789 	struct inode *inode;
790 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
791 	struct btrfs_key key;
792 	int err;
793 
794 	sb->s_maxbytes = MAX_LFS_FILESIZE;
795 	sb->s_magic = BTRFS_SUPER_MAGIC;
796 	sb->s_op = &btrfs_super_ops;
797 	sb->s_d_op = &btrfs_dentry_operations;
798 	sb->s_export_op = &btrfs_export_ops;
799 	sb->s_xattr = btrfs_xattr_handlers;
800 	sb->s_time_gran = 1;
801 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
802 	sb->s_flags |= MS_POSIXACL;
803 #endif
804 	sb->s_flags |= MS_I_VERSION;
805 	err = open_ctree(sb, fs_devices, (char *)data);
806 	if (err) {
807 		printk("btrfs: open_ctree failed\n");
808 		return err;
809 	}
810 
811 	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
812 	key.type = BTRFS_INODE_ITEM_KEY;
813 	key.offset = 0;
814 	inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
815 	if (IS_ERR(inode)) {
816 		err = PTR_ERR(inode);
817 		goto fail_close;
818 	}
819 
820 	sb->s_root = d_make_root(inode);
821 	if (!sb->s_root) {
822 		err = -ENOMEM;
823 		goto fail_close;
824 	}
825 
826 	save_mount_options(sb, data);
827 	cleancache_init_fs(sb);
828 	sb->s_flags |= MS_ACTIVE;
829 	return 0;
830 
831 fail_close:
832 	close_ctree(fs_info->tree_root);
833 	return err;
834 }
835 
836 int btrfs_sync_fs(struct super_block *sb, int wait)
837 {
838 	struct btrfs_trans_handle *trans;
839 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
840 	struct btrfs_root *root = fs_info->tree_root;
841 	int ret;
842 
843 	trace_btrfs_sync_fs(wait);
844 
845 	if (!wait) {
846 		filemap_flush(fs_info->btree_inode->i_mapping);
847 		return 0;
848 	}
849 
850 	btrfs_wait_ordered_extents(root, 0, 0);
851 
852 	trans = btrfs_start_transaction(root, 0);
853 	if (IS_ERR(trans))
854 		return PTR_ERR(trans);
855 	ret = btrfs_commit_transaction(trans, root);
856 	return ret;
857 }
858 
859 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
860 {
861 	struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
862 	struct btrfs_root *root = info->tree_root;
863 	char *compress_type;
864 
865 	if (btrfs_test_opt(root, DEGRADED))
866 		seq_puts(seq, ",degraded");
867 	if (btrfs_test_opt(root, NODATASUM))
868 		seq_puts(seq, ",nodatasum");
869 	if (btrfs_test_opt(root, NODATACOW))
870 		seq_puts(seq, ",nodatacow");
871 	if (btrfs_test_opt(root, NOBARRIER))
872 		seq_puts(seq, ",nobarrier");
873 	if (info->max_inline != 8192 * 1024)
874 		seq_printf(seq, ",max_inline=%llu",
875 			   (unsigned long long)info->max_inline);
876 	if (info->alloc_start != 0)
877 		seq_printf(seq, ",alloc_start=%llu",
878 			   (unsigned long long)info->alloc_start);
879 	if (info->thread_pool_size !=  min_t(unsigned long,
880 					     num_online_cpus() + 2, 8))
881 		seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
882 	if (btrfs_test_opt(root, COMPRESS)) {
883 		if (info->compress_type == BTRFS_COMPRESS_ZLIB)
884 			compress_type = "zlib";
885 		else
886 			compress_type = "lzo";
887 		if (btrfs_test_opt(root, FORCE_COMPRESS))
888 			seq_printf(seq, ",compress-force=%s", compress_type);
889 		else
890 			seq_printf(seq, ",compress=%s", compress_type);
891 	}
892 	if (btrfs_test_opt(root, NOSSD))
893 		seq_puts(seq, ",nossd");
894 	if (btrfs_test_opt(root, SSD_SPREAD))
895 		seq_puts(seq, ",ssd_spread");
896 	else if (btrfs_test_opt(root, SSD))
897 		seq_puts(seq, ",ssd");
898 	if (btrfs_test_opt(root, NOTREELOG))
899 		seq_puts(seq, ",notreelog");
900 	if (btrfs_test_opt(root, FLUSHONCOMMIT))
901 		seq_puts(seq, ",flushoncommit");
902 	if (btrfs_test_opt(root, DISCARD))
903 		seq_puts(seq, ",discard");
904 	if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
905 		seq_puts(seq, ",noacl");
906 	if (btrfs_test_opt(root, SPACE_CACHE))
907 		seq_puts(seq, ",space_cache");
908 	else
909 		seq_puts(seq, ",nospace_cache");
910 	if (btrfs_test_opt(root, CLEAR_CACHE))
911 		seq_puts(seq, ",clear_cache");
912 	if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
913 		seq_puts(seq, ",user_subvol_rm_allowed");
914 	if (btrfs_test_opt(root, ENOSPC_DEBUG))
915 		seq_puts(seq, ",enospc_debug");
916 	if (btrfs_test_opt(root, AUTO_DEFRAG))
917 		seq_puts(seq, ",autodefrag");
918 	if (btrfs_test_opt(root, INODE_MAP_CACHE))
919 		seq_puts(seq, ",inode_cache");
920 	if (btrfs_test_opt(root, SKIP_BALANCE))
921 		seq_puts(seq, ",skip_balance");
922 	if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
923 		seq_puts(seq, ",fatal_errors=panic");
924 	return 0;
925 }
926 
927 static int btrfs_test_super(struct super_block *s, void *data)
928 {
929 	struct btrfs_fs_info *p = data;
930 	struct btrfs_fs_info *fs_info = btrfs_sb(s);
931 
932 	return fs_info->fs_devices == p->fs_devices;
933 }
934 
935 static int btrfs_set_super(struct super_block *s, void *data)
936 {
937 	int err = set_anon_super(s, data);
938 	if (!err)
939 		s->s_fs_info = data;
940 	return err;
941 }
942 
943 /*
944  * subvolumes are identified by ino 256
945  */
946 static inline int is_subvolume_inode(struct inode *inode)
947 {
948 	if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
949 		return 1;
950 	return 0;
951 }
952 
953 /*
954  * This will strip out the subvol=%s argument for an argument string and add
955  * subvolid=0 to make sure we get the actual tree root for path walking to the
956  * subvol we want.
957  */
958 static char *setup_root_args(char *args)
959 {
960 	unsigned len = strlen(args) + 2 + 1;
961 	char *src, *dst, *buf;
962 
963 	/*
964 	 * We need the same args as before, but with this substitution:
965 	 * s!subvol=[^,]+!subvolid=0!
966 	 *
967 	 * Since the replacement string is up to 2 bytes longer than the
968 	 * original, allocate strlen(args) + 2 + 1 bytes.
969 	 */
970 
971 	src = strstr(args, "subvol=");
972 	/* This shouldn't happen, but just in case.. */
973 	if (!src)
974 		return NULL;
975 
976 	buf = dst = kmalloc(len, GFP_NOFS);
977 	if (!buf)
978 		return NULL;
979 
980 	/*
981 	 * If the subvol= arg is not at the start of the string,
982 	 * copy whatever precedes it into buf.
983 	 */
984 	if (src != args) {
985 		*src++ = '\0';
986 		strcpy(buf, args);
987 		dst += strlen(args);
988 	}
989 
990 	strcpy(dst, "subvolid=0");
991 	dst += strlen("subvolid=0");
992 
993 	/*
994 	 * If there is a "," after the original subvol=... string,
995 	 * copy that suffix into our buffer.  Otherwise, we're done.
996 	 */
997 	src = strchr(src, ',');
998 	if (src)
999 		strcpy(dst, src);
1000 
1001 	return buf;
1002 }
1003 
1004 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1005 				   const char *device_name, char *data)
1006 {
1007 	struct dentry *root;
1008 	struct vfsmount *mnt;
1009 	char *newargs;
1010 
1011 	newargs = setup_root_args(data);
1012 	if (!newargs)
1013 		return ERR_PTR(-ENOMEM);
1014 	mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1015 			     newargs);
1016 	kfree(newargs);
1017 	if (IS_ERR(mnt))
1018 		return ERR_CAST(mnt);
1019 
1020 	root = mount_subtree(mnt, subvol_name);
1021 
1022 	if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1023 		struct super_block *s = root->d_sb;
1024 		dput(root);
1025 		root = ERR_PTR(-EINVAL);
1026 		deactivate_locked_super(s);
1027 		printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1028 				subvol_name);
1029 	}
1030 
1031 	return root;
1032 }
1033 
1034 /*
1035  * Find a superblock for the given device / mount point.
1036  *
1037  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1038  *	  for multiple device setup.  Make sure to keep it in sync.
1039  */
1040 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1041 		const char *device_name, void *data)
1042 {
1043 	struct block_device *bdev = NULL;
1044 	struct super_block *s;
1045 	struct dentry *root;
1046 	struct btrfs_fs_devices *fs_devices = NULL;
1047 	struct btrfs_fs_info *fs_info = NULL;
1048 	fmode_t mode = FMODE_READ;
1049 	char *subvol_name = NULL;
1050 	u64 subvol_objectid = 0;
1051 	u64 subvol_rootid = 0;
1052 	int error = 0;
1053 
1054 	if (!(flags & MS_RDONLY))
1055 		mode |= FMODE_WRITE;
1056 
1057 	error = btrfs_parse_early_options(data, mode, fs_type,
1058 					  &subvol_name, &subvol_objectid,
1059 					  &subvol_rootid, &fs_devices);
1060 	if (error) {
1061 		kfree(subvol_name);
1062 		return ERR_PTR(error);
1063 	}
1064 
1065 	if (subvol_name) {
1066 		root = mount_subvol(subvol_name, flags, device_name, data);
1067 		kfree(subvol_name);
1068 		return root;
1069 	}
1070 
1071 	error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1072 	if (error)
1073 		return ERR_PTR(error);
1074 
1075 	/*
1076 	 * Setup a dummy root and fs_info for test/set super.  This is because
1077 	 * we don't actually fill this stuff out until open_ctree, but we need
1078 	 * it for searching for existing supers, so this lets us do that and
1079 	 * then open_ctree will properly initialize everything later.
1080 	 */
1081 	fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1082 	if (!fs_info)
1083 		return ERR_PTR(-ENOMEM);
1084 
1085 	fs_info->fs_devices = fs_devices;
1086 
1087 	fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1088 	fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1089 	if (!fs_info->super_copy || !fs_info->super_for_commit) {
1090 		error = -ENOMEM;
1091 		goto error_fs_info;
1092 	}
1093 
1094 	error = btrfs_open_devices(fs_devices, mode, fs_type);
1095 	if (error)
1096 		goto error_fs_info;
1097 
1098 	if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1099 		error = -EACCES;
1100 		goto error_close_devices;
1101 	}
1102 
1103 	bdev = fs_devices->latest_bdev;
1104 	s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1105 		 fs_info);
1106 	if (IS_ERR(s)) {
1107 		error = PTR_ERR(s);
1108 		goto error_close_devices;
1109 	}
1110 
1111 	if (s->s_root) {
1112 		btrfs_close_devices(fs_devices);
1113 		free_fs_info(fs_info);
1114 		if ((flags ^ s->s_flags) & MS_RDONLY)
1115 			error = -EBUSY;
1116 	} else {
1117 		char b[BDEVNAME_SIZE];
1118 
1119 		strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1120 		btrfs_sb(s)->bdev_holder = fs_type;
1121 		error = btrfs_fill_super(s, fs_devices, data,
1122 					 flags & MS_SILENT ? 1 : 0);
1123 	}
1124 
1125 	root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1126 	if (IS_ERR(root))
1127 		deactivate_locked_super(s);
1128 
1129 	return root;
1130 
1131 error_close_devices:
1132 	btrfs_close_devices(fs_devices);
1133 error_fs_info:
1134 	free_fs_info(fs_info);
1135 	return ERR_PTR(error);
1136 }
1137 
1138 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1139 {
1140 	spin_lock_irq(&workers->lock);
1141 	workers->max_workers = new_limit;
1142 	spin_unlock_irq(&workers->lock);
1143 }
1144 
1145 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1146 				     int new_pool_size, int old_pool_size)
1147 {
1148 	if (new_pool_size == old_pool_size)
1149 		return;
1150 
1151 	fs_info->thread_pool_size = new_pool_size;
1152 
1153 	printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1154 	       old_pool_size, new_pool_size);
1155 
1156 	btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1157 	btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1158 	btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1159 	btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1160 	btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1161 	btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1162 	btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1163 	btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1164 	btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1165 	btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1166 	btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1167 	btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1168 	btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1169 	btrfs_set_max_workers(&fs_info->scrub_workers, new_pool_size);
1170 }
1171 
1172 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1173 {
1174 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1175 	struct btrfs_root *root = fs_info->tree_root;
1176 	unsigned old_flags = sb->s_flags;
1177 	unsigned long old_opts = fs_info->mount_opt;
1178 	unsigned long old_compress_type = fs_info->compress_type;
1179 	u64 old_max_inline = fs_info->max_inline;
1180 	u64 old_alloc_start = fs_info->alloc_start;
1181 	int old_thread_pool_size = fs_info->thread_pool_size;
1182 	unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1183 	int ret;
1184 
1185 	ret = btrfs_parse_options(root, data);
1186 	if (ret) {
1187 		ret = -EINVAL;
1188 		goto restore;
1189 	}
1190 
1191 	btrfs_resize_thread_pool(fs_info,
1192 		fs_info->thread_pool_size, old_thread_pool_size);
1193 
1194 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1195 		return 0;
1196 
1197 	if (*flags & MS_RDONLY) {
1198 		sb->s_flags |= MS_RDONLY;
1199 
1200 		ret = btrfs_commit_super(root);
1201 		if (ret)
1202 			goto restore;
1203 	} else {
1204 		if (fs_info->fs_devices->rw_devices == 0) {
1205 			ret = -EACCES;
1206 			goto restore;
1207 		}
1208 
1209 		if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1210 			ret = -EINVAL;
1211 			goto restore;
1212 		}
1213 
1214 		ret = btrfs_cleanup_fs_roots(fs_info);
1215 		if (ret)
1216 			goto restore;
1217 
1218 		/* recover relocation */
1219 		ret = btrfs_recover_relocation(root);
1220 		if (ret)
1221 			goto restore;
1222 
1223 		ret = btrfs_resume_balance_async(fs_info);
1224 		if (ret)
1225 			goto restore;
1226 
1227 		sb->s_flags &= ~MS_RDONLY;
1228 	}
1229 
1230 	return 0;
1231 
1232 restore:
1233 	/* We've hit an error - don't reset MS_RDONLY */
1234 	if (sb->s_flags & MS_RDONLY)
1235 		old_flags |= MS_RDONLY;
1236 	sb->s_flags = old_flags;
1237 	fs_info->mount_opt = old_opts;
1238 	fs_info->compress_type = old_compress_type;
1239 	fs_info->max_inline = old_max_inline;
1240 	fs_info->alloc_start = old_alloc_start;
1241 	btrfs_resize_thread_pool(fs_info,
1242 		old_thread_pool_size, fs_info->thread_pool_size);
1243 	fs_info->metadata_ratio = old_metadata_ratio;
1244 	return ret;
1245 }
1246 
1247 /* Used to sort the devices by max_avail(descending sort) */
1248 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1249 				       const void *dev_info2)
1250 {
1251 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
1252 	    ((struct btrfs_device_info *)dev_info2)->max_avail)
1253 		return -1;
1254 	else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1255 		 ((struct btrfs_device_info *)dev_info2)->max_avail)
1256 		return 1;
1257 	else
1258 	return 0;
1259 }
1260 
1261 /*
1262  * sort the devices by max_avail, in which max free extent size of each device
1263  * is stored.(Descending Sort)
1264  */
1265 static inline void btrfs_descending_sort_devices(
1266 					struct btrfs_device_info *devices,
1267 					size_t nr_devices)
1268 {
1269 	sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1270 	     btrfs_cmp_device_free_bytes, NULL);
1271 }
1272 
1273 /*
1274  * The helper to calc the free space on the devices that can be used to store
1275  * file data.
1276  */
1277 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1278 {
1279 	struct btrfs_fs_info *fs_info = root->fs_info;
1280 	struct btrfs_device_info *devices_info;
1281 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1282 	struct btrfs_device *device;
1283 	u64 skip_space;
1284 	u64 type;
1285 	u64 avail_space;
1286 	u64 used_space;
1287 	u64 min_stripe_size;
1288 	int min_stripes = 1, num_stripes = 1;
1289 	int i = 0, nr_devices;
1290 	int ret;
1291 
1292 	nr_devices = fs_info->fs_devices->open_devices;
1293 	BUG_ON(!nr_devices);
1294 
1295 	devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1296 			       GFP_NOFS);
1297 	if (!devices_info)
1298 		return -ENOMEM;
1299 
1300 	/* calc min stripe number for data space alloction */
1301 	type = btrfs_get_alloc_profile(root, 1);
1302 	if (type & BTRFS_BLOCK_GROUP_RAID0) {
1303 		min_stripes = 2;
1304 		num_stripes = nr_devices;
1305 	} else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1306 		min_stripes = 2;
1307 		num_stripes = 2;
1308 	} else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1309 		min_stripes = 4;
1310 		num_stripes = 4;
1311 	}
1312 
1313 	if (type & BTRFS_BLOCK_GROUP_DUP)
1314 		min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1315 	else
1316 		min_stripe_size = BTRFS_STRIPE_LEN;
1317 
1318 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
1319 		if (!device->in_fs_metadata || !device->bdev)
1320 			continue;
1321 
1322 		avail_space = device->total_bytes - device->bytes_used;
1323 
1324 		/* align with stripe_len */
1325 		do_div(avail_space, BTRFS_STRIPE_LEN);
1326 		avail_space *= BTRFS_STRIPE_LEN;
1327 
1328 		/*
1329 		 * In order to avoid overwritting the superblock on the drive,
1330 		 * btrfs starts at an offset of at least 1MB when doing chunk
1331 		 * allocation.
1332 		 */
1333 		skip_space = 1024 * 1024;
1334 
1335 		/* user can set the offset in fs_info->alloc_start. */
1336 		if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1337 		    device->total_bytes)
1338 			skip_space = max(fs_info->alloc_start, skip_space);
1339 
1340 		/*
1341 		 * btrfs can not use the free space in [0, skip_space - 1],
1342 		 * we must subtract it from the total. In order to implement
1343 		 * it, we account the used space in this range first.
1344 		 */
1345 		ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1346 						     &used_space);
1347 		if (ret) {
1348 			kfree(devices_info);
1349 			return ret;
1350 		}
1351 
1352 		/* calc the free space in [0, skip_space - 1] */
1353 		skip_space -= used_space;
1354 
1355 		/*
1356 		 * we can use the free space in [0, skip_space - 1], subtract
1357 		 * it from the total.
1358 		 */
1359 		if (avail_space && avail_space >= skip_space)
1360 			avail_space -= skip_space;
1361 		else
1362 			avail_space = 0;
1363 
1364 		if (avail_space < min_stripe_size)
1365 			continue;
1366 
1367 		devices_info[i].dev = device;
1368 		devices_info[i].max_avail = avail_space;
1369 
1370 		i++;
1371 	}
1372 
1373 	nr_devices = i;
1374 
1375 	btrfs_descending_sort_devices(devices_info, nr_devices);
1376 
1377 	i = nr_devices - 1;
1378 	avail_space = 0;
1379 	while (nr_devices >= min_stripes) {
1380 		if (num_stripes > nr_devices)
1381 			num_stripes = nr_devices;
1382 
1383 		if (devices_info[i].max_avail >= min_stripe_size) {
1384 			int j;
1385 			u64 alloc_size;
1386 
1387 			avail_space += devices_info[i].max_avail * num_stripes;
1388 			alloc_size = devices_info[i].max_avail;
1389 			for (j = i + 1 - num_stripes; j <= i; j++)
1390 				devices_info[j].max_avail -= alloc_size;
1391 		}
1392 		i--;
1393 		nr_devices--;
1394 	}
1395 
1396 	kfree(devices_info);
1397 	*free_bytes = avail_space;
1398 	return 0;
1399 }
1400 
1401 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1402 {
1403 	struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1404 	struct btrfs_super_block *disk_super = fs_info->super_copy;
1405 	struct list_head *head = &fs_info->space_info;
1406 	struct btrfs_space_info *found;
1407 	u64 total_used = 0;
1408 	u64 total_free_data = 0;
1409 	int bits = dentry->d_sb->s_blocksize_bits;
1410 	__be32 *fsid = (__be32 *)fs_info->fsid;
1411 	int ret;
1412 
1413 	/* holding chunk_muext to avoid allocating new chunks */
1414 	mutex_lock(&fs_info->chunk_mutex);
1415 	rcu_read_lock();
1416 	list_for_each_entry_rcu(found, head, list) {
1417 		if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1418 			total_free_data += found->disk_total - found->disk_used;
1419 			total_free_data -=
1420 				btrfs_account_ro_block_groups_free_space(found);
1421 		}
1422 
1423 		total_used += found->disk_used;
1424 	}
1425 	rcu_read_unlock();
1426 
1427 	buf->f_namelen = BTRFS_NAME_LEN;
1428 	buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1429 	buf->f_bfree = buf->f_blocks - (total_used >> bits);
1430 	buf->f_bsize = dentry->d_sb->s_blocksize;
1431 	buf->f_type = BTRFS_SUPER_MAGIC;
1432 	buf->f_bavail = total_free_data;
1433 	ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1434 	if (ret) {
1435 		mutex_unlock(&fs_info->chunk_mutex);
1436 		return ret;
1437 	}
1438 	buf->f_bavail += total_free_data;
1439 	buf->f_bavail = buf->f_bavail >> bits;
1440 	mutex_unlock(&fs_info->chunk_mutex);
1441 
1442 	/* We treat it as constant endianness (it doesn't matter _which_)
1443 	   because we want the fsid to come out the same whether mounted
1444 	   on a big-endian or little-endian host */
1445 	buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1446 	buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1447 	/* Mask in the root object ID too, to disambiguate subvols */
1448 	buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1449 	buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1450 
1451 	return 0;
1452 }
1453 
1454 static void btrfs_kill_super(struct super_block *sb)
1455 {
1456 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1457 	kill_anon_super(sb);
1458 	free_fs_info(fs_info);
1459 }
1460 
1461 static struct file_system_type btrfs_fs_type = {
1462 	.owner		= THIS_MODULE,
1463 	.name		= "btrfs",
1464 	.mount		= btrfs_mount,
1465 	.kill_sb	= btrfs_kill_super,
1466 	.fs_flags	= FS_REQUIRES_DEV,
1467 };
1468 
1469 /*
1470  * used by btrfsctl to scan devices when no FS is mounted
1471  */
1472 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1473 				unsigned long arg)
1474 {
1475 	struct btrfs_ioctl_vol_args *vol;
1476 	struct btrfs_fs_devices *fs_devices;
1477 	int ret = -ENOTTY;
1478 
1479 	if (!capable(CAP_SYS_ADMIN))
1480 		return -EPERM;
1481 
1482 	vol = memdup_user((void __user *)arg, sizeof(*vol));
1483 	if (IS_ERR(vol))
1484 		return PTR_ERR(vol);
1485 
1486 	switch (cmd) {
1487 	case BTRFS_IOC_SCAN_DEV:
1488 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1489 					    &btrfs_fs_type, &fs_devices);
1490 		break;
1491 	case BTRFS_IOC_DEVICES_READY:
1492 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1493 					    &btrfs_fs_type, &fs_devices);
1494 		if (ret)
1495 			break;
1496 		ret = !(fs_devices->num_devices == fs_devices->total_devices);
1497 		break;
1498 	}
1499 
1500 	kfree(vol);
1501 	return ret;
1502 }
1503 
1504 static int btrfs_freeze(struct super_block *sb)
1505 {
1506 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1507 	mutex_lock(&fs_info->transaction_kthread_mutex);
1508 	mutex_lock(&fs_info->cleaner_mutex);
1509 	return 0;
1510 }
1511 
1512 static int btrfs_unfreeze(struct super_block *sb)
1513 {
1514 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1515 	mutex_unlock(&fs_info->cleaner_mutex);
1516 	mutex_unlock(&fs_info->transaction_kthread_mutex);
1517 	return 0;
1518 }
1519 
1520 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1521 {
1522 	struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1523 	struct btrfs_fs_devices *cur_devices;
1524 	struct btrfs_device *dev, *first_dev = NULL;
1525 	struct list_head *head;
1526 	struct rcu_string *name;
1527 
1528 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1529 	cur_devices = fs_info->fs_devices;
1530 	while (cur_devices) {
1531 		head = &cur_devices->devices;
1532 		list_for_each_entry(dev, head, dev_list) {
1533 			if (!first_dev || dev->devid < first_dev->devid)
1534 				first_dev = dev;
1535 		}
1536 		cur_devices = cur_devices->seed;
1537 	}
1538 
1539 	if (first_dev) {
1540 		rcu_read_lock();
1541 		name = rcu_dereference(first_dev->name);
1542 		seq_escape(m, name->str, " \t\n\\");
1543 		rcu_read_unlock();
1544 	} else {
1545 		WARN_ON(1);
1546 	}
1547 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1548 	return 0;
1549 }
1550 
1551 static const struct super_operations btrfs_super_ops = {
1552 	.drop_inode	= btrfs_drop_inode,
1553 	.evict_inode	= btrfs_evict_inode,
1554 	.put_super	= btrfs_put_super,
1555 	.sync_fs	= btrfs_sync_fs,
1556 	.show_options	= btrfs_show_options,
1557 	.show_devname	= btrfs_show_devname,
1558 	.write_inode	= btrfs_write_inode,
1559 	.alloc_inode	= btrfs_alloc_inode,
1560 	.destroy_inode	= btrfs_destroy_inode,
1561 	.statfs		= btrfs_statfs,
1562 	.remount_fs	= btrfs_remount,
1563 	.freeze_fs	= btrfs_freeze,
1564 	.unfreeze_fs	= btrfs_unfreeze,
1565 };
1566 
1567 static const struct file_operations btrfs_ctl_fops = {
1568 	.unlocked_ioctl	 = btrfs_control_ioctl,
1569 	.compat_ioctl = btrfs_control_ioctl,
1570 	.owner	 = THIS_MODULE,
1571 	.llseek = noop_llseek,
1572 };
1573 
1574 static struct miscdevice btrfs_misc = {
1575 	.minor		= BTRFS_MINOR,
1576 	.name		= "btrfs-control",
1577 	.fops		= &btrfs_ctl_fops
1578 };
1579 
1580 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1581 MODULE_ALIAS("devname:btrfs-control");
1582 
1583 static int btrfs_interface_init(void)
1584 {
1585 	return misc_register(&btrfs_misc);
1586 }
1587 
1588 static void btrfs_interface_exit(void)
1589 {
1590 	if (misc_deregister(&btrfs_misc) < 0)
1591 		printk(KERN_INFO "misc_deregister failed for control device");
1592 }
1593 
1594 static int __init init_btrfs_fs(void)
1595 {
1596 	int err;
1597 
1598 	err = btrfs_init_sysfs();
1599 	if (err)
1600 		return err;
1601 
1602 	btrfs_init_compress();
1603 
1604 	err = btrfs_init_cachep();
1605 	if (err)
1606 		goto free_compress;
1607 
1608 	err = extent_io_init();
1609 	if (err)
1610 		goto free_cachep;
1611 
1612 	err = extent_map_init();
1613 	if (err)
1614 		goto free_extent_io;
1615 
1616 	err = btrfs_delayed_inode_init();
1617 	if (err)
1618 		goto free_extent_map;
1619 
1620 	err = btrfs_interface_init();
1621 	if (err)
1622 		goto free_delayed_inode;
1623 
1624 	err = register_filesystem(&btrfs_fs_type);
1625 	if (err)
1626 		goto unregister_ioctl;
1627 
1628 	btrfs_init_lockdep();
1629 
1630 	printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1631 	return 0;
1632 
1633 unregister_ioctl:
1634 	btrfs_interface_exit();
1635 free_delayed_inode:
1636 	btrfs_delayed_inode_exit();
1637 free_extent_map:
1638 	extent_map_exit();
1639 free_extent_io:
1640 	extent_io_exit();
1641 free_cachep:
1642 	btrfs_destroy_cachep();
1643 free_compress:
1644 	btrfs_exit_compress();
1645 	btrfs_exit_sysfs();
1646 	return err;
1647 }
1648 
1649 static void __exit exit_btrfs_fs(void)
1650 {
1651 	btrfs_destroy_cachep();
1652 	btrfs_delayed_inode_exit();
1653 	extent_map_exit();
1654 	extent_io_exit();
1655 	btrfs_interface_exit();
1656 	unregister_filesystem(&btrfs_fs_type);
1657 	btrfs_exit_sysfs();
1658 	btrfs_cleanup_fs_uuids();
1659 	btrfs_exit_compress();
1660 }
1661 
1662 module_init(init_btrfs_fs)
1663 module_exit(exit_btrfs_fs)
1664 
1665 MODULE_LICENSE("GPL");
1666