xref: /linux/fs/btrfs/super.c (revision 1525e06e4414dbe053ea10cbd8cc03bb150e8e27)
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 
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/btrfs.h>
60 
61 static const struct super_operations btrfs_super_ops;
62 static struct file_system_type btrfs_fs_type;
63 
64 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
65 				      char nbuf[16])
66 {
67 	char *errstr = NULL;
68 
69 	switch (errno) {
70 	case -EIO:
71 		errstr = "IO failure";
72 		break;
73 	case -ENOMEM:
74 		errstr = "Out of memory";
75 		break;
76 	case -EROFS:
77 		errstr = "Readonly filesystem";
78 		break;
79 	case -EEXIST:
80 		errstr = "Object already exists";
81 		break;
82 	default:
83 		if (nbuf) {
84 			if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
85 				errstr = nbuf;
86 		}
87 		break;
88 	}
89 
90 	return errstr;
91 }
92 
93 static void __save_error_info(struct btrfs_fs_info *fs_info)
94 {
95 	/*
96 	 * today we only save the error info into ram.  Long term we'll
97 	 * also send it down to the disk
98 	 */
99 	fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
100 }
101 
102 /* NOTE:
103  *	We move write_super stuff at umount in order to avoid deadlock
104  *	for umount hold all lock.
105  */
106 static void save_error_info(struct btrfs_fs_info *fs_info)
107 {
108 	__save_error_info(fs_info);
109 }
110 
111 /* btrfs handle error by forcing the filesystem readonly */
112 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
113 {
114 	struct super_block *sb = fs_info->sb;
115 
116 	if (sb->s_flags & MS_RDONLY)
117 		return;
118 
119 	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
120 		sb->s_flags |= MS_RDONLY;
121 		printk(KERN_INFO "btrfs is forced readonly\n");
122 		__btrfs_scrub_cancel(fs_info);
123 //		WARN_ON(1);
124 	}
125 }
126 
127 /*
128  * __btrfs_std_error decodes expected errors from the caller and
129  * invokes the approciate error response.
130  */
131 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
132 		       unsigned int line, int errno, const char *fmt, ...)
133 {
134 	struct super_block *sb = fs_info->sb;
135 	char nbuf[16];
136 	const char *errstr;
137 	va_list args;
138 	va_start(args, fmt);
139 
140 	/*
141 	 * Special case: if the error is EROFS, and we're already
142 	 * under MS_RDONLY, then it is safe here.
143 	 */
144 	if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
145   		return;
146 
147   	errstr = btrfs_decode_error(fs_info, errno, nbuf);
148 	if (fmt) {
149 		struct va_format vaf = {
150 			.fmt = fmt,
151 			.va = &args,
152 		};
153 
154 		printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
155 			sb->s_id, function, line, errstr, &vaf);
156 	} else {
157 		printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
158 			sb->s_id, function, line, errstr);
159 	}
160 
161 	/* Don't go through full error handling during mount */
162 	if (sb->s_flags & MS_BORN) {
163 		save_error_info(fs_info);
164 		btrfs_handle_error(fs_info);
165 	}
166 	va_end(args);
167 }
168 
169 const char *logtypes[] = {
170 	"emergency",
171 	"alert",
172 	"critical",
173 	"error",
174 	"warning",
175 	"notice",
176 	"info",
177 	"debug",
178 };
179 
180 void btrfs_printk(struct btrfs_fs_info *fs_info, const char *fmt, ...)
181 {
182 	struct super_block *sb = fs_info->sb;
183 	char lvl[4];
184 	struct va_format vaf;
185 	va_list args;
186 	const char *type = logtypes[4];
187 
188 	va_start(args, fmt);
189 
190 	if (fmt[0] == '<' && isdigit(fmt[1]) && fmt[2] == '>') {
191 		memcpy(lvl, fmt, 3);
192 		lvl[3] = '\0';
193 		fmt += 3;
194 		type = logtypes[fmt[1] - '0'];
195 	} else
196 		*lvl = '\0';
197 
198 	vaf.fmt = fmt;
199 	vaf.va = &args;
200 	printk("%sBTRFS %s (device %s): %pV", lvl, type, sb->s_id, &vaf);
201 }
202 
203 /*
204  * We only mark the transaction aborted and then set the file system read-only.
205  * This will prevent new transactions from starting or trying to join this
206  * one.
207  *
208  * This means that error recovery at the call site is limited to freeing
209  * any local memory allocations and passing the error code up without
210  * further cleanup. The transaction should complete as it normally would
211  * in the call path but will return -EIO.
212  *
213  * We'll complete the cleanup in btrfs_end_transaction and
214  * btrfs_commit_transaction.
215  */
216 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
217 			       struct btrfs_root *root, const char *function,
218 			       unsigned int line, int errno)
219 {
220 	WARN_ONCE(1, KERN_DEBUG "btrfs: Transaction aborted");
221 	trans->aborted = errno;
222 	/* Nothing used. The other threads that have joined this
223 	 * transaction may be able to continue. */
224 	if (!trans->blocks_used) {
225 		btrfs_printk(root->fs_info, "Aborting unused transaction.\n");
226 		return;
227 	}
228 	trans->transaction->aborted = errno;
229 	__btrfs_std_error(root->fs_info, function, line, errno, NULL);
230 }
231 /*
232  * __btrfs_panic decodes unexpected, fatal errors from the caller,
233  * issues an alert, and either panics or BUGs, depending on mount options.
234  */
235 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
236 		   unsigned int line, int errno, const char *fmt, ...)
237 {
238 	char nbuf[16];
239 	char *s_id = "<unknown>";
240 	const char *errstr;
241 	struct va_format vaf = { .fmt = fmt };
242 	va_list args;
243 
244 	if (fs_info)
245 		s_id = fs_info->sb->s_id;
246 
247 	va_start(args, fmt);
248 	vaf.va = &args;
249 
250 	errstr = btrfs_decode_error(fs_info, errno, nbuf);
251 	if (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR)
252 		panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
253 			s_id, function, line, &vaf, errstr);
254 
255 	printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
256 	       s_id, function, line, &vaf, errstr);
257 	va_end(args);
258 	/* Caller calls BUG() */
259 }
260 
261 static void btrfs_put_super(struct super_block *sb)
262 {
263 	(void)close_ctree(btrfs_sb(sb)->tree_root);
264 	/* FIXME: need to fix VFS to return error? */
265 	/* AV: return it _where_?  ->put_super() can be triggered by any number
266 	 * of async events, up to and including delivery of SIGKILL to the
267 	 * last process that kept it busy.  Or segfault in the aforementioned
268 	 * process...  Whom would you report that to?
269 	 */
270 }
271 
272 enum {
273 	Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
274 	Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
275 	Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
276 	Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
277 	Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
278 	Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
279 	Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
280 	Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
281 	Opt_check_integrity, Opt_check_integrity_including_extent_data,
282 	Opt_check_integrity_print_mask, Opt_fatal_errors,
283 	Opt_err,
284 };
285 
286 static match_table_t tokens = {
287 	{Opt_degraded, "degraded"},
288 	{Opt_subvol, "subvol=%s"},
289 	{Opt_subvolid, "subvolid=%d"},
290 	{Opt_device, "device=%s"},
291 	{Opt_nodatasum, "nodatasum"},
292 	{Opt_nodatacow, "nodatacow"},
293 	{Opt_nobarrier, "nobarrier"},
294 	{Opt_max_inline, "max_inline=%s"},
295 	{Opt_alloc_start, "alloc_start=%s"},
296 	{Opt_thread_pool, "thread_pool=%d"},
297 	{Opt_compress, "compress"},
298 	{Opt_compress_type, "compress=%s"},
299 	{Opt_compress_force, "compress-force"},
300 	{Opt_compress_force_type, "compress-force=%s"},
301 	{Opt_ssd, "ssd"},
302 	{Opt_ssd_spread, "ssd_spread"},
303 	{Opt_nossd, "nossd"},
304 	{Opt_noacl, "noacl"},
305 	{Opt_notreelog, "notreelog"},
306 	{Opt_flushoncommit, "flushoncommit"},
307 	{Opt_ratio, "metadata_ratio=%d"},
308 	{Opt_discard, "discard"},
309 	{Opt_space_cache, "space_cache"},
310 	{Opt_clear_cache, "clear_cache"},
311 	{Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
312 	{Opt_enospc_debug, "enospc_debug"},
313 	{Opt_subvolrootid, "subvolrootid=%d"},
314 	{Opt_defrag, "autodefrag"},
315 	{Opt_inode_cache, "inode_cache"},
316 	{Opt_no_space_cache, "nospace_cache"},
317 	{Opt_recovery, "recovery"},
318 	{Opt_skip_balance, "skip_balance"},
319 	{Opt_check_integrity, "check_int"},
320 	{Opt_check_integrity_including_extent_data, "check_int_data"},
321 	{Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
322 	{Opt_fatal_errors, "fatal_errors=%s"},
323 	{Opt_err, NULL},
324 };
325 
326 /*
327  * Regular mount options parser.  Everything that is needed only when
328  * reading in a new superblock is parsed here.
329  * XXX JDM: This needs to be cleaned up for remount.
330  */
331 int btrfs_parse_options(struct btrfs_root *root, char *options)
332 {
333 	struct btrfs_fs_info *info = root->fs_info;
334 	substring_t args[MAX_OPT_ARGS];
335 	char *p, *num, *orig = NULL;
336 	u64 cache_gen;
337 	int intarg;
338 	int ret = 0;
339 	char *compress_type;
340 	bool compress_force = false;
341 
342 	cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
343 	if (cache_gen)
344 		btrfs_set_opt(info->mount_opt, SPACE_CACHE);
345 
346 	if (!options)
347 		goto out;
348 
349 	/*
350 	 * strsep changes the string, duplicate it because parse_options
351 	 * gets called twice
352 	 */
353 	options = kstrdup(options, GFP_NOFS);
354 	if (!options)
355 		return -ENOMEM;
356 
357 	orig = options;
358 
359 	while ((p = strsep(&options, ",")) != NULL) {
360 		int token;
361 		if (!*p)
362 			continue;
363 
364 		token = match_token(p, tokens, args);
365 		switch (token) {
366 		case Opt_degraded:
367 			printk(KERN_INFO "btrfs: allowing degraded mounts\n");
368 			btrfs_set_opt(info->mount_opt, DEGRADED);
369 			break;
370 		case Opt_subvol:
371 		case Opt_subvolid:
372 		case Opt_subvolrootid:
373 		case Opt_device:
374 			/*
375 			 * These are parsed by btrfs_parse_early_options
376 			 * and can be happily ignored here.
377 			 */
378 			break;
379 		case Opt_nodatasum:
380 			printk(KERN_INFO "btrfs: setting nodatasum\n");
381 			btrfs_set_opt(info->mount_opt, NODATASUM);
382 			break;
383 		case Opt_nodatacow:
384 			printk(KERN_INFO "btrfs: setting nodatacow\n");
385 			btrfs_set_opt(info->mount_opt, NODATACOW);
386 			btrfs_set_opt(info->mount_opt, NODATASUM);
387 			break;
388 		case Opt_compress_force:
389 		case Opt_compress_force_type:
390 			compress_force = true;
391 		case Opt_compress:
392 		case Opt_compress_type:
393 			if (token == Opt_compress ||
394 			    token == Opt_compress_force ||
395 			    strcmp(args[0].from, "zlib") == 0) {
396 				compress_type = "zlib";
397 				info->compress_type = BTRFS_COMPRESS_ZLIB;
398 			} else if (strcmp(args[0].from, "lzo") == 0) {
399 				compress_type = "lzo";
400 				info->compress_type = BTRFS_COMPRESS_LZO;
401 			} else {
402 				ret = -EINVAL;
403 				goto out;
404 			}
405 
406 			btrfs_set_opt(info->mount_opt, COMPRESS);
407 			if (compress_force) {
408 				btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
409 				pr_info("btrfs: force %s compression\n",
410 					compress_type);
411 			} else
412 				pr_info("btrfs: use %s compression\n",
413 					compress_type);
414 			break;
415 		case Opt_ssd:
416 			printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
417 			btrfs_set_opt(info->mount_opt, SSD);
418 			break;
419 		case Opt_ssd_spread:
420 			printk(KERN_INFO "btrfs: use spread ssd "
421 			       "allocation scheme\n");
422 			btrfs_set_opt(info->mount_opt, SSD);
423 			btrfs_set_opt(info->mount_opt, SSD_SPREAD);
424 			break;
425 		case Opt_nossd:
426 			printk(KERN_INFO "btrfs: not using ssd allocation "
427 			       "scheme\n");
428 			btrfs_set_opt(info->mount_opt, NOSSD);
429 			btrfs_clear_opt(info->mount_opt, SSD);
430 			btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
431 			break;
432 		case Opt_nobarrier:
433 			printk(KERN_INFO "btrfs: turning off barriers\n");
434 			btrfs_set_opt(info->mount_opt, NOBARRIER);
435 			break;
436 		case Opt_thread_pool:
437 			intarg = 0;
438 			match_int(&args[0], &intarg);
439 			if (intarg)
440 				info->thread_pool_size = intarg;
441 			break;
442 		case Opt_max_inline:
443 			num = match_strdup(&args[0]);
444 			if (num) {
445 				info->max_inline = memparse(num, NULL);
446 				kfree(num);
447 
448 				if (info->max_inline) {
449 					info->max_inline = max_t(u64,
450 						info->max_inline,
451 						root->sectorsize);
452 				}
453 				printk(KERN_INFO "btrfs: max_inline at %llu\n",
454 					(unsigned long long)info->max_inline);
455 			}
456 			break;
457 		case Opt_alloc_start:
458 			num = match_strdup(&args[0]);
459 			if (num) {
460 				info->alloc_start = memparse(num, NULL);
461 				kfree(num);
462 				printk(KERN_INFO
463 					"btrfs: allocations start at %llu\n",
464 					(unsigned long long)info->alloc_start);
465 			}
466 			break;
467 		case Opt_noacl:
468 			root->fs_info->sb->s_flags &= ~MS_POSIXACL;
469 			break;
470 		case Opt_notreelog:
471 			printk(KERN_INFO "btrfs: disabling tree log\n");
472 			btrfs_set_opt(info->mount_opt, NOTREELOG);
473 			break;
474 		case Opt_flushoncommit:
475 			printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
476 			btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
477 			break;
478 		case Opt_ratio:
479 			intarg = 0;
480 			match_int(&args[0], &intarg);
481 			if (intarg) {
482 				info->metadata_ratio = intarg;
483 				printk(KERN_INFO "btrfs: metadata ratio %d\n",
484 				       info->metadata_ratio);
485 			}
486 			break;
487 		case Opt_discard:
488 			btrfs_set_opt(info->mount_opt, DISCARD);
489 			break;
490 		case Opt_space_cache:
491 			btrfs_set_opt(info->mount_opt, SPACE_CACHE);
492 			break;
493 		case Opt_no_space_cache:
494 			printk(KERN_INFO "btrfs: disabling disk space caching\n");
495 			btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
496 			break;
497 		case Opt_inode_cache:
498 			printk(KERN_INFO "btrfs: enabling inode map caching\n");
499 			btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
500 			break;
501 		case Opt_clear_cache:
502 			printk(KERN_INFO "btrfs: force clearing of disk cache\n");
503 			btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
504 			break;
505 		case Opt_user_subvol_rm_allowed:
506 			btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
507 			break;
508 		case Opt_enospc_debug:
509 			btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
510 			break;
511 		case Opt_defrag:
512 			printk(KERN_INFO "btrfs: enabling auto defrag");
513 			btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
514 			break;
515 		case Opt_recovery:
516 			printk(KERN_INFO "btrfs: enabling auto recovery");
517 			btrfs_set_opt(info->mount_opt, RECOVERY);
518 			break;
519 		case Opt_skip_balance:
520 			btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
521 			break;
522 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
523 		case Opt_check_integrity_including_extent_data:
524 			printk(KERN_INFO "btrfs: enabling check integrity"
525 			       " including extent data\n");
526 			btrfs_set_opt(info->mount_opt,
527 				      CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
528 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
529 			break;
530 		case Opt_check_integrity:
531 			printk(KERN_INFO "btrfs: enabling check integrity\n");
532 			btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
533 			break;
534 		case Opt_check_integrity_print_mask:
535 			intarg = 0;
536 			match_int(&args[0], &intarg);
537 			if (intarg) {
538 				info->check_integrity_print_mask = intarg;
539 				printk(KERN_INFO "btrfs:"
540 				       " check_integrity_print_mask 0x%x\n",
541 				       info->check_integrity_print_mask);
542 			}
543 			break;
544 #else
545 		case Opt_check_integrity_including_extent_data:
546 		case Opt_check_integrity:
547 		case Opt_check_integrity_print_mask:
548 			printk(KERN_ERR "btrfs: support for check_integrity*"
549 			       " not compiled in!\n");
550 			ret = -EINVAL;
551 			goto out;
552 #endif
553 		case Opt_fatal_errors:
554 			if (strcmp(args[0].from, "panic") == 0)
555 				btrfs_set_opt(info->mount_opt,
556 					      PANIC_ON_FATAL_ERROR);
557 			else if (strcmp(args[0].from, "bug") == 0)
558 				btrfs_clear_opt(info->mount_opt,
559 					      PANIC_ON_FATAL_ERROR);
560 			else {
561 				ret = -EINVAL;
562 				goto out;
563 			}
564 			break;
565 		case Opt_err:
566 			printk(KERN_INFO "btrfs: unrecognized mount option "
567 			       "'%s'\n", p);
568 			ret = -EINVAL;
569 			goto out;
570 		default:
571 			break;
572 		}
573 	}
574 out:
575 	if (!ret && btrfs_test_opt(root, SPACE_CACHE))
576 		printk(KERN_INFO "btrfs: disk space caching is enabled\n");
577 	kfree(orig);
578 	return ret;
579 }
580 
581 /*
582  * Parse mount options that are required early in the mount process.
583  *
584  * All other options will be parsed on much later in the mount process and
585  * only when we need to allocate a new super block.
586  */
587 static int btrfs_parse_early_options(const char *options, fmode_t flags,
588 		void *holder, char **subvol_name, u64 *subvol_objectid,
589 		u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
590 {
591 	substring_t args[MAX_OPT_ARGS];
592 	char *device_name, *opts, *orig, *p;
593 	int error = 0;
594 	int intarg;
595 
596 	if (!options)
597 		return 0;
598 
599 	/*
600 	 * strsep changes the string, duplicate it because parse_options
601 	 * gets called twice
602 	 */
603 	opts = kstrdup(options, GFP_KERNEL);
604 	if (!opts)
605 		return -ENOMEM;
606 	orig = opts;
607 
608 	while ((p = strsep(&opts, ",")) != NULL) {
609 		int token;
610 		if (!*p)
611 			continue;
612 
613 		token = match_token(p, tokens, args);
614 		switch (token) {
615 		case Opt_subvol:
616 			kfree(*subvol_name);
617 			*subvol_name = match_strdup(&args[0]);
618 			break;
619 		case Opt_subvolid:
620 			intarg = 0;
621 			error = match_int(&args[0], &intarg);
622 			if (!error) {
623 				/* we want the original fs_tree */
624 				if (!intarg)
625 					*subvol_objectid =
626 						BTRFS_FS_TREE_OBJECTID;
627 				else
628 					*subvol_objectid = intarg;
629 			}
630 			break;
631 		case Opt_subvolrootid:
632 			intarg = 0;
633 			error = match_int(&args[0], &intarg);
634 			if (!error) {
635 				/* we want the original fs_tree */
636 				if (!intarg)
637 					*subvol_rootid =
638 						BTRFS_FS_TREE_OBJECTID;
639 				else
640 					*subvol_rootid = intarg;
641 			}
642 			break;
643 		case Opt_device:
644 			device_name = match_strdup(&args[0]);
645 			if (!device_name) {
646 				error = -ENOMEM;
647 				goto out;
648 			}
649 			error = btrfs_scan_one_device(device_name,
650 					flags, holder, fs_devices);
651 			kfree(device_name);
652 			if (error)
653 				goto out;
654 			break;
655 		default:
656 			break;
657 		}
658 	}
659 
660 out:
661 	kfree(orig);
662 	return error;
663 }
664 
665 static struct dentry *get_default_root(struct super_block *sb,
666 				       u64 subvol_objectid)
667 {
668 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
669 	struct btrfs_root *root = fs_info->tree_root;
670 	struct btrfs_root *new_root;
671 	struct btrfs_dir_item *di;
672 	struct btrfs_path *path;
673 	struct btrfs_key location;
674 	struct inode *inode;
675 	u64 dir_id;
676 	int new = 0;
677 
678 	/*
679 	 * We have a specific subvol we want to mount, just setup location and
680 	 * go look up the root.
681 	 */
682 	if (subvol_objectid) {
683 		location.objectid = subvol_objectid;
684 		location.type = BTRFS_ROOT_ITEM_KEY;
685 		location.offset = (u64)-1;
686 		goto find_root;
687 	}
688 
689 	path = btrfs_alloc_path();
690 	if (!path)
691 		return ERR_PTR(-ENOMEM);
692 	path->leave_spinning = 1;
693 
694 	/*
695 	 * Find the "default" dir item which points to the root item that we
696 	 * will mount by default if we haven't been given a specific subvolume
697 	 * to mount.
698 	 */
699 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
700 	di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
701 	if (IS_ERR(di)) {
702 		btrfs_free_path(path);
703 		return ERR_CAST(di);
704 	}
705 	if (!di) {
706 		/*
707 		 * Ok the default dir item isn't there.  This is weird since
708 		 * it's always been there, but don't freak out, just try and
709 		 * mount to root most subvolume.
710 		 */
711 		btrfs_free_path(path);
712 		dir_id = BTRFS_FIRST_FREE_OBJECTID;
713 		new_root = fs_info->fs_root;
714 		goto setup_root;
715 	}
716 
717 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
718 	btrfs_free_path(path);
719 
720 find_root:
721 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
722 	if (IS_ERR(new_root))
723 		return ERR_CAST(new_root);
724 
725 	if (btrfs_root_refs(&new_root->root_item) == 0)
726 		return ERR_PTR(-ENOENT);
727 
728 	dir_id = btrfs_root_dirid(&new_root->root_item);
729 setup_root:
730 	location.objectid = dir_id;
731 	location.type = BTRFS_INODE_ITEM_KEY;
732 	location.offset = 0;
733 
734 	inode = btrfs_iget(sb, &location, new_root, &new);
735 	if (IS_ERR(inode))
736 		return ERR_CAST(inode);
737 
738 	/*
739 	 * If we're just mounting the root most subvol put the inode and return
740 	 * a reference to the dentry.  We will have already gotten a reference
741 	 * to the inode in btrfs_fill_super so we're good to go.
742 	 */
743 	if (!new && sb->s_root->d_inode == inode) {
744 		iput(inode);
745 		return dget(sb->s_root);
746 	}
747 
748 	return d_obtain_alias(inode);
749 }
750 
751 static int btrfs_fill_super(struct super_block *sb,
752 			    struct btrfs_fs_devices *fs_devices,
753 			    void *data, int silent)
754 {
755 	struct inode *inode;
756 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
757 	struct btrfs_key key;
758 	int err;
759 
760 	sb->s_maxbytes = MAX_LFS_FILESIZE;
761 	sb->s_magic = BTRFS_SUPER_MAGIC;
762 	sb->s_op = &btrfs_super_ops;
763 	sb->s_d_op = &btrfs_dentry_operations;
764 	sb->s_export_op = &btrfs_export_ops;
765 	sb->s_xattr = btrfs_xattr_handlers;
766 	sb->s_time_gran = 1;
767 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
768 	sb->s_flags |= MS_POSIXACL;
769 #endif
770 	sb->s_flags |= MS_I_VERSION;
771 	err = open_ctree(sb, fs_devices, (char *)data);
772 	if (err) {
773 		printk("btrfs: open_ctree failed\n");
774 		return err;
775 	}
776 
777 	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
778 	key.type = BTRFS_INODE_ITEM_KEY;
779 	key.offset = 0;
780 	inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
781 	if (IS_ERR(inode)) {
782 		err = PTR_ERR(inode);
783 		goto fail_close;
784 	}
785 
786 	sb->s_root = d_make_root(inode);
787 	if (!sb->s_root) {
788 		err = -ENOMEM;
789 		goto fail_close;
790 	}
791 
792 	save_mount_options(sb, data);
793 	cleancache_init_fs(sb);
794 	sb->s_flags |= MS_ACTIVE;
795 	return 0;
796 
797 fail_close:
798 	close_ctree(fs_info->tree_root);
799 	return err;
800 }
801 
802 int btrfs_sync_fs(struct super_block *sb, int wait)
803 {
804 	struct btrfs_trans_handle *trans;
805 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
806 	struct btrfs_root *root = fs_info->tree_root;
807 	int ret;
808 
809 	trace_btrfs_sync_fs(wait);
810 
811 	if (!wait) {
812 		filemap_flush(fs_info->btree_inode->i_mapping);
813 		return 0;
814 	}
815 
816 	btrfs_wait_ordered_extents(root, 0, 0);
817 
818 	trans = btrfs_start_transaction(root, 0);
819 	if (IS_ERR(trans))
820 		return PTR_ERR(trans);
821 	ret = btrfs_commit_transaction(trans, root);
822 	return ret;
823 }
824 
825 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
826 {
827 	struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
828 	struct btrfs_root *root = info->tree_root;
829 	char *compress_type;
830 
831 	if (btrfs_test_opt(root, DEGRADED))
832 		seq_puts(seq, ",degraded");
833 	if (btrfs_test_opt(root, NODATASUM))
834 		seq_puts(seq, ",nodatasum");
835 	if (btrfs_test_opt(root, NODATACOW))
836 		seq_puts(seq, ",nodatacow");
837 	if (btrfs_test_opt(root, NOBARRIER))
838 		seq_puts(seq, ",nobarrier");
839 	if (info->max_inline != 8192 * 1024)
840 		seq_printf(seq, ",max_inline=%llu",
841 			   (unsigned long long)info->max_inline);
842 	if (info->alloc_start != 0)
843 		seq_printf(seq, ",alloc_start=%llu",
844 			   (unsigned long long)info->alloc_start);
845 	if (info->thread_pool_size !=  min_t(unsigned long,
846 					     num_online_cpus() + 2, 8))
847 		seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
848 	if (btrfs_test_opt(root, COMPRESS)) {
849 		if (info->compress_type == BTRFS_COMPRESS_ZLIB)
850 			compress_type = "zlib";
851 		else
852 			compress_type = "lzo";
853 		if (btrfs_test_opt(root, FORCE_COMPRESS))
854 			seq_printf(seq, ",compress-force=%s", compress_type);
855 		else
856 			seq_printf(seq, ",compress=%s", compress_type);
857 	}
858 	if (btrfs_test_opt(root, NOSSD))
859 		seq_puts(seq, ",nossd");
860 	if (btrfs_test_opt(root, SSD_SPREAD))
861 		seq_puts(seq, ",ssd_spread");
862 	else if (btrfs_test_opt(root, SSD))
863 		seq_puts(seq, ",ssd");
864 	if (btrfs_test_opt(root, NOTREELOG))
865 		seq_puts(seq, ",notreelog");
866 	if (btrfs_test_opt(root, FLUSHONCOMMIT))
867 		seq_puts(seq, ",flushoncommit");
868 	if (btrfs_test_opt(root, DISCARD))
869 		seq_puts(seq, ",discard");
870 	if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
871 		seq_puts(seq, ",noacl");
872 	if (btrfs_test_opt(root, SPACE_CACHE))
873 		seq_puts(seq, ",space_cache");
874 	else
875 		seq_puts(seq, ",nospace_cache");
876 	if (btrfs_test_opt(root, CLEAR_CACHE))
877 		seq_puts(seq, ",clear_cache");
878 	if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
879 		seq_puts(seq, ",user_subvol_rm_allowed");
880 	if (btrfs_test_opt(root, ENOSPC_DEBUG))
881 		seq_puts(seq, ",enospc_debug");
882 	if (btrfs_test_opt(root, AUTO_DEFRAG))
883 		seq_puts(seq, ",autodefrag");
884 	if (btrfs_test_opt(root, INODE_MAP_CACHE))
885 		seq_puts(seq, ",inode_cache");
886 	if (btrfs_test_opt(root, SKIP_BALANCE))
887 		seq_puts(seq, ",skip_balance");
888 	if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
889 		seq_puts(seq, ",fatal_errors=panic");
890 	return 0;
891 }
892 
893 static int btrfs_test_super(struct super_block *s, void *data)
894 {
895 	struct btrfs_fs_info *p = data;
896 	struct btrfs_fs_info *fs_info = btrfs_sb(s);
897 
898 	return fs_info->fs_devices == p->fs_devices;
899 }
900 
901 static int btrfs_set_super(struct super_block *s, void *data)
902 {
903 	int err = set_anon_super(s, data);
904 	if (!err)
905 		s->s_fs_info = data;
906 	return err;
907 }
908 
909 /*
910  * subvolumes are identified by ino 256
911  */
912 static inline int is_subvolume_inode(struct inode *inode)
913 {
914 	if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
915 		return 1;
916 	return 0;
917 }
918 
919 /*
920  * This will strip out the subvol=%s argument for an argument string and add
921  * subvolid=0 to make sure we get the actual tree root for path walking to the
922  * subvol we want.
923  */
924 static char *setup_root_args(char *args)
925 {
926 	unsigned len = strlen(args) + 2 + 1;
927 	char *src, *dst, *buf;
928 
929 	/*
930 	 * We need the same args as before, but with this substitution:
931 	 * s!subvol=[^,]+!subvolid=0!
932 	 *
933 	 * Since the replacement string is up to 2 bytes longer than the
934 	 * original, allocate strlen(args) + 2 + 1 bytes.
935 	 */
936 
937 	src = strstr(args, "subvol=");
938 	/* This shouldn't happen, but just in case.. */
939 	if (!src)
940 		return NULL;
941 
942 	buf = dst = kmalloc(len, GFP_NOFS);
943 	if (!buf)
944 		return NULL;
945 
946 	/*
947 	 * If the subvol= arg is not at the start of the string,
948 	 * copy whatever precedes it into buf.
949 	 */
950 	if (src != args) {
951 		*src++ = '\0';
952 		strcpy(buf, args);
953 		dst += strlen(args);
954 	}
955 
956 	strcpy(dst, "subvolid=0");
957 	dst += strlen("subvolid=0");
958 
959 	/*
960 	 * If there is a "," after the original subvol=... string,
961 	 * copy that suffix into our buffer.  Otherwise, we're done.
962 	 */
963 	src = strchr(src, ',');
964 	if (src)
965 		strcpy(dst, src);
966 
967 	return buf;
968 }
969 
970 static struct dentry *mount_subvol(const char *subvol_name, int flags,
971 				   const char *device_name, char *data)
972 {
973 	struct dentry *root;
974 	struct vfsmount *mnt;
975 	char *newargs;
976 
977 	newargs = setup_root_args(data);
978 	if (!newargs)
979 		return ERR_PTR(-ENOMEM);
980 	mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
981 			     newargs);
982 	kfree(newargs);
983 	if (IS_ERR(mnt))
984 		return ERR_CAST(mnt);
985 
986 	root = mount_subtree(mnt, subvol_name);
987 
988 	if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
989 		struct super_block *s = root->d_sb;
990 		dput(root);
991 		root = ERR_PTR(-EINVAL);
992 		deactivate_locked_super(s);
993 		printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
994 				subvol_name);
995 	}
996 
997 	return root;
998 }
999 
1000 /*
1001  * Find a superblock for the given device / mount point.
1002  *
1003  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1004  *	  for multiple device setup.  Make sure to keep it in sync.
1005  */
1006 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1007 		const char *device_name, void *data)
1008 {
1009 	struct block_device *bdev = NULL;
1010 	struct super_block *s;
1011 	struct dentry *root;
1012 	struct btrfs_fs_devices *fs_devices = NULL;
1013 	struct btrfs_fs_info *fs_info = NULL;
1014 	fmode_t mode = FMODE_READ;
1015 	char *subvol_name = NULL;
1016 	u64 subvol_objectid = 0;
1017 	u64 subvol_rootid = 0;
1018 	int error = 0;
1019 
1020 	if (!(flags & MS_RDONLY))
1021 		mode |= FMODE_WRITE;
1022 
1023 	error = btrfs_parse_early_options(data, mode, fs_type,
1024 					  &subvol_name, &subvol_objectid,
1025 					  &subvol_rootid, &fs_devices);
1026 	if (error) {
1027 		kfree(subvol_name);
1028 		return ERR_PTR(error);
1029 	}
1030 
1031 	if (subvol_name) {
1032 		root = mount_subvol(subvol_name, flags, device_name, data);
1033 		kfree(subvol_name);
1034 		return root;
1035 	}
1036 
1037 	error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1038 	if (error)
1039 		return ERR_PTR(error);
1040 
1041 	/*
1042 	 * Setup a dummy root and fs_info for test/set super.  This is because
1043 	 * we don't actually fill this stuff out until open_ctree, but we need
1044 	 * it for searching for existing supers, so this lets us do that and
1045 	 * then open_ctree will properly initialize everything later.
1046 	 */
1047 	fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1048 	if (!fs_info)
1049 		return ERR_PTR(-ENOMEM);
1050 
1051 	fs_info->fs_devices = fs_devices;
1052 
1053 	fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1054 	fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1055 	if (!fs_info->super_copy || !fs_info->super_for_commit) {
1056 		error = -ENOMEM;
1057 		goto error_fs_info;
1058 	}
1059 
1060 	error = btrfs_open_devices(fs_devices, mode, fs_type);
1061 	if (error)
1062 		goto error_fs_info;
1063 
1064 	if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1065 		error = -EACCES;
1066 		goto error_close_devices;
1067 	}
1068 
1069 	bdev = fs_devices->latest_bdev;
1070 	s = sget(fs_type, btrfs_test_super, btrfs_set_super, fs_info);
1071 	if (IS_ERR(s)) {
1072 		error = PTR_ERR(s);
1073 		goto error_close_devices;
1074 	}
1075 
1076 	if (s->s_root) {
1077 		btrfs_close_devices(fs_devices);
1078 		free_fs_info(fs_info);
1079 		if ((flags ^ s->s_flags) & MS_RDONLY)
1080 			error = -EBUSY;
1081 	} else {
1082 		char b[BDEVNAME_SIZE];
1083 
1084 		s->s_flags = flags | MS_NOSEC;
1085 		strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1086 		btrfs_sb(s)->bdev_holder = fs_type;
1087 		error = btrfs_fill_super(s, fs_devices, data,
1088 					 flags & MS_SILENT ? 1 : 0);
1089 	}
1090 
1091 	root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1092 	if (IS_ERR(root))
1093 		deactivate_locked_super(s);
1094 
1095 	return root;
1096 
1097 error_close_devices:
1098 	btrfs_close_devices(fs_devices);
1099 error_fs_info:
1100 	free_fs_info(fs_info);
1101 	return ERR_PTR(error);
1102 }
1103 
1104 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1105 {
1106 	spin_lock_irq(&workers->lock);
1107 	workers->max_workers = new_limit;
1108 	spin_unlock_irq(&workers->lock);
1109 }
1110 
1111 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1112 				     int new_pool_size, int old_pool_size)
1113 {
1114 	if (new_pool_size == old_pool_size)
1115 		return;
1116 
1117 	fs_info->thread_pool_size = new_pool_size;
1118 
1119 	printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1120 	       old_pool_size, new_pool_size);
1121 
1122 	btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1123 	btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1124 	btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1125 	btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1126 	btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1127 	btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1128 	btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1129 	btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1130 	btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1131 	btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1132 	btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1133 	btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1134 	btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1135 	btrfs_set_max_workers(&fs_info->scrub_workers, new_pool_size);
1136 }
1137 
1138 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1139 {
1140 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1141 	struct btrfs_root *root = fs_info->tree_root;
1142 	unsigned old_flags = sb->s_flags;
1143 	unsigned long old_opts = fs_info->mount_opt;
1144 	unsigned long old_compress_type = fs_info->compress_type;
1145 	u64 old_max_inline = fs_info->max_inline;
1146 	u64 old_alloc_start = fs_info->alloc_start;
1147 	int old_thread_pool_size = fs_info->thread_pool_size;
1148 	unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1149 	int ret;
1150 
1151 	ret = btrfs_parse_options(root, data);
1152 	if (ret) {
1153 		ret = -EINVAL;
1154 		goto restore;
1155 	}
1156 
1157 	btrfs_resize_thread_pool(fs_info,
1158 		fs_info->thread_pool_size, old_thread_pool_size);
1159 
1160 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1161 		return 0;
1162 
1163 	if (*flags & MS_RDONLY) {
1164 		sb->s_flags |= MS_RDONLY;
1165 
1166 		ret = btrfs_commit_super(root);
1167 		if (ret)
1168 			goto restore;
1169 	} else {
1170 		if (fs_info->fs_devices->rw_devices == 0) {
1171 			ret = -EACCES;
1172 			goto restore;
1173 		}
1174 
1175 		if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1176 			ret = -EINVAL;
1177 			goto restore;
1178 		}
1179 
1180 		ret = btrfs_cleanup_fs_roots(fs_info);
1181 		if (ret)
1182 			goto restore;
1183 
1184 		/* recover relocation */
1185 		ret = btrfs_recover_relocation(root);
1186 		if (ret)
1187 			goto restore;
1188 
1189 		sb->s_flags &= ~MS_RDONLY;
1190 	}
1191 
1192 	return 0;
1193 
1194 restore:
1195 	/* We've hit an error - don't reset MS_RDONLY */
1196 	if (sb->s_flags & MS_RDONLY)
1197 		old_flags |= MS_RDONLY;
1198 	sb->s_flags = old_flags;
1199 	fs_info->mount_opt = old_opts;
1200 	fs_info->compress_type = old_compress_type;
1201 	fs_info->max_inline = old_max_inline;
1202 	fs_info->alloc_start = old_alloc_start;
1203 	btrfs_resize_thread_pool(fs_info,
1204 		old_thread_pool_size, fs_info->thread_pool_size);
1205 	fs_info->metadata_ratio = old_metadata_ratio;
1206 	return ret;
1207 }
1208 
1209 /* Used to sort the devices by max_avail(descending sort) */
1210 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1211 				       const void *dev_info2)
1212 {
1213 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
1214 	    ((struct btrfs_device_info *)dev_info2)->max_avail)
1215 		return -1;
1216 	else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1217 		 ((struct btrfs_device_info *)dev_info2)->max_avail)
1218 		return 1;
1219 	else
1220 	return 0;
1221 }
1222 
1223 /*
1224  * sort the devices by max_avail, in which max free extent size of each device
1225  * is stored.(Descending Sort)
1226  */
1227 static inline void btrfs_descending_sort_devices(
1228 					struct btrfs_device_info *devices,
1229 					size_t nr_devices)
1230 {
1231 	sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1232 	     btrfs_cmp_device_free_bytes, NULL);
1233 }
1234 
1235 /*
1236  * The helper to calc the free space on the devices that can be used to store
1237  * file data.
1238  */
1239 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1240 {
1241 	struct btrfs_fs_info *fs_info = root->fs_info;
1242 	struct btrfs_device_info *devices_info;
1243 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1244 	struct btrfs_device *device;
1245 	u64 skip_space;
1246 	u64 type;
1247 	u64 avail_space;
1248 	u64 used_space;
1249 	u64 min_stripe_size;
1250 	int min_stripes = 1, num_stripes = 1;
1251 	int i = 0, nr_devices;
1252 	int ret;
1253 
1254 	nr_devices = fs_info->fs_devices->open_devices;
1255 	BUG_ON(!nr_devices);
1256 
1257 	devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1258 			       GFP_NOFS);
1259 	if (!devices_info)
1260 		return -ENOMEM;
1261 
1262 	/* calc min stripe number for data space alloction */
1263 	type = btrfs_get_alloc_profile(root, 1);
1264 	if (type & BTRFS_BLOCK_GROUP_RAID0) {
1265 		min_stripes = 2;
1266 		num_stripes = nr_devices;
1267 	} else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1268 		min_stripes = 2;
1269 		num_stripes = 2;
1270 	} else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1271 		min_stripes = 4;
1272 		num_stripes = 4;
1273 	}
1274 
1275 	if (type & BTRFS_BLOCK_GROUP_DUP)
1276 		min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1277 	else
1278 		min_stripe_size = BTRFS_STRIPE_LEN;
1279 
1280 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
1281 		if (!device->in_fs_metadata || !device->bdev)
1282 			continue;
1283 
1284 		avail_space = device->total_bytes - device->bytes_used;
1285 
1286 		/* align with stripe_len */
1287 		do_div(avail_space, BTRFS_STRIPE_LEN);
1288 		avail_space *= BTRFS_STRIPE_LEN;
1289 
1290 		/*
1291 		 * In order to avoid overwritting the superblock on the drive,
1292 		 * btrfs starts at an offset of at least 1MB when doing chunk
1293 		 * allocation.
1294 		 */
1295 		skip_space = 1024 * 1024;
1296 
1297 		/* user can set the offset in fs_info->alloc_start. */
1298 		if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1299 		    device->total_bytes)
1300 			skip_space = max(fs_info->alloc_start, skip_space);
1301 
1302 		/*
1303 		 * btrfs can not use the free space in [0, skip_space - 1],
1304 		 * we must subtract it from the total. In order to implement
1305 		 * it, we account the used space in this range first.
1306 		 */
1307 		ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1308 						     &used_space);
1309 		if (ret) {
1310 			kfree(devices_info);
1311 			return ret;
1312 		}
1313 
1314 		/* calc the free space in [0, skip_space - 1] */
1315 		skip_space -= used_space;
1316 
1317 		/*
1318 		 * we can use the free space in [0, skip_space - 1], subtract
1319 		 * it from the total.
1320 		 */
1321 		if (avail_space && avail_space >= skip_space)
1322 			avail_space -= skip_space;
1323 		else
1324 			avail_space = 0;
1325 
1326 		if (avail_space < min_stripe_size)
1327 			continue;
1328 
1329 		devices_info[i].dev = device;
1330 		devices_info[i].max_avail = avail_space;
1331 
1332 		i++;
1333 	}
1334 
1335 	nr_devices = i;
1336 
1337 	btrfs_descending_sort_devices(devices_info, nr_devices);
1338 
1339 	i = nr_devices - 1;
1340 	avail_space = 0;
1341 	while (nr_devices >= min_stripes) {
1342 		if (num_stripes > nr_devices)
1343 			num_stripes = nr_devices;
1344 
1345 		if (devices_info[i].max_avail >= min_stripe_size) {
1346 			int j;
1347 			u64 alloc_size;
1348 
1349 			avail_space += devices_info[i].max_avail * num_stripes;
1350 			alloc_size = devices_info[i].max_avail;
1351 			for (j = i + 1 - num_stripes; j <= i; j++)
1352 				devices_info[j].max_avail -= alloc_size;
1353 		}
1354 		i--;
1355 		nr_devices--;
1356 	}
1357 
1358 	kfree(devices_info);
1359 	*free_bytes = avail_space;
1360 	return 0;
1361 }
1362 
1363 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1364 {
1365 	struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1366 	struct btrfs_super_block *disk_super = fs_info->super_copy;
1367 	struct list_head *head = &fs_info->space_info;
1368 	struct btrfs_space_info *found;
1369 	u64 total_used = 0;
1370 	u64 total_free_data = 0;
1371 	int bits = dentry->d_sb->s_blocksize_bits;
1372 	__be32 *fsid = (__be32 *)fs_info->fsid;
1373 	int ret;
1374 
1375 	/* holding chunk_muext to avoid allocating new chunks */
1376 	mutex_lock(&fs_info->chunk_mutex);
1377 	rcu_read_lock();
1378 	list_for_each_entry_rcu(found, head, list) {
1379 		if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1380 			total_free_data += found->disk_total - found->disk_used;
1381 			total_free_data -=
1382 				btrfs_account_ro_block_groups_free_space(found);
1383 		}
1384 
1385 		total_used += found->disk_used;
1386 	}
1387 	rcu_read_unlock();
1388 
1389 	buf->f_namelen = BTRFS_NAME_LEN;
1390 	buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1391 	buf->f_bfree = buf->f_blocks - (total_used >> bits);
1392 	buf->f_bsize = dentry->d_sb->s_blocksize;
1393 	buf->f_type = BTRFS_SUPER_MAGIC;
1394 	buf->f_bavail = total_free_data;
1395 	ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1396 	if (ret) {
1397 		mutex_unlock(&fs_info->chunk_mutex);
1398 		return ret;
1399 	}
1400 	buf->f_bavail += total_free_data;
1401 	buf->f_bavail = buf->f_bavail >> bits;
1402 	mutex_unlock(&fs_info->chunk_mutex);
1403 
1404 	/* We treat it as constant endianness (it doesn't matter _which_)
1405 	   because we want the fsid to come out the same whether mounted
1406 	   on a big-endian or little-endian host */
1407 	buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1408 	buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1409 	/* Mask in the root object ID too, to disambiguate subvols */
1410 	buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1411 	buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1412 
1413 	return 0;
1414 }
1415 
1416 static void btrfs_kill_super(struct super_block *sb)
1417 {
1418 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1419 	kill_anon_super(sb);
1420 	free_fs_info(fs_info);
1421 }
1422 
1423 static struct file_system_type btrfs_fs_type = {
1424 	.owner		= THIS_MODULE,
1425 	.name		= "btrfs",
1426 	.mount		= btrfs_mount,
1427 	.kill_sb	= btrfs_kill_super,
1428 	.fs_flags	= FS_REQUIRES_DEV,
1429 };
1430 
1431 /*
1432  * used by btrfsctl to scan devices when no FS is mounted
1433  */
1434 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1435 				unsigned long arg)
1436 {
1437 	struct btrfs_ioctl_vol_args *vol;
1438 	struct btrfs_fs_devices *fs_devices;
1439 	int ret = -ENOTTY;
1440 
1441 	if (!capable(CAP_SYS_ADMIN))
1442 		return -EPERM;
1443 
1444 	vol = memdup_user((void __user *)arg, sizeof(*vol));
1445 	if (IS_ERR(vol))
1446 		return PTR_ERR(vol);
1447 
1448 	switch (cmd) {
1449 	case BTRFS_IOC_SCAN_DEV:
1450 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1451 					    &btrfs_fs_type, &fs_devices);
1452 		break;
1453 	}
1454 
1455 	kfree(vol);
1456 	return ret;
1457 }
1458 
1459 static int btrfs_freeze(struct super_block *sb)
1460 {
1461 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1462 	mutex_lock(&fs_info->transaction_kthread_mutex);
1463 	mutex_lock(&fs_info->cleaner_mutex);
1464 	return 0;
1465 }
1466 
1467 static int btrfs_unfreeze(struct super_block *sb)
1468 {
1469 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1470 	mutex_unlock(&fs_info->cleaner_mutex);
1471 	mutex_unlock(&fs_info->transaction_kthread_mutex);
1472 	return 0;
1473 }
1474 
1475 static void btrfs_fs_dirty_inode(struct inode *inode, int flags)
1476 {
1477 	int ret;
1478 
1479 	ret = btrfs_dirty_inode(inode);
1480 	if (ret)
1481 		printk_ratelimited(KERN_ERR "btrfs: fail to dirty inode %Lu "
1482 				   "error %d\n", btrfs_ino(inode), ret);
1483 }
1484 
1485 static const struct super_operations btrfs_super_ops = {
1486 	.drop_inode	= btrfs_drop_inode,
1487 	.evict_inode	= btrfs_evict_inode,
1488 	.put_super	= btrfs_put_super,
1489 	.sync_fs	= btrfs_sync_fs,
1490 	.show_options	= btrfs_show_options,
1491 	.write_inode	= btrfs_write_inode,
1492 	.dirty_inode	= btrfs_fs_dirty_inode,
1493 	.alloc_inode	= btrfs_alloc_inode,
1494 	.destroy_inode	= btrfs_destroy_inode,
1495 	.statfs		= btrfs_statfs,
1496 	.remount_fs	= btrfs_remount,
1497 	.freeze_fs	= btrfs_freeze,
1498 	.unfreeze_fs	= btrfs_unfreeze,
1499 };
1500 
1501 static const struct file_operations btrfs_ctl_fops = {
1502 	.unlocked_ioctl	 = btrfs_control_ioctl,
1503 	.compat_ioctl = btrfs_control_ioctl,
1504 	.owner	 = THIS_MODULE,
1505 	.llseek = noop_llseek,
1506 };
1507 
1508 static struct miscdevice btrfs_misc = {
1509 	.minor		= BTRFS_MINOR,
1510 	.name		= "btrfs-control",
1511 	.fops		= &btrfs_ctl_fops
1512 };
1513 
1514 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1515 MODULE_ALIAS("devname:btrfs-control");
1516 
1517 static int btrfs_interface_init(void)
1518 {
1519 	return misc_register(&btrfs_misc);
1520 }
1521 
1522 static void btrfs_interface_exit(void)
1523 {
1524 	if (misc_deregister(&btrfs_misc) < 0)
1525 		printk(KERN_INFO "misc_deregister failed for control device");
1526 }
1527 
1528 static int __init init_btrfs_fs(void)
1529 {
1530 	int err;
1531 
1532 	err = btrfs_init_sysfs();
1533 	if (err)
1534 		return err;
1535 
1536 	btrfs_init_compress();
1537 
1538 	err = btrfs_init_cachep();
1539 	if (err)
1540 		goto free_compress;
1541 
1542 	err = extent_io_init();
1543 	if (err)
1544 		goto free_cachep;
1545 
1546 	err = extent_map_init();
1547 	if (err)
1548 		goto free_extent_io;
1549 
1550 	err = btrfs_delayed_inode_init();
1551 	if (err)
1552 		goto free_extent_map;
1553 
1554 	err = btrfs_interface_init();
1555 	if (err)
1556 		goto free_delayed_inode;
1557 
1558 	err = register_filesystem(&btrfs_fs_type);
1559 	if (err)
1560 		goto unregister_ioctl;
1561 
1562 	btrfs_init_lockdep();
1563 
1564 	printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1565 	return 0;
1566 
1567 unregister_ioctl:
1568 	btrfs_interface_exit();
1569 free_delayed_inode:
1570 	btrfs_delayed_inode_exit();
1571 free_extent_map:
1572 	extent_map_exit();
1573 free_extent_io:
1574 	extent_io_exit();
1575 free_cachep:
1576 	btrfs_destroy_cachep();
1577 free_compress:
1578 	btrfs_exit_compress();
1579 	btrfs_exit_sysfs();
1580 	return err;
1581 }
1582 
1583 static void __exit exit_btrfs_fs(void)
1584 {
1585 	btrfs_destroy_cachep();
1586 	btrfs_delayed_inode_exit();
1587 	extent_map_exit();
1588 	extent_io_exit();
1589 	btrfs_interface_exit();
1590 	unregister_filesystem(&btrfs_fs_type);
1591 	btrfs_exit_sysfs();
1592 	btrfs_cleanup_fs_uuids();
1593 	btrfs_exit_compress();
1594 }
1595 
1596 module_init(init_btrfs_fs)
1597 module_exit(exit_btrfs_fs)
1598 
1599 MODULE_LICENSE("GPL");
1600