xref: /openbmc/linux/fs/btrfs/super.c (revision 565d76cb)
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 "compat.h"
43 #include "ctree.h"
44 #include "disk-io.h"
45 #include "transaction.h"
46 #include "btrfs_inode.h"
47 #include "ioctl.h"
48 #include "print-tree.h"
49 #include "xattr.h"
50 #include "volumes.h"
51 #include "version.h"
52 #include "export.h"
53 #include "compression.h"
54 
55 static const struct super_operations btrfs_super_ops;
56 
57 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
58 				      char nbuf[16])
59 {
60 	char *errstr = NULL;
61 
62 	switch (errno) {
63 	case -EIO:
64 		errstr = "IO failure";
65 		break;
66 	case -ENOMEM:
67 		errstr = "Out of memory";
68 		break;
69 	case -EROFS:
70 		errstr = "Readonly filesystem";
71 		break;
72 	default:
73 		if (nbuf) {
74 			if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
75 				errstr = nbuf;
76 		}
77 		break;
78 	}
79 
80 	return errstr;
81 }
82 
83 static void __save_error_info(struct btrfs_fs_info *fs_info)
84 {
85 	/*
86 	 * today we only save the error info into ram.  Long term we'll
87 	 * also send it down to the disk
88 	 */
89 	fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
90 }
91 
92 /* NOTE:
93  *	We move write_super stuff at umount in order to avoid deadlock
94  *	for umount hold all lock.
95  */
96 static void save_error_info(struct btrfs_fs_info *fs_info)
97 {
98 	__save_error_info(fs_info);
99 }
100 
101 /* btrfs handle error by forcing the filesystem readonly */
102 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
103 {
104 	struct super_block *sb = fs_info->sb;
105 
106 	if (sb->s_flags & MS_RDONLY)
107 		return;
108 
109 	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
110 		sb->s_flags |= MS_RDONLY;
111 		printk(KERN_INFO "btrfs is forced readonly\n");
112 	}
113 }
114 
115 /*
116  * __btrfs_std_error decodes expected errors from the caller and
117  * invokes the approciate error response.
118  */
119 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
120 		     unsigned int line, int errno)
121 {
122 	struct super_block *sb = fs_info->sb;
123 	char nbuf[16];
124 	const char *errstr;
125 
126 	/*
127 	 * Special case: if the error is EROFS, and we're already
128 	 * under MS_RDONLY, then it is safe here.
129 	 */
130 	if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
131 		return;
132 
133 	errstr = btrfs_decode_error(fs_info, errno, nbuf);
134 	printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
135 		sb->s_id, function, line, errstr);
136 	save_error_info(fs_info);
137 
138 	btrfs_handle_error(fs_info);
139 }
140 
141 static void btrfs_put_super(struct super_block *sb)
142 {
143 	struct btrfs_root *root = btrfs_sb(sb);
144 	int ret;
145 
146 	ret = close_ctree(root);
147 	sb->s_fs_info = NULL;
148 
149 	(void)ret; /* FIXME: need to fix VFS to return error? */
150 }
151 
152 enum {
153 	Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
154 	Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
155 	Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
156 	Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
157 	Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
158 	Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
159 	Opt_enospc_debug, Opt_err,
160 };
161 
162 static match_table_t tokens = {
163 	{Opt_degraded, "degraded"},
164 	{Opt_subvol, "subvol=%s"},
165 	{Opt_subvolid, "subvolid=%d"},
166 	{Opt_device, "device=%s"},
167 	{Opt_nodatasum, "nodatasum"},
168 	{Opt_nodatacow, "nodatacow"},
169 	{Opt_nobarrier, "nobarrier"},
170 	{Opt_max_inline, "max_inline=%s"},
171 	{Opt_alloc_start, "alloc_start=%s"},
172 	{Opt_thread_pool, "thread_pool=%d"},
173 	{Opt_compress, "compress"},
174 	{Opt_compress_type, "compress=%s"},
175 	{Opt_compress_force, "compress-force"},
176 	{Opt_compress_force_type, "compress-force=%s"},
177 	{Opt_ssd, "ssd"},
178 	{Opt_ssd_spread, "ssd_spread"},
179 	{Opt_nossd, "nossd"},
180 	{Opt_noacl, "noacl"},
181 	{Opt_notreelog, "notreelog"},
182 	{Opt_flushoncommit, "flushoncommit"},
183 	{Opt_ratio, "metadata_ratio=%d"},
184 	{Opt_discard, "discard"},
185 	{Opt_space_cache, "space_cache"},
186 	{Opt_clear_cache, "clear_cache"},
187 	{Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
188 	{Opt_enospc_debug, "enospc_debug"},
189 	{Opt_err, NULL},
190 };
191 
192 /*
193  * Regular mount options parser.  Everything that is needed only when
194  * reading in a new superblock is parsed here.
195  */
196 int btrfs_parse_options(struct btrfs_root *root, char *options)
197 {
198 	struct btrfs_fs_info *info = root->fs_info;
199 	substring_t args[MAX_OPT_ARGS];
200 	char *p, *num, *orig;
201 	int intarg;
202 	int ret = 0;
203 	char *compress_type;
204 	bool compress_force = false;
205 
206 	if (!options)
207 		return 0;
208 
209 	/*
210 	 * strsep changes the string, duplicate it because parse_options
211 	 * gets called twice
212 	 */
213 	options = kstrdup(options, GFP_NOFS);
214 	if (!options)
215 		return -ENOMEM;
216 
217 	orig = options;
218 
219 	while ((p = strsep(&options, ",")) != NULL) {
220 		int token;
221 		if (!*p)
222 			continue;
223 
224 		token = match_token(p, tokens, args);
225 		switch (token) {
226 		case Opt_degraded:
227 			printk(KERN_INFO "btrfs: allowing degraded mounts\n");
228 			btrfs_set_opt(info->mount_opt, DEGRADED);
229 			break;
230 		case Opt_subvol:
231 		case Opt_subvolid:
232 		case Opt_device:
233 			/*
234 			 * These are parsed by btrfs_parse_early_options
235 			 * and can be happily ignored here.
236 			 */
237 			break;
238 		case Opt_nodatasum:
239 			printk(KERN_INFO "btrfs: setting nodatasum\n");
240 			btrfs_set_opt(info->mount_opt, NODATASUM);
241 			break;
242 		case Opt_nodatacow:
243 			printk(KERN_INFO "btrfs: setting nodatacow\n");
244 			btrfs_set_opt(info->mount_opt, NODATACOW);
245 			btrfs_set_opt(info->mount_opt, NODATASUM);
246 			break;
247 		case Opt_compress_force:
248 		case Opt_compress_force_type:
249 			compress_force = true;
250 		case Opt_compress:
251 		case Opt_compress_type:
252 			if (token == Opt_compress ||
253 			    token == Opt_compress_force ||
254 			    strcmp(args[0].from, "zlib") == 0) {
255 				compress_type = "zlib";
256 				info->compress_type = BTRFS_COMPRESS_ZLIB;
257 			} else if (strcmp(args[0].from, "lzo") == 0) {
258 				compress_type = "lzo";
259 				info->compress_type = BTRFS_COMPRESS_LZO;
260 			} else {
261 				ret = -EINVAL;
262 				goto out;
263 			}
264 
265 			btrfs_set_opt(info->mount_opt, COMPRESS);
266 			if (compress_force) {
267 				btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
268 				pr_info("btrfs: force %s compression\n",
269 					compress_type);
270 			} else
271 				pr_info("btrfs: use %s compression\n",
272 					compress_type);
273 			break;
274 		case Opt_ssd:
275 			printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
276 			btrfs_set_opt(info->mount_opt, SSD);
277 			break;
278 		case Opt_ssd_spread:
279 			printk(KERN_INFO "btrfs: use spread ssd "
280 			       "allocation scheme\n");
281 			btrfs_set_opt(info->mount_opt, SSD);
282 			btrfs_set_opt(info->mount_opt, SSD_SPREAD);
283 			break;
284 		case Opt_nossd:
285 			printk(KERN_INFO "btrfs: not using ssd allocation "
286 			       "scheme\n");
287 			btrfs_set_opt(info->mount_opt, NOSSD);
288 			btrfs_clear_opt(info->mount_opt, SSD);
289 			btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
290 			break;
291 		case Opt_nobarrier:
292 			printk(KERN_INFO "btrfs: turning off barriers\n");
293 			btrfs_set_opt(info->mount_opt, NOBARRIER);
294 			break;
295 		case Opt_thread_pool:
296 			intarg = 0;
297 			match_int(&args[0], &intarg);
298 			if (intarg) {
299 				info->thread_pool_size = intarg;
300 				printk(KERN_INFO "btrfs: thread pool %d\n",
301 				       info->thread_pool_size);
302 			}
303 			break;
304 		case Opt_max_inline:
305 			num = match_strdup(&args[0]);
306 			if (num) {
307 				info->max_inline = memparse(num, NULL);
308 				kfree(num);
309 
310 				if (info->max_inline) {
311 					info->max_inline = max_t(u64,
312 						info->max_inline,
313 						root->sectorsize);
314 				}
315 				printk(KERN_INFO "btrfs: max_inline at %llu\n",
316 					(unsigned long long)info->max_inline);
317 			}
318 			break;
319 		case Opt_alloc_start:
320 			num = match_strdup(&args[0]);
321 			if (num) {
322 				info->alloc_start = memparse(num, NULL);
323 				kfree(num);
324 				printk(KERN_INFO
325 					"btrfs: allocations start at %llu\n",
326 					(unsigned long long)info->alloc_start);
327 			}
328 			break;
329 		case Opt_noacl:
330 			root->fs_info->sb->s_flags &= ~MS_POSIXACL;
331 			break;
332 		case Opt_notreelog:
333 			printk(KERN_INFO "btrfs: disabling tree log\n");
334 			btrfs_set_opt(info->mount_opt, NOTREELOG);
335 			break;
336 		case Opt_flushoncommit:
337 			printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
338 			btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
339 			break;
340 		case Opt_ratio:
341 			intarg = 0;
342 			match_int(&args[0], &intarg);
343 			if (intarg) {
344 				info->metadata_ratio = intarg;
345 				printk(KERN_INFO "btrfs: metadata ratio %d\n",
346 				       info->metadata_ratio);
347 			}
348 			break;
349 		case Opt_discard:
350 			btrfs_set_opt(info->mount_opt, DISCARD);
351 			break;
352 		case Opt_space_cache:
353 			printk(KERN_INFO "btrfs: enabling disk space caching\n");
354 			btrfs_set_opt(info->mount_opt, SPACE_CACHE);
355 			break;
356 		case Opt_clear_cache:
357 			printk(KERN_INFO "btrfs: force clearing of disk cache\n");
358 			btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
359 			break;
360 		case Opt_user_subvol_rm_allowed:
361 			btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
362 			break;
363 		case Opt_enospc_debug:
364 			btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
365 			break;
366 		case Opt_err:
367 			printk(KERN_INFO "btrfs: unrecognized mount option "
368 			       "'%s'\n", p);
369 			ret = -EINVAL;
370 			goto out;
371 		default:
372 			break;
373 		}
374 	}
375 out:
376 	kfree(orig);
377 	return ret;
378 }
379 
380 /*
381  * Parse mount options that are required early in the mount process.
382  *
383  * All other options will be parsed on much later in the mount process and
384  * only when we need to allocate a new super block.
385  */
386 static int btrfs_parse_early_options(const char *options, fmode_t flags,
387 		void *holder, char **subvol_name, u64 *subvol_objectid,
388 		struct btrfs_fs_devices **fs_devices)
389 {
390 	substring_t args[MAX_OPT_ARGS];
391 	char *opts, *orig, *p;
392 	int error = 0;
393 	int intarg;
394 
395 	if (!options)
396 		goto out;
397 
398 	/*
399 	 * strsep changes the string, duplicate it because parse_options
400 	 * gets called twice
401 	 */
402 	opts = kstrdup(options, GFP_KERNEL);
403 	if (!opts)
404 		return -ENOMEM;
405 	orig = opts;
406 
407 	while ((p = strsep(&opts, ",")) != NULL) {
408 		int token;
409 		if (!*p)
410 			continue;
411 
412 		token = match_token(p, tokens, args);
413 		switch (token) {
414 		case Opt_subvol:
415 			*subvol_name = match_strdup(&args[0]);
416 			break;
417 		case Opt_subvolid:
418 			intarg = 0;
419 			error = match_int(&args[0], &intarg);
420 			if (!error) {
421 				/* we want the original fs_tree */
422 				if (!intarg)
423 					*subvol_objectid =
424 						BTRFS_FS_TREE_OBJECTID;
425 				else
426 					*subvol_objectid = intarg;
427 			}
428 			break;
429 		case Opt_device:
430 			error = btrfs_scan_one_device(match_strdup(&args[0]),
431 					flags, holder, fs_devices);
432 			if (error)
433 				goto out_free_opts;
434 			break;
435 		default:
436 			break;
437 		}
438 	}
439 
440  out_free_opts:
441 	kfree(orig);
442  out:
443 	/*
444 	 * If no subvolume name is specified we use the default one.  Allocate
445 	 * a copy of the string "." here so that code later in the
446 	 * mount path doesn't care if it's the default volume or another one.
447 	 */
448 	if (!*subvol_name) {
449 		*subvol_name = kstrdup(".", GFP_KERNEL);
450 		if (!*subvol_name)
451 			return -ENOMEM;
452 	}
453 	return error;
454 }
455 
456 static struct dentry *get_default_root(struct super_block *sb,
457 				       u64 subvol_objectid)
458 {
459 	struct btrfs_root *root = sb->s_fs_info;
460 	struct btrfs_root *new_root;
461 	struct btrfs_dir_item *di;
462 	struct btrfs_path *path;
463 	struct btrfs_key location;
464 	struct inode *inode;
465 	struct dentry *dentry;
466 	u64 dir_id;
467 	int new = 0;
468 
469 	/*
470 	 * We have a specific subvol we want to mount, just setup location and
471 	 * go look up the root.
472 	 */
473 	if (subvol_objectid) {
474 		location.objectid = subvol_objectid;
475 		location.type = BTRFS_ROOT_ITEM_KEY;
476 		location.offset = (u64)-1;
477 		goto find_root;
478 	}
479 
480 	path = btrfs_alloc_path();
481 	if (!path)
482 		return ERR_PTR(-ENOMEM);
483 	path->leave_spinning = 1;
484 
485 	/*
486 	 * Find the "default" dir item which points to the root item that we
487 	 * will mount by default if we haven't been given a specific subvolume
488 	 * to mount.
489 	 */
490 	dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
491 	di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
492 	if (IS_ERR(di))
493 		return ERR_CAST(di);
494 	if (!di) {
495 		/*
496 		 * Ok the default dir item isn't there.  This is weird since
497 		 * it's always been there, but don't freak out, just try and
498 		 * mount to root most subvolume.
499 		 */
500 		btrfs_free_path(path);
501 		dir_id = BTRFS_FIRST_FREE_OBJECTID;
502 		new_root = root->fs_info->fs_root;
503 		goto setup_root;
504 	}
505 
506 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
507 	btrfs_free_path(path);
508 
509 find_root:
510 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
511 	if (IS_ERR(new_root))
512 		return ERR_CAST(new_root);
513 
514 	if (btrfs_root_refs(&new_root->root_item) == 0)
515 		return ERR_PTR(-ENOENT);
516 
517 	dir_id = btrfs_root_dirid(&new_root->root_item);
518 setup_root:
519 	location.objectid = dir_id;
520 	location.type = BTRFS_INODE_ITEM_KEY;
521 	location.offset = 0;
522 
523 	inode = btrfs_iget(sb, &location, new_root, &new);
524 	if (IS_ERR(inode))
525 		return ERR_CAST(inode);
526 
527 	/*
528 	 * If we're just mounting the root most subvol put the inode and return
529 	 * a reference to the dentry.  We will have already gotten a reference
530 	 * to the inode in btrfs_fill_super so we're good to go.
531 	 */
532 	if (!new && sb->s_root->d_inode == inode) {
533 		iput(inode);
534 		return dget(sb->s_root);
535 	}
536 
537 	if (new) {
538 		const struct qstr name = { .name = "/", .len = 1 };
539 
540 		/*
541 		 * New inode, we need to make the dentry a sibling of s_root so
542 		 * everything gets cleaned up properly on unmount.
543 		 */
544 		dentry = d_alloc(sb->s_root, &name);
545 		if (!dentry) {
546 			iput(inode);
547 			return ERR_PTR(-ENOMEM);
548 		}
549 		d_splice_alias(inode, dentry);
550 	} else {
551 		/*
552 		 * We found the inode in cache, just find a dentry for it and
553 		 * put the reference to the inode we just got.
554 		 */
555 		dentry = d_find_alias(inode);
556 		iput(inode);
557 	}
558 
559 	return dentry;
560 }
561 
562 static int btrfs_fill_super(struct super_block *sb,
563 			    struct btrfs_fs_devices *fs_devices,
564 			    void *data, int silent)
565 {
566 	struct inode *inode;
567 	struct dentry *root_dentry;
568 	struct btrfs_root *tree_root;
569 	struct btrfs_key key;
570 	int err;
571 
572 	sb->s_maxbytes = MAX_LFS_FILESIZE;
573 	sb->s_magic = BTRFS_SUPER_MAGIC;
574 	sb->s_op = &btrfs_super_ops;
575 	sb->s_d_op = &btrfs_dentry_operations;
576 	sb->s_export_op = &btrfs_export_ops;
577 	sb->s_xattr = btrfs_xattr_handlers;
578 	sb->s_time_gran = 1;
579 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
580 	sb->s_flags |= MS_POSIXACL;
581 #endif
582 
583 	tree_root = open_ctree(sb, fs_devices, (char *)data);
584 
585 	if (IS_ERR(tree_root)) {
586 		printk("btrfs: open_ctree failed\n");
587 		return PTR_ERR(tree_root);
588 	}
589 	sb->s_fs_info = tree_root;
590 
591 	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
592 	key.type = BTRFS_INODE_ITEM_KEY;
593 	key.offset = 0;
594 	inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
595 	if (IS_ERR(inode)) {
596 		err = PTR_ERR(inode);
597 		goto fail_close;
598 	}
599 
600 	root_dentry = d_alloc_root(inode);
601 	if (!root_dentry) {
602 		iput(inode);
603 		err = -ENOMEM;
604 		goto fail_close;
605 	}
606 
607 	sb->s_root = root_dentry;
608 
609 	save_mount_options(sb, data);
610 	return 0;
611 
612 fail_close:
613 	close_ctree(tree_root);
614 	return err;
615 }
616 
617 int btrfs_sync_fs(struct super_block *sb, int wait)
618 {
619 	struct btrfs_trans_handle *trans;
620 	struct btrfs_root *root = btrfs_sb(sb);
621 	int ret;
622 
623 	if (!wait) {
624 		filemap_flush(root->fs_info->btree_inode->i_mapping);
625 		return 0;
626 	}
627 
628 	btrfs_start_delalloc_inodes(root, 0);
629 	btrfs_wait_ordered_extents(root, 0, 0);
630 
631 	trans = btrfs_start_transaction(root, 0);
632 	if (IS_ERR(trans))
633 		return PTR_ERR(trans);
634 	ret = btrfs_commit_transaction(trans, root);
635 	return ret;
636 }
637 
638 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
639 {
640 	struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
641 	struct btrfs_fs_info *info = root->fs_info;
642 
643 	if (btrfs_test_opt(root, DEGRADED))
644 		seq_puts(seq, ",degraded");
645 	if (btrfs_test_opt(root, NODATASUM))
646 		seq_puts(seq, ",nodatasum");
647 	if (btrfs_test_opt(root, NODATACOW))
648 		seq_puts(seq, ",nodatacow");
649 	if (btrfs_test_opt(root, NOBARRIER))
650 		seq_puts(seq, ",nobarrier");
651 	if (info->max_inline != 8192 * 1024)
652 		seq_printf(seq, ",max_inline=%llu",
653 			   (unsigned long long)info->max_inline);
654 	if (info->alloc_start != 0)
655 		seq_printf(seq, ",alloc_start=%llu",
656 			   (unsigned long long)info->alloc_start);
657 	if (info->thread_pool_size !=  min_t(unsigned long,
658 					     num_online_cpus() + 2, 8))
659 		seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
660 	if (btrfs_test_opt(root, COMPRESS))
661 		seq_puts(seq, ",compress");
662 	if (btrfs_test_opt(root, NOSSD))
663 		seq_puts(seq, ",nossd");
664 	if (btrfs_test_opt(root, SSD_SPREAD))
665 		seq_puts(seq, ",ssd_spread");
666 	else if (btrfs_test_opt(root, SSD))
667 		seq_puts(seq, ",ssd");
668 	if (btrfs_test_opt(root, NOTREELOG))
669 		seq_puts(seq, ",notreelog");
670 	if (btrfs_test_opt(root, FLUSHONCOMMIT))
671 		seq_puts(seq, ",flushoncommit");
672 	if (btrfs_test_opt(root, DISCARD))
673 		seq_puts(seq, ",discard");
674 	if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
675 		seq_puts(seq, ",noacl");
676 	return 0;
677 }
678 
679 static int btrfs_test_super(struct super_block *s, void *data)
680 {
681 	struct btrfs_root *test_root = data;
682 	struct btrfs_root *root = btrfs_sb(s);
683 
684 	/*
685 	 * If this super block is going away, return false as it
686 	 * can't match as an existing super block.
687 	 */
688 	if (!atomic_read(&s->s_active))
689 		return 0;
690 	return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
691 }
692 
693 static int btrfs_set_super(struct super_block *s, void *data)
694 {
695 	s->s_fs_info = data;
696 
697 	return set_anon_super(s, data);
698 }
699 
700 
701 /*
702  * Find a superblock for the given device / mount point.
703  *
704  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
705  *	  for multiple device setup.  Make sure to keep it in sync.
706  */
707 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
708 		const char *dev_name, void *data)
709 {
710 	struct block_device *bdev = NULL;
711 	struct super_block *s;
712 	struct dentry *root;
713 	struct btrfs_fs_devices *fs_devices = NULL;
714 	struct btrfs_root *tree_root = NULL;
715 	struct btrfs_fs_info *fs_info = NULL;
716 	fmode_t mode = FMODE_READ;
717 	char *subvol_name = NULL;
718 	u64 subvol_objectid = 0;
719 	int error = 0;
720 
721 	if (!(flags & MS_RDONLY))
722 		mode |= FMODE_WRITE;
723 
724 	error = btrfs_parse_early_options(data, mode, fs_type,
725 					  &subvol_name, &subvol_objectid,
726 					  &fs_devices);
727 	if (error)
728 		return ERR_PTR(error);
729 
730 	error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
731 	if (error)
732 		goto error_free_subvol_name;
733 
734 	error = btrfs_open_devices(fs_devices, mode, fs_type);
735 	if (error)
736 		goto error_free_subvol_name;
737 
738 	if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
739 		error = -EACCES;
740 		goto error_close_devices;
741 	}
742 
743 	/*
744 	 * Setup a dummy root and fs_info for test/set super.  This is because
745 	 * we don't actually fill this stuff out until open_ctree, but we need
746 	 * it for searching for existing supers, so this lets us do that and
747 	 * then open_ctree will properly initialize everything later.
748 	 */
749 	fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
750 	tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
751 	if (!fs_info || !tree_root) {
752 		error = -ENOMEM;
753 		goto error_close_devices;
754 	}
755 	fs_info->tree_root = tree_root;
756 	fs_info->fs_devices = fs_devices;
757 	tree_root->fs_info = fs_info;
758 
759 	bdev = fs_devices->latest_bdev;
760 	s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
761 	if (IS_ERR(s))
762 		goto error_s;
763 
764 	if (s->s_root) {
765 		if ((flags ^ s->s_flags) & MS_RDONLY) {
766 			deactivate_locked_super(s);
767 			error = -EBUSY;
768 			goto error_close_devices;
769 		}
770 
771 		btrfs_close_devices(fs_devices);
772 		kfree(fs_info);
773 		kfree(tree_root);
774 	} else {
775 		char b[BDEVNAME_SIZE];
776 
777 		s->s_flags = flags;
778 		strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
779 		error = btrfs_fill_super(s, fs_devices, data,
780 					 flags & MS_SILENT ? 1 : 0);
781 		if (error) {
782 			deactivate_locked_super(s);
783 			goto error_free_subvol_name;
784 		}
785 
786 		btrfs_sb(s)->fs_info->bdev_holder = fs_type;
787 		s->s_flags |= MS_ACTIVE;
788 	}
789 
790 	root = get_default_root(s, subvol_objectid);
791 	if (IS_ERR(root)) {
792 		error = PTR_ERR(root);
793 		deactivate_locked_super(s);
794 		goto error_free_subvol_name;
795 	}
796 	/* if they gave us a subvolume name bind mount into that */
797 	if (strcmp(subvol_name, ".")) {
798 		struct dentry *new_root;
799 		mutex_lock(&root->d_inode->i_mutex);
800 		new_root = lookup_one_len(subvol_name, root,
801 				      strlen(subvol_name));
802 		mutex_unlock(&root->d_inode->i_mutex);
803 
804 		if (IS_ERR(new_root)) {
805 			dput(root);
806 			deactivate_locked_super(s);
807 			error = PTR_ERR(new_root);
808 			goto error_free_subvol_name;
809 		}
810 		if (!new_root->d_inode) {
811 			dput(root);
812 			dput(new_root);
813 			deactivate_locked_super(s);
814 			error = -ENXIO;
815 			goto error_free_subvol_name;
816 		}
817 		dput(root);
818 		root = new_root;
819 	}
820 
821 	kfree(subvol_name);
822 	return root;
823 
824 error_s:
825 	error = PTR_ERR(s);
826 error_close_devices:
827 	btrfs_close_devices(fs_devices);
828 	kfree(fs_info);
829 	kfree(tree_root);
830 error_free_subvol_name:
831 	kfree(subvol_name);
832 	return ERR_PTR(error);
833 }
834 
835 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
836 {
837 	struct btrfs_root *root = btrfs_sb(sb);
838 	int ret;
839 
840 	ret = btrfs_parse_options(root, data);
841 	if (ret)
842 		return -EINVAL;
843 
844 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
845 		return 0;
846 
847 	if (*flags & MS_RDONLY) {
848 		sb->s_flags |= MS_RDONLY;
849 
850 		ret =  btrfs_commit_super(root);
851 		WARN_ON(ret);
852 	} else {
853 		if (root->fs_info->fs_devices->rw_devices == 0)
854 			return -EACCES;
855 
856 		if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
857 			return -EINVAL;
858 
859 		ret = btrfs_cleanup_fs_roots(root->fs_info);
860 		WARN_ON(ret);
861 
862 		/* recover relocation */
863 		ret = btrfs_recover_relocation(root);
864 		WARN_ON(ret);
865 
866 		sb->s_flags &= ~MS_RDONLY;
867 	}
868 
869 	return 0;
870 }
871 
872 /*
873  * The helper to calc the free space on the devices that can be used to store
874  * file data.
875  */
876 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
877 {
878 	struct btrfs_fs_info *fs_info = root->fs_info;
879 	struct btrfs_device_info *devices_info;
880 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
881 	struct btrfs_device *device;
882 	u64 skip_space;
883 	u64 type;
884 	u64 avail_space;
885 	u64 used_space;
886 	u64 min_stripe_size;
887 	int min_stripes = 1;
888 	int i = 0, nr_devices;
889 	int ret;
890 
891 	nr_devices = fs_info->fs_devices->rw_devices;
892 	BUG_ON(!nr_devices);
893 
894 	devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
895 			       GFP_NOFS);
896 	if (!devices_info)
897 		return -ENOMEM;
898 
899 	/* calc min stripe number for data space alloction */
900 	type = btrfs_get_alloc_profile(root, 1);
901 	if (type & BTRFS_BLOCK_GROUP_RAID0)
902 		min_stripes = 2;
903 	else if (type & BTRFS_BLOCK_GROUP_RAID1)
904 		min_stripes = 2;
905 	else if (type & BTRFS_BLOCK_GROUP_RAID10)
906 		min_stripes = 4;
907 
908 	if (type & BTRFS_BLOCK_GROUP_DUP)
909 		min_stripe_size = 2 * BTRFS_STRIPE_LEN;
910 	else
911 		min_stripe_size = BTRFS_STRIPE_LEN;
912 
913 	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
914 		if (!device->in_fs_metadata)
915 			continue;
916 
917 		avail_space = device->total_bytes - device->bytes_used;
918 
919 		/* align with stripe_len */
920 		do_div(avail_space, BTRFS_STRIPE_LEN);
921 		avail_space *= BTRFS_STRIPE_LEN;
922 
923 		/*
924 		 * In order to avoid overwritting the superblock on the drive,
925 		 * btrfs starts at an offset of at least 1MB when doing chunk
926 		 * allocation.
927 		 */
928 		skip_space = 1024 * 1024;
929 
930 		/* user can set the offset in fs_info->alloc_start. */
931 		if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
932 		    device->total_bytes)
933 			skip_space = max(fs_info->alloc_start, skip_space);
934 
935 		/*
936 		 * btrfs can not use the free space in [0, skip_space - 1],
937 		 * we must subtract it from the total. In order to implement
938 		 * it, we account the used space in this range first.
939 		 */
940 		ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
941 						     &used_space);
942 		if (ret) {
943 			kfree(devices_info);
944 			return ret;
945 		}
946 
947 		/* calc the free space in [0, skip_space - 1] */
948 		skip_space -= used_space;
949 
950 		/*
951 		 * we can use the free space in [0, skip_space - 1], subtract
952 		 * it from the total.
953 		 */
954 		if (avail_space && avail_space >= skip_space)
955 			avail_space -= skip_space;
956 		else
957 			avail_space = 0;
958 
959 		if (avail_space < min_stripe_size)
960 			continue;
961 
962 		devices_info[i].dev = device;
963 		devices_info[i].max_avail = avail_space;
964 
965 		i++;
966 	}
967 
968 	nr_devices = i;
969 
970 	btrfs_descending_sort_devices(devices_info, nr_devices);
971 
972 	i = nr_devices - 1;
973 	avail_space = 0;
974 	while (nr_devices >= min_stripes) {
975 		if (devices_info[i].max_avail >= min_stripe_size) {
976 			int j;
977 			u64 alloc_size;
978 
979 			avail_space += devices_info[i].max_avail * min_stripes;
980 			alloc_size = devices_info[i].max_avail;
981 			for (j = i + 1 - min_stripes; j <= i; j++)
982 				devices_info[j].max_avail -= alloc_size;
983 		}
984 		i--;
985 		nr_devices--;
986 	}
987 
988 	kfree(devices_info);
989 	*free_bytes = avail_space;
990 	return 0;
991 }
992 
993 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
994 {
995 	struct btrfs_root *root = btrfs_sb(dentry->d_sb);
996 	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
997 	struct list_head *head = &root->fs_info->space_info;
998 	struct btrfs_space_info *found;
999 	u64 total_used = 0;
1000 	u64 total_free_data = 0;
1001 	int bits = dentry->d_sb->s_blocksize_bits;
1002 	__be32 *fsid = (__be32 *)root->fs_info->fsid;
1003 	int ret;
1004 
1005 	/* holding chunk_muext to avoid allocating new chunks */
1006 	mutex_lock(&root->fs_info->chunk_mutex);
1007 	rcu_read_lock();
1008 	list_for_each_entry_rcu(found, head, list) {
1009 		if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1010 			total_free_data += found->disk_total - found->disk_used;
1011 			total_free_data -=
1012 				btrfs_account_ro_block_groups_free_space(found);
1013 		}
1014 
1015 		total_used += found->disk_used;
1016 	}
1017 	rcu_read_unlock();
1018 
1019 	buf->f_namelen = BTRFS_NAME_LEN;
1020 	buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1021 	buf->f_bfree = buf->f_blocks - (total_used >> bits);
1022 	buf->f_bsize = dentry->d_sb->s_blocksize;
1023 	buf->f_type = BTRFS_SUPER_MAGIC;
1024 	buf->f_bavail = total_free_data;
1025 	ret = btrfs_calc_avail_data_space(root, &total_free_data);
1026 	if (ret) {
1027 		mutex_unlock(&root->fs_info->chunk_mutex);
1028 		return ret;
1029 	}
1030 	buf->f_bavail += total_free_data;
1031 	buf->f_bavail = buf->f_bavail >> bits;
1032 	mutex_unlock(&root->fs_info->chunk_mutex);
1033 
1034 	/* We treat it as constant endianness (it doesn't matter _which_)
1035 	   because we want the fsid to come out the same whether mounted
1036 	   on a big-endian or little-endian host */
1037 	buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1038 	buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1039 	/* Mask in the root object ID too, to disambiguate subvols */
1040 	buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1041 	buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1042 
1043 	return 0;
1044 }
1045 
1046 static struct file_system_type btrfs_fs_type = {
1047 	.owner		= THIS_MODULE,
1048 	.name		= "btrfs",
1049 	.mount		= btrfs_mount,
1050 	.kill_sb	= kill_anon_super,
1051 	.fs_flags	= FS_REQUIRES_DEV,
1052 };
1053 
1054 /*
1055  * used by btrfsctl to scan devices when no FS is mounted
1056  */
1057 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1058 				unsigned long arg)
1059 {
1060 	struct btrfs_ioctl_vol_args *vol;
1061 	struct btrfs_fs_devices *fs_devices;
1062 	int ret = -ENOTTY;
1063 
1064 	if (!capable(CAP_SYS_ADMIN))
1065 		return -EPERM;
1066 
1067 	vol = memdup_user((void __user *)arg, sizeof(*vol));
1068 	if (IS_ERR(vol))
1069 		return PTR_ERR(vol);
1070 
1071 	switch (cmd) {
1072 	case BTRFS_IOC_SCAN_DEV:
1073 		ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1074 					    &btrfs_fs_type, &fs_devices);
1075 		break;
1076 	}
1077 
1078 	kfree(vol);
1079 	return ret;
1080 }
1081 
1082 static int btrfs_freeze(struct super_block *sb)
1083 {
1084 	struct btrfs_root *root = btrfs_sb(sb);
1085 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
1086 	mutex_lock(&root->fs_info->cleaner_mutex);
1087 	return 0;
1088 }
1089 
1090 static int btrfs_unfreeze(struct super_block *sb)
1091 {
1092 	struct btrfs_root *root = btrfs_sb(sb);
1093 	mutex_unlock(&root->fs_info->cleaner_mutex);
1094 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1095 	return 0;
1096 }
1097 
1098 static const struct super_operations btrfs_super_ops = {
1099 	.drop_inode	= btrfs_drop_inode,
1100 	.evict_inode	= btrfs_evict_inode,
1101 	.put_super	= btrfs_put_super,
1102 	.sync_fs	= btrfs_sync_fs,
1103 	.show_options	= btrfs_show_options,
1104 	.write_inode	= btrfs_write_inode,
1105 	.dirty_inode	= btrfs_dirty_inode,
1106 	.alloc_inode	= btrfs_alloc_inode,
1107 	.destroy_inode	= btrfs_destroy_inode,
1108 	.statfs		= btrfs_statfs,
1109 	.remount_fs	= btrfs_remount,
1110 	.freeze_fs	= btrfs_freeze,
1111 	.unfreeze_fs	= btrfs_unfreeze,
1112 };
1113 
1114 static const struct file_operations btrfs_ctl_fops = {
1115 	.unlocked_ioctl	 = btrfs_control_ioctl,
1116 	.compat_ioctl = btrfs_control_ioctl,
1117 	.owner	 = THIS_MODULE,
1118 	.llseek = noop_llseek,
1119 };
1120 
1121 static struct miscdevice btrfs_misc = {
1122 	.minor		= BTRFS_MINOR,
1123 	.name		= "btrfs-control",
1124 	.fops		= &btrfs_ctl_fops
1125 };
1126 
1127 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1128 MODULE_ALIAS("devname:btrfs-control");
1129 
1130 static int btrfs_interface_init(void)
1131 {
1132 	return misc_register(&btrfs_misc);
1133 }
1134 
1135 static void btrfs_interface_exit(void)
1136 {
1137 	if (misc_deregister(&btrfs_misc) < 0)
1138 		printk(KERN_INFO "misc_deregister failed for control device");
1139 }
1140 
1141 static int __init init_btrfs_fs(void)
1142 {
1143 	int err;
1144 
1145 	err = btrfs_init_sysfs();
1146 	if (err)
1147 		return err;
1148 
1149 	err = btrfs_init_compress();
1150 	if (err)
1151 		goto free_sysfs;
1152 
1153 	err = btrfs_init_cachep();
1154 	if (err)
1155 		goto free_compress;
1156 
1157 	err = extent_io_init();
1158 	if (err)
1159 		goto free_cachep;
1160 
1161 	err = extent_map_init();
1162 	if (err)
1163 		goto free_extent_io;
1164 
1165 	err = btrfs_interface_init();
1166 	if (err)
1167 		goto free_extent_map;
1168 
1169 	err = register_filesystem(&btrfs_fs_type);
1170 	if (err)
1171 		goto unregister_ioctl;
1172 
1173 	printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1174 	return 0;
1175 
1176 unregister_ioctl:
1177 	btrfs_interface_exit();
1178 free_extent_map:
1179 	extent_map_exit();
1180 free_extent_io:
1181 	extent_io_exit();
1182 free_cachep:
1183 	btrfs_destroy_cachep();
1184 free_compress:
1185 	btrfs_exit_compress();
1186 free_sysfs:
1187 	btrfs_exit_sysfs();
1188 	return err;
1189 }
1190 
1191 static void __exit exit_btrfs_fs(void)
1192 {
1193 	btrfs_destroy_cachep();
1194 	extent_map_exit();
1195 	extent_io_exit();
1196 	btrfs_interface_exit();
1197 	unregister_filesystem(&btrfs_fs_type);
1198 	btrfs_exit_sysfs();
1199 	btrfs_cleanup_fs_uuids();
1200 	btrfs_exit_compress();
1201 }
1202 
1203 module_init(init_btrfs_fs)
1204 module_exit(exit_btrfs_fs)
1205 
1206 MODULE_LICENSE("GPL");
1207