Merge branch 'for-linus-4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/mason...
[sfrench/cifs-2.6.git] / fs / btrfs / super.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/buffer_head.h>
22 #include <linux/fs.h>
23 #include <linux/pagemap.h>
24 #include <linux/highmem.h>
25 #include <linux/time.h>
26 #include <linux/init.h>
27 #include <linux/seq_file.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mount.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/parser.h>
37 #include <linux/ctype.h>
38 #include <linux/namei.h>
39 #include <linux/miscdevice.h>
40 #include <linux/magic.h>
41 #include <linux/slab.h>
42 #include <linux/cleancache.h>
43 #include <linux/ratelimit.h>
44 #include <linux/btrfs.h>
45 #include "delayed-inode.h"
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "hash.h"
52 #include "props.h"
53 #include "xattr.h"
54 #include "volumes.h"
55 #include "export.h"
56 #include "compression.h"
57 #include "rcu-string.h"
58 #include "dev-replace.h"
59 #include "free-space-cache.h"
60 #include "backref.h"
61 #include "tests/btrfs-tests.h"
62
63 #include "qgroup.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/btrfs.h>
66
67 static const struct super_operations btrfs_super_ops;
68 static struct file_system_type btrfs_fs_type;
69
70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
71
72 const char *btrfs_decode_error(int errno)
73 {
74         char *errstr = "unknown";
75
76         switch (errno) {
77         case -EIO:
78                 errstr = "IO failure";
79                 break;
80         case -ENOMEM:
81                 errstr = "Out of memory";
82                 break;
83         case -EROFS:
84                 errstr = "Readonly filesystem";
85                 break;
86         case -EEXIST:
87                 errstr = "Object already exists";
88                 break;
89         case -ENOSPC:
90                 errstr = "No space left";
91                 break;
92         case -ENOENT:
93                 errstr = "No such entry";
94                 break;
95         }
96
97         return errstr;
98 }
99
100 /* btrfs handle error by forcing the filesystem readonly */
101 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
102 {
103         struct super_block *sb = fs_info->sb;
104
105         if (sb->s_flags & MS_RDONLY)
106                 return;
107
108         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
109                 sb->s_flags |= MS_RDONLY;
110                 btrfs_info(fs_info, "forced readonly");
111                 /*
112                  * Note that a running device replace operation is not
113                  * canceled here although there is no way to update
114                  * the progress. It would add the risk of a deadlock,
115                  * therefore the canceling is omitted. The only penalty
116                  * is that some I/O remains active until the procedure
117                  * completes. The next time when the filesystem is
118                  * mounted writeable again, the device replace
119                  * operation continues.
120                  */
121         }
122 }
123
124 /*
125  * __btrfs_handle_fs_error decodes expected errors from the caller and
126  * invokes the approciate error response.
127  */
128 __cold
129 void __btrfs_handle_fs_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 #ifdef CONFIG_PRINTK
134         const char *errstr;
135 #endif
136
137         /*
138          * Special case: if the error is EROFS, and we're already
139          * under MS_RDONLY, then it is safe here.
140          */
141         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
142                 return;
143
144 #ifdef CONFIG_PRINTK
145         errstr = btrfs_decode_error(errno);
146         if (fmt) {
147                 struct va_format vaf;
148                 va_list args;
149
150                 va_start(args, fmt);
151                 vaf.fmt = fmt;
152                 vaf.va = &args;
153
154                 printk(KERN_CRIT
155                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
156                         sb->s_id, function, line, errno, errstr, &vaf);
157                 va_end(args);
158         } else {
159                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
160                         sb->s_id, function, line, errno, errstr);
161         }
162 #endif
163
164         /*
165          * Today we only save the error info to memory.  Long term we'll
166          * also send it down to the disk
167          */
168         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
169
170         /* Don't go through full error handling during mount */
171         if (sb->s_flags & MS_BORN)
172                 btrfs_handle_error(fs_info);
173 }
174
175 #ifdef CONFIG_PRINTK
176 static const char * const logtypes[] = {
177         "emergency",
178         "alert",
179         "critical",
180         "error",
181         "warning",
182         "notice",
183         "info",
184         "debug",
185 };
186
187
188 /*
189  * Use one ratelimit state per log level so that a flood of less important
190  * messages doesn't cause more important ones to be dropped.
191  */
192 static struct ratelimit_state printk_limits[] = {
193         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
194         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
195         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
196         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
197         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
198         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
199         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
200         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
201 };
202
203 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
204 {
205         struct super_block *sb = fs_info->sb;
206         char lvl[4];
207         struct va_format vaf;
208         va_list args;
209         const char *type = logtypes[4];
210         int kern_level;
211         struct ratelimit_state *ratelimit;
212
213         va_start(args, fmt);
214
215         kern_level = printk_get_level(fmt);
216         if (kern_level) {
217                 size_t size = printk_skip_level(fmt) - fmt;
218                 memcpy(lvl, fmt,  size);
219                 lvl[size] = '\0';
220                 fmt += size;
221                 type = logtypes[kern_level - '0'];
222                 ratelimit = &printk_limits[kern_level - '0'];
223         } else {
224                 *lvl = '\0';
225                 /* Default to debug output */
226                 ratelimit = &printk_limits[7];
227         }
228
229         vaf.fmt = fmt;
230         vaf.va = &args;
231
232         if (__ratelimit(ratelimit))
233                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
234
235         va_end(args);
236 }
237 #endif
238
239 /*
240  * We only mark the transaction aborted and then set the file system read-only.
241  * This will prevent new transactions from starting or trying to join this
242  * one.
243  *
244  * This means that error recovery at the call site is limited to freeing
245  * any local memory allocations and passing the error code up without
246  * further cleanup. The transaction should complete as it normally would
247  * in the call path but will return -EIO.
248  *
249  * We'll complete the cleanup in btrfs_end_transaction and
250  * btrfs_commit_transaction.
251  */
252 __cold
253 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
254                                const char *function,
255                                unsigned int line, int errno)
256 {
257         struct btrfs_fs_info *fs_info = trans->fs_info;
258
259         trans->aborted = errno;
260         /* Nothing used. The other threads that have joined this
261          * transaction may be able to continue. */
262         if (!trans->dirty && list_empty(&trans->new_bgs)) {
263                 const char *errstr;
264
265                 errstr = btrfs_decode_error(errno);
266                 btrfs_warn(fs_info,
267                            "%s:%d: Aborting unused transaction(%s).",
268                            function, line, errstr);
269                 return;
270         }
271         ACCESS_ONCE(trans->transaction->aborted) = errno;
272         /* Wake up anybody who may be waiting on this transaction */
273         wake_up(&fs_info->transaction_wait);
274         wake_up(&fs_info->transaction_blocked_wait);
275         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
276 }
277 /*
278  * __btrfs_panic decodes unexpected, fatal errors from the caller,
279  * issues an alert, and either panics or BUGs, depending on mount options.
280  */
281 __cold
282 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
283                    unsigned int line, int errno, const char *fmt, ...)
284 {
285         char *s_id = "<unknown>";
286         const char *errstr;
287         struct va_format vaf = { .fmt = fmt };
288         va_list args;
289
290         if (fs_info)
291                 s_id = fs_info->sb->s_id;
292
293         va_start(args, fmt);
294         vaf.va = &args;
295
296         errstr = btrfs_decode_error(errno);
297         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
298                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
299                         s_id, function, line, &vaf, errno, errstr);
300
301         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
302                    function, line, &vaf, errno, errstr);
303         va_end(args);
304         /* Caller calls BUG() */
305 }
306
307 static void btrfs_put_super(struct super_block *sb)
308 {
309         close_ctree(btrfs_sb(sb)->tree_root);
310 }
311
312 enum {
313         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
314         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
315         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
316         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
317         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
318         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
319         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
320         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
321         Opt_skip_balance, Opt_check_integrity,
322         Opt_check_integrity_including_extent_data,
323         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
324         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
325         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
326         Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
327         Opt_nologreplay, Opt_norecovery,
328 #ifdef CONFIG_BTRFS_DEBUG
329         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
330 #endif
331         Opt_err,
332 };
333
334 static const match_table_t tokens = {
335         {Opt_degraded, "degraded"},
336         {Opt_subvol, "subvol=%s"},
337         {Opt_subvolid, "subvolid=%s"},
338         {Opt_device, "device=%s"},
339         {Opt_nodatasum, "nodatasum"},
340         {Opt_datasum, "datasum"},
341         {Opt_nodatacow, "nodatacow"},
342         {Opt_datacow, "datacow"},
343         {Opt_nobarrier, "nobarrier"},
344         {Opt_barrier, "barrier"},
345         {Opt_max_inline, "max_inline=%s"},
346         {Opt_alloc_start, "alloc_start=%s"},
347         {Opt_thread_pool, "thread_pool=%d"},
348         {Opt_compress, "compress"},
349         {Opt_compress_type, "compress=%s"},
350         {Opt_compress_force, "compress-force"},
351         {Opt_compress_force_type, "compress-force=%s"},
352         {Opt_ssd, "ssd"},
353         {Opt_ssd_spread, "ssd_spread"},
354         {Opt_nossd, "nossd"},
355         {Opt_acl, "acl"},
356         {Opt_noacl, "noacl"},
357         {Opt_notreelog, "notreelog"},
358         {Opt_treelog, "treelog"},
359         {Opt_nologreplay, "nologreplay"},
360         {Opt_norecovery, "norecovery"},
361         {Opt_flushoncommit, "flushoncommit"},
362         {Opt_noflushoncommit, "noflushoncommit"},
363         {Opt_ratio, "metadata_ratio=%d"},
364         {Opt_discard, "discard"},
365         {Opt_nodiscard, "nodiscard"},
366         {Opt_space_cache, "space_cache"},
367         {Opt_space_cache_version, "space_cache=%s"},
368         {Opt_clear_cache, "clear_cache"},
369         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
370         {Opt_enospc_debug, "enospc_debug"},
371         {Opt_noenospc_debug, "noenospc_debug"},
372         {Opt_subvolrootid, "subvolrootid=%d"},
373         {Opt_defrag, "autodefrag"},
374         {Opt_nodefrag, "noautodefrag"},
375         {Opt_inode_cache, "inode_cache"},
376         {Opt_noinode_cache, "noinode_cache"},
377         {Opt_no_space_cache, "nospace_cache"},
378         {Opt_recovery, "recovery"}, /* deprecated */
379         {Opt_usebackuproot, "usebackuproot"},
380         {Opt_skip_balance, "skip_balance"},
381         {Opt_check_integrity, "check_int"},
382         {Opt_check_integrity_including_extent_data, "check_int_data"},
383         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
384         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
385         {Opt_fatal_errors, "fatal_errors=%s"},
386         {Opt_commit_interval, "commit=%d"},
387 #ifdef CONFIG_BTRFS_DEBUG
388         {Opt_fragment_data, "fragment=data"},
389         {Opt_fragment_metadata, "fragment=metadata"},
390         {Opt_fragment_all, "fragment=all"},
391 #endif
392         {Opt_err, NULL},
393 };
394
395 /*
396  * Regular mount options parser.  Everything that is needed only when
397  * reading in a new superblock is parsed here.
398  * XXX JDM: This needs to be cleaned up for remount.
399  */
400 int btrfs_parse_options(struct btrfs_root *root, char *options,
401                         unsigned long new_flags)
402 {
403         struct btrfs_fs_info *info = root->fs_info;
404         substring_t args[MAX_OPT_ARGS];
405         char *p, *num, *orig = NULL;
406         u64 cache_gen;
407         int intarg;
408         int ret = 0;
409         char *compress_type;
410         bool compress_force = false;
411         enum btrfs_compression_type saved_compress_type;
412         bool saved_compress_force;
413         int no_compress = 0;
414
415         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
416         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
417                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
418         else if (cache_gen)
419                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
420
421         /*
422          * Even the options are empty, we still need to do extra check
423          * against new flags
424          */
425         if (!options)
426                 goto check;
427
428         /*
429          * strsep changes the string, duplicate it because parse_options
430          * gets called twice
431          */
432         options = kstrdup(options, GFP_NOFS);
433         if (!options)
434                 return -ENOMEM;
435
436         orig = options;
437
438         while ((p = strsep(&options, ",")) != NULL) {
439                 int token;
440                 if (!*p)
441                         continue;
442
443                 token = match_token(p, tokens, args);
444                 switch (token) {
445                 case Opt_degraded:
446                         btrfs_info(root->fs_info, "allowing degraded mounts");
447                         btrfs_set_opt(info->mount_opt, DEGRADED);
448                         break;
449                 case Opt_subvol:
450                 case Opt_subvolid:
451                 case Opt_subvolrootid:
452                 case Opt_device:
453                         /*
454                          * These are parsed by btrfs_parse_early_options
455                          * and can be happily ignored here.
456                          */
457                         break;
458                 case Opt_nodatasum:
459                         btrfs_set_and_info(info, NODATASUM,
460                                            "setting nodatasum");
461                         break;
462                 case Opt_datasum:
463                         if (btrfs_test_opt(info, NODATASUM)) {
464                                 if (btrfs_test_opt(info, NODATACOW))
465                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
466                                 else
467                                         btrfs_info(root->fs_info, "setting datasum");
468                         }
469                         btrfs_clear_opt(info->mount_opt, NODATACOW);
470                         btrfs_clear_opt(info->mount_opt, NODATASUM);
471                         break;
472                 case Opt_nodatacow:
473                         if (!btrfs_test_opt(info, NODATACOW)) {
474                                 if (!btrfs_test_opt(info, COMPRESS) ||
475                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
476                                         btrfs_info(root->fs_info,
477                                                    "setting nodatacow, compression disabled");
478                                 } else {
479                                         btrfs_info(root->fs_info, "setting nodatacow");
480                                 }
481                         }
482                         btrfs_clear_opt(info->mount_opt, COMPRESS);
483                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
484                         btrfs_set_opt(info->mount_opt, NODATACOW);
485                         btrfs_set_opt(info->mount_opt, NODATASUM);
486                         break;
487                 case Opt_datacow:
488                         btrfs_clear_and_info(info, NODATACOW,
489                                              "setting datacow");
490                         break;
491                 case Opt_compress_force:
492                 case Opt_compress_force_type:
493                         compress_force = true;
494                         /* Fallthrough */
495                 case Opt_compress:
496                 case Opt_compress_type:
497                         saved_compress_type = btrfs_test_opt(info,
498                                                              COMPRESS) ?
499                                 info->compress_type : BTRFS_COMPRESS_NONE;
500                         saved_compress_force =
501                                 btrfs_test_opt(info, FORCE_COMPRESS);
502                         if (token == Opt_compress ||
503                             token == Opt_compress_force ||
504                             strcmp(args[0].from, "zlib") == 0) {
505                                 compress_type = "zlib";
506                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
507                                 btrfs_set_opt(info->mount_opt, COMPRESS);
508                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
509                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
510                                 no_compress = 0;
511                         } else if (strcmp(args[0].from, "lzo") == 0) {
512                                 compress_type = "lzo";
513                                 info->compress_type = BTRFS_COMPRESS_LZO;
514                                 btrfs_set_opt(info->mount_opt, COMPRESS);
515                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
516                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
517                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
518                                 no_compress = 0;
519                         } else if (strncmp(args[0].from, "no", 2) == 0) {
520                                 compress_type = "no";
521                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
522                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
523                                 compress_force = false;
524                                 no_compress++;
525                         } else {
526                                 ret = -EINVAL;
527                                 goto out;
528                         }
529
530                         if (compress_force) {
531                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
532                         } else {
533                                 /*
534                                  * If we remount from compress-force=xxx to
535                                  * compress=xxx, we need clear FORCE_COMPRESS
536                                  * flag, otherwise, there is no way for users
537                                  * to disable forcible compression separately.
538                                  */
539                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
540                         }
541                         if ((btrfs_test_opt(info, COMPRESS) &&
542                              (info->compress_type != saved_compress_type ||
543                               compress_force != saved_compress_force)) ||
544                             (!btrfs_test_opt(info, COMPRESS) &&
545                              no_compress == 1)) {
546                                 btrfs_info(root->fs_info,
547                                            "%s %s compression",
548                                            (compress_force) ? "force" : "use",
549                                            compress_type);
550                         }
551                         compress_force = false;
552                         break;
553                 case Opt_ssd:
554                         btrfs_set_and_info(info, SSD,
555                                            "use ssd allocation scheme");
556                         break;
557                 case Opt_ssd_spread:
558                         btrfs_set_and_info(info, SSD_SPREAD,
559                                            "use spread ssd allocation scheme");
560                         btrfs_set_opt(info->mount_opt, SSD);
561                         break;
562                 case Opt_nossd:
563                         btrfs_set_and_info(info, NOSSD,
564                                              "not using ssd allocation scheme");
565                         btrfs_clear_opt(info->mount_opt, SSD);
566                         break;
567                 case Opt_barrier:
568                         btrfs_clear_and_info(info, NOBARRIER,
569                                              "turning on barriers");
570                         break;
571                 case Opt_nobarrier:
572                         btrfs_set_and_info(info, NOBARRIER,
573                                            "turning off barriers");
574                         break;
575                 case Opt_thread_pool:
576                         ret = match_int(&args[0], &intarg);
577                         if (ret) {
578                                 goto out;
579                         } else if (intarg > 0) {
580                                 info->thread_pool_size = intarg;
581                         } else {
582                                 ret = -EINVAL;
583                                 goto out;
584                         }
585                         break;
586                 case Opt_max_inline:
587                         num = match_strdup(&args[0]);
588                         if (num) {
589                                 info->max_inline = memparse(num, NULL);
590                                 kfree(num);
591
592                                 if (info->max_inline) {
593                                         info->max_inline = min_t(u64,
594                                                 info->max_inline,
595                                                 root->sectorsize);
596                                 }
597                                 btrfs_info(root->fs_info, "max_inline at %llu",
598                                         info->max_inline);
599                         } else {
600                                 ret = -ENOMEM;
601                                 goto out;
602                         }
603                         break;
604                 case Opt_alloc_start:
605                         num = match_strdup(&args[0]);
606                         if (num) {
607                                 mutex_lock(&info->chunk_mutex);
608                                 info->alloc_start = memparse(num, NULL);
609                                 mutex_unlock(&info->chunk_mutex);
610                                 kfree(num);
611                                 btrfs_info(root->fs_info, "allocations start at %llu",
612                                         info->alloc_start);
613                         } else {
614                                 ret = -ENOMEM;
615                                 goto out;
616                         }
617                         break;
618                 case Opt_acl:
619 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
620                         root->fs_info->sb->s_flags |= MS_POSIXACL;
621                         break;
622 #else
623                         btrfs_err(root->fs_info,
624                                 "support for ACL not compiled in!");
625                         ret = -EINVAL;
626                         goto out;
627 #endif
628                 case Opt_noacl:
629                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
630                         break;
631                 case Opt_notreelog:
632                         btrfs_set_and_info(info, NOTREELOG,
633                                            "disabling tree log");
634                         break;
635                 case Opt_treelog:
636                         btrfs_clear_and_info(info, NOTREELOG,
637                                              "enabling tree log");
638                         break;
639                 case Opt_norecovery:
640                 case Opt_nologreplay:
641                         btrfs_set_and_info(info, NOLOGREPLAY,
642                                            "disabling log replay at mount time");
643                         break;
644                 case Opt_flushoncommit:
645                         btrfs_set_and_info(info, FLUSHONCOMMIT,
646                                            "turning on flush-on-commit");
647                         break;
648                 case Opt_noflushoncommit:
649                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
650                                              "turning off flush-on-commit");
651                         break;
652                 case Opt_ratio:
653                         ret = match_int(&args[0], &intarg);
654                         if (ret) {
655                                 goto out;
656                         } else if (intarg >= 0) {
657                                 info->metadata_ratio = intarg;
658                                 btrfs_info(root->fs_info, "metadata ratio %d",
659                                        info->metadata_ratio);
660                         } else {
661                                 ret = -EINVAL;
662                                 goto out;
663                         }
664                         break;
665                 case Opt_discard:
666                         btrfs_set_and_info(info, DISCARD,
667                                            "turning on discard");
668                         break;
669                 case Opt_nodiscard:
670                         btrfs_clear_and_info(info, DISCARD,
671                                              "turning off discard");
672                         break;
673                 case Opt_space_cache:
674                 case Opt_space_cache_version:
675                         if (token == Opt_space_cache ||
676                             strcmp(args[0].from, "v1") == 0) {
677                                 btrfs_clear_opt(root->fs_info->mount_opt,
678                                                 FREE_SPACE_TREE);
679                                 btrfs_set_and_info(info, SPACE_CACHE,
680                                                    "enabling disk space caching");
681                         } else if (strcmp(args[0].from, "v2") == 0) {
682                                 btrfs_clear_opt(root->fs_info->mount_opt,
683                                                 SPACE_CACHE);
684                                 btrfs_set_and_info(info,
685                                                    FREE_SPACE_TREE,
686                                                    "enabling free space tree");
687                         } else {
688                                 ret = -EINVAL;
689                                 goto out;
690                         }
691                         break;
692                 case Opt_rescan_uuid_tree:
693                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
694                         break;
695                 case Opt_no_space_cache:
696                         if (btrfs_test_opt(info, SPACE_CACHE)) {
697                                 btrfs_clear_and_info(info,
698                                                      SPACE_CACHE,
699                                                      "disabling disk space caching");
700                         }
701                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
702                                 btrfs_clear_and_info(info,
703                                                      FREE_SPACE_TREE,
704                                                      "disabling free space tree");
705                         }
706                         break;
707                 case Opt_inode_cache:
708                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
709                                            "enabling inode map caching");
710                         break;
711                 case Opt_noinode_cache:
712                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
713                                              "disabling inode map caching");
714                         break;
715                 case Opt_clear_cache:
716                         btrfs_set_and_info(info, CLEAR_CACHE,
717                                            "force clearing of disk cache");
718                         break;
719                 case Opt_user_subvol_rm_allowed:
720                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
721                         break;
722                 case Opt_enospc_debug:
723                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
724                         break;
725                 case Opt_noenospc_debug:
726                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
727                         break;
728                 case Opt_defrag:
729                         btrfs_set_and_info(info, AUTO_DEFRAG,
730                                            "enabling auto defrag");
731                         break;
732                 case Opt_nodefrag:
733                         btrfs_clear_and_info(info, AUTO_DEFRAG,
734                                              "disabling auto defrag");
735                         break;
736                 case Opt_recovery:
737                         btrfs_warn(root->fs_info,
738                                    "'recovery' is deprecated, use 'usebackuproot' instead");
739                 case Opt_usebackuproot:
740                         btrfs_info(root->fs_info,
741                                    "trying to use backup root at mount time");
742                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
743                         break;
744                 case Opt_skip_balance:
745                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
746                         break;
747 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
748                 case Opt_check_integrity_including_extent_data:
749                         btrfs_info(root->fs_info,
750                                    "enabling check integrity including extent data");
751                         btrfs_set_opt(info->mount_opt,
752                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
753                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
754                         break;
755                 case Opt_check_integrity:
756                         btrfs_info(root->fs_info, "enabling check integrity");
757                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
758                         break;
759                 case Opt_check_integrity_print_mask:
760                         ret = match_int(&args[0], &intarg);
761                         if (ret) {
762                                 goto out;
763                         } else if (intarg >= 0) {
764                                 info->check_integrity_print_mask = intarg;
765                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
766                                        info->check_integrity_print_mask);
767                         } else {
768                                 ret = -EINVAL;
769                                 goto out;
770                         }
771                         break;
772 #else
773                 case Opt_check_integrity_including_extent_data:
774                 case Opt_check_integrity:
775                 case Opt_check_integrity_print_mask:
776                         btrfs_err(root->fs_info,
777                                 "support for check_integrity* not compiled in!");
778                         ret = -EINVAL;
779                         goto out;
780 #endif
781                 case Opt_fatal_errors:
782                         if (strcmp(args[0].from, "panic") == 0)
783                                 btrfs_set_opt(info->mount_opt,
784                                               PANIC_ON_FATAL_ERROR);
785                         else if (strcmp(args[0].from, "bug") == 0)
786                                 btrfs_clear_opt(info->mount_opt,
787                                               PANIC_ON_FATAL_ERROR);
788                         else {
789                                 ret = -EINVAL;
790                                 goto out;
791                         }
792                         break;
793                 case Opt_commit_interval:
794                         intarg = 0;
795                         ret = match_int(&args[0], &intarg);
796                         if (ret < 0) {
797                                 btrfs_err(root->fs_info, "invalid commit interval");
798                                 ret = -EINVAL;
799                                 goto out;
800                         }
801                         if (intarg > 0) {
802                                 if (intarg > 300) {
803                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
804                                                         intarg);
805                                 }
806                                 info->commit_interval = intarg;
807                         } else {
808                                 btrfs_info(root->fs_info, "using default commit interval %ds",
809                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
810                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
811                         }
812                         break;
813 #ifdef CONFIG_BTRFS_DEBUG
814                 case Opt_fragment_all:
815                         btrfs_info(root->fs_info, "fragmenting all space");
816                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
817                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
818                         break;
819                 case Opt_fragment_metadata:
820                         btrfs_info(root->fs_info, "fragmenting metadata");
821                         btrfs_set_opt(info->mount_opt,
822                                       FRAGMENT_METADATA);
823                         break;
824                 case Opt_fragment_data:
825                         btrfs_info(root->fs_info, "fragmenting data");
826                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
827                         break;
828 #endif
829                 case Opt_err:
830                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
831                         ret = -EINVAL;
832                         goto out;
833                 default:
834                         break;
835                 }
836         }
837 check:
838         /*
839          * Extra check for current option against current flag
840          */
841         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
842                 btrfs_err(root->fs_info,
843                           "nologreplay must be used with ro mount option");
844                 ret = -EINVAL;
845         }
846 out:
847         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
848             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
849             !btrfs_test_opt(info, CLEAR_CACHE)) {
850                 btrfs_err(root->fs_info, "cannot disable free space tree");
851                 ret = -EINVAL;
852
853         }
854         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
855                 btrfs_info(root->fs_info, "disk space caching is enabled");
856         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
857                 btrfs_info(root->fs_info, "using free space tree");
858         kfree(orig);
859         return ret;
860 }
861
862 /*
863  * Parse mount options that are required early in the mount process.
864  *
865  * All other options will be parsed on much later in the mount process and
866  * only when we need to allocate a new super block.
867  */
868 static int btrfs_parse_early_options(const char *options, fmode_t flags,
869                 void *holder, char **subvol_name, u64 *subvol_objectid,
870                 struct btrfs_fs_devices **fs_devices)
871 {
872         substring_t args[MAX_OPT_ARGS];
873         char *device_name, *opts, *orig, *p;
874         char *num = NULL;
875         int error = 0;
876
877         if (!options)
878                 return 0;
879
880         /*
881          * strsep changes the string, duplicate it because parse_options
882          * gets called twice
883          */
884         opts = kstrdup(options, GFP_KERNEL);
885         if (!opts)
886                 return -ENOMEM;
887         orig = opts;
888
889         while ((p = strsep(&opts, ",")) != NULL) {
890                 int token;
891                 if (!*p)
892                         continue;
893
894                 token = match_token(p, tokens, args);
895                 switch (token) {
896                 case Opt_subvol:
897                         kfree(*subvol_name);
898                         *subvol_name = match_strdup(&args[0]);
899                         if (!*subvol_name) {
900                                 error = -ENOMEM;
901                                 goto out;
902                         }
903                         break;
904                 case Opt_subvolid:
905                         num = match_strdup(&args[0]);
906                         if (num) {
907                                 *subvol_objectid = memparse(num, NULL);
908                                 kfree(num);
909                                 /* we want the original fs_tree */
910                                 if (!*subvol_objectid)
911                                         *subvol_objectid =
912                                                 BTRFS_FS_TREE_OBJECTID;
913                         } else {
914                                 error = -EINVAL;
915                                 goto out;
916                         }
917                         break;
918                 case Opt_subvolrootid:
919                         printk(KERN_WARNING
920                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
921                                 "no effect\n");
922                         break;
923                 case Opt_device:
924                         device_name = match_strdup(&args[0]);
925                         if (!device_name) {
926                                 error = -ENOMEM;
927                                 goto out;
928                         }
929                         error = btrfs_scan_one_device(device_name,
930                                         flags, holder, fs_devices);
931                         kfree(device_name);
932                         if (error)
933                                 goto out;
934                         break;
935                 default:
936                         break;
937                 }
938         }
939
940 out:
941         kfree(orig);
942         return error;
943 }
944
945 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
946                                            u64 subvol_objectid)
947 {
948         struct btrfs_root *root = fs_info->tree_root;
949         struct btrfs_root *fs_root;
950         struct btrfs_root_ref *root_ref;
951         struct btrfs_inode_ref *inode_ref;
952         struct btrfs_key key;
953         struct btrfs_path *path = NULL;
954         char *name = NULL, *ptr;
955         u64 dirid;
956         int len;
957         int ret;
958
959         path = btrfs_alloc_path();
960         if (!path) {
961                 ret = -ENOMEM;
962                 goto err;
963         }
964         path->leave_spinning = 1;
965
966         name = kmalloc(PATH_MAX, GFP_NOFS);
967         if (!name) {
968                 ret = -ENOMEM;
969                 goto err;
970         }
971         ptr = name + PATH_MAX - 1;
972         ptr[0] = '\0';
973
974         /*
975          * Walk up the subvolume trees in the tree of tree roots by root
976          * backrefs until we hit the top-level subvolume.
977          */
978         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
979                 key.objectid = subvol_objectid;
980                 key.type = BTRFS_ROOT_BACKREF_KEY;
981                 key.offset = (u64)-1;
982
983                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
984                 if (ret < 0) {
985                         goto err;
986                 } else if (ret > 0) {
987                         ret = btrfs_previous_item(root, path, subvol_objectid,
988                                                   BTRFS_ROOT_BACKREF_KEY);
989                         if (ret < 0) {
990                                 goto err;
991                         } else if (ret > 0) {
992                                 ret = -ENOENT;
993                                 goto err;
994                         }
995                 }
996
997                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
998                 subvol_objectid = key.offset;
999
1000                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1001                                           struct btrfs_root_ref);
1002                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1003                 ptr -= len + 1;
1004                 if (ptr < name) {
1005                         ret = -ENAMETOOLONG;
1006                         goto err;
1007                 }
1008                 read_extent_buffer(path->nodes[0], ptr + 1,
1009                                    (unsigned long)(root_ref + 1), len);
1010                 ptr[0] = '/';
1011                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1012                 btrfs_release_path(path);
1013
1014                 key.objectid = subvol_objectid;
1015                 key.type = BTRFS_ROOT_ITEM_KEY;
1016                 key.offset = (u64)-1;
1017                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1018                 if (IS_ERR(fs_root)) {
1019                         ret = PTR_ERR(fs_root);
1020                         goto err;
1021                 }
1022
1023                 /*
1024                  * Walk up the filesystem tree by inode refs until we hit the
1025                  * root directory.
1026                  */
1027                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1028                         key.objectid = dirid;
1029                         key.type = BTRFS_INODE_REF_KEY;
1030                         key.offset = (u64)-1;
1031
1032                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1033                         if (ret < 0) {
1034                                 goto err;
1035                         } else if (ret > 0) {
1036                                 ret = btrfs_previous_item(fs_root, path, dirid,
1037                                                           BTRFS_INODE_REF_KEY);
1038                                 if (ret < 0) {
1039                                         goto err;
1040                                 } else if (ret > 0) {
1041                                         ret = -ENOENT;
1042                                         goto err;
1043                                 }
1044                         }
1045
1046                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1047                         dirid = key.offset;
1048
1049                         inode_ref = btrfs_item_ptr(path->nodes[0],
1050                                                    path->slots[0],
1051                                                    struct btrfs_inode_ref);
1052                         len = btrfs_inode_ref_name_len(path->nodes[0],
1053                                                        inode_ref);
1054                         ptr -= len + 1;
1055                         if (ptr < name) {
1056                                 ret = -ENAMETOOLONG;
1057                                 goto err;
1058                         }
1059                         read_extent_buffer(path->nodes[0], ptr + 1,
1060                                            (unsigned long)(inode_ref + 1), len);
1061                         ptr[0] = '/';
1062                         btrfs_release_path(path);
1063                 }
1064         }
1065
1066         btrfs_free_path(path);
1067         if (ptr == name + PATH_MAX - 1) {
1068                 name[0] = '/';
1069                 name[1] = '\0';
1070         } else {
1071                 memmove(name, ptr, name + PATH_MAX - ptr);
1072         }
1073         return name;
1074
1075 err:
1076         btrfs_free_path(path);
1077         kfree(name);
1078         return ERR_PTR(ret);
1079 }
1080
1081 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1082 {
1083         struct btrfs_root *root = fs_info->tree_root;
1084         struct btrfs_dir_item *di;
1085         struct btrfs_path *path;
1086         struct btrfs_key location;
1087         u64 dir_id;
1088
1089         path = btrfs_alloc_path();
1090         if (!path)
1091                 return -ENOMEM;
1092         path->leave_spinning = 1;
1093
1094         /*
1095          * Find the "default" dir item which points to the root item that we
1096          * will mount by default if we haven't been given a specific subvolume
1097          * to mount.
1098          */
1099         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1100         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1101         if (IS_ERR(di)) {
1102                 btrfs_free_path(path);
1103                 return PTR_ERR(di);
1104         }
1105         if (!di) {
1106                 /*
1107                  * Ok the default dir item isn't there.  This is weird since
1108                  * it's always been there, but don't freak out, just try and
1109                  * mount the top-level subvolume.
1110                  */
1111                 btrfs_free_path(path);
1112                 *objectid = BTRFS_FS_TREE_OBJECTID;
1113                 return 0;
1114         }
1115
1116         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1117         btrfs_free_path(path);
1118         *objectid = location.objectid;
1119         return 0;
1120 }
1121
1122 static int btrfs_fill_super(struct super_block *sb,
1123                             struct btrfs_fs_devices *fs_devices,
1124                             void *data, int silent)
1125 {
1126         struct inode *inode;
1127         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1128         struct btrfs_key key;
1129         int err;
1130
1131         sb->s_maxbytes = MAX_LFS_FILESIZE;
1132         sb->s_magic = BTRFS_SUPER_MAGIC;
1133         sb->s_op = &btrfs_super_ops;
1134         sb->s_d_op = &btrfs_dentry_operations;
1135         sb->s_export_op = &btrfs_export_ops;
1136         sb->s_xattr = btrfs_xattr_handlers;
1137         sb->s_time_gran = 1;
1138 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1139         sb->s_flags |= MS_POSIXACL;
1140 #endif
1141         sb->s_flags |= MS_I_VERSION;
1142         sb->s_iflags |= SB_I_CGROUPWB;
1143         err = open_ctree(sb, fs_devices, (char *)data);
1144         if (err) {
1145                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1146                 return err;
1147         }
1148
1149         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1150         key.type = BTRFS_INODE_ITEM_KEY;
1151         key.offset = 0;
1152         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1153         if (IS_ERR(inode)) {
1154                 err = PTR_ERR(inode);
1155                 goto fail_close;
1156         }
1157
1158         sb->s_root = d_make_root(inode);
1159         if (!sb->s_root) {
1160                 err = -ENOMEM;
1161                 goto fail_close;
1162         }
1163
1164         save_mount_options(sb, data);
1165         cleancache_init_fs(sb);
1166         sb->s_flags |= MS_ACTIVE;
1167         return 0;
1168
1169 fail_close:
1170         close_ctree(fs_info->tree_root);
1171         return err;
1172 }
1173
1174 int btrfs_sync_fs(struct super_block *sb, int wait)
1175 {
1176         struct btrfs_trans_handle *trans;
1177         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1178         struct btrfs_root *root = fs_info->tree_root;
1179
1180         trace_btrfs_sync_fs(fs_info, wait);
1181
1182         if (!wait) {
1183                 filemap_flush(fs_info->btree_inode->i_mapping);
1184                 return 0;
1185         }
1186
1187         btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
1188
1189         trans = btrfs_attach_transaction_barrier(root);
1190         if (IS_ERR(trans)) {
1191                 /* no transaction, don't bother */
1192                 if (PTR_ERR(trans) == -ENOENT) {
1193                         /*
1194                          * Exit unless we have some pending changes
1195                          * that need to go through commit
1196                          */
1197                         if (fs_info->pending_changes == 0)
1198                                 return 0;
1199                         /*
1200                          * A non-blocking test if the fs is frozen. We must not
1201                          * start a new transaction here otherwise a deadlock
1202                          * happens. The pending operations are delayed to the
1203                          * next commit after thawing.
1204                          */
1205                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1206                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1207                         else
1208                                 return 0;
1209                         trans = btrfs_start_transaction(root, 0);
1210                 }
1211                 if (IS_ERR(trans))
1212                         return PTR_ERR(trans);
1213         }
1214         return btrfs_commit_transaction(trans, root);
1215 }
1216
1217 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1218 {
1219         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1220         struct btrfs_root *root = info->tree_root;
1221         char *compress_type;
1222
1223         if (btrfs_test_opt(info, DEGRADED))
1224                 seq_puts(seq, ",degraded");
1225         if (btrfs_test_opt(info, NODATASUM))
1226                 seq_puts(seq, ",nodatasum");
1227         if (btrfs_test_opt(info, NODATACOW))
1228                 seq_puts(seq, ",nodatacow");
1229         if (btrfs_test_opt(info, NOBARRIER))
1230                 seq_puts(seq, ",nobarrier");
1231         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1232                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1233         if (info->alloc_start != 0)
1234                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1235         if (info->thread_pool_size !=  min_t(unsigned long,
1236                                              num_online_cpus() + 2, 8))
1237                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1238         if (btrfs_test_opt(info, COMPRESS)) {
1239                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1240                         compress_type = "zlib";
1241                 else
1242                         compress_type = "lzo";
1243                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1244                         seq_printf(seq, ",compress-force=%s", compress_type);
1245                 else
1246                         seq_printf(seq, ",compress=%s", compress_type);
1247         }
1248         if (btrfs_test_opt(info, NOSSD))
1249                 seq_puts(seq, ",nossd");
1250         if (btrfs_test_opt(info, SSD_SPREAD))
1251                 seq_puts(seq, ",ssd_spread");
1252         else if (btrfs_test_opt(info, SSD))
1253                 seq_puts(seq, ",ssd");
1254         if (btrfs_test_opt(info, NOTREELOG))
1255                 seq_puts(seq, ",notreelog");
1256         if (btrfs_test_opt(info, NOLOGREPLAY))
1257                 seq_puts(seq, ",nologreplay");
1258         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1259                 seq_puts(seq, ",flushoncommit");
1260         if (btrfs_test_opt(info, DISCARD))
1261                 seq_puts(seq, ",discard");
1262         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1263                 seq_puts(seq, ",noacl");
1264         if (btrfs_test_opt(info, SPACE_CACHE))
1265                 seq_puts(seq, ",space_cache");
1266         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1267                 seq_puts(seq, ",space_cache=v2");
1268         else
1269                 seq_puts(seq, ",nospace_cache");
1270         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1271                 seq_puts(seq, ",rescan_uuid_tree");
1272         if (btrfs_test_opt(info, CLEAR_CACHE))
1273                 seq_puts(seq, ",clear_cache");
1274         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1275                 seq_puts(seq, ",user_subvol_rm_allowed");
1276         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1277                 seq_puts(seq, ",enospc_debug");
1278         if (btrfs_test_opt(info, AUTO_DEFRAG))
1279                 seq_puts(seq, ",autodefrag");
1280         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1281                 seq_puts(seq, ",inode_cache");
1282         if (btrfs_test_opt(info, SKIP_BALANCE))
1283                 seq_puts(seq, ",skip_balance");
1284 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1285         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1286                 seq_puts(seq, ",check_int_data");
1287         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1288                 seq_puts(seq, ",check_int");
1289         if (info->check_integrity_print_mask)
1290                 seq_printf(seq, ",check_int_print_mask=%d",
1291                                 info->check_integrity_print_mask);
1292 #endif
1293         if (info->metadata_ratio)
1294                 seq_printf(seq, ",metadata_ratio=%d",
1295                                 info->metadata_ratio);
1296         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1297                 seq_puts(seq, ",fatal_errors=panic");
1298         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1299                 seq_printf(seq, ",commit=%d", info->commit_interval);
1300 #ifdef CONFIG_BTRFS_DEBUG
1301         if (btrfs_test_opt(info, FRAGMENT_DATA))
1302                 seq_puts(seq, ",fragment=data");
1303         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1304                 seq_puts(seq, ",fragment=metadata");
1305 #endif
1306         seq_printf(seq, ",subvolid=%llu",
1307                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1308         seq_puts(seq, ",subvol=");
1309         seq_dentry(seq, dentry, " \t\n\\");
1310         return 0;
1311 }
1312
1313 static int btrfs_test_super(struct super_block *s, void *data)
1314 {
1315         struct btrfs_fs_info *p = data;
1316         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1317
1318         return fs_info->fs_devices == p->fs_devices;
1319 }
1320
1321 static int btrfs_set_super(struct super_block *s, void *data)
1322 {
1323         int err = set_anon_super(s, data);
1324         if (!err)
1325                 s->s_fs_info = data;
1326         return err;
1327 }
1328
1329 /*
1330  * subvolumes are identified by ino 256
1331  */
1332 static inline int is_subvolume_inode(struct inode *inode)
1333 {
1334         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1335                 return 1;
1336         return 0;
1337 }
1338
1339 /*
1340  * This will add subvolid=0 to the argument string while removing any subvol=
1341  * and subvolid= arguments to make sure we get the top-level root for path
1342  * walking to the subvol we want.
1343  */
1344 static char *setup_root_args(char *args)
1345 {
1346         char *buf, *dst, *sep;
1347
1348         if (!args)
1349                 return kstrdup("subvolid=0", GFP_NOFS);
1350
1351         /* The worst case is that we add ",subvolid=0" to the end. */
1352         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1353         if (!buf)
1354                 return NULL;
1355
1356         while (1) {
1357                 sep = strchrnul(args, ',');
1358                 if (!strstarts(args, "subvol=") &&
1359                     !strstarts(args, "subvolid=")) {
1360                         memcpy(dst, args, sep - args);
1361                         dst += sep - args;
1362                         *dst++ = ',';
1363                 }
1364                 if (*sep)
1365                         args = sep + 1;
1366                 else
1367                         break;
1368         }
1369         strcpy(dst, "subvolid=0");
1370
1371         return buf;
1372 }
1373
1374 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1375                                    int flags, const char *device_name,
1376                                    char *data)
1377 {
1378         struct dentry *root;
1379         struct vfsmount *mnt = NULL;
1380         char *newargs;
1381         int ret;
1382
1383         newargs = setup_root_args(data);
1384         if (!newargs) {
1385                 root = ERR_PTR(-ENOMEM);
1386                 goto out;
1387         }
1388
1389         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1390         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1391                 if (flags & MS_RDONLY) {
1392                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1393                                              device_name, newargs);
1394                 } else {
1395                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1396                                              device_name, newargs);
1397                         if (IS_ERR(mnt)) {
1398                                 root = ERR_CAST(mnt);
1399                                 mnt = NULL;
1400                                 goto out;
1401                         }
1402
1403                         down_write(&mnt->mnt_sb->s_umount);
1404                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1405                         up_write(&mnt->mnt_sb->s_umount);
1406                         if (ret < 0) {
1407                                 root = ERR_PTR(ret);
1408                                 goto out;
1409                         }
1410                 }
1411         }
1412         if (IS_ERR(mnt)) {
1413                 root = ERR_CAST(mnt);
1414                 mnt = NULL;
1415                 goto out;
1416         }
1417
1418         if (!subvol_name) {
1419                 if (!subvol_objectid) {
1420                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1421                                                           &subvol_objectid);
1422                         if (ret) {
1423                                 root = ERR_PTR(ret);
1424                                 goto out;
1425                         }
1426                 }
1427                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1428                                                             subvol_objectid);
1429                 if (IS_ERR(subvol_name)) {
1430                         root = ERR_CAST(subvol_name);
1431                         subvol_name = NULL;
1432                         goto out;
1433                 }
1434
1435         }
1436
1437         root = mount_subtree(mnt, subvol_name);
1438         /* mount_subtree() drops our reference on the vfsmount. */
1439         mnt = NULL;
1440
1441         if (!IS_ERR(root)) {
1442                 struct super_block *s = root->d_sb;
1443                 struct inode *root_inode = d_inode(root);
1444                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1445
1446                 ret = 0;
1447                 if (!is_subvolume_inode(root_inode)) {
1448                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1449                                subvol_name);
1450                         ret = -EINVAL;
1451                 }
1452                 if (subvol_objectid && root_objectid != subvol_objectid) {
1453                         /*
1454                          * This will also catch a race condition where a
1455                          * subvolume which was passed by ID is renamed and
1456                          * another subvolume is renamed over the old location.
1457                          */
1458                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1459                                subvol_name, subvol_objectid);
1460                         ret = -EINVAL;
1461                 }
1462                 if (ret) {
1463                         dput(root);
1464                         root = ERR_PTR(ret);
1465                         deactivate_locked_super(s);
1466                 }
1467         }
1468
1469 out:
1470         mntput(mnt);
1471         kfree(newargs);
1472         kfree(subvol_name);
1473         return root;
1474 }
1475
1476 static int parse_security_options(char *orig_opts,
1477                                   struct security_mnt_opts *sec_opts)
1478 {
1479         char *secdata = NULL;
1480         int ret = 0;
1481
1482         secdata = alloc_secdata();
1483         if (!secdata)
1484                 return -ENOMEM;
1485         ret = security_sb_copy_data(orig_opts, secdata);
1486         if (ret) {
1487                 free_secdata(secdata);
1488                 return ret;
1489         }
1490         ret = security_sb_parse_opts_str(secdata, sec_opts);
1491         free_secdata(secdata);
1492         return ret;
1493 }
1494
1495 static int setup_security_options(struct btrfs_fs_info *fs_info,
1496                                   struct super_block *sb,
1497                                   struct security_mnt_opts *sec_opts)
1498 {
1499         int ret = 0;
1500
1501         /*
1502          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1503          * is valid.
1504          */
1505         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1506         if (ret)
1507                 return ret;
1508
1509 #ifdef CONFIG_SECURITY
1510         if (!fs_info->security_opts.num_mnt_opts) {
1511                 /* first time security setup, copy sec_opts to fs_info */
1512                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1513         } else {
1514                 /*
1515                  * Since SELinux (the only one supporting security_mnt_opts)
1516                  * does NOT support changing context during remount/mount of
1517                  * the same sb, this must be the same or part of the same
1518                  * security options, just free it.
1519                  */
1520                 security_free_mnt_opts(sec_opts);
1521         }
1522 #endif
1523         return ret;
1524 }
1525
1526 /*
1527  * Find a superblock for the given device / mount point.
1528  *
1529  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1530  *        for multiple device setup.  Make sure to keep it in sync.
1531  */
1532 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1533                 const char *device_name, void *data)
1534 {
1535         struct block_device *bdev = NULL;
1536         struct super_block *s;
1537         struct btrfs_fs_devices *fs_devices = NULL;
1538         struct btrfs_fs_info *fs_info = NULL;
1539         struct security_mnt_opts new_sec_opts;
1540         fmode_t mode = FMODE_READ;
1541         char *subvol_name = NULL;
1542         u64 subvol_objectid = 0;
1543         int error = 0;
1544
1545         if (!(flags & MS_RDONLY))
1546                 mode |= FMODE_WRITE;
1547
1548         error = btrfs_parse_early_options(data, mode, fs_type,
1549                                           &subvol_name, &subvol_objectid,
1550                                           &fs_devices);
1551         if (error) {
1552                 kfree(subvol_name);
1553                 return ERR_PTR(error);
1554         }
1555
1556         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1557                 /* mount_subvol() will free subvol_name. */
1558                 return mount_subvol(subvol_name, subvol_objectid, flags,
1559                                     device_name, data);
1560         }
1561
1562         security_init_mnt_opts(&new_sec_opts);
1563         if (data) {
1564                 error = parse_security_options(data, &new_sec_opts);
1565                 if (error)
1566                         return ERR_PTR(error);
1567         }
1568
1569         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1570         if (error)
1571                 goto error_sec_opts;
1572
1573         /*
1574          * Setup a dummy root and fs_info for test/set super.  This is because
1575          * we don't actually fill this stuff out until open_ctree, but we need
1576          * it for searching for existing supers, so this lets us do that and
1577          * then open_ctree will properly initialize everything later.
1578          */
1579         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1580         if (!fs_info) {
1581                 error = -ENOMEM;
1582                 goto error_sec_opts;
1583         }
1584
1585         fs_info->fs_devices = fs_devices;
1586
1587         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1588         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1589         security_init_mnt_opts(&fs_info->security_opts);
1590         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1591                 error = -ENOMEM;
1592                 goto error_fs_info;
1593         }
1594
1595         error = btrfs_open_devices(fs_devices, mode, fs_type);
1596         if (error)
1597                 goto error_fs_info;
1598
1599         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1600                 error = -EACCES;
1601                 goto error_close_devices;
1602         }
1603
1604         bdev = fs_devices->latest_bdev;
1605         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1606                  fs_info);
1607         if (IS_ERR(s)) {
1608                 error = PTR_ERR(s);
1609                 goto error_close_devices;
1610         }
1611
1612         if (s->s_root) {
1613                 btrfs_close_devices(fs_devices);
1614                 free_fs_info(fs_info);
1615                 if ((flags ^ s->s_flags) & MS_RDONLY)
1616                         error = -EBUSY;
1617         } else {
1618                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1619                 btrfs_sb(s)->bdev_holder = fs_type;
1620                 error = btrfs_fill_super(s, fs_devices, data,
1621                                          flags & MS_SILENT ? 1 : 0);
1622         }
1623         if (error) {
1624                 deactivate_locked_super(s);
1625                 goto error_sec_opts;
1626         }
1627
1628         fs_info = btrfs_sb(s);
1629         error = setup_security_options(fs_info, s, &new_sec_opts);
1630         if (error) {
1631                 deactivate_locked_super(s);
1632                 goto error_sec_opts;
1633         }
1634
1635         return dget(s->s_root);
1636
1637 error_close_devices:
1638         btrfs_close_devices(fs_devices);
1639 error_fs_info:
1640         free_fs_info(fs_info);
1641 error_sec_opts:
1642         security_free_mnt_opts(&new_sec_opts);
1643         return ERR_PTR(error);
1644 }
1645
1646 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1647                                      int new_pool_size, int old_pool_size)
1648 {
1649         if (new_pool_size == old_pool_size)
1650                 return;
1651
1652         fs_info->thread_pool_size = new_pool_size;
1653
1654         btrfs_info(fs_info, "resize thread pool %d -> %d",
1655                old_pool_size, new_pool_size);
1656
1657         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1658         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1659         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1660         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1661         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1662         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1663         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1664                                 new_pool_size);
1665         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1666         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1667         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1668         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1669         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1670                                 new_pool_size);
1671 }
1672
1673 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1674 {
1675         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1676 }
1677
1678 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1679                                        unsigned long old_opts, int flags)
1680 {
1681         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1682             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1683              (flags & MS_RDONLY))) {
1684                 /* wait for any defraggers to finish */
1685                 wait_event(fs_info->transaction_wait,
1686                            (atomic_read(&fs_info->defrag_running) == 0));
1687                 if (flags & MS_RDONLY)
1688                         sync_filesystem(fs_info->sb);
1689         }
1690 }
1691
1692 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1693                                          unsigned long old_opts)
1694 {
1695         /*
1696          * We need to cleanup all defragable inodes if the autodefragment is
1697          * close or the filesystem is read only.
1698          */
1699         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1700             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1701              (fs_info->sb->s_flags & MS_RDONLY))) {
1702                 btrfs_cleanup_defrag_inodes(fs_info);
1703         }
1704
1705         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1706 }
1707
1708 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1709 {
1710         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1711         struct btrfs_root *root = fs_info->tree_root;
1712         unsigned old_flags = sb->s_flags;
1713         unsigned long old_opts = fs_info->mount_opt;
1714         unsigned long old_compress_type = fs_info->compress_type;
1715         u64 old_max_inline = fs_info->max_inline;
1716         u64 old_alloc_start = fs_info->alloc_start;
1717         int old_thread_pool_size = fs_info->thread_pool_size;
1718         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1719         int ret;
1720
1721         sync_filesystem(sb);
1722         btrfs_remount_prepare(fs_info);
1723
1724         if (data) {
1725                 struct security_mnt_opts new_sec_opts;
1726
1727                 security_init_mnt_opts(&new_sec_opts);
1728                 ret = parse_security_options(data, &new_sec_opts);
1729                 if (ret)
1730                         goto restore;
1731                 ret = setup_security_options(fs_info, sb,
1732                                              &new_sec_opts);
1733                 if (ret) {
1734                         security_free_mnt_opts(&new_sec_opts);
1735                         goto restore;
1736                 }
1737         }
1738
1739         ret = btrfs_parse_options(root, data, *flags);
1740         if (ret) {
1741                 ret = -EINVAL;
1742                 goto restore;
1743         }
1744
1745         btrfs_remount_begin(fs_info, old_opts, *flags);
1746         btrfs_resize_thread_pool(fs_info,
1747                 fs_info->thread_pool_size, old_thread_pool_size);
1748
1749         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1750                 goto out;
1751
1752         if (*flags & MS_RDONLY) {
1753                 /*
1754                  * this also happens on 'umount -rf' or on shutdown, when
1755                  * the filesystem is busy.
1756                  */
1757                 cancel_work_sync(&fs_info->async_reclaim_work);
1758
1759                 /* wait for the uuid_scan task to finish */
1760                 down(&fs_info->uuid_tree_rescan_sem);
1761                 /* avoid complains from lockdep et al. */
1762                 up(&fs_info->uuid_tree_rescan_sem);
1763
1764                 sb->s_flags |= MS_RDONLY;
1765
1766                 /*
1767                  * Setting MS_RDONLY will put the cleaner thread to
1768                  * sleep at the next loop if it's already active.
1769                  * If it's already asleep, we'll leave unused block
1770                  * groups on disk until we're mounted read-write again
1771                  * unless we clean them up here.
1772                  */
1773                 btrfs_delete_unused_bgs(fs_info);
1774
1775                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1776                 btrfs_scrub_cancel(fs_info);
1777                 btrfs_pause_balance(fs_info);
1778
1779                 ret = btrfs_commit_super(root);
1780                 if (ret)
1781                         goto restore;
1782         } else {
1783                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1784                         btrfs_err(fs_info,
1785                                 "Remounting read-write after error is not allowed");
1786                         ret = -EINVAL;
1787                         goto restore;
1788                 }
1789                 if (fs_info->fs_devices->rw_devices == 0) {
1790                         ret = -EACCES;
1791                         goto restore;
1792                 }
1793
1794                 if (fs_info->fs_devices->missing_devices >
1795                      fs_info->num_tolerated_disk_barrier_failures &&
1796                     !(*flags & MS_RDONLY)) {
1797                         btrfs_warn(fs_info,
1798                                 "too many missing devices, writeable remount is not allowed");
1799                         ret = -EACCES;
1800                         goto restore;
1801                 }
1802
1803                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1804                         ret = -EINVAL;
1805                         goto restore;
1806                 }
1807
1808                 ret = btrfs_cleanup_fs_roots(fs_info);
1809                 if (ret)
1810                         goto restore;
1811
1812                 /* recover relocation */
1813                 mutex_lock(&fs_info->cleaner_mutex);
1814                 ret = btrfs_recover_relocation(root);
1815                 mutex_unlock(&fs_info->cleaner_mutex);
1816                 if (ret)
1817                         goto restore;
1818
1819                 ret = btrfs_resume_balance_async(fs_info);
1820                 if (ret)
1821                         goto restore;
1822
1823                 ret = btrfs_resume_dev_replace_async(fs_info);
1824                 if (ret) {
1825                         btrfs_warn(fs_info, "failed to resume dev_replace");
1826                         goto restore;
1827                 }
1828
1829                 if (!fs_info->uuid_root) {
1830                         btrfs_info(fs_info, "creating UUID tree");
1831                         ret = btrfs_create_uuid_tree(fs_info);
1832                         if (ret) {
1833                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1834                                 goto restore;
1835                         }
1836                 }
1837                 sb->s_flags &= ~MS_RDONLY;
1838
1839                 fs_info->open = 1;
1840         }
1841 out:
1842         wake_up_process(fs_info->transaction_kthread);
1843         btrfs_remount_cleanup(fs_info, old_opts);
1844         return 0;
1845
1846 restore:
1847         /* We've hit an error - don't reset MS_RDONLY */
1848         if (sb->s_flags & MS_RDONLY)
1849                 old_flags |= MS_RDONLY;
1850         sb->s_flags = old_flags;
1851         fs_info->mount_opt = old_opts;
1852         fs_info->compress_type = old_compress_type;
1853         fs_info->max_inline = old_max_inline;
1854         mutex_lock(&fs_info->chunk_mutex);
1855         fs_info->alloc_start = old_alloc_start;
1856         mutex_unlock(&fs_info->chunk_mutex);
1857         btrfs_resize_thread_pool(fs_info,
1858                 old_thread_pool_size, fs_info->thread_pool_size);
1859         fs_info->metadata_ratio = old_metadata_ratio;
1860         btrfs_remount_cleanup(fs_info, old_opts);
1861         return ret;
1862 }
1863
1864 /* Used to sort the devices by max_avail(descending sort) */
1865 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1866                                        const void *dev_info2)
1867 {
1868         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1869             ((struct btrfs_device_info *)dev_info2)->max_avail)
1870                 return -1;
1871         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1872                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1873                 return 1;
1874         else
1875         return 0;
1876 }
1877
1878 /*
1879  * sort the devices by max_avail, in which max free extent size of each device
1880  * is stored.(Descending Sort)
1881  */
1882 static inline void btrfs_descending_sort_devices(
1883                                         struct btrfs_device_info *devices,
1884                                         size_t nr_devices)
1885 {
1886         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1887              btrfs_cmp_device_free_bytes, NULL);
1888 }
1889
1890 /*
1891  * The helper to calc the free space on the devices that can be used to store
1892  * file data.
1893  */
1894 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1895 {
1896         struct btrfs_fs_info *fs_info = root->fs_info;
1897         struct btrfs_device_info *devices_info;
1898         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1899         struct btrfs_device *device;
1900         u64 skip_space;
1901         u64 type;
1902         u64 avail_space;
1903         u64 used_space;
1904         u64 min_stripe_size;
1905         int min_stripes = 1, num_stripes = 1;
1906         int i = 0, nr_devices;
1907         int ret;
1908
1909         /*
1910          * We aren't under the device list lock, so this is racy-ish, but good
1911          * enough for our purposes.
1912          */
1913         nr_devices = fs_info->fs_devices->open_devices;
1914         if (!nr_devices) {
1915                 smp_mb();
1916                 nr_devices = fs_info->fs_devices->open_devices;
1917                 ASSERT(nr_devices);
1918                 if (!nr_devices) {
1919                         *free_bytes = 0;
1920                         return 0;
1921                 }
1922         }
1923
1924         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1925                                GFP_NOFS);
1926         if (!devices_info)
1927                 return -ENOMEM;
1928
1929         /* calc min stripe number for data space allocation */
1930         type = btrfs_get_alloc_profile(root, 1);
1931         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1932                 min_stripes = 2;
1933                 num_stripes = nr_devices;
1934         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1935                 min_stripes = 2;
1936                 num_stripes = 2;
1937         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1938                 min_stripes = 4;
1939                 num_stripes = 4;
1940         }
1941
1942         if (type & BTRFS_BLOCK_GROUP_DUP)
1943                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1944         else
1945                 min_stripe_size = BTRFS_STRIPE_LEN;
1946
1947         if (fs_info->alloc_start)
1948                 mutex_lock(&fs_devices->device_list_mutex);
1949         rcu_read_lock();
1950         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1951                 if (!device->in_fs_metadata || !device->bdev ||
1952                     device->is_tgtdev_for_dev_replace)
1953                         continue;
1954
1955                 if (i >= nr_devices)
1956                         break;
1957
1958                 avail_space = device->total_bytes - device->bytes_used;
1959
1960                 /* align with stripe_len */
1961                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1962                 avail_space *= BTRFS_STRIPE_LEN;
1963
1964                 /*
1965                  * In order to avoid overwriting the superblock on the drive,
1966                  * btrfs starts at an offset of at least 1MB when doing chunk
1967                  * allocation.
1968                  */
1969                 skip_space = SZ_1M;
1970
1971                 /* user can set the offset in fs_info->alloc_start. */
1972                 if (fs_info->alloc_start &&
1973                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1974                     device->total_bytes) {
1975                         rcu_read_unlock();
1976                         skip_space = max(fs_info->alloc_start, skip_space);
1977
1978                         /*
1979                          * btrfs can not use the free space in
1980                          * [0, skip_space - 1], we must subtract it from the
1981                          * total. In order to implement it, we account the used
1982                          * space in this range first.
1983                          */
1984                         ret = btrfs_account_dev_extents_size(device, 0,
1985                                                              skip_space - 1,
1986                                                              &used_space);
1987                         if (ret) {
1988                                 kfree(devices_info);
1989                                 mutex_unlock(&fs_devices->device_list_mutex);
1990                                 return ret;
1991                         }
1992
1993                         rcu_read_lock();
1994
1995                         /* calc the free space in [0, skip_space - 1] */
1996                         skip_space -= used_space;
1997                 }
1998
1999                 /*
2000                  * we can use the free space in [0, skip_space - 1], subtract
2001                  * it from the total.
2002                  */
2003                 if (avail_space && avail_space >= skip_space)
2004                         avail_space -= skip_space;
2005                 else
2006                         avail_space = 0;
2007
2008                 if (avail_space < min_stripe_size)
2009                         continue;
2010
2011                 devices_info[i].dev = device;
2012                 devices_info[i].max_avail = avail_space;
2013
2014                 i++;
2015         }
2016         rcu_read_unlock();
2017         if (fs_info->alloc_start)
2018                 mutex_unlock(&fs_devices->device_list_mutex);
2019
2020         nr_devices = i;
2021
2022         btrfs_descending_sort_devices(devices_info, nr_devices);
2023
2024         i = nr_devices - 1;
2025         avail_space = 0;
2026         while (nr_devices >= min_stripes) {
2027                 if (num_stripes > nr_devices)
2028                         num_stripes = nr_devices;
2029
2030                 if (devices_info[i].max_avail >= min_stripe_size) {
2031                         int j;
2032                         u64 alloc_size;
2033
2034                         avail_space += devices_info[i].max_avail * num_stripes;
2035                         alloc_size = devices_info[i].max_avail;
2036                         for (j = i + 1 - num_stripes; j <= i; j++)
2037                                 devices_info[j].max_avail -= alloc_size;
2038                 }
2039                 i--;
2040                 nr_devices--;
2041         }
2042
2043         kfree(devices_info);
2044         *free_bytes = avail_space;
2045         return 0;
2046 }
2047
2048 /*
2049  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2050  *
2051  * If there's a redundant raid level at DATA block groups, use the respective
2052  * multiplier to scale the sizes.
2053  *
2054  * Unused device space usage is based on simulating the chunk allocator
2055  * algorithm that respects the device sizes, order of allocations and the
2056  * 'alloc_start' value, this is a close approximation of the actual use but
2057  * there are other factors that may change the result (like a new metadata
2058  * chunk).
2059  *
2060  * If metadata is exhausted, f_bavail will be 0.
2061  */
2062 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2063 {
2064         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2065         struct btrfs_super_block *disk_super = fs_info->super_copy;
2066         struct list_head *head = &fs_info->space_info;
2067         struct btrfs_space_info *found;
2068         u64 total_used = 0;
2069         u64 total_free_data = 0;
2070         u64 total_free_meta = 0;
2071         int bits = dentry->d_sb->s_blocksize_bits;
2072         __be32 *fsid = (__be32 *)fs_info->fsid;
2073         unsigned factor = 1;
2074         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2075         int ret;
2076         u64 thresh = 0;
2077         int mixed = 0;
2078
2079         /*
2080          * holding chunk_mutex to avoid allocating new chunks, holding
2081          * device_list_mutex to avoid the device being removed
2082          */
2083         rcu_read_lock();
2084         list_for_each_entry_rcu(found, head, list) {
2085                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2086                         int i;
2087
2088                         total_free_data += found->disk_total - found->disk_used;
2089                         total_free_data -=
2090                                 btrfs_account_ro_block_groups_free_space(found);
2091
2092                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2093                                 if (!list_empty(&found->block_groups[i])) {
2094                                         switch (i) {
2095                                         case BTRFS_RAID_DUP:
2096                                         case BTRFS_RAID_RAID1:
2097                                         case BTRFS_RAID_RAID10:
2098                                                 factor = 2;
2099                                         }
2100                                 }
2101                         }
2102                 }
2103
2104                 /*
2105                  * Metadata in mixed block goup profiles are accounted in data
2106                  */
2107                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2108                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2109                                 mixed = 1;
2110                         else
2111                                 total_free_meta += found->disk_total -
2112                                         found->disk_used;
2113                 }
2114
2115                 total_used += found->disk_used;
2116         }
2117
2118         rcu_read_unlock();
2119
2120         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2121         buf->f_blocks >>= bits;
2122         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2123
2124         /* Account global block reserve as used, it's in logical size already */
2125         spin_lock(&block_rsv->lock);
2126         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2127         if (buf->f_bfree >= block_rsv->size >> bits)
2128                 buf->f_bfree -= block_rsv->size >> bits;
2129         else
2130                 buf->f_bfree = 0;
2131         spin_unlock(&block_rsv->lock);
2132
2133         buf->f_bavail = div_u64(total_free_data, factor);
2134         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2135         if (ret)
2136                 return ret;
2137         buf->f_bavail += div_u64(total_free_data, factor);
2138         buf->f_bavail = buf->f_bavail >> bits;
2139
2140         /*
2141          * We calculate the remaining metadata space minus global reserve. If
2142          * this is (supposedly) smaller than zero, there's no space. But this
2143          * does not hold in practice, the exhausted state happens where's still
2144          * some positive delta. So we apply some guesswork and compare the
2145          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2146          *
2147          * We probably cannot calculate the exact threshold value because this
2148          * depends on the internal reservations requested by various
2149          * operations, so some operations that consume a few metadata will
2150          * succeed even if the Avail is zero. But this is better than the other
2151          * way around.
2152          */
2153         thresh = 4 * 1024 * 1024;
2154
2155         if (!mixed && total_free_meta - thresh < block_rsv->size)
2156                 buf->f_bavail = 0;
2157
2158         buf->f_type = BTRFS_SUPER_MAGIC;
2159         buf->f_bsize = dentry->d_sb->s_blocksize;
2160         buf->f_namelen = BTRFS_NAME_LEN;
2161
2162         /* We treat it as constant endianness (it doesn't matter _which_)
2163            because we want the fsid to come out the same whether mounted
2164            on a big-endian or little-endian host */
2165         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2166         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2167         /* Mask in the root object ID too, to disambiguate subvols */
2168         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2169         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2170
2171         return 0;
2172 }
2173
2174 static void btrfs_kill_super(struct super_block *sb)
2175 {
2176         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2177         kill_anon_super(sb);
2178         free_fs_info(fs_info);
2179 }
2180
2181 static struct file_system_type btrfs_fs_type = {
2182         .owner          = THIS_MODULE,
2183         .name           = "btrfs",
2184         .mount          = btrfs_mount,
2185         .kill_sb        = btrfs_kill_super,
2186         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2187 };
2188 MODULE_ALIAS_FS("btrfs");
2189
2190 static int btrfs_control_open(struct inode *inode, struct file *file)
2191 {
2192         /*
2193          * The control file's private_data is used to hold the
2194          * transaction when it is started and is used to keep
2195          * track of whether a transaction is already in progress.
2196          */
2197         file->private_data = NULL;
2198         return 0;
2199 }
2200
2201 /*
2202  * used by btrfsctl to scan devices when no FS is mounted
2203  */
2204 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2205                                 unsigned long arg)
2206 {
2207         struct btrfs_ioctl_vol_args *vol;
2208         struct btrfs_fs_devices *fs_devices;
2209         int ret = -ENOTTY;
2210
2211         if (!capable(CAP_SYS_ADMIN))
2212                 return -EPERM;
2213
2214         vol = memdup_user((void __user *)arg, sizeof(*vol));
2215         if (IS_ERR(vol))
2216                 return PTR_ERR(vol);
2217
2218         switch (cmd) {
2219         case BTRFS_IOC_SCAN_DEV:
2220                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2221                                             &btrfs_fs_type, &fs_devices);
2222                 break;
2223         case BTRFS_IOC_DEVICES_READY:
2224                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2225                                             &btrfs_fs_type, &fs_devices);
2226                 if (ret)
2227                         break;
2228                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2229                 break;
2230         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2231                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2232                 break;
2233         }
2234
2235         kfree(vol);
2236         return ret;
2237 }
2238
2239 static int btrfs_freeze(struct super_block *sb)
2240 {
2241         struct btrfs_trans_handle *trans;
2242         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2243
2244         trans = btrfs_attach_transaction_barrier(root);
2245         if (IS_ERR(trans)) {
2246                 /* no transaction, don't bother */
2247                 if (PTR_ERR(trans) == -ENOENT)
2248                         return 0;
2249                 return PTR_ERR(trans);
2250         }
2251         return btrfs_commit_transaction(trans, root);
2252 }
2253
2254 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2255 {
2256         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2257         struct btrfs_fs_devices *cur_devices;
2258         struct btrfs_device *dev, *first_dev = NULL;
2259         struct list_head *head;
2260         struct rcu_string *name;
2261
2262         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2263         cur_devices = fs_info->fs_devices;
2264         while (cur_devices) {
2265                 head = &cur_devices->devices;
2266                 list_for_each_entry(dev, head, dev_list) {
2267                         if (dev->missing)
2268                                 continue;
2269                         if (!dev->name)
2270                                 continue;
2271                         if (!first_dev || dev->devid < first_dev->devid)
2272                                 first_dev = dev;
2273                 }
2274                 cur_devices = cur_devices->seed;
2275         }
2276
2277         if (first_dev) {
2278                 rcu_read_lock();
2279                 name = rcu_dereference(first_dev->name);
2280                 seq_escape(m, name->str, " \t\n\\");
2281                 rcu_read_unlock();
2282         } else {
2283                 WARN_ON(1);
2284         }
2285         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2286         return 0;
2287 }
2288
2289 static const struct super_operations btrfs_super_ops = {
2290         .drop_inode     = btrfs_drop_inode,
2291         .evict_inode    = btrfs_evict_inode,
2292         .put_super      = btrfs_put_super,
2293         .sync_fs        = btrfs_sync_fs,
2294         .show_options   = btrfs_show_options,
2295         .show_devname   = btrfs_show_devname,
2296         .write_inode    = btrfs_write_inode,
2297         .alloc_inode    = btrfs_alloc_inode,
2298         .destroy_inode  = btrfs_destroy_inode,
2299         .statfs         = btrfs_statfs,
2300         .remount_fs     = btrfs_remount,
2301         .freeze_fs      = btrfs_freeze,
2302 };
2303
2304 static const struct file_operations btrfs_ctl_fops = {
2305         .open = btrfs_control_open,
2306         .unlocked_ioctl  = btrfs_control_ioctl,
2307         .compat_ioctl = btrfs_control_ioctl,
2308         .owner   = THIS_MODULE,
2309         .llseek = noop_llseek,
2310 };
2311
2312 static struct miscdevice btrfs_misc = {
2313         .minor          = BTRFS_MINOR,
2314         .name           = "btrfs-control",
2315         .fops           = &btrfs_ctl_fops
2316 };
2317
2318 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2319 MODULE_ALIAS("devname:btrfs-control");
2320
2321 static int btrfs_interface_init(void)
2322 {
2323         return misc_register(&btrfs_misc);
2324 }
2325
2326 static void btrfs_interface_exit(void)
2327 {
2328         misc_deregister(&btrfs_misc);
2329 }
2330
2331 static void btrfs_print_mod_info(void)
2332 {
2333         printk(KERN_INFO "Btrfs loaded, crc32c=%s"
2334 #ifdef CONFIG_BTRFS_DEBUG
2335                         ", debug=on"
2336 #endif
2337 #ifdef CONFIG_BTRFS_ASSERT
2338                         ", assert=on"
2339 #endif
2340 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2341                         ", integrity-checker=on"
2342 #endif
2343                         "\n",
2344                         btrfs_crc32c_impl());
2345 }
2346
2347 static int __init init_btrfs_fs(void)
2348 {
2349         int err;
2350
2351         err = btrfs_hash_init();
2352         if (err)
2353                 return err;
2354
2355         btrfs_props_init();
2356
2357         err = btrfs_init_sysfs();
2358         if (err)
2359                 goto free_hash;
2360
2361         btrfs_init_compress();
2362
2363         err = btrfs_init_cachep();
2364         if (err)
2365                 goto free_compress;
2366
2367         err = extent_io_init();
2368         if (err)
2369                 goto free_cachep;
2370
2371         err = extent_map_init();
2372         if (err)
2373                 goto free_extent_io;
2374
2375         err = ordered_data_init();
2376         if (err)
2377                 goto free_extent_map;
2378
2379         err = btrfs_delayed_inode_init();
2380         if (err)
2381                 goto free_ordered_data;
2382
2383         err = btrfs_auto_defrag_init();
2384         if (err)
2385                 goto free_delayed_inode;
2386
2387         err = btrfs_delayed_ref_init();
2388         if (err)
2389                 goto free_auto_defrag;
2390
2391         err = btrfs_prelim_ref_init();
2392         if (err)
2393                 goto free_delayed_ref;
2394
2395         err = btrfs_end_io_wq_init();
2396         if (err)
2397                 goto free_prelim_ref;
2398
2399         err = btrfs_interface_init();
2400         if (err)
2401                 goto free_end_io_wq;
2402
2403         btrfs_init_lockdep();
2404
2405         btrfs_print_mod_info();
2406
2407         err = btrfs_run_sanity_tests();
2408         if (err)
2409                 goto unregister_ioctl;
2410
2411         err = register_filesystem(&btrfs_fs_type);
2412         if (err)
2413                 goto unregister_ioctl;
2414
2415         return 0;
2416
2417 unregister_ioctl:
2418         btrfs_interface_exit();
2419 free_end_io_wq:
2420         btrfs_end_io_wq_exit();
2421 free_prelim_ref:
2422         btrfs_prelim_ref_exit();
2423 free_delayed_ref:
2424         btrfs_delayed_ref_exit();
2425 free_auto_defrag:
2426         btrfs_auto_defrag_exit();
2427 free_delayed_inode:
2428         btrfs_delayed_inode_exit();
2429 free_ordered_data:
2430         ordered_data_exit();
2431 free_extent_map:
2432         extent_map_exit();
2433 free_extent_io:
2434         extent_io_exit();
2435 free_cachep:
2436         btrfs_destroy_cachep();
2437 free_compress:
2438         btrfs_exit_compress();
2439         btrfs_exit_sysfs();
2440 free_hash:
2441         btrfs_hash_exit();
2442         return err;
2443 }
2444
2445 static void __exit exit_btrfs_fs(void)
2446 {
2447         btrfs_destroy_cachep();
2448         btrfs_delayed_ref_exit();
2449         btrfs_auto_defrag_exit();
2450         btrfs_delayed_inode_exit();
2451         btrfs_prelim_ref_exit();
2452         ordered_data_exit();
2453         extent_map_exit();
2454         extent_io_exit();
2455         btrfs_interface_exit();
2456         btrfs_end_io_wq_exit();
2457         unregister_filesystem(&btrfs_fs_type);
2458         btrfs_exit_sysfs();
2459         btrfs_cleanup_fs_uuids();
2460         btrfs_exit_compress();
2461         btrfs_hash_exit();
2462 }
2463
2464 late_initcall(init_btrfs_fs);
2465 module_exit(exit_btrfs_fs)
2466
2467 MODULE_LICENSE("GPL");