Merge branch 'for-4.14/block-postmerge' of git://git.kernel.dk/linux-block
[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                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
155                         sb->s_id, function, line, errno, errstr, &vaf);
156                 va_end(args);
157         } else {
158                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
159                         sb->s_id, function, line, errno, errstr);
160         }
161 #endif
162
163         /*
164          * Today we only save the error info to memory.  Long term we'll
165          * also send it down to the disk
166          */
167         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
168
169         /* Don't go through full error handling during mount */
170         if (sb->s_flags & MS_BORN)
171                 btrfs_handle_error(fs_info);
172 }
173
174 #ifdef CONFIG_PRINTK
175 static const char * const logtypes[] = {
176         "emergency",
177         "alert",
178         "critical",
179         "error",
180         "warning",
181         "notice",
182         "info",
183         "debug",
184 };
185
186
187 /*
188  * Use one ratelimit state per log level so that a flood of less important
189  * messages doesn't cause more important ones to be dropped.
190  */
191 static struct ratelimit_state printk_limits[] = {
192         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
193         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
194         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
195         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
196         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
197         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
198         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
199         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
200 };
201
202 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
203 {
204         struct super_block *sb = fs_info->sb;
205         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
206         struct va_format vaf;
207         va_list args;
208         int kern_level;
209         const char *type = logtypes[4];
210         struct ratelimit_state *ratelimit = &printk_limits[4];
211
212         va_start(args, fmt);
213
214         while ((kern_level = printk_get_level(fmt)) != 0) {
215                 size_t size = printk_skip_level(fmt) - fmt;
216
217                 if (kern_level >= '0' && kern_level <= '7') {
218                         memcpy(lvl, fmt,  size);
219                         lvl[size] = '\0';
220                         type = logtypes[kern_level - '0'];
221                         ratelimit = &printk_limits[kern_level - '0'];
222                 }
223                 fmt += size;
224         }
225
226         vaf.fmt = fmt;
227         vaf.va = &args;
228
229         if (__ratelimit(ratelimit))
230                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
231
232         va_end(args);
233 }
234 #endif
235
236 /*
237  * We only mark the transaction aborted and then set the file system read-only.
238  * This will prevent new transactions from starting or trying to join this
239  * one.
240  *
241  * This means that error recovery at the call site is limited to freeing
242  * any local memory allocations and passing the error code up without
243  * further cleanup. The transaction should complete as it normally would
244  * in the call path but will return -EIO.
245  *
246  * We'll complete the cleanup in btrfs_end_transaction and
247  * btrfs_commit_transaction.
248  */
249 __cold
250 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
251                                const char *function,
252                                unsigned int line, int errno)
253 {
254         struct btrfs_fs_info *fs_info = trans->fs_info;
255
256         trans->aborted = errno;
257         /* Nothing used. The other threads that have joined this
258          * transaction may be able to continue. */
259         if (!trans->dirty && list_empty(&trans->new_bgs)) {
260                 const char *errstr;
261
262                 errstr = btrfs_decode_error(errno);
263                 btrfs_warn(fs_info,
264                            "%s:%d: Aborting unused transaction(%s).",
265                            function, line, errstr);
266                 return;
267         }
268         WRITE_ONCE(trans->transaction->aborted, errno);
269         /* Wake up anybody who may be waiting on this transaction */
270         wake_up(&fs_info->transaction_wait);
271         wake_up(&fs_info->transaction_blocked_wait);
272         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
273 }
274 /*
275  * __btrfs_panic decodes unexpected, fatal errors from the caller,
276  * issues an alert, and either panics or BUGs, depending on mount options.
277  */
278 __cold
279 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
280                    unsigned int line, int errno, const char *fmt, ...)
281 {
282         char *s_id = "<unknown>";
283         const char *errstr;
284         struct va_format vaf = { .fmt = fmt };
285         va_list args;
286
287         if (fs_info)
288                 s_id = fs_info->sb->s_id;
289
290         va_start(args, fmt);
291         vaf.va = &args;
292
293         errstr = btrfs_decode_error(errno);
294         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
295                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
296                         s_id, function, line, &vaf, errno, errstr);
297
298         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
299                    function, line, &vaf, errno, errstr);
300         va_end(args);
301         /* Caller calls BUG() */
302 }
303
304 static void btrfs_put_super(struct super_block *sb)
305 {
306         close_ctree(btrfs_sb(sb));
307 }
308
309 enum {
310         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
311         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
312         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
313         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
314         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
315         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
316         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
317         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
318         Opt_skip_balance, Opt_check_integrity,
319         Opt_check_integrity_including_extent_data,
320         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
321         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
322         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
323         Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
324         Opt_nologreplay, Opt_norecovery,
325 #ifdef CONFIG_BTRFS_DEBUG
326         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
327 #endif
328         Opt_err,
329 };
330
331 static const match_table_t tokens = {
332         {Opt_degraded, "degraded"},
333         {Opt_subvol, "subvol=%s"},
334         {Opt_subvolid, "subvolid=%s"},
335         {Opt_device, "device=%s"},
336         {Opt_nodatasum, "nodatasum"},
337         {Opt_datasum, "datasum"},
338         {Opt_nodatacow, "nodatacow"},
339         {Opt_datacow, "datacow"},
340         {Opt_nobarrier, "nobarrier"},
341         {Opt_barrier, "barrier"},
342         {Opt_max_inline, "max_inline=%s"},
343         {Opt_alloc_start, "alloc_start=%s"},
344         {Opt_thread_pool, "thread_pool=%d"},
345         {Opt_compress, "compress"},
346         {Opt_compress_type, "compress=%s"},
347         {Opt_compress_force, "compress-force"},
348         {Opt_compress_force_type, "compress-force=%s"},
349         {Opt_ssd, "ssd"},
350         {Opt_ssd_spread, "ssd_spread"},
351         {Opt_nossd, "nossd"},
352         {Opt_acl, "acl"},
353         {Opt_noacl, "noacl"},
354         {Opt_notreelog, "notreelog"},
355         {Opt_treelog, "treelog"},
356         {Opt_nologreplay, "nologreplay"},
357         {Opt_norecovery, "norecovery"},
358         {Opt_flushoncommit, "flushoncommit"},
359         {Opt_noflushoncommit, "noflushoncommit"},
360         {Opt_ratio, "metadata_ratio=%d"},
361         {Opt_discard, "discard"},
362         {Opt_nodiscard, "nodiscard"},
363         {Opt_space_cache, "space_cache"},
364         {Opt_space_cache_version, "space_cache=%s"},
365         {Opt_clear_cache, "clear_cache"},
366         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
367         {Opt_enospc_debug, "enospc_debug"},
368         {Opt_noenospc_debug, "noenospc_debug"},
369         {Opt_subvolrootid, "subvolrootid=%d"},
370         {Opt_defrag, "autodefrag"},
371         {Opt_nodefrag, "noautodefrag"},
372         {Opt_inode_cache, "inode_cache"},
373         {Opt_noinode_cache, "noinode_cache"},
374         {Opt_no_space_cache, "nospace_cache"},
375         {Opt_recovery, "recovery"}, /* deprecated */
376         {Opt_usebackuproot, "usebackuproot"},
377         {Opt_skip_balance, "skip_balance"},
378         {Opt_check_integrity, "check_int"},
379         {Opt_check_integrity_including_extent_data, "check_int_data"},
380         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
381         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
382         {Opt_fatal_errors, "fatal_errors=%s"},
383         {Opt_commit_interval, "commit=%d"},
384 #ifdef CONFIG_BTRFS_DEBUG
385         {Opt_fragment_data, "fragment=data"},
386         {Opt_fragment_metadata, "fragment=metadata"},
387         {Opt_fragment_all, "fragment=all"},
388 #endif
389         {Opt_err, NULL},
390 };
391
392 /*
393  * Regular mount options parser.  Everything that is needed only when
394  * reading in a new superblock is parsed here.
395  * XXX JDM: This needs to be cleaned up for remount.
396  */
397 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
398                         unsigned long new_flags)
399 {
400         substring_t args[MAX_OPT_ARGS];
401         char *p, *num, *orig = NULL;
402         u64 cache_gen;
403         int intarg;
404         int ret = 0;
405         char *compress_type;
406         bool compress_force = false;
407         enum btrfs_compression_type saved_compress_type;
408         bool saved_compress_force;
409         int no_compress = 0;
410
411         cache_gen = btrfs_super_cache_generation(info->super_copy);
412         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
413                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
414         else if (cache_gen)
415                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
416
417         /*
418          * Even the options are empty, we still need to do extra check
419          * against new flags
420          */
421         if (!options)
422                 goto check;
423
424         /*
425          * strsep changes the string, duplicate it because parse_options
426          * gets called twice
427          */
428         options = kstrdup(options, GFP_NOFS);
429         if (!options)
430                 return -ENOMEM;
431
432         orig = options;
433
434         while ((p = strsep(&options, ",")) != NULL) {
435                 int token;
436                 if (!*p)
437                         continue;
438
439                 token = match_token(p, tokens, args);
440                 switch (token) {
441                 case Opt_degraded:
442                         btrfs_info(info, "allowing degraded mounts");
443                         btrfs_set_opt(info->mount_opt, DEGRADED);
444                         break;
445                 case Opt_subvol:
446                 case Opt_subvolid:
447                 case Opt_subvolrootid:
448                 case Opt_device:
449                         /*
450                          * These are parsed by btrfs_parse_early_options
451                          * and can be happily ignored here.
452                          */
453                         break;
454                 case Opt_nodatasum:
455                         btrfs_set_and_info(info, NODATASUM,
456                                            "setting nodatasum");
457                         break;
458                 case Opt_datasum:
459                         if (btrfs_test_opt(info, NODATASUM)) {
460                                 if (btrfs_test_opt(info, NODATACOW))
461                                         btrfs_info(info,
462                                                    "setting datasum, datacow enabled");
463                                 else
464                                         btrfs_info(info, "setting datasum");
465                         }
466                         btrfs_clear_opt(info->mount_opt, NODATACOW);
467                         btrfs_clear_opt(info->mount_opt, NODATASUM);
468                         break;
469                 case Opt_nodatacow:
470                         if (!btrfs_test_opt(info, NODATACOW)) {
471                                 if (!btrfs_test_opt(info, COMPRESS) ||
472                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
473                                         btrfs_info(info,
474                                                    "setting nodatacow, compression disabled");
475                                 } else {
476                                         btrfs_info(info, "setting nodatacow");
477                                 }
478                         }
479                         btrfs_clear_opt(info->mount_opt, COMPRESS);
480                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
481                         btrfs_set_opt(info->mount_opt, NODATACOW);
482                         btrfs_set_opt(info->mount_opt, NODATASUM);
483                         break;
484                 case Opt_datacow:
485                         btrfs_clear_and_info(info, NODATACOW,
486                                              "setting datacow");
487                         break;
488                 case Opt_compress_force:
489                 case Opt_compress_force_type:
490                         compress_force = true;
491                         /* Fallthrough */
492                 case Opt_compress:
493                 case Opt_compress_type:
494                         saved_compress_type = btrfs_test_opt(info,
495                                                              COMPRESS) ?
496                                 info->compress_type : BTRFS_COMPRESS_NONE;
497                         saved_compress_force =
498                                 btrfs_test_opt(info, FORCE_COMPRESS);
499                         if (token == Opt_compress ||
500                             token == Opt_compress_force ||
501                             strcmp(args[0].from, "zlib") == 0) {
502                                 compress_type = "zlib";
503                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
504                                 btrfs_set_opt(info->mount_opt, COMPRESS);
505                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
506                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
507                                 no_compress = 0;
508                         } else if (strcmp(args[0].from, "lzo") == 0) {
509                                 compress_type = "lzo";
510                                 info->compress_type = BTRFS_COMPRESS_LZO;
511                                 btrfs_set_opt(info->mount_opt, COMPRESS);
512                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
513                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
514                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
515                                 no_compress = 0;
516                         } else if (strncmp(args[0].from, "no", 2) == 0) {
517                                 compress_type = "no";
518                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
519                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
520                                 compress_force = false;
521                                 no_compress++;
522                         } else {
523                                 ret = -EINVAL;
524                                 goto out;
525                         }
526
527                         if (compress_force) {
528                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
529                         } else {
530                                 /*
531                                  * If we remount from compress-force=xxx to
532                                  * compress=xxx, we need clear FORCE_COMPRESS
533                                  * flag, otherwise, there is no way for users
534                                  * to disable forcible compression separately.
535                                  */
536                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
537                         }
538                         if ((btrfs_test_opt(info, COMPRESS) &&
539                              (info->compress_type != saved_compress_type ||
540                               compress_force != saved_compress_force)) ||
541                             (!btrfs_test_opt(info, COMPRESS) &&
542                              no_compress == 1)) {
543                                 btrfs_info(info, "%s %s compression",
544                                            (compress_force) ? "force" : "use",
545                                            compress_type);
546                         }
547                         compress_force = false;
548                         break;
549                 case Opt_ssd:
550                         btrfs_set_and_info(info, SSD,
551                                            "use ssd allocation scheme");
552                         btrfs_clear_opt(info->mount_opt, NOSSD);
553                         break;
554                 case Opt_ssd_spread:
555                         btrfs_set_and_info(info, SSD_SPREAD,
556                                            "use spread ssd allocation scheme");
557                         btrfs_set_opt(info->mount_opt, SSD);
558                         btrfs_clear_opt(info->mount_opt, NOSSD);
559                         break;
560                 case Opt_nossd:
561                         btrfs_set_and_info(info, NOSSD,
562                                              "not using ssd allocation scheme");
563                         btrfs_clear_opt(info->mount_opt, SSD);
564                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
565                         break;
566                 case Opt_barrier:
567                         btrfs_clear_and_info(info, NOBARRIER,
568                                              "turning on barriers");
569                         break;
570                 case Opt_nobarrier:
571                         btrfs_set_and_info(info, NOBARRIER,
572                                            "turning off barriers");
573                         break;
574                 case Opt_thread_pool:
575                         ret = match_int(&args[0], &intarg);
576                         if (ret) {
577                                 goto out;
578                         } else if (intarg > 0) {
579                                 info->thread_pool_size = intarg;
580                         } else {
581                                 ret = -EINVAL;
582                                 goto out;
583                         }
584                         break;
585                 case Opt_max_inline:
586                         num = match_strdup(&args[0]);
587                         if (num) {
588                                 info->max_inline = memparse(num, NULL);
589                                 kfree(num);
590
591                                 if (info->max_inline) {
592                                         info->max_inline = min_t(u64,
593                                                 info->max_inline,
594                                                 info->sectorsize);
595                                 }
596                                 btrfs_info(info, "max_inline at %llu",
597                                            info->max_inline);
598                         } else {
599                                 ret = -ENOMEM;
600                                 goto out;
601                         }
602                         break;
603                 case Opt_alloc_start:
604                         btrfs_info(info,
605                                 "option alloc_start is obsolete, ignored");
606                         break;
607                 case Opt_acl:
608 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
609                         info->sb->s_flags |= MS_POSIXACL;
610                         break;
611 #else
612                         btrfs_err(info, "support for ACL not compiled in!");
613                         ret = -EINVAL;
614                         goto out;
615 #endif
616                 case Opt_noacl:
617                         info->sb->s_flags &= ~MS_POSIXACL;
618                         break;
619                 case Opt_notreelog:
620                         btrfs_set_and_info(info, NOTREELOG,
621                                            "disabling tree log");
622                         break;
623                 case Opt_treelog:
624                         btrfs_clear_and_info(info, NOTREELOG,
625                                              "enabling tree log");
626                         break;
627                 case Opt_norecovery:
628                 case Opt_nologreplay:
629                         btrfs_set_and_info(info, NOLOGREPLAY,
630                                            "disabling log replay at mount time");
631                         break;
632                 case Opt_flushoncommit:
633                         btrfs_set_and_info(info, FLUSHONCOMMIT,
634                                            "turning on flush-on-commit");
635                         break;
636                 case Opt_noflushoncommit:
637                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
638                                              "turning off flush-on-commit");
639                         break;
640                 case Opt_ratio:
641                         ret = match_int(&args[0], &intarg);
642                         if (ret) {
643                                 goto out;
644                         } else if (intarg >= 0) {
645                                 info->metadata_ratio = intarg;
646                                 btrfs_info(info, "metadata ratio %d",
647                                            info->metadata_ratio);
648                         } else {
649                                 ret = -EINVAL;
650                                 goto out;
651                         }
652                         break;
653                 case Opt_discard:
654                         btrfs_set_and_info(info, DISCARD,
655                                            "turning on discard");
656                         break;
657                 case Opt_nodiscard:
658                         btrfs_clear_and_info(info, DISCARD,
659                                              "turning off discard");
660                         break;
661                 case Opt_space_cache:
662                 case Opt_space_cache_version:
663                         if (token == Opt_space_cache ||
664                             strcmp(args[0].from, "v1") == 0) {
665                                 btrfs_clear_opt(info->mount_opt,
666                                                 FREE_SPACE_TREE);
667                                 btrfs_set_and_info(info, SPACE_CACHE,
668                                            "enabling disk space caching");
669                         } else if (strcmp(args[0].from, "v2") == 0) {
670                                 btrfs_clear_opt(info->mount_opt,
671                                                 SPACE_CACHE);
672                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
673                                                    "enabling free space tree");
674                         } else {
675                                 ret = -EINVAL;
676                                 goto out;
677                         }
678                         break;
679                 case Opt_rescan_uuid_tree:
680                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
681                         break;
682                 case Opt_no_space_cache:
683                         if (btrfs_test_opt(info, SPACE_CACHE)) {
684                                 btrfs_clear_and_info(info, SPACE_CACHE,
685                                              "disabling disk space caching");
686                         }
687                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
688                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
689                                              "disabling free space tree");
690                         }
691                         break;
692                 case Opt_inode_cache:
693                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
694                                            "enabling inode map caching");
695                         break;
696                 case Opt_noinode_cache:
697                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
698                                              "disabling inode map caching");
699                         break;
700                 case Opt_clear_cache:
701                         btrfs_set_and_info(info, CLEAR_CACHE,
702                                            "force clearing of disk cache");
703                         break;
704                 case Opt_user_subvol_rm_allowed:
705                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
706                         break;
707                 case Opt_enospc_debug:
708                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
709                         break;
710                 case Opt_noenospc_debug:
711                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
712                         break;
713                 case Opt_defrag:
714                         btrfs_set_and_info(info, AUTO_DEFRAG,
715                                            "enabling auto defrag");
716                         break;
717                 case Opt_nodefrag:
718                         btrfs_clear_and_info(info, AUTO_DEFRAG,
719                                              "disabling auto defrag");
720                         break;
721                 case Opt_recovery:
722                         btrfs_warn(info,
723                                    "'recovery' is deprecated, use 'usebackuproot' instead");
724                 case Opt_usebackuproot:
725                         btrfs_info(info,
726                                    "trying to use backup root at mount time");
727                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
728                         break;
729                 case Opt_skip_balance:
730                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
731                         break;
732 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
733                 case Opt_check_integrity_including_extent_data:
734                         btrfs_info(info,
735                                    "enabling check integrity including extent data");
736                         btrfs_set_opt(info->mount_opt,
737                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
738                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
739                         break;
740                 case Opt_check_integrity:
741                         btrfs_info(info, "enabling check integrity");
742                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
743                         break;
744                 case Opt_check_integrity_print_mask:
745                         ret = match_int(&args[0], &intarg);
746                         if (ret) {
747                                 goto out;
748                         } else if (intarg >= 0) {
749                                 info->check_integrity_print_mask = intarg;
750                                 btrfs_info(info,
751                                            "check_integrity_print_mask 0x%x",
752                                            info->check_integrity_print_mask);
753                         } else {
754                                 ret = -EINVAL;
755                                 goto out;
756                         }
757                         break;
758 #else
759                 case Opt_check_integrity_including_extent_data:
760                 case Opt_check_integrity:
761                 case Opt_check_integrity_print_mask:
762                         btrfs_err(info,
763                                   "support for check_integrity* not compiled in!");
764                         ret = -EINVAL;
765                         goto out;
766 #endif
767                 case Opt_fatal_errors:
768                         if (strcmp(args[0].from, "panic") == 0)
769                                 btrfs_set_opt(info->mount_opt,
770                                               PANIC_ON_FATAL_ERROR);
771                         else if (strcmp(args[0].from, "bug") == 0)
772                                 btrfs_clear_opt(info->mount_opt,
773                                               PANIC_ON_FATAL_ERROR);
774                         else {
775                                 ret = -EINVAL;
776                                 goto out;
777                         }
778                         break;
779                 case Opt_commit_interval:
780                         intarg = 0;
781                         ret = match_int(&args[0], &intarg);
782                         if (ret < 0) {
783                                 btrfs_err(info, "invalid commit interval");
784                                 ret = -EINVAL;
785                                 goto out;
786                         }
787                         if (intarg > 0) {
788                                 if (intarg > 300) {
789                                         btrfs_warn(info,
790                                                 "excessive commit interval %d",
791                                                 intarg);
792                                 }
793                                 info->commit_interval = intarg;
794                         } else {
795                                 btrfs_info(info,
796                                            "using default commit interval %ds",
797                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
798                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
799                         }
800                         break;
801 #ifdef CONFIG_BTRFS_DEBUG
802                 case Opt_fragment_all:
803                         btrfs_info(info, "fragmenting all space");
804                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
805                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
806                         break;
807                 case Opt_fragment_metadata:
808                         btrfs_info(info, "fragmenting metadata");
809                         btrfs_set_opt(info->mount_opt,
810                                       FRAGMENT_METADATA);
811                         break;
812                 case Opt_fragment_data:
813                         btrfs_info(info, "fragmenting data");
814                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
815                         break;
816 #endif
817                 case Opt_err:
818                         btrfs_info(info, "unrecognized mount option '%s'", p);
819                         ret = -EINVAL;
820                         goto out;
821                 default:
822                         break;
823                 }
824         }
825 check:
826         /*
827          * Extra check for current option against current flag
828          */
829         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
830                 btrfs_err(info,
831                           "nologreplay must be used with ro mount option");
832                 ret = -EINVAL;
833         }
834 out:
835         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
836             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
837             !btrfs_test_opt(info, CLEAR_CACHE)) {
838                 btrfs_err(info, "cannot disable free space tree");
839                 ret = -EINVAL;
840
841         }
842         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
843                 btrfs_info(info, "disk space caching is enabled");
844         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
845                 btrfs_info(info, "using free space tree");
846         kfree(orig);
847         return ret;
848 }
849
850 /*
851  * Parse mount options that are required early in the mount process.
852  *
853  * All other options will be parsed on much later in the mount process and
854  * only when we need to allocate a new super block.
855  */
856 static int btrfs_parse_early_options(const char *options, fmode_t flags,
857                 void *holder, char **subvol_name, u64 *subvol_objectid,
858                 struct btrfs_fs_devices **fs_devices)
859 {
860         substring_t args[MAX_OPT_ARGS];
861         char *device_name, *opts, *orig, *p;
862         char *num = NULL;
863         int error = 0;
864
865         if (!options)
866                 return 0;
867
868         /*
869          * strsep changes the string, duplicate it because parse_options
870          * gets called twice
871          */
872         opts = kstrdup(options, GFP_KERNEL);
873         if (!opts)
874                 return -ENOMEM;
875         orig = opts;
876
877         while ((p = strsep(&opts, ",")) != NULL) {
878                 int token;
879                 if (!*p)
880                         continue;
881
882                 token = match_token(p, tokens, args);
883                 switch (token) {
884                 case Opt_subvol:
885                         kfree(*subvol_name);
886                         *subvol_name = match_strdup(&args[0]);
887                         if (!*subvol_name) {
888                                 error = -ENOMEM;
889                                 goto out;
890                         }
891                         break;
892                 case Opt_subvolid:
893                         num = match_strdup(&args[0]);
894                         if (num) {
895                                 *subvol_objectid = memparse(num, NULL);
896                                 kfree(num);
897                                 /* we want the original fs_tree */
898                                 if (!*subvol_objectid)
899                                         *subvol_objectid =
900                                                 BTRFS_FS_TREE_OBJECTID;
901                         } else {
902                                 error = -EINVAL;
903                                 goto out;
904                         }
905                         break;
906                 case Opt_subvolrootid:
907                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
908                         break;
909                 case Opt_device:
910                         device_name = match_strdup(&args[0]);
911                         if (!device_name) {
912                                 error = -ENOMEM;
913                                 goto out;
914                         }
915                         error = btrfs_scan_one_device(device_name,
916                                         flags, holder, fs_devices);
917                         kfree(device_name);
918                         if (error)
919                                 goto out;
920                         break;
921                 default:
922                         break;
923                 }
924         }
925
926 out:
927         kfree(orig);
928         return error;
929 }
930
931 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
932                                            u64 subvol_objectid)
933 {
934         struct btrfs_root *root = fs_info->tree_root;
935         struct btrfs_root *fs_root;
936         struct btrfs_root_ref *root_ref;
937         struct btrfs_inode_ref *inode_ref;
938         struct btrfs_key key;
939         struct btrfs_path *path = NULL;
940         char *name = NULL, *ptr;
941         u64 dirid;
942         int len;
943         int ret;
944
945         path = btrfs_alloc_path();
946         if (!path) {
947                 ret = -ENOMEM;
948                 goto err;
949         }
950         path->leave_spinning = 1;
951
952         name = kmalloc(PATH_MAX, GFP_NOFS);
953         if (!name) {
954                 ret = -ENOMEM;
955                 goto err;
956         }
957         ptr = name + PATH_MAX - 1;
958         ptr[0] = '\0';
959
960         /*
961          * Walk up the subvolume trees in the tree of tree roots by root
962          * backrefs until we hit the top-level subvolume.
963          */
964         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
965                 key.objectid = subvol_objectid;
966                 key.type = BTRFS_ROOT_BACKREF_KEY;
967                 key.offset = (u64)-1;
968
969                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
970                 if (ret < 0) {
971                         goto err;
972                 } else if (ret > 0) {
973                         ret = btrfs_previous_item(root, path, subvol_objectid,
974                                                   BTRFS_ROOT_BACKREF_KEY);
975                         if (ret < 0) {
976                                 goto err;
977                         } else if (ret > 0) {
978                                 ret = -ENOENT;
979                                 goto err;
980                         }
981                 }
982
983                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
984                 subvol_objectid = key.offset;
985
986                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
987                                           struct btrfs_root_ref);
988                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
989                 ptr -= len + 1;
990                 if (ptr < name) {
991                         ret = -ENAMETOOLONG;
992                         goto err;
993                 }
994                 read_extent_buffer(path->nodes[0], ptr + 1,
995                                    (unsigned long)(root_ref + 1), len);
996                 ptr[0] = '/';
997                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
998                 btrfs_release_path(path);
999
1000                 key.objectid = subvol_objectid;
1001                 key.type = BTRFS_ROOT_ITEM_KEY;
1002                 key.offset = (u64)-1;
1003                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1004                 if (IS_ERR(fs_root)) {
1005                         ret = PTR_ERR(fs_root);
1006                         goto err;
1007                 }
1008
1009                 /*
1010                  * Walk up the filesystem tree by inode refs until we hit the
1011                  * root directory.
1012                  */
1013                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1014                         key.objectid = dirid;
1015                         key.type = BTRFS_INODE_REF_KEY;
1016                         key.offset = (u64)-1;
1017
1018                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1019                         if (ret < 0) {
1020                                 goto err;
1021                         } else if (ret > 0) {
1022                                 ret = btrfs_previous_item(fs_root, path, dirid,
1023                                                           BTRFS_INODE_REF_KEY);
1024                                 if (ret < 0) {
1025                                         goto err;
1026                                 } else if (ret > 0) {
1027                                         ret = -ENOENT;
1028                                         goto err;
1029                                 }
1030                         }
1031
1032                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1033                         dirid = key.offset;
1034
1035                         inode_ref = btrfs_item_ptr(path->nodes[0],
1036                                                    path->slots[0],
1037                                                    struct btrfs_inode_ref);
1038                         len = btrfs_inode_ref_name_len(path->nodes[0],
1039                                                        inode_ref);
1040                         ptr -= len + 1;
1041                         if (ptr < name) {
1042                                 ret = -ENAMETOOLONG;
1043                                 goto err;
1044                         }
1045                         read_extent_buffer(path->nodes[0], ptr + 1,
1046                                            (unsigned long)(inode_ref + 1), len);
1047                         ptr[0] = '/';
1048                         btrfs_release_path(path);
1049                 }
1050         }
1051
1052         btrfs_free_path(path);
1053         if (ptr == name + PATH_MAX - 1) {
1054                 name[0] = '/';
1055                 name[1] = '\0';
1056         } else {
1057                 memmove(name, ptr, name + PATH_MAX - ptr);
1058         }
1059         return name;
1060
1061 err:
1062         btrfs_free_path(path);
1063         kfree(name);
1064         return ERR_PTR(ret);
1065 }
1066
1067 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1068 {
1069         struct btrfs_root *root = fs_info->tree_root;
1070         struct btrfs_dir_item *di;
1071         struct btrfs_path *path;
1072         struct btrfs_key location;
1073         u64 dir_id;
1074
1075         path = btrfs_alloc_path();
1076         if (!path)
1077                 return -ENOMEM;
1078         path->leave_spinning = 1;
1079
1080         /*
1081          * Find the "default" dir item which points to the root item that we
1082          * will mount by default if we haven't been given a specific subvolume
1083          * to mount.
1084          */
1085         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1086         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1087         if (IS_ERR(di)) {
1088                 btrfs_free_path(path);
1089                 return PTR_ERR(di);
1090         }
1091         if (!di) {
1092                 /*
1093                  * Ok the default dir item isn't there.  This is weird since
1094                  * it's always been there, but don't freak out, just try and
1095                  * mount the top-level subvolume.
1096                  */
1097                 btrfs_free_path(path);
1098                 *objectid = BTRFS_FS_TREE_OBJECTID;
1099                 return 0;
1100         }
1101
1102         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1103         btrfs_free_path(path);
1104         *objectid = location.objectid;
1105         return 0;
1106 }
1107
1108 static int btrfs_fill_super(struct super_block *sb,
1109                             struct btrfs_fs_devices *fs_devices,
1110                             void *data)
1111 {
1112         struct inode *inode;
1113         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1114         struct btrfs_key key;
1115         int err;
1116
1117         sb->s_maxbytes = MAX_LFS_FILESIZE;
1118         sb->s_magic = BTRFS_SUPER_MAGIC;
1119         sb->s_op = &btrfs_super_ops;
1120         sb->s_d_op = &btrfs_dentry_operations;
1121         sb->s_export_op = &btrfs_export_ops;
1122         sb->s_xattr = btrfs_xattr_handlers;
1123         sb->s_time_gran = 1;
1124 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1125         sb->s_flags |= MS_POSIXACL;
1126 #endif
1127         sb->s_flags |= MS_I_VERSION;
1128         sb->s_iflags |= SB_I_CGROUPWB;
1129
1130         err = super_setup_bdi(sb);
1131         if (err) {
1132                 btrfs_err(fs_info, "super_setup_bdi failed");
1133                 return err;
1134         }
1135
1136         err = open_ctree(sb, fs_devices, (char *)data);
1137         if (err) {
1138                 btrfs_err(fs_info, "open_ctree failed");
1139                 return err;
1140         }
1141
1142         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1143         key.type = BTRFS_INODE_ITEM_KEY;
1144         key.offset = 0;
1145         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1146         if (IS_ERR(inode)) {
1147                 err = PTR_ERR(inode);
1148                 goto fail_close;
1149         }
1150
1151         sb->s_root = d_make_root(inode);
1152         if (!sb->s_root) {
1153                 err = -ENOMEM;
1154                 goto fail_close;
1155         }
1156
1157         cleancache_init_fs(sb);
1158         sb->s_flags |= MS_ACTIVE;
1159         return 0;
1160
1161 fail_close:
1162         close_ctree(fs_info);
1163         return err;
1164 }
1165
1166 int btrfs_sync_fs(struct super_block *sb, int wait)
1167 {
1168         struct btrfs_trans_handle *trans;
1169         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1170         struct btrfs_root *root = fs_info->tree_root;
1171
1172         trace_btrfs_sync_fs(fs_info, wait);
1173
1174         if (!wait) {
1175                 filemap_flush(fs_info->btree_inode->i_mapping);
1176                 return 0;
1177         }
1178
1179         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1180
1181         trans = btrfs_attach_transaction_barrier(root);
1182         if (IS_ERR(trans)) {
1183                 /* no transaction, don't bother */
1184                 if (PTR_ERR(trans) == -ENOENT) {
1185                         /*
1186                          * Exit unless we have some pending changes
1187                          * that need to go through commit
1188                          */
1189                         if (fs_info->pending_changes == 0)
1190                                 return 0;
1191                         /*
1192                          * A non-blocking test if the fs is frozen. We must not
1193                          * start a new transaction here otherwise a deadlock
1194                          * happens. The pending operations are delayed to the
1195                          * next commit after thawing.
1196                          */
1197                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1198                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1199                         else
1200                                 return 0;
1201                         trans = btrfs_start_transaction(root, 0);
1202                 }
1203                 if (IS_ERR(trans))
1204                         return PTR_ERR(trans);
1205         }
1206         return btrfs_commit_transaction(trans);
1207 }
1208
1209 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1210 {
1211         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1212         char *compress_type;
1213
1214         if (btrfs_test_opt(info, DEGRADED))
1215                 seq_puts(seq, ",degraded");
1216         if (btrfs_test_opt(info, NODATASUM))
1217                 seq_puts(seq, ",nodatasum");
1218         if (btrfs_test_opt(info, NODATACOW))
1219                 seq_puts(seq, ",nodatacow");
1220         if (btrfs_test_opt(info, NOBARRIER))
1221                 seq_puts(seq, ",nobarrier");
1222         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1223                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1224         if (info->thread_pool_size !=  min_t(unsigned long,
1225                                              num_online_cpus() + 2, 8))
1226                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1227         if (btrfs_test_opt(info, COMPRESS)) {
1228                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1229                         compress_type = "zlib";
1230                 else
1231                         compress_type = "lzo";
1232                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1233                         seq_printf(seq, ",compress-force=%s", compress_type);
1234                 else
1235                         seq_printf(seq, ",compress=%s", compress_type);
1236         }
1237         if (btrfs_test_opt(info, NOSSD))
1238                 seq_puts(seq, ",nossd");
1239         if (btrfs_test_opt(info, SSD_SPREAD))
1240                 seq_puts(seq, ",ssd_spread");
1241         else if (btrfs_test_opt(info, SSD))
1242                 seq_puts(seq, ",ssd");
1243         if (btrfs_test_opt(info, NOTREELOG))
1244                 seq_puts(seq, ",notreelog");
1245         if (btrfs_test_opt(info, NOLOGREPLAY))
1246                 seq_puts(seq, ",nologreplay");
1247         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1248                 seq_puts(seq, ",flushoncommit");
1249         if (btrfs_test_opt(info, DISCARD))
1250                 seq_puts(seq, ",discard");
1251         if (!(info->sb->s_flags & MS_POSIXACL))
1252                 seq_puts(seq, ",noacl");
1253         if (btrfs_test_opt(info, SPACE_CACHE))
1254                 seq_puts(seq, ",space_cache");
1255         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1256                 seq_puts(seq, ",space_cache=v2");
1257         else
1258                 seq_puts(seq, ",nospace_cache");
1259         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1260                 seq_puts(seq, ",rescan_uuid_tree");
1261         if (btrfs_test_opt(info, CLEAR_CACHE))
1262                 seq_puts(seq, ",clear_cache");
1263         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1264                 seq_puts(seq, ",user_subvol_rm_allowed");
1265         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1266                 seq_puts(seq, ",enospc_debug");
1267         if (btrfs_test_opt(info, AUTO_DEFRAG))
1268                 seq_puts(seq, ",autodefrag");
1269         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1270                 seq_puts(seq, ",inode_cache");
1271         if (btrfs_test_opt(info, SKIP_BALANCE))
1272                 seq_puts(seq, ",skip_balance");
1273 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1274         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1275                 seq_puts(seq, ",check_int_data");
1276         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1277                 seq_puts(seq, ",check_int");
1278         if (info->check_integrity_print_mask)
1279                 seq_printf(seq, ",check_int_print_mask=%d",
1280                                 info->check_integrity_print_mask);
1281 #endif
1282         if (info->metadata_ratio)
1283                 seq_printf(seq, ",metadata_ratio=%d",
1284                                 info->metadata_ratio);
1285         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1286                 seq_puts(seq, ",fatal_errors=panic");
1287         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1288                 seq_printf(seq, ",commit=%d", info->commit_interval);
1289 #ifdef CONFIG_BTRFS_DEBUG
1290         if (btrfs_test_opt(info, FRAGMENT_DATA))
1291                 seq_puts(seq, ",fragment=data");
1292         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1293                 seq_puts(seq, ",fragment=metadata");
1294 #endif
1295         seq_printf(seq, ",subvolid=%llu",
1296                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1297         seq_puts(seq, ",subvol=");
1298         seq_dentry(seq, dentry, " \t\n\\");
1299         return 0;
1300 }
1301
1302 static int btrfs_test_super(struct super_block *s, void *data)
1303 {
1304         struct btrfs_fs_info *p = data;
1305         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1306
1307         return fs_info->fs_devices == p->fs_devices;
1308 }
1309
1310 static int btrfs_set_super(struct super_block *s, void *data)
1311 {
1312         int err = set_anon_super(s, data);
1313         if (!err)
1314                 s->s_fs_info = data;
1315         return err;
1316 }
1317
1318 /*
1319  * subvolumes are identified by ino 256
1320  */
1321 static inline int is_subvolume_inode(struct inode *inode)
1322 {
1323         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1324                 return 1;
1325         return 0;
1326 }
1327
1328 /*
1329  * This will add subvolid=0 to the argument string while removing any subvol=
1330  * and subvolid= arguments to make sure we get the top-level root for path
1331  * walking to the subvol we want.
1332  */
1333 static char *setup_root_args(char *args)
1334 {
1335         char *buf, *dst, *sep;
1336
1337         if (!args)
1338                 return kstrdup("subvolid=0", GFP_NOFS);
1339
1340         /* The worst case is that we add ",subvolid=0" to the end. */
1341         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1342         if (!buf)
1343                 return NULL;
1344
1345         while (1) {
1346                 sep = strchrnul(args, ',');
1347                 if (!strstarts(args, "subvol=") &&
1348                     !strstarts(args, "subvolid=")) {
1349                         memcpy(dst, args, sep - args);
1350                         dst += sep - args;
1351                         *dst++ = ',';
1352                 }
1353                 if (*sep)
1354                         args = sep + 1;
1355                 else
1356                         break;
1357         }
1358         strcpy(dst, "subvolid=0");
1359
1360         return buf;
1361 }
1362
1363 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1364                                    int flags, const char *device_name,
1365                                    char *data)
1366 {
1367         struct dentry *root;
1368         struct vfsmount *mnt = NULL;
1369         char *newargs;
1370         int ret;
1371
1372         newargs = setup_root_args(data);
1373         if (!newargs) {
1374                 root = ERR_PTR(-ENOMEM);
1375                 goto out;
1376         }
1377
1378         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1379         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1380                 if (flags & MS_RDONLY) {
1381                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1382                                              device_name, newargs);
1383                 } else {
1384                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1385                                              device_name, newargs);
1386                         if (IS_ERR(mnt)) {
1387                                 root = ERR_CAST(mnt);
1388                                 mnt = NULL;
1389                                 goto out;
1390                         }
1391
1392                         down_write(&mnt->mnt_sb->s_umount);
1393                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1394                         up_write(&mnt->mnt_sb->s_umount);
1395                         if (ret < 0) {
1396                                 root = ERR_PTR(ret);
1397                                 goto out;
1398                         }
1399                 }
1400         }
1401         if (IS_ERR(mnt)) {
1402                 root = ERR_CAST(mnt);
1403                 mnt = NULL;
1404                 goto out;
1405         }
1406
1407         if (!subvol_name) {
1408                 if (!subvol_objectid) {
1409                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1410                                                           &subvol_objectid);
1411                         if (ret) {
1412                                 root = ERR_PTR(ret);
1413                                 goto out;
1414                         }
1415                 }
1416                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1417                                                             subvol_objectid);
1418                 if (IS_ERR(subvol_name)) {
1419                         root = ERR_CAST(subvol_name);
1420                         subvol_name = NULL;
1421                         goto out;
1422                 }
1423
1424         }
1425
1426         root = mount_subtree(mnt, subvol_name);
1427         /* mount_subtree() drops our reference on the vfsmount. */
1428         mnt = NULL;
1429
1430         if (!IS_ERR(root)) {
1431                 struct super_block *s = root->d_sb;
1432                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1433                 struct inode *root_inode = d_inode(root);
1434                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1435
1436                 ret = 0;
1437                 if (!is_subvolume_inode(root_inode)) {
1438                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1439                                subvol_name);
1440                         ret = -EINVAL;
1441                 }
1442                 if (subvol_objectid && root_objectid != subvol_objectid) {
1443                         /*
1444                          * This will also catch a race condition where a
1445                          * subvolume which was passed by ID is renamed and
1446                          * another subvolume is renamed over the old location.
1447                          */
1448                         btrfs_err(fs_info,
1449                                   "subvol '%s' does not match subvolid %llu",
1450                                   subvol_name, subvol_objectid);
1451                         ret = -EINVAL;
1452                 }
1453                 if (ret) {
1454                         dput(root);
1455                         root = ERR_PTR(ret);
1456                         deactivate_locked_super(s);
1457                 }
1458         }
1459
1460 out:
1461         mntput(mnt);
1462         kfree(newargs);
1463         kfree(subvol_name);
1464         return root;
1465 }
1466
1467 static int parse_security_options(char *orig_opts,
1468                                   struct security_mnt_opts *sec_opts)
1469 {
1470         char *secdata = NULL;
1471         int ret = 0;
1472
1473         secdata = alloc_secdata();
1474         if (!secdata)
1475                 return -ENOMEM;
1476         ret = security_sb_copy_data(orig_opts, secdata);
1477         if (ret) {
1478                 free_secdata(secdata);
1479                 return ret;
1480         }
1481         ret = security_sb_parse_opts_str(secdata, sec_opts);
1482         free_secdata(secdata);
1483         return ret;
1484 }
1485
1486 static int setup_security_options(struct btrfs_fs_info *fs_info,
1487                                   struct super_block *sb,
1488                                   struct security_mnt_opts *sec_opts)
1489 {
1490         int ret = 0;
1491
1492         /*
1493          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1494          * is valid.
1495          */
1496         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1497         if (ret)
1498                 return ret;
1499
1500 #ifdef CONFIG_SECURITY
1501         if (!fs_info->security_opts.num_mnt_opts) {
1502                 /* first time security setup, copy sec_opts to fs_info */
1503                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1504         } else {
1505                 /*
1506                  * Since SELinux (the only one supporting security_mnt_opts)
1507                  * does NOT support changing context during remount/mount of
1508                  * the same sb, this must be the same or part of the same
1509                  * security options, just free it.
1510                  */
1511                 security_free_mnt_opts(sec_opts);
1512         }
1513 #endif
1514         return ret;
1515 }
1516
1517 /*
1518  * Find a superblock for the given device / mount point.
1519  *
1520  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1521  *        for multiple device setup.  Make sure to keep it in sync.
1522  */
1523 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1524                 const char *device_name, void *data)
1525 {
1526         struct block_device *bdev = NULL;
1527         struct super_block *s;
1528         struct btrfs_fs_devices *fs_devices = NULL;
1529         struct btrfs_fs_info *fs_info = NULL;
1530         struct security_mnt_opts new_sec_opts;
1531         fmode_t mode = FMODE_READ;
1532         char *subvol_name = NULL;
1533         u64 subvol_objectid = 0;
1534         int error = 0;
1535
1536         if (!(flags & MS_RDONLY))
1537                 mode |= FMODE_WRITE;
1538
1539         error = btrfs_parse_early_options(data, mode, fs_type,
1540                                           &subvol_name, &subvol_objectid,
1541                                           &fs_devices);
1542         if (error) {
1543                 kfree(subvol_name);
1544                 return ERR_PTR(error);
1545         }
1546
1547         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1548                 /* mount_subvol() will free subvol_name. */
1549                 return mount_subvol(subvol_name, subvol_objectid, flags,
1550                                     device_name, data);
1551         }
1552
1553         security_init_mnt_opts(&new_sec_opts);
1554         if (data) {
1555                 error = parse_security_options(data, &new_sec_opts);
1556                 if (error)
1557                         return ERR_PTR(error);
1558         }
1559
1560         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1561         if (error)
1562                 goto error_sec_opts;
1563
1564         /*
1565          * Setup a dummy root and fs_info for test/set super.  This is because
1566          * we don't actually fill this stuff out until open_ctree, but we need
1567          * it for searching for existing supers, so this lets us do that and
1568          * then open_ctree will properly initialize everything later.
1569          */
1570         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1571         if (!fs_info) {
1572                 error = -ENOMEM;
1573                 goto error_sec_opts;
1574         }
1575
1576         fs_info->fs_devices = fs_devices;
1577
1578         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1579         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1580         security_init_mnt_opts(&fs_info->security_opts);
1581         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1582                 error = -ENOMEM;
1583                 goto error_fs_info;
1584         }
1585
1586         error = btrfs_open_devices(fs_devices, mode, fs_type);
1587         if (error)
1588                 goto error_fs_info;
1589
1590         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1591                 error = -EACCES;
1592                 goto error_close_devices;
1593         }
1594
1595         bdev = fs_devices->latest_bdev;
1596         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1597                  fs_info);
1598         if (IS_ERR(s)) {
1599                 error = PTR_ERR(s);
1600                 goto error_close_devices;
1601         }
1602
1603         if (s->s_root) {
1604                 btrfs_close_devices(fs_devices);
1605                 free_fs_info(fs_info);
1606                 if ((flags ^ s->s_flags) & MS_RDONLY)
1607                         error = -EBUSY;
1608         } else {
1609                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1610                 btrfs_sb(s)->bdev_holder = fs_type;
1611                 error = btrfs_fill_super(s, fs_devices, data);
1612         }
1613         if (error) {
1614                 deactivate_locked_super(s);
1615                 goto error_sec_opts;
1616         }
1617
1618         fs_info = btrfs_sb(s);
1619         error = setup_security_options(fs_info, s, &new_sec_opts);
1620         if (error) {
1621                 deactivate_locked_super(s);
1622                 goto error_sec_opts;
1623         }
1624
1625         return dget(s->s_root);
1626
1627 error_close_devices:
1628         btrfs_close_devices(fs_devices);
1629 error_fs_info:
1630         free_fs_info(fs_info);
1631 error_sec_opts:
1632         security_free_mnt_opts(&new_sec_opts);
1633         return ERR_PTR(error);
1634 }
1635
1636 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1637                                      int new_pool_size, int old_pool_size)
1638 {
1639         if (new_pool_size == old_pool_size)
1640                 return;
1641
1642         fs_info->thread_pool_size = new_pool_size;
1643
1644         btrfs_info(fs_info, "resize thread pool %d -> %d",
1645                old_pool_size, new_pool_size);
1646
1647         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1648         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1649         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1650         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1651         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1652         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1653         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1654                                 new_pool_size);
1655         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1656         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1657         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1658         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1659         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1660                                 new_pool_size);
1661 }
1662
1663 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1664 {
1665         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1666 }
1667
1668 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1669                                        unsigned long old_opts, int flags)
1670 {
1671         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1672             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1673              (flags & MS_RDONLY))) {
1674                 /* wait for any defraggers to finish */
1675                 wait_event(fs_info->transaction_wait,
1676                            (atomic_read(&fs_info->defrag_running) == 0));
1677                 if (flags & MS_RDONLY)
1678                         sync_filesystem(fs_info->sb);
1679         }
1680 }
1681
1682 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1683                                          unsigned long old_opts)
1684 {
1685         /*
1686          * We need to cleanup all defragable inodes if the autodefragment is
1687          * close or the filesystem is read only.
1688          */
1689         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1690             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1691              (fs_info->sb->s_flags & MS_RDONLY))) {
1692                 btrfs_cleanup_defrag_inodes(fs_info);
1693         }
1694
1695         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1696 }
1697
1698 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1699 {
1700         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1701         struct btrfs_root *root = fs_info->tree_root;
1702         unsigned old_flags = sb->s_flags;
1703         unsigned long old_opts = fs_info->mount_opt;
1704         unsigned long old_compress_type = fs_info->compress_type;
1705         u64 old_max_inline = fs_info->max_inline;
1706         int old_thread_pool_size = fs_info->thread_pool_size;
1707         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1708         int ret;
1709
1710         sync_filesystem(sb);
1711         btrfs_remount_prepare(fs_info);
1712
1713         if (data) {
1714                 struct security_mnt_opts new_sec_opts;
1715
1716                 security_init_mnt_opts(&new_sec_opts);
1717                 ret = parse_security_options(data, &new_sec_opts);
1718                 if (ret)
1719                         goto restore;
1720                 ret = setup_security_options(fs_info, sb,
1721                                              &new_sec_opts);
1722                 if (ret) {
1723                         security_free_mnt_opts(&new_sec_opts);
1724                         goto restore;
1725                 }
1726         }
1727
1728         ret = btrfs_parse_options(fs_info, data, *flags);
1729         if (ret) {
1730                 ret = -EINVAL;
1731                 goto restore;
1732         }
1733
1734         btrfs_remount_begin(fs_info, old_opts, *flags);
1735         btrfs_resize_thread_pool(fs_info,
1736                 fs_info->thread_pool_size, old_thread_pool_size);
1737
1738         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1739                 goto out;
1740
1741         if (*flags & MS_RDONLY) {
1742                 /*
1743                  * this also happens on 'umount -rf' or on shutdown, when
1744                  * the filesystem is busy.
1745                  */
1746                 cancel_work_sync(&fs_info->async_reclaim_work);
1747
1748                 /* wait for the uuid_scan task to finish */
1749                 down(&fs_info->uuid_tree_rescan_sem);
1750                 /* avoid complains from lockdep et al. */
1751                 up(&fs_info->uuid_tree_rescan_sem);
1752
1753                 sb->s_flags |= MS_RDONLY;
1754
1755                 /*
1756                  * Setting MS_RDONLY will put the cleaner thread to
1757                  * sleep at the next loop if it's already active.
1758                  * If it's already asleep, we'll leave unused block
1759                  * groups on disk until we're mounted read-write again
1760                  * unless we clean them up here.
1761                  */
1762                 btrfs_delete_unused_bgs(fs_info);
1763
1764                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1765                 btrfs_scrub_cancel(fs_info);
1766                 btrfs_pause_balance(fs_info);
1767
1768                 ret = btrfs_commit_super(fs_info);
1769                 if (ret)
1770                         goto restore;
1771         } else {
1772                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1773                         btrfs_err(fs_info,
1774                                 "Remounting read-write after error is not allowed");
1775                         ret = -EINVAL;
1776                         goto restore;
1777                 }
1778                 if (fs_info->fs_devices->rw_devices == 0) {
1779                         ret = -EACCES;
1780                         goto restore;
1781                 }
1782
1783                 if (fs_info->fs_devices->missing_devices >
1784                      fs_info->num_tolerated_disk_barrier_failures) {
1785                         btrfs_warn(fs_info,
1786                                 "too many missing devices, writeable remount is not allowed");
1787                         ret = -EACCES;
1788                         goto restore;
1789                 }
1790
1791                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1792                         ret = -EINVAL;
1793                         goto restore;
1794                 }
1795
1796                 ret = btrfs_cleanup_fs_roots(fs_info);
1797                 if (ret)
1798                         goto restore;
1799
1800                 /* recover relocation */
1801                 mutex_lock(&fs_info->cleaner_mutex);
1802                 ret = btrfs_recover_relocation(root);
1803                 mutex_unlock(&fs_info->cleaner_mutex);
1804                 if (ret)
1805                         goto restore;
1806
1807                 ret = btrfs_resume_balance_async(fs_info);
1808                 if (ret)
1809                         goto restore;
1810
1811                 ret = btrfs_resume_dev_replace_async(fs_info);
1812                 if (ret) {
1813                         btrfs_warn(fs_info, "failed to resume dev_replace");
1814                         goto restore;
1815                 }
1816
1817                 if (!fs_info->uuid_root) {
1818                         btrfs_info(fs_info, "creating UUID tree");
1819                         ret = btrfs_create_uuid_tree(fs_info);
1820                         if (ret) {
1821                                 btrfs_warn(fs_info,
1822                                            "failed to create the UUID tree %d",
1823                                            ret);
1824                                 goto restore;
1825                         }
1826                 }
1827                 sb->s_flags &= ~MS_RDONLY;
1828
1829                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1830         }
1831 out:
1832         wake_up_process(fs_info->transaction_kthread);
1833         btrfs_remount_cleanup(fs_info, old_opts);
1834         return 0;
1835
1836 restore:
1837         /* We've hit an error - don't reset MS_RDONLY */
1838         if (sb->s_flags & MS_RDONLY)
1839                 old_flags |= MS_RDONLY;
1840         sb->s_flags = old_flags;
1841         fs_info->mount_opt = old_opts;
1842         fs_info->compress_type = old_compress_type;
1843         fs_info->max_inline = old_max_inline;
1844         btrfs_resize_thread_pool(fs_info,
1845                 old_thread_pool_size, fs_info->thread_pool_size);
1846         fs_info->metadata_ratio = old_metadata_ratio;
1847         btrfs_remount_cleanup(fs_info, old_opts);
1848         return ret;
1849 }
1850
1851 /* Used to sort the devices by max_avail(descending sort) */
1852 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1853                                        const void *dev_info2)
1854 {
1855         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1856             ((struct btrfs_device_info *)dev_info2)->max_avail)
1857                 return -1;
1858         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1859                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1860                 return 1;
1861         else
1862         return 0;
1863 }
1864
1865 /*
1866  * sort the devices by max_avail, in which max free extent size of each device
1867  * is stored.(Descending Sort)
1868  */
1869 static inline void btrfs_descending_sort_devices(
1870                                         struct btrfs_device_info *devices,
1871                                         size_t nr_devices)
1872 {
1873         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1874              btrfs_cmp_device_free_bytes, NULL);
1875 }
1876
1877 /*
1878  * The helper to calc the free space on the devices that can be used to store
1879  * file data.
1880  */
1881 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1882                                        u64 *free_bytes)
1883 {
1884         struct btrfs_device_info *devices_info;
1885         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1886         struct btrfs_device *device;
1887         u64 skip_space;
1888         u64 type;
1889         u64 avail_space;
1890         u64 min_stripe_size;
1891         int min_stripes = 1, num_stripes = 1;
1892         int i = 0, nr_devices;
1893
1894         /*
1895          * We aren't under the device list lock, so this is racy-ish, but good
1896          * enough for our purposes.
1897          */
1898         nr_devices = fs_info->fs_devices->open_devices;
1899         if (!nr_devices) {
1900                 smp_mb();
1901                 nr_devices = fs_info->fs_devices->open_devices;
1902                 ASSERT(nr_devices);
1903                 if (!nr_devices) {
1904                         *free_bytes = 0;
1905                         return 0;
1906                 }
1907         }
1908
1909         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1910                                GFP_KERNEL);
1911         if (!devices_info)
1912                 return -ENOMEM;
1913
1914         /* calc min stripe number for data space allocation */
1915         type = btrfs_data_alloc_profile(fs_info);
1916         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1917                 min_stripes = 2;
1918                 num_stripes = nr_devices;
1919         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1920                 min_stripes = 2;
1921                 num_stripes = 2;
1922         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1923                 min_stripes = 4;
1924                 num_stripes = 4;
1925         }
1926
1927         if (type & BTRFS_BLOCK_GROUP_DUP)
1928                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1929         else
1930                 min_stripe_size = BTRFS_STRIPE_LEN;
1931
1932         rcu_read_lock();
1933         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1934                 if (!device->in_fs_metadata || !device->bdev ||
1935                     device->is_tgtdev_for_dev_replace)
1936                         continue;
1937
1938                 if (i >= nr_devices)
1939                         break;
1940
1941                 avail_space = device->total_bytes - device->bytes_used;
1942
1943                 /* align with stripe_len */
1944                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1945                 avail_space *= BTRFS_STRIPE_LEN;
1946
1947                 /*
1948                  * In order to avoid overwriting the superblock on the drive,
1949                  * btrfs starts at an offset of at least 1MB when doing chunk
1950                  * allocation.
1951                  */
1952                 skip_space = SZ_1M;
1953
1954                 /*
1955                  * we can use the free space in [0, skip_space - 1], subtract
1956                  * it from the total.
1957                  */
1958                 if (avail_space && avail_space >= skip_space)
1959                         avail_space -= skip_space;
1960                 else
1961                         avail_space = 0;
1962
1963                 if (avail_space < min_stripe_size)
1964                         continue;
1965
1966                 devices_info[i].dev = device;
1967                 devices_info[i].max_avail = avail_space;
1968
1969                 i++;
1970         }
1971         rcu_read_unlock();
1972
1973         nr_devices = i;
1974
1975         btrfs_descending_sort_devices(devices_info, nr_devices);
1976
1977         i = nr_devices - 1;
1978         avail_space = 0;
1979         while (nr_devices >= min_stripes) {
1980                 if (num_stripes > nr_devices)
1981                         num_stripes = nr_devices;
1982
1983                 if (devices_info[i].max_avail >= min_stripe_size) {
1984                         int j;
1985                         u64 alloc_size;
1986
1987                         avail_space += devices_info[i].max_avail * num_stripes;
1988                         alloc_size = devices_info[i].max_avail;
1989                         for (j = i + 1 - num_stripes; j <= i; j++)
1990                                 devices_info[j].max_avail -= alloc_size;
1991                 }
1992                 i--;
1993                 nr_devices--;
1994         }
1995
1996         kfree(devices_info);
1997         *free_bytes = avail_space;
1998         return 0;
1999 }
2000
2001 /*
2002  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2003  *
2004  * If there's a redundant raid level at DATA block groups, use the respective
2005  * multiplier to scale the sizes.
2006  *
2007  * Unused device space usage is based on simulating the chunk allocator
2008  * algorithm that respects the device sizes and order of allocations.  This is
2009  * a close approximation of the actual use but there are other factors that may
2010  * change the result (like a new metadata chunk).
2011  *
2012  * If metadata is exhausted, f_bavail will be 0.
2013  */
2014 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2015 {
2016         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2017         struct btrfs_super_block *disk_super = fs_info->super_copy;
2018         struct list_head *head = &fs_info->space_info;
2019         struct btrfs_space_info *found;
2020         u64 total_used = 0;
2021         u64 total_free_data = 0;
2022         u64 total_free_meta = 0;
2023         int bits = dentry->d_sb->s_blocksize_bits;
2024         __be32 *fsid = (__be32 *)fs_info->fsid;
2025         unsigned factor = 1;
2026         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2027         int ret;
2028         u64 thresh = 0;
2029         int mixed = 0;
2030
2031         rcu_read_lock();
2032         list_for_each_entry_rcu(found, head, list) {
2033                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2034                         int i;
2035
2036                         total_free_data += found->disk_total - found->disk_used;
2037                         total_free_data -=
2038                                 btrfs_account_ro_block_groups_free_space(found);
2039
2040                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2041                                 if (!list_empty(&found->block_groups[i])) {
2042                                         switch (i) {
2043                                         case BTRFS_RAID_DUP:
2044                                         case BTRFS_RAID_RAID1:
2045                                         case BTRFS_RAID_RAID10:
2046                                                 factor = 2;
2047                                         }
2048                                 }
2049                         }
2050                 }
2051
2052                 /*
2053                  * Metadata in mixed block goup profiles are accounted in data
2054                  */
2055                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2056                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2057                                 mixed = 1;
2058                         else
2059                                 total_free_meta += found->disk_total -
2060                                         found->disk_used;
2061                 }
2062
2063                 total_used += found->disk_used;
2064         }
2065
2066         rcu_read_unlock();
2067
2068         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2069         buf->f_blocks >>= bits;
2070         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2071
2072         /* Account global block reserve as used, it's in logical size already */
2073         spin_lock(&block_rsv->lock);
2074         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2075         if (buf->f_bfree >= block_rsv->size >> bits)
2076                 buf->f_bfree -= block_rsv->size >> bits;
2077         else
2078                 buf->f_bfree = 0;
2079         spin_unlock(&block_rsv->lock);
2080
2081         buf->f_bavail = div_u64(total_free_data, factor);
2082         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2083         if (ret)
2084                 return ret;
2085         buf->f_bavail += div_u64(total_free_data, factor);
2086         buf->f_bavail = buf->f_bavail >> bits;
2087
2088         /*
2089          * We calculate the remaining metadata space minus global reserve. If
2090          * this is (supposedly) smaller than zero, there's no space. But this
2091          * does not hold in practice, the exhausted state happens where's still
2092          * some positive delta. So we apply some guesswork and compare the
2093          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2094          *
2095          * We probably cannot calculate the exact threshold value because this
2096          * depends on the internal reservations requested by various
2097          * operations, so some operations that consume a few metadata will
2098          * succeed even if the Avail is zero. But this is better than the other
2099          * way around.
2100          */
2101         thresh = 4 * 1024 * 1024;
2102
2103         if (!mixed && total_free_meta - thresh < block_rsv->size)
2104                 buf->f_bavail = 0;
2105
2106         buf->f_type = BTRFS_SUPER_MAGIC;
2107         buf->f_bsize = dentry->d_sb->s_blocksize;
2108         buf->f_namelen = BTRFS_NAME_LEN;
2109
2110         /* We treat it as constant endianness (it doesn't matter _which_)
2111            because we want the fsid to come out the same whether mounted
2112            on a big-endian or little-endian host */
2113         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2114         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2115         /* Mask in the root object ID too, to disambiguate subvols */
2116         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2117         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2118
2119         return 0;
2120 }
2121
2122 static void btrfs_kill_super(struct super_block *sb)
2123 {
2124         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2125         kill_anon_super(sb);
2126         free_fs_info(fs_info);
2127 }
2128
2129 static struct file_system_type btrfs_fs_type = {
2130         .owner          = THIS_MODULE,
2131         .name           = "btrfs",
2132         .mount          = btrfs_mount,
2133         .kill_sb        = btrfs_kill_super,
2134         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2135 };
2136 MODULE_ALIAS_FS("btrfs");
2137
2138 static int btrfs_control_open(struct inode *inode, struct file *file)
2139 {
2140         /*
2141          * The control file's private_data is used to hold the
2142          * transaction when it is started and is used to keep
2143          * track of whether a transaction is already in progress.
2144          */
2145         file->private_data = NULL;
2146         return 0;
2147 }
2148
2149 /*
2150  * used by btrfsctl to scan devices when no FS is mounted
2151  */
2152 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2153                                 unsigned long arg)
2154 {
2155         struct btrfs_ioctl_vol_args *vol;
2156         struct btrfs_fs_devices *fs_devices;
2157         int ret = -ENOTTY;
2158
2159         if (!capable(CAP_SYS_ADMIN))
2160                 return -EPERM;
2161
2162         vol = memdup_user((void __user *)arg, sizeof(*vol));
2163         if (IS_ERR(vol))
2164                 return PTR_ERR(vol);
2165
2166         switch (cmd) {
2167         case BTRFS_IOC_SCAN_DEV:
2168                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2169                                             &btrfs_fs_type, &fs_devices);
2170                 break;
2171         case BTRFS_IOC_DEVICES_READY:
2172                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2173                                             &btrfs_fs_type, &fs_devices);
2174                 if (ret)
2175                         break;
2176                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2177                 break;
2178         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2179                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2180                 break;
2181         }
2182
2183         kfree(vol);
2184         return ret;
2185 }
2186
2187 static int btrfs_freeze(struct super_block *sb)
2188 {
2189         struct btrfs_trans_handle *trans;
2190         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2191         struct btrfs_root *root = fs_info->tree_root;
2192
2193         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2194         /*
2195          * We don't need a barrier here, we'll wait for any transaction that
2196          * could be in progress on other threads (and do delayed iputs that
2197          * we want to avoid on a frozen filesystem), or do the commit
2198          * ourselves.
2199          */
2200         trans = btrfs_attach_transaction_barrier(root);
2201         if (IS_ERR(trans)) {
2202                 /* no transaction, don't bother */
2203                 if (PTR_ERR(trans) == -ENOENT)
2204                         return 0;
2205                 return PTR_ERR(trans);
2206         }
2207         return btrfs_commit_transaction(trans);
2208 }
2209
2210 static int btrfs_unfreeze(struct super_block *sb)
2211 {
2212         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2213
2214         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2215         return 0;
2216 }
2217
2218 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2219 {
2220         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2221         struct btrfs_fs_devices *cur_devices;
2222         struct btrfs_device *dev, *first_dev = NULL;
2223         struct list_head *head;
2224         struct rcu_string *name;
2225
2226         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2227         cur_devices = fs_info->fs_devices;
2228         while (cur_devices) {
2229                 head = &cur_devices->devices;
2230                 list_for_each_entry(dev, head, dev_list) {
2231                         if (dev->missing)
2232                                 continue;
2233                         if (!dev->name)
2234                                 continue;
2235                         if (!first_dev || dev->devid < first_dev->devid)
2236                                 first_dev = dev;
2237                 }
2238                 cur_devices = cur_devices->seed;
2239         }
2240
2241         if (first_dev) {
2242                 rcu_read_lock();
2243                 name = rcu_dereference(first_dev->name);
2244                 seq_escape(m, name->str, " \t\n\\");
2245                 rcu_read_unlock();
2246         } else {
2247                 WARN_ON(1);
2248         }
2249         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2250         return 0;
2251 }
2252
2253 static const struct super_operations btrfs_super_ops = {
2254         .drop_inode     = btrfs_drop_inode,
2255         .evict_inode    = btrfs_evict_inode,
2256         .put_super      = btrfs_put_super,
2257         .sync_fs        = btrfs_sync_fs,
2258         .show_options   = btrfs_show_options,
2259         .show_devname   = btrfs_show_devname,
2260         .write_inode    = btrfs_write_inode,
2261         .alloc_inode    = btrfs_alloc_inode,
2262         .destroy_inode  = btrfs_destroy_inode,
2263         .statfs         = btrfs_statfs,
2264         .remount_fs     = btrfs_remount,
2265         .freeze_fs      = btrfs_freeze,
2266         .unfreeze_fs    = btrfs_unfreeze,
2267 };
2268
2269 static const struct file_operations btrfs_ctl_fops = {
2270         .open = btrfs_control_open,
2271         .unlocked_ioctl  = btrfs_control_ioctl,
2272         .compat_ioctl = btrfs_control_ioctl,
2273         .owner   = THIS_MODULE,
2274         .llseek = noop_llseek,
2275 };
2276
2277 static struct miscdevice btrfs_misc = {
2278         .minor          = BTRFS_MINOR,
2279         .name           = "btrfs-control",
2280         .fops           = &btrfs_ctl_fops
2281 };
2282
2283 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2284 MODULE_ALIAS("devname:btrfs-control");
2285
2286 static int btrfs_interface_init(void)
2287 {
2288         return misc_register(&btrfs_misc);
2289 }
2290
2291 static void btrfs_interface_exit(void)
2292 {
2293         misc_deregister(&btrfs_misc);
2294 }
2295
2296 static void btrfs_print_mod_info(void)
2297 {
2298         pr_info("Btrfs loaded, crc32c=%s"
2299 #ifdef CONFIG_BTRFS_DEBUG
2300                         ", debug=on"
2301 #endif
2302 #ifdef CONFIG_BTRFS_ASSERT
2303                         ", assert=on"
2304 #endif
2305 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2306                         ", integrity-checker=on"
2307 #endif
2308                         "\n",
2309                         btrfs_crc32c_impl());
2310 }
2311
2312 static int __init init_btrfs_fs(void)
2313 {
2314         int err;
2315
2316         err = btrfs_hash_init();
2317         if (err)
2318                 return err;
2319
2320         btrfs_props_init();
2321
2322         err = btrfs_init_sysfs();
2323         if (err)
2324                 goto free_hash;
2325
2326         btrfs_init_compress();
2327
2328         err = btrfs_init_cachep();
2329         if (err)
2330                 goto free_compress;
2331
2332         err = extent_io_init();
2333         if (err)
2334                 goto free_cachep;
2335
2336         err = extent_map_init();
2337         if (err)
2338                 goto free_extent_io;
2339
2340         err = ordered_data_init();
2341         if (err)
2342                 goto free_extent_map;
2343
2344         err = btrfs_delayed_inode_init();
2345         if (err)
2346                 goto free_ordered_data;
2347
2348         err = btrfs_auto_defrag_init();
2349         if (err)
2350                 goto free_delayed_inode;
2351
2352         err = btrfs_delayed_ref_init();
2353         if (err)
2354                 goto free_auto_defrag;
2355
2356         err = btrfs_prelim_ref_init();
2357         if (err)
2358                 goto free_delayed_ref;
2359
2360         err = btrfs_end_io_wq_init();
2361         if (err)
2362                 goto free_prelim_ref;
2363
2364         err = btrfs_interface_init();
2365         if (err)
2366                 goto free_end_io_wq;
2367
2368         btrfs_init_lockdep();
2369
2370         btrfs_print_mod_info();
2371
2372         err = btrfs_run_sanity_tests();
2373         if (err)
2374                 goto unregister_ioctl;
2375
2376         err = register_filesystem(&btrfs_fs_type);
2377         if (err)
2378                 goto unregister_ioctl;
2379
2380         return 0;
2381
2382 unregister_ioctl:
2383         btrfs_interface_exit();
2384 free_end_io_wq:
2385         btrfs_end_io_wq_exit();
2386 free_prelim_ref:
2387         btrfs_prelim_ref_exit();
2388 free_delayed_ref:
2389         btrfs_delayed_ref_exit();
2390 free_auto_defrag:
2391         btrfs_auto_defrag_exit();
2392 free_delayed_inode:
2393         btrfs_delayed_inode_exit();
2394 free_ordered_data:
2395         ordered_data_exit();
2396 free_extent_map:
2397         extent_map_exit();
2398 free_extent_io:
2399         extent_io_exit();
2400 free_cachep:
2401         btrfs_destroy_cachep();
2402 free_compress:
2403         btrfs_exit_compress();
2404         btrfs_exit_sysfs();
2405 free_hash:
2406         btrfs_hash_exit();
2407         return err;
2408 }
2409
2410 static void __exit exit_btrfs_fs(void)
2411 {
2412         btrfs_destroy_cachep();
2413         btrfs_delayed_ref_exit();
2414         btrfs_auto_defrag_exit();
2415         btrfs_delayed_inode_exit();
2416         btrfs_prelim_ref_exit();
2417         ordered_data_exit();
2418         extent_map_exit();
2419         extent_io_exit();
2420         btrfs_interface_exit();
2421         btrfs_end_io_wq_exit();
2422         unregister_filesystem(&btrfs_fs_type);
2423         btrfs_exit_sysfs();
2424         btrfs_cleanup_fs_uuids();
2425         btrfs_exit_compress();
2426         btrfs_hash_exit();
2427 }
2428
2429 late_initcall(init_btrfs_fs);
2430 module_exit(exit_btrfs_fs)
2431
2432 MODULE_LICENSE("GPL");