3e298f26a38300e014695997011d8ba47a4ac464
[sfrench/cifs-2.6.git] / fs / btrfs / super.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/blkdev.h>
7 #include <linux/module.h>
8 #include <linux/buffer_head.h>
9 #include <linux/fs.h>
10 #include <linux/pagemap.h>
11 #include <linux/highmem.h>
12 #include <linux/time.h>
13 #include <linux/init.h>
14 #include <linux/seq_file.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/mount.h>
18 #include <linux/mpage.h>
19 #include <linux/swap.h>
20 #include <linux/writeback.h>
21 #include <linux/statfs.h>
22 #include <linux/compat.h>
23 #include <linux/parser.h>
24 #include <linux/ctype.h>
25 #include <linux/namei.h>
26 #include <linux/miscdevice.h>
27 #include <linux/magic.h>
28 #include <linux/slab.h>
29 #include <linux/cleancache.h>
30 #include <linux/ratelimit.h>
31 #include <linux/crc32c.h>
32 #include <linux/btrfs.h>
33 #include "delayed-inode.h"
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "print-tree.h"
39 #include "props.h"
40 #include "xattr.h"
41 #include "volumes.h"
42 #include "export.h"
43 #include "compression.h"
44 #include "rcu-string.h"
45 #include "dev-replace.h"
46 #include "free-space-cache.h"
47 #include "backref.h"
48 #include "tests/btrfs-tests.h"
49
50 #include "qgroup.h"
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/btrfs.h>
53
54 static const struct super_operations btrfs_super_ops;
55
56 /*
57  * Types for mounting the default subvolume and a subvolume explicitly
58  * requested by subvol=/path. That way the callchain is straightforward and we
59  * don't have to play tricks with the mount options and recursive calls to
60  * btrfs_mount.
61  *
62  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
63  */
64 static struct file_system_type btrfs_fs_type;
65 static struct file_system_type btrfs_root_fs_type;
66
67 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
68
69 const char *btrfs_decode_error(int errno)
70 {
71         char *errstr = "unknown";
72
73         switch (errno) {
74         case -EIO:
75                 errstr = "IO failure";
76                 break;
77         case -ENOMEM:
78                 errstr = "Out of memory";
79                 break;
80         case -EROFS:
81                 errstr = "Readonly filesystem";
82                 break;
83         case -EEXIST:
84                 errstr = "Object already exists";
85                 break;
86         case -ENOSPC:
87                 errstr = "No space left";
88                 break;
89         case -ENOENT:
90                 errstr = "No such entry";
91                 break;
92         }
93
94         return errstr;
95 }
96
97 /*
98  * __btrfs_handle_fs_error decodes expected errors from the caller and
99  * invokes the approciate error response.
100  */
101 __cold
102 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
103                        unsigned int line, int errno, const char *fmt, ...)
104 {
105         struct super_block *sb = fs_info->sb;
106 #ifdef CONFIG_PRINTK
107         const char *errstr;
108 #endif
109
110         /*
111          * Special case: if the error is EROFS, and we're already
112          * under SB_RDONLY, then it is safe here.
113          */
114         if (errno == -EROFS && sb_rdonly(sb))
115                 return;
116
117 #ifdef CONFIG_PRINTK
118         errstr = btrfs_decode_error(errno);
119         if (fmt) {
120                 struct va_format vaf;
121                 va_list args;
122
123                 va_start(args, fmt);
124                 vaf.fmt = fmt;
125                 vaf.va = &args;
126
127                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
128                         sb->s_id, function, line, errno, errstr, &vaf);
129                 va_end(args);
130         } else {
131                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
132                         sb->s_id, function, line, errno, errstr);
133         }
134 #endif
135
136         /*
137          * Today we only save the error info to memory.  Long term we'll
138          * also send it down to the disk
139          */
140         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
141
142         /* Don't go through full error handling during mount */
143         if (!(sb->s_flags & SB_BORN))
144                 return;
145
146         if (sb_rdonly(sb))
147                 return;
148
149         /* btrfs handle error by forcing the filesystem readonly */
150         sb->s_flags |= SB_RDONLY;
151         btrfs_info(fs_info, "forced readonly");
152         /*
153          * Note that a running device replace operation is not canceled here
154          * although there is no way to update the progress. It would add the
155          * risk of a deadlock, therefore the canceling is omitted. The only
156          * penalty is that some I/O remains active until the procedure
157          * completes. The next time when the filesystem is mounted writeable
158          * again, the device replace operation continues.
159          */
160 }
161
162 #ifdef CONFIG_PRINTK
163 static const char * const logtypes[] = {
164         "emergency",
165         "alert",
166         "critical",
167         "error",
168         "warning",
169         "notice",
170         "info",
171         "debug",
172 };
173
174
175 /*
176  * Use one ratelimit state per log level so that a flood of less important
177  * messages doesn't cause more important ones to be dropped.
178  */
179 static struct ratelimit_state printk_limits[] = {
180         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
181         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
182         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
183         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
184         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
185         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
186         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
187         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
188 };
189
190 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
191 {
192         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
193         struct va_format vaf;
194         va_list args;
195         int kern_level;
196         const char *type = logtypes[4];
197         struct ratelimit_state *ratelimit = &printk_limits[4];
198
199         va_start(args, fmt);
200
201         while ((kern_level = printk_get_level(fmt)) != 0) {
202                 size_t size = printk_skip_level(fmt) - fmt;
203
204                 if (kern_level >= '0' && kern_level <= '7') {
205                         memcpy(lvl, fmt,  size);
206                         lvl[size] = '\0';
207                         type = logtypes[kern_level - '0'];
208                         ratelimit = &printk_limits[kern_level - '0'];
209                 }
210                 fmt += size;
211         }
212
213         vaf.fmt = fmt;
214         vaf.va = &args;
215
216         if (__ratelimit(ratelimit))
217                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
218                         fs_info ? fs_info->sb->s_id : "<unknown>", &vaf);
219
220         va_end(args);
221 }
222 #endif
223
224 /*
225  * We only mark the transaction aborted and then set the file system read-only.
226  * This will prevent new transactions from starting or trying to join this
227  * one.
228  *
229  * This means that error recovery at the call site is limited to freeing
230  * any local memory allocations and passing the error code up without
231  * further cleanup. The transaction should complete as it normally would
232  * in the call path but will return -EIO.
233  *
234  * We'll complete the cleanup in btrfs_end_transaction and
235  * btrfs_commit_transaction.
236  */
237 __cold
238 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
239                                const char *function,
240                                unsigned int line, int errno)
241 {
242         struct btrfs_fs_info *fs_info = trans->fs_info;
243
244         trans->aborted = errno;
245         /* Nothing used. The other threads that have joined this
246          * transaction may be able to continue. */
247         if (!trans->dirty && list_empty(&trans->new_bgs)) {
248                 const char *errstr;
249
250                 errstr = btrfs_decode_error(errno);
251                 btrfs_warn(fs_info,
252                            "%s:%d: Aborting unused transaction(%s).",
253                            function, line, errstr);
254                 return;
255         }
256         WRITE_ONCE(trans->transaction->aborted, errno);
257         /* Wake up anybody who may be waiting on this transaction */
258         wake_up(&fs_info->transaction_wait);
259         wake_up(&fs_info->transaction_blocked_wait);
260         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
261 }
262 /*
263  * __btrfs_panic decodes unexpected, fatal errors from the caller,
264  * issues an alert, and either panics or BUGs, depending on mount options.
265  */
266 __cold
267 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
268                    unsigned int line, int errno, const char *fmt, ...)
269 {
270         char *s_id = "<unknown>";
271         const char *errstr;
272         struct va_format vaf = { .fmt = fmt };
273         va_list args;
274
275         if (fs_info)
276                 s_id = fs_info->sb->s_id;
277
278         va_start(args, fmt);
279         vaf.va = &args;
280
281         errstr = btrfs_decode_error(errno);
282         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
283                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
284                         s_id, function, line, &vaf, errno, errstr);
285
286         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
287                    function, line, &vaf, errno, errstr);
288         va_end(args);
289         /* Caller calls BUG() */
290 }
291
292 static void btrfs_put_super(struct super_block *sb)
293 {
294         close_ctree(btrfs_sb(sb));
295 }
296
297 enum {
298         Opt_acl, Opt_noacl,
299         Opt_clear_cache,
300         Opt_commit_interval,
301         Opt_compress,
302         Opt_compress_force,
303         Opt_compress_force_type,
304         Opt_compress_type,
305         Opt_degraded,
306         Opt_device,
307         Opt_fatal_errors,
308         Opt_flushoncommit, Opt_noflushoncommit,
309         Opt_inode_cache, Opt_noinode_cache,
310         Opt_max_inline,
311         Opt_barrier, Opt_nobarrier,
312         Opt_datacow, Opt_nodatacow,
313         Opt_datasum, Opt_nodatasum,
314         Opt_defrag, Opt_nodefrag,
315         Opt_discard, Opt_nodiscard,
316         Opt_nologreplay,
317         Opt_norecovery,
318         Opt_ratio,
319         Opt_rescan_uuid_tree,
320         Opt_skip_balance,
321         Opt_space_cache, Opt_no_space_cache,
322         Opt_space_cache_version,
323         Opt_ssd, Opt_nossd,
324         Opt_ssd_spread, Opt_nossd_spread,
325         Opt_subvol,
326         Opt_subvol_empty,
327         Opt_subvolid,
328         Opt_thread_pool,
329         Opt_treelog, Opt_notreelog,
330         Opt_usebackuproot,
331         Opt_user_subvol_rm_allowed,
332
333         /* Deprecated options */
334         Opt_alloc_start,
335         Opt_recovery,
336         Opt_subvolrootid,
337
338         /* Debugging options */
339         Opt_check_integrity,
340         Opt_check_integrity_including_extent_data,
341         Opt_check_integrity_print_mask,
342         Opt_enospc_debug, Opt_noenospc_debug,
343 #ifdef CONFIG_BTRFS_DEBUG
344         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
345 #endif
346 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
347         Opt_ref_verify,
348 #endif
349         Opt_err,
350 };
351
352 static const match_table_t tokens = {
353         {Opt_acl, "acl"},
354         {Opt_noacl, "noacl"},
355         {Opt_clear_cache, "clear_cache"},
356         {Opt_commit_interval, "commit=%u"},
357         {Opt_compress, "compress"},
358         {Opt_compress_type, "compress=%s"},
359         {Opt_compress_force, "compress-force"},
360         {Opt_compress_force_type, "compress-force=%s"},
361         {Opt_degraded, "degraded"},
362         {Opt_device, "device=%s"},
363         {Opt_fatal_errors, "fatal_errors=%s"},
364         {Opt_flushoncommit, "flushoncommit"},
365         {Opt_noflushoncommit, "noflushoncommit"},
366         {Opt_inode_cache, "inode_cache"},
367         {Opt_noinode_cache, "noinode_cache"},
368         {Opt_max_inline, "max_inline=%s"},
369         {Opt_barrier, "barrier"},
370         {Opt_nobarrier, "nobarrier"},
371         {Opt_datacow, "datacow"},
372         {Opt_nodatacow, "nodatacow"},
373         {Opt_datasum, "datasum"},
374         {Opt_nodatasum, "nodatasum"},
375         {Opt_defrag, "autodefrag"},
376         {Opt_nodefrag, "noautodefrag"},
377         {Opt_discard, "discard"},
378         {Opt_nodiscard, "nodiscard"},
379         {Opt_nologreplay, "nologreplay"},
380         {Opt_norecovery, "norecovery"},
381         {Opt_ratio, "metadata_ratio=%u"},
382         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
383         {Opt_skip_balance, "skip_balance"},
384         {Opt_space_cache, "space_cache"},
385         {Opt_no_space_cache, "nospace_cache"},
386         {Opt_space_cache_version, "space_cache=%s"},
387         {Opt_ssd, "ssd"},
388         {Opt_nossd, "nossd"},
389         {Opt_ssd_spread, "ssd_spread"},
390         {Opt_nossd_spread, "nossd_spread"},
391         {Opt_subvol, "subvol=%s"},
392         {Opt_subvol_empty, "subvol="},
393         {Opt_subvolid, "subvolid=%s"},
394         {Opt_thread_pool, "thread_pool=%u"},
395         {Opt_treelog, "treelog"},
396         {Opt_notreelog, "notreelog"},
397         {Opt_usebackuproot, "usebackuproot"},
398         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
399
400         /* Deprecated options */
401         {Opt_alloc_start, "alloc_start=%s"},
402         {Opt_recovery, "recovery"},
403         {Opt_subvolrootid, "subvolrootid=%d"},
404
405         /* Debugging options */
406         {Opt_check_integrity, "check_int"},
407         {Opt_check_integrity_including_extent_data, "check_int_data"},
408         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
409         {Opt_enospc_debug, "enospc_debug"},
410         {Opt_noenospc_debug, "noenospc_debug"},
411 #ifdef CONFIG_BTRFS_DEBUG
412         {Opt_fragment_data, "fragment=data"},
413         {Opt_fragment_metadata, "fragment=metadata"},
414         {Opt_fragment_all, "fragment=all"},
415 #endif
416 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
417         {Opt_ref_verify, "ref_verify"},
418 #endif
419         {Opt_err, NULL},
420 };
421
422 /*
423  * Regular mount options parser.  Everything that is needed only when
424  * reading in a new superblock is parsed here.
425  * XXX JDM: This needs to be cleaned up for remount.
426  */
427 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
428                         unsigned long new_flags)
429 {
430         substring_t args[MAX_OPT_ARGS];
431         char *p, *num;
432         u64 cache_gen;
433         int intarg;
434         int ret = 0;
435         char *compress_type;
436         bool compress_force = false;
437         enum btrfs_compression_type saved_compress_type;
438         bool saved_compress_force;
439         int no_compress = 0;
440
441         cache_gen = btrfs_super_cache_generation(info->super_copy);
442         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
443                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
444         else if (cache_gen)
445                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
446
447         /*
448          * Even the options are empty, we still need to do extra check
449          * against new flags
450          */
451         if (!options)
452                 goto check;
453
454         while ((p = strsep(&options, ",")) != NULL) {
455                 int token;
456                 if (!*p)
457                         continue;
458
459                 token = match_token(p, tokens, args);
460                 switch (token) {
461                 case Opt_degraded:
462                         btrfs_info(info, "allowing degraded mounts");
463                         btrfs_set_opt(info->mount_opt, DEGRADED);
464                         break;
465                 case Opt_subvol:
466                 case Opt_subvol_empty:
467                 case Opt_subvolid:
468                 case Opt_subvolrootid:
469                 case Opt_device:
470                         /*
471                          * These are parsed by btrfs_parse_subvol_options
472                          * and btrfs_parse_early_options
473                          * and can be happily ignored here.
474                          */
475                         break;
476                 case Opt_nodatasum:
477                         btrfs_set_and_info(info, NODATASUM,
478                                            "setting nodatasum");
479                         break;
480                 case Opt_datasum:
481                         if (btrfs_test_opt(info, NODATASUM)) {
482                                 if (btrfs_test_opt(info, NODATACOW))
483                                         btrfs_info(info,
484                                                    "setting datasum, datacow enabled");
485                                 else
486                                         btrfs_info(info, "setting datasum");
487                         }
488                         btrfs_clear_opt(info->mount_opt, NODATACOW);
489                         btrfs_clear_opt(info->mount_opt, NODATASUM);
490                         break;
491                 case Opt_nodatacow:
492                         if (!btrfs_test_opt(info, NODATACOW)) {
493                                 if (!btrfs_test_opt(info, COMPRESS) ||
494                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
495                                         btrfs_info(info,
496                                                    "setting nodatacow, compression disabled");
497                                 } else {
498                                         btrfs_info(info, "setting nodatacow");
499                                 }
500                         }
501                         btrfs_clear_opt(info->mount_opt, COMPRESS);
502                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
503                         btrfs_set_opt(info->mount_opt, NODATACOW);
504                         btrfs_set_opt(info->mount_opt, NODATASUM);
505                         break;
506                 case Opt_datacow:
507                         btrfs_clear_and_info(info, NODATACOW,
508                                              "setting datacow");
509                         break;
510                 case Opt_compress_force:
511                 case Opt_compress_force_type:
512                         compress_force = true;
513                         /* Fallthrough */
514                 case Opt_compress:
515                 case Opt_compress_type:
516                         saved_compress_type = btrfs_test_opt(info,
517                                                              COMPRESS) ?
518                                 info->compress_type : BTRFS_COMPRESS_NONE;
519                         saved_compress_force =
520                                 btrfs_test_opt(info, FORCE_COMPRESS);
521                         if (token == Opt_compress ||
522                             token == Opt_compress_force ||
523                             strncmp(args[0].from, "zlib", 4) == 0) {
524                                 compress_type = "zlib";
525
526                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
527                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
528                                 /*
529                                  * args[0] contains uninitialized data since
530                                  * for these tokens we don't expect any
531                                  * parameter.
532                                  */
533                                 if (token != Opt_compress &&
534                                     token != Opt_compress_force)
535                                         info->compress_level =
536                                           btrfs_compress_str2level(args[0].from);
537                                 btrfs_set_opt(info->mount_opt, COMPRESS);
538                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
539                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
540                                 no_compress = 0;
541                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
542                                 compress_type = "lzo";
543                                 info->compress_type = BTRFS_COMPRESS_LZO;
544                                 btrfs_set_opt(info->mount_opt, COMPRESS);
545                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
546                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
547                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
548                                 no_compress = 0;
549                         } else if (strcmp(args[0].from, "zstd") == 0) {
550                                 compress_type = "zstd";
551                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
552                                 btrfs_set_opt(info->mount_opt, COMPRESS);
553                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
554                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
555                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
556                                 no_compress = 0;
557                         } else if (strncmp(args[0].from, "no", 2) == 0) {
558                                 compress_type = "no";
559                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
560                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
561                                 compress_force = false;
562                                 no_compress++;
563                         } else {
564                                 ret = -EINVAL;
565                                 goto out;
566                         }
567
568                         if (compress_force) {
569                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
570                         } else {
571                                 /*
572                                  * If we remount from compress-force=xxx to
573                                  * compress=xxx, we need clear FORCE_COMPRESS
574                                  * flag, otherwise, there is no way for users
575                                  * to disable forcible compression separately.
576                                  */
577                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
578                         }
579                         if ((btrfs_test_opt(info, COMPRESS) &&
580                              (info->compress_type != saved_compress_type ||
581                               compress_force != saved_compress_force)) ||
582                             (!btrfs_test_opt(info, COMPRESS) &&
583                              no_compress == 1)) {
584                                 btrfs_info(info, "%s %s compression, level %d",
585                                            (compress_force) ? "force" : "use",
586                                            compress_type, info->compress_level);
587                         }
588                         compress_force = false;
589                         break;
590                 case Opt_ssd:
591                         btrfs_set_and_info(info, SSD,
592                                            "enabling ssd optimizations");
593                         btrfs_clear_opt(info->mount_opt, NOSSD);
594                         break;
595                 case Opt_ssd_spread:
596                         btrfs_set_and_info(info, SSD,
597                                            "enabling ssd optimizations");
598                         btrfs_set_and_info(info, SSD_SPREAD,
599                                            "using spread ssd allocation scheme");
600                         btrfs_clear_opt(info->mount_opt, NOSSD);
601                         break;
602                 case Opt_nossd:
603                         btrfs_set_opt(info->mount_opt, NOSSD);
604                         btrfs_clear_and_info(info, SSD,
605                                              "not using ssd optimizations");
606                         /* Fallthrough */
607                 case Opt_nossd_spread:
608                         btrfs_clear_and_info(info, SSD_SPREAD,
609                                              "not using spread ssd allocation scheme");
610                         break;
611                 case Opt_barrier:
612                         btrfs_clear_and_info(info, NOBARRIER,
613                                              "turning on barriers");
614                         break;
615                 case Opt_nobarrier:
616                         btrfs_set_and_info(info, NOBARRIER,
617                                            "turning off barriers");
618                         break;
619                 case Opt_thread_pool:
620                         ret = match_int(&args[0], &intarg);
621                         if (ret) {
622                                 goto out;
623                         } else if (intarg == 0) {
624                                 ret = -EINVAL;
625                                 goto out;
626                         }
627                         info->thread_pool_size = intarg;
628                         break;
629                 case Opt_max_inline:
630                         num = match_strdup(&args[0]);
631                         if (num) {
632                                 info->max_inline = memparse(num, NULL);
633                                 kfree(num);
634
635                                 if (info->max_inline) {
636                                         info->max_inline = min_t(u64,
637                                                 info->max_inline,
638                                                 info->sectorsize);
639                                 }
640                                 btrfs_info(info, "max_inline at %llu",
641                                            info->max_inline);
642                         } else {
643                                 ret = -ENOMEM;
644                                 goto out;
645                         }
646                         break;
647                 case Opt_alloc_start:
648                         btrfs_info(info,
649                                 "option alloc_start is obsolete, ignored");
650                         break;
651                 case Opt_acl:
652 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
653                         info->sb->s_flags |= SB_POSIXACL;
654                         break;
655 #else
656                         btrfs_err(info, "support for ACL not compiled in!");
657                         ret = -EINVAL;
658                         goto out;
659 #endif
660                 case Opt_noacl:
661                         info->sb->s_flags &= ~SB_POSIXACL;
662                         break;
663                 case Opt_notreelog:
664                         btrfs_set_and_info(info, NOTREELOG,
665                                            "disabling tree log");
666                         break;
667                 case Opt_treelog:
668                         btrfs_clear_and_info(info, NOTREELOG,
669                                              "enabling tree log");
670                         break;
671                 case Opt_norecovery:
672                 case Opt_nologreplay:
673                         btrfs_set_and_info(info, NOLOGREPLAY,
674                                            "disabling log replay at mount time");
675                         break;
676                 case Opt_flushoncommit:
677                         btrfs_set_and_info(info, FLUSHONCOMMIT,
678                                            "turning on flush-on-commit");
679                         break;
680                 case Opt_noflushoncommit:
681                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
682                                              "turning off flush-on-commit");
683                         break;
684                 case Opt_ratio:
685                         ret = match_int(&args[0], &intarg);
686                         if (ret)
687                                 goto out;
688                         info->metadata_ratio = intarg;
689                         btrfs_info(info, "metadata ratio %u",
690                                    info->metadata_ratio);
691                         break;
692                 case Opt_discard:
693                         btrfs_set_and_info(info, DISCARD,
694                                            "turning on discard");
695                         break;
696                 case Opt_nodiscard:
697                         btrfs_clear_and_info(info, DISCARD,
698                                              "turning off discard");
699                         break;
700                 case Opt_space_cache:
701                 case Opt_space_cache_version:
702                         if (token == Opt_space_cache ||
703                             strcmp(args[0].from, "v1") == 0) {
704                                 btrfs_clear_opt(info->mount_opt,
705                                                 FREE_SPACE_TREE);
706                                 btrfs_set_and_info(info, SPACE_CACHE,
707                                            "enabling disk space caching");
708                         } else if (strcmp(args[0].from, "v2") == 0) {
709                                 btrfs_clear_opt(info->mount_opt,
710                                                 SPACE_CACHE);
711                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
712                                                    "enabling free space tree");
713                         } else {
714                                 ret = -EINVAL;
715                                 goto out;
716                         }
717                         break;
718                 case Opt_rescan_uuid_tree:
719                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
720                         break;
721                 case Opt_no_space_cache:
722                         if (btrfs_test_opt(info, SPACE_CACHE)) {
723                                 btrfs_clear_and_info(info, SPACE_CACHE,
724                                              "disabling disk space caching");
725                         }
726                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
727                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
728                                              "disabling free space tree");
729                         }
730                         break;
731                 case Opt_inode_cache:
732                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
733                                            "enabling inode map caching");
734                         break;
735                 case Opt_noinode_cache:
736                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
737                                              "disabling inode map caching");
738                         break;
739                 case Opt_clear_cache:
740                         btrfs_set_and_info(info, CLEAR_CACHE,
741                                            "force clearing of disk cache");
742                         break;
743                 case Opt_user_subvol_rm_allowed:
744                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
745                         break;
746                 case Opt_enospc_debug:
747                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
748                         break;
749                 case Opt_noenospc_debug:
750                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
751                         break;
752                 case Opt_defrag:
753                         btrfs_set_and_info(info, AUTO_DEFRAG,
754                                            "enabling auto defrag");
755                         break;
756                 case Opt_nodefrag:
757                         btrfs_clear_and_info(info, AUTO_DEFRAG,
758                                              "disabling auto defrag");
759                         break;
760                 case Opt_recovery:
761                         btrfs_warn(info,
762                                    "'recovery' is deprecated, use 'usebackuproot' instead");
763                         /* fall through */
764                 case Opt_usebackuproot:
765                         btrfs_info(info,
766                                    "trying to use backup root at mount time");
767                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
768                         break;
769                 case Opt_skip_balance:
770                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
771                         break;
772 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
773                 case Opt_check_integrity_including_extent_data:
774                         btrfs_info(info,
775                                    "enabling check integrity including extent data");
776                         btrfs_set_opt(info->mount_opt,
777                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
778                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
779                         break;
780                 case Opt_check_integrity:
781                         btrfs_info(info, "enabling check integrity");
782                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
783                         break;
784                 case Opt_check_integrity_print_mask:
785                         ret = match_int(&args[0], &intarg);
786                         if (ret)
787                                 goto out;
788                         info->check_integrity_print_mask = intarg;
789                         btrfs_info(info, "check_integrity_print_mask 0x%x",
790                                    info->check_integrity_print_mask);
791                         break;
792 #else
793                 case Opt_check_integrity_including_extent_data:
794                 case Opt_check_integrity:
795                 case Opt_check_integrity_print_mask:
796                         btrfs_err(info,
797                                   "support for check_integrity* not compiled in!");
798                         ret = -EINVAL;
799                         goto out;
800 #endif
801                 case Opt_fatal_errors:
802                         if (strcmp(args[0].from, "panic") == 0)
803                                 btrfs_set_opt(info->mount_opt,
804                                               PANIC_ON_FATAL_ERROR);
805                         else if (strcmp(args[0].from, "bug") == 0)
806                                 btrfs_clear_opt(info->mount_opt,
807                                               PANIC_ON_FATAL_ERROR);
808                         else {
809                                 ret = -EINVAL;
810                                 goto out;
811                         }
812                         break;
813                 case Opt_commit_interval:
814                         intarg = 0;
815                         ret = match_int(&args[0], &intarg);
816                         if (ret)
817                                 goto out;
818                         if (intarg == 0) {
819                                 btrfs_info(info,
820                                            "using default commit interval %us",
821                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
822                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
823                         } else if (intarg > 300) {
824                                 btrfs_warn(info, "excessive commit interval %d",
825                                            intarg);
826                         }
827                         info->commit_interval = intarg;
828                         break;
829 #ifdef CONFIG_BTRFS_DEBUG
830                 case Opt_fragment_all:
831                         btrfs_info(info, "fragmenting all space");
832                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
833                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
834                         break;
835                 case Opt_fragment_metadata:
836                         btrfs_info(info, "fragmenting metadata");
837                         btrfs_set_opt(info->mount_opt,
838                                       FRAGMENT_METADATA);
839                         break;
840                 case Opt_fragment_data:
841                         btrfs_info(info, "fragmenting data");
842                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
843                         break;
844 #endif
845 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
846                 case Opt_ref_verify:
847                         btrfs_info(info, "doing ref verification");
848                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
849                         break;
850 #endif
851                 case Opt_err:
852                         btrfs_info(info, "unrecognized mount option '%s'", p);
853                         ret = -EINVAL;
854                         goto out;
855                 default:
856                         break;
857                 }
858         }
859 check:
860         /*
861          * Extra check for current option against current flag
862          */
863         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & SB_RDONLY)) {
864                 btrfs_err(info,
865                           "nologreplay must be used with ro mount option");
866                 ret = -EINVAL;
867         }
868 out:
869         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
870             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
871             !btrfs_test_opt(info, CLEAR_CACHE)) {
872                 btrfs_err(info, "cannot disable free space tree");
873                 ret = -EINVAL;
874
875         }
876         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
877                 btrfs_info(info, "disk space caching is enabled");
878         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
879                 btrfs_info(info, "using free space tree");
880         return ret;
881 }
882
883 /*
884  * Parse mount options that are required early in the mount process.
885  *
886  * All other options will be parsed on much later in the mount process and
887  * only when we need to allocate a new super block.
888  */
889 static int btrfs_parse_early_options(const char *options, fmode_t flags,
890                 void *holder, struct btrfs_fs_devices **fs_devices)
891 {
892         substring_t args[MAX_OPT_ARGS];
893         char *device_name, *opts, *orig, *p;
894         int error = 0;
895
896         if (!options)
897                 return 0;
898
899         /*
900          * strsep changes the string, duplicate it because btrfs_parse_options
901          * gets called later
902          */
903         opts = kstrdup(options, GFP_KERNEL);
904         if (!opts)
905                 return -ENOMEM;
906         orig = opts;
907
908         while ((p = strsep(&opts, ",")) != NULL) {
909                 int token;
910
911                 if (!*p)
912                         continue;
913
914                 token = match_token(p, tokens, args);
915                 if (token == Opt_device) {
916                         device_name = match_strdup(&args[0]);
917                         if (!device_name) {
918                                 error = -ENOMEM;
919                                 goto out;
920                         }
921                         error = btrfs_scan_one_device(device_name,
922                                         flags, holder, fs_devices);
923                         kfree(device_name);
924                         if (error)
925                                 goto out;
926                 }
927         }
928
929 out:
930         kfree(orig);
931         return error;
932 }
933
934 /*
935  * Parse mount options that are related to subvolume id
936  *
937  * The value is later passed to mount_subvol()
938  */
939 static int btrfs_parse_subvol_options(const char *options, fmode_t flags,
940                 char **subvol_name, u64 *subvol_objectid)
941 {
942         substring_t args[MAX_OPT_ARGS];
943         char *opts, *orig, *p;
944         int error = 0;
945         u64 subvolid;
946
947         if (!options)
948                 return 0;
949
950         /*
951          * strsep changes the string, duplicate it because
952          * btrfs_parse_early_options gets called later
953          */
954         opts = kstrdup(options, GFP_KERNEL);
955         if (!opts)
956                 return -ENOMEM;
957         orig = opts;
958
959         while ((p = strsep(&opts, ",")) != NULL) {
960                 int token;
961                 if (!*p)
962                         continue;
963
964                 token = match_token(p, tokens, args);
965                 switch (token) {
966                 case Opt_subvol:
967                         kfree(*subvol_name);
968                         *subvol_name = match_strdup(&args[0]);
969                         if (!*subvol_name) {
970                                 error = -ENOMEM;
971                                 goto out;
972                         }
973                         break;
974                 case Opt_subvolid:
975                         error = match_u64(&args[0], &subvolid);
976                         if (error)
977                                 goto out;
978
979                         /* we want the original fs_tree */
980                         if (subvolid == 0)
981                                 subvolid = BTRFS_FS_TREE_OBJECTID;
982
983                         *subvol_objectid = subvolid;
984                         break;
985                 case Opt_subvolrootid:
986                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
987                         break;
988                 default:
989                         break;
990                 }
991         }
992
993 out:
994         kfree(orig);
995         return error;
996 }
997
998 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
999                                            u64 subvol_objectid)
1000 {
1001         struct btrfs_root *root = fs_info->tree_root;
1002         struct btrfs_root *fs_root;
1003         struct btrfs_root_ref *root_ref;
1004         struct btrfs_inode_ref *inode_ref;
1005         struct btrfs_key key;
1006         struct btrfs_path *path = NULL;
1007         char *name = NULL, *ptr;
1008         u64 dirid;
1009         int len;
1010         int ret;
1011
1012         path = btrfs_alloc_path();
1013         if (!path) {
1014                 ret = -ENOMEM;
1015                 goto err;
1016         }
1017         path->leave_spinning = 1;
1018
1019         name = kmalloc(PATH_MAX, GFP_KERNEL);
1020         if (!name) {
1021                 ret = -ENOMEM;
1022                 goto err;
1023         }
1024         ptr = name + PATH_MAX - 1;
1025         ptr[0] = '\0';
1026
1027         /*
1028          * Walk up the subvolume trees in the tree of tree roots by root
1029          * backrefs until we hit the top-level subvolume.
1030          */
1031         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1032                 key.objectid = subvol_objectid;
1033                 key.type = BTRFS_ROOT_BACKREF_KEY;
1034                 key.offset = (u64)-1;
1035
1036                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1037                 if (ret < 0) {
1038                         goto err;
1039                 } else if (ret > 0) {
1040                         ret = btrfs_previous_item(root, path, subvol_objectid,
1041                                                   BTRFS_ROOT_BACKREF_KEY);
1042                         if (ret < 0) {
1043                                 goto err;
1044                         } else if (ret > 0) {
1045                                 ret = -ENOENT;
1046                                 goto err;
1047                         }
1048                 }
1049
1050                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1051                 subvol_objectid = key.offset;
1052
1053                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1054                                           struct btrfs_root_ref);
1055                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1056                 ptr -= len + 1;
1057                 if (ptr < name) {
1058                         ret = -ENAMETOOLONG;
1059                         goto err;
1060                 }
1061                 read_extent_buffer(path->nodes[0], ptr + 1,
1062                                    (unsigned long)(root_ref + 1), len);
1063                 ptr[0] = '/';
1064                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1065                 btrfs_release_path(path);
1066
1067                 key.objectid = subvol_objectid;
1068                 key.type = BTRFS_ROOT_ITEM_KEY;
1069                 key.offset = (u64)-1;
1070                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1071                 if (IS_ERR(fs_root)) {
1072                         ret = PTR_ERR(fs_root);
1073                         goto err;
1074                 }
1075
1076                 /*
1077                  * Walk up the filesystem tree by inode refs until we hit the
1078                  * root directory.
1079                  */
1080                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1081                         key.objectid = dirid;
1082                         key.type = BTRFS_INODE_REF_KEY;
1083                         key.offset = (u64)-1;
1084
1085                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1086                         if (ret < 0) {
1087                                 goto err;
1088                         } else if (ret > 0) {
1089                                 ret = btrfs_previous_item(fs_root, path, dirid,
1090                                                           BTRFS_INODE_REF_KEY);
1091                                 if (ret < 0) {
1092                                         goto err;
1093                                 } else if (ret > 0) {
1094                                         ret = -ENOENT;
1095                                         goto err;
1096                                 }
1097                         }
1098
1099                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1100                         dirid = key.offset;
1101
1102                         inode_ref = btrfs_item_ptr(path->nodes[0],
1103                                                    path->slots[0],
1104                                                    struct btrfs_inode_ref);
1105                         len = btrfs_inode_ref_name_len(path->nodes[0],
1106                                                        inode_ref);
1107                         ptr -= len + 1;
1108                         if (ptr < name) {
1109                                 ret = -ENAMETOOLONG;
1110                                 goto err;
1111                         }
1112                         read_extent_buffer(path->nodes[0], ptr + 1,
1113                                            (unsigned long)(inode_ref + 1), len);
1114                         ptr[0] = '/';
1115                         btrfs_release_path(path);
1116                 }
1117         }
1118
1119         btrfs_free_path(path);
1120         if (ptr == name + PATH_MAX - 1) {
1121                 name[0] = '/';
1122                 name[1] = '\0';
1123         } else {
1124                 memmove(name, ptr, name + PATH_MAX - ptr);
1125         }
1126         return name;
1127
1128 err:
1129         btrfs_free_path(path);
1130         kfree(name);
1131         return ERR_PTR(ret);
1132 }
1133
1134 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1135 {
1136         struct btrfs_root *root = fs_info->tree_root;
1137         struct btrfs_dir_item *di;
1138         struct btrfs_path *path;
1139         struct btrfs_key location;
1140         u64 dir_id;
1141
1142         path = btrfs_alloc_path();
1143         if (!path)
1144                 return -ENOMEM;
1145         path->leave_spinning = 1;
1146
1147         /*
1148          * Find the "default" dir item which points to the root item that we
1149          * will mount by default if we haven't been given a specific subvolume
1150          * to mount.
1151          */
1152         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1153         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1154         if (IS_ERR(di)) {
1155                 btrfs_free_path(path);
1156                 return PTR_ERR(di);
1157         }
1158         if (!di) {
1159                 /*
1160                  * Ok the default dir item isn't there.  This is weird since
1161                  * it's always been there, but don't freak out, just try and
1162                  * mount the top-level subvolume.
1163                  */
1164                 btrfs_free_path(path);
1165                 *objectid = BTRFS_FS_TREE_OBJECTID;
1166                 return 0;
1167         }
1168
1169         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1170         btrfs_free_path(path);
1171         *objectid = location.objectid;
1172         return 0;
1173 }
1174
1175 static int btrfs_fill_super(struct super_block *sb,
1176                             struct btrfs_fs_devices *fs_devices,
1177                             void *data)
1178 {
1179         struct inode *inode;
1180         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1181         struct btrfs_key key;
1182         int err;
1183
1184         sb->s_maxbytes = MAX_LFS_FILESIZE;
1185         sb->s_magic = BTRFS_SUPER_MAGIC;
1186         sb->s_op = &btrfs_super_ops;
1187         sb->s_d_op = &btrfs_dentry_operations;
1188         sb->s_export_op = &btrfs_export_ops;
1189         sb->s_xattr = btrfs_xattr_handlers;
1190         sb->s_time_gran = 1;
1191 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1192         sb->s_flags |= SB_POSIXACL;
1193 #endif
1194         sb->s_flags |= SB_I_VERSION;
1195         sb->s_iflags |= SB_I_CGROUPWB;
1196
1197         err = super_setup_bdi(sb);
1198         if (err) {
1199                 btrfs_err(fs_info, "super_setup_bdi failed");
1200                 return err;
1201         }
1202
1203         err = open_ctree(sb, fs_devices, (char *)data);
1204         if (err) {
1205                 btrfs_err(fs_info, "open_ctree failed");
1206                 return err;
1207         }
1208
1209         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1210         key.type = BTRFS_INODE_ITEM_KEY;
1211         key.offset = 0;
1212         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1213         if (IS_ERR(inode)) {
1214                 err = PTR_ERR(inode);
1215                 goto fail_close;
1216         }
1217
1218         sb->s_root = d_make_root(inode);
1219         if (!sb->s_root) {
1220                 err = -ENOMEM;
1221                 goto fail_close;
1222         }
1223
1224         cleancache_init_fs(sb);
1225         sb->s_flags |= SB_ACTIVE;
1226         return 0;
1227
1228 fail_close:
1229         close_ctree(fs_info);
1230         return err;
1231 }
1232
1233 int btrfs_sync_fs(struct super_block *sb, int wait)
1234 {
1235         struct btrfs_trans_handle *trans;
1236         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1237         struct btrfs_root *root = fs_info->tree_root;
1238
1239         trace_btrfs_sync_fs(fs_info, wait);
1240
1241         if (!wait) {
1242                 filemap_flush(fs_info->btree_inode->i_mapping);
1243                 return 0;
1244         }
1245
1246         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1247
1248         trans = btrfs_attach_transaction_barrier(root);
1249         if (IS_ERR(trans)) {
1250                 /* no transaction, don't bother */
1251                 if (PTR_ERR(trans) == -ENOENT) {
1252                         /*
1253                          * Exit unless we have some pending changes
1254                          * that need to go through commit
1255                          */
1256                         if (fs_info->pending_changes == 0)
1257                                 return 0;
1258                         /*
1259                          * A non-blocking test if the fs is frozen. We must not
1260                          * start a new transaction here otherwise a deadlock
1261                          * happens. The pending operations are delayed to the
1262                          * next commit after thawing.
1263                          */
1264                         if (sb_start_write_trylock(sb))
1265                                 sb_end_write(sb);
1266                         else
1267                                 return 0;
1268                         trans = btrfs_start_transaction(root, 0);
1269                 }
1270                 if (IS_ERR(trans))
1271                         return PTR_ERR(trans);
1272         }
1273         return btrfs_commit_transaction(trans);
1274 }
1275
1276 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1277 {
1278         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1279         const char *compress_type;
1280
1281         if (btrfs_test_opt(info, DEGRADED))
1282                 seq_puts(seq, ",degraded");
1283         if (btrfs_test_opt(info, NODATASUM))
1284                 seq_puts(seq, ",nodatasum");
1285         if (btrfs_test_opt(info, NODATACOW))
1286                 seq_puts(seq, ",nodatacow");
1287         if (btrfs_test_opt(info, NOBARRIER))
1288                 seq_puts(seq, ",nobarrier");
1289         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1290                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1291         if (info->thread_pool_size !=  min_t(unsigned long,
1292                                              num_online_cpus() + 2, 8))
1293                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1294         if (btrfs_test_opt(info, COMPRESS)) {
1295                 compress_type = btrfs_compress_type2str(info->compress_type);
1296                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1297                         seq_printf(seq, ",compress-force=%s", compress_type);
1298                 else
1299                         seq_printf(seq, ",compress=%s", compress_type);
1300                 if (info->compress_level)
1301                         seq_printf(seq, ":%d", info->compress_level);
1302         }
1303         if (btrfs_test_opt(info, NOSSD))
1304                 seq_puts(seq, ",nossd");
1305         if (btrfs_test_opt(info, SSD_SPREAD))
1306                 seq_puts(seq, ",ssd_spread");
1307         else if (btrfs_test_opt(info, SSD))
1308                 seq_puts(seq, ",ssd");
1309         if (btrfs_test_opt(info, NOTREELOG))
1310                 seq_puts(seq, ",notreelog");
1311         if (btrfs_test_opt(info, NOLOGREPLAY))
1312                 seq_puts(seq, ",nologreplay");
1313         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1314                 seq_puts(seq, ",flushoncommit");
1315         if (btrfs_test_opt(info, DISCARD))
1316                 seq_puts(seq, ",discard");
1317         if (!(info->sb->s_flags & SB_POSIXACL))
1318                 seq_puts(seq, ",noacl");
1319         if (btrfs_test_opt(info, SPACE_CACHE))
1320                 seq_puts(seq, ",space_cache");
1321         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1322                 seq_puts(seq, ",space_cache=v2");
1323         else
1324                 seq_puts(seq, ",nospace_cache");
1325         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1326                 seq_puts(seq, ",rescan_uuid_tree");
1327         if (btrfs_test_opt(info, CLEAR_CACHE))
1328                 seq_puts(seq, ",clear_cache");
1329         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1330                 seq_puts(seq, ",user_subvol_rm_allowed");
1331         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1332                 seq_puts(seq, ",enospc_debug");
1333         if (btrfs_test_opt(info, AUTO_DEFRAG))
1334                 seq_puts(seq, ",autodefrag");
1335         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1336                 seq_puts(seq, ",inode_cache");
1337         if (btrfs_test_opt(info, SKIP_BALANCE))
1338                 seq_puts(seq, ",skip_balance");
1339 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1340         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1341                 seq_puts(seq, ",check_int_data");
1342         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1343                 seq_puts(seq, ",check_int");
1344         if (info->check_integrity_print_mask)
1345                 seq_printf(seq, ",check_int_print_mask=%d",
1346                                 info->check_integrity_print_mask);
1347 #endif
1348         if (info->metadata_ratio)
1349                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1350         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1351                 seq_puts(seq, ",fatal_errors=panic");
1352         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1353                 seq_printf(seq, ",commit=%u", info->commit_interval);
1354 #ifdef CONFIG_BTRFS_DEBUG
1355         if (btrfs_test_opt(info, FRAGMENT_DATA))
1356                 seq_puts(seq, ",fragment=data");
1357         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1358                 seq_puts(seq, ",fragment=metadata");
1359 #endif
1360         if (btrfs_test_opt(info, REF_VERIFY))
1361                 seq_puts(seq, ",ref_verify");
1362         seq_printf(seq, ",subvolid=%llu",
1363                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1364         seq_puts(seq, ",subvol=");
1365         seq_dentry(seq, dentry, " \t\n\\");
1366         return 0;
1367 }
1368
1369 static int btrfs_test_super(struct super_block *s, void *data)
1370 {
1371         struct btrfs_fs_info *p = data;
1372         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1373
1374         return fs_info->fs_devices == p->fs_devices;
1375 }
1376
1377 static int btrfs_set_super(struct super_block *s, void *data)
1378 {
1379         int err = set_anon_super(s, data);
1380         if (!err)
1381                 s->s_fs_info = data;
1382         return err;
1383 }
1384
1385 /*
1386  * subvolumes are identified by ino 256
1387  */
1388 static inline int is_subvolume_inode(struct inode *inode)
1389 {
1390         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1391                 return 1;
1392         return 0;
1393 }
1394
1395 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1396                                    const char *device_name, struct vfsmount *mnt)
1397 {
1398         struct dentry *root;
1399         int ret;
1400
1401         if (!subvol_name) {
1402                 if (!subvol_objectid) {
1403                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1404                                                           &subvol_objectid);
1405                         if (ret) {
1406                                 root = ERR_PTR(ret);
1407                                 goto out;
1408                         }
1409                 }
1410                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1411                                                             subvol_objectid);
1412                 if (IS_ERR(subvol_name)) {
1413                         root = ERR_CAST(subvol_name);
1414                         subvol_name = NULL;
1415                         goto out;
1416                 }
1417
1418         }
1419
1420         root = mount_subtree(mnt, subvol_name);
1421         /* mount_subtree() drops our reference on the vfsmount. */
1422         mnt = NULL;
1423
1424         if (!IS_ERR(root)) {
1425                 struct super_block *s = root->d_sb;
1426                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1427                 struct inode *root_inode = d_inode(root);
1428                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1429
1430                 ret = 0;
1431                 if (!is_subvolume_inode(root_inode)) {
1432                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1433                                subvol_name);
1434                         ret = -EINVAL;
1435                 }
1436                 if (subvol_objectid && root_objectid != subvol_objectid) {
1437                         /*
1438                          * This will also catch a race condition where a
1439                          * subvolume which was passed by ID is renamed and
1440                          * another subvolume is renamed over the old location.
1441                          */
1442                         btrfs_err(fs_info,
1443                                   "subvol '%s' does not match subvolid %llu",
1444                                   subvol_name, subvol_objectid);
1445                         ret = -EINVAL;
1446                 }
1447                 if (ret) {
1448                         dput(root);
1449                         root = ERR_PTR(ret);
1450                         deactivate_locked_super(s);
1451                 }
1452         }
1453
1454 out:
1455         mntput(mnt);
1456         kfree(subvol_name);
1457         return root;
1458 }
1459
1460 static int parse_security_options(char *orig_opts,
1461                                   struct security_mnt_opts *sec_opts)
1462 {
1463         char *secdata = NULL;
1464         int ret = 0;
1465
1466         secdata = alloc_secdata();
1467         if (!secdata)
1468                 return -ENOMEM;
1469         ret = security_sb_copy_data(orig_opts, secdata);
1470         if (ret) {
1471                 free_secdata(secdata);
1472                 return ret;
1473         }
1474         ret = security_sb_parse_opts_str(secdata, sec_opts);
1475         free_secdata(secdata);
1476         return ret;
1477 }
1478
1479 static int setup_security_options(struct btrfs_fs_info *fs_info,
1480                                   struct super_block *sb,
1481                                   struct security_mnt_opts *sec_opts)
1482 {
1483         int ret = 0;
1484
1485         /*
1486          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1487          * is valid.
1488          */
1489         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1490         if (ret)
1491                 return ret;
1492
1493 #ifdef CONFIG_SECURITY
1494         if (!fs_info->security_opts.num_mnt_opts) {
1495                 /* first time security setup, copy sec_opts to fs_info */
1496                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1497         } else {
1498                 /*
1499                  * Since SELinux (the only one supporting security_mnt_opts)
1500                  * does NOT support changing context during remount/mount of
1501                  * the same sb, this must be the same or part of the same
1502                  * security options, just free it.
1503                  */
1504                 security_free_mnt_opts(sec_opts);
1505         }
1506 #endif
1507         return ret;
1508 }
1509
1510 /*
1511  * Find a superblock for the given device / mount point.
1512  *
1513  * Note: This is based on mount_bdev from fs/super.c with a few additions
1514  *       for multiple device setup.  Make sure to keep it in sync.
1515  */
1516 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1517                 int flags, const char *device_name, void *data)
1518 {
1519         struct block_device *bdev = NULL;
1520         struct super_block *s;
1521         struct btrfs_fs_devices *fs_devices = NULL;
1522         struct btrfs_fs_info *fs_info = NULL;
1523         struct security_mnt_opts new_sec_opts;
1524         fmode_t mode = FMODE_READ;
1525         int error = 0;
1526
1527         if (!(flags & SB_RDONLY))
1528                 mode |= FMODE_WRITE;
1529
1530         error = btrfs_parse_early_options(data, mode, fs_type,
1531                                           &fs_devices);
1532         if (error) {
1533                 return ERR_PTR(error);
1534         }
1535
1536         security_init_mnt_opts(&new_sec_opts);
1537         if (data) {
1538                 error = parse_security_options(data, &new_sec_opts);
1539                 if (error)
1540                         return ERR_PTR(error);
1541         }
1542
1543         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1544         if (error)
1545                 goto error_sec_opts;
1546
1547         /*
1548          * Setup a dummy root and fs_info for test/set super.  This is because
1549          * we don't actually fill this stuff out until open_ctree, but we need
1550          * it for searching for existing supers, so this lets us do that and
1551          * then open_ctree will properly initialize everything later.
1552          */
1553         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1554         if (!fs_info) {
1555                 error = -ENOMEM;
1556                 goto error_sec_opts;
1557         }
1558
1559         fs_info->fs_devices = fs_devices;
1560
1561         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1562         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1563         security_init_mnt_opts(&fs_info->security_opts);
1564         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1565                 error = -ENOMEM;
1566                 goto error_fs_info;
1567         }
1568
1569         error = btrfs_open_devices(fs_devices, mode, fs_type);
1570         if (error)
1571                 goto error_fs_info;
1572
1573         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1574                 error = -EACCES;
1575                 goto error_close_devices;
1576         }
1577
1578         bdev = fs_devices->latest_bdev;
1579         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1580                  fs_info);
1581         if (IS_ERR(s)) {
1582                 error = PTR_ERR(s);
1583                 goto error_close_devices;
1584         }
1585
1586         if (s->s_root) {
1587                 btrfs_close_devices(fs_devices);
1588                 free_fs_info(fs_info);
1589                 if ((flags ^ s->s_flags) & SB_RDONLY)
1590                         error = -EBUSY;
1591         } else {
1592                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1593                 btrfs_sb(s)->bdev_holder = fs_type;
1594                 error = btrfs_fill_super(s, fs_devices, data);
1595         }
1596         if (error) {
1597                 deactivate_locked_super(s);
1598                 goto error_sec_opts;
1599         }
1600
1601         fs_info = btrfs_sb(s);
1602         error = setup_security_options(fs_info, s, &new_sec_opts);
1603         if (error) {
1604                 deactivate_locked_super(s);
1605                 goto error_sec_opts;
1606         }
1607
1608         return dget(s->s_root);
1609
1610 error_close_devices:
1611         btrfs_close_devices(fs_devices);
1612 error_fs_info:
1613         free_fs_info(fs_info);
1614 error_sec_opts:
1615         security_free_mnt_opts(&new_sec_opts);
1616         return ERR_PTR(error);
1617 }
1618
1619 /*
1620  * Mount function which is called by VFS layer.
1621  *
1622  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1623  * which needs vfsmount* of device's root (/).  This means device's root has to
1624  * be mounted internally in any case.
1625  *
1626  * Operation flow:
1627  *   1. Parse subvol id related options for later use in mount_subvol().
1628  *
1629  *   2. Mount device's root (/) by calling vfs_kern_mount().
1630  *
1631  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1632  *      first place. In order to avoid calling btrfs_mount() again, we use
1633  *      different file_system_type which is not registered to VFS by
1634  *      register_filesystem() (btrfs_root_fs_type). As a result,
1635  *      btrfs_mount_root() is called. The return value will be used by
1636  *      mount_subtree() in mount_subvol().
1637  *
1638  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1639  *      "btrfs subvolume set-default", mount_subvol() is called always.
1640  */
1641 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1642                 const char *device_name, void *data)
1643 {
1644         struct vfsmount *mnt_root;
1645         struct dentry *root;
1646         fmode_t mode = FMODE_READ;
1647         char *subvol_name = NULL;
1648         u64 subvol_objectid = 0;
1649         int error = 0;
1650
1651         if (!(flags & SB_RDONLY))
1652                 mode |= FMODE_WRITE;
1653
1654         error = btrfs_parse_subvol_options(data, mode,
1655                                           &subvol_name, &subvol_objectid);
1656         if (error) {
1657                 kfree(subvol_name);
1658                 return ERR_PTR(error);
1659         }
1660
1661         /* mount device's root (/) */
1662         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1663         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1664                 if (flags & SB_RDONLY) {
1665                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1666                                 flags & ~SB_RDONLY, device_name, data);
1667                 } else {
1668                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1669                                 flags | SB_RDONLY, device_name, data);
1670                         if (IS_ERR(mnt_root)) {
1671                                 root = ERR_CAST(mnt_root);
1672                                 goto out;
1673                         }
1674
1675                         down_write(&mnt_root->mnt_sb->s_umount);
1676                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1677                         up_write(&mnt_root->mnt_sb->s_umount);
1678                         if (error < 0) {
1679                                 root = ERR_PTR(error);
1680                                 mntput(mnt_root);
1681                                 goto out;
1682                         }
1683                 }
1684         }
1685         if (IS_ERR(mnt_root)) {
1686                 root = ERR_CAST(mnt_root);
1687                 goto out;
1688         }
1689
1690         /* mount_subvol() will free subvol_name and mnt_root */
1691         root = mount_subvol(subvol_name, subvol_objectid, device_name, mnt_root);
1692
1693 out:
1694         return root;
1695 }
1696
1697 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1698                                      u32 new_pool_size, u32 old_pool_size)
1699 {
1700         if (new_pool_size == old_pool_size)
1701                 return;
1702
1703         fs_info->thread_pool_size = new_pool_size;
1704
1705         btrfs_info(fs_info, "resize thread pool %d -> %d",
1706                old_pool_size, new_pool_size);
1707
1708         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1709         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1710         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1711         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1712         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1713         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1714         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1715                                 new_pool_size);
1716         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1717         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1718         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1719         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1720         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1721                                 new_pool_size);
1722 }
1723
1724 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1725 {
1726         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1727 }
1728
1729 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1730                                        unsigned long old_opts, int flags)
1731 {
1732         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1733             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1734              (flags & SB_RDONLY))) {
1735                 /* wait for any defraggers to finish */
1736                 wait_event(fs_info->transaction_wait,
1737                            (atomic_read(&fs_info->defrag_running) == 0));
1738                 if (flags & SB_RDONLY)
1739                         sync_filesystem(fs_info->sb);
1740         }
1741 }
1742
1743 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1744                                          unsigned long old_opts)
1745 {
1746         /*
1747          * We need to cleanup all defragable inodes if the autodefragment is
1748          * close or the filesystem is read only.
1749          */
1750         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1751             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1752                 btrfs_cleanup_defrag_inodes(fs_info);
1753         }
1754
1755         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1756 }
1757
1758 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1759 {
1760         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1761         struct btrfs_root *root = fs_info->tree_root;
1762         unsigned old_flags = sb->s_flags;
1763         unsigned long old_opts = fs_info->mount_opt;
1764         unsigned long old_compress_type = fs_info->compress_type;
1765         u64 old_max_inline = fs_info->max_inline;
1766         u32 old_thread_pool_size = fs_info->thread_pool_size;
1767         u32 old_metadata_ratio = fs_info->metadata_ratio;
1768         int ret;
1769
1770         sync_filesystem(sb);
1771         btrfs_remount_prepare(fs_info);
1772
1773         if (data) {
1774                 struct security_mnt_opts new_sec_opts;
1775
1776                 security_init_mnt_opts(&new_sec_opts);
1777                 ret = parse_security_options(data, &new_sec_opts);
1778                 if (ret)
1779                         goto restore;
1780                 ret = setup_security_options(fs_info, sb,
1781                                              &new_sec_opts);
1782                 if (ret) {
1783                         security_free_mnt_opts(&new_sec_opts);
1784                         goto restore;
1785                 }
1786         }
1787
1788         ret = btrfs_parse_options(fs_info, data, *flags);
1789         if (ret)
1790                 goto restore;
1791
1792         btrfs_remount_begin(fs_info, old_opts, *flags);
1793         btrfs_resize_thread_pool(fs_info,
1794                 fs_info->thread_pool_size, old_thread_pool_size);
1795
1796         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1797                 goto out;
1798
1799         if (*flags & SB_RDONLY) {
1800                 /*
1801                  * this also happens on 'umount -rf' or on shutdown, when
1802                  * the filesystem is busy.
1803                  */
1804                 cancel_work_sync(&fs_info->async_reclaim_work);
1805
1806                 /* wait for the uuid_scan task to finish */
1807                 down(&fs_info->uuid_tree_rescan_sem);
1808                 /* avoid complains from lockdep et al. */
1809                 up(&fs_info->uuid_tree_rescan_sem);
1810
1811                 sb->s_flags |= SB_RDONLY;
1812
1813                 /*
1814                  * Setting SB_RDONLY will put the cleaner thread to
1815                  * sleep at the next loop if it's already active.
1816                  * If it's already asleep, we'll leave unused block
1817                  * groups on disk until we're mounted read-write again
1818                  * unless we clean them up here.
1819                  */
1820                 btrfs_delete_unused_bgs(fs_info);
1821
1822                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1823                 btrfs_scrub_cancel(fs_info);
1824                 btrfs_pause_balance(fs_info);
1825
1826                 ret = btrfs_commit_super(fs_info);
1827                 if (ret)
1828                         goto restore;
1829         } else {
1830                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1831                         btrfs_err(fs_info,
1832                                 "Remounting read-write after error is not allowed");
1833                         ret = -EINVAL;
1834                         goto restore;
1835                 }
1836                 if (fs_info->fs_devices->rw_devices == 0) {
1837                         ret = -EACCES;
1838                         goto restore;
1839                 }
1840
1841                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1842                         btrfs_warn(fs_info,
1843                                 "too many missing devices, writeable remount is not allowed");
1844                         ret = -EACCES;
1845                         goto restore;
1846                 }
1847
1848                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1849                         ret = -EINVAL;
1850                         goto restore;
1851                 }
1852
1853                 ret = btrfs_cleanup_fs_roots(fs_info);
1854                 if (ret)
1855                         goto restore;
1856
1857                 /* recover relocation */
1858                 mutex_lock(&fs_info->cleaner_mutex);
1859                 ret = btrfs_recover_relocation(root);
1860                 mutex_unlock(&fs_info->cleaner_mutex);
1861                 if (ret)
1862                         goto restore;
1863
1864                 ret = btrfs_resume_balance_async(fs_info);
1865                 if (ret)
1866                         goto restore;
1867
1868                 ret = btrfs_resume_dev_replace_async(fs_info);
1869                 if (ret) {
1870                         btrfs_warn(fs_info, "failed to resume dev_replace");
1871                         goto restore;
1872                 }
1873
1874                 btrfs_qgroup_rescan_resume(fs_info);
1875
1876                 if (!fs_info->uuid_root) {
1877                         btrfs_info(fs_info, "creating UUID tree");
1878                         ret = btrfs_create_uuid_tree(fs_info);
1879                         if (ret) {
1880                                 btrfs_warn(fs_info,
1881                                            "failed to create the UUID tree %d",
1882                                            ret);
1883                                 goto restore;
1884                         }
1885                 }
1886                 sb->s_flags &= ~SB_RDONLY;
1887
1888                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1889         }
1890 out:
1891         wake_up_process(fs_info->transaction_kthread);
1892         btrfs_remount_cleanup(fs_info, old_opts);
1893         return 0;
1894
1895 restore:
1896         /* We've hit an error - don't reset SB_RDONLY */
1897         if (sb_rdonly(sb))
1898                 old_flags |= SB_RDONLY;
1899         sb->s_flags = old_flags;
1900         fs_info->mount_opt = old_opts;
1901         fs_info->compress_type = old_compress_type;
1902         fs_info->max_inline = old_max_inline;
1903         btrfs_resize_thread_pool(fs_info,
1904                 old_thread_pool_size, fs_info->thread_pool_size);
1905         fs_info->metadata_ratio = old_metadata_ratio;
1906         btrfs_remount_cleanup(fs_info, old_opts);
1907         return ret;
1908 }
1909
1910 /* Used to sort the devices by max_avail(descending sort) */
1911 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1912                                        const void *dev_info2)
1913 {
1914         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1915             ((struct btrfs_device_info *)dev_info2)->max_avail)
1916                 return -1;
1917         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1918                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1919                 return 1;
1920         else
1921         return 0;
1922 }
1923
1924 /*
1925  * sort the devices by max_avail, in which max free extent size of each device
1926  * is stored.(Descending Sort)
1927  */
1928 static inline void btrfs_descending_sort_devices(
1929                                         struct btrfs_device_info *devices,
1930                                         size_t nr_devices)
1931 {
1932         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1933              btrfs_cmp_device_free_bytes, NULL);
1934 }
1935
1936 /*
1937  * The helper to calc the free space on the devices that can be used to store
1938  * file data.
1939  */
1940 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1941                                        u64 *free_bytes)
1942 {
1943         struct btrfs_device_info *devices_info;
1944         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1945         struct btrfs_device *device;
1946         u64 skip_space;
1947         u64 type;
1948         u64 avail_space;
1949         u64 min_stripe_size;
1950         int min_stripes = 1, num_stripes = 1;
1951         int i = 0, nr_devices;
1952
1953         /*
1954          * We aren't under the device list lock, so this is racy-ish, but good
1955          * enough for our purposes.
1956          */
1957         nr_devices = fs_info->fs_devices->open_devices;
1958         if (!nr_devices) {
1959                 smp_mb();
1960                 nr_devices = fs_info->fs_devices->open_devices;
1961                 ASSERT(nr_devices);
1962                 if (!nr_devices) {
1963                         *free_bytes = 0;
1964                         return 0;
1965                 }
1966         }
1967
1968         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1969                                GFP_KERNEL);
1970         if (!devices_info)
1971                 return -ENOMEM;
1972
1973         /* calc min stripe number for data space allocation */
1974         type = btrfs_data_alloc_profile(fs_info);
1975         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1976                 min_stripes = 2;
1977                 num_stripes = nr_devices;
1978         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1979                 min_stripes = 2;
1980                 num_stripes = 2;
1981         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1982                 min_stripes = 4;
1983                 num_stripes = 4;
1984         }
1985
1986         if (type & BTRFS_BLOCK_GROUP_DUP)
1987                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1988         else
1989                 min_stripe_size = BTRFS_STRIPE_LEN;
1990
1991         rcu_read_lock();
1992         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1993                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1994                                                 &device->dev_state) ||
1995                     !device->bdev ||
1996                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1997                         continue;
1998
1999                 if (i >= nr_devices)
2000                         break;
2001
2002                 avail_space = device->total_bytes - device->bytes_used;
2003
2004                 /* align with stripe_len */
2005                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
2006                 avail_space *= BTRFS_STRIPE_LEN;
2007
2008                 /*
2009                  * In order to avoid overwriting the superblock on the drive,
2010                  * btrfs starts at an offset of at least 1MB when doing chunk
2011                  * allocation.
2012                  */
2013                 skip_space = SZ_1M;
2014
2015                 /*
2016                  * we can use the free space in [0, skip_space - 1], subtract
2017                  * it from the total.
2018                  */
2019                 if (avail_space && avail_space >= skip_space)
2020                         avail_space -= skip_space;
2021                 else
2022                         avail_space = 0;
2023
2024                 if (avail_space < min_stripe_size)
2025                         continue;
2026
2027                 devices_info[i].dev = device;
2028                 devices_info[i].max_avail = avail_space;
2029
2030                 i++;
2031         }
2032         rcu_read_unlock();
2033
2034         nr_devices = i;
2035
2036         btrfs_descending_sort_devices(devices_info, nr_devices);
2037
2038         i = nr_devices - 1;
2039         avail_space = 0;
2040         while (nr_devices >= min_stripes) {
2041                 if (num_stripes > nr_devices)
2042                         num_stripes = nr_devices;
2043
2044                 if (devices_info[i].max_avail >= min_stripe_size) {
2045                         int j;
2046                         u64 alloc_size;
2047
2048                         avail_space += devices_info[i].max_avail * num_stripes;
2049                         alloc_size = devices_info[i].max_avail;
2050                         for (j = i + 1 - num_stripes; j <= i; j++)
2051                                 devices_info[j].max_avail -= alloc_size;
2052                 }
2053                 i--;
2054                 nr_devices--;
2055         }
2056
2057         kfree(devices_info);
2058         *free_bytes = avail_space;
2059         return 0;
2060 }
2061
2062 /*
2063  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2064  *
2065  * If there's a redundant raid level at DATA block groups, use the respective
2066  * multiplier to scale the sizes.
2067  *
2068  * Unused device space usage is based on simulating the chunk allocator
2069  * algorithm that respects the device sizes and order of allocations.  This is
2070  * a close approximation of the actual use but there are other factors that may
2071  * change the result (like a new metadata chunk).
2072  *
2073  * If metadata is exhausted, f_bavail will be 0.
2074  */
2075 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2076 {
2077         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2078         struct btrfs_super_block *disk_super = fs_info->super_copy;
2079         struct list_head *head = &fs_info->space_info;
2080         struct btrfs_space_info *found;
2081         u64 total_used = 0;
2082         u64 total_free_data = 0;
2083         u64 total_free_meta = 0;
2084         int bits = dentry->d_sb->s_blocksize_bits;
2085         __be32 *fsid = (__be32 *)fs_info->fsid;
2086         unsigned factor = 1;
2087         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2088         int ret;
2089         u64 thresh = 0;
2090         int mixed = 0;
2091
2092         rcu_read_lock();
2093         list_for_each_entry_rcu(found, head, list) {
2094                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2095                         int i;
2096
2097                         total_free_data += found->disk_total - found->disk_used;
2098                         total_free_data -=
2099                                 btrfs_account_ro_block_groups_free_space(found);
2100
2101                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2102                                 if (!list_empty(&found->block_groups[i])) {
2103                                         switch (i) {
2104                                         case BTRFS_RAID_DUP:
2105                                         case BTRFS_RAID_RAID1:
2106                                         case BTRFS_RAID_RAID10:
2107                                                 factor = 2;
2108                                         }
2109                                 }
2110                         }
2111                 }
2112
2113                 /*
2114                  * Metadata in mixed block goup profiles are accounted in data
2115                  */
2116                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2117                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2118                                 mixed = 1;
2119                         else
2120                                 total_free_meta += found->disk_total -
2121                                         found->disk_used;
2122                 }
2123
2124                 total_used += found->disk_used;
2125         }
2126
2127         rcu_read_unlock();
2128
2129         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2130         buf->f_blocks >>= bits;
2131         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2132
2133         /* Account global block reserve as used, it's in logical size already */
2134         spin_lock(&block_rsv->lock);
2135         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2136         if (buf->f_bfree >= block_rsv->size >> bits)
2137                 buf->f_bfree -= block_rsv->size >> bits;
2138         else
2139                 buf->f_bfree = 0;
2140         spin_unlock(&block_rsv->lock);
2141
2142         buf->f_bavail = div_u64(total_free_data, factor);
2143         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2144         if (ret)
2145                 return ret;
2146         buf->f_bavail += div_u64(total_free_data, factor);
2147         buf->f_bavail = buf->f_bavail >> bits;
2148
2149         /*
2150          * We calculate the remaining metadata space minus global reserve. If
2151          * this is (supposedly) smaller than zero, there's no space. But this
2152          * does not hold in practice, the exhausted state happens where's still
2153          * some positive delta. So we apply some guesswork and compare the
2154          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2155          *
2156          * We probably cannot calculate the exact threshold value because this
2157          * depends on the internal reservations requested by various
2158          * operations, so some operations that consume a few metadata will
2159          * succeed even if the Avail is zero. But this is better than the other
2160          * way around.
2161          */
2162         thresh = SZ_4M;
2163
2164         if (!mixed && total_free_meta - thresh < block_rsv->size)
2165                 buf->f_bavail = 0;
2166
2167         buf->f_type = BTRFS_SUPER_MAGIC;
2168         buf->f_bsize = dentry->d_sb->s_blocksize;
2169         buf->f_namelen = BTRFS_NAME_LEN;
2170
2171         /* We treat it as constant endianness (it doesn't matter _which_)
2172            because we want the fsid to come out the same whether mounted
2173            on a big-endian or little-endian host */
2174         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2175         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2176         /* Mask in the root object ID too, to disambiguate subvols */
2177         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2178         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2179
2180         return 0;
2181 }
2182
2183 static void btrfs_kill_super(struct super_block *sb)
2184 {
2185         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2186         kill_anon_super(sb);
2187         free_fs_info(fs_info);
2188 }
2189
2190 static struct file_system_type btrfs_fs_type = {
2191         .owner          = THIS_MODULE,
2192         .name           = "btrfs",
2193         .mount          = btrfs_mount,
2194         .kill_sb        = btrfs_kill_super,
2195         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2196 };
2197
2198 static struct file_system_type btrfs_root_fs_type = {
2199         .owner          = THIS_MODULE,
2200         .name           = "btrfs",
2201         .mount          = btrfs_mount_root,
2202         .kill_sb        = btrfs_kill_super,
2203         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2204 };
2205
2206 MODULE_ALIAS_FS("btrfs");
2207
2208 static int btrfs_control_open(struct inode *inode, struct file *file)
2209 {
2210         /*
2211          * The control file's private_data is used to hold the
2212          * transaction when it is started and is used to keep
2213          * track of whether a transaction is already in progress.
2214          */
2215         file->private_data = NULL;
2216         return 0;
2217 }
2218
2219 /*
2220  * used by btrfsctl to scan devices when no FS is mounted
2221  */
2222 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2223                                 unsigned long arg)
2224 {
2225         struct btrfs_ioctl_vol_args *vol;
2226         struct btrfs_fs_devices *fs_devices;
2227         int ret = -ENOTTY;
2228
2229         if (!capable(CAP_SYS_ADMIN))
2230                 return -EPERM;
2231
2232         vol = memdup_user((void __user *)arg, sizeof(*vol));
2233         if (IS_ERR(vol))
2234                 return PTR_ERR(vol);
2235
2236         switch (cmd) {
2237         case BTRFS_IOC_SCAN_DEV:
2238                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2239                                             &btrfs_root_fs_type, &fs_devices);
2240                 break;
2241         case BTRFS_IOC_DEVICES_READY:
2242                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2243                                             &btrfs_root_fs_type, &fs_devices);
2244                 if (ret)
2245                         break;
2246                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2247                 break;
2248         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2249                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2250                 break;
2251         }
2252
2253         kfree(vol);
2254         return ret;
2255 }
2256
2257 static int btrfs_freeze(struct super_block *sb)
2258 {
2259         struct btrfs_trans_handle *trans;
2260         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2261         struct btrfs_root *root = fs_info->tree_root;
2262
2263         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2264         /*
2265          * We don't need a barrier here, we'll wait for any transaction that
2266          * could be in progress on other threads (and do delayed iputs that
2267          * we want to avoid on a frozen filesystem), or do the commit
2268          * ourselves.
2269          */
2270         trans = btrfs_attach_transaction_barrier(root);
2271         if (IS_ERR(trans)) {
2272                 /* no transaction, don't bother */
2273                 if (PTR_ERR(trans) == -ENOENT)
2274                         return 0;
2275                 return PTR_ERR(trans);
2276         }
2277         return btrfs_commit_transaction(trans);
2278 }
2279
2280 static int btrfs_unfreeze(struct super_block *sb)
2281 {
2282         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2283
2284         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2285         return 0;
2286 }
2287
2288 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2289 {
2290         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2291         struct btrfs_fs_devices *cur_devices;
2292         struct btrfs_device *dev, *first_dev = NULL;
2293         struct list_head *head;
2294         struct rcu_string *name;
2295
2296         /*
2297          * Lightweight locking of the devices. We should not need
2298          * device_list_mutex here as we only read the device data and the list
2299          * is protected by RCU.  Even if a device is deleted during the list
2300          * traversals, we'll get valid data, the freeing callback will wait at
2301          * least until until the rcu_read_unlock.
2302          */
2303         rcu_read_lock();
2304         cur_devices = fs_info->fs_devices;
2305         while (cur_devices) {
2306                 head = &cur_devices->devices;
2307                 list_for_each_entry_rcu(dev, head, dev_list) {
2308                         if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2309                                 continue;
2310                         if (!dev->name)
2311                                 continue;
2312                         if (!first_dev || dev->devid < first_dev->devid)
2313                                 first_dev = dev;
2314                 }
2315                 cur_devices = cur_devices->seed;
2316         }
2317
2318         if (first_dev) {
2319                 name = rcu_dereference(first_dev->name);
2320                 seq_escape(m, name->str, " \t\n\\");
2321         } else {
2322                 WARN_ON(1);
2323         }
2324         rcu_read_unlock();
2325         return 0;
2326 }
2327
2328 static const struct super_operations btrfs_super_ops = {
2329         .drop_inode     = btrfs_drop_inode,
2330         .evict_inode    = btrfs_evict_inode,
2331         .put_super      = btrfs_put_super,
2332         .sync_fs        = btrfs_sync_fs,
2333         .show_options   = btrfs_show_options,
2334         .show_devname   = btrfs_show_devname,
2335         .write_inode    = btrfs_write_inode,
2336         .alloc_inode    = btrfs_alloc_inode,
2337         .destroy_inode  = btrfs_destroy_inode,
2338         .statfs         = btrfs_statfs,
2339         .remount_fs     = btrfs_remount,
2340         .freeze_fs      = btrfs_freeze,
2341         .unfreeze_fs    = btrfs_unfreeze,
2342 };
2343
2344 static const struct file_operations btrfs_ctl_fops = {
2345         .open = btrfs_control_open,
2346         .unlocked_ioctl  = btrfs_control_ioctl,
2347         .compat_ioctl = btrfs_control_ioctl,
2348         .owner   = THIS_MODULE,
2349         .llseek = noop_llseek,
2350 };
2351
2352 static struct miscdevice btrfs_misc = {
2353         .minor          = BTRFS_MINOR,
2354         .name           = "btrfs-control",
2355         .fops           = &btrfs_ctl_fops
2356 };
2357
2358 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2359 MODULE_ALIAS("devname:btrfs-control");
2360
2361 static int __init btrfs_interface_init(void)
2362 {
2363         return misc_register(&btrfs_misc);
2364 }
2365
2366 static __cold void btrfs_interface_exit(void)
2367 {
2368         misc_deregister(&btrfs_misc);
2369 }
2370
2371 static void __init btrfs_print_mod_info(void)
2372 {
2373         pr_info("Btrfs loaded, crc32c=%s"
2374 #ifdef CONFIG_BTRFS_DEBUG
2375                         ", debug=on"
2376 #endif
2377 #ifdef CONFIG_BTRFS_ASSERT
2378                         ", assert=on"
2379 #endif
2380 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2381                         ", integrity-checker=on"
2382 #endif
2383 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2384                         ", ref-verify=on"
2385 #endif
2386                         "\n",
2387                         crc32c_impl());
2388 }
2389
2390 static int __init init_btrfs_fs(void)
2391 {
2392         int err;
2393
2394         btrfs_props_init();
2395
2396         err = btrfs_init_sysfs();
2397         if (err)
2398                 return err;
2399
2400         btrfs_init_compress();
2401
2402         err = btrfs_init_cachep();
2403         if (err)
2404                 goto free_compress;
2405
2406         err = extent_io_init();
2407         if (err)
2408                 goto free_cachep;
2409
2410         err = extent_map_init();
2411         if (err)
2412                 goto free_extent_io;
2413
2414         err = ordered_data_init();
2415         if (err)
2416                 goto free_extent_map;
2417
2418         err = btrfs_delayed_inode_init();
2419         if (err)
2420                 goto free_ordered_data;
2421
2422         err = btrfs_auto_defrag_init();
2423         if (err)
2424                 goto free_delayed_inode;
2425
2426         err = btrfs_delayed_ref_init();
2427         if (err)
2428                 goto free_auto_defrag;
2429
2430         err = btrfs_prelim_ref_init();
2431         if (err)
2432                 goto free_delayed_ref;
2433
2434         err = btrfs_end_io_wq_init();
2435         if (err)
2436                 goto free_prelim_ref;
2437
2438         err = btrfs_interface_init();
2439         if (err)
2440                 goto free_end_io_wq;
2441
2442         btrfs_init_lockdep();
2443
2444         btrfs_print_mod_info();
2445
2446         err = btrfs_run_sanity_tests();
2447         if (err)
2448                 goto unregister_ioctl;
2449
2450         err = register_filesystem(&btrfs_fs_type);
2451         if (err)
2452                 goto unregister_ioctl;
2453
2454         return 0;
2455
2456 unregister_ioctl:
2457         btrfs_interface_exit();
2458 free_end_io_wq:
2459         btrfs_end_io_wq_exit();
2460 free_prelim_ref:
2461         btrfs_prelim_ref_exit();
2462 free_delayed_ref:
2463         btrfs_delayed_ref_exit();
2464 free_auto_defrag:
2465         btrfs_auto_defrag_exit();
2466 free_delayed_inode:
2467         btrfs_delayed_inode_exit();
2468 free_ordered_data:
2469         ordered_data_exit();
2470 free_extent_map:
2471         extent_map_exit();
2472 free_extent_io:
2473         extent_io_exit();
2474 free_cachep:
2475         btrfs_destroy_cachep();
2476 free_compress:
2477         btrfs_exit_compress();
2478         btrfs_exit_sysfs();
2479
2480         return err;
2481 }
2482
2483 static void __exit exit_btrfs_fs(void)
2484 {
2485         btrfs_destroy_cachep();
2486         btrfs_delayed_ref_exit();
2487         btrfs_auto_defrag_exit();
2488         btrfs_delayed_inode_exit();
2489         btrfs_prelim_ref_exit();
2490         ordered_data_exit();
2491         extent_map_exit();
2492         extent_io_exit();
2493         btrfs_interface_exit();
2494         btrfs_end_io_wq_exit();
2495         unregister_filesystem(&btrfs_fs_type);
2496         btrfs_exit_sysfs();
2497         btrfs_cleanup_fs_uuids();
2498         btrfs_exit_compress();
2499 }
2500
2501 late_initcall(init_btrfs_fs);
2502 module_exit(exit_btrfs_fs)
2503
2504 MODULE_LICENSE("GPL");