de87b7b0367d719e1adb4ce91937b509e28ee7f9
[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/fs.h>
9 #include <linux/pagemap.h>
10 #include <linux/highmem.h>
11 #include <linux/time.h>
12 #include <linux/init.h>
13 #include <linux/seq_file.h>
14 #include <linux/string.h>
15 #include <linux/backing-dev.h>
16 #include <linux/mount.h>
17 #include <linux/writeback.h>
18 #include <linux/statfs.h>
19 #include <linux/compat.h>
20 #include <linux/parser.h>
21 #include <linux/ctype.h>
22 #include <linux/namei.h>
23 #include <linux/miscdevice.h>
24 #include <linux/magic.h>
25 #include <linux/slab.h>
26 #include <linux/cleancache.h>
27 #include <linux/ratelimit.h>
28 #include <linux/crc32c.h>
29 #include <linux/btrfs.h>
30 #include "delayed-inode.h"
31 #include "ctree.h"
32 #include "disk-io.h"
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "print-tree.h"
36 #include "props.h"
37 #include "xattr.h"
38 #include "volumes.h"
39 #include "export.h"
40 #include "compression.h"
41 #include "rcu-string.h"
42 #include "dev-replace.h"
43 #include "free-space-cache.h"
44 #include "backref.h"
45 #include "tests/btrfs-tests.h"
46
47 #include "qgroup.h"
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/btrfs.h>
50
51 static const struct super_operations btrfs_super_ops;
52
53 /*
54  * Types for mounting the default subvolume and a subvolume explicitly
55  * requested by subvol=/path. That way the callchain is straightforward and we
56  * don't have to play tricks with the mount options and recursive calls to
57  * btrfs_mount.
58  *
59  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
60  */
61 static struct file_system_type btrfs_fs_type;
62 static struct file_system_type btrfs_root_fs_type;
63
64 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
65
66 const char *btrfs_decode_error(int errno)
67 {
68         char *errstr = "unknown";
69
70         switch (errno) {
71         case -EIO:
72                 errstr = "IO failure";
73                 break;
74         case -ENOMEM:
75                 errstr = "Out of memory";
76                 break;
77         case -EROFS:
78                 errstr = "Readonly filesystem";
79                 break;
80         case -EEXIST:
81                 errstr = "Object already exists";
82                 break;
83         case -ENOSPC:
84                 errstr = "No space left";
85                 break;
86         case -ENOENT:
87                 errstr = "No such entry";
88                 break;
89         }
90
91         return errstr;
92 }
93
94 /*
95  * __btrfs_handle_fs_error decodes expected errors from the caller and
96  * invokes the approciate error response.
97  */
98 __cold
99 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
100                        unsigned int line, int errno, const char *fmt, ...)
101 {
102         struct super_block *sb = fs_info->sb;
103 #ifdef CONFIG_PRINTK
104         const char *errstr;
105 #endif
106
107         /*
108          * Special case: if the error is EROFS, and we're already
109          * under SB_RDONLY, then it is safe here.
110          */
111         if (errno == -EROFS && sb_rdonly(sb))
112                 return;
113
114 #ifdef CONFIG_PRINTK
115         errstr = btrfs_decode_error(errno);
116         if (fmt) {
117                 struct va_format vaf;
118                 va_list args;
119
120                 va_start(args, fmt);
121                 vaf.fmt = fmt;
122                 vaf.va = &args;
123
124                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
125                         sb->s_id, function, line, errno, errstr, &vaf);
126                 va_end(args);
127         } else {
128                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
129                         sb->s_id, function, line, errno, errstr);
130         }
131 #endif
132
133         /*
134          * Today we only save the error info to memory.  Long term we'll
135          * also send it down to the disk
136          */
137         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
138
139         /* Don't go through full error handling during mount */
140         if (!(sb->s_flags & SB_BORN))
141                 return;
142
143         if (sb_rdonly(sb))
144                 return;
145
146         /* btrfs handle error by forcing the filesystem readonly */
147         sb->s_flags |= SB_RDONLY;
148         btrfs_info(fs_info, "forced readonly");
149         /*
150          * Note that a running device replace operation is not canceled here
151          * although there is no way to update the progress. It would add the
152          * risk of a deadlock, therefore the canceling is omitted. The only
153          * penalty is that some I/O remains active until the procedure
154          * completes. The next time when the filesystem is mounted writeable
155          * again, the device replace operation continues.
156          */
157 }
158
159 #ifdef CONFIG_PRINTK
160 static const char * const logtypes[] = {
161         "emergency",
162         "alert",
163         "critical",
164         "error",
165         "warning",
166         "notice",
167         "info",
168         "debug",
169 };
170
171
172 /*
173  * Use one ratelimit state per log level so that a flood of less important
174  * messages doesn't cause more important ones to be dropped.
175  */
176 static struct ratelimit_state printk_limits[] = {
177         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
178         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
179         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
180         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
181         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
182         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
183         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
184         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
185 };
186
187 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
188 {
189         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
190         struct va_format vaf;
191         va_list args;
192         int kern_level;
193         const char *type = logtypes[4];
194         struct ratelimit_state *ratelimit = &printk_limits[4];
195
196         va_start(args, fmt);
197
198         while ((kern_level = printk_get_level(fmt)) != 0) {
199                 size_t size = printk_skip_level(fmt) - fmt;
200
201                 if (kern_level >= '0' && kern_level <= '7') {
202                         memcpy(lvl, fmt,  size);
203                         lvl[size] = '\0';
204                         type = logtypes[kern_level - '0'];
205                         ratelimit = &printk_limits[kern_level - '0'];
206                 }
207                 fmt += size;
208         }
209
210         vaf.fmt = fmt;
211         vaf.va = &args;
212
213         if (__ratelimit(ratelimit))
214                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
215                         fs_info ? fs_info->sb->s_id : "<unknown>", &vaf);
216
217         va_end(args);
218 }
219 #endif
220
221 /*
222  * We only mark the transaction aborted and then set the file system read-only.
223  * This will prevent new transactions from starting or trying to join this
224  * one.
225  *
226  * This means that error recovery at the call site is limited to freeing
227  * any local memory allocations and passing the error code up without
228  * further cleanup. The transaction should complete as it normally would
229  * in the call path but will return -EIO.
230  *
231  * We'll complete the cleanup in btrfs_end_transaction and
232  * btrfs_commit_transaction.
233  */
234 __cold
235 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
236                                const char *function,
237                                unsigned int line, int errno)
238 {
239         struct btrfs_fs_info *fs_info = trans->fs_info;
240
241         trans->aborted = errno;
242         /* Nothing used. The other threads that have joined this
243          * transaction may be able to continue. */
244         if (!trans->dirty && list_empty(&trans->new_bgs)) {
245                 const char *errstr;
246
247                 errstr = btrfs_decode_error(errno);
248                 btrfs_warn(fs_info,
249                            "%s:%d: Aborting unused transaction(%s).",
250                            function, line, errstr);
251                 return;
252         }
253         WRITE_ONCE(trans->transaction->aborted, errno);
254         /* Wake up anybody who may be waiting on this transaction */
255         wake_up(&fs_info->transaction_wait);
256         wake_up(&fs_info->transaction_blocked_wait);
257         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
258 }
259 /*
260  * __btrfs_panic decodes unexpected, fatal errors from the caller,
261  * issues an alert, and either panics or BUGs, depending on mount options.
262  */
263 __cold
264 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
265                    unsigned int line, int errno, const char *fmt, ...)
266 {
267         char *s_id = "<unknown>";
268         const char *errstr;
269         struct va_format vaf = { .fmt = fmt };
270         va_list args;
271
272         if (fs_info)
273                 s_id = fs_info->sb->s_id;
274
275         va_start(args, fmt);
276         vaf.va = &args;
277
278         errstr = btrfs_decode_error(errno);
279         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
280                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
281                         s_id, function, line, &vaf, errno, errstr);
282
283         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
284                    function, line, &vaf, errno, errstr);
285         va_end(args);
286         /* Caller calls BUG() */
287 }
288
289 static void btrfs_put_super(struct super_block *sb)
290 {
291         close_ctree(btrfs_sb(sb));
292 }
293
294 enum {
295         Opt_acl, Opt_noacl,
296         Opt_clear_cache,
297         Opt_commit_interval,
298         Opt_compress,
299         Opt_compress_force,
300         Opt_compress_force_type,
301         Opt_compress_type,
302         Opt_degraded,
303         Opt_device,
304         Opt_fatal_errors,
305         Opt_flushoncommit, Opt_noflushoncommit,
306         Opt_inode_cache, Opt_noinode_cache,
307         Opt_max_inline,
308         Opt_barrier, Opt_nobarrier,
309         Opt_datacow, Opt_nodatacow,
310         Opt_datasum, Opt_nodatasum,
311         Opt_defrag, Opt_nodefrag,
312         Opt_discard, Opt_nodiscard,
313         Opt_nologreplay,
314         Opt_norecovery,
315         Opt_ratio,
316         Opt_rescan_uuid_tree,
317         Opt_skip_balance,
318         Opt_space_cache, Opt_no_space_cache,
319         Opt_space_cache_version,
320         Opt_ssd, Opt_nossd,
321         Opt_ssd_spread, Opt_nossd_spread,
322         Opt_subvol,
323         Opt_subvol_empty,
324         Opt_subvolid,
325         Opt_thread_pool,
326         Opt_treelog, Opt_notreelog,
327         Opt_usebackuproot,
328         Opt_user_subvol_rm_allowed,
329
330         /* Deprecated options */
331         Opt_alloc_start,
332         Opt_recovery,
333         Opt_subvolrootid,
334
335         /* Debugging options */
336         Opt_check_integrity,
337         Opt_check_integrity_including_extent_data,
338         Opt_check_integrity_print_mask,
339         Opt_enospc_debug, Opt_noenospc_debug,
340 #ifdef CONFIG_BTRFS_DEBUG
341         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
342 #endif
343 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
344         Opt_ref_verify,
345 #endif
346         Opt_err,
347 };
348
349 static const match_table_t tokens = {
350         {Opt_acl, "acl"},
351         {Opt_noacl, "noacl"},
352         {Opt_clear_cache, "clear_cache"},
353         {Opt_commit_interval, "commit=%u"},
354         {Opt_compress, "compress"},
355         {Opt_compress_type, "compress=%s"},
356         {Opt_compress_force, "compress-force"},
357         {Opt_compress_force_type, "compress-force=%s"},
358         {Opt_degraded, "degraded"},
359         {Opt_device, "device=%s"},
360         {Opt_fatal_errors, "fatal_errors=%s"},
361         {Opt_flushoncommit, "flushoncommit"},
362         {Opt_noflushoncommit, "noflushoncommit"},
363         {Opt_inode_cache, "inode_cache"},
364         {Opt_noinode_cache, "noinode_cache"},
365         {Opt_max_inline, "max_inline=%s"},
366         {Opt_barrier, "barrier"},
367         {Opt_nobarrier, "nobarrier"},
368         {Opt_datacow, "datacow"},
369         {Opt_nodatacow, "nodatacow"},
370         {Opt_datasum, "datasum"},
371         {Opt_nodatasum, "nodatasum"},
372         {Opt_defrag, "autodefrag"},
373         {Opt_nodefrag, "noautodefrag"},
374         {Opt_discard, "discard"},
375         {Opt_nodiscard, "nodiscard"},
376         {Opt_nologreplay, "nologreplay"},
377         {Opt_norecovery, "norecovery"},
378         {Opt_ratio, "metadata_ratio=%u"},
379         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
380         {Opt_skip_balance, "skip_balance"},
381         {Opt_space_cache, "space_cache"},
382         {Opt_no_space_cache, "nospace_cache"},
383         {Opt_space_cache_version, "space_cache=%s"},
384         {Opt_ssd, "ssd"},
385         {Opt_nossd, "nossd"},
386         {Opt_ssd_spread, "ssd_spread"},
387         {Opt_nossd_spread, "nossd_spread"},
388         {Opt_subvol, "subvol=%s"},
389         {Opt_subvol_empty, "subvol="},
390         {Opt_subvolid, "subvolid=%s"},
391         {Opt_thread_pool, "thread_pool=%u"},
392         {Opt_treelog, "treelog"},
393         {Opt_notreelog, "notreelog"},
394         {Opt_usebackuproot, "usebackuproot"},
395         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
396
397         /* Deprecated options */
398         {Opt_alloc_start, "alloc_start=%s"},
399         {Opt_recovery, "recovery"},
400         {Opt_subvolrootid, "subvolrootid=%d"},
401
402         /* Debugging options */
403         {Opt_check_integrity, "check_int"},
404         {Opt_check_integrity_including_extent_data, "check_int_data"},
405         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
406         {Opt_enospc_debug, "enospc_debug"},
407         {Opt_noenospc_debug, "noenospc_debug"},
408 #ifdef CONFIG_BTRFS_DEBUG
409         {Opt_fragment_data, "fragment=data"},
410         {Opt_fragment_metadata, "fragment=metadata"},
411         {Opt_fragment_all, "fragment=all"},
412 #endif
413 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
414         {Opt_ref_verify, "ref_verify"},
415 #endif
416         {Opt_err, NULL},
417 };
418
419 /*
420  * Regular mount options parser.  Everything that is needed only when
421  * reading in a new superblock is parsed here.
422  * XXX JDM: This needs to be cleaned up for remount.
423  */
424 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
425                         unsigned long new_flags)
426 {
427         substring_t args[MAX_OPT_ARGS];
428         char *p, *num;
429         u64 cache_gen;
430         int intarg;
431         int ret = 0;
432         char *compress_type;
433         bool compress_force = false;
434         enum btrfs_compression_type saved_compress_type;
435         bool saved_compress_force;
436         int no_compress = 0;
437
438         cache_gen = btrfs_super_cache_generation(info->super_copy);
439         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
440                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
441         else if (cache_gen)
442                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
443
444         /*
445          * Even the options are empty, we still need to do extra check
446          * against new flags
447          */
448         if (!options)
449                 goto check;
450
451         while ((p = strsep(&options, ",")) != NULL) {
452                 int token;
453                 if (!*p)
454                         continue;
455
456                 token = match_token(p, tokens, args);
457                 switch (token) {
458                 case Opt_degraded:
459                         btrfs_info(info, "allowing degraded mounts");
460                         btrfs_set_opt(info->mount_opt, DEGRADED);
461                         break;
462                 case Opt_subvol:
463                 case Opt_subvol_empty:
464                 case Opt_subvolid:
465                 case Opt_subvolrootid:
466                 case Opt_device:
467                         /*
468                          * These are parsed by btrfs_parse_subvol_options
469                          * and btrfs_parse_early_options
470                          * and can be happily ignored here.
471                          */
472                         break;
473                 case Opt_nodatasum:
474                         btrfs_set_and_info(info, NODATASUM,
475                                            "setting nodatasum");
476                         break;
477                 case Opt_datasum:
478                         if (btrfs_test_opt(info, NODATASUM)) {
479                                 if (btrfs_test_opt(info, NODATACOW))
480                                         btrfs_info(info,
481                                                    "setting datasum, datacow enabled");
482                                 else
483                                         btrfs_info(info, "setting datasum");
484                         }
485                         btrfs_clear_opt(info->mount_opt, NODATACOW);
486                         btrfs_clear_opt(info->mount_opt, NODATASUM);
487                         break;
488                 case Opt_nodatacow:
489                         if (!btrfs_test_opt(info, NODATACOW)) {
490                                 if (!btrfs_test_opt(info, COMPRESS) ||
491                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
492                                         btrfs_info(info,
493                                                    "setting nodatacow, compression disabled");
494                                 } else {
495                                         btrfs_info(info, "setting nodatacow");
496                                 }
497                         }
498                         btrfs_clear_opt(info->mount_opt, COMPRESS);
499                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
500                         btrfs_set_opt(info->mount_opt, NODATACOW);
501                         btrfs_set_opt(info->mount_opt, NODATASUM);
502                         break;
503                 case Opt_datacow:
504                         btrfs_clear_and_info(info, NODATACOW,
505                                              "setting datacow");
506                         break;
507                 case Opt_compress_force:
508                 case Opt_compress_force_type:
509                         compress_force = true;
510                         /* Fallthrough */
511                 case Opt_compress:
512                 case Opt_compress_type:
513                         saved_compress_type = btrfs_test_opt(info,
514                                                              COMPRESS) ?
515                                 info->compress_type : BTRFS_COMPRESS_NONE;
516                         saved_compress_force =
517                                 btrfs_test_opt(info, FORCE_COMPRESS);
518                         if (token == Opt_compress ||
519                             token == Opt_compress_force ||
520                             strncmp(args[0].from, "zlib", 4) == 0) {
521                                 compress_type = "zlib";
522
523                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
524                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
525                                 /*
526                                  * args[0] contains uninitialized data since
527                                  * for these tokens we don't expect any
528                                  * parameter.
529                                  */
530                                 if (token != Opt_compress &&
531                                     token != Opt_compress_force)
532                                         info->compress_level =
533                                           btrfs_compress_str2level(args[0].from);
534                                 btrfs_set_opt(info->mount_opt, COMPRESS);
535                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
536                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
537                                 no_compress = 0;
538                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
539                                 compress_type = "lzo";
540                                 info->compress_type = BTRFS_COMPRESS_LZO;
541                                 btrfs_set_opt(info->mount_opt, COMPRESS);
542                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
543                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
544                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
545                                 no_compress = 0;
546                         } else if (strcmp(args[0].from, "zstd") == 0) {
547                                 compress_type = "zstd";
548                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
549                                 btrfs_set_opt(info->mount_opt, COMPRESS);
550                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
551                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
552                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
553                                 no_compress = 0;
554                         } else if (strncmp(args[0].from, "no", 2) == 0) {
555                                 compress_type = "no";
556                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
557                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
558                                 compress_force = false;
559                                 no_compress++;
560                         } else {
561                                 ret = -EINVAL;
562                                 goto out;
563                         }
564
565                         if (compress_force) {
566                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
567                         } else {
568                                 /*
569                                  * If we remount from compress-force=xxx to
570                                  * compress=xxx, we need clear FORCE_COMPRESS
571                                  * flag, otherwise, there is no way for users
572                                  * to disable forcible compression separately.
573                                  */
574                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
575                         }
576                         if ((btrfs_test_opt(info, COMPRESS) &&
577                              (info->compress_type != saved_compress_type ||
578                               compress_force != saved_compress_force)) ||
579                             (!btrfs_test_opt(info, COMPRESS) &&
580                              no_compress == 1)) {
581                                 btrfs_info(info, "%s %s compression, level %d",
582                                            (compress_force) ? "force" : "use",
583                                            compress_type, info->compress_level);
584                         }
585                         compress_force = false;
586                         break;
587                 case Opt_ssd:
588                         btrfs_set_and_info(info, SSD,
589                                            "enabling ssd optimizations");
590                         btrfs_clear_opt(info->mount_opt, NOSSD);
591                         break;
592                 case Opt_ssd_spread:
593                         btrfs_set_and_info(info, SSD,
594                                            "enabling ssd optimizations");
595                         btrfs_set_and_info(info, SSD_SPREAD,
596                                            "using spread ssd allocation scheme");
597                         btrfs_clear_opt(info->mount_opt, NOSSD);
598                         break;
599                 case Opt_nossd:
600                         btrfs_set_opt(info->mount_opt, NOSSD);
601                         btrfs_clear_and_info(info, SSD,
602                                              "not using ssd optimizations");
603                         /* Fallthrough */
604                 case Opt_nossd_spread:
605                         btrfs_clear_and_info(info, SSD_SPREAD,
606                                              "not using spread ssd allocation scheme");
607                         break;
608                 case Opt_barrier:
609                         btrfs_clear_and_info(info, NOBARRIER,
610                                              "turning on barriers");
611                         break;
612                 case Opt_nobarrier:
613                         btrfs_set_and_info(info, NOBARRIER,
614                                            "turning off barriers");
615                         break;
616                 case Opt_thread_pool:
617                         ret = match_int(&args[0], &intarg);
618                         if (ret) {
619                                 goto out;
620                         } else if (intarg == 0) {
621                                 ret = -EINVAL;
622                                 goto out;
623                         }
624                         info->thread_pool_size = intarg;
625                         break;
626                 case Opt_max_inline:
627                         num = match_strdup(&args[0]);
628                         if (num) {
629                                 info->max_inline = memparse(num, NULL);
630                                 kfree(num);
631
632                                 if (info->max_inline) {
633                                         info->max_inline = min_t(u64,
634                                                 info->max_inline,
635                                                 info->sectorsize);
636                                 }
637                                 btrfs_info(info, "max_inline at %llu",
638                                            info->max_inline);
639                         } else {
640                                 ret = -ENOMEM;
641                                 goto out;
642                         }
643                         break;
644                 case Opt_alloc_start:
645                         btrfs_info(info,
646                                 "option alloc_start is obsolete, ignored");
647                         break;
648                 case Opt_acl:
649 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
650                         info->sb->s_flags |= SB_POSIXACL;
651                         break;
652 #else
653                         btrfs_err(info, "support for ACL not compiled in!");
654                         ret = -EINVAL;
655                         goto out;
656 #endif
657                 case Opt_noacl:
658                         info->sb->s_flags &= ~SB_POSIXACL;
659                         break;
660                 case Opt_notreelog:
661                         btrfs_set_and_info(info, NOTREELOG,
662                                            "disabling tree log");
663                         break;
664                 case Opt_treelog:
665                         btrfs_clear_and_info(info, NOTREELOG,
666                                              "enabling tree log");
667                         break;
668                 case Opt_norecovery:
669                 case Opt_nologreplay:
670                         btrfs_set_and_info(info, NOLOGREPLAY,
671                                            "disabling log replay at mount time");
672                         break;
673                 case Opt_flushoncommit:
674                         btrfs_set_and_info(info, FLUSHONCOMMIT,
675                                            "turning on flush-on-commit");
676                         break;
677                 case Opt_noflushoncommit:
678                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
679                                              "turning off flush-on-commit");
680                         break;
681                 case Opt_ratio:
682                         ret = match_int(&args[0], &intarg);
683                         if (ret)
684                                 goto out;
685                         info->metadata_ratio = intarg;
686                         btrfs_info(info, "metadata ratio %u",
687                                    info->metadata_ratio);
688                         break;
689                 case Opt_discard:
690                         btrfs_set_and_info(info, DISCARD,
691                                            "turning on discard");
692                         break;
693                 case Opt_nodiscard:
694                         btrfs_clear_and_info(info, DISCARD,
695                                              "turning off discard");
696                         break;
697                 case Opt_space_cache:
698                 case Opt_space_cache_version:
699                         if (token == Opt_space_cache ||
700                             strcmp(args[0].from, "v1") == 0) {
701                                 btrfs_clear_opt(info->mount_opt,
702                                                 FREE_SPACE_TREE);
703                                 btrfs_set_and_info(info, SPACE_CACHE,
704                                            "enabling disk space caching");
705                         } else if (strcmp(args[0].from, "v2") == 0) {
706                                 btrfs_clear_opt(info->mount_opt,
707                                                 SPACE_CACHE);
708                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
709                                                    "enabling free space tree");
710                         } else {
711                                 ret = -EINVAL;
712                                 goto out;
713                         }
714                         break;
715                 case Opt_rescan_uuid_tree:
716                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
717                         break;
718                 case Opt_no_space_cache:
719                         if (btrfs_test_opt(info, SPACE_CACHE)) {
720                                 btrfs_clear_and_info(info, SPACE_CACHE,
721                                              "disabling disk space caching");
722                         }
723                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
724                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
725                                              "disabling free space tree");
726                         }
727                         break;
728                 case Opt_inode_cache:
729                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
730                                            "enabling inode map caching");
731                         break;
732                 case Opt_noinode_cache:
733                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
734                                              "disabling inode map caching");
735                         break;
736                 case Opt_clear_cache:
737                         btrfs_set_and_info(info, CLEAR_CACHE,
738                                            "force clearing of disk cache");
739                         break;
740                 case Opt_user_subvol_rm_allowed:
741                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
742                         break;
743                 case Opt_enospc_debug:
744                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
745                         break;
746                 case Opt_noenospc_debug:
747                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
748                         break;
749                 case Opt_defrag:
750                         btrfs_set_and_info(info, AUTO_DEFRAG,
751                                            "enabling auto defrag");
752                         break;
753                 case Opt_nodefrag:
754                         btrfs_clear_and_info(info, AUTO_DEFRAG,
755                                              "disabling auto defrag");
756                         break;
757                 case Opt_recovery:
758                         btrfs_warn(info,
759                                    "'recovery' is deprecated, use 'usebackuproot' instead");
760                         /* fall through */
761                 case Opt_usebackuproot:
762                         btrfs_info(info,
763                                    "trying to use backup root at mount time");
764                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
765                         break;
766                 case Opt_skip_balance:
767                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
768                         break;
769 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
770                 case Opt_check_integrity_including_extent_data:
771                         btrfs_info(info,
772                                    "enabling check integrity including extent data");
773                         btrfs_set_opt(info->mount_opt,
774                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
775                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
776                         break;
777                 case Opt_check_integrity:
778                         btrfs_info(info, "enabling check integrity");
779                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
780                         break;
781                 case Opt_check_integrity_print_mask:
782                         ret = match_int(&args[0], &intarg);
783                         if (ret)
784                                 goto out;
785                         info->check_integrity_print_mask = intarg;
786                         btrfs_info(info, "check_integrity_print_mask 0x%x",
787                                    info->check_integrity_print_mask);
788                         break;
789 #else
790                 case Opt_check_integrity_including_extent_data:
791                 case Opt_check_integrity:
792                 case Opt_check_integrity_print_mask:
793                         btrfs_err(info,
794                                   "support for check_integrity* not compiled in!");
795                         ret = -EINVAL;
796                         goto out;
797 #endif
798                 case Opt_fatal_errors:
799                         if (strcmp(args[0].from, "panic") == 0)
800                                 btrfs_set_opt(info->mount_opt,
801                                               PANIC_ON_FATAL_ERROR);
802                         else if (strcmp(args[0].from, "bug") == 0)
803                                 btrfs_clear_opt(info->mount_opt,
804                                               PANIC_ON_FATAL_ERROR);
805                         else {
806                                 ret = -EINVAL;
807                                 goto out;
808                         }
809                         break;
810                 case Opt_commit_interval:
811                         intarg = 0;
812                         ret = match_int(&args[0], &intarg);
813                         if (ret)
814                                 goto out;
815                         if (intarg == 0) {
816                                 btrfs_info(info,
817                                            "using default commit interval %us",
818                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
819                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
820                         } else if (intarg > 300) {
821                                 btrfs_warn(info, "excessive commit interval %d",
822                                            intarg);
823                         }
824                         info->commit_interval = intarg;
825                         break;
826 #ifdef CONFIG_BTRFS_DEBUG
827                 case Opt_fragment_all:
828                         btrfs_info(info, "fragmenting all space");
829                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
830                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
831                         break;
832                 case Opt_fragment_metadata:
833                         btrfs_info(info, "fragmenting metadata");
834                         btrfs_set_opt(info->mount_opt,
835                                       FRAGMENT_METADATA);
836                         break;
837                 case Opt_fragment_data:
838                         btrfs_info(info, "fragmenting data");
839                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
840                         break;
841 #endif
842 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
843                 case Opt_ref_verify:
844                         btrfs_info(info, "doing ref verification");
845                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
846                         break;
847 #endif
848                 case Opt_err:
849                         btrfs_info(info, "unrecognized mount option '%s'", p);
850                         ret = -EINVAL;
851                         goto out;
852                 default:
853                         break;
854                 }
855         }
856 check:
857         /*
858          * Extra check for current option against current flag
859          */
860         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & SB_RDONLY)) {
861                 btrfs_err(info,
862                           "nologreplay must be used with ro mount option");
863                 ret = -EINVAL;
864         }
865 out:
866         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
867             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
868             !btrfs_test_opt(info, CLEAR_CACHE)) {
869                 btrfs_err(info, "cannot disable free space tree");
870                 ret = -EINVAL;
871
872         }
873         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
874                 btrfs_info(info, "disk space caching is enabled");
875         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
876                 btrfs_info(info, "using free space tree");
877         return ret;
878 }
879
880 /*
881  * Parse mount options that are required early in the mount process.
882  *
883  * All other options will be parsed on much later in the mount process and
884  * only when we need to allocate a new super block.
885  */
886 static int btrfs_parse_early_options(const char *options, fmode_t flags,
887                 void *holder, struct btrfs_fs_devices **fs_devices)
888 {
889         substring_t args[MAX_OPT_ARGS];
890         char *device_name, *opts, *orig, *p;
891         int error = 0;
892
893         lockdep_assert_held(&uuid_mutex);
894
895         if (!options)
896                 return 0;
897
898         /*
899          * strsep changes the string, duplicate it because btrfs_parse_options
900          * gets called later
901          */
902         opts = kstrdup(options, GFP_KERNEL);
903         if (!opts)
904                 return -ENOMEM;
905         orig = opts;
906
907         while ((p = strsep(&opts, ",")) != NULL) {
908                 int token;
909
910                 if (!*p)
911                         continue;
912
913                 token = match_token(p, tokens, args);
914                 if (token == Opt_device) {
915                         device_name = match_strdup(&args[0]);
916                         if (!device_name) {
917                                 error = -ENOMEM;
918                                 goto out;
919                         }
920                         error = btrfs_scan_one_device(device_name,
921                                         flags, holder, fs_devices);
922                         kfree(device_name);
923                         if (error)
924                                 goto out;
925                 }
926         }
927
928 out:
929         kfree(orig);
930         return error;
931 }
932
933 /*
934  * Parse mount options that are related to subvolume id
935  *
936  * The value is later passed to mount_subvol()
937  */
938 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
939                 u64 *subvol_objectid)
940 {
941         substring_t args[MAX_OPT_ARGS];
942         char *opts, *orig, *p;
943         int error = 0;
944         u64 subvolid;
945
946         if (!options)
947                 return 0;
948
949         /*
950          * strsep changes the string, duplicate it because
951          * btrfs_parse_early_options gets called later
952          */
953         opts = kstrdup(options, GFP_KERNEL);
954         if (!opts)
955                 return -ENOMEM;
956         orig = opts;
957
958         while ((p = strsep(&opts, ",")) != NULL) {
959                 int token;
960                 if (!*p)
961                         continue;
962
963                 token = match_token(p, tokens, args);
964                 switch (token) {
965                 case Opt_subvol:
966                         kfree(*subvol_name);
967                         *subvol_name = match_strdup(&args[0]);
968                         if (!*subvol_name) {
969                                 error = -ENOMEM;
970                                 goto out;
971                         }
972                         break;
973                 case Opt_subvolid:
974                         error = match_u64(&args[0], &subvolid);
975                         if (error)
976                                 goto out;
977
978                         /* we want the original fs_tree */
979                         if (subvolid == 0)
980                                 subvolid = BTRFS_FS_TREE_OBJECTID;
981
982                         *subvol_objectid = subvolid;
983                         break;
984                 case Opt_subvolrootid:
985                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
986                         break;
987                 default:
988                         break;
989                 }
990         }
991
992 out:
993         kfree(orig);
994         return error;
995 }
996
997 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
998                                            u64 subvol_objectid)
999 {
1000         struct btrfs_root *root = fs_info->tree_root;
1001         struct btrfs_root *fs_root;
1002         struct btrfs_root_ref *root_ref;
1003         struct btrfs_inode_ref *inode_ref;
1004         struct btrfs_key key;
1005         struct btrfs_path *path = NULL;
1006         char *name = NULL, *ptr;
1007         u64 dirid;
1008         int len;
1009         int ret;
1010
1011         path = btrfs_alloc_path();
1012         if (!path) {
1013                 ret = -ENOMEM;
1014                 goto err;
1015         }
1016         path->leave_spinning = 1;
1017
1018         name = kmalloc(PATH_MAX, GFP_KERNEL);
1019         if (!name) {
1020                 ret = -ENOMEM;
1021                 goto err;
1022         }
1023         ptr = name + PATH_MAX - 1;
1024         ptr[0] = '\0';
1025
1026         /*
1027          * Walk up the subvolume trees in the tree of tree roots by root
1028          * backrefs until we hit the top-level subvolume.
1029          */
1030         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1031                 key.objectid = subvol_objectid;
1032                 key.type = BTRFS_ROOT_BACKREF_KEY;
1033                 key.offset = (u64)-1;
1034
1035                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1036                 if (ret < 0) {
1037                         goto err;
1038                 } else if (ret > 0) {
1039                         ret = btrfs_previous_item(root, path, subvol_objectid,
1040                                                   BTRFS_ROOT_BACKREF_KEY);
1041                         if (ret < 0) {
1042                                 goto err;
1043                         } else if (ret > 0) {
1044                                 ret = -ENOENT;
1045                                 goto err;
1046                         }
1047                 }
1048
1049                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1050                 subvol_objectid = key.offset;
1051
1052                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1053                                           struct btrfs_root_ref);
1054                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1055                 ptr -= len + 1;
1056                 if (ptr < name) {
1057                         ret = -ENAMETOOLONG;
1058                         goto err;
1059                 }
1060                 read_extent_buffer(path->nodes[0], ptr + 1,
1061                                    (unsigned long)(root_ref + 1), len);
1062                 ptr[0] = '/';
1063                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1064                 btrfs_release_path(path);
1065
1066                 key.objectid = subvol_objectid;
1067                 key.type = BTRFS_ROOT_ITEM_KEY;
1068                 key.offset = (u64)-1;
1069                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1070                 if (IS_ERR(fs_root)) {
1071                         ret = PTR_ERR(fs_root);
1072                         goto err;
1073                 }
1074
1075                 /*
1076                  * Walk up the filesystem tree by inode refs until we hit the
1077                  * root directory.
1078                  */
1079                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1080                         key.objectid = dirid;
1081                         key.type = BTRFS_INODE_REF_KEY;
1082                         key.offset = (u64)-1;
1083
1084                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1085                         if (ret < 0) {
1086                                 goto err;
1087                         } else if (ret > 0) {
1088                                 ret = btrfs_previous_item(fs_root, path, dirid,
1089                                                           BTRFS_INODE_REF_KEY);
1090                                 if (ret < 0) {
1091                                         goto err;
1092                                 } else if (ret > 0) {
1093                                         ret = -ENOENT;
1094                                         goto err;
1095                                 }
1096                         }
1097
1098                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1099                         dirid = key.offset;
1100
1101                         inode_ref = btrfs_item_ptr(path->nodes[0],
1102                                                    path->slots[0],
1103                                                    struct btrfs_inode_ref);
1104                         len = btrfs_inode_ref_name_len(path->nodes[0],
1105                                                        inode_ref);
1106                         ptr -= len + 1;
1107                         if (ptr < name) {
1108                                 ret = -ENAMETOOLONG;
1109                                 goto err;
1110                         }
1111                         read_extent_buffer(path->nodes[0], ptr + 1,
1112                                            (unsigned long)(inode_ref + 1), len);
1113                         ptr[0] = '/';
1114                         btrfs_release_path(path);
1115                 }
1116         }
1117
1118         btrfs_free_path(path);
1119         if (ptr == name + PATH_MAX - 1) {
1120                 name[0] = '/';
1121                 name[1] = '\0';
1122         } else {
1123                 memmove(name, ptr, name + PATH_MAX - ptr);
1124         }
1125         return name;
1126
1127 err:
1128         btrfs_free_path(path);
1129         kfree(name);
1130         return ERR_PTR(ret);
1131 }
1132
1133 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1134 {
1135         struct btrfs_root *root = fs_info->tree_root;
1136         struct btrfs_dir_item *di;
1137         struct btrfs_path *path;
1138         struct btrfs_key location;
1139         u64 dir_id;
1140
1141         path = btrfs_alloc_path();
1142         if (!path)
1143                 return -ENOMEM;
1144         path->leave_spinning = 1;
1145
1146         /*
1147          * Find the "default" dir item which points to the root item that we
1148          * will mount by default if we haven't been given a specific subvolume
1149          * to mount.
1150          */
1151         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1152         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1153         if (IS_ERR(di)) {
1154                 btrfs_free_path(path);
1155                 return PTR_ERR(di);
1156         }
1157         if (!di) {
1158                 /*
1159                  * Ok the default dir item isn't there.  This is weird since
1160                  * it's always been there, but don't freak out, just try and
1161                  * mount the top-level subvolume.
1162                  */
1163                 btrfs_free_path(path);
1164                 *objectid = BTRFS_FS_TREE_OBJECTID;
1165                 return 0;
1166         }
1167
1168         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1169         btrfs_free_path(path);
1170         *objectid = location.objectid;
1171         return 0;
1172 }
1173
1174 static int btrfs_fill_super(struct super_block *sb,
1175                             struct btrfs_fs_devices *fs_devices,
1176                             void *data)
1177 {
1178         struct inode *inode;
1179         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1180         struct btrfs_key key;
1181         int err;
1182
1183         sb->s_maxbytes = MAX_LFS_FILESIZE;
1184         sb->s_magic = BTRFS_SUPER_MAGIC;
1185         sb->s_op = &btrfs_super_ops;
1186         sb->s_d_op = &btrfs_dentry_operations;
1187         sb->s_export_op = &btrfs_export_ops;
1188         sb->s_xattr = btrfs_xattr_handlers;
1189         sb->s_time_gran = 1;
1190 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1191         sb->s_flags |= SB_POSIXACL;
1192 #endif
1193         sb->s_flags |= SB_I_VERSION;
1194         sb->s_iflags |= SB_I_CGROUPWB;
1195
1196         err = super_setup_bdi(sb);
1197         if (err) {
1198                 btrfs_err(fs_info, "super_setup_bdi failed");
1199                 return err;
1200         }
1201
1202         err = open_ctree(sb, fs_devices, (char *)data);
1203         if (err) {
1204                 btrfs_err(fs_info, "open_ctree failed");
1205                 return err;
1206         }
1207
1208         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1209         key.type = BTRFS_INODE_ITEM_KEY;
1210         key.offset = 0;
1211         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1212         if (IS_ERR(inode)) {
1213                 err = PTR_ERR(inode);
1214                 goto fail_close;
1215         }
1216
1217         sb->s_root = d_make_root(inode);
1218         if (!sb->s_root) {
1219                 err = -ENOMEM;
1220                 goto fail_close;
1221         }
1222
1223         cleancache_init_fs(sb);
1224         sb->s_flags |= SB_ACTIVE;
1225         return 0;
1226
1227 fail_close:
1228         close_ctree(fs_info);
1229         return err;
1230 }
1231
1232 int btrfs_sync_fs(struct super_block *sb, int wait)
1233 {
1234         struct btrfs_trans_handle *trans;
1235         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1236         struct btrfs_root *root = fs_info->tree_root;
1237
1238         trace_btrfs_sync_fs(fs_info, wait);
1239
1240         if (!wait) {
1241                 filemap_flush(fs_info->btree_inode->i_mapping);
1242                 return 0;
1243         }
1244
1245         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1246
1247         trans = btrfs_attach_transaction_barrier(root);
1248         if (IS_ERR(trans)) {
1249                 /* no transaction, don't bother */
1250                 if (PTR_ERR(trans) == -ENOENT) {
1251                         /*
1252                          * Exit unless we have some pending changes
1253                          * that need to go through commit
1254                          */
1255                         if (fs_info->pending_changes == 0)
1256                                 return 0;
1257                         /*
1258                          * A non-blocking test if the fs is frozen. We must not
1259                          * start a new transaction here otherwise a deadlock
1260                          * happens. The pending operations are delayed to the
1261                          * next commit after thawing.
1262                          */
1263                         if (sb_start_write_trylock(sb))
1264                                 sb_end_write(sb);
1265                         else
1266                                 return 0;
1267                         trans = btrfs_start_transaction(root, 0);
1268                 }
1269                 if (IS_ERR(trans))
1270                         return PTR_ERR(trans);
1271         }
1272         return btrfs_commit_transaction(trans);
1273 }
1274
1275 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1276 {
1277         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1278         const char *compress_type;
1279
1280         if (btrfs_test_opt(info, DEGRADED))
1281                 seq_puts(seq, ",degraded");
1282         if (btrfs_test_opt(info, NODATASUM))
1283                 seq_puts(seq, ",nodatasum");
1284         if (btrfs_test_opt(info, NODATACOW))
1285                 seq_puts(seq, ",nodatacow");
1286         if (btrfs_test_opt(info, NOBARRIER))
1287                 seq_puts(seq, ",nobarrier");
1288         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1289                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1290         if (info->thread_pool_size !=  min_t(unsigned long,
1291                                              num_online_cpus() + 2, 8))
1292                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1293         if (btrfs_test_opt(info, COMPRESS)) {
1294                 compress_type = btrfs_compress_type2str(info->compress_type);
1295                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1296                         seq_printf(seq, ",compress-force=%s", compress_type);
1297                 else
1298                         seq_printf(seq, ",compress=%s", compress_type);
1299                 if (info->compress_level)
1300                         seq_printf(seq, ":%d", info->compress_level);
1301         }
1302         if (btrfs_test_opt(info, NOSSD))
1303                 seq_puts(seq, ",nossd");
1304         if (btrfs_test_opt(info, SSD_SPREAD))
1305                 seq_puts(seq, ",ssd_spread");
1306         else if (btrfs_test_opt(info, SSD))
1307                 seq_puts(seq, ",ssd");
1308         if (btrfs_test_opt(info, NOTREELOG))
1309                 seq_puts(seq, ",notreelog");
1310         if (btrfs_test_opt(info, NOLOGREPLAY))
1311                 seq_puts(seq, ",nologreplay");
1312         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1313                 seq_puts(seq, ",flushoncommit");
1314         if (btrfs_test_opt(info, DISCARD))
1315                 seq_puts(seq, ",discard");
1316         if (!(info->sb->s_flags & SB_POSIXACL))
1317                 seq_puts(seq, ",noacl");
1318         if (btrfs_test_opt(info, SPACE_CACHE))
1319                 seq_puts(seq, ",space_cache");
1320         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1321                 seq_puts(seq, ",space_cache=v2");
1322         else
1323                 seq_puts(seq, ",nospace_cache");
1324         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1325                 seq_puts(seq, ",rescan_uuid_tree");
1326         if (btrfs_test_opt(info, CLEAR_CACHE))
1327                 seq_puts(seq, ",clear_cache");
1328         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1329                 seq_puts(seq, ",user_subvol_rm_allowed");
1330         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1331                 seq_puts(seq, ",enospc_debug");
1332         if (btrfs_test_opt(info, AUTO_DEFRAG))
1333                 seq_puts(seq, ",autodefrag");
1334         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1335                 seq_puts(seq, ",inode_cache");
1336         if (btrfs_test_opt(info, SKIP_BALANCE))
1337                 seq_puts(seq, ",skip_balance");
1338 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1339         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1340                 seq_puts(seq, ",check_int_data");
1341         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1342                 seq_puts(seq, ",check_int");
1343         if (info->check_integrity_print_mask)
1344                 seq_printf(seq, ",check_int_print_mask=%d",
1345                                 info->check_integrity_print_mask);
1346 #endif
1347         if (info->metadata_ratio)
1348                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1349         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1350                 seq_puts(seq, ",fatal_errors=panic");
1351         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1352                 seq_printf(seq, ",commit=%u", info->commit_interval);
1353 #ifdef CONFIG_BTRFS_DEBUG
1354         if (btrfs_test_opt(info, FRAGMENT_DATA))
1355                 seq_puts(seq, ",fragment=data");
1356         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1357                 seq_puts(seq, ",fragment=metadata");
1358 #endif
1359         if (btrfs_test_opt(info, REF_VERIFY))
1360                 seq_puts(seq, ",ref_verify");
1361         seq_printf(seq, ",subvolid=%llu",
1362                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1363         seq_puts(seq, ",subvol=");
1364         seq_dentry(seq, dentry, " \t\n\\");
1365         return 0;
1366 }
1367
1368 static int btrfs_test_super(struct super_block *s, void *data)
1369 {
1370         struct btrfs_fs_info *p = data;
1371         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1372
1373         return fs_info->fs_devices == p->fs_devices;
1374 }
1375
1376 static int btrfs_set_super(struct super_block *s, void *data)
1377 {
1378         int err = set_anon_super(s, data);
1379         if (!err)
1380                 s->s_fs_info = data;
1381         return err;
1382 }
1383
1384 /*
1385  * subvolumes are identified by ino 256
1386  */
1387 static inline int is_subvolume_inode(struct inode *inode)
1388 {
1389         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1390                 return 1;
1391         return 0;
1392 }
1393
1394 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1395                                    const char *device_name, struct vfsmount *mnt)
1396 {
1397         struct dentry *root;
1398         int ret;
1399
1400         if (!subvol_name) {
1401                 if (!subvol_objectid) {
1402                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1403                                                           &subvol_objectid);
1404                         if (ret) {
1405                                 root = ERR_PTR(ret);
1406                                 goto out;
1407                         }
1408                 }
1409                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1410                                                             subvol_objectid);
1411                 if (IS_ERR(subvol_name)) {
1412                         root = ERR_CAST(subvol_name);
1413                         subvol_name = NULL;
1414                         goto out;
1415                 }
1416
1417         }
1418
1419         root = mount_subtree(mnt, subvol_name);
1420         /* mount_subtree() drops our reference on the vfsmount. */
1421         mnt = NULL;
1422
1423         if (!IS_ERR(root)) {
1424                 struct super_block *s = root->d_sb;
1425                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1426                 struct inode *root_inode = d_inode(root);
1427                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1428
1429                 ret = 0;
1430                 if (!is_subvolume_inode(root_inode)) {
1431                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1432                                subvol_name);
1433                         ret = -EINVAL;
1434                 }
1435                 if (subvol_objectid && root_objectid != subvol_objectid) {
1436                         /*
1437                          * This will also catch a race condition where a
1438                          * subvolume which was passed by ID is renamed and
1439                          * another subvolume is renamed over the old location.
1440                          */
1441                         btrfs_err(fs_info,
1442                                   "subvol '%s' does not match subvolid %llu",
1443                                   subvol_name, subvol_objectid);
1444                         ret = -EINVAL;
1445                 }
1446                 if (ret) {
1447                         dput(root);
1448                         root = ERR_PTR(ret);
1449                         deactivate_locked_super(s);
1450                 }
1451         }
1452
1453 out:
1454         mntput(mnt);
1455         kfree(subvol_name);
1456         return root;
1457 }
1458
1459 static int parse_security_options(char *orig_opts,
1460                                   struct security_mnt_opts *sec_opts)
1461 {
1462         char *secdata = NULL;
1463         int ret = 0;
1464
1465         secdata = alloc_secdata();
1466         if (!secdata)
1467                 return -ENOMEM;
1468         ret = security_sb_copy_data(orig_opts, secdata);
1469         if (ret) {
1470                 free_secdata(secdata);
1471                 return ret;
1472         }
1473         ret = security_sb_parse_opts_str(secdata, sec_opts);
1474         free_secdata(secdata);
1475         return ret;
1476 }
1477
1478 static int setup_security_options(struct btrfs_fs_info *fs_info,
1479                                   struct super_block *sb,
1480                                   struct security_mnt_opts *sec_opts)
1481 {
1482         int ret = 0;
1483
1484         /*
1485          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1486          * is valid.
1487          */
1488         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1489         if (ret)
1490                 return ret;
1491
1492 #ifdef CONFIG_SECURITY
1493         if (!fs_info->security_opts.num_mnt_opts) {
1494                 /* first time security setup, copy sec_opts to fs_info */
1495                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1496         } else {
1497                 /*
1498                  * Since SELinux (the only one supporting security_mnt_opts)
1499                  * does NOT support changing context during remount/mount of
1500                  * the same sb, this must be the same or part of the same
1501                  * security options, just free it.
1502                  */
1503                 security_free_mnt_opts(sec_opts);
1504         }
1505 #endif
1506         return ret;
1507 }
1508
1509 /*
1510  * Find a superblock for the given device / mount point.
1511  *
1512  * Note: This is based on mount_bdev from fs/super.c with a few additions
1513  *       for multiple device setup.  Make sure to keep it in sync.
1514  */
1515 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1516                 int flags, const char *device_name, void *data)
1517 {
1518         struct block_device *bdev = NULL;
1519         struct super_block *s;
1520         struct btrfs_fs_devices *fs_devices = NULL;
1521         struct btrfs_fs_info *fs_info = NULL;
1522         struct security_mnt_opts new_sec_opts;
1523         fmode_t mode = FMODE_READ;
1524         int error = 0;
1525
1526         if (!(flags & SB_RDONLY))
1527                 mode |= FMODE_WRITE;
1528
1529         mutex_lock(&uuid_mutex);
1530         error = btrfs_parse_early_options(data, mode, fs_type,
1531                                           &fs_devices);
1532         mutex_unlock(&uuid_mutex);
1533         if (error) {
1534                 return ERR_PTR(error);
1535         }
1536
1537         security_init_mnt_opts(&new_sec_opts);
1538         if (data) {
1539                 error = parse_security_options(data, &new_sec_opts);
1540                 if (error)
1541                         return ERR_PTR(error);
1542         }
1543
1544         mutex_lock(&uuid_mutex);
1545         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1546         mutex_unlock(&uuid_mutex);
1547         if (error)
1548                 goto error_sec_opts;
1549
1550         /*
1551          * Setup a dummy root and fs_info for test/set super.  This is because
1552          * we don't actually fill this stuff out until open_ctree, but we need
1553          * it for searching for existing supers, so this lets us do that and
1554          * then open_ctree will properly initialize everything later.
1555          */
1556         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1557         if (!fs_info) {
1558                 error = -ENOMEM;
1559                 goto error_sec_opts;
1560         }
1561
1562         fs_info->fs_devices = fs_devices;
1563
1564         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1565         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1566         security_init_mnt_opts(&fs_info->security_opts);
1567         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1568                 error = -ENOMEM;
1569                 goto error_fs_info;
1570         }
1571
1572         mutex_lock(&uuid_mutex);
1573         error = btrfs_open_devices(fs_devices, mode, fs_type);
1574         mutex_unlock(&uuid_mutex);
1575         if (error)
1576                 goto error_fs_info;
1577
1578         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1579                 error = -EACCES;
1580                 goto error_close_devices;
1581         }
1582
1583         bdev = fs_devices->latest_bdev;
1584         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1585                  fs_info);
1586         if (IS_ERR(s)) {
1587                 error = PTR_ERR(s);
1588                 goto error_close_devices;
1589         }
1590
1591         if (s->s_root) {
1592                 btrfs_close_devices(fs_devices);
1593                 free_fs_info(fs_info);
1594                 if ((flags ^ s->s_flags) & SB_RDONLY)
1595                         error = -EBUSY;
1596         } else {
1597                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1598                 btrfs_sb(s)->bdev_holder = fs_type;
1599                 error = btrfs_fill_super(s, fs_devices, data);
1600         }
1601         if (error) {
1602                 deactivate_locked_super(s);
1603                 goto error_sec_opts;
1604         }
1605
1606         fs_info = btrfs_sb(s);
1607         error = setup_security_options(fs_info, s, &new_sec_opts);
1608         if (error) {
1609                 deactivate_locked_super(s);
1610                 goto error_sec_opts;
1611         }
1612
1613         return dget(s->s_root);
1614
1615 error_close_devices:
1616         btrfs_close_devices(fs_devices);
1617 error_fs_info:
1618         free_fs_info(fs_info);
1619 error_sec_opts:
1620         security_free_mnt_opts(&new_sec_opts);
1621         return ERR_PTR(error);
1622 }
1623
1624 /*
1625  * Mount function which is called by VFS layer.
1626  *
1627  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1628  * which needs vfsmount* of device's root (/).  This means device's root has to
1629  * be mounted internally in any case.
1630  *
1631  * Operation flow:
1632  *   1. Parse subvol id related options for later use in mount_subvol().
1633  *
1634  *   2. Mount device's root (/) by calling vfs_kern_mount().
1635  *
1636  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1637  *      first place. In order to avoid calling btrfs_mount() again, we use
1638  *      different file_system_type which is not registered to VFS by
1639  *      register_filesystem() (btrfs_root_fs_type). As a result,
1640  *      btrfs_mount_root() is called. The return value will be used by
1641  *      mount_subtree() in mount_subvol().
1642  *
1643  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1644  *      "btrfs subvolume set-default", mount_subvol() is called always.
1645  */
1646 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1647                 const char *device_name, void *data)
1648 {
1649         struct vfsmount *mnt_root;
1650         struct dentry *root;
1651         fmode_t mode = FMODE_READ;
1652         char *subvol_name = NULL;
1653         u64 subvol_objectid = 0;
1654         int error = 0;
1655
1656         if (!(flags & SB_RDONLY))
1657                 mode |= FMODE_WRITE;
1658
1659         error = btrfs_parse_subvol_options(data, &subvol_name,
1660                                         &subvol_objectid);
1661         if (error) {
1662                 kfree(subvol_name);
1663                 return ERR_PTR(error);
1664         }
1665
1666         /* mount device's root (/) */
1667         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1668         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1669                 if (flags & SB_RDONLY) {
1670                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1671                                 flags & ~SB_RDONLY, device_name, data);
1672                 } else {
1673                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1674                                 flags | SB_RDONLY, device_name, data);
1675                         if (IS_ERR(mnt_root)) {
1676                                 root = ERR_CAST(mnt_root);
1677                                 goto out;
1678                         }
1679
1680                         down_write(&mnt_root->mnt_sb->s_umount);
1681                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1682                         up_write(&mnt_root->mnt_sb->s_umount);
1683                         if (error < 0) {
1684                                 root = ERR_PTR(error);
1685                                 mntput(mnt_root);
1686                                 goto out;
1687                         }
1688                 }
1689         }
1690         if (IS_ERR(mnt_root)) {
1691                 root = ERR_CAST(mnt_root);
1692                 goto out;
1693         }
1694
1695         /* mount_subvol() will free subvol_name and mnt_root */
1696         root = mount_subvol(subvol_name, subvol_objectid, device_name, mnt_root);
1697
1698 out:
1699         return root;
1700 }
1701
1702 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1703                                      u32 new_pool_size, u32 old_pool_size)
1704 {
1705         if (new_pool_size == old_pool_size)
1706                 return;
1707
1708         fs_info->thread_pool_size = new_pool_size;
1709
1710         btrfs_info(fs_info, "resize thread pool %d -> %d",
1711                old_pool_size, new_pool_size);
1712
1713         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1714         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1715         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1716         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1717         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1718         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1719         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1720                                 new_pool_size);
1721         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1722         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1723         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1724         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1725         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1726                                 new_pool_size);
1727 }
1728
1729 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1730 {
1731         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1732 }
1733
1734 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1735                                        unsigned long old_opts, int flags)
1736 {
1737         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1738             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1739              (flags & SB_RDONLY))) {
1740                 /* wait for any defraggers to finish */
1741                 wait_event(fs_info->transaction_wait,
1742                            (atomic_read(&fs_info->defrag_running) == 0));
1743                 if (flags & SB_RDONLY)
1744                         sync_filesystem(fs_info->sb);
1745         }
1746 }
1747
1748 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1749                                          unsigned long old_opts)
1750 {
1751         /*
1752          * We need to cleanup all defragable inodes if the autodefragment is
1753          * close or the filesystem is read only.
1754          */
1755         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1756             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1757                 btrfs_cleanup_defrag_inodes(fs_info);
1758         }
1759
1760         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1761 }
1762
1763 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1764 {
1765         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1766         struct btrfs_root *root = fs_info->tree_root;
1767         unsigned old_flags = sb->s_flags;
1768         unsigned long old_opts = fs_info->mount_opt;
1769         unsigned long old_compress_type = fs_info->compress_type;
1770         u64 old_max_inline = fs_info->max_inline;
1771         u32 old_thread_pool_size = fs_info->thread_pool_size;
1772         u32 old_metadata_ratio = fs_info->metadata_ratio;
1773         int ret;
1774
1775         sync_filesystem(sb);
1776         btrfs_remount_prepare(fs_info);
1777
1778         if (data) {
1779                 struct security_mnt_opts new_sec_opts;
1780
1781                 security_init_mnt_opts(&new_sec_opts);
1782                 ret = parse_security_options(data, &new_sec_opts);
1783                 if (ret)
1784                         goto restore;
1785                 ret = setup_security_options(fs_info, sb,
1786                                              &new_sec_opts);
1787                 if (ret) {
1788                         security_free_mnt_opts(&new_sec_opts);
1789                         goto restore;
1790                 }
1791         }
1792
1793         ret = btrfs_parse_options(fs_info, data, *flags);
1794         if (ret)
1795                 goto restore;
1796
1797         btrfs_remount_begin(fs_info, old_opts, *flags);
1798         btrfs_resize_thread_pool(fs_info,
1799                 fs_info->thread_pool_size, old_thread_pool_size);
1800
1801         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1802                 goto out;
1803
1804         if (*flags & SB_RDONLY) {
1805                 /*
1806                  * this also happens on 'umount -rf' or on shutdown, when
1807                  * the filesystem is busy.
1808                  */
1809                 cancel_work_sync(&fs_info->async_reclaim_work);
1810
1811                 /* wait for the uuid_scan task to finish */
1812                 down(&fs_info->uuid_tree_rescan_sem);
1813                 /* avoid complains from lockdep et al. */
1814                 up(&fs_info->uuid_tree_rescan_sem);
1815
1816                 sb->s_flags |= SB_RDONLY;
1817
1818                 /*
1819                  * Setting SB_RDONLY will put the cleaner thread to
1820                  * sleep at the next loop if it's already active.
1821                  * If it's already asleep, we'll leave unused block
1822                  * groups on disk until we're mounted read-write again
1823                  * unless we clean them up here.
1824                  */
1825                 btrfs_delete_unused_bgs(fs_info);
1826
1827                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1828                 btrfs_scrub_cancel(fs_info);
1829                 btrfs_pause_balance(fs_info);
1830
1831                 ret = btrfs_commit_super(fs_info);
1832                 if (ret)
1833                         goto restore;
1834         } else {
1835                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1836                         btrfs_err(fs_info,
1837                                 "Remounting read-write after error is not allowed");
1838                         ret = -EINVAL;
1839                         goto restore;
1840                 }
1841                 if (fs_info->fs_devices->rw_devices == 0) {
1842                         ret = -EACCES;
1843                         goto restore;
1844                 }
1845
1846                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1847                         btrfs_warn(fs_info,
1848                                 "too many missing devices, writeable remount is not allowed");
1849                         ret = -EACCES;
1850                         goto restore;
1851                 }
1852
1853                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1854                         ret = -EINVAL;
1855                         goto restore;
1856                 }
1857
1858                 ret = btrfs_cleanup_fs_roots(fs_info);
1859                 if (ret)
1860                         goto restore;
1861
1862                 /* recover relocation */
1863                 mutex_lock(&fs_info->cleaner_mutex);
1864                 ret = btrfs_recover_relocation(root);
1865                 mutex_unlock(&fs_info->cleaner_mutex);
1866                 if (ret)
1867                         goto restore;
1868
1869                 ret = btrfs_resume_balance_async(fs_info);
1870                 if (ret)
1871                         goto restore;
1872
1873                 ret = btrfs_resume_dev_replace_async(fs_info);
1874                 if (ret) {
1875                         btrfs_warn(fs_info, "failed to resume dev_replace");
1876                         goto restore;
1877                 }
1878
1879                 btrfs_qgroup_rescan_resume(fs_info);
1880
1881                 if (!fs_info->uuid_root) {
1882                         btrfs_info(fs_info, "creating UUID tree");
1883                         ret = btrfs_create_uuid_tree(fs_info);
1884                         if (ret) {
1885                                 btrfs_warn(fs_info,
1886                                            "failed to create the UUID tree %d",
1887                                            ret);
1888                                 goto restore;
1889                         }
1890                 }
1891                 sb->s_flags &= ~SB_RDONLY;
1892
1893                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1894         }
1895 out:
1896         wake_up_process(fs_info->transaction_kthread);
1897         btrfs_remount_cleanup(fs_info, old_opts);
1898         return 0;
1899
1900 restore:
1901         /* We've hit an error - don't reset SB_RDONLY */
1902         if (sb_rdonly(sb))
1903                 old_flags |= SB_RDONLY;
1904         sb->s_flags = old_flags;
1905         fs_info->mount_opt = old_opts;
1906         fs_info->compress_type = old_compress_type;
1907         fs_info->max_inline = old_max_inline;
1908         btrfs_resize_thread_pool(fs_info,
1909                 old_thread_pool_size, fs_info->thread_pool_size);
1910         fs_info->metadata_ratio = old_metadata_ratio;
1911         btrfs_remount_cleanup(fs_info, old_opts);
1912         return ret;
1913 }
1914
1915 /* Used to sort the devices by max_avail(descending sort) */
1916 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1917                                        const void *dev_info2)
1918 {
1919         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1920             ((struct btrfs_device_info *)dev_info2)->max_avail)
1921                 return -1;
1922         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1923                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1924                 return 1;
1925         else
1926         return 0;
1927 }
1928
1929 /*
1930  * sort the devices by max_avail, in which max free extent size of each device
1931  * is stored.(Descending Sort)
1932  */
1933 static inline void btrfs_descending_sort_devices(
1934                                         struct btrfs_device_info *devices,
1935                                         size_t nr_devices)
1936 {
1937         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1938              btrfs_cmp_device_free_bytes, NULL);
1939 }
1940
1941 /*
1942  * The helper to calc the free space on the devices that can be used to store
1943  * file data.
1944  */
1945 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1946                                        u64 *free_bytes)
1947 {
1948         struct btrfs_device_info *devices_info;
1949         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1950         struct btrfs_device *device;
1951         u64 skip_space;
1952         u64 type;
1953         u64 avail_space;
1954         u64 min_stripe_size;
1955         int min_stripes = 1, num_stripes = 1;
1956         int i = 0, nr_devices;
1957
1958         /*
1959          * We aren't under the device list lock, so this is racy-ish, but good
1960          * enough for our purposes.
1961          */
1962         nr_devices = fs_info->fs_devices->open_devices;
1963         if (!nr_devices) {
1964                 smp_mb();
1965                 nr_devices = fs_info->fs_devices->open_devices;
1966                 ASSERT(nr_devices);
1967                 if (!nr_devices) {
1968                         *free_bytes = 0;
1969                         return 0;
1970                 }
1971         }
1972
1973         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1974                                GFP_KERNEL);
1975         if (!devices_info)
1976                 return -ENOMEM;
1977
1978         /* calc min stripe number for data space allocation */
1979         type = btrfs_data_alloc_profile(fs_info);
1980         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1981                 min_stripes = 2;
1982                 num_stripes = nr_devices;
1983         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1984                 min_stripes = 2;
1985                 num_stripes = 2;
1986         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1987                 min_stripes = 4;
1988                 num_stripes = 4;
1989         }
1990
1991         if (type & BTRFS_BLOCK_GROUP_DUP)
1992                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1993         else
1994                 min_stripe_size = BTRFS_STRIPE_LEN;
1995
1996         rcu_read_lock();
1997         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1998                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1999                                                 &device->dev_state) ||
2000                     !device->bdev ||
2001                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2002                         continue;
2003
2004                 if (i >= nr_devices)
2005                         break;
2006
2007                 avail_space = device->total_bytes - device->bytes_used;
2008
2009                 /* align with stripe_len */
2010                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
2011                 avail_space *= BTRFS_STRIPE_LEN;
2012
2013                 /*
2014                  * In order to avoid overwriting the superblock on the drive,
2015                  * btrfs starts at an offset of at least 1MB when doing chunk
2016                  * allocation.
2017                  */
2018                 skip_space = SZ_1M;
2019
2020                 /*
2021                  * we can use the free space in [0, skip_space - 1], subtract
2022                  * it from the total.
2023                  */
2024                 if (avail_space && avail_space >= skip_space)
2025                         avail_space -= skip_space;
2026                 else
2027                         avail_space = 0;
2028
2029                 if (avail_space < min_stripe_size)
2030                         continue;
2031
2032                 devices_info[i].dev = device;
2033                 devices_info[i].max_avail = avail_space;
2034
2035                 i++;
2036         }
2037         rcu_read_unlock();
2038
2039         nr_devices = i;
2040
2041         btrfs_descending_sort_devices(devices_info, nr_devices);
2042
2043         i = nr_devices - 1;
2044         avail_space = 0;
2045         while (nr_devices >= min_stripes) {
2046                 if (num_stripes > nr_devices)
2047                         num_stripes = nr_devices;
2048
2049                 if (devices_info[i].max_avail >= min_stripe_size) {
2050                         int j;
2051                         u64 alloc_size;
2052
2053                         avail_space += devices_info[i].max_avail * num_stripes;
2054                         alloc_size = devices_info[i].max_avail;
2055                         for (j = i + 1 - num_stripes; j <= i; j++)
2056                                 devices_info[j].max_avail -= alloc_size;
2057                 }
2058                 i--;
2059                 nr_devices--;
2060         }
2061
2062         kfree(devices_info);
2063         *free_bytes = avail_space;
2064         return 0;
2065 }
2066
2067 /*
2068  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2069  *
2070  * If there's a redundant raid level at DATA block groups, use the respective
2071  * multiplier to scale the sizes.
2072  *
2073  * Unused device space usage is based on simulating the chunk allocator
2074  * algorithm that respects the device sizes and order of allocations.  This is
2075  * a close approximation of the actual use but there are other factors that may
2076  * change the result (like a new metadata chunk).
2077  *
2078  * If metadata is exhausted, f_bavail will be 0.
2079  */
2080 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2081 {
2082         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2083         struct btrfs_super_block *disk_super = fs_info->super_copy;
2084         struct list_head *head = &fs_info->space_info;
2085         struct btrfs_space_info *found;
2086         u64 total_used = 0;
2087         u64 total_free_data = 0;
2088         u64 total_free_meta = 0;
2089         int bits = dentry->d_sb->s_blocksize_bits;
2090         __be32 *fsid = (__be32 *)fs_info->fsid;
2091         unsigned factor = 1;
2092         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2093         int ret;
2094         u64 thresh = 0;
2095         int mixed = 0;
2096
2097         rcu_read_lock();
2098         list_for_each_entry_rcu(found, head, list) {
2099                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2100                         int i;
2101
2102                         total_free_data += found->disk_total - found->disk_used;
2103                         total_free_data -=
2104                                 btrfs_account_ro_block_groups_free_space(found);
2105
2106                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2107                                 if (!list_empty(&found->block_groups[i])) {
2108                                         switch (i) {
2109                                         case BTRFS_RAID_DUP:
2110                                         case BTRFS_RAID_RAID1:
2111                                         case BTRFS_RAID_RAID10:
2112                                                 factor = 2;
2113                                         }
2114                                 }
2115                         }
2116                 }
2117
2118                 /*
2119                  * Metadata in mixed block goup profiles are accounted in data
2120                  */
2121                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2122                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2123                                 mixed = 1;
2124                         else
2125                                 total_free_meta += found->disk_total -
2126                                         found->disk_used;
2127                 }
2128
2129                 total_used += found->disk_used;
2130         }
2131
2132         rcu_read_unlock();
2133
2134         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2135         buf->f_blocks >>= bits;
2136         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2137
2138         /* Account global block reserve as used, it's in logical size already */
2139         spin_lock(&block_rsv->lock);
2140         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2141         if (buf->f_bfree >= block_rsv->size >> bits)
2142                 buf->f_bfree -= block_rsv->size >> bits;
2143         else
2144                 buf->f_bfree = 0;
2145         spin_unlock(&block_rsv->lock);
2146
2147         buf->f_bavail = div_u64(total_free_data, factor);
2148         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2149         if (ret)
2150                 return ret;
2151         buf->f_bavail += div_u64(total_free_data, factor);
2152         buf->f_bavail = buf->f_bavail >> bits;
2153
2154         /*
2155          * We calculate the remaining metadata space minus global reserve. If
2156          * this is (supposedly) smaller than zero, there's no space. But this
2157          * does not hold in practice, the exhausted state happens where's still
2158          * some positive delta. So we apply some guesswork and compare the
2159          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2160          *
2161          * We probably cannot calculate the exact threshold value because this
2162          * depends on the internal reservations requested by various
2163          * operations, so some operations that consume a few metadata will
2164          * succeed even if the Avail is zero. But this is better than the other
2165          * way around.
2166          */
2167         thresh = SZ_4M;
2168
2169         if (!mixed && total_free_meta - thresh < block_rsv->size)
2170                 buf->f_bavail = 0;
2171
2172         buf->f_type = BTRFS_SUPER_MAGIC;
2173         buf->f_bsize = dentry->d_sb->s_blocksize;
2174         buf->f_namelen = BTRFS_NAME_LEN;
2175
2176         /* We treat it as constant endianness (it doesn't matter _which_)
2177            because we want the fsid to come out the same whether mounted
2178            on a big-endian or little-endian host */
2179         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2180         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2181         /* Mask in the root object ID too, to disambiguate subvols */
2182         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2183         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2184
2185         return 0;
2186 }
2187
2188 static void btrfs_kill_super(struct super_block *sb)
2189 {
2190         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2191         kill_anon_super(sb);
2192         free_fs_info(fs_info);
2193 }
2194
2195 static struct file_system_type btrfs_fs_type = {
2196         .owner          = THIS_MODULE,
2197         .name           = "btrfs",
2198         .mount          = btrfs_mount,
2199         .kill_sb        = btrfs_kill_super,
2200         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2201 };
2202
2203 static struct file_system_type btrfs_root_fs_type = {
2204         .owner          = THIS_MODULE,
2205         .name           = "btrfs",
2206         .mount          = btrfs_mount_root,
2207         .kill_sb        = btrfs_kill_super,
2208         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2209 };
2210
2211 MODULE_ALIAS_FS("btrfs");
2212
2213 static int btrfs_control_open(struct inode *inode, struct file *file)
2214 {
2215         /*
2216          * The control file's private_data is used to hold the
2217          * transaction when it is started and is used to keep
2218          * track of whether a transaction is already in progress.
2219          */
2220         file->private_data = NULL;
2221         return 0;
2222 }
2223
2224 /*
2225  * used by btrfsctl to scan devices when no FS is mounted
2226  */
2227 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2228                                 unsigned long arg)
2229 {
2230         struct btrfs_ioctl_vol_args *vol;
2231         struct btrfs_fs_devices *fs_devices;
2232         int ret = -ENOTTY;
2233
2234         if (!capable(CAP_SYS_ADMIN))
2235                 return -EPERM;
2236
2237         vol = memdup_user((void __user *)arg, sizeof(*vol));
2238         if (IS_ERR(vol))
2239                 return PTR_ERR(vol);
2240
2241         switch (cmd) {
2242         case BTRFS_IOC_SCAN_DEV:
2243                 mutex_lock(&uuid_mutex);
2244                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2245                                             &btrfs_root_fs_type, &fs_devices);
2246                 mutex_unlock(&uuid_mutex);
2247                 break;
2248         case BTRFS_IOC_DEVICES_READY:
2249                 mutex_lock(&uuid_mutex);
2250                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2251                                             &btrfs_root_fs_type, &fs_devices);
2252                 if (ret) {
2253                         mutex_unlock(&uuid_mutex);
2254                         break;
2255                 }
2256                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2257                 mutex_unlock(&uuid_mutex);
2258                 break;
2259         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2260                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2261                 break;
2262         }
2263
2264         kfree(vol);
2265         return ret;
2266 }
2267
2268 static int btrfs_freeze(struct super_block *sb)
2269 {
2270         struct btrfs_trans_handle *trans;
2271         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2272         struct btrfs_root *root = fs_info->tree_root;
2273
2274         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2275         /*
2276          * We don't need a barrier here, we'll wait for any transaction that
2277          * could be in progress on other threads (and do delayed iputs that
2278          * we want to avoid on a frozen filesystem), or do the commit
2279          * ourselves.
2280          */
2281         trans = btrfs_attach_transaction_barrier(root);
2282         if (IS_ERR(trans)) {
2283                 /* no transaction, don't bother */
2284                 if (PTR_ERR(trans) == -ENOENT)
2285                         return 0;
2286                 return PTR_ERR(trans);
2287         }
2288         return btrfs_commit_transaction(trans);
2289 }
2290
2291 static int btrfs_unfreeze(struct super_block *sb)
2292 {
2293         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2294
2295         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2296         return 0;
2297 }
2298
2299 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2300 {
2301         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2302         struct btrfs_fs_devices *cur_devices;
2303         struct btrfs_device *dev, *first_dev = NULL;
2304         struct list_head *head;
2305         struct rcu_string *name;
2306
2307         /*
2308          * Lightweight locking of the devices. We should not need
2309          * device_list_mutex here as we only read the device data and the list
2310          * is protected by RCU.  Even if a device is deleted during the list
2311          * traversals, we'll get valid data, the freeing callback will wait at
2312          * least until until the rcu_read_unlock.
2313          */
2314         rcu_read_lock();
2315         cur_devices = fs_info->fs_devices;
2316         while (cur_devices) {
2317                 head = &cur_devices->devices;
2318                 list_for_each_entry_rcu(dev, head, dev_list) {
2319                         if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2320                                 continue;
2321                         if (!dev->name)
2322                                 continue;
2323                         if (!first_dev || dev->devid < first_dev->devid)
2324                                 first_dev = dev;
2325                 }
2326                 cur_devices = cur_devices->seed;
2327         }
2328
2329         if (first_dev) {
2330                 name = rcu_dereference(first_dev->name);
2331                 seq_escape(m, name->str, " \t\n\\");
2332         } else {
2333                 WARN_ON(1);
2334         }
2335         rcu_read_unlock();
2336         return 0;
2337 }
2338
2339 static const struct super_operations btrfs_super_ops = {
2340         .drop_inode     = btrfs_drop_inode,
2341         .evict_inode    = btrfs_evict_inode,
2342         .put_super      = btrfs_put_super,
2343         .sync_fs        = btrfs_sync_fs,
2344         .show_options   = btrfs_show_options,
2345         .show_devname   = btrfs_show_devname,
2346         .write_inode    = btrfs_write_inode,
2347         .alloc_inode    = btrfs_alloc_inode,
2348         .destroy_inode  = btrfs_destroy_inode,
2349         .statfs         = btrfs_statfs,
2350         .remount_fs     = btrfs_remount,
2351         .freeze_fs      = btrfs_freeze,
2352         .unfreeze_fs    = btrfs_unfreeze,
2353 };
2354
2355 static const struct file_operations btrfs_ctl_fops = {
2356         .open = btrfs_control_open,
2357         .unlocked_ioctl  = btrfs_control_ioctl,
2358         .compat_ioctl = btrfs_control_ioctl,
2359         .owner   = THIS_MODULE,
2360         .llseek = noop_llseek,
2361 };
2362
2363 static struct miscdevice btrfs_misc = {
2364         .minor          = BTRFS_MINOR,
2365         .name           = "btrfs-control",
2366         .fops           = &btrfs_ctl_fops
2367 };
2368
2369 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2370 MODULE_ALIAS("devname:btrfs-control");
2371
2372 static int __init btrfs_interface_init(void)
2373 {
2374         return misc_register(&btrfs_misc);
2375 }
2376
2377 static __cold void btrfs_interface_exit(void)
2378 {
2379         misc_deregister(&btrfs_misc);
2380 }
2381
2382 static void __init btrfs_print_mod_info(void)
2383 {
2384         static const char options[] = ""
2385 #ifdef CONFIG_BTRFS_DEBUG
2386                         ", debug=on"
2387 #endif
2388 #ifdef CONFIG_BTRFS_ASSERT
2389                         ", assert=on"
2390 #endif
2391 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2392                         ", integrity-checker=on"
2393 #endif
2394 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2395                         ", ref-verify=on"
2396 #endif
2397                         ;
2398         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2399 }
2400
2401 static int __init init_btrfs_fs(void)
2402 {
2403         int err;
2404
2405         btrfs_props_init();
2406
2407         err = btrfs_init_sysfs();
2408         if (err)
2409                 return err;
2410
2411         btrfs_init_compress();
2412
2413         err = btrfs_init_cachep();
2414         if (err)
2415                 goto free_compress;
2416
2417         err = extent_io_init();
2418         if (err)
2419                 goto free_cachep;
2420
2421         err = extent_map_init();
2422         if (err)
2423                 goto free_extent_io;
2424
2425         err = ordered_data_init();
2426         if (err)
2427                 goto free_extent_map;
2428
2429         err = btrfs_delayed_inode_init();
2430         if (err)
2431                 goto free_ordered_data;
2432
2433         err = btrfs_auto_defrag_init();
2434         if (err)
2435                 goto free_delayed_inode;
2436
2437         err = btrfs_delayed_ref_init();
2438         if (err)
2439                 goto free_auto_defrag;
2440
2441         err = btrfs_prelim_ref_init();
2442         if (err)
2443                 goto free_delayed_ref;
2444
2445         err = btrfs_end_io_wq_init();
2446         if (err)
2447                 goto free_prelim_ref;
2448
2449         err = btrfs_interface_init();
2450         if (err)
2451                 goto free_end_io_wq;
2452
2453         btrfs_init_lockdep();
2454
2455         btrfs_print_mod_info();
2456
2457         err = btrfs_run_sanity_tests();
2458         if (err)
2459                 goto unregister_ioctl;
2460
2461         err = register_filesystem(&btrfs_fs_type);
2462         if (err)
2463                 goto unregister_ioctl;
2464
2465         return 0;
2466
2467 unregister_ioctl:
2468         btrfs_interface_exit();
2469 free_end_io_wq:
2470         btrfs_end_io_wq_exit();
2471 free_prelim_ref:
2472         btrfs_prelim_ref_exit();
2473 free_delayed_ref:
2474         btrfs_delayed_ref_exit();
2475 free_auto_defrag:
2476         btrfs_auto_defrag_exit();
2477 free_delayed_inode:
2478         btrfs_delayed_inode_exit();
2479 free_ordered_data:
2480         ordered_data_exit();
2481 free_extent_map:
2482         extent_map_exit();
2483 free_extent_io:
2484         extent_io_exit();
2485 free_cachep:
2486         btrfs_destroy_cachep();
2487 free_compress:
2488         btrfs_exit_compress();
2489         btrfs_exit_sysfs();
2490
2491         return err;
2492 }
2493
2494 static void __exit exit_btrfs_fs(void)
2495 {
2496         btrfs_destroy_cachep();
2497         btrfs_delayed_ref_exit();
2498         btrfs_auto_defrag_exit();
2499         btrfs_delayed_inode_exit();
2500         btrfs_prelim_ref_exit();
2501         ordered_data_exit();
2502         extent_map_exit();
2503         extent_io_exit();
2504         btrfs_interface_exit();
2505         btrfs_end_io_wq_exit();
2506         unregister_filesystem(&btrfs_fs_type);
2507         btrfs_exit_sysfs();
2508         btrfs_cleanup_fs_uuids();
2509         btrfs_exit_compress();
2510 }
2511
2512 late_initcall(init_btrfs_fs);
2513 module_exit(exit_btrfs_fs)
2514
2515 MODULE_LICENSE("GPL");