39d8e39b2fe13d3db8c29b500a101d2c4151b865
[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         if (!options)
894                 return 0;
895
896         /*
897          * strsep changes the string, duplicate it because btrfs_parse_options
898          * gets called later
899          */
900         opts = kstrdup(options, GFP_KERNEL);
901         if (!opts)
902                 return -ENOMEM;
903         orig = opts;
904
905         while ((p = strsep(&opts, ",")) != NULL) {
906                 int token;
907
908                 if (!*p)
909                         continue;
910
911                 token = match_token(p, tokens, args);
912                 if (token == Opt_device) {
913                         device_name = match_strdup(&args[0]);
914                         if (!device_name) {
915                                 error = -ENOMEM;
916                                 goto out;
917                         }
918                         error = btrfs_scan_one_device(device_name,
919                                         flags, holder, fs_devices);
920                         kfree(device_name);
921                         if (error)
922                                 goto out;
923                 }
924         }
925
926 out:
927         kfree(orig);
928         return error;
929 }
930
931 /*
932  * Parse mount options that are related to subvolume id
933  *
934  * The value is later passed to mount_subvol()
935  */
936 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
937                 u64 *subvol_objectid)
938 {
939         substring_t args[MAX_OPT_ARGS];
940         char *opts, *orig, *p;
941         int error = 0;
942         u64 subvolid;
943
944         if (!options)
945                 return 0;
946
947         /*
948          * strsep changes the string, duplicate it because
949          * btrfs_parse_early_options gets called later
950          */
951         opts = kstrdup(options, GFP_KERNEL);
952         if (!opts)
953                 return -ENOMEM;
954         orig = opts;
955
956         while ((p = strsep(&opts, ",")) != NULL) {
957                 int token;
958                 if (!*p)
959                         continue;
960
961                 token = match_token(p, tokens, args);
962                 switch (token) {
963                 case Opt_subvol:
964                         kfree(*subvol_name);
965                         *subvol_name = match_strdup(&args[0]);
966                         if (!*subvol_name) {
967                                 error = -ENOMEM;
968                                 goto out;
969                         }
970                         break;
971                 case Opt_subvolid:
972                         error = match_u64(&args[0], &subvolid);
973                         if (error)
974                                 goto out;
975
976                         /* we want the original fs_tree */
977                         if (subvolid == 0)
978                                 subvolid = BTRFS_FS_TREE_OBJECTID;
979
980                         *subvol_objectid = subvolid;
981                         break;
982                 case Opt_subvolrootid:
983                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
984                         break;
985                 default:
986                         break;
987                 }
988         }
989
990 out:
991         kfree(orig);
992         return error;
993 }
994
995 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
996                                            u64 subvol_objectid)
997 {
998         struct btrfs_root *root = fs_info->tree_root;
999         struct btrfs_root *fs_root;
1000         struct btrfs_root_ref *root_ref;
1001         struct btrfs_inode_ref *inode_ref;
1002         struct btrfs_key key;
1003         struct btrfs_path *path = NULL;
1004         char *name = NULL, *ptr;
1005         u64 dirid;
1006         int len;
1007         int ret;
1008
1009         path = btrfs_alloc_path();
1010         if (!path) {
1011                 ret = -ENOMEM;
1012                 goto err;
1013         }
1014         path->leave_spinning = 1;
1015
1016         name = kmalloc(PATH_MAX, GFP_KERNEL);
1017         if (!name) {
1018                 ret = -ENOMEM;
1019                 goto err;
1020         }
1021         ptr = name + PATH_MAX - 1;
1022         ptr[0] = '\0';
1023
1024         /*
1025          * Walk up the subvolume trees in the tree of tree roots by root
1026          * backrefs until we hit the top-level subvolume.
1027          */
1028         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1029                 key.objectid = subvol_objectid;
1030                 key.type = BTRFS_ROOT_BACKREF_KEY;
1031                 key.offset = (u64)-1;
1032
1033                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1034                 if (ret < 0) {
1035                         goto err;
1036                 } else if (ret > 0) {
1037                         ret = btrfs_previous_item(root, path, subvol_objectid,
1038                                                   BTRFS_ROOT_BACKREF_KEY);
1039                         if (ret < 0) {
1040                                 goto err;
1041                         } else if (ret > 0) {
1042                                 ret = -ENOENT;
1043                                 goto err;
1044                         }
1045                 }
1046
1047                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1048                 subvol_objectid = key.offset;
1049
1050                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1051                                           struct btrfs_root_ref);
1052                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1053                 ptr -= len + 1;
1054                 if (ptr < name) {
1055                         ret = -ENAMETOOLONG;
1056                         goto err;
1057                 }
1058                 read_extent_buffer(path->nodes[0], ptr + 1,
1059                                    (unsigned long)(root_ref + 1), len);
1060                 ptr[0] = '/';
1061                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1062                 btrfs_release_path(path);
1063
1064                 key.objectid = subvol_objectid;
1065                 key.type = BTRFS_ROOT_ITEM_KEY;
1066                 key.offset = (u64)-1;
1067                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1068                 if (IS_ERR(fs_root)) {
1069                         ret = PTR_ERR(fs_root);
1070                         goto err;
1071                 }
1072
1073                 /*
1074                  * Walk up the filesystem tree by inode refs until we hit the
1075                  * root directory.
1076                  */
1077                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1078                         key.objectid = dirid;
1079                         key.type = BTRFS_INODE_REF_KEY;
1080                         key.offset = (u64)-1;
1081
1082                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1083                         if (ret < 0) {
1084                                 goto err;
1085                         } else if (ret > 0) {
1086                                 ret = btrfs_previous_item(fs_root, path, dirid,
1087                                                           BTRFS_INODE_REF_KEY);
1088                                 if (ret < 0) {
1089                                         goto err;
1090                                 } else if (ret > 0) {
1091                                         ret = -ENOENT;
1092                                         goto err;
1093                                 }
1094                         }
1095
1096                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1097                         dirid = key.offset;
1098
1099                         inode_ref = btrfs_item_ptr(path->nodes[0],
1100                                                    path->slots[0],
1101                                                    struct btrfs_inode_ref);
1102                         len = btrfs_inode_ref_name_len(path->nodes[0],
1103                                                        inode_ref);
1104                         ptr -= len + 1;
1105                         if (ptr < name) {
1106                                 ret = -ENAMETOOLONG;
1107                                 goto err;
1108                         }
1109                         read_extent_buffer(path->nodes[0], ptr + 1,
1110                                            (unsigned long)(inode_ref + 1), len);
1111                         ptr[0] = '/';
1112                         btrfs_release_path(path);
1113                 }
1114         }
1115
1116         btrfs_free_path(path);
1117         if (ptr == name + PATH_MAX - 1) {
1118                 name[0] = '/';
1119                 name[1] = '\0';
1120         } else {
1121                 memmove(name, ptr, name + PATH_MAX - ptr);
1122         }
1123         return name;
1124
1125 err:
1126         btrfs_free_path(path);
1127         kfree(name);
1128         return ERR_PTR(ret);
1129 }
1130
1131 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1132 {
1133         struct btrfs_root *root = fs_info->tree_root;
1134         struct btrfs_dir_item *di;
1135         struct btrfs_path *path;
1136         struct btrfs_key location;
1137         u64 dir_id;
1138
1139         path = btrfs_alloc_path();
1140         if (!path)
1141                 return -ENOMEM;
1142         path->leave_spinning = 1;
1143
1144         /*
1145          * Find the "default" dir item which points to the root item that we
1146          * will mount by default if we haven't been given a specific subvolume
1147          * to mount.
1148          */
1149         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1150         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1151         if (IS_ERR(di)) {
1152                 btrfs_free_path(path);
1153                 return PTR_ERR(di);
1154         }
1155         if (!di) {
1156                 /*
1157                  * Ok the default dir item isn't there.  This is weird since
1158                  * it's always been there, but don't freak out, just try and
1159                  * mount the top-level subvolume.
1160                  */
1161                 btrfs_free_path(path);
1162                 *objectid = BTRFS_FS_TREE_OBJECTID;
1163                 return 0;
1164         }
1165
1166         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1167         btrfs_free_path(path);
1168         *objectid = location.objectid;
1169         return 0;
1170 }
1171
1172 static int btrfs_fill_super(struct super_block *sb,
1173                             struct btrfs_fs_devices *fs_devices,
1174                             void *data)
1175 {
1176         struct inode *inode;
1177         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1178         struct btrfs_key key;
1179         int err;
1180
1181         sb->s_maxbytes = MAX_LFS_FILESIZE;
1182         sb->s_magic = BTRFS_SUPER_MAGIC;
1183         sb->s_op = &btrfs_super_ops;
1184         sb->s_d_op = &btrfs_dentry_operations;
1185         sb->s_export_op = &btrfs_export_ops;
1186         sb->s_xattr = btrfs_xattr_handlers;
1187         sb->s_time_gran = 1;
1188 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1189         sb->s_flags |= SB_POSIXACL;
1190 #endif
1191         sb->s_flags |= SB_I_VERSION;
1192         sb->s_iflags |= SB_I_CGROUPWB;
1193
1194         err = super_setup_bdi(sb);
1195         if (err) {
1196                 btrfs_err(fs_info, "super_setup_bdi failed");
1197                 return err;
1198         }
1199
1200         err = open_ctree(sb, fs_devices, (char *)data);
1201         if (err) {
1202                 btrfs_err(fs_info, "open_ctree failed");
1203                 return err;
1204         }
1205
1206         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1207         key.type = BTRFS_INODE_ITEM_KEY;
1208         key.offset = 0;
1209         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1210         if (IS_ERR(inode)) {
1211                 err = PTR_ERR(inode);
1212                 goto fail_close;
1213         }
1214
1215         sb->s_root = d_make_root(inode);
1216         if (!sb->s_root) {
1217                 err = -ENOMEM;
1218                 goto fail_close;
1219         }
1220
1221         cleancache_init_fs(sb);
1222         sb->s_flags |= SB_ACTIVE;
1223         return 0;
1224
1225 fail_close:
1226         close_ctree(fs_info);
1227         return err;
1228 }
1229
1230 int btrfs_sync_fs(struct super_block *sb, int wait)
1231 {
1232         struct btrfs_trans_handle *trans;
1233         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1234         struct btrfs_root *root = fs_info->tree_root;
1235
1236         trace_btrfs_sync_fs(fs_info, wait);
1237
1238         if (!wait) {
1239                 filemap_flush(fs_info->btree_inode->i_mapping);
1240                 return 0;
1241         }
1242
1243         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1244
1245         trans = btrfs_attach_transaction_barrier(root);
1246         if (IS_ERR(trans)) {
1247                 /* no transaction, don't bother */
1248                 if (PTR_ERR(trans) == -ENOENT) {
1249                         /*
1250                          * Exit unless we have some pending changes
1251                          * that need to go through commit
1252                          */
1253                         if (fs_info->pending_changes == 0)
1254                                 return 0;
1255                         /*
1256                          * A non-blocking test if the fs is frozen. We must not
1257                          * start a new transaction here otherwise a deadlock
1258                          * happens. The pending operations are delayed to the
1259                          * next commit after thawing.
1260                          */
1261                         if (sb_start_write_trylock(sb))
1262                                 sb_end_write(sb);
1263                         else
1264                                 return 0;
1265                         trans = btrfs_start_transaction(root, 0);
1266                 }
1267                 if (IS_ERR(trans))
1268                         return PTR_ERR(trans);
1269         }
1270         return btrfs_commit_transaction(trans);
1271 }
1272
1273 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1274 {
1275         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1276         const char *compress_type;
1277
1278         if (btrfs_test_opt(info, DEGRADED))
1279                 seq_puts(seq, ",degraded");
1280         if (btrfs_test_opt(info, NODATASUM))
1281                 seq_puts(seq, ",nodatasum");
1282         if (btrfs_test_opt(info, NODATACOW))
1283                 seq_puts(seq, ",nodatacow");
1284         if (btrfs_test_opt(info, NOBARRIER))
1285                 seq_puts(seq, ",nobarrier");
1286         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1287                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1288         if (info->thread_pool_size !=  min_t(unsigned long,
1289                                              num_online_cpus() + 2, 8))
1290                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1291         if (btrfs_test_opt(info, COMPRESS)) {
1292                 compress_type = btrfs_compress_type2str(info->compress_type);
1293                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1294                         seq_printf(seq, ",compress-force=%s", compress_type);
1295                 else
1296                         seq_printf(seq, ",compress=%s", compress_type);
1297                 if (info->compress_level)
1298                         seq_printf(seq, ":%d", info->compress_level);
1299         }
1300         if (btrfs_test_opt(info, NOSSD))
1301                 seq_puts(seq, ",nossd");
1302         if (btrfs_test_opt(info, SSD_SPREAD))
1303                 seq_puts(seq, ",ssd_spread");
1304         else if (btrfs_test_opt(info, SSD))
1305                 seq_puts(seq, ",ssd");
1306         if (btrfs_test_opt(info, NOTREELOG))
1307                 seq_puts(seq, ",notreelog");
1308         if (btrfs_test_opt(info, NOLOGREPLAY))
1309                 seq_puts(seq, ",nologreplay");
1310         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1311                 seq_puts(seq, ",flushoncommit");
1312         if (btrfs_test_opt(info, DISCARD))
1313                 seq_puts(seq, ",discard");
1314         if (!(info->sb->s_flags & SB_POSIXACL))
1315                 seq_puts(seq, ",noacl");
1316         if (btrfs_test_opt(info, SPACE_CACHE))
1317                 seq_puts(seq, ",space_cache");
1318         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1319                 seq_puts(seq, ",space_cache=v2");
1320         else
1321                 seq_puts(seq, ",nospace_cache");
1322         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1323                 seq_puts(seq, ",rescan_uuid_tree");
1324         if (btrfs_test_opt(info, CLEAR_CACHE))
1325                 seq_puts(seq, ",clear_cache");
1326         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1327                 seq_puts(seq, ",user_subvol_rm_allowed");
1328         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1329                 seq_puts(seq, ",enospc_debug");
1330         if (btrfs_test_opt(info, AUTO_DEFRAG))
1331                 seq_puts(seq, ",autodefrag");
1332         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1333                 seq_puts(seq, ",inode_cache");
1334         if (btrfs_test_opt(info, SKIP_BALANCE))
1335                 seq_puts(seq, ",skip_balance");
1336 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1337         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1338                 seq_puts(seq, ",check_int_data");
1339         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1340                 seq_puts(seq, ",check_int");
1341         if (info->check_integrity_print_mask)
1342                 seq_printf(seq, ",check_int_print_mask=%d",
1343                                 info->check_integrity_print_mask);
1344 #endif
1345         if (info->metadata_ratio)
1346                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1347         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1348                 seq_puts(seq, ",fatal_errors=panic");
1349         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1350                 seq_printf(seq, ",commit=%u", info->commit_interval);
1351 #ifdef CONFIG_BTRFS_DEBUG
1352         if (btrfs_test_opt(info, FRAGMENT_DATA))
1353                 seq_puts(seq, ",fragment=data");
1354         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1355                 seq_puts(seq, ",fragment=metadata");
1356 #endif
1357         if (btrfs_test_opt(info, REF_VERIFY))
1358                 seq_puts(seq, ",ref_verify");
1359         seq_printf(seq, ",subvolid=%llu",
1360                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1361         seq_puts(seq, ",subvol=");
1362         seq_dentry(seq, dentry, " \t\n\\");
1363         return 0;
1364 }
1365
1366 static int btrfs_test_super(struct super_block *s, void *data)
1367 {
1368         struct btrfs_fs_info *p = data;
1369         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1370
1371         return fs_info->fs_devices == p->fs_devices;
1372 }
1373
1374 static int btrfs_set_super(struct super_block *s, void *data)
1375 {
1376         int err = set_anon_super(s, data);
1377         if (!err)
1378                 s->s_fs_info = data;
1379         return err;
1380 }
1381
1382 /*
1383  * subvolumes are identified by ino 256
1384  */
1385 static inline int is_subvolume_inode(struct inode *inode)
1386 {
1387         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1388                 return 1;
1389         return 0;
1390 }
1391
1392 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1393                                    const char *device_name, struct vfsmount *mnt)
1394 {
1395         struct dentry *root;
1396         int ret;
1397
1398         if (!subvol_name) {
1399                 if (!subvol_objectid) {
1400                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1401                                                           &subvol_objectid);
1402                         if (ret) {
1403                                 root = ERR_PTR(ret);
1404                                 goto out;
1405                         }
1406                 }
1407                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1408                                                             subvol_objectid);
1409                 if (IS_ERR(subvol_name)) {
1410                         root = ERR_CAST(subvol_name);
1411                         subvol_name = NULL;
1412                         goto out;
1413                 }
1414
1415         }
1416
1417         root = mount_subtree(mnt, subvol_name);
1418         /* mount_subtree() drops our reference on the vfsmount. */
1419         mnt = NULL;
1420
1421         if (!IS_ERR(root)) {
1422                 struct super_block *s = root->d_sb;
1423                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1424                 struct inode *root_inode = d_inode(root);
1425                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1426
1427                 ret = 0;
1428                 if (!is_subvolume_inode(root_inode)) {
1429                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1430                                subvol_name);
1431                         ret = -EINVAL;
1432                 }
1433                 if (subvol_objectid && root_objectid != subvol_objectid) {
1434                         /*
1435                          * This will also catch a race condition where a
1436                          * subvolume which was passed by ID is renamed and
1437                          * another subvolume is renamed over the old location.
1438                          */
1439                         btrfs_err(fs_info,
1440                                   "subvol '%s' does not match subvolid %llu",
1441                                   subvol_name, subvol_objectid);
1442                         ret = -EINVAL;
1443                 }
1444                 if (ret) {
1445                         dput(root);
1446                         root = ERR_PTR(ret);
1447                         deactivate_locked_super(s);
1448                 }
1449         }
1450
1451 out:
1452         mntput(mnt);
1453         kfree(subvol_name);
1454         return root;
1455 }
1456
1457 static int parse_security_options(char *orig_opts,
1458                                   struct security_mnt_opts *sec_opts)
1459 {
1460         char *secdata = NULL;
1461         int ret = 0;
1462
1463         secdata = alloc_secdata();
1464         if (!secdata)
1465                 return -ENOMEM;
1466         ret = security_sb_copy_data(orig_opts, secdata);
1467         if (ret) {
1468                 free_secdata(secdata);
1469                 return ret;
1470         }
1471         ret = security_sb_parse_opts_str(secdata, sec_opts);
1472         free_secdata(secdata);
1473         return ret;
1474 }
1475
1476 static int setup_security_options(struct btrfs_fs_info *fs_info,
1477                                   struct super_block *sb,
1478                                   struct security_mnt_opts *sec_opts)
1479 {
1480         int ret = 0;
1481
1482         /*
1483          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1484          * is valid.
1485          */
1486         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1487         if (ret)
1488                 return ret;
1489
1490 #ifdef CONFIG_SECURITY
1491         if (!fs_info->security_opts.num_mnt_opts) {
1492                 /* first time security setup, copy sec_opts to fs_info */
1493                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1494         } else {
1495                 /*
1496                  * Since SELinux (the only one supporting security_mnt_opts)
1497                  * does NOT support changing context during remount/mount of
1498                  * the same sb, this must be the same or part of the same
1499                  * security options, just free it.
1500                  */
1501                 security_free_mnt_opts(sec_opts);
1502         }
1503 #endif
1504         return ret;
1505 }
1506
1507 /*
1508  * Find a superblock for the given device / mount point.
1509  *
1510  * Note: This is based on mount_bdev from fs/super.c with a few additions
1511  *       for multiple device setup.  Make sure to keep it in sync.
1512  */
1513 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1514                 int flags, const char *device_name, void *data)
1515 {
1516         struct block_device *bdev = NULL;
1517         struct super_block *s;
1518         struct btrfs_fs_devices *fs_devices = NULL;
1519         struct btrfs_fs_info *fs_info = NULL;
1520         struct security_mnt_opts new_sec_opts;
1521         fmode_t mode = FMODE_READ;
1522         int error = 0;
1523
1524         if (!(flags & SB_RDONLY))
1525                 mode |= FMODE_WRITE;
1526
1527         error = btrfs_parse_early_options(data, mode, fs_type,
1528                                           &fs_devices);
1529         if (error) {
1530                 return ERR_PTR(error);
1531         }
1532
1533         security_init_mnt_opts(&new_sec_opts);
1534         if (data) {
1535                 error = parse_security_options(data, &new_sec_opts);
1536                 if (error)
1537                         return ERR_PTR(error);
1538         }
1539
1540         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1541         if (error)
1542                 goto error_sec_opts;
1543
1544         /*
1545          * Setup a dummy root and fs_info for test/set super.  This is because
1546          * we don't actually fill this stuff out until open_ctree, but we need
1547          * it for searching for existing supers, so this lets us do that and
1548          * then open_ctree will properly initialize everything later.
1549          */
1550         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1551         if (!fs_info) {
1552                 error = -ENOMEM;
1553                 goto error_sec_opts;
1554         }
1555
1556         fs_info->fs_devices = fs_devices;
1557
1558         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1559         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1560         security_init_mnt_opts(&fs_info->security_opts);
1561         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1562                 error = -ENOMEM;
1563                 goto error_fs_info;
1564         }
1565
1566         error = btrfs_open_devices(fs_devices, mode, fs_type);
1567         if (error)
1568                 goto error_fs_info;
1569
1570         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1571                 error = -EACCES;
1572                 goto error_close_devices;
1573         }
1574
1575         bdev = fs_devices->latest_bdev;
1576         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1577                  fs_info);
1578         if (IS_ERR(s)) {
1579                 error = PTR_ERR(s);
1580                 goto error_close_devices;
1581         }
1582
1583         if (s->s_root) {
1584                 btrfs_close_devices(fs_devices);
1585                 free_fs_info(fs_info);
1586                 if ((flags ^ s->s_flags) & SB_RDONLY)
1587                         error = -EBUSY;
1588         } else {
1589                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1590                 btrfs_sb(s)->bdev_holder = fs_type;
1591                 error = btrfs_fill_super(s, fs_devices, data);
1592         }
1593         if (error) {
1594                 deactivate_locked_super(s);
1595                 goto error_sec_opts;
1596         }
1597
1598         fs_info = btrfs_sb(s);
1599         error = setup_security_options(fs_info, s, &new_sec_opts);
1600         if (error) {
1601                 deactivate_locked_super(s);
1602                 goto error_sec_opts;
1603         }
1604
1605         return dget(s->s_root);
1606
1607 error_close_devices:
1608         btrfs_close_devices(fs_devices);
1609 error_fs_info:
1610         free_fs_info(fs_info);
1611 error_sec_opts:
1612         security_free_mnt_opts(&new_sec_opts);
1613         return ERR_PTR(error);
1614 }
1615
1616 /*
1617  * Mount function which is called by VFS layer.
1618  *
1619  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1620  * which needs vfsmount* of device's root (/).  This means device's root has to
1621  * be mounted internally in any case.
1622  *
1623  * Operation flow:
1624  *   1. Parse subvol id related options for later use in mount_subvol().
1625  *
1626  *   2. Mount device's root (/) by calling vfs_kern_mount().
1627  *
1628  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1629  *      first place. In order to avoid calling btrfs_mount() again, we use
1630  *      different file_system_type which is not registered to VFS by
1631  *      register_filesystem() (btrfs_root_fs_type). As a result,
1632  *      btrfs_mount_root() is called. The return value will be used by
1633  *      mount_subtree() in mount_subvol().
1634  *
1635  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1636  *      "btrfs subvolume set-default", mount_subvol() is called always.
1637  */
1638 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1639                 const char *device_name, void *data)
1640 {
1641         struct vfsmount *mnt_root;
1642         struct dentry *root;
1643         fmode_t mode = FMODE_READ;
1644         char *subvol_name = NULL;
1645         u64 subvol_objectid = 0;
1646         int error = 0;
1647
1648         if (!(flags & SB_RDONLY))
1649                 mode |= FMODE_WRITE;
1650
1651         error = btrfs_parse_subvol_options(data, &subvol_name,
1652                                         &subvol_objectid);
1653         if (error) {
1654                 kfree(subvol_name);
1655                 return ERR_PTR(error);
1656         }
1657
1658         /* mount device's root (/) */
1659         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1660         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1661                 if (flags & SB_RDONLY) {
1662                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1663                                 flags & ~SB_RDONLY, device_name, data);
1664                 } else {
1665                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1666                                 flags | SB_RDONLY, device_name, data);
1667                         if (IS_ERR(mnt_root)) {
1668                                 root = ERR_CAST(mnt_root);
1669                                 goto out;
1670                         }
1671
1672                         down_write(&mnt_root->mnt_sb->s_umount);
1673                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1674                         up_write(&mnt_root->mnt_sb->s_umount);
1675                         if (error < 0) {
1676                                 root = ERR_PTR(error);
1677                                 mntput(mnt_root);
1678                                 goto out;
1679                         }
1680                 }
1681         }
1682         if (IS_ERR(mnt_root)) {
1683                 root = ERR_CAST(mnt_root);
1684                 goto out;
1685         }
1686
1687         /* mount_subvol() will free subvol_name and mnt_root */
1688         root = mount_subvol(subvol_name, subvol_objectid, device_name, mnt_root);
1689
1690 out:
1691         return root;
1692 }
1693
1694 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1695                                      u32 new_pool_size, u32 old_pool_size)
1696 {
1697         if (new_pool_size == old_pool_size)
1698                 return;
1699
1700         fs_info->thread_pool_size = new_pool_size;
1701
1702         btrfs_info(fs_info, "resize thread pool %d -> %d",
1703                old_pool_size, new_pool_size);
1704
1705         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1706         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1707         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1708         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1709         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1710         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1711         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1712                                 new_pool_size);
1713         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1714         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1715         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1716         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1717         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1718                                 new_pool_size);
1719 }
1720
1721 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1722 {
1723         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1724 }
1725
1726 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1727                                        unsigned long old_opts, int flags)
1728 {
1729         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1730             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1731              (flags & SB_RDONLY))) {
1732                 /* wait for any defraggers to finish */
1733                 wait_event(fs_info->transaction_wait,
1734                            (atomic_read(&fs_info->defrag_running) == 0));
1735                 if (flags & SB_RDONLY)
1736                         sync_filesystem(fs_info->sb);
1737         }
1738 }
1739
1740 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1741                                          unsigned long old_opts)
1742 {
1743         /*
1744          * We need to cleanup all defragable inodes if the autodefragment is
1745          * close or the filesystem is read only.
1746          */
1747         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1748             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1749                 btrfs_cleanup_defrag_inodes(fs_info);
1750         }
1751
1752         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1753 }
1754
1755 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1756 {
1757         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1758         struct btrfs_root *root = fs_info->tree_root;
1759         unsigned old_flags = sb->s_flags;
1760         unsigned long old_opts = fs_info->mount_opt;
1761         unsigned long old_compress_type = fs_info->compress_type;
1762         u64 old_max_inline = fs_info->max_inline;
1763         u32 old_thread_pool_size = fs_info->thread_pool_size;
1764         u32 old_metadata_ratio = fs_info->metadata_ratio;
1765         int ret;
1766
1767         sync_filesystem(sb);
1768         btrfs_remount_prepare(fs_info);
1769
1770         if (data) {
1771                 struct security_mnt_opts new_sec_opts;
1772
1773                 security_init_mnt_opts(&new_sec_opts);
1774                 ret = parse_security_options(data, &new_sec_opts);
1775                 if (ret)
1776                         goto restore;
1777                 ret = setup_security_options(fs_info, sb,
1778                                              &new_sec_opts);
1779                 if (ret) {
1780                         security_free_mnt_opts(&new_sec_opts);
1781                         goto restore;
1782                 }
1783         }
1784
1785         ret = btrfs_parse_options(fs_info, data, *flags);
1786         if (ret)
1787                 goto restore;
1788
1789         btrfs_remount_begin(fs_info, old_opts, *flags);
1790         btrfs_resize_thread_pool(fs_info,
1791                 fs_info->thread_pool_size, old_thread_pool_size);
1792
1793         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1794                 goto out;
1795
1796         if (*flags & SB_RDONLY) {
1797                 /*
1798                  * this also happens on 'umount -rf' or on shutdown, when
1799                  * the filesystem is busy.
1800                  */
1801                 cancel_work_sync(&fs_info->async_reclaim_work);
1802
1803                 /* wait for the uuid_scan task to finish */
1804                 down(&fs_info->uuid_tree_rescan_sem);
1805                 /* avoid complains from lockdep et al. */
1806                 up(&fs_info->uuid_tree_rescan_sem);
1807
1808                 sb->s_flags |= SB_RDONLY;
1809
1810                 /*
1811                  * Setting SB_RDONLY will put the cleaner thread to
1812                  * sleep at the next loop if it's already active.
1813                  * If it's already asleep, we'll leave unused block
1814                  * groups on disk until we're mounted read-write again
1815                  * unless we clean them up here.
1816                  */
1817                 btrfs_delete_unused_bgs(fs_info);
1818
1819                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1820                 btrfs_scrub_cancel(fs_info);
1821                 btrfs_pause_balance(fs_info);
1822
1823                 ret = btrfs_commit_super(fs_info);
1824                 if (ret)
1825                         goto restore;
1826         } else {
1827                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1828                         btrfs_err(fs_info,
1829                                 "Remounting read-write after error is not allowed");
1830                         ret = -EINVAL;
1831                         goto restore;
1832                 }
1833                 if (fs_info->fs_devices->rw_devices == 0) {
1834                         ret = -EACCES;
1835                         goto restore;
1836                 }
1837
1838                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1839                         btrfs_warn(fs_info,
1840                                 "too many missing devices, writeable remount is not allowed");
1841                         ret = -EACCES;
1842                         goto restore;
1843                 }
1844
1845                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1846                         ret = -EINVAL;
1847                         goto restore;
1848                 }
1849
1850                 ret = btrfs_cleanup_fs_roots(fs_info);
1851                 if (ret)
1852                         goto restore;
1853
1854                 /* recover relocation */
1855                 mutex_lock(&fs_info->cleaner_mutex);
1856                 ret = btrfs_recover_relocation(root);
1857                 mutex_unlock(&fs_info->cleaner_mutex);
1858                 if (ret)
1859                         goto restore;
1860
1861                 ret = btrfs_resume_balance_async(fs_info);
1862                 if (ret)
1863                         goto restore;
1864
1865                 ret = btrfs_resume_dev_replace_async(fs_info);
1866                 if (ret) {
1867                         btrfs_warn(fs_info, "failed to resume dev_replace");
1868                         goto restore;
1869                 }
1870
1871                 btrfs_qgroup_rescan_resume(fs_info);
1872
1873                 if (!fs_info->uuid_root) {
1874                         btrfs_info(fs_info, "creating UUID tree");
1875                         ret = btrfs_create_uuid_tree(fs_info);
1876                         if (ret) {
1877                                 btrfs_warn(fs_info,
1878                                            "failed to create the UUID tree %d",
1879                                            ret);
1880                                 goto restore;
1881                         }
1882                 }
1883                 sb->s_flags &= ~SB_RDONLY;
1884
1885                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1886         }
1887 out:
1888         wake_up_process(fs_info->transaction_kthread);
1889         btrfs_remount_cleanup(fs_info, old_opts);
1890         return 0;
1891
1892 restore:
1893         /* We've hit an error - don't reset SB_RDONLY */
1894         if (sb_rdonly(sb))
1895                 old_flags |= SB_RDONLY;
1896         sb->s_flags = old_flags;
1897         fs_info->mount_opt = old_opts;
1898         fs_info->compress_type = old_compress_type;
1899         fs_info->max_inline = old_max_inline;
1900         btrfs_resize_thread_pool(fs_info,
1901                 old_thread_pool_size, fs_info->thread_pool_size);
1902         fs_info->metadata_ratio = old_metadata_ratio;
1903         btrfs_remount_cleanup(fs_info, old_opts);
1904         return ret;
1905 }
1906
1907 /* Used to sort the devices by max_avail(descending sort) */
1908 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1909                                        const void *dev_info2)
1910 {
1911         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1912             ((struct btrfs_device_info *)dev_info2)->max_avail)
1913                 return -1;
1914         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1915                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1916                 return 1;
1917         else
1918         return 0;
1919 }
1920
1921 /*
1922  * sort the devices by max_avail, in which max free extent size of each device
1923  * is stored.(Descending Sort)
1924  */
1925 static inline void btrfs_descending_sort_devices(
1926                                         struct btrfs_device_info *devices,
1927                                         size_t nr_devices)
1928 {
1929         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1930              btrfs_cmp_device_free_bytes, NULL);
1931 }
1932
1933 /*
1934  * The helper to calc the free space on the devices that can be used to store
1935  * file data.
1936  */
1937 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1938                                        u64 *free_bytes)
1939 {
1940         struct btrfs_device_info *devices_info;
1941         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1942         struct btrfs_device *device;
1943         u64 skip_space;
1944         u64 type;
1945         u64 avail_space;
1946         u64 min_stripe_size;
1947         int min_stripes = 1, num_stripes = 1;
1948         int i = 0, nr_devices;
1949
1950         /*
1951          * We aren't under the device list lock, so this is racy-ish, but good
1952          * enough for our purposes.
1953          */
1954         nr_devices = fs_info->fs_devices->open_devices;
1955         if (!nr_devices) {
1956                 smp_mb();
1957                 nr_devices = fs_info->fs_devices->open_devices;
1958                 ASSERT(nr_devices);
1959                 if (!nr_devices) {
1960                         *free_bytes = 0;
1961                         return 0;
1962                 }
1963         }
1964
1965         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1966                                GFP_KERNEL);
1967         if (!devices_info)
1968                 return -ENOMEM;
1969
1970         /* calc min stripe number for data space allocation */
1971         type = btrfs_data_alloc_profile(fs_info);
1972         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1973                 min_stripes = 2;
1974                 num_stripes = nr_devices;
1975         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1976                 min_stripes = 2;
1977                 num_stripes = 2;
1978         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1979                 min_stripes = 4;
1980                 num_stripes = 4;
1981         }
1982
1983         if (type & BTRFS_BLOCK_GROUP_DUP)
1984                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1985         else
1986                 min_stripe_size = BTRFS_STRIPE_LEN;
1987
1988         rcu_read_lock();
1989         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1990                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1991                                                 &device->dev_state) ||
1992                     !device->bdev ||
1993                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1994                         continue;
1995
1996                 if (i >= nr_devices)
1997                         break;
1998
1999                 avail_space = device->total_bytes - device->bytes_used;
2000
2001                 /* align with stripe_len */
2002                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
2003                 avail_space *= BTRFS_STRIPE_LEN;
2004
2005                 /*
2006                  * In order to avoid overwriting the superblock on the drive,
2007                  * btrfs starts at an offset of at least 1MB when doing chunk
2008                  * allocation.
2009                  */
2010                 skip_space = SZ_1M;
2011
2012                 /*
2013                  * we can use the free space in [0, skip_space - 1], subtract
2014                  * it from the total.
2015                  */
2016                 if (avail_space && avail_space >= skip_space)
2017                         avail_space -= skip_space;
2018                 else
2019                         avail_space = 0;
2020
2021                 if (avail_space < min_stripe_size)
2022                         continue;
2023
2024                 devices_info[i].dev = device;
2025                 devices_info[i].max_avail = avail_space;
2026
2027                 i++;
2028         }
2029         rcu_read_unlock();
2030
2031         nr_devices = i;
2032
2033         btrfs_descending_sort_devices(devices_info, nr_devices);
2034
2035         i = nr_devices - 1;
2036         avail_space = 0;
2037         while (nr_devices >= min_stripes) {
2038                 if (num_stripes > nr_devices)
2039                         num_stripes = nr_devices;
2040
2041                 if (devices_info[i].max_avail >= min_stripe_size) {
2042                         int j;
2043                         u64 alloc_size;
2044
2045                         avail_space += devices_info[i].max_avail * num_stripes;
2046                         alloc_size = devices_info[i].max_avail;
2047                         for (j = i + 1 - num_stripes; j <= i; j++)
2048                                 devices_info[j].max_avail -= alloc_size;
2049                 }
2050                 i--;
2051                 nr_devices--;
2052         }
2053
2054         kfree(devices_info);
2055         *free_bytes = avail_space;
2056         return 0;
2057 }
2058
2059 /*
2060  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2061  *
2062  * If there's a redundant raid level at DATA block groups, use the respective
2063  * multiplier to scale the sizes.
2064  *
2065  * Unused device space usage is based on simulating the chunk allocator
2066  * algorithm that respects the device sizes and order of allocations.  This is
2067  * a close approximation of the actual use but there are other factors that may
2068  * change the result (like a new metadata chunk).
2069  *
2070  * If metadata is exhausted, f_bavail will be 0.
2071  */
2072 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2073 {
2074         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2075         struct btrfs_super_block *disk_super = fs_info->super_copy;
2076         struct list_head *head = &fs_info->space_info;
2077         struct btrfs_space_info *found;
2078         u64 total_used = 0;
2079         u64 total_free_data = 0;
2080         u64 total_free_meta = 0;
2081         int bits = dentry->d_sb->s_blocksize_bits;
2082         __be32 *fsid = (__be32 *)fs_info->fsid;
2083         unsigned factor = 1;
2084         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2085         int ret;
2086         u64 thresh = 0;
2087         int mixed = 0;
2088
2089         rcu_read_lock();
2090         list_for_each_entry_rcu(found, head, list) {
2091                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2092                         int i;
2093
2094                         total_free_data += found->disk_total - found->disk_used;
2095                         total_free_data -=
2096                                 btrfs_account_ro_block_groups_free_space(found);
2097
2098                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2099                                 if (!list_empty(&found->block_groups[i])) {
2100                                         switch (i) {
2101                                         case BTRFS_RAID_DUP:
2102                                         case BTRFS_RAID_RAID1:
2103                                         case BTRFS_RAID_RAID10:
2104                                                 factor = 2;
2105                                         }
2106                                 }
2107                         }
2108                 }
2109
2110                 /*
2111                  * Metadata in mixed block goup profiles are accounted in data
2112                  */
2113                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2114                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2115                                 mixed = 1;
2116                         else
2117                                 total_free_meta += found->disk_total -
2118                                         found->disk_used;
2119                 }
2120
2121                 total_used += found->disk_used;
2122         }
2123
2124         rcu_read_unlock();
2125
2126         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2127         buf->f_blocks >>= bits;
2128         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2129
2130         /* Account global block reserve as used, it's in logical size already */
2131         spin_lock(&block_rsv->lock);
2132         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2133         if (buf->f_bfree >= block_rsv->size >> bits)
2134                 buf->f_bfree -= block_rsv->size >> bits;
2135         else
2136                 buf->f_bfree = 0;
2137         spin_unlock(&block_rsv->lock);
2138
2139         buf->f_bavail = div_u64(total_free_data, factor);
2140         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2141         if (ret)
2142                 return ret;
2143         buf->f_bavail += div_u64(total_free_data, factor);
2144         buf->f_bavail = buf->f_bavail >> bits;
2145
2146         /*
2147          * We calculate the remaining metadata space minus global reserve. If
2148          * this is (supposedly) smaller than zero, there's no space. But this
2149          * does not hold in practice, the exhausted state happens where's still
2150          * some positive delta. So we apply some guesswork and compare the
2151          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2152          *
2153          * We probably cannot calculate the exact threshold value because this
2154          * depends on the internal reservations requested by various
2155          * operations, so some operations that consume a few metadata will
2156          * succeed even if the Avail is zero. But this is better than the other
2157          * way around.
2158          */
2159         thresh = SZ_4M;
2160
2161         if (!mixed && total_free_meta - thresh < block_rsv->size)
2162                 buf->f_bavail = 0;
2163
2164         buf->f_type = BTRFS_SUPER_MAGIC;
2165         buf->f_bsize = dentry->d_sb->s_blocksize;
2166         buf->f_namelen = BTRFS_NAME_LEN;
2167
2168         /* We treat it as constant endianness (it doesn't matter _which_)
2169            because we want the fsid to come out the same whether mounted
2170            on a big-endian or little-endian host */
2171         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2172         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2173         /* Mask in the root object ID too, to disambiguate subvols */
2174         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2175         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2176
2177         return 0;
2178 }
2179
2180 static void btrfs_kill_super(struct super_block *sb)
2181 {
2182         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2183         kill_anon_super(sb);
2184         free_fs_info(fs_info);
2185 }
2186
2187 static struct file_system_type btrfs_fs_type = {
2188         .owner          = THIS_MODULE,
2189         .name           = "btrfs",
2190         .mount          = btrfs_mount,
2191         .kill_sb        = btrfs_kill_super,
2192         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2193 };
2194
2195 static struct file_system_type btrfs_root_fs_type = {
2196         .owner          = THIS_MODULE,
2197         .name           = "btrfs",
2198         .mount          = btrfs_mount_root,
2199         .kill_sb        = btrfs_kill_super,
2200         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2201 };
2202
2203 MODULE_ALIAS_FS("btrfs");
2204
2205 static int btrfs_control_open(struct inode *inode, struct file *file)
2206 {
2207         /*
2208          * The control file's private_data is used to hold the
2209          * transaction when it is started and is used to keep
2210          * track of whether a transaction is already in progress.
2211          */
2212         file->private_data = NULL;
2213         return 0;
2214 }
2215
2216 /*
2217  * used by btrfsctl to scan devices when no FS is mounted
2218  */
2219 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2220                                 unsigned long arg)
2221 {
2222         struct btrfs_ioctl_vol_args *vol;
2223         struct btrfs_fs_devices *fs_devices;
2224         int ret = -ENOTTY;
2225
2226         if (!capable(CAP_SYS_ADMIN))
2227                 return -EPERM;
2228
2229         vol = memdup_user((void __user *)arg, sizeof(*vol));
2230         if (IS_ERR(vol))
2231                 return PTR_ERR(vol);
2232
2233         switch (cmd) {
2234         case BTRFS_IOC_SCAN_DEV:
2235                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2236                                             &btrfs_root_fs_type, &fs_devices);
2237                 break;
2238         case BTRFS_IOC_DEVICES_READY:
2239                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2240                                             &btrfs_root_fs_type, &fs_devices);
2241                 if (ret)
2242                         break;
2243                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2244                 break;
2245         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2246                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2247                 break;
2248         }
2249
2250         kfree(vol);
2251         return ret;
2252 }
2253
2254 static int btrfs_freeze(struct super_block *sb)
2255 {
2256         struct btrfs_trans_handle *trans;
2257         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2258         struct btrfs_root *root = fs_info->tree_root;
2259
2260         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2261         /*
2262          * We don't need a barrier here, we'll wait for any transaction that
2263          * could be in progress on other threads (and do delayed iputs that
2264          * we want to avoid on a frozen filesystem), or do the commit
2265          * ourselves.
2266          */
2267         trans = btrfs_attach_transaction_barrier(root);
2268         if (IS_ERR(trans)) {
2269                 /* no transaction, don't bother */
2270                 if (PTR_ERR(trans) == -ENOENT)
2271                         return 0;
2272                 return PTR_ERR(trans);
2273         }
2274         return btrfs_commit_transaction(trans);
2275 }
2276
2277 static int btrfs_unfreeze(struct super_block *sb)
2278 {
2279         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2280
2281         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2282         return 0;
2283 }
2284
2285 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2286 {
2287         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2288         struct btrfs_fs_devices *cur_devices;
2289         struct btrfs_device *dev, *first_dev = NULL;
2290         struct list_head *head;
2291         struct rcu_string *name;
2292
2293         /*
2294          * Lightweight locking of the devices. We should not need
2295          * device_list_mutex here as we only read the device data and the list
2296          * is protected by RCU.  Even if a device is deleted during the list
2297          * traversals, we'll get valid data, the freeing callback will wait at
2298          * least until until the rcu_read_unlock.
2299          */
2300         rcu_read_lock();
2301         cur_devices = fs_info->fs_devices;
2302         while (cur_devices) {
2303                 head = &cur_devices->devices;
2304                 list_for_each_entry_rcu(dev, head, dev_list) {
2305                         if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2306                                 continue;
2307                         if (!dev->name)
2308                                 continue;
2309                         if (!first_dev || dev->devid < first_dev->devid)
2310                                 first_dev = dev;
2311                 }
2312                 cur_devices = cur_devices->seed;
2313         }
2314
2315         if (first_dev) {
2316                 name = rcu_dereference(first_dev->name);
2317                 seq_escape(m, name->str, " \t\n\\");
2318         } else {
2319                 WARN_ON(1);
2320         }
2321         rcu_read_unlock();
2322         return 0;
2323 }
2324
2325 static const struct super_operations btrfs_super_ops = {
2326         .drop_inode     = btrfs_drop_inode,
2327         .evict_inode    = btrfs_evict_inode,
2328         .put_super      = btrfs_put_super,
2329         .sync_fs        = btrfs_sync_fs,
2330         .show_options   = btrfs_show_options,
2331         .show_devname   = btrfs_show_devname,
2332         .write_inode    = btrfs_write_inode,
2333         .alloc_inode    = btrfs_alloc_inode,
2334         .destroy_inode  = btrfs_destroy_inode,
2335         .statfs         = btrfs_statfs,
2336         .remount_fs     = btrfs_remount,
2337         .freeze_fs      = btrfs_freeze,
2338         .unfreeze_fs    = btrfs_unfreeze,
2339 };
2340
2341 static const struct file_operations btrfs_ctl_fops = {
2342         .open = btrfs_control_open,
2343         .unlocked_ioctl  = btrfs_control_ioctl,
2344         .compat_ioctl = btrfs_control_ioctl,
2345         .owner   = THIS_MODULE,
2346         .llseek = noop_llseek,
2347 };
2348
2349 static struct miscdevice btrfs_misc = {
2350         .minor          = BTRFS_MINOR,
2351         .name           = "btrfs-control",
2352         .fops           = &btrfs_ctl_fops
2353 };
2354
2355 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2356 MODULE_ALIAS("devname:btrfs-control");
2357
2358 static int __init btrfs_interface_init(void)
2359 {
2360         return misc_register(&btrfs_misc);
2361 }
2362
2363 static __cold void btrfs_interface_exit(void)
2364 {
2365         misc_deregister(&btrfs_misc);
2366 }
2367
2368 static void __init btrfs_print_mod_info(void)
2369 {
2370         static const char options[] = ""
2371 #ifdef CONFIG_BTRFS_DEBUG
2372                         ", debug=on"
2373 #endif
2374 #ifdef CONFIG_BTRFS_ASSERT
2375                         ", assert=on"
2376 #endif
2377 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2378                         ", integrity-checker=on"
2379 #endif
2380 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2381                         ", ref-verify=on"
2382 #endif
2383                         ;
2384         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2385 }
2386
2387 static int __init init_btrfs_fs(void)
2388 {
2389         int err;
2390
2391         btrfs_props_init();
2392
2393         err = btrfs_init_sysfs();
2394         if (err)
2395                 return err;
2396
2397         btrfs_init_compress();
2398
2399         err = btrfs_init_cachep();
2400         if (err)
2401                 goto free_compress;
2402
2403         err = extent_io_init();
2404         if (err)
2405                 goto free_cachep;
2406
2407         err = extent_map_init();
2408         if (err)
2409                 goto free_extent_io;
2410
2411         err = ordered_data_init();
2412         if (err)
2413                 goto free_extent_map;
2414
2415         err = btrfs_delayed_inode_init();
2416         if (err)
2417                 goto free_ordered_data;
2418
2419         err = btrfs_auto_defrag_init();
2420         if (err)
2421                 goto free_delayed_inode;
2422
2423         err = btrfs_delayed_ref_init();
2424         if (err)
2425                 goto free_auto_defrag;
2426
2427         err = btrfs_prelim_ref_init();
2428         if (err)
2429                 goto free_delayed_ref;
2430
2431         err = btrfs_end_io_wq_init();
2432         if (err)
2433                 goto free_prelim_ref;
2434
2435         err = btrfs_interface_init();
2436         if (err)
2437                 goto free_end_io_wq;
2438
2439         btrfs_init_lockdep();
2440
2441         btrfs_print_mod_info();
2442
2443         err = btrfs_run_sanity_tests();
2444         if (err)
2445                 goto unregister_ioctl;
2446
2447         err = register_filesystem(&btrfs_fs_type);
2448         if (err)
2449                 goto unregister_ioctl;
2450
2451         return 0;
2452
2453 unregister_ioctl:
2454         btrfs_interface_exit();
2455 free_end_io_wq:
2456         btrfs_end_io_wq_exit();
2457 free_prelim_ref:
2458         btrfs_prelim_ref_exit();
2459 free_delayed_ref:
2460         btrfs_delayed_ref_exit();
2461 free_auto_defrag:
2462         btrfs_auto_defrag_exit();
2463 free_delayed_inode:
2464         btrfs_delayed_inode_exit();
2465 free_ordered_data:
2466         ordered_data_exit();
2467 free_extent_map:
2468         extent_map_exit();
2469 free_extent_io:
2470         extent_io_exit();
2471 free_cachep:
2472         btrfs_destroy_cachep();
2473 free_compress:
2474         btrfs_exit_compress();
2475         btrfs_exit_sysfs();
2476
2477         return err;
2478 }
2479
2480 static void __exit exit_btrfs_fs(void)
2481 {
2482         btrfs_destroy_cachep();
2483         btrfs_delayed_ref_exit();
2484         btrfs_auto_defrag_exit();
2485         btrfs_delayed_inode_exit();
2486         btrfs_prelim_ref_exit();
2487         ordered_data_exit();
2488         extent_map_exit();
2489         extent_io_exit();
2490         btrfs_interface_exit();
2491         btrfs_end_io_wq_exit();
2492         unregister_filesystem(&btrfs_fs_type);
2493         btrfs_exit_sysfs();
2494         btrfs_cleanup_fs_uuids();
2495         btrfs_exit_compress();
2496 }
2497
2498 late_initcall(init_btrfs_fs);
2499 module_exit(exit_btrfs_fs)
2500
2501 MODULE_LICENSE("GPL");