540c18511e7a8663065832f9b37574e173207918
[sfrench/cifs-2.6.git] / fs / btrfs / super.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/buffer_head.h>
22 #include <linux/fs.h>
23 #include <linux/pagemap.h>
24 #include <linux/highmem.h>
25 #include <linux/time.h>
26 #include <linux/init.h>
27 #include <linux/seq_file.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mount.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/parser.h>
37 #include <linux/ctype.h>
38 #include <linux/namei.h>
39 #include <linux/miscdevice.h>
40 #include <linux/magic.h>
41 #include <linux/slab.h>
42 #include <linux/cleancache.h>
43 #include <linux/ratelimit.h>
44 #include <linux/btrfs.h>
45 #include "delayed-inode.h"
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "hash.h"
52 #include "props.h"
53 #include "xattr.h"
54 #include "volumes.h"
55 #include "export.h"
56 #include "compression.h"
57 #include "rcu-string.h"
58 #include "dev-replace.h"
59 #include "free-space-cache.h"
60 #include "backref.h"
61 #include "tests/btrfs-tests.h"
62
63 #include "qgroup.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/btrfs.h>
66
67 static const struct super_operations btrfs_super_ops;
68
69 /*
70  * Types for mounting the default subvolume and a subvolume explicitly
71  * requested by subvol=/path. That way the callchain is straightforward and we
72  * don't have to play tricks with the mount options and recursive calls to
73  * btrfs_mount.
74  *
75  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
76  */
77 static struct file_system_type btrfs_fs_type;
78 static struct file_system_type btrfs_root_fs_type;
79
80 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
81
82 const char *btrfs_decode_error(int errno)
83 {
84         char *errstr = "unknown";
85
86         switch (errno) {
87         case -EIO:
88                 errstr = "IO failure";
89                 break;
90         case -ENOMEM:
91                 errstr = "Out of memory";
92                 break;
93         case -EROFS:
94                 errstr = "Readonly filesystem";
95                 break;
96         case -EEXIST:
97                 errstr = "Object already exists";
98                 break;
99         case -ENOSPC:
100                 errstr = "No space left";
101                 break;
102         case -ENOENT:
103                 errstr = "No such entry";
104                 break;
105         }
106
107         return errstr;
108 }
109
110 /*
111  * __btrfs_handle_fs_error decodes expected errors from the caller and
112  * invokes the approciate error response.
113  */
114 __cold
115 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
116                        unsigned int line, int errno, const char *fmt, ...)
117 {
118         struct super_block *sb = fs_info->sb;
119 #ifdef CONFIG_PRINTK
120         const char *errstr;
121 #endif
122
123         /*
124          * Special case: if the error is EROFS, and we're already
125          * under SB_RDONLY, then it is safe here.
126          */
127         if (errno == -EROFS && sb_rdonly(sb))
128                 return;
129
130 #ifdef CONFIG_PRINTK
131         errstr = btrfs_decode_error(errno);
132         if (fmt) {
133                 struct va_format vaf;
134                 va_list args;
135
136                 va_start(args, fmt);
137                 vaf.fmt = fmt;
138                 vaf.va = &args;
139
140                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
141                         sb->s_id, function, line, errno, errstr, &vaf);
142                 va_end(args);
143         } else {
144                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
145                         sb->s_id, function, line, errno, errstr);
146         }
147 #endif
148
149         /*
150          * Today we only save the error info to memory.  Long term we'll
151          * also send it down to the disk
152          */
153         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
154
155         /* Don't go through full error handling during mount */
156         if (!(sb->s_flags & SB_BORN))
157                 return;
158
159         if (sb_rdonly(sb))
160                 return;
161
162         /* btrfs handle error by forcing the filesystem readonly */
163         sb->s_flags |= SB_RDONLY;
164         btrfs_info(fs_info, "forced readonly");
165         /*
166          * Note that a running device replace operation is not canceled here
167          * although there is no way to update the progress. It would add the
168          * risk of a deadlock, therefore the canceling is omitted. The only
169          * penalty is that some I/O remains active until the procedure
170          * completes. The next time when the filesystem is mounted writeable
171          * again, the device replace operation continues.
172          */
173 }
174
175 #ifdef CONFIG_PRINTK
176 static const char * const logtypes[] = {
177         "emergency",
178         "alert",
179         "critical",
180         "error",
181         "warning",
182         "notice",
183         "info",
184         "debug",
185 };
186
187
188 /*
189  * Use one ratelimit state per log level so that a flood of less important
190  * messages doesn't cause more important ones to be dropped.
191  */
192 static struct ratelimit_state printk_limits[] = {
193         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
194         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
195         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
196         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
197         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
198         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
199         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
200         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
201 };
202
203 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
204 {
205         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
206         struct va_format vaf;
207         va_list args;
208         int kern_level;
209         const char *type = logtypes[4];
210         struct ratelimit_state *ratelimit = &printk_limits[4];
211
212         va_start(args, fmt);
213
214         while ((kern_level = printk_get_level(fmt)) != 0) {
215                 size_t size = printk_skip_level(fmt) - fmt;
216
217                 if (kern_level >= '0' && kern_level <= '7') {
218                         memcpy(lvl, fmt,  size);
219                         lvl[size] = '\0';
220                         type = logtypes[kern_level - '0'];
221                         ratelimit = &printk_limits[kern_level - '0'];
222                 }
223                 fmt += size;
224         }
225
226         vaf.fmt = fmt;
227         vaf.va = &args;
228
229         if (__ratelimit(ratelimit))
230                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
231                         fs_info ? fs_info->sb->s_id : "<unknown>", &vaf);
232
233         va_end(args);
234 }
235 #endif
236
237 /*
238  * We only mark the transaction aborted and then set the file system read-only.
239  * This will prevent new transactions from starting or trying to join this
240  * one.
241  *
242  * This means that error recovery at the call site is limited to freeing
243  * any local memory allocations and passing the error code up without
244  * further cleanup. The transaction should complete as it normally would
245  * in the call path but will return -EIO.
246  *
247  * We'll complete the cleanup in btrfs_end_transaction and
248  * btrfs_commit_transaction.
249  */
250 __cold
251 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
252                                const char *function,
253                                unsigned int line, int errno)
254 {
255         struct btrfs_fs_info *fs_info = trans->fs_info;
256
257         trans->aborted = errno;
258         /* Nothing used. The other threads that have joined this
259          * transaction may be able to continue. */
260         if (!trans->dirty && list_empty(&trans->new_bgs)) {
261                 const char *errstr;
262
263                 errstr = btrfs_decode_error(errno);
264                 btrfs_warn(fs_info,
265                            "%s:%d: Aborting unused transaction(%s).",
266                            function, line, errstr);
267                 return;
268         }
269         WRITE_ONCE(trans->transaction->aborted, errno);
270         /* Wake up anybody who may be waiting on this transaction */
271         wake_up(&fs_info->transaction_wait);
272         wake_up(&fs_info->transaction_blocked_wait);
273         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
274 }
275 /*
276  * __btrfs_panic decodes unexpected, fatal errors from the caller,
277  * issues an alert, and either panics or BUGs, depending on mount options.
278  */
279 __cold
280 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
281                    unsigned int line, int errno, const char *fmt, ...)
282 {
283         char *s_id = "<unknown>";
284         const char *errstr;
285         struct va_format vaf = { .fmt = fmt };
286         va_list args;
287
288         if (fs_info)
289                 s_id = fs_info->sb->s_id;
290
291         va_start(args, fmt);
292         vaf.va = &args;
293
294         errstr = btrfs_decode_error(errno);
295         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
296                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
297                         s_id, function, line, &vaf, errno, errstr);
298
299         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
300                    function, line, &vaf, errno, errstr);
301         va_end(args);
302         /* Caller calls BUG() */
303 }
304
305 static void btrfs_put_super(struct super_block *sb)
306 {
307         close_ctree(btrfs_sb(sb));
308 }
309
310 enum {
311         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
312         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
313         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
314         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
315         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
316         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
317         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
318         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
319         Opt_skip_balance, Opt_check_integrity,
320         Opt_check_integrity_including_extent_data,
321         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
322         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
323         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
324         Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
325         Opt_nologreplay, Opt_norecovery,
326 #ifdef CONFIG_BTRFS_DEBUG
327         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
328 #endif
329 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
330         Opt_ref_verify,
331 #endif
332         Opt_err,
333 };
334
335 static const match_table_t tokens = {
336         {Opt_degraded, "degraded"},
337         {Opt_subvol, "subvol=%s"},
338         {Opt_subvolid, "subvolid=%s"},
339         {Opt_device, "device=%s"},
340         {Opt_nodatasum, "nodatasum"},
341         {Opt_datasum, "datasum"},
342         {Opt_nodatacow, "nodatacow"},
343         {Opt_datacow, "datacow"},
344         {Opt_nobarrier, "nobarrier"},
345         {Opt_barrier, "barrier"},
346         {Opt_max_inline, "max_inline=%s"},
347         {Opt_alloc_start, "alloc_start=%s"},            /* deprecated */
348         {Opt_thread_pool, "thread_pool=%u"},
349         {Opt_compress, "compress"},
350         {Opt_compress_type, "compress=%s"},
351         {Opt_compress_force, "compress-force"},
352         {Opt_compress_force_type, "compress-force=%s"},
353         {Opt_ssd, "ssd"},
354         {Opt_ssd_spread, "ssd_spread"},
355         {Opt_nossd, "nossd"},
356         {Opt_acl, "acl"},
357         {Opt_noacl, "noacl"},
358         {Opt_notreelog, "notreelog"},
359         {Opt_treelog, "treelog"},
360         {Opt_nologreplay, "nologreplay"},
361         {Opt_norecovery, "norecovery"},
362         {Opt_flushoncommit, "flushoncommit"},
363         {Opt_noflushoncommit, "noflushoncommit"},
364         {Opt_ratio, "metadata_ratio=%u"},
365         {Opt_discard, "discard"},
366         {Opt_nodiscard, "nodiscard"},
367         {Opt_space_cache, "space_cache"},
368         {Opt_space_cache_version, "space_cache=%s"},
369         {Opt_clear_cache, "clear_cache"},
370         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
371         {Opt_enospc_debug, "enospc_debug"},
372         {Opt_noenospc_debug, "noenospc_debug"},
373         {Opt_subvolrootid, "subvolrootid=%d"},          /* deprecated */
374         {Opt_defrag, "autodefrag"},
375         {Opt_nodefrag, "noautodefrag"},
376         {Opt_inode_cache, "inode_cache"},
377         {Opt_noinode_cache, "noinode_cache"},
378         {Opt_no_space_cache, "nospace_cache"},
379         {Opt_recovery, "recovery"}, /* deprecated */
380         {Opt_usebackuproot, "usebackuproot"},
381         {Opt_skip_balance, "skip_balance"},
382         {Opt_check_integrity, "check_int"},
383         {Opt_check_integrity_including_extent_data, "check_int_data"},
384         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
385         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
386         {Opt_fatal_errors, "fatal_errors=%s"},
387         {Opt_commit_interval, "commit=%u"},
388 #ifdef CONFIG_BTRFS_DEBUG
389         {Opt_fragment_data, "fragment=data"},
390         {Opt_fragment_metadata, "fragment=metadata"},
391         {Opt_fragment_all, "fragment=all"},
392 #endif
393 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
394         {Opt_ref_verify, "ref_verify"},
395 #endif
396         {Opt_err, NULL},
397 };
398
399 /*
400  * Regular mount options parser.  Everything that is needed only when
401  * reading in a new superblock is parsed here.
402  * XXX JDM: This needs to be cleaned up for remount.
403  */
404 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
405                         unsigned long new_flags)
406 {
407         substring_t args[MAX_OPT_ARGS];
408         char *p, *num;
409         u64 cache_gen;
410         int intarg;
411         int ret = 0;
412         char *compress_type;
413         bool compress_force = false;
414         enum btrfs_compression_type saved_compress_type;
415         bool saved_compress_force;
416         int no_compress = 0;
417
418         cache_gen = btrfs_super_cache_generation(info->super_copy);
419         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
420                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
421         else if (cache_gen)
422                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
423
424         /*
425          * Even the options are empty, we still need to do extra check
426          * against new flags
427          */
428         if (!options)
429                 goto check;
430
431         while ((p = strsep(&options, ",")) != NULL) {
432                 int token;
433                 if (!*p)
434                         continue;
435
436                 token = match_token(p, tokens, args);
437                 switch (token) {
438                 case Opt_degraded:
439                         btrfs_info(info, "allowing degraded mounts");
440                         btrfs_set_opt(info->mount_opt, DEGRADED);
441                         break;
442                 case Opt_subvol:
443                 case Opt_subvolid:
444                 case Opt_subvolrootid:
445                 case Opt_device:
446                         /*
447                          * These are parsed by btrfs_parse_subvol_options
448                          * and btrfs_parse_early_options
449                          * and can be happily ignored here.
450                          */
451                         break;
452                 case Opt_nodatasum:
453                         btrfs_set_and_info(info, NODATASUM,
454                                            "setting nodatasum");
455                         break;
456                 case Opt_datasum:
457                         if (btrfs_test_opt(info, NODATASUM)) {
458                                 if (btrfs_test_opt(info, NODATACOW))
459                                         btrfs_info(info,
460                                                    "setting datasum, datacow enabled");
461                                 else
462                                         btrfs_info(info, "setting datasum");
463                         }
464                         btrfs_clear_opt(info->mount_opt, NODATACOW);
465                         btrfs_clear_opt(info->mount_opt, NODATASUM);
466                         break;
467                 case Opt_nodatacow:
468                         if (!btrfs_test_opt(info, NODATACOW)) {
469                                 if (!btrfs_test_opt(info, COMPRESS) ||
470                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
471                                         btrfs_info(info,
472                                                    "setting nodatacow, compression disabled");
473                                 } else {
474                                         btrfs_info(info, "setting nodatacow");
475                                 }
476                         }
477                         btrfs_clear_opt(info->mount_opt, COMPRESS);
478                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
479                         btrfs_set_opt(info->mount_opt, NODATACOW);
480                         btrfs_set_opt(info->mount_opt, NODATASUM);
481                         break;
482                 case Opt_datacow:
483                         btrfs_clear_and_info(info, NODATACOW,
484                                              "setting datacow");
485                         break;
486                 case Opt_compress_force:
487                 case Opt_compress_force_type:
488                         compress_force = true;
489                         /* Fallthrough */
490                 case Opt_compress:
491                 case Opt_compress_type:
492                         saved_compress_type = btrfs_test_opt(info,
493                                                              COMPRESS) ?
494                                 info->compress_type : BTRFS_COMPRESS_NONE;
495                         saved_compress_force =
496                                 btrfs_test_opt(info, FORCE_COMPRESS);
497                         if (token == Opt_compress ||
498                             token == Opt_compress_force ||
499                             strncmp(args[0].from, "zlib", 4) == 0) {
500                                 compress_type = "zlib";
501
502                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
503                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
504                                 /*
505                                  * args[0] contains uninitialized data since
506                                  * for these tokens we don't expect any
507                                  * parameter.
508                                  */
509                                 if (token != Opt_compress &&
510                                     token != Opt_compress_force)
511                                         info->compress_level =
512                                           btrfs_compress_str2level(args[0].from);
513                                 btrfs_set_opt(info->mount_opt, COMPRESS);
514                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
515                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
516                                 no_compress = 0;
517                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
518                                 compress_type = "lzo";
519                                 info->compress_type = BTRFS_COMPRESS_LZO;
520                                 btrfs_set_opt(info->mount_opt, COMPRESS);
521                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
522                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
523                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
524                                 no_compress = 0;
525                         } else if (strcmp(args[0].from, "zstd") == 0) {
526                                 compress_type = "zstd";
527                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
528                                 btrfs_set_opt(info->mount_opt, COMPRESS);
529                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
530                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
531                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
532                                 no_compress = 0;
533                         } else if (strncmp(args[0].from, "no", 2) == 0) {
534                                 compress_type = "no";
535                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
536                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
537                                 compress_force = false;
538                                 no_compress++;
539                         } else {
540                                 ret = -EINVAL;
541                                 goto out;
542                         }
543
544                         if (compress_force) {
545                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
546                         } else {
547                                 /*
548                                  * If we remount from compress-force=xxx to
549                                  * compress=xxx, we need clear FORCE_COMPRESS
550                                  * flag, otherwise, there is no way for users
551                                  * to disable forcible compression separately.
552                                  */
553                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
554                         }
555                         if ((btrfs_test_opt(info, COMPRESS) &&
556                              (info->compress_type != saved_compress_type ||
557                               compress_force != saved_compress_force)) ||
558                             (!btrfs_test_opt(info, COMPRESS) &&
559                              no_compress == 1)) {
560                                 btrfs_info(info, "%s %s compression, level %d",
561                                            (compress_force) ? "force" : "use",
562                                            compress_type, info->compress_level);
563                         }
564                         compress_force = false;
565                         break;
566                 case Opt_ssd:
567                         btrfs_set_and_info(info, SSD,
568                                            "enabling ssd optimizations");
569                         btrfs_clear_opt(info->mount_opt, NOSSD);
570                         break;
571                 case Opt_ssd_spread:
572                         btrfs_set_and_info(info, SSD,
573                                            "enabling ssd optimizations");
574                         btrfs_set_and_info(info, SSD_SPREAD,
575                                            "using spread ssd allocation scheme");
576                         btrfs_clear_opt(info->mount_opt, NOSSD);
577                         break;
578                 case Opt_nossd:
579                         btrfs_set_opt(info->mount_opt, NOSSD);
580                         btrfs_clear_and_info(info, SSD,
581                                              "not using ssd optimizations");
582                         btrfs_clear_and_info(info, SSD_SPREAD,
583                                              "not using spread ssd allocation scheme");
584                         break;
585                 case Opt_barrier:
586                         btrfs_clear_and_info(info, NOBARRIER,
587                                              "turning on barriers");
588                         break;
589                 case Opt_nobarrier:
590                         btrfs_set_and_info(info, NOBARRIER,
591                                            "turning off barriers");
592                         break;
593                 case Opt_thread_pool:
594                         ret = match_int(&args[0], &intarg);
595                         if (ret) {
596                                 goto out;
597                         } else if (intarg == 0) {
598                                 ret = -EINVAL;
599                                 goto out;
600                         }
601                         info->thread_pool_size = intarg;
602                         break;
603                 case Opt_max_inline:
604                         num = match_strdup(&args[0]);
605                         if (num) {
606                                 info->max_inline = memparse(num, NULL);
607                                 kfree(num);
608
609                                 if (info->max_inline) {
610                                         info->max_inline = min_t(u64,
611                                                 info->max_inline,
612                                                 info->sectorsize);
613                                 }
614                                 btrfs_info(info, "max_inline at %llu",
615                                            info->max_inline);
616                         } else {
617                                 ret = -ENOMEM;
618                                 goto out;
619                         }
620                         break;
621                 case Opt_alloc_start:
622                         btrfs_info(info,
623                                 "option alloc_start is obsolete, ignored");
624                         break;
625                 case Opt_acl:
626 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
627                         info->sb->s_flags |= SB_POSIXACL;
628                         break;
629 #else
630                         btrfs_err(info, "support for ACL not compiled in!");
631                         ret = -EINVAL;
632                         goto out;
633 #endif
634                 case Opt_noacl:
635                         info->sb->s_flags &= ~SB_POSIXACL;
636                         break;
637                 case Opt_notreelog:
638                         btrfs_set_and_info(info, NOTREELOG,
639                                            "disabling tree log");
640                         break;
641                 case Opt_treelog:
642                         btrfs_clear_and_info(info, NOTREELOG,
643                                              "enabling tree log");
644                         break;
645                 case Opt_norecovery:
646                 case Opt_nologreplay:
647                         btrfs_set_and_info(info, NOLOGREPLAY,
648                                            "disabling log replay at mount time");
649                         break;
650                 case Opt_flushoncommit:
651                         btrfs_set_and_info(info, FLUSHONCOMMIT,
652                                            "turning on flush-on-commit");
653                         break;
654                 case Opt_noflushoncommit:
655                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
656                                              "turning off flush-on-commit");
657                         break;
658                 case Opt_ratio:
659                         ret = match_int(&args[0], &intarg);
660                         if (ret)
661                                 goto out;
662                         info->metadata_ratio = intarg;
663                         btrfs_info(info, "metadata ratio %u",
664                                    info->metadata_ratio);
665                         break;
666                 case Opt_discard:
667                         btrfs_set_and_info(info, DISCARD,
668                                            "turning on discard");
669                         break;
670                 case Opt_nodiscard:
671                         btrfs_clear_and_info(info, DISCARD,
672                                              "turning off discard");
673                         break;
674                 case Opt_space_cache:
675                 case Opt_space_cache_version:
676                         if (token == Opt_space_cache ||
677                             strcmp(args[0].from, "v1") == 0) {
678                                 btrfs_clear_opt(info->mount_opt,
679                                                 FREE_SPACE_TREE);
680                                 btrfs_set_and_info(info, SPACE_CACHE,
681                                            "enabling disk space caching");
682                         } else if (strcmp(args[0].from, "v2") == 0) {
683                                 btrfs_clear_opt(info->mount_opt,
684                                                 SPACE_CACHE);
685                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
686                                                    "enabling free space tree");
687                         } else {
688                                 ret = -EINVAL;
689                                 goto out;
690                         }
691                         break;
692                 case Opt_rescan_uuid_tree:
693                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
694                         break;
695                 case Opt_no_space_cache:
696                         if (btrfs_test_opt(info, SPACE_CACHE)) {
697                                 btrfs_clear_and_info(info, SPACE_CACHE,
698                                              "disabling disk space caching");
699                         }
700                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
701                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
702                                              "disabling free space tree");
703                         }
704                         break;
705                 case Opt_inode_cache:
706                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
707                                            "enabling inode map caching");
708                         break;
709                 case Opt_noinode_cache:
710                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
711                                              "disabling inode map caching");
712                         break;
713                 case Opt_clear_cache:
714                         btrfs_set_and_info(info, CLEAR_CACHE,
715                                            "force clearing of disk cache");
716                         break;
717                 case Opt_user_subvol_rm_allowed:
718                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
719                         break;
720                 case Opt_enospc_debug:
721                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
722                         break;
723                 case Opt_noenospc_debug:
724                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
725                         break;
726                 case Opt_defrag:
727                         btrfs_set_and_info(info, AUTO_DEFRAG,
728                                            "enabling auto defrag");
729                         break;
730                 case Opt_nodefrag:
731                         btrfs_clear_and_info(info, AUTO_DEFRAG,
732                                              "disabling auto defrag");
733                         break;
734                 case Opt_recovery:
735                         btrfs_warn(info,
736                                    "'recovery' is deprecated, use 'usebackuproot' instead");
737                 case Opt_usebackuproot:
738                         btrfs_info(info,
739                                    "trying to use backup root at mount time");
740                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
741                         break;
742                 case Opt_skip_balance:
743                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
744                         break;
745 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
746                 case Opt_check_integrity_including_extent_data:
747                         btrfs_info(info,
748                                    "enabling check integrity including extent data");
749                         btrfs_set_opt(info->mount_opt,
750                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
751                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
752                         break;
753                 case Opt_check_integrity:
754                         btrfs_info(info, "enabling check integrity");
755                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
756                         break;
757                 case Opt_check_integrity_print_mask:
758                         ret = match_int(&args[0], &intarg);
759                         if (ret)
760                                 goto out;
761                         info->check_integrity_print_mask = intarg;
762                         btrfs_info(info, "check_integrity_print_mask 0x%x",
763                                    info->check_integrity_print_mask);
764                         break;
765 #else
766                 case Opt_check_integrity_including_extent_data:
767                 case Opt_check_integrity:
768                 case Opt_check_integrity_print_mask:
769                         btrfs_err(info,
770                                   "support for check_integrity* not compiled in!");
771                         ret = -EINVAL;
772                         goto out;
773 #endif
774                 case Opt_fatal_errors:
775                         if (strcmp(args[0].from, "panic") == 0)
776                                 btrfs_set_opt(info->mount_opt,
777                                               PANIC_ON_FATAL_ERROR);
778                         else if (strcmp(args[0].from, "bug") == 0)
779                                 btrfs_clear_opt(info->mount_opt,
780                                               PANIC_ON_FATAL_ERROR);
781                         else {
782                                 ret = -EINVAL;
783                                 goto out;
784                         }
785                         break;
786                 case Opt_commit_interval:
787                         intarg = 0;
788                         ret = match_int(&args[0], &intarg);
789                         if (ret)
790                                 goto out;
791                         if (intarg == 0) {
792                                 btrfs_info(info,
793                                            "using default commit interval %us",
794                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
795                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
796                         } else if (intarg > 300) {
797                                 btrfs_warn(info, "excessive commit interval %d",
798                                            intarg);
799                         }
800                         info->commit_interval = intarg;
801                         break;
802 #ifdef CONFIG_BTRFS_DEBUG
803                 case Opt_fragment_all:
804                         btrfs_info(info, "fragmenting all space");
805                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
806                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
807                         break;
808                 case Opt_fragment_metadata:
809                         btrfs_info(info, "fragmenting metadata");
810                         btrfs_set_opt(info->mount_opt,
811                                       FRAGMENT_METADATA);
812                         break;
813                 case Opt_fragment_data:
814                         btrfs_info(info, "fragmenting data");
815                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
816                         break;
817 #endif
818 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
819                 case Opt_ref_verify:
820                         btrfs_info(info, "doing ref verification");
821                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
822                         break;
823 #endif
824                 case Opt_err:
825                         btrfs_info(info, "unrecognized mount option '%s'", p);
826                         ret = -EINVAL;
827                         goto out;
828                 default:
829                         break;
830                 }
831         }
832 check:
833         /*
834          * Extra check for current option against current flag
835          */
836         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & SB_RDONLY)) {
837                 btrfs_err(info,
838                           "nologreplay must be used with ro mount option");
839                 ret = -EINVAL;
840         }
841 out:
842         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
843             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
844             !btrfs_test_opt(info, CLEAR_CACHE)) {
845                 btrfs_err(info, "cannot disable free space tree");
846                 ret = -EINVAL;
847
848         }
849         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
850                 btrfs_info(info, "disk space caching is enabled");
851         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
852                 btrfs_info(info, "using free space tree");
853         return ret;
854 }
855
856 /*
857  * Parse mount options that are required early in the mount process.
858  *
859  * All other options will be parsed on much later in the mount process and
860  * only when we need to allocate a new super block.
861  */
862 static int btrfs_parse_early_options(const char *options, fmode_t flags,
863                 void *holder, struct btrfs_fs_devices **fs_devices)
864 {
865         substring_t args[MAX_OPT_ARGS];
866         char *device_name, *opts, *orig, *p;
867         int error = 0;
868
869         if (!options)
870                 return 0;
871
872         /*
873          * strsep changes the string, duplicate it because btrfs_parse_options
874          * gets called later
875          */
876         opts = kstrdup(options, GFP_KERNEL);
877         if (!opts)
878                 return -ENOMEM;
879         orig = opts;
880
881         while ((p = strsep(&opts, ",")) != NULL) {
882                 int token;
883
884                 if (!*p)
885                         continue;
886
887                 token = match_token(p, tokens, args);
888                 if (token == Opt_device) {
889                         device_name = match_strdup(&args[0]);
890                         if (!device_name) {
891                                 error = -ENOMEM;
892                                 goto out;
893                         }
894                         error = btrfs_scan_one_device(device_name,
895                                         flags, holder, fs_devices);
896                         kfree(device_name);
897                         if (error)
898                                 goto out;
899                 }
900         }
901
902 out:
903         kfree(orig);
904         return error;
905 }
906
907 /*
908  * Parse mount options that are related to subvolume id
909  *
910  * The value is later passed to mount_subvol()
911  */
912 static int btrfs_parse_subvol_options(const char *options, fmode_t flags,
913                 char **subvol_name, u64 *subvol_objectid)
914 {
915         substring_t args[MAX_OPT_ARGS];
916         char *opts, *orig, *p;
917         char *num = NULL;
918         int error = 0;
919
920         if (!options)
921                 return 0;
922
923         /*
924          * strsep changes the string, duplicate it because
925          * btrfs_parse_early_options gets called later
926          */
927         opts = kstrdup(options, GFP_KERNEL);
928         if (!opts)
929                 return -ENOMEM;
930         orig = opts;
931
932         while ((p = strsep(&opts, ",")) != NULL) {
933                 int token;
934                 if (!*p)
935                         continue;
936
937                 token = match_token(p, tokens, args);
938                 switch (token) {
939                 case Opt_subvol:
940                         kfree(*subvol_name);
941                         *subvol_name = match_strdup(&args[0]);
942                         if (!*subvol_name) {
943                                 error = -ENOMEM;
944                                 goto out;
945                         }
946                         break;
947                 case Opt_subvolid:
948                         num = match_strdup(&args[0]);
949                         if (num) {
950                                 *subvol_objectid = memparse(num, NULL);
951                                 kfree(num);
952                                 /* we want the original fs_tree */
953                                 if (!*subvol_objectid)
954                                         *subvol_objectid =
955                                                 BTRFS_FS_TREE_OBJECTID;
956                         } else {
957                                 error = -EINVAL;
958                                 goto out;
959                         }
960                         break;
961                 case Opt_subvolrootid:
962                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
963                         break;
964                 default:
965                         break;
966                 }
967         }
968
969 out:
970         kfree(orig);
971         return error;
972 }
973
974 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
975                                            u64 subvol_objectid)
976 {
977         struct btrfs_root *root = fs_info->tree_root;
978         struct btrfs_root *fs_root;
979         struct btrfs_root_ref *root_ref;
980         struct btrfs_inode_ref *inode_ref;
981         struct btrfs_key key;
982         struct btrfs_path *path = NULL;
983         char *name = NULL, *ptr;
984         u64 dirid;
985         int len;
986         int ret;
987
988         path = btrfs_alloc_path();
989         if (!path) {
990                 ret = -ENOMEM;
991                 goto err;
992         }
993         path->leave_spinning = 1;
994
995         name = kmalloc(PATH_MAX, GFP_KERNEL);
996         if (!name) {
997                 ret = -ENOMEM;
998                 goto err;
999         }
1000         ptr = name + PATH_MAX - 1;
1001         ptr[0] = '\0';
1002
1003         /*
1004          * Walk up the subvolume trees in the tree of tree roots by root
1005          * backrefs until we hit the top-level subvolume.
1006          */
1007         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1008                 key.objectid = subvol_objectid;
1009                 key.type = BTRFS_ROOT_BACKREF_KEY;
1010                 key.offset = (u64)-1;
1011
1012                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1013                 if (ret < 0) {
1014                         goto err;
1015                 } else if (ret > 0) {
1016                         ret = btrfs_previous_item(root, path, subvol_objectid,
1017                                                   BTRFS_ROOT_BACKREF_KEY);
1018                         if (ret < 0) {
1019                                 goto err;
1020                         } else if (ret > 0) {
1021                                 ret = -ENOENT;
1022                                 goto err;
1023                         }
1024                 }
1025
1026                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1027                 subvol_objectid = key.offset;
1028
1029                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1030                                           struct btrfs_root_ref);
1031                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1032                 ptr -= len + 1;
1033                 if (ptr < name) {
1034                         ret = -ENAMETOOLONG;
1035                         goto err;
1036                 }
1037                 read_extent_buffer(path->nodes[0], ptr + 1,
1038                                    (unsigned long)(root_ref + 1), len);
1039                 ptr[0] = '/';
1040                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1041                 btrfs_release_path(path);
1042
1043                 key.objectid = subvol_objectid;
1044                 key.type = BTRFS_ROOT_ITEM_KEY;
1045                 key.offset = (u64)-1;
1046                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1047                 if (IS_ERR(fs_root)) {
1048                         ret = PTR_ERR(fs_root);
1049                         goto err;
1050                 }
1051
1052                 /*
1053                  * Walk up the filesystem tree by inode refs until we hit the
1054                  * root directory.
1055                  */
1056                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1057                         key.objectid = dirid;
1058                         key.type = BTRFS_INODE_REF_KEY;
1059                         key.offset = (u64)-1;
1060
1061                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1062                         if (ret < 0) {
1063                                 goto err;
1064                         } else if (ret > 0) {
1065                                 ret = btrfs_previous_item(fs_root, path, dirid,
1066                                                           BTRFS_INODE_REF_KEY);
1067                                 if (ret < 0) {
1068                                         goto err;
1069                                 } else if (ret > 0) {
1070                                         ret = -ENOENT;
1071                                         goto err;
1072                                 }
1073                         }
1074
1075                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1076                         dirid = key.offset;
1077
1078                         inode_ref = btrfs_item_ptr(path->nodes[0],
1079                                                    path->slots[0],
1080                                                    struct btrfs_inode_ref);
1081                         len = btrfs_inode_ref_name_len(path->nodes[0],
1082                                                        inode_ref);
1083                         ptr -= len + 1;
1084                         if (ptr < name) {
1085                                 ret = -ENAMETOOLONG;
1086                                 goto err;
1087                         }
1088                         read_extent_buffer(path->nodes[0], ptr + 1,
1089                                            (unsigned long)(inode_ref + 1), len);
1090                         ptr[0] = '/';
1091                         btrfs_release_path(path);
1092                 }
1093         }
1094
1095         btrfs_free_path(path);
1096         if (ptr == name + PATH_MAX - 1) {
1097                 name[0] = '/';
1098                 name[1] = '\0';
1099         } else {
1100                 memmove(name, ptr, name + PATH_MAX - ptr);
1101         }
1102         return name;
1103
1104 err:
1105         btrfs_free_path(path);
1106         kfree(name);
1107         return ERR_PTR(ret);
1108 }
1109
1110 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1111 {
1112         struct btrfs_root *root = fs_info->tree_root;
1113         struct btrfs_dir_item *di;
1114         struct btrfs_path *path;
1115         struct btrfs_key location;
1116         u64 dir_id;
1117
1118         path = btrfs_alloc_path();
1119         if (!path)
1120                 return -ENOMEM;
1121         path->leave_spinning = 1;
1122
1123         /*
1124          * Find the "default" dir item which points to the root item that we
1125          * will mount by default if we haven't been given a specific subvolume
1126          * to mount.
1127          */
1128         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1129         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1130         if (IS_ERR(di)) {
1131                 btrfs_free_path(path);
1132                 return PTR_ERR(di);
1133         }
1134         if (!di) {
1135                 /*
1136                  * Ok the default dir item isn't there.  This is weird since
1137                  * it's always been there, but don't freak out, just try and
1138                  * mount the top-level subvolume.
1139                  */
1140                 btrfs_free_path(path);
1141                 *objectid = BTRFS_FS_TREE_OBJECTID;
1142                 return 0;
1143         }
1144
1145         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1146         btrfs_free_path(path);
1147         *objectid = location.objectid;
1148         return 0;
1149 }
1150
1151 static int btrfs_fill_super(struct super_block *sb,
1152                             struct btrfs_fs_devices *fs_devices,
1153                             void *data)
1154 {
1155         struct inode *inode;
1156         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1157         struct btrfs_key key;
1158         int err;
1159
1160         sb->s_maxbytes = MAX_LFS_FILESIZE;
1161         sb->s_magic = BTRFS_SUPER_MAGIC;
1162         sb->s_op = &btrfs_super_ops;
1163         sb->s_d_op = &btrfs_dentry_operations;
1164         sb->s_export_op = &btrfs_export_ops;
1165         sb->s_xattr = btrfs_xattr_handlers;
1166         sb->s_time_gran = 1;
1167 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1168         sb->s_flags |= SB_POSIXACL;
1169 #endif
1170         sb->s_flags |= SB_I_VERSION;
1171         sb->s_iflags |= SB_I_CGROUPWB;
1172
1173         err = super_setup_bdi(sb);
1174         if (err) {
1175                 btrfs_err(fs_info, "super_setup_bdi failed");
1176                 return err;
1177         }
1178
1179         err = open_ctree(sb, fs_devices, (char *)data);
1180         if (err) {
1181                 btrfs_err(fs_info, "open_ctree failed");
1182                 return err;
1183         }
1184
1185         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1186         key.type = BTRFS_INODE_ITEM_KEY;
1187         key.offset = 0;
1188         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1189         if (IS_ERR(inode)) {
1190                 err = PTR_ERR(inode);
1191                 goto fail_close;
1192         }
1193
1194         sb->s_root = d_make_root(inode);
1195         if (!sb->s_root) {
1196                 err = -ENOMEM;
1197                 goto fail_close;
1198         }
1199
1200         cleancache_init_fs(sb);
1201         sb->s_flags |= SB_ACTIVE;
1202         return 0;
1203
1204 fail_close:
1205         close_ctree(fs_info);
1206         return err;
1207 }
1208
1209 int btrfs_sync_fs(struct super_block *sb, int wait)
1210 {
1211         struct btrfs_trans_handle *trans;
1212         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1213         struct btrfs_root *root = fs_info->tree_root;
1214
1215         trace_btrfs_sync_fs(fs_info, wait);
1216
1217         if (!wait) {
1218                 filemap_flush(fs_info->btree_inode->i_mapping);
1219                 return 0;
1220         }
1221
1222         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1223
1224         trans = btrfs_attach_transaction_barrier(root);
1225         if (IS_ERR(trans)) {
1226                 /* no transaction, don't bother */
1227                 if (PTR_ERR(trans) == -ENOENT) {
1228                         /*
1229                          * Exit unless we have some pending changes
1230                          * that need to go through commit
1231                          */
1232                         if (fs_info->pending_changes == 0)
1233                                 return 0;
1234                         /*
1235                          * A non-blocking test if the fs is frozen. We must not
1236                          * start a new transaction here otherwise a deadlock
1237                          * happens. The pending operations are delayed to the
1238                          * next commit after thawing.
1239                          */
1240                         if (sb_start_write_trylock(sb))
1241                                 sb_end_write(sb);
1242                         else
1243                                 return 0;
1244                         trans = btrfs_start_transaction(root, 0);
1245                 }
1246                 if (IS_ERR(trans))
1247                         return PTR_ERR(trans);
1248         }
1249         return btrfs_commit_transaction(trans);
1250 }
1251
1252 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1253 {
1254         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1255         const char *compress_type;
1256
1257         if (btrfs_test_opt(info, DEGRADED))
1258                 seq_puts(seq, ",degraded");
1259         if (btrfs_test_opt(info, NODATASUM))
1260                 seq_puts(seq, ",nodatasum");
1261         if (btrfs_test_opt(info, NODATACOW))
1262                 seq_puts(seq, ",nodatacow");
1263         if (btrfs_test_opt(info, NOBARRIER))
1264                 seq_puts(seq, ",nobarrier");
1265         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1266                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1267         if (info->thread_pool_size !=  min_t(unsigned long,
1268                                              num_online_cpus() + 2, 8))
1269                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1270         if (btrfs_test_opt(info, COMPRESS)) {
1271                 compress_type = btrfs_compress_type2str(info->compress_type);
1272                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1273                         seq_printf(seq, ",compress-force=%s", compress_type);
1274                 else
1275                         seq_printf(seq, ",compress=%s", compress_type);
1276                 if (info->compress_level)
1277                         seq_printf(seq, ":%d", info->compress_level);
1278         }
1279         if (btrfs_test_opt(info, NOSSD))
1280                 seq_puts(seq, ",nossd");
1281         if (btrfs_test_opt(info, SSD_SPREAD))
1282                 seq_puts(seq, ",ssd_spread");
1283         else if (btrfs_test_opt(info, SSD))
1284                 seq_puts(seq, ",ssd");
1285         if (btrfs_test_opt(info, NOTREELOG))
1286                 seq_puts(seq, ",notreelog");
1287         if (btrfs_test_opt(info, NOLOGREPLAY))
1288                 seq_puts(seq, ",nologreplay");
1289         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1290                 seq_puts(seq, ",flushoncommit");
1291         if (btrfs_test_opt(info, DISCARD))
1292                 seq_puts(seq, ",discard");
1293         if (!(info->sb->s_flags & SB_POSIXACL))
1294                 seq_puts(seq, ",noacl");
1295         if (btrfs_test_opt(info, SPACE_CACHE))
1296                 seq_puts(seq, ",space_cache");
1297         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1298                 seq_puts(seq, ",space_cache=v2");
1299         else
1300                 seq_puts(seq, ",nospace_cache");
1301         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1302                 seq_puts(seq, ",rescan_uuid_tree");
1303         if (btrfs_test_opt(info, CLEAR_CACHE))
1304                 seq_puts(seq, ",clear_cache");
1305         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1306                 seq_puts(seq, ",user_subvol_rm_allowed");
1307         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1308                 seq_puts(seq, ",enospc_debug");
1309         if (btrfs_test_opt(info, AUTO_DEFRAG))
1310                 seq_puts(seq, ",autodefrag");
1311         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1312                 seq_puts(seq, ",inode_cache");
1313         if (btrfs_test_opt(info, SKIP_BALANCE))
1314                 seq_puts(seq, ",skip_balance");
1315 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1316         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1317                 seq_puts(seq, ",check_int_data");
1318         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1319                 seq_puts(seq, ",check_int");
1320         if (info->check_integrity_print_mask)
1321                 seq_printf(seq, ",check_int_print_mask=%d",
1322                                 info->check_integrity_print_mask);
1323 #endif
1324         if (info->metadata_ratio)
1325                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1326         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1327                 seq_puts(seq, ",fatal_errors=panic");
1328         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1329                 seq_printf(seq, ",commit=%u", info->commit_interval);
1330 #ifdef CONFIG_BTRFS_DEBUG
1331         if (btrfs_test_opt(info, FRAGMENT_DATA))
1332                 seq_puts(seq, ",fragment=data");
1333         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1334                 seq_puts(seq, ",fragment=metadata");
1335 #endif
1336         if (btrfs_test_opt(info, REF_VERIFY))
1337                 seq_puts(seq, ",ref_verify");
1338         seq_printf(seq, ",subvolid=%llu",
1339                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1340         seq_puts(seq, ",subvol=");
1341         seq_dentry(seq, dentry, " \t\n\\");
1342         return 0;
1343 }
1344
1345 static int btrfs_test_super(struct super_block *s, void *data)
1346 {
1347         struct btrfs_fs_info *p = data;
1348         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1349
1350         return fs_info->fs_devices == p->fs_devices;
1351 }
1352
1353 static int btrfs_set_super(struct super_block *s, void *data)
1354 {
1355         int err = set_anon_super(s, data);
1356         if (!err)
1357                 s->s_fs_info = data;
1358         return err;
1359 }
1360
1361 /*
1362  * subvolumes are identified by ino 256
1363  */
1364 static inline int is_subvolume_inode(struct inode *inode)
1365 {
1366         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1367                 return 1;
1368         return 0;
1369 }
1370
1371 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1372                                    const char *device_name, struct vfsmount *mnt)
1373 {
1374         struct dentry *root;
1375         int ret;
1376
1377         if (!subvol_name) {
1378                 if (!subvol_objectid) {
1379                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1380                                                           &subvol_objectid);
1381                         if (ret) {
1382                                 root = ERR_PTR(ret);
1383                                 goto out;
1384                         }
1385                 }
1386                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1387                                                             subvol_objectid);
1388                 if (IS_ERR(subvol_name)) {
1389                         root = ERR_CAST(subvol_name);
1390                         subvol_name = NULL;
1391                         goto out;
1392                 }
1393
1394         }
1395
1396         root = mount_subtree(mnt, subvol_name);
1397         /* mount_subtree() drops our reference on the vfsmount. */
1398         mnt = NULL;
1399
1400         if (!IS_ERR(root)) {
1401                 struct super_block *s = root->d_sb;
1402                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1403                 struct inode *root_inode = d_inode(root);
1404                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1405
1406                 ret = 0;
1407                 if (!is_subvolume_inode(root_inode)) {
1408                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1409                                subvol_name);
1410                         ret = -EINVAL;
1411                 }
1412                 if (subvol_objectid && root_objectid != subvol_objectid) {
1413                         /*
1414                          * This will also catch a race condition where a
1415                          * subvolume which was passed by ID is renamed and
1416                          * another subvolume is renamed over the old location.
1417                          */
1418                         btrfs_err(fs_info,
1419                                   "subvol '%s' does not match subvolid %llu",
1420                                   subvol_name, subvol_objectid);
1421                         ret = -EINVAL;
1422                 }
1423                 if (ret) {
1424                         dput(root);
1425                         root = ERR_PTR(ret);
1426                         deactivate_locked_super(s);
1427                 }
1428         }
1429
1430 out:
1431         mntput(mnt);
1432         kfree(subvol_name);
1433         return root;
1434 }
1435
1436 static int parse_security_options(char *orig_opts,
1437                                   struct security_mnt_opts *sec_opts)
1438 {
1439         char *secdata = NULL;
1440         int ret = 0;
1441
1442         secdata = alloc_secdata();
1443         if (!secdata)
1444                 return -ENOMEM;
1445         ret = security_sb_copy_data(orig_opts, secdata);
1446         if (ret) {
1447                 free_secdata(secdata);
1448                 return ret;
1449         }
1450         ret = security_sb_parse_opts_str(secdata, sec_opts);
1451         free_secdata(secdata);
1452         return ret;
1453 }
1454
1455 static int setup_security_options(struct btrfs_fs_info *fs_info,
1456                                   struct super_block *sb,
1457                                   struct security_mnt_opts *sec_opts)
1458 {
1459         int ret = 0;
1460
1461         /*
1462          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1463          * is valid.
1464          */
1465         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1466         if (ret)
1467                 return ret;
1468
1469 #ifdef CONFIG_SECURITY
1470         if (!fs_info->security_opts.num_mnt_opts) {
1471                 /* first time security setup, copy sec_opts to fs_info */
1472                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1473         } else {
1474                 /*
1475                  * Since SELinux (the only one supporting security_mnt_opts)
1476                  * does NOT support changing context during remount/mount of
1477                  * the same sb, this must be the same or part of the same
1478                  * security options, just free it.
1479                  */
1480                 security_free_mnt_opts(sec_opts);
1481         }
1482 #endif
1483         return ret;
1484 }
1485
1486 /*
1487  * Find a superblock for the given device / mount point.
1488  *
1489  * Note: This is based on mount_bdev from fs/super.c with a few additions
1490  *       for multiple device setup.  Make sure to keep it in sync.
1491  */
1492 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1493                 int flags, const char *device_name, void *data)
1494 {
1495         struct block_device *bdev = NULL;
1496         struct super_block *s;
1497         struct btrfs_fs_devices *fs_devices = NULL;
1498         struct btrfs_fs_info *fs_info = NULL;
1499         struct security_mnt_opts new_sec_opts;
1500         fmode_t mode = FMODE_READ;
1501         int error = 0;
1502
1503         if (!(flags & SB_RDONLY))
1504                 mode |= FMODE_WRITE;
1505
1506         error = btrfs_parse_early_options(data, mode, fs_type,
1507                                           &fs_devices);
1508         if (error) {
1509                 return ERR_PTR(error);
1510         }
1511
1512         security_init_mnt_opts(&new_sec_opts);
1513         if (data) {
1514                 error = parse_security_options(data, &new_sec_opts);
1515                 if (error)
1516                         return ERR_PTR(error);
1517         }
1518
1519         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1520         if (error)
1521                 goto error_sec_opts;
1522
1523         /*
1524          * Setup a dummy root and fs_info for test/set super.  This is because
1525          * we don't actually fill this stuff out until open_ctree, but we need
1526          * it for searching for existing supers, so this lets us do that and
1527          * then open_ctree will properly initialize everything later.
1528          */
1529         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1530         if (!fs_info) {
1531                 error = -ENOMEM;
1532                 goto error_sec_opts;
1533         }
1534
1535         fs_info->fs_devices = fs_devices;
1536
1537         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1538         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1539         security_init_mnt_opts(&fs_info->security_opts);
1540         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1541                 error = -ENOMEM;
1542                 goto error_fs_info;
1543         }
1544
1545         error = btrfs_open_devices(fs_devices, mode, fs_type);
1546         if (error)
1547                 goto error_fs_info;
1548
1549         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1550                 error = -EACCES;
1551                 goto error_close_devices;
1552         }
1553
1554         bdev = fs_devices->latest_bdev;
1555         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1556                  fs_info);
1557         if (IS_ERR(s)) {
1558                 error = PTR_ERR(s);
1559                 goto error_close_devices;
1560         }
1561
1562         if (s->s_root) {
1563                 btrfs_close_devices(fs_devices);
1564                 free_fs_info(fs_info);
1565                 if ((flags ^ s->s_flags) & SB_RDONLY)
1566                         error = -EBUSY;
1567         } else {
1568                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1569                 btrfs_sb(s)->bdev_holder = fs_type;
1570                 error = btrfs_fill_super(s, fs_devices, data);
1571         }
1572         if (error) {
1573                 deactivate_locked_super(s);
1574                 goto error_sec_opts;
1575         }
1576
1577         fs_info = btrfs_sb(s);
1578         error = setup_security_options(fs_info, s, &new_sec_opts);
1579         if (error) {
1580                 deactivate_locked_super(s);
1581                 goto error_sec_opts;
1582         }
1583
1584         return dget(s->s_root);
1585
1586 error_close_devices:
1587         btrfs_close_devices(fs_devices);
1588 error_fs_info:
1589         free_fs_info(fs_info);
1590 error_sec_opts:
1591         security_free_mnt_opts(&new_sec_opts);
1592         return ERR_PTR(error);
1593 }
1594
1595 /*
1596  * Mount function which is called by VFS layer.
1597  *
1598  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1599  * which needs vfsmount* of device's root (/).  This means device's root has to
1600  * be mounted internally in any case.
1601  *
1602  * Operation flow:
1603  *   1. Parse subvol id related options for later use in mount_subvol().
1604  *
1605  *   2. Mount device's root (/) by calling vfs_kern_mount().
1606  *
1607  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1608  *      first place. In order to avoid calling btrfs_mount() again, we use
1609  *      different file_system_type which is not registered to VFS by
1610  *      register_filesystem() (btrfs_root_fs_type). As a result,
1611  *      btrfs_mount_root() is called. The return value will be used by
1612  *      mount_subtree() in mount_subvol().
1613  *
1614  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1615  *      "btrfs subvolume set-default", mount_subvol() is called always.
1616  */
1617 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1618                 const char *device_name, void *data)
1619 {
1620         struct vfsmount *mnt_root;
1621         struct dentry *root;
1622         fmode_t mode = FMODE_READ;
1623         char *subvol_name = NULL;
1624         u64 subvol_objectid = 0;
1625         int error = 0;
1626
1627         if (!(flags & SB_RDONLY))
1628                 mode |= FMODE_WRITE;
1629
1630         error = btrfs_parse_subvol_options(data, mode,
1631                                           &subvol_name, &subvol_objectid);
1632         if (error) {
1633                 kfree(subvol_name);
1634                 return ERR_PTR(error);
1635         }
1636
1637         /* mount device's root (/) */
1638         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1639         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1640                 if (flags & SB_RDONLY) {
1641                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1642                                 flags & ~SB_RDONLY, device_name, data);
1643                 } else {
1644                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1645                                 flags | SB_RDONLY, device_name, data);
1646                         if (IS_ERR(mnt_root)) {
1647                                 root = ERR_CAST(mnt_root);
1648                                 goto out;
1649                         }
1650
1651                         down_write(&mnt_root->mnt_sb->s_umount);
1652                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1653                         up_write(&mnt_root->mnt_sb->s_umount);
1654                         if (error < 0) {
1655                                 root = ERR_PTR(error);
1656                                 mntput(mnt_root);
1657                                 goto out;
1658                         }
1659                 }
1660         }
1661         if (IS_ERR(mnt_root)) {
1662                 root = ERR_CAST(mnt_root);
1663                 goto out;
1664         }
1665
1666         /* mount_subvol() will free subvol_name and mnt_root */
1667         root = mount_subvol(subvol_name, subvol_objectid, device_name, mnt_root);
1668
1669 out:
1670         return root;
1671 }
1672
1673 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1674                                      u32 new_pool_size, u32 old_pool_size)
1675 {
1676         if (new_pool_size == old_pool_size)
1677                 return;
1678
1679         fs_info->thread_pool_size = new_pool_size;
1680
1681         btrfs_info(fs_info, "resize thread pool %d -> %d",
1682                old_pool_size, new_pool_size);
1683
1684         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1685         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1686         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1687         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1688         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1689         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1690         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1691                                 new_pool_size);
1692         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1693         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1694         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1695         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1696         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1697                                 new_pool_size);
1698 }
1699
1700 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1701 {
1702         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1703 }
1704
1705 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1706                                        unsigned long old_opts, int flags)
1707 {
1708         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1709             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1710              (flags & SB_RDONLY))) {
1711                 /* wait for any defraggers to finish */
1712                 wait_event(fs_info->transaction_wait,
1713                            (atomic_read(&fs_info->defrag_running) == 0));
1714                 if (flags & SB_RDONLY)
1715                         sync_filesystem(fs_info->sb);
1716         }
1717 }
1718
1719 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1720                                          unsigned long old_opts)
1721 {
1722         /*
1723          * We need to cleanup all defragable inodes if the autodefragment is
1724          * close or the filesystem is read only.
1725          */
1726         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1727             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1728                 btrfs_cleanup_defrag_inodes(fs_info);
1729         }
1730
1731         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1732 }
1733
1734 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1735 {
1736         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1737         struct btrfs_root *root = fs_info->tree_root;
1738         unsigned old_flags = sb->s_flags;
1739         unsigned long old_opts = fs_info->mount_opt;
1740         unsigned long old_compress_type = fs_info->compress_type;
1741         u64 old_max_inline = fs_info->max_inline;
1742         u32 old_thread_pool_size = fs_info->thread_pool_size;
1743         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1744         int ret;
1745
1746         sync_filesystem(sb);
1747         btrfs_remount_prepare(fs_info);
1748
1749         if (data) {
1750                 struct security_mnt_opts new_sec_opts;
1751
1752                 security_init_mnt_opts(&new_sec_opts);
1753                 ret = parse_security_options(data, &new_sec_opts);
1754                 if (ret)
1755                         goto restore;
1756                 ret = setup_security_options(fs_info, sb,
1757                                              &new_sec_opts);
1758                 if (ret) {
1759                         security_free_mnt_opts(&new_sec_opts);
1760                         goto restore;
1761                 }
1762         }
1763
1764         ret = btrfs_parse_options(fs_info, data, *flags);
1765         if (ret) {
1766                 ret = -EINVAL;
1767                 goto restore;
1768         }
1769
1770         btrfs_remount_begin(fs_info, old_opts, *flags);
1771         btrfs_resize_thread_pool(fs_info,
1772                 fs_info->thread_pool_size, old_thread_pool_size);
1773
1774         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1775                 goto out;
1776
1777         if (*flags & SB_RDONLY) {
1778                 /*
1779                  * this also happens on 'umount -rf' or on shutdown, when
1780                  * the filesystem is busy.
1781                  */
1782                 cancel_work_sync(&fs_info->async_reclaim_work);
1783
1784                 /* wait for the uuid_scan task to finish */
1785                 down(&fs_info->uuid_tree_rescan_sem);
1786                 /* avoid complains from lockdep et al. */
1787                 up(&fs_info->uuid_tree_rescan_sem);
1788
1789                 sb->s_flags |= SB_RDONLY;
1790
1791                 /*
1792                  * Setting SB_RDONLY will put the cleaner thread to
1793                  * sleep at the next loop if it's already active.
1794                  * If it's already asleep, we'll leave unused block
1795                  * groups on disk until we're mounted read-write again
1796                  * unless we clean them up here.
1797                  */
1798                 btrfs_delete_unused_bgs(fs_info);
1799
1800                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1801                 btrfs_scrub_cancel(fs_info);
1802                 btrfs_pause_balance(fs_info);
1803
1804                 ret = btrfs_commit_super(fs_info);
1805                 if (ret)
1806                         goto restore;
1807         } else {
1808                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1809                         btrfs_err(fs_info,
1810                                 "Remounting read-write after error is not allowed");
1811                         ret = -EINVAL;
1812                         goto restore;
1813                 }
1814                 if (fs_info->fs_devices->rw_devices == 0) {
1815                         ret = -EACCES;
1816                         goto restore;
1817                 }
1818
1819                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1820                         btrfs_warn(fs_info,
1821                                 "too many missing devices, writeable remount is not allowed");
1822                         ret = -EACCES;
1823                         goto restore;
1824                 }
1825
1826                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1827                         ret = -EINVAL;
1828                         goto restore;
1829                 }
1830
1831                 ret = btrfs_cleanup_fs_roots(fs_info);
1832                 if (ret)
1833                         goto restore;
1834
1835                 /* recover relocation */
1836                 mutex_lock(&fs_info->cleaner_mutex);
1837                 ret = btrfs_recover_relocation(root);
1838                 mutex_unlock(&fs_info->cleaner_mutex);
1839                 if (ret)
1840                         goto restore;
1841
1842                 ret = btrfs_resume_balance_async(fs_info);
1843                 if (ret)
1844                         goto restore;
1845
1846                 ret = btrfs_resume_dev_replace_async(fs_info);
1847                 if (ret) {
1848                         btrfs_warn(fs_info, "failed to resume dev_replace");
1849                         goto restore;
1850                 }
1851
1852                 btrfs_qgroup_rescan_resume(fs_info);
1853
1854                 if (!fs_info->uuid_root) {
1855                         btrfs_info(fs_info, "creating UUID tree");
1856                         ret = btrfs_create_uuid_tree(fs_info);
1857                         if (ret) {
1858                                 btrfs_warn(fs_info,
1859                                            "failed to create the UUID tree %d",
1860                                            ret);
1861                                 goto restore;
1862                         }
1863                 }
1864                 sb->s_flags &= ~SB_RDONLY;
1865
1866                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1867         }
1868 out:
1869         wake_up_process(fs_info->transaction_kthread);
1870         btrfs_remount_cleanup(fs_info, old_opts);
1871         return 0;
1872
1873 restore:
1874         /* We've hit an error - don't reset SB_RDONLY */
1875         if (sb_rdonly(sb))
1876                 old_flags |= SB_RDONLY;
1877         sb->s_flags = old_flags;
1878         fs_info->mount_opt = old_opts;
1879         fs_info->compress_type = old_compress_type;
1880         fs_info->max_inline = old_max_inline;
1881         btrfs_resize_thread_pool(fs_info,
1882                 old_thread_pool_size, fs_info->thread_pool_size);
1883         fs_info->metadata_ratio = old_metadata_ratio;
1884         btrfs_remount_cleanup(fs_info, old_opts);
1885         return ret;
1886 }
1887
1888 /* Used to sort the devices by max_avail(descending sort) */
1889 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1890                                        const void *dev_info2)
1891 {
1892         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1893             ((struct btrfs_device_info *)dev_info2)->max_avail)
1894                 return -1;
1895         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1896                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1897                 return 1;
1898         else
1899         return 0;
1900 }
1901
1902 /*
1903  * sort the devices by max_avail, in which max free extent size of each device
1904  * is stored.(Descending Sort)
1905  */
1906 static inline void btrfs_descending_sort_devices(
1907                                         struct btrfs_device_info *devices,
1908                                         size_t nr_devices)
1909 {
1910         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1911              btrfs_cmp_device_free_bytes, NULL);
1912 }
1913
1914 /*
1915  * The helper to calc the free space on the devices that can be used to store
1916  * file data.
1917  */
1918 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1919                                        u64 *free_bytes)
1920 {
1921         struct btrfs_device_info *devices_info;
1922         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1923         struct btrfs_device *device;
1924         u64 skip_space;
1925         u64 type;
1926         u64 avail_space;
1927         u64 min_stripe_size;
1928         int min_stripes = 1, num_stripes = 1;
1929         int i = 0, nr_devices;
1930
1931         /*
1932          * We aren't under the device list lock, so this is racy-ish, but good
1933          * enough for our purposes.
1934          */
1935         nr_devices = fs_info->fs_devices->open_devices;
1936         if (!nr_devices) {
1937                 smp_mb();
1938                 nr_devices = fs_info->fs_devices->open_devices;
1939                 ASSERT(nr_devices);
1940                 if (!nr_devices) {
1941                         *free_bytes = 0;
1942                         return 0;
1943                 }
1944         }
1945
1946         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1947                                GFP_KERNEL);
1948         if (!devices_info)
1949                 return -ENOMEM;
1950
1951         /* calc min stripe number for data space allocation */
1952         type = btrfs_data_alloc_profile(fs_info);
1953         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1954                 min_stripes = 2;
1955                 num_stripes = nr_devices;
1956         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1957                 min_stripes = 2;
1958                 num_stripes = 2;
1959         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1960                 min_stripes = 4;
1961                 num_stripes = 4;
1962         }
1963
1964         if (type & BTRFS_BLOCK_GROUP_DUP)
1965                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1966         else
1967                 min_stripe_size = BTRFS_STRIPE_LEN;
1968
1969         rcu_read_lock();
1970         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1971                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1972                                                 &device->dev_state) ||
1973                     !device->bdev ||
1974                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1975                         continue;
1976
1977                 if (i >= nr_devices)
1978                         break;
1979
1980                 avail_space = device->total_bytes - device->bytes_used;
1981
1982                 /* align with stripe_len */
1983                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1984                 avail_space *= BTRFS_STRIPE_LEN;
1985
1986                 /*
1987                  * In order to avoid overwriting the superblock on the drive,
1988                  * btrfs starts at an offset of at least 1MB when doing chunk
1989                  * allocation.
1990                  */
1991                 skip_space = SZ_1M;
1992
1993                 /*
1994                  * we can use the free space in [0, skip_space - 1], subtract
1995                  * it from the total.
1996                  */
1997                 if (avail_space && avail_space >= skip_space)
1998                         avail_space -= skip_space;
1999                 else
2000                         avail_space = 0;
2001
2002                 if (avail_space < min_stripe_size)
2003                         continue;
2004
2005                 devices_info[i].dev = device;
2006                 devices_info[i].max_avail = avail_space;
2007
2008                 i++;
2009         }
2010         rcu_read_unlock();
2011
2012         nr_devices = i;
2013
2014         btrfs_descending_sort_devices(devices_info, nr_devices);
2015
2016         i = nr_devices - 1;
2017         avail_space = 0;
2018         while (nr_devices >= min_stripes) {
2019                 if (num_stripes > nr_devices)
2020                         num_stripes = nr_devices;
2021
2022                 if (devices_info[i].max_avail >= min_stripe_size) {
2023                         int j;
2024                         u64 alloc_size;
2025
2026                         avail_space += devices_info[i].max_avail * num_stripes;
2027                         alloc_size = devices_info[i].max_avail;
2028                         for (j = i + 1 - num_stripes; j <= i; j++)
2029                                 devices_info[j].max_avail -= alloc_size;
2030                 }
2031                 i--;
2032                 nr_devices--;
2033         }
2034
2035         kfree(devices_info);
2036         *free_bytes = avail_space;
2037         return 0;
2038 }
2039
2040 /*
2041  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2042  *
2043  * If there's a redundant raid level at DATA block groups, use the respective
2044  * multiplier to scale the sizes.
2045  *
2046  * Unused device space usage is based on simulating the chunk allocator
2047  * algorithm that respects the device sizes and order of allocations.  This is
2048  * a close approximation of the actual use but there are other factors that may
2049  * change the result (like a new metadata chunk).
2050  *
2051  * If metadata is exhausted, f_bavail will be 0.
2052  */
2053 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2054 {
2055         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2056         struct btrfs_super_block *disk_super = fs_info->super_copy;
2057         struct list_head *head = &fs_info->space_info;
2058         struct btrfs_space_info *found;
2059         u64 total_used = 0;
2060         u64 total_free_data = 0;
2061         u64 total_free_meta = 0;
2062         int bits = dentry->d_sb->s_blocksize_bits;
2063         __be32 *fsid = (__be32 *)fs_info->fsid;
2064         unsigned factor = 1;
2065         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2066         int ret;
2067         u64 thresh = 0;
2068         int mixed = 0;
2069
2070         rcu_read_lock();
2071         list_for_each_entry_rcu(found, head, list) {
2072                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2073                         int i;
2074
2075                         total_free_data += found->disk_total - found->disk_used;
2076                         total_free_data -=
2077                                 btrfs_account_ro_block_groups_free_space(found);
2078
2079                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2080                                 if (!list_empty(&found->block_groups[i])) {
2081                                         switch (i) {
2082                                         case BTRFS_RAID_DUP:
2083                                         case BTRFS_RAID_RAID1:
2084                                         case BTRFS_RAID_RAID10:
2085                                                 factor = 2;
2086                                         }
2087                                 }
2088                         }
2089                 }
2090
2091                 /*
2092                  * Metadata in mixed block goup profiles are accounted in data
2093                  */
2094                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2095                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2096                                 mixed = 1;
2097                         else
2098                                 total_free_meta += found->disk_total -
2099                                         found->disk_used;
2100                 }
2101
2102                 total_used += found->disk_used;
2103         }
2104
2105         rcu_read_unlock();
2106
2107         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2108         buf->f_blocks >>= bits;
2109         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2110
2111         /* Account global block reserve as used, it's in logical size already */
2112         spin_lock(&block_rsv->lock);
2113         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2114         if (buf->f_bfree >= block_rsv->size >> bits)
2115                 buf->f_bfree -= block_rsv->size >> bits;
2116         else
2117                 buf->f_bfree = 0;
2118         spin_unlock(&block_rsv->lock);
2119
2120         buf->f_bavail = div_u64(total_free_data, factor);
2121         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2122         if (ret)
2123                 return ret;
2124         buf->f_bavail += div_u64(total_free_data, factor);
2125         buf->f_bavail = buf->f_bavail >> bits;
2126
2127         /*
2128          * We calculate the remaining metadata space minus global reserve. If
2129          * this is (supposedly) smaller than zero, there's no space. But this
2130          * does not hold in practice, the exhausted state happens where's still
2131          * some positive delta. So we apply some guesswork and compare the
2132          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2133          *
2134          * We probably cannot calculate the exact threshold value because this
2135          * depends on the internal reservations requested by various
2136          * operations, so some operations that consume a few metadata will
2137          * succeed even if the Avail is zero. But this is better than the other
2138          * way around.
2139          */
2140         thresh = SZ_4M;
2141
2142         if (!mixed && total_free_meta - thresh < block_rsv->size)
2143                 buf->f_bavail = 0;
2144
2145         buf->f_type = BTRFS_SUPER_MAGIC;
2146         buf->f_bsize = dentry->d_sb->s_blocksize;
2147         buf->f_namelen = BTRFS_NAME_LEN;
2148
2149         /* We treat it as constant endianness (it doesn't matter _which_)
2150            because we want the fsid to come out the same whether mounted
2151            on a big-endian or little-endian host */
2152         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2153         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2154         /* Mask in the root object ID too, to disambiguate subvols */
2155         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2156         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2157
2158         return 0;
2159 }
2160
2161 static void btrfs_kill_super(struct super_block *sb)
2162 {
2163         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2164         kill_anon_super(sb);
2165         free_fs_info(fs_info);
2166 }
2167
2168 static struct file_system_type btrfs_fs_type = {
2169         .owner          = THIS_MODULE,
2170         .name           = "btrfs",
2171         .mount          = btrfs_mount,
2172         .kill_sb        = btrfs_kill_super,
2173         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2174 };
2175
2176 static struct file_system_type btrfs_root_fs_type = {
2177         .owner          = THIS_MODULE,
2178         .name           = "btrfs",
2179         .mount          = btrfs_mount_root,
2180         .kill_sb        = btrfs_kill_super,
2181         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2182 };
2183
2184 MODULE_ALIAS_FS("btrfs");
2185
2186 static int btrfs_control_open(struct inode *inode, struct file *file)
2187 {
2188         /*
2189          * The control file's private_data is used to hold the
2190          * transaction when it is started and is used to keep
2191          * track of whether a transaction is already in progress.
2192          */
2193         file->private_data = NULL;
2194         return 0;
2195 }
2196
2197 /*
2198  * used by btrfsctl to scan devices when no FS is mounted
2199  */
2200 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2201                                 unsigned long arg)
2202 {
2203         struct btrfs_ioctl_vol_args *vol;
2204         struct btrfs_fs_devices *fs_devices;
2205         int ret = -ENOTTY;
2206
2207         if (!capable(CAP_SYS_ADMIN))
2208                 return -EPERM;
2209
2210         vol = memdup_user((void __user *)arg, sizeof(*vol));
2211         if (IS_ERR(vol))
2212                 return PTR_ERR(vol);
2213
2214         switch (cmd) {
2215         case BTRFS_IOC_SCAN_DEV:
2216                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2217                                             &btrfs_root_fs_type, &fs_devices);
2218                 break;
2219         case BTRFS_IOC_DEVICES_READY:
2220                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2221                                             &btrfs_root_fs_type, &fs_devices);
2222                 if (ret)
2223                         break;
2224                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2225                 break;
2226         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2227                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2228                 break;
2229         }
2230
2231         kfree(vol);
2232         return ret;
2233 }
2234
2235 static int btrfs_freeze(struct super_block *sb)
2236 {
2237         struct btrfs_trans_handle *trans;
2238         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2239         struct btrfs_root *root = fs_info->tree_root;
2240
2241         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2242         /*
2243          * We don't need a barrier here, we'll wait for any transaction that
2244          * could be in progress on other threads (and do delayed iputs that
2245          * we want to avoid on a frozen filesystem), or do the commit
2246          * ourselves.
2247          */
2248         trans = btrfs_attach_transaction_barrier(root);
2249         if (IS_ERR(trans)) {
2250                 /* no transaction, don't bother */
2251                 if (PTR_ERR(trans) == -ENOENT)
2252                         return 0;
2253                 return PTR_ERR(trans);
2254         }
2255         return btrfs_commit_transaction(trans);
2256 }
2257
2258 static int btrfs_unfreeze(struct super_block *sb)
2259 {
2260         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2261
2262         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2263         return 0;
2264 }
2265
2266 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2267 {
2268         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2269         struct btrfs_fs_devices *cur_devices;
2270         struct btrfs_device *dev, *first_dev = NULL;
2271         struct list_head *head;
2272         struct rcu_string *name;
2273
2274         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2275         cur_devices = fs_info->fs_devices;
2276         while (cur_devices) {
2277                 head = &cur_devices->devices;
2278                 list_for_each_entry(dev, head, dev_list) {
2279                         if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2280                                 continue;
2281                         if (!dev->name)
2282                                 continue;
2283                         if (!first_dev || dev->devid < first_dev->devid)
2284                                 first_dev = dev;
2285                 }
2286                 cur_devices = cur_devices->seed;
2287         }
2288
2289         if (first_dev) {
2290                 rcu_read_lock();
2291                 name = rcu_dereference(first_dev->name);
2292                 seq_escape(m, name->str, " \t\n\\");
2293                 rcu_read_unlock();
2294         } else {
2295                 WARN_ON(1);
2296         }
2297         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2298         return 0;
2299 }
2300
2301 static const struct super_operations btrfs_super_ops = {
2302         .drop_inode     = btrfs_drop_inode,
2303         .evict_inode    = btrfs_evict_inode,
2304         .put_super      = btrfs_put_super,
2305         .sync_fs        = btrfs_sync_fs,
2306         .show_options   = btrfs_show_options,
2307         .show_devname   = btrfs_show_devname,
2308         .write_inode    = btrfs_write_inode,
2309         .alloc_inode    = btrfs_alloc_inode,
2310         .destroy_inode  = btrfs_destroy_inode,
2311         .statfs         = btrfs_statfs,
2312         .remount_fs     = btrfs_remount,
2313         .freeze_fs      = btrfs_freeze,
2314         .unfreeze_fs    = btrfs_unfreeze,
2315 };
2316
2317 static const struct file_operations btrfs_ctl_fops = {
2318         .open = btrfs_control_open,
2319         .unlocked_ioctl  = btrfs_control_ioctl,
2320         .compat_ioctl = btrfs_control_ioctl,
2321         .owner   = THIS_MODULE,
2322         .llseek = noop_llseek,
2323 };
2324
2325 static struct miscdevice btrfs_misc = {
2326         .minor          = BTRFS_MINOR,
2327         .name           = "btrfs-control",
2328         .fops           = &btrfs_ctl_fops
2329 };
2330
2331 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2332 MODULE_ALIAS("devname:btrfs-control");
2333
2334 static int __init btrfs_interface_init(void)
2335 {
2336         return misc_register(&btrfs_misc);
2337 }
2338
2339 static void btrfs_interface_exit(void)
2340 {
2341         misc_deregister(&btrfs_misc);
2342 }
2343
2344 static void __init btrfs_print_mod_info(void)
2345 {
2346         pr_info("Btrfs loaded, crc32c=%s"
2347 #ifdef CONFIG_BTRFS_DEBUG
2348                         ", debug=on"
2349 #endif
2350 #ifdef CONFIG_BTRFS_ASSERT
2351                         ", assert=on"
2352 #endif
2353 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2354                         ", integrity-checker=on"
2355 #endif
2356 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2357                         ", ref-verify=on"
2358 #endif
2359                         "\n",
2360                         btrfs_crc32c_impl());
2361 }
2362
2363 static int __init init_btrfs_fs(void)
2364 {
2365         int err;
2366
2367         err = btrfs_hash_init();
2368         if (err)
2369                 return err;
2370
2371         btrfs_props_init();
2372
2373         err = btrfs_init_sysfs();
2374         if (err)
2375                 goto free_hash;
2376
2377         btrfs_init_compress();
2378
2379         err = btrfs_init_cachep();
2380         if (err)
2381                 goto free_compress;
2382
2383         err = extent_io_init();
2384         if (err)
2385                 goto free_cachep;
2386
2387         err = extent_map_init();
2388         if (err)
2389                 goto free_extent_io;
2390
2391         err = ordered_data_init();
2392         if (err)
2393                 goto free_extent_map;
2394
2395         err = btrfs_delayed_inode_init();
2396         if (err)
2397                 goto free_ordered_data;
2398
2399         err = btrfs_auto_defrag_init();
2400         if (err)
2401                 goto free_delayed_inode;
2402
2403         err = btrfs_delayed_ref_init();
2404         if (err)
2405                 goto free_auto_defrag;
2406
2407         err = btrfs_prelim_ref_init();
2408         if (err)
2409                 goto free_delayed_ref;
2410
2411         err = btrfs_end_io_wq_init();
2412         if (err)
2413                 goto free_prelim_ref;
2414
2415         err = btrfs_interface_init();
2416         if (err)
2417                 goto free_end_io_wq;
2418
2419         btrfs_init_lockdep();
2420
2421         btrfs_print_mod_info();
2422
2423         err = btrfs_run_sanity_tests();
2424         if (err)
2425                 goto unregister_ioctl;
2426
2427         err = register_filesystem(&btrfs_fs_type);
2428         if (err)
2429                 goto unregister_ioctl;
2430
2431         return 0;
2432
2433 unregister_ioctl:
2434         btrfs_interface_exit();
2435 free_end_io_wq:
2436         btrfs_end_io_wq_exit();
2437 free_prelim_ref:
2438         btrfs_prelim_ref_exit();
2439 free_delayed_ref:
2440         btrfs_delayed_ref_exit();
2441 free_auto_defrag:
2442         btrfs_auto_defrag_exit();
2443 free_delayed_inode:
2444         btrfs_delayed_inode_exit();
2445 free_ordered_data:
2446         ordered_data_exit();
2447 free_extent_map:
2448         extent_map_exit();
2449 free_extent_io:
2450         extent_io_exit();
2451 free_cachep:
2452         btrfs_destroy_cachep();
2453 free_compress:
2454         btrfs_exit_compress();
2455         btrfs_exit_sysfs();
2456 free_hash:
2457         btrfs_hash_exit();
2458         return err;
2459 }
2460
2461 static void __exit exit_btrfs_fs(void)
2462 {
2463         btrfs_destroy_cachep();
2464         btrfs_delayed_ref_exit();
2465         btrfs_auto_defrag_exit();
2466         btrfs_delayed_inode_exit();
2467         btrfs_prelim_ref_exit();
2468         ordered_data_exit();
2469         extent_map_exit();
2470         extent_io_exit();
2471         btrfs_interface_exit();
2472         btrfs_end_io_wq_exit();
2473         unregister_filesystem(&btrfs_fs_type);
2474         btrfs_exit_sysfs();
2475         btrfs_cleanup_fs_uuids();
2476         btrfs_exit_compress();
2477         btrfs_hash_exit();
2478 }
2479
2480 late_initcall(init_btrfs_fs);
2481 module_exit(exit_btrfs_fs)
2482
2483 MODULE_LICENSE("GPL");