Btrfs: create the uuid tree on remount rw
[cascardo/linux.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 "compat.h"
46 #include "delayed-inode.h"
47 #include "ctree.h"
48 #include "disk-io.h"
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "xattr.h"
53 #include "volumes.h"
54 #include "export.h"
55 #include "compression.h"
56 #include "rcu-string.h"
57 #include "dev-replace.h"
58 #include "free-space-cache.h"
59 #include "backref.h"
60 #include "tests/btrfs-tests.h"
61
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/btrfs.h>
64
65 static const struct super_operations btrfs_super_ops;
66 static struct file_system_type btrfs_fs_type;
67
68 static const char *btrfs_decode_error(int errno)
69 {
70         char *errstr = "unknown";
71
72         switch (errno) {
73         case -EIO:
74                 errstr = "IO failure";
75                 break;
76         case -ENOMEM:
77                 errstr = "Out of memory";
78                 break;
79         case -EROFS:
80                 errstr = "Readonly filesystem";
81                 break;
82         case -EEXIST:
83                 errstr = "Object already exists";
84                 break;
85         case -ENOSPC:
86                 errstr = "No space left";
87                 break;
88         case -ENOENT:
89                 errstr = "No such entry";
90                 break;
91         }
92
93         return errstr;
94 }
95
96 static void save_error_info(struct btrfs_fs_info *fs_info)
97 {
98         /*
99          * today we only save the error info into ram.  Long term we'll
100          * also send it down to the disk
101          */
102         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
103 }
104
105 /* btrfs handle error by forcing the filesystem readonly */
106 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
107 {
108         struct super_block *sb = fs_info->sb;
109
110         if (sb->s_flags & MS_RDONLY)
111                 return;
112
113         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
114                 sb->s_flags |= MS_RDONLY;
115                 btrfs_info(fs_info, "forced readonly");
116                 /*
117                  * Note that a running device replace operation is not
118                  * canceled here although there is no way to update
119                  * the progress. It would add the risk of a deadlock,
120                  * therefore the canceling is ommited. The only penalty
121                  * is that some I/O remains active until the procedure
122                  * completes. The next time when the filesystem is
123                  * mounted writeable again, the device replace
124                  * operation continues.
125                  */
126         }
127 }
128
129 #ifdef CONFIG_PRINTK
130 /*
131  * __btrfs_std_error decodes expected errors from the caller and
132  * invokes the approciate error response.
133  */
134 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
135                        unsigned int line, int errno, const char *fmt, ...)
136 {
137         struct super_block *sb = fs_info->sb;
138         const char *errstr;
139
140         /*
141          * Special case: if the error is EROFS, and we're already
142          * under MS_RDONLY, then it is safe here.
143          */
144         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
145                 return;
146
147         errstr = btrfs_decode_error(errno);
148         if (fmt) {
149                 struct va_format vaf;
150                 va_list args;
151
152                 va_start(args, fmt);
153                 vaf.fmt = fmt;
154                 vaf.va = &args;
155
156                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s (%pV)\n",
157                         sb->s_id, function, line, errno, errstr, &vaf);
158                 va_end(args);
159         } else {
160                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s\n",
161                         sb->s_id, function, line, errno, errstr);
162         }
163
164         /* Don't go through full error handling during mount */
165         save_error_info(fs_info);
166         if (sb->s_flags & MS_BORN)
167                 btrfs_handle_error(fs_info);
168 }
169
170 static const char * const logtypes[] = {
171         "emergency",
172         "alert",
173         "critical",
174         "error",
175         "warning",
176         "notice",
177         "info",
178         "debug",
179 };
180
181 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
182 {
183         struct super_block *sb = fs_info->sb;
184         char lvl[4];
185         struct va_format vaf;
186         va_list args;
187         const char *type = logtypes[4];
188         int kern_level;
189
190         va_start(args, fmt);
191
192         kern_level = printk_get_level(fmt);
193         if (kern_level) {
194                 size_t size = printk_skip_level(fmt) - fmt;
195                 memcpy(lvl, fmt,  size);
196                 lvl[size] = '\0';
197                 fmt += size;
198                 type = logtypes[kern_level - '0'];
199         } else
200                 *lvl = '\0';
201
202         vaf.fmt = fmt;
203         vaf.va = &args;
204
205         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
206
207         va_end(args);
208 }
209
210 #else
211
212 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
213                        unsigned int line, int errno, const char *fmt, ...)
214 {
215         struct super_block *sb = fs_info->sb;
216
217         /*
218          * Special case: if the error is EROFS, and we're already
219          * under MS_RDONLY, then it is safe here.
220          */
221         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
222                 return;
223
224         /* Don't go through full error handling during mount */
225         if (sb->s_flags & MS_BORN) {
226                 save_error_info(fs_info);
227                 btrfs_handle_error(fs_info);
228         }
229 }
230 #endif
231
232 /*
233  * We only mark the transaction aborted and then set the file system read-only.
234  * This will prevent new transactions from starting or trying to join this
235  * one.
236  *
237  * This means that error recovery at the call site is limited to freeing
238  * any local memory allocations and passing the error code up without
239  * further cleanup. The transaction should complete as it normally would
240  * in the call path but will return -EIO.
241  *
242  * We'll complete the cleanup in btrfs_end_transaction and
243  * btrfs_commit_transaction.
244  */
245 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
246                                struct btrfs_root *root, const char *function,
247                                unsigned int line, int errno)
248 {
249         /*
250          * Report first abort since mount
251          */
252         if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,
253                                 &root->fs_info->fs_state)) {
254                 WARN(1, KERN_DEBUG "btrfs: Transaction aborted (error %d)\n",
255                                 errno);
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->blocks_used) {
261                 const char *errstr;
262
263                 errstr = btrfs_decode_error(errno);
264                 btrfs_warn(root->fs_info,
265                            "%s:%d: Aborting unused transaction(%s).",
266                            function, line, errstr);
267                 return;
268         }
269         ACCESS_ONCE(trans->transaction->aborted) = errno;
270         /* Wake up anybody who may be waiting on this transaction */
271         wake_up(&root->fs_info->transaction_wait);
272         wake_up(&root->fs_info->transaction_blocked_wait);
273         __btrfs_std_error(root->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 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
280                    unsigned int line, int errno, const char *fmt, ...)
281 {
282         char *s_id = "<unknown>";
283         const char *errstr;
284         struct va_format vaf = { .fmt = fmt };
285         va_list args;
286
287         if (fs_info)
288                 s_id = fs_info->sb->s_id;
289
290         va_start(args, fmt);
291         vaf.va = &args;
292
293         errstr = btrfs_decode_error(errno);
294         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
295                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
296                         s_id, function, line, &vaf, errno, errstr);
297
298         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
299                s_id, function, line, &vaf, errno, errstr);
300         va_end(args);
301         /* Caller calls BUG() */
302 }
303
304 static void btrfs_put_super(struct super_block *sb)
305 {
306         (void)close_ctree(btrfs_sb(sb)->tree_root);
307         /* FIXME: need to fix VFS to return error? */
308         /* AV: return it _where_?  ->put_super() can be triggered by any number
309          * of async events, up to and including delivery of SIGKILL to the
310          * last process that kept it busy.  Or segfault in the aforementioned
311          * process...  Whom would you report that to?
312          */
313 }
314
315 enum {
316         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
317         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
318         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
319         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
320         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
321         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
322         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
323         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
324         Opt_check_integrity, Opt_check_integrity_including_extent_data,
325         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
326         Opt_commit_interval,
327         Opt_err,
328 };
329
330 static match_table_t tokens = {
331         {Opt_degraded, "degraded"},
332         {Opt_subvol, "subvol=%s"},
333         {Opt_subvolid, "subvolid=%s"},
334         {Opt_device, "device=%s"},
335         {Opt_nodatasum, "nodatasum"},
336         {Opt_nodatacow, "nodatacow"},
337         {Opt_nobarrier, "nobarrier"},
338         {Opt_max_inline, "max_inline=%s"},
339         {Opt_alloc_start, "alloc_start=%s"},
340         {Opt_thread_pool, "thread_pool=%d"},
341         {Opt_compress, "compress"},
342         {Opt_compress_type, "compress=%s"},
343         {Opt_compress_force, "compress-force"},
344         {Opt_compress_force_type, "compress-force=%s"},
345         {Opt_ssd, "ssd"},
346         {Opt_ssd_spread, "ssd_spread"},
347         {Opt_nossd, "nossd"},
348         {Opt_noacl, "noacl"},
349         {Opt_notreelog, "notreelog"},
350         {Opt_flushoncommit, "flushoncommit"},
351         {Opt_ratio, "metadata_ratio=%d"},
352         {Opt_discard, "discard"},
353         {Opt_space_cache, "space_cache"},
354         {Opt_clear_cache, "clear_cache"},
355         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
356         {Opt_enospc_debug, "enospc_debug"},
357         {Opt_subvolrootid, "subvolrootid=%d"},
358         {Opt_defrag, "autodefrag"},
359         {Opt_inode_cache, "inode_cache"},
360         {Opt_no_space_cache, "nospace_cache"},
361         {Opt_recovery, "recovery"},
362         {Opt_skip_balance, "skip_balance"},
363         {Opt_check_integrity, "check_int"},
364         {Opt_check_integrity_including_extent_data, "check_int_data"},
365         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
366         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
367         {Opt_fatal_errors, "fatal_errors=%s"},
368         {Opt_commit_interval, "commit=%d"},
369         {Opt_err, NULL},
370 };
371
372 /*
373  * Regular mount options parser.  Everything that is needed only when
374  * reading in a new superblock is parsed here.
375  * XXX JDM: This needs to be cleaned up for remount.
376  */
377 int btrfs_parse_options(struct btrfs_root *root, char *options)
378 {
379         struct btrfs_fs_info *info = root->fs_info;
380         substring_t args[MAX_OPT_ARGS];
381         char *p, *num, *orig = NULL;
382         u64 cache_gen;
383         int intarg;
384         int ret = 0;
385         char *compress_type;
386         bool compress_force = false;
387
388         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
389         if (cache_gen)
390                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
391
392         if (!options)
393                 goto out;
394
395         /*
396          * strsep changes the string, duplicate it because parse_options
397          * gets called twice
398          */
399         options = kstrdup(options, GFP_NOFS);
400         if (!options)
401                 return -ENOMEM;
402
403         orig = options;
404
405         while ((p = strsep(&options, ",")) != NULL) {
406                 int token;
407                 if (!*p)
408                         continue;
409
410                 token = match_token(p, tokens, args);
411                 switch (token) {
412                 case Opt_degraded:
413                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
414                         btrfs_set_opt(info->mount_opt, DEGRADED);
415                         break;
416                 case Opt_subvol:
417                 case Opt_subvolid:
418                 case Opt_subvolrootid:
419                 case Opt_device:
420                         /*
421                          * These are parsed by btrfs_parse_early_options
422                          * and can be happily ignored here.
423                          */
424                         break;
425                 case Opt_nodatasum:
426                         printk(KERN_INFO "btrfs: setting nodatasum\n");
427                         btrfs_set_opt(info->mount_opt, NODATASUM);
428                         break;
429                 case Opt_nodatacow:
430                         if (!btrfs_test_opt(root, COMPRESS) ||
431                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
432                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
433                         } else {
434                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
435                         }
436                         info->compress_type = BTRFS_COMPRESS_NONE;
437                         btrfs_clear_opt(info->mount_opt, COMPRESS);
438                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
439                         btrfs_set_opt(info->mount_opt, NODATACOW);
440                         btrfs_set_opt(info->mount_opt, NODATASUM);
441                         break;
442                 case Opt_compress_force:
443                 case Opt_compress_force_type:
444                         compress_force = true;
445                         /* Fallthrough */
446                 case Opt_compress:
447                 case Opt_compress_type:
448                         if (token == Opt_compress ||
449                             token == Opt_compress_force ||
450                             strcmp(args[0].from, "zlib") == 0) {
451                                 compress_type = "zlib";
452                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
453                                 btrfs_set_opt(info->mount_opt, COMPRESS);
454                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
455                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
456                         } else if (strcmp(args[0].from, "lzo") == 0) {
457                                 compress_type = "lzo";
458                                 info->compress_type = BTRFS_COMPRESS_LZO;
459                                 btrfs_set_opt(info->mount_opt, COMPRESS);
460                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
461                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
462                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
463                         } else if (strncmp(args[0].from, "no", 2) == 0) {
464                                 compress_type = "no";
465                                 info->compress_type = BTRFS_COMPRESS_NONE;
466                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
467                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
468                                 compress_force = false;
469                         } else {
470                                 ret = -EINVAL;
471                                 goto out;
472                         }
473
474                         if (compress_force) {
475                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
476                                 pr_info("btrfs: force %s compression\n",
477                                         compress_type);
478                         } else
479                                 pr_info("btrfs: use %s compression\n",
480                                         compress_type);
481                         break;
482                 case Opt_ssd:
483                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
484                         btrfs_set_opt(info->mount_opt, SSD);
485                         break;
486                 case Opt_ssd_spread:
487                         printk(KERN_INFO "btrfs: use spread ssd "
488                                "allocation scheme\n");
489                         btrfs_set_opt(info->mount_opt, SSD);
490                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
491                         break;
492                 case Opt_nossd:
493                         printk(KERN_INFO "btrfs: not using ssd allocation "
494                                "scheme\n");
495                         btrfs_set_opt(info->mount_opt, NOSSD);
496                         btrfs_clear_opt(info->mount_opt, SSD);
497                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
498                         break;
499                 case Opt_nobarrier:
500                         printk(KERN_INFO "btrfs: turning off barriers\n");
501                         btrfs_set_opt(info->mount_opt, NOBARRIER);
502                         break;
503                 case Opt_thread_pool:
504                         ret = match_int(&args[0], &intarg);
505                         if (ret) {
506                                 goto out;
507                         } else if (intarg > 0) {
508                                 info->thread_pool_size = intarg;
509                         } else {
510                                 ret = -EINVAL;
511                                 goto out;
512                         }
513                         break;
514                 case Opt_max_inline:
515                         num = match_strdup(&args[0]);
516                         if (num) {
517                                 info->max_inline = memparse(num, NULL);
518                                 kfree(num);
519
520                                 if (info->max_inline) {
521                                         info->max_inline = max_t(u64,
522                                                 info->max_inline,
523                                                 root->sectorsize);
524                                 }
525                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
526                                         info->max_inline);
527                         } else {
528                                 ret = -ENOMEM;
529                                 goto out;
530                         }
531                         break;
532                 case Opt_alloc_start:
533                         num = match_strdup(&args[0]);
534                         if (num) {
535                                 mutex_lock(&info->chunk_mutex);
536                                 info->alloc_start = memparse(num, NULL);
537                                 mutex_unlock(&info->chunk_mutex);
538                                 kfree(num);
539                                 printk(KERN_INFO
540                                         "btrfs: allocations start at %llu\n",
541                                         info->alloc_start);
542                         } else {
543                                 ret = -ENOMEM;
544                                 goto out;
545                         }
546                         break;
547                 case Opt_noacl:
548                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
549                         break;
550                 case Opt_notreelog:
551                         printk(KERN_INFO "btrfs: disabling tree log\n");
552                         btrfs_set_opt(info->mount_opt, NOTREELOG);
553                         break;
554                 case Opt_flushoncommit:
555                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
556                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
557                         break;
558                 case Opt_ratio:
559                         ret = match_int(&args[0], &intarg);
560                         if (ret) {
561                                 goto out;
562                         } else if (intarg >= 0) {
563                                 info->metadata_ratio = intarg;
564                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
565                                        info->metadata_ratio);
566                         } else {
567                                 ret = -EINVAL;
568                                 goto out;
569                         }
570                         break;
571                 case Opt_discard:
572                         btrfs_set_opt(info->mount_opt, DISCARD);
573                         break;
574                 case Opt_space_cache:
575                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
576                         break;
577                 case Opt_rescan_uuid_tree:
578                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
579                         break;
580                 case Opt_no_space_cache:
581                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
582                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
583                         break;
584                 case Opt_inode_cache:
585                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
586                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
587                         break;
588                 case Opt_clear_cache:
589                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
590                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
591                         break;
592                 case Opt_user_subvol_rm_allowed:
593                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
594                         break;
595                 case Opt_enospc_debug:
596                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
597                         break;
598                 case Opt_defrag:
599                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
600                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
601                         break;
602                 case Opt_recovery:
603                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
604                         btrfs_set_opt(info->mount_opt, RECOVERY);
605                         break;
606                 case Opt_skip_balance:
607                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
608                         break;
609 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
610                 case Opt_check_integrity_including_extent_data:
611                         printk(KERN_INFO "btrfs: enabling check integrity"
612                                " including extent data\n");
613                         btrfs_set_opt(info->mount_opt,
614                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
615                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
616                         break;
617                 case Opt_check_integrity:
618                         printk(KERN_INFO "btrfs: enabling check integrity\n");
619                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
620                         break;
621                 case Opt_check_integrity_print_mask:
622                         ret = match_int(&args[0], &intarg);
623                         if (ret) {
624                                 goto out;
625                         } else if (intarg >= 0) {
626                                 info->check_integrity_print_mask = intarg;
627                                 printk(KERN_INFO "btrfs:"
628                                        " check_integrity_print_mask 0x%x\n",
629                                        info->check_integrity_print_mask);
630                         } else {
631                                 ret = -EINVAL;
632                                 goto out;
633                         }
634                         break;
635 #else
636                 case Opt_check_integrity_including_extent_data:
637                 case Opt_check_integrity:
638                 case Opt_check_integrity_print_mask:
639                         printk(KERN_ERR "btrfs: support for check_integrity*"
640                                " not compiled in!\n");
641                         ret = -EINVAL;
642                         goto out;
643 #endif
644                 case Opt_fatal_errors:
645                         if (strcmp(args[0].from, "panic") == 0)
646                                 btrfs_set_opt(info->mount_opt,
647                                               PANIC_ON_FATAL_ERROR);
648                         else if (strcmp(args[0].from, "bug") == 0)
649                                 btrfs_clear_opt(info->mount_opt,
650                                               PANIC_ON_FATAL_ERROR);
651                         else {
652                                 ret = -EINVAL;
653                                 goto out;
654                         }
655                         break;
656                 case Opt_commit_interval:
657                         intarg = 0;
658                         ret = match_int(&args[0], &intarg);
659                         if (ret < 0) {
660                                 printk(KERN_ERR
661                                         "btrfs: invalid commit interval\n");
662                                 ret = -EINVAL;
663                                 goto out;
664                         }
665                         if (intarg > 0) {
666                                 if (intarg > 300) {
667                                         printk(KERN_WARNING
668                                             "btrfs: excessive commit interval %d\n",
669                                                         intarg);
670                                 }
671                                 info->commit_interval = intarg;
672                         } else {
673                                 printk(KERN_INFO
674                                     "btrfs: using default commit interval %ds\n",
675                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
676                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
677                         }
678                         break;
679                 case Opt_err:
680                         printk(KERN_INFO "btrfs: unrecognized mount option "
681                                "'%s'\n", p);
682                         ret = -EINVAL;
683                         goto out;
684                 default:
685                         break;
686                 }
687         }
688 out:
689         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
690                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
691         kfree(orig);
692         return ret;
693 }
694
695 /*
696  * Parse mount options that are required early in the mount process.
697  *
698  * All other options will be parsed on much later in the mount process and
699  * only when we need to allocate a new super block.
700  */
701 static int btrfs_parse_early_options(const char *options, fmode_t flags,
702                 void *holder, char **subvol_name, u64 *subvol_objectid,
703                 struct btrfs_fs_devices **fs_devices)
704 {
705         substring_t args[MAX_OPT_ARGS];
706         char *device_name, *opts, *orig, *p;
707         char *num = NULL;
708         int error = 0;
709
710         if (!options)
711                 return 0;
712
713         /*
714          * strsep changes the string, duplicate it because parse_options
715          * gets called twice
716          */
717         opts = kstrdup(options, GFP_KERNEL);
718         if (!opts)
719                 return -ENOMEM;
720         orig = opts;
721
722         while ((p = strsep(&opts, ",")) != NULL) {
723                 int token;
724                 if (!*p)
725                         continue;
726
727                 token = match_token(p, tokens, args);
728                 switch (token) {
729                 case Opt_subvol:
730                         kfree(*subvol_name);
731                         *subvol_name = match_strdup(&args[0]);
732                         if (!*subvol_name) {
733                                 error = -ENOMEM;
734                                 goto out;
735                         }
736                         break;
737                 case Opt_subvolid:
738                         num = match_strdup(&args[0]);
739                         if (num) {
740                                 *subvol_objectid = memparse(num, NULL);
741                                 kfree(num);
742                                 /* we want the original fs_tree */
743                                 if (!*subvol_objectid)
744                                         *subvol_objectid =
745                                                 BTRFS_FS_TREE_OBJECTID;
746                         } else {
747                                 error = -EINVAL;
748                                 goto out;
749                         }
750                         break;
751                 case Opt_subvolrootid:
752                         printk(KERN_WARNING
753                                 "btrfs: 'subvolrootid' mount option is deprecated and has no effect\n");
754                         break;
755                 case Opt_device:
756                         device_name = match_strdup(&args[0]);
757                         if (!device_name) {
758                                 error = -ENOMEM;
759                                 goto out;
760                         }
761                         error = btrfs_scan_one_device(device_name,
762                                         flags, holder, fs_devices);
763                         kfree(device_name);
764                         if (error)
765                                 goto out;
766                         break;
767                 default:
768                         break;
769                 }
770         }
771
772 out:
773         kfree(orig);
774         return error;
775 }
776
777 static struct dentry *get_default_root(struct super_block *sb,
778                                        u64 subvol_objectid)
779 {
780         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
781         struct btrfs_root *root = fs_info->tree_root;
782         struct btrfs_root *new_root;
783         struct btrfs_dir_item *di;
784         struct btrfs_path *path;
785         struct btrfs_key location;
786         struct inode *inode;
787         u64 dir_id;
788         int new = 0;
789
790         /*
791          * We have a specific subvol we want to mount, just setup location and
792          * go look up the root.
793          */
794         if (subvol_objectid) {
795                 location.objectid = subvol_objectid;
796                 location.type = BTRFS_ROOT_ITEM_KEY;
797                 location.offset = (u64)-1;
798                 goto find_root;
799         }
800
801         path = btrfs_alloc_path();
802         if (!path)
803                 return ERR_PTR(-ENOMEM);
804         path->leave_spinning = 1;
805
806         /*
807          * Find the "default" dir item which points to the root item that we
808          * will mount by default if we haven't been given a specific subvolume
809          * to mount.
810          */
811         dir_id = btrfs_super_root_dir(fs_info->super_copy);
812         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
813         if (IS_ERR(di)) {
814                 btrfs_free_path(path);
815                 return ERR_CAST(di);
816         }
817         if (!di) {
818                 /*
819                  * Ok the default dir item isn't there.  This is weird since
820                  * it's always been there, but don't freak out, just try and
821                  * mount to root most subvolume.
822                  */
823                 btrfs_free_path(path);
824                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
825                 new_root = fs_info->fs_root;
826                 goto setup_root;
827         }
828
829         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
830         btrfs_free_path(path);
831
832 find_root:
833         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
834         if (IS_ERR(new_root))
835                 return ERR_CAST(new_root);
836
837         dir_id = btrfs_root_dirid(&new_root->root_item);
838 setup_root:
839         location.objectid = dir_id;
840         location.type = BTRFS_INODE_ITEM_KEY;
841         location.offset = 0;
842
843         inode = btrfs_iget(sb, &location, new_root, &new);
844         if (IS_ERR(inode))
845                 return ERR_CAST(inode);
846
847         /*
848          * If we're just mounting the root most subvol put the inode and return
849          * a reference to the dentry.  We will have already gotten a reference
850          * to the inode in btrfs_fill_super so we're good to go.
851          */
852         if (!new && sb->s_root->d_inode == inode) {
853                 iput(inode);
854                 return dget(sb->s_root);
855         }
856
857         return d_obtain_alias(inode);
858 }
859
860 static int btrfs_fill_super(struct super_block *sb,
861                             struct btrfs_fs_devices *fs_devices,
862                             void *data, int silent)
863 {
864         struct inode *inode;
865         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
866         struct btrfs_key key;
867         int err;
868
869         sb->s_maxbytes = MAX_LFS_FILESIZE;
870         sb->s_magic = BTRFS_SUPER_MAGIC;
871         sb->s_op = &btrfs_super_ops;
872         sb->s_d_op = &btrfs_dentry_operations;
873         sb->s_export_op = &btrfs_export_ops;
874         sb->s_xattr = btrfs_xattr_handlers;
875         sb->s_time_gran = 1;
876 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
877         sb->s_flags |= MS_POSIXACL;
878 #endif
879         sb->s_flags |= MS_I_VERSION;
880         err = open_ctree(sb, fs_devices, (char *)data);
881         if (err) {
882                 printk("btrfs: open_ctree failed\n");
883                 return err;
884         }
885
886         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
887         key.type = BTRFS_INODE_ITEM_KEY;
888         key.offset = 0;
889         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
890         if (IS_ERR(inode)) {
891                 err = PTR_ERR(inode);
892                 goto fail_close;
893         }
894
895         sb->s_root = d_make_root(inode);
896         if (!sb->s_root) {
897                 err = -ENOMEM;
898                 goto fail_close;
899         }
900
901         save_mount_options(sb, data);
902         cleancache_init_fs(sb);
903         sb->s_flags |= MS_ACTIVE;
904         return 0;
905
906 fail_close:
907         close_ctree(fs_info->tree_root);
908         return err;
909 }
910
911 int btrfs_sync_fs(struct super_block *sb, int wait)
912 {
913         struct btrfs_trans_handle *trans;
914         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
915         struct btrfs_root *root = fs_info->tree_root;
916
917         trace_btrfs_sync_fs(wait);
918
919         if (!wait) {
920                 filemap_flush(fs_info->btree_inode->i_mapping);
921                 return 0;
922         }
923
924         btrfs_wait_all_ordered_extents(fs_info);
925
926         trans = btrfs_attach_transaction_barrier(root);
927         if (IS_ERR(trans)) {
928                 /* no transaction, don't bother */
929                 if (PTR_ERR(trans) == -ENOENT)
930                         return 0;
931                 return PTR_ERR(trans);
932         }
933         return btrfs_commit_transaction(trans, root);
934 }
935
936 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
937 {
938         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
939         struct btrfs_root *root = info->tree_root;
940         char *compress_type;
941
942         if (btrfs_test_opt(root, DEGRADED))
943                 seq_puts(seq, ",degraded");
944         if (btrfs_test_opt(root, NODATASUM))
945                 seq_puts(seq, ",nodatasum");
946         if (btrfs_test_opt(root, NODATACOW))
947                 seq_puts(seq, ",nodatacow");
948         if (btrfs_test_opt(root, NOBARRIER))
949                 seq_puts(seq, ",nobarrier");
950         if (info->max_inline != 8192 * 1024)
951                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
952         if (info->alloc_start != 0)
953                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
954         if (info->thread_pool_size !=  min_t(unsigned long,
955                                              num_online_cpus() + 2, 8))
956                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
957         if (btrfs_test_opt(root, COMPRESS)) {
958                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
959                         compress_type = "zlib";
960                 else
961                         compress_type = "lzo";
962                 if (btrfs_test_opt(root, FORCE_COMPRESS))
963                         seq_printf(seq, ",compress-force=%s", compress_type);
964                 else
965                         seq_printf(seq, ",compress=%s", compress_type);
966         }
967         if (btrfs_test_opt(root, NOSSD))
968                 seq_puts(seq, ",nossd");
969         if (btrfs_test_opt(root, SSD_SPREAD))
970                 seq_puts(seq, ",ssd_spread");
971         else if (btrfs_test_opt(root, SSD))
972                 seq_puts(seq, ",ssd");
973         if (btrfs_test_opt(root, NOTREELOG))
974                 seq_puts(seq, ",notreelog");
975         if (btrfs_test_opt(root, FLUSHONCOMMIT))
976                 seq_puts(seq, ",flushoncommit");
977         if (btrfs_test_opt(root, DISCARD))
978                 seq_puts(seq, ",discard");
979         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
980                 seq_puts(seq, ",noacl");
981         if (btrfs_test_opt(root, SPACE_CACHE))
982                 seq_puts(seq, ",space_cache");
983         else
984                 seq_puts(seq, ",nospace_cache");
985         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
986                 seq_puts(seq, ",rescan_uuid_tree");
987         if (btrfs_test_opt(root, CLEAR_CACHE))
988                 seq_puts(seq, ",clear_cache");
989         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
990                 seq_puts(seq, ",user_subvol_rm_allowed");
991         if (btrfs_test_opt(root, ENOSPC_DEBUG))
992                 seq_puts(seq, ",enospc_debug");
993         if (btrfs_test_opt(root, AUTO_DEFRAG))
994                 seq_puts(seq, ",autodefrag");
995         if (btrfs_test_opt(root, INODE_MAP_CACHE))
996                 seq_puts(seq, ",inode_cache");
997         if (btrfs_test_opt(root, SKIP_BALANCE))
998                 seq_puts(seq, ",skip_balance");
999         if (btrfs_test_opt(root, RECOVERY))
1000                 seq_puts(seq, ",recovery");
1001 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1002         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1003                 seq_puts(seq, ",check_int_data");
1004         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1005                 seq_puts(seq, ",check_int");
1006         if (info->check_integrity_print_mask)
1007                 seq_printf(seq, ",check_int_print_mask=%d",
1008                                 info->check_integrity_print_mask);
1009 #endif
1010         if (info->metadata_ratio)
1011                 seq_printf(seq, ",metadata_ratio=%d",
1012                                 info->metadata_ratio);
1013         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1014                 seq_puts(seq, ",fatal_errors=panic");
1015         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1016                 seq_printf(seq, ",commit=%d", info->commit_interval);
1017         return 0;
1018 }
1019
1020 static int btrfs_test_super(struct super_block *s, void *data)
1021 {
1022         struct btrfs_fs_info *p = data;
1023         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1024
1025         return fs_info->fs_devices == p->fs_devices;
1026 }
1027
1028 static int btrfs_set_super(struct super_block *s, void *data)
1029 {
1030         int err = set_anon_super(s, data);
1031         if (!err)
1032                 s->s_fs_info = data;
1033         return err;
1034 }
1035
1036 /*
1037  * subvolumes are identified by ino 256
1038  */
1039 static inline int is_subvolume_inode(struct inode *inode)
1040 {
1041         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1042                 return 1;
1043         return 0;
1044 }
1045
1046 /*
1047  * This will strip out the subvol=%s argument for an argument string and add
1048  * subvolid=0 to make sure we get the actual tree root for path walking to the
1049  * subvol we want.
1050  */
1051 static char *setup_root_args(char *args)
1052 {
1053         unsigned len = strlen(args) + 2 + 1;
1054         char *src, *dst, *buf;
1055
1056         /*
1057          * We need the same args as before, but with this substitution:
1058          * s!subvol=[^,]+!subvolid=0!
1059          *
1060          * Since the replacement string is up to 2 bytes longer than the
1061          * original, allocate strlen(args) + 2 + 1 bytes.
1062          */
1063
1064         src = strstr(args, "subvol=");
1065         /* This shouldn't happen, but just in case.. */
1066         if (!src)
1067                 return NULL;
1068
1069         buf = dst = kmalloc(len, GFP_NOFS);
1070         if (!buf)
1071                 return NULL;
1072
1073         /*
1074          * If the subvol= arg is not at the start of the string,
1075          * copy whatever precedes it into buf.
1076          */
1077         if (src != args) {
1078                 *src++ = '\0';
1079                 strcpy(buf, args);
1080                 dst += strlen(args);
1081         }
1082
1083         strcpy(dst, "subvolid=0");
1084         dst += strlen("subvolid=0");
1085
1086         /*
1087          * If there is a "," after the original subvol=... string,
1088          * copy that suffix into our buffer.  Otherwise, we're done.
1089          */
1090         src = strchr(src, ',');
1091         if (src)
1092                 strcpy(dst, src);
1093
1094         return buf;
1095 }
1096
1097 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1098                                    const char *device_name, char *data)
1099 {
1100         struct dentry *root;
1101         struct vfsmount *mnt;
1102         char *newargs;
1103
1104         newargs = setup_root_args(data);
1105         if (!newargs)
1106                 return ERR_PTR(-ENOMEM);
1107         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1108                              newargs);
1109         kfree(newargs);
1110         if (IS_ERR(mnt))
1111                 return ERR_CAST(mnt);
1112
1113         root = mount_subtree(mnt, subvol_name);
1114
1115         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1116                 struct super_block *s = root->d_sb;
1117                 dput(root);
1118                 root = ERR_PTR(-EINVAL);
1119                 deactivate_locked_super(s);
1120                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1121                                 subvol_name);
1122         }
1123
1124         return root;
1125 }
1126
1127 /*
1128  * Find a superblock for the given device / mount point.
1129  *
1130  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1131  *        for multiple device setup.  Make sure to keep it in sync.
1132  */
1133 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1134                 const char *device_name, void *data)
1135 {
1136         struct block_device *bdev = NULL;
1137         struct super_block *s;
1138         struct dentry *root;
1139         struct btrfs_fs_devices *fs_devices = NULL;
1140         struct btrfs_fs_info *fs_info = NULL;
1141         fmode_t mode = FMODE_READ;
1142         char *subvol_name = NULL;
1143         u64 subvol_objectid = 0;
1144         int error = 0;
1145
1146         if (!(flags & MS_RDONLY))
1147                 mode |= FMODE_WRITE;
1148
1149         error = btrfs_parse_early_options(data, mode, fs_type,
1150                                           &subvol_name, &subvol_objectid,
1151                                           &fs_devices);
1152         if (error) {
1153                 kfree(subvol_name);
1154                 return ERR_PTR(error);
1155         }
1156
1157         if (subvol_name) {
1158                 root = mount_subvol(subvol_name, flags, device_name, data);
1159                 kfree(subvol_name);
1160                 return root;
1161         }
1162
1163         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1164         if (error)
1165                 return ERR_PTR(error);
1166
1167         /*
1168          * Setup a dummy root and fs_info for test/set super.  This is because
1169          * we don't actually fill this stuff out until open_ctree, but we need
1170          * it for searching for existing supers, so this lets us do that and
1171          * then open_ctree will properly initialize everything later.
1172          */
1173         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1174         if (!fs_info)
1175                 return ERR_PTR(-ENOMEM);
1176
1177         fs_info->fs_devices = fs_devices;
1178
1179         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1180         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1181         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1182                 error = -ENOMEM;
1183                 goto error_fs_info;
1184         }
1185
1186         error = btrfs_open_devices(fs_devices, mode, fs_type);
1187         if (error)
1188                 goto error_fs_info;
1189
1190         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1191                 error = -EACCES;
1192                 goto error_close_devices;
1193         }
1194
1195         bdev = fs_devices->latest_bdev;
1196         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1197                  fs_info);
1198         if (IS_ERR(s)) {
1199                 error = PTR_ERR(s);
1200                 goto error_close_devices;
1201         }
1202
1203         if (s->s_root) {
1204                 btrfs_close_devices(fs_devices);
1205                 free_fs_info(fs_info);
1206                 if ((flags ^ s->s_flags) & MS_RDONLY)
1207                         error = -EBUSY;
1208         } else {
1209                 char b[BDEVNAME_SIZE];
1210
1211                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1212                 btrfs_sb(s)->bdev_holder = fs_type;
1213                 error = btrfs_fill_super(s, fs_devices, data,
1214                                          flags & MS_SILENT ? 1 : 0);
1215         }
1216
1217         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1218         if (IS_ERR(root))
1219                 deactivate_locked_super(s);
1220
1221         return root;
1222
1223 error_close_devices:
1224         btrfs_close_devices(fs_devices);
1225 error_fs_info:
1226         free_fs_info(fs_info);
1227         return ERR_PTR(error);
1228 }
1229
1230 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1231 {
1232         spin_lock_irq(&workers->lock);
1233         workers->max_workers = new_limit;
1234         spin_unlock_irq(&workers->lock);
1235 }
1236
1237 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1238                                      int new_pool_size, int old_pool_size)
1239 {
1240         if (new_pool_size == old_pool_size)
1241                 return;
1242
1243         fs_info->thread_pool_size = new_pool_size;
1244
1245         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1246                old_pool_size, new_pool_size);
1247
1248         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1249         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1250         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1251         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1252         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1253         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1254         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1255         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1256         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1257         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1258         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1259         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1260         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1261         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1262                               new_pool_size);
1263 }
1264
1265 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1266 {
1267         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1268 }
1269
1270 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1271                                        unsigned long old_opts, int flags)
1272 {
1273         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1274             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1275              (flags & MS_RDONLY))) {
1276                 /* wait for any defraggers to finish */
1277                 wait_event(fs_info->transaction_wait,
1278                            (atomic_read(&fs_info->defrag_running) == 0));
1279                 if (flags & MS_RDONLY)
1280                         sync_filesystem(fs_info->sb);
1281         }
1282 }
1283
1284 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1285                                          unsigned long old_opts)
1286 {
1287         /*
1288          * We need cleanup all defragable inodes if the autodefragment is
1289          * close or the fs is R/O.
1290          */
1291         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1292             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1293              (fs_info->sb->s_flags & MS_RDONLY))) {
1294                 btrfs_cleanup_defrag_inodes(fs_info);
1295         }
1296
1297         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1298 }
1299
1300 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1301 {
1302         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1303         struct btrfs_root *root = fs_info->tree_root;
1304         unsigned old_flags = sb->s_flags;
1305         unsigned long old_opts = fs_info->mount_opt;
1306         unsigned long old_compress_type = fs_info->compress_type;
1307         u64 old_max_inline = fs_info->max_inline;
1308         u64 old_alloc_start = fs_info->alloc_start;
1309         int old_thread_pool_size = fs_info->thread_pool_size;
1310         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1311         int ret;
1312
1313         btrfs_remount_prepare(fs_info);
1314
1315         ret = btrfs_parse_options(root, data);
1316         if (ret) {
1317                 ret = -EINVAL;
1318                 goto restore;
1319         }
1320
1321         btrfs_remount_begin(fs_info, old_opts, *flags);
1322         btrfs_resize_thread_pool(fs_info,
1323                 fs_info->thread_pool_size, old_thread_pool_size);
1324
1325         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1326                 goto out;
1327
1328         if (*flags & MS_RDONLY) {
1329                 /*
1330                  * this also happens on 'umount -rf' or on shutdown, when
1331                  * the filesystem is busy.
1332                  */
1333                 sb->s_flags |= MS_RDONLY;
1334
1335                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1336                 btrfs_scrub_cancel(fs_info);
1337                 btrfs_pause_balance(fs_info);
1338
1339                 ret = btrfs_commit_super(root);
1340                 if (ret)
1341                         goto restore;
1342         } else {
1343                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1344                         btrfs_err(fs_info,
1345                                 "Remounting read-write after error is not allowed\n");
1346                         ret = -EINVAL;
1347                         goto restore;
1348                 }
1349                 if (fs_info->fs_devices->rw_devices == 0) {
1350                         ret = -EACCES;
1351                         goto restore;
1352                 }
1353
1354                 if (fs_info->fs_devices->missing_devices >
1355                      fs_info->num_tolerated_disk_barrier_failures &&
1356                     !(*flags & MS_RDONLY)) {
1357                         printk(KERN_WARNING
1358                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1359                         ret = -EACCES;
1360                         goto restore;
1361                 }
1362
1363                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1364                         ret = -EINVAL;
1365                         goto restore;
1366                 }
1367
1368                 ret = btrfs_cleanup_fs_roots(fs_info);
1369                 if (ret)
1370                         goto restore;
1371
1372                 /* recover relocation */
1373                 ret = btrfs_recover_relocation(root);
1374                 if (ret)
1375                         goto restore;
1376
1377                 ret = btrfs_resume_balance_async(fs_info);
1378                 if (ret)
1379                         goto restore;
1380
1381                 ret = btrfs_resume_dev_replace_async(fs_info);
1382                 if (ret) {
1383                         pr_warn("btrfs: failed to resume dev_replace\n");
1384                         goto restore;
1385                 }
1386
1387                 if (!fs_info->uuid_root) {
1388                         pr_info("btrfs: creating UUID tree\n");
1389                         ret = btrfs_create_uuid_tree(fs_info);
1390                         if (ret) {
1391                                 pr_warn("btrfs: failed to create the uuid tree"
1392                                         "%d\n", ret);
1393                                 goto restore;
1394                         }
1395                 }
1396                 sb->s_flags &= ~MS_RDONLY;
1397         }
1398 out:
1399         btrfs_remount_cleanup(fs_info, old_opts);
1400         return 0;
1401
1402 restore:
1403         /* We've hit an error - don't reset MS_RDONLY */
1404         if (sb->s_flags & MS_RDONLY)
1405                 old_flags |= MS_RDONLY;
1406         sb->s_flags = old_flags;
1407         fs_info->mount_opt = old_opts;
1408         fs_info->compress_type = old_compress_type;
1409         fs_info->max_inline = old_max_inline;
1410         mutex_lock(&fs_info->chunk_mutex);
1411         fs_info->alloc_start = old_alloc_start;
1412         mutex_unlock(&fs_info->chunk_mutex);
1413         btrfs_resize_thread_pool(fs_info,
1414                 old_thread_pool_size, fs_info->thread_pool_size);
1415         fs_info->metadata_ratio = old_metadata_ratio;
1416         btrfs_remount_cleanup(fs_info, old_opts);
1417         return ret;
1418 }
1419
1420 /* Used to sort the devices by max_avail(descending sort) */
1421 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1422                                        const void *dev_info2)
1423 {
1424         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1425             ((struct btrfs_device_info *)dev_info2)->max_avail)
1426                 return -1;
1427         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1428                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1429                 return 1;
1430         else
1431         return 0;
1432 }
1433
1434 /*
1435  * sort the devices by max_avail, in which max free extent size of each device
1436  * is stored.(Descending Sort)
1437  */
1438 static inline void btrfs_descending_sort_devices(
1439                                         struct btrfs_device_info *devices,
1440                                         size_t nr_devices)
1441 {
1442         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1443              btrfs_cmp_device_free_bytes, NULL);
1444 }
1445
1446 /*
1447  * The helper to calc the free space on the devices that can be used to store
1448  * file data.
1449  */
1450 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1451 {
1452         struct btrfs_fs_info *fs_info = root->fs_info;
1453         struct btrfs_device_info *devices_info;
1454         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1455         struct btrfs_device *device;
1456         u64 skip_space;
1457         u64 type;
1458         u64 avail_space;
1459         u64 used_space;
1460         u64 min_stripe_size;
1461         int min_stripes = 1, num_stripes = 1;
1462         int i = 0, nr_devices;
1463         int ret;
1464
1465         nr_devices = fs_info->fs_devices->open_devices;
1466         BUG_ON(!nr_devices);
1467
1468         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1469                                GFP_NOFS);
1470         if (!devices_info)
1471                 return -ENOMEM;
1472
1473         /* calc min stripe number for data space alloction */
1474         type = btrfs_get_alloc_profile(root, 1);
1475         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1476                 min_stripes = 2;
1477                 num_stripes = nr_devices;
1478         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1479                 min_stripes = 2;
1480                 num_stripes = 2;
1481         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1482                 min_stripes = 4;
1483                 num_stripes = 4;
1484         }
1485
1486         if (type & BTRFS_BLOCK_GROUP_DUP)
1487                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1488         else
1489                 min_stripe_size = BTRFS_STRIPE_LEN;
1490
1491         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1492                 if (!device->in_fs_metadata || !device->bdev ||
1493                     device->is_tgtdev_for_dev_replace)
1494                         continue;
1495
1496                 avail_space = device->total_bytes - device->bytes_used;
1497
1498                 /* align with stripe_len */
1499                 do_div(avail_space, BTRFS_STRIPE_LEN);
1500                 avail_space *= BTRFS_STRIPE_LEN;
1501
1502                 /*
1503                  * In order to avoid overwritting the superblock on the drive,
1504                  * btrfs starts at an offset of at least 1MB when doing chunk
1505                  * allocation.
1506                  */
1507                 skip_space = 1024 * 1024;
1508
1509                 /* user can set the offset in fs_info->alloc_start. */
1510                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1511                     device->total_bytes)
1512                         skip_space = max(fs_info->alloc_start, skip_space);
1513
1514                 /*
1515                  * btrfs can not use the free space in [0, skip_space - 1],
1516                  * we must subtract it from the total. In order to implement
1517                  * it, we account the used space in this range first.
1518                  */
1519                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1520                                                      &used_space);
1521                 if (ret) {
1522                         kfree(devices_info);
1523                         return ret;
1524                 }
1525
1526                 /* calc the free space in [0, skip_space - 1] */
1527                 skip_space -= used_space;
1528
1529                 /*
1530                  * we can use the free space in [0, skip_space - 1], subtract
1531                  * it from the total.
1532                  */
1533                 if (avail_space && avail_space >= skip_space)
1534                         avail_space -= skip_space;
1535                 else
1536                         avail_space = 0;
1537
1538                 if (avail_space < min_stripe_size)
1539                         continue;
1540
1541                 devices_info[i].dev = device;
1542                 devices_info[i].max_avail = avail_space;
1543
1544                 i++;
1545         }
1546
1547         nr_devices = i;
1548
1549         btrfs_descending_sort_devices(devices_info, nr_devices);
1550
1551         i = nr_devices - 1;
1552         avail_space = 0;
1553         while (nr_devices >= min_stripes) {
1554                 if (num_stripes > nr_devices)
1555                         num_stripes = nr_devices;
1556
1557                 if (devices_info[i].max_avail >= min_stripe_size) {
1558                         int j;
1559                         u64 alloc_size;
1560
1561                         avail_space += devices_info[i].max_avail * num_stripes;
1562                         alloc_size = devices_info[i].max_avail;
1563                         for (j = i + 1 - num_stripes; j <= i; j++)
1564                                 devices_info[j].max_avail -= alloc_size;
1565                 }
1566                 i--;
1567                 nr_devices--;
1568         }
1569
1570         kfree(devices_info);
1571         *free_bytes = avail_space;
1572         return 0;
1573 }
1574
1575 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1576 {
1577         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1578         struct btrfs_super_block *disk_super = fs_info->super_copy;
1579         struct list_head *head = &fs_info->space_info;
1580         struct btrfs_space_info *found;
1581         u64 total_used = 0;
1582         u64 total_free_data = 0;
1583         int bits = dentry->d_sb->s_blocksize_bits;
1584         __be32 *fsid = (__be32 *)fs_info->fsid;
1585         int ret;
1586
1587         /* holding chunk_muext to avoid allocating new chunks */
1588         mutex_lock(&fs_info->chunk_mutex);
1589         rcu_read_lock();
1590         list_for_each_entry_rcu(found, head, list) {
1591                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1592                         total_free_data += found->disk_total - found->disk_used;
1593                         total_free_data -=
1594                                 btrfs_account_ro_block_groups_free_space(found);
1595                 }
1596
1597                 total_used += found->disk_used;
1598         }
1599         rcu_read_unlock();
1600
1601         buf->f_namelen = BTRFS_NAME_LEN;
1602         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1603         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1604         buf->f_bsize = dentry->d_sb->s_blocksize;
1605         buf->f_type = BTRFS_SUPER_MAGIC;
1606         buf->f_bavail = total_free_data;
1607         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1608         if (ret) {
1609                 mutex_unlock(&fs_info->chunk_mutex);
1610                 return ret;
1611         }
1612         buf->f_bavail += total_free_data;
1613         buf->f_bavail = buf->f_bavail >> bits;
1614         mutex_unlock(&fs_info->chunk_mutex);
1615
1616         /* We treat it as constant endianness (it doesn't matter _which_)
1617            because we want the fsid to come out the same whether mounted
1618            on a big-endian or little-endian host */
1619         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1620         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1621         /* Mask in the root object ID too, to disambiguate subvols */
1622         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1623         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1624
1625         return 0;
1626 }
1627
1628 static void btrfs_kill_super(struct super_block *sb)
1629 {
1630         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1631         kill_anon_super(sb);
1632         free_fs_info(fs_info);
1633 }
1634
1635 static struct file_system_type btrfs_fs_type = {
1636         .owner          = THIS_MODULE,
1637         .name           = "btrfs",
1638         .mount          = btrfs_mount,
1639         .kill_sb        = btrfs_kill_super,
1640         .fs_flags       = FS_REQUIRES_DEV,
1641 };
1642 MODULE_ALIAS_FS("btrfs");
1643
1644 /*
1645  * used by btrfsctl to scan devices when no FS is mounted
1646  */
1647 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1648                                 unsigned long arg)
1649 {
1650         struct btrfs_ioctl_vol_args *vol;
1651         struct btrfs_fs_devices *fs_devices;
1652         int ret = -ENOTTY;
1653
1654         if (!capable(CAP_SYS_ADMIN))
1655                 return -EPERM;
1656
1657         vol = memdup_user((void __user *)arg, sizeof(*vol));
1658         if (IS_ERR(vol))
1659                 return PTR_ERR(vol);
1660
1661         switch (cmd) {
1662         case BTRFS_IOC_SCAN_DEV:
1663                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1664                                             &btrfs_fs_type, &fs_devices);
1665                 break;
1666         case BTRFS_IOC_DEVICES_READY:
1667                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1668                                             &btrfs_fs_type, &fs_devices);
1669                 if (ret)
1670                         break;
1671                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1672                 break;
1673         }
1674
1675         kfree(vol);
1676         return ret;
1677 }
1678
1679 static int btrfs_freeze(struct super_block *sb)
1680 {
1681         struct btrfs_trans_handle *trans;
1682         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1683
1684         trans = btrfs_attach_transaction_barrier(root);
1685         if (IS_ERR(trans)) {
1686                 /* no transaction, don't bother */
1687                 if (PTR_ERR(trans) == -ENOENT)
1688                         return 0;
1689                 return PTR_ERR(trans);
1690         }
1691         return btrfs_commit_transaction(trans, root);
1692 }
1693
1694 static int btrfs_unfreeze(struct super_block *sb)
1695 {
1696         return 0;
1697 }
1698
1699 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1700 {
1701         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1702         struct btrfs_fs_devices *cur_devices;
1703         struct btrfs_device *dev, *first_dev = NULL;
1704         struct list_head *head;
1705         struct rcu_string *name;
1706
1707         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1708         cur_devices = fs_info->fs_devices;
1709         while (cur_devices) {
1710                 head = &cur_devices->devices;
1711                 list_for_each_entry(dev, head, dev_list) {
1712                         if (dev->missing)
1713                                 continue;
1714                         if (!first_dev || dev->devid < first_dev->devid)
1715                                 first_dev = dev;
1716                 }
1717                 cur_devices = cur_devices->seed;
1718         }
1719
1720         if (first_dev) {
1721                 rcu_read_lock();
1722                 name = rcu_dereference(first_dev->name);
1723                 seq_escape(m, name->str, " \t\n\\");
1724                 rcu_read_unlock();
1725         } else {
1726                 WARN_ON(1);
1727         }
1728         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1729         return 0;
1730 }
1731
1732 static const struct super_operations btrfs_super_ops = {
1733         .drop_inode     = btrfs_drop_inode,
1734         .evict_inode    = btrfs_evict_inode,
1735         .put_super      = btrfs_put_super,
1736         .sync_fs        = btrfs_sync_fs,
1737         .show_options   = btrfs_show_options,
1738         .show_devname   = btrfs_show_devname,
1739         .write_inode    = btrfs_write_inode,
1740         .alloc_inode    = btrfs_alloc_inode,
1741         .destroy_inode  = btrfs_destroy_inode,
1742         .statfs         = btrfs_statfs,
1743         .remount_fs     = btrfs_remount,
1744         .freeze_fs      = btrfs_freeze,
1745         .unfreeze_fs    = btrfs_unfreeze,
1746 };
1747
1748 static const struct file_operations btrfs_ctl_fops = {
1749         .unlocked_ioctl  = btrfs_control_ioctl,
1750         .compat_ioctl = btrfs_control_ioctl,
1751         .owner   = THIS_MODULE,
1752         .llseek = noop_llseek,
1753 };
1754
1755 static struct miscdevice btrfs_misc = {
1756         .minor          = BTRFS_MINOR,
1757         .name           = "btrfs-control",
1758         .fops           = &btrfs_ctl_fops
1759 };
1760
1761 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1762 MODULE_ALIAS("devname:btrfs-control");
1763
1764 static int btrfs_interface_init(void)
1765 {
1766         return misc_register(&btrfs_misc);
1767 }
1768
1769 static void btrfs_interface_exit(void)
1770 {
1771         if (misc_deregister(&btrfs_misc) < 0)
1772                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1773 }
1774
1775 static void btrfs_print_info(void)
1776 {
1777         printk(KERN_INFO "Btrfs loaded"
1778 #ifdef CONFIG_BTRFS_DEBUG
1779                         ", debug=on"
1780 #endif
1781 #ifdef CONFIG_BTRFS_ASSERT
1782                         ", assert=on"
1783 #endif
1784 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1785                         ", integrity-checker=on"
1786 #endif
1787                         "\n");
1788 }
1789
1790 static int btrfs_run_sanity_tests(void)
1791 {
1792         return btrfs_test_free_space_cache();
1793 }
1794
1795 static int __init init_btrfs_fs(void)
1796 {
1797         int err;
1798
1799         err = btrfs_init_sysfs();
1800         if (err)
1801                 return err;
1802
1803         btrfs_init_compress();
1804
1805         err = btrfs_init_cachep();
1806         if (err)
1807                 goto free_compress;
1808
1809         err = extent_io_init();
1810         if (err)
1811                 goto free_cachep;
1812
1813         err = extent_map_init();
1814         if (err)
1815                 goto free_extent_io;
1816
1817         err = ordered_data_init();
1818         if (err)
1819                 goto free_extent_map;
1820
1821         err = btrfs_delayed_inode_init();
1822         if (err)
1823                 goto free_ordered_data;
1824
1825         err = btrfs_auto_defrag_init();
1826         if (err)
1827                 goto free_delayed_inode;
1828
1829         err = btrfs_delayed_ref_init();
1830         if (err)
1831                 goto free_auto_defrag;
1832
1833         err = btrfs_prelim_ref_init();
1834         if (err)
1835                 goto free_prelim_ref;
1836
1837         err = btrfs_interface_init();
1838         if (err)
1839                 goto free_delayed_ref;
1840
1841         btrfs_init_lockdep();
1842
1843         btrfs_print_info();
1844
1845         err = btrfs_run_sanity_tests();
1846         if (err)
1847                 goto unregister_ioctl;
1848
1849         err = register_filesystem(&btrfs_fs_type);
1850         if (err)
1851                 goto unregister_ioctl;
1852
1853         return 0;
1854
1855 unregister_ioctl:
1856         btrfs_interface_exit();
1857 free_prelim_ref:
1858         btrfs_prelim_ref_exit();
1859 free_delayed_ref:
1860         btrfs_delayed_ref_exit();
1861 free_auto_defrag:
1862         btrfs_auto_defrag_exit();
1863 free_delayed_inode:
1864         btrfs_delayed_inode_exit();
1865 free_ordered_data:
1866         ordered_data_exit();
1867 free_extent_map:
1868         extent_map_exit();
1869 free_extent_io:
1870         extent_io_exit();
1871 free_cachep:
1872         btrfs_destroy_cachep();
1873 free_compress:
1874         btrfs_exit_compress();
1875         btrfs_exit_sysfs();
1876         return err;
1877 }
1878
1879 static void __exit exit_btrfs_fs(void)
1880 {
1881         btrfs_destroy_cachep();
1882         btrfs_delayed_ref_exit();
1883         btrfs_auto_defrag_exit();
1884         btrfs_delayed_inode_exit();
1885         btrfs_prelim_ref_exit();
1886         ordered_data_exit();
1887         extent_map_exit();
1888         extent_io_exit();
1889         btrfs_interface_exit();
1890         unregister_filesystem(&btrfs_fs_type);
1891         btrfs_exit_sysfs();
1892         btrfs_cleanup_fs_uuids();
1893         btrfs_exit_compress();
1894 }
1895
1896 module_init(init_btrfs_fs)
1897 module_exit(exit_btrfs_fs)
1898
1899 MODULE_LICENSE("GPL");