1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6 #ifndef BTRFS_VOLUMES_H
7 #define BTRFS_VOLUMES_H
8
9 #include <linux/bio.h>
10 #include <linux/sort.h>
11 #include <linux/btrfs.h>
12 #include "async-thread.h"
13
14 #define BTRFS_MAX_DATA_CHUNK_SIZE (10ULL * SZ_1G)
15
16 extern struct mutex uuid_mutex;
17
18 #define BTRFS_STRIPE_LEN SZ_64K
19
20 struct buffer_head;
21 struct btrfs_pending_bios {
22 struct bio *head;
23 struct bio *tail;
24 };
25
26 /*
27 * Use sequence counter to get consistent device stat data on
28 * 32-bit processors.
29 */
30 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
31 #include <linux/seqlock.h>
32 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
33 #define btrfs_device_data_ordered_init(device) \
34 seqcount_init(&device->data_seqcount)
35 #else
36 #define btrfs_device_data_ordered_init(device) do { } while (0)
37 #endif
38
39 #define BTRFS_DEV_STATE_WRITEABLE (0)
40 #define BTRFS_DEV_STATE_IN_FS_METADATA (1)
41 #define BTRFS_DEV_STATE_MISSING (2)
42 #define BTRFS_DEV_STATE_REPLACE_TGT (3)
43 #define BTRFS_DEV_STATE_FLUSH_SENT (4)
44
45 struct btrfs_device {
46 struct list_head dev_list;
47 struct list_head dev_alloc_list;
48 struct btrfs_fs_devices *fs_devices;
49 struct btrfs_fs_info *fs_info;
50
51 struct rcu_string *name;
52
53 u64 generation;
54
55 spinlock_t io_lock ____cacheline_aligned;
56 int running_pending;
57 /* regular prio bios */
58 struct btrfs_pending_bios pending_bios;
59 /* sync bios */
60 struct btrfs_pending_bios pending_sync_bios;
61
62 struct block_device *bdev;
63
64 /* the mode sent to blkdev_get */
65 fmode_t mode;
66
67 unsigned long dev_state;
68 blk_status_t last_flush_error;
69 int flush_bio_sent;
70
71 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
72 seqcount_t data_seqcount;
73 #endif
74
75 /* the internal btrfs device id */
76 u64 devid;
77
78 /* size of the device in memory */
79 u64 total_bytes;
80
81 /* size of the device on disk */
82 u64 disk_total_bytes;
83
84 /* bytes used */
85 u64 bytes_used;
86
87 /* optimal io alignment for this device */
88 u32 io_align;
89
90 /* optimal io width for this device */
91 u32 io_width;
92 /* type and info about this device */
93 u64 type;
94
95 /* minimal io size for this device */
96 u32 sector_size;
97
98 /* physical drive uuid (or lvm uuid) */
99 u8 uuid[BTRFS_UUID_SIZE];
100
101 /*
102 * size of the device on the current transaction
103 *
104 * This variant is update when committing the transaction,
105 * and protected by device_list_mutex
106 */
107 u64 commit_total_bytes;
108
109 /* bytes used on the current transaction */
110 u64 commit_bytes_used;
111 /*
112 * used to manage the device which is resized
113 *
114 * It is protected by chunk_lock.
115 */
116 struct list_head resized_list;
117
118 /* for sending down flush barriers */
119 struct bio *flush_bio;
120 struct completion flush_wait;
121
122 /* per-device scrub information */
123 struct scrub_ctx *scrub_ctx;
124
125 struct btrfs_work work;
126 struct rcu_head rcu;
127
128 /* readahead state */
129 atomic_t reada_in_flight;
130 u64 reada_next;
131 struct reada_zone *reada_curr_zone;
132 struct radix_tree_root reada_zones;
133 struct radix_tree_root reada_extents;
134
135 /* disk I/O failure stats. For detailed description refer to
136 * enum btrfs_dev_stat_values in ioctl.h */
137 int dev_stats_valid;
138
139 /* Counter to record the change of device stats */
140 atomic_t dev_stats_ccnt;
141 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
142 };
143
144 /*
145 * If we read those variants at the context of their own lock, we needn't
146 * use the following helpers, reading them directly is safe.
147 */
148 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
149 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
150 static inline u64 \
151 btrfs_device_get_##name(const struct btrfs_device *dev) \
152 { \
153 u64 size; \
154 unsigned int seq; \
155 \
156 do { \
157 seq = read_seqcount_begin(&dev->data_seqcount); \
158 size = dev->name; \
159 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
160 return size; \
161 } \
162 \
163 static inline void \
164 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
165 { \
166 preempt_disable(); \
167 write_seqcount_begin(&dev->data_seqcount); \
168 dev->name = size; \
169 write_seqcount_end(&dev->data_seqcount); \
170 preempt_enable(); \
171 }
172 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
173 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
174 static inline u64 \
175 btrfs_device_get_##name(const struct btrfs_device *dev) \
176 { \
177 u64 size; \
178 \
179 preempt_disable(); \
180 size = dev->name; \
181 preempt_enable(); \
182 return size; \
183 } \
184 \
185 static inline void \
186 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
187 { \
188 preempt_disable(); \
189 dev->name = size; \
190 preempt_enable(); \
191 }
192 #else
193 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
194 static inline u64 \
195 btrfs_device_get_##name(const struct btrfs_device *dev) \
196 { \
197 return dev->name; \
198 } \
199 \
200 static inline void \
201 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
202 { \
203 dev->name = size; \
204 }
205 #endif
206
207 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
208 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
209 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
210
211 struct btrfs_fs_devices {
212 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
213 struct list_head fs_list;
214
215 u64 num_devices;
216 u64 open_devices;
217 u64 rw_devices;
218 u64 missing_devices;
219 u64 total_rw_bytes;
220 u64 total_devices;
221 struct block_device *latest_bdev;
222
223 /* all of the devices in the FS, protected by a mutex
224 * so we can safely walk it to write out the supers without
225 * worrying about add/remove by the multi-device code.
226 * Scrubbing super can kick off supers writing by holding
227 * this mutex lock.
228 */
229 struct mutex device_list_mutex;
230 struct list_head devices;
231
232 struct list_head resized_devices;
233 /* devices not currently being allocated */
234 struct list_head alloc_list;
235
236 struct btrfs_fs_devices *seed;
237 int seeding;
238
239 int opened;
240
241 /* set when we find or add a device that doesn't have the
242 * nonrot flag set
243 */
244 int rotating;
245
246 struct btrfs_fs_info *fs_info;
247 /* sysfs kobjects */
248 struct kobject fsid_kobj;
249 struct kobject *device_dir_kobj;
250 struct completion kobj_unregister;
251 };
252
253 #define BTRFS_BIO_INLINE_CSUM_SIZE 64
254
255 /*
256 * we need the mirror number and stripe index to be passed around
257 * the call chain while we are processing end_io (especially errors).
258 * Really, what we need is a btrfs_bio structure that has this info
259 * and is properly sized with its stripe array, but we're not there
260 * quite yet. We have our own btrfs bioset, and all of the bios
261 * we allocate are actually btrfs_io_bios. We'll cram as much of
262 * struct btrfs_bio as we can into this over time.
263 */
264 typedef void (btrfs_io_bio_end_io_t) (struct btrfs_io_bio *bio, int err);
265 struct btrfs_io_bio {
266 unsigned int mirror_num;
267 unsigned int stripe_index;
268 u64 logical;
269 u8 *csum;
270 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
271 u8 *csum_allocated;
272 btrfs_io_bio_end_io_t *end_io;
273 struct bvec_iter iter;
274 /*
275 * This member must come last, bio_alloc_bioset will allocate enough
276 * bytes for entire btrfs_io_bio but relies on bio being last.
277 */
278 struct bio bio;
279 };
280
btrfs_io_bio(struct bio * bio)281 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
282 {
283 return container_of(bio, struct btrfs_io_bio, bio);
284 }
285
286 struct btrfs_bio_stripe {
287 struct btrfs_device *dev;
288 u64 physical;
289 u64 length; /* only used for discard mappings */
290 };
291
292 struct btrfs_bio;
293 typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
294
295 struct btrfs_bio {
296 refcount_t refs;
297 atomic_t stripes_pending;
298 struct btrfs_fs_info *fs_info;
299 u64 map_type; /* get from map_lookup->type */
300 bio_end_io_t *end_io;
301 struct bio *orig_bio;
302 unsigned long flags;
303 void *private;
304 atomic_t error;
305 int max_errors;
306 int num_stripes;
307 int mirror_num;
308 int num_tgtdevs;
309 int *tgtdev_map;
310 /*
311 * logical block numbers for the start of each stripe
312 * The last one or two are p/q. These are sorted,
313 * so raid_map[0] is the start of our full stripe
314 */
315 u64 *raid_map;
316 struct btrfs_bio_stripe stripes[];
317 };
318
319 struct btrfs_device_info {
320 struct btrfs_device *dev;
321 u64 dev_offset;
322 u64 max_avail;
323 u64 total_avail;
324 };
325
326 struct btrfs_raid_attr {
327 int sub_stripes; /* sub_stripes info for map */
328 int dev_stripes; /* stripes per dev */
329 int devs_max; /* max devs to use */
330 int devs_min; /* min devs needed */
331 int tolerated_failures; /* max tolerated fail devs */
332 int devs_increment; /* ndevs has to be a multiple of this */
333 int ncopies; /* how many copies to data has */
334 int mindev_error; /* error code if min devs requisite is unmet */
335 const char raid_name[8]; /* name of the raid */
336 u64 bg_flag; /* block group flag of the raid */
337 };
338
339 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
340
341 struct map_lookup {
342 u64 type;
343 int io_align;
344 int io_width;
345 u64 stripe_len;
346 int num_stripes;
347 int sub_stripes;
348 int verified_stripes; /* For mount time dev extent verification */
349 struct btrfs_bio_stripe stripes[];
350 };
351
352 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
353 (sizeof(struct btrfs_bio_stripe) * (n)))
354
355 struct btrfs_balance_args;
356 struct btrfs_balance_progress;
357 struct btrfs_balance_control {
358 struct btrfs_balance_args data;
359 struct btrfs_balance_args meta;
360 struct btrfs_balance_args sys;
361
362 u64 flags;
363
364 struct btrfs_balance_progress stat;
365 };
366
367 enum btrfs_map_op {
368 BTRFS_MAP_READ,
369 BTRFS_MAP_WRITE,
370 BTRFS_MAP_DISCARD,
371 BTRFS_MAP_GET_READ_MIRRORS,
372 };
373
btrfs_op(struct bio * bio)374 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
375 {
376 switch (bio_op(bio)) {
377 case REQ_OP_DISCARD:
378 return BTRFS_MAP_DISCARD;
379 case REQ_OP_WRITE:
380 return BTRFS_MAP_WRITE;
381 default:
382 WARN_ON_ONCE(1);
383 case REQ_OP_READ:
384 return BTRFS_MAP_READ;
385 }
386 }
387
388 void btrfs_get_bbio(struct btrfs_bio *bbio);
389 void btrfs_put_bbio(struct btrfs_bio *bbio);
390 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
391 u64 logical, u64 *length,
392 struct btrfs_bio **bbio_ret, int mirror_num);
393 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
394 u64 logical, u64 *length,
395 struct btrfs_bio **bbio_ret);
396 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
397 u64 physical, u64 **logical, int *naddrs, int *stripe_len);
398 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
399 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
400 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
401 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
402 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
403 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
404 int mirror_num, int async_submit);
405 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
406 fmode_t flags, void *holder);
407 struct btrfs_device *btrfs_scan_one_device(const char *path,
408 fmode_t flags, void *holder);
409 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
410 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
411 void btrfs_assign_next_active_device(struct btrfs_device *device,
412 struct btrfs_device *this_dev);
413 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
414 const char *device_path,
415 struct btrfs_device **device);
416 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
417 const char *devpath,
418 struct btrfs_device **device);
419 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
420 const u64 *devid,
421 const u8 *uuid);
422 void btrfs_free_device(struct btrfs_device *device);
423 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
424 const char *device_path, u64 devid);
425 void __exit btrfs_cleanup_fs_uuids(void);
426 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
427 int btrfs_grow_device(struct btrfs_trans_handle *trans,
428 struct btrfs_device *device, u64 new_size);
429 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
430 u8 *uuid, u8 *fsid);
431 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
432 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
433 int btrfs_balance(struct btrfs_fs_info *fs_info,
434 struct btrfs_balance_control *bctl,
435 struct btrfs_ioctl_balance_args *bargs);
436 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
437 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
438 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
439 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
440 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
441 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
442 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
443 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
444 struct btrfs_device *device, u64 num_bytes,
445 u64 search_start, u64 *start, u64 *max_avail);
446 int find_free_dev_extent(struct btrfs_trans_handle *trans,
447 struct btrfs_device *device, u64 num_bytes,
448 u64 *start, u64 *max_avail);
449 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
450 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
451 struct btrfs_ioctl_get_dev_stats *stats);
452 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
453 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
454 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
455 struct btrfs_fs_info *fs_info);
456 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
457 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
458 struct btrfs_device *srcdev);
459 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
460 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
461 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
462 u64 logical, u64 len);
463 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
464 u64 logical);
465 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
466 u64 chunk_offset, u64 chunk_size);
467 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
468
btrfs_dev_stat_inc(struct btrfs_device * dev,int index)469 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
470 int index)
471 {
472 atomic_inc(dev->dev_stat_values + index);
473 /*
474 * This memory barrier orders stores updating statistics before stores
475 * updating dev_stats_ccnt.
476 *
477 * It pairs with smp_rmb() in btrfs_run_dev_stats().
478 */
479 smp_mb__before_atomic();
480 atomic_inc(&dev->dev_stats_ccnt);
481 }
482
btrfs_dev_stat_read(struct btrfs_device * dev,int index)483 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
484 int index)
485 {
486 return atomic_read(dev->dev_stat_values + index);
487 }
488
btrfs_dev_stat_read_and_reset(struct btrfs_device * dev,int index)489 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
490 int index)
491 {
492 int ret;
493
494 ret = atomic_xchg(dev->dev_stat_values + index, 0);
495 /*
496 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
497 * - RMW operations that have a return value are fully ordered;
498 *
499 * This implicit memory barriers is paired with the smp_rmb in
500 * btrfs_run_dev_stats
501 */
502 atomic_inc(&dev->dev_stats_ccnt);
503 return ret;
504 }
505
btrfs_dev_stat_set(struct btrfs_device * dev,int index,unsigned long val)506 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
507 int index, unsigned long val)
508 {
509 atomic_set(dev->dev_stat_values + index, val);
510 /*
511 * This memory barrier orders stores updating statistics before stores
512 * updating dev_stats_ccnt.
513 *
514 * It pairs with smp_rmb() in btrfs_run_dev_stats().
515 */
516 smp_mb__before_atomic();
517 atomic_inc(&dev->dev_stats_ccnt);
518 }
519
btrfs_dev_stat_reset(struct btrfs_device * dev,int index)520 static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
521 int index)
522 {
523 btrfs_dev_stat_set(dev, index, 0);
524 }
525
526 /*
527 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
528 * can be used as index to access btrfs_raid_array[].
529 */
btrfs_bg_flags_to_raid_index(u64 flags)530 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
531 {
532 if (flags & BTRFS_BLOCK_GROUP_RAID10)
533 return BTRFS_RAID_RAID10;
534 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
535 return BTRFS_RAID_RAID1;
536 else if (flags & BTRFS_BLOCK_GROUP_DUP)
537 return BTRFS_RAID_DUP;
538 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
539 return BTRFS_RAID_RAID0;
540 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
541 return BTRFS_RAID_RAID5;
542 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
543 return BTRFS_RAID_RAID6;
544
545 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
546 }
547
548 const char *get_raid_name(enum btrfs_raid_types type);
549
550 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info);
551 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans);
552
553 struct list_head *btrfs_get_fs_uuids(void);
554 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
555 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
556 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
557 struct btrfs_device *failing_dev);
558
559 int btrfs_bg_type_to_factor(u64 flags);
560 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
561
562 #endif
563
