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 btrfs_io_geometry {
21 /* remaining bytes before crossing a stripe */
22 u64 len;
23 /* offset of logical address in chunk */
24 u64 offset;
25 /* length of single IO stripe */
26 u64 stripe_len;
27 /* number of stripe where address falls */
28 u64 stripe_nr;
29 /* offset of address in stripe */
30 u64 stripe_offset;
31 /* offset of raid56 stripe into the chunk */
32 u64 raid56_stripe_offset;
33 };
34
35 /*
36 * Use sequence counter to get consistent device stat data on
37 * 32-bit processors.
38 */
39 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
40 #include <linux/seqlock.h>
41 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
42 #define btrfs_device_data_ordered_init(device) \
43 seqcount_init(&device->data_seqcount)
44 #else
45 #define btrfs_device_data_ordered_init(device) do { } while (0)
46 #endif
47
48 #define BTRFS_DEV_STATE_WRITEABLE (0)
49 #define BTRFS_DEV_STATE_IN_FS_METADATA (1)
50 #define BTRFS_DEV_STATE_MISSING (2)
51 #define BTRFS_DEV_STATE_REPLACE_TGT (3)
52 #define BTRFS_DEV_STATE_FLUSH_SENT (4)
53 #define BTRFS_DEV_STATE_NO_READA (5)
54
55 struct btrfs_zoned_device_info;
56
57 struct btrfs_device {
58 struct list_head dev_list; /* device_list_mutex */
59 struct list_head dev_alloc_list; /* chunk mutex */
60 struct list_head post_commit_list; /* chunk mutex */
61 struct btrfs_fs_devices *fs_devices;
62 struct btrfs_fs_info *fs_info;
63
64 struct rcu_string __rcu *name;
65
66 u64 generation;
67
68 struct block_device *bdev;
69
70 struct btrfs_zoned_device_info *zone_info;
71
72 /* the mode sent to blkdev_get */
73 fmode_t mode;
74
75 unsigned long dev_state;
76 blk_status_t last_flush_error;
77
78 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
79 seqcount_t data_seqcount;
80 #endif
81
82 /* the internal btrfs device id */
83 u64 devid;
84
85 /* size of the device in memory */
86 u64 total_bytes;
87
88 /* size of the device on disk */
89 u64 disk_total_bytes;
90
91 /* bytes used */
92 u64 bytes_used;
93
94 /* optimal io alignment for this device */
95 u32 io_align;
96
97 /* optimal io width for this device */
98 u32 io_width;
99 /* type and info about this device */
100 u64 type;
101
102 /* minimal io size for this device */
103 u32 sector_size;
104
105 /* physical drive uuid (or lvm uuid) */
106 u8 uuid[BTRFS_UUID_SIZE];
107
108 /*
109 * size of the device on the current transaction
110 *
111 * This variant is update when committing the transaction,
112 * and protected by chunk mutex
113 */
114 u64 commit_total_bytes;
115
116 /* bytes used on the current transaction */
117 u64 commit_bytes_used;
118
119 /* for sending down flush barriers */
120 struct bio *flush_bio;
121 struct completion flush_wait;
122
123 /* per-device scrub information */
124 struct scrub_ctx *scrub_ctx;
125
126 /* readahead state */
127 atomic_t reada_in_flight;
128 u64 reada_next;
129 struct reada_zone *reada_curr_zone;
130 struct radix_tree_root reada_zones;
131 struct radix_tree_root reada_extents;
132
133 /* disk I/O failure stats. For detailed description refer to
134 * enum btrfs_dev_stat_values in ioctl.h */
135 int dev_stats_valid;
136
137 /* Counter to record the change of device stats */
138 atomic_t dev_stats_ccnt;
139 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
140
141 struct extent_io_tree alloc_state;
142
143 struct completion kobj_unregister;
144 /* For sysfs/FSID/devinfo/devid/ */
145 struct kobject devid_kobj;
146
147 /* Bandwidth limit for scrub, in bytes */
148 u64 scrub_speed_max;
149 };
150
151 /*
152 * If we read those variants at the context of their own lock, we needn't
153 * use the following helpers, reading them directly is safe.
154 */
155 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
156 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
157 static inline u64 \
158 btrfs_device_get_##name(const struct btrfs_device *dev) \
159 { \
160 u64 size; \
161 unsigned int seq; \
162 \
163 do { \
164 seq = read_seqcount_begin(&dev->data_seqcount); \
165 size = dev->name; \
166 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
167 return size; \
168 } \
169 \
170 static inline void \
171 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
172 { \
173 preempt_disable(); \
174 write_seqcount_begin(&dev->data_seqcount); \
175 dev->name = size; \
176 write_seqcount_end(&dev->data_seqcount); \
177 preempt_enable(); \
178 }
179 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
180 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
181 static inline u64 \
182 btrfs_device_get_##name(const struct btrfs_device *dev) \
183 { \
184 u64 size; \
185 \
186 preempt_disable(); \
187 size = dev->name; \
188 preempt_enable(); \
189 return size; \
190 } \
191 \
192 static inline void \
193 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
194 { \
195 preempt_disable(); \
196 dev->name = size; \
197 preempt_enable(); \
198 }
199 #else
200 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
201 static inline u64 \
202 btrfs_device_get_##name(const struct btrfs_device *dev) \
203 { \
204 return dev->name; \
205 } \
206 \
207 static inline void \
208 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
209 { \
210 dev->name = size; \
211 }
212 #endif
213
214 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
215 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
216 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
217
218 enum btrfs_chunk_allocation_policy {
219 BTRFS_CHUNK_ALLOC_REGULAR,
220 BTRFS_CHUNK_ALLOC_ZONED,
221 };
222
223 /*
224 * Read policies for mirrored block group profiles, read picks the stripe based
225 * on these policies.
226 */
227 enum btrfs_read_policy {
228 /* Use process PID to choose the stripe */
229 BTRFS_READ_POLICY_PID,
230 BTRFS_NR_READ_POLICY,
231 };
232
233 struct btrfs_fs_devices {
234 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
235 u8 metadata_uuid[BTRFS_FSID_SIZE];
236 bool fsid_change;
237 struct list_head fs_list;
238
239 /*
240 * Number of devices under this fsid including missing and
241 * replace-target device and excludes seed devices.
242 */
243 u64 num_devices;
244
245 /*
246 * The number of devices that successfully opened, including
247 * replace-target, excludes seed devices.
248 */
249 u64 open_devices;
250
251 /* The number of devices that are under the chunk allocation list. */
252 u64 rw_devices;
253
254 /* Count of missing devices under this fsid excluding seed device. */
255 u64 missing_devices;
256 u64 total_rw_bytes;
257
258 /*
259 * Count of devices from btrfs_super_block::num_devices for this fsid,
260 * which includes the seed device, excludes the transient replace-target
261 * device.
262 */
263 u64 total_devices;
264
265 /* Highest generation number of seen devices */
266 u64 latest_generation;
267
268 /*
269 * The mount device or a device with highest generation after removal
270 * or replace.
271 */
272 struct btrfs_device *latest_dev;
273
274 /* all of the devices in the FS, protected by a mutex
275 * so we can safely walk it to write out the supers without
276 * worrying about add/remove by the multi-device code.
277 * Scrubbing super can kick off supers writing by holding
278 * this mutex lock.
279 */
280 struct mutex device_list_mutex;
281
282 /* List of all devices, protected by device_list_mutex */
283 struct list_head devices;
284
285 /*
286 * Devices which can satisfy space allocation. Protected by
287 * chunk_mutex
288 */
289 struct list_head alloc_list;
290
291 struct list_head seed_list;
292 bool seeding;
293
294 int opened;
295
296 /* set when we find or add a device that doesn't have the
297 * nonrot flag set
298 */
299 bool rotating;
300
301 struct btrfs_fs_info *fs_info;
302 /* sysfs kobjects */
303 struct kobject fsid_kobj;
304 struct kobject *devices_kobj;
305 struct kobject *devinfo_kobj;
306 struct completion kobj_unregister;
307
308 enum btrfs_chunk_allocation_policy chunk_alloc_policy;
309
310 /* Policy used to read the mirrored stripes */
311 enum btrfs_read_policy read_policy;
312 };
313
314 #define BTRFS_BIO_INLINE_CSUM_SIZE 64
315
316 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
317 - sizeof(struct btrfs_chunk)) \
318 / sizeof(struct btrfs_stripe) + 1)
319
320 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
321 - 2 * sizeof(struct btrfs_disk_key) \
322 - 2 * sizeof(struct btrfs_chunk)) \
323 / sizeof(struct btrfs_stripe) + 1)
324
325 /*
326 * Additional info to pass along bio.
327 *
328 * Mostly for btrfs specific features like csum and mirror_num.
329 */
330 struct btrfs_bio {
331 unsigned int mirror_num;
332
333 /* @device is for stripe IO submission. */
334 struct btrfs_device *device;
335 u8 *csum;
336 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
337 struct bvec_iter iter;
338
339 /*
340 * This member must come last, bio_alloc_bioset will allocate enough
341 * bytes for entire btrfs_bio but relies on bio being last.
342 */
343 struct bio bio;
344 };
345
btrfs_bio(struct bio * bio)346 static inline struct btrfs_bio *btrfs_bio(struct bio *bio)
347 {
348 return container_of(bio, struct btrfs_bio, bio);
349 }
350
btrfs_bio_free_csum(struct btrfs_bio * bbio)351 static inline void btrfs_bio_free_csum(struct btrfs_bio *bbio)
352 {
353 if (bbio->csum != bbio->csum_inline) {
354 kfree(bbio->csum);
355 bbio->csum = NULL;
356 }
357 }
358
359 struct btrfs_io_stripe {
360 struct btrfs_device *dev;
361 u64 physical;
362 u64 length; /* only used for discard mappings */
363 };
364
365 /*
366 * Context for IO subsmission for device stripe.
367 *
368 * - Track the unfinished mirrors for mirror based profiles
369 * Mirror based profiles are SINGLE/DUP/RAID1/RAID10.
370 *
371 * - Contain the logical -> physical mapping info
372 * Used by submit_stripe_bio() for mapping logical bio
373 * into physical device address.
374 *
375 * - Contain device replace info
376 * Used by handle_ops_on_dev_replace() to copy logical bios
377 * into the new device.
378 *
379 * - Contain RAID56 full stripe logical bytenrs
380 */
381 struct btrfs_io_context {
382 refcount_t refs;
383 atomic_t stripes_pending;
384 struct btrfs_fs_info *fs_info;
385 u64 map_type; /* get from map_lookup->type */
386 bio_end_io_t *end_io;
387 struct bio *orig_bio;
388 void *private;
389 atomic_t error;
390 int max_errors;
391 int num_stripes;
392 int mirror_num;
393 int num_tgtdevs;
394 int *tgtdev_map;
395 /*
396 * logical block numbers for the start of each stripe
397 * The last one or two are p/q. These are sorted,
398 * so raid_map[0] is the start of our full stripe
399 */
400 u64 *raid_map;
401 struct btrfs_io_stripe stripes[];
402 };
403
404 struct btrfs_device_info {
405 struct btrfs_device *dev;
406 u64 dev_offset;
407 u64 max_avail;
408 u64 total_avail;
409 };
410
411 struct btrfs_raid_attr {
412 u8 sub_stripes; /* sub_stripes info for map */
413 u8 dev_stripes; /* stripes per dev */
414 u8 devs_max; /* max devs to use */
415 u8 devs_min; /* min devs needed */
416 u8 tolerated_failures; /* max tolerated fail devs */
417 u8 devs_increment; /* ndevs has to be a multiple of this */
418 u8 ncopies; /* how many copies to data has */
419 u8 nparity; /* number of stripes worth of bytes to store
420 * parity information */
421 u8 mindev_error; /* error code if min devs requisite is unmet */
422 const char raid_name[8]; /* name of the raid */
423 u64 bg_flag; /* block group flag of the raid */
424 };
425
426 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
427
428 struct map_lookup {
429 u64 type;
430 int io_align;
431 int io_width;
432 u64 stripe_len;
433 int num_stripes;
434 int sub_stripes;
435 int verified_stripes; /* For mount time dev extent verification */
436 struct btrfs_io_stripe stripes[];
437 };
438
439 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
440 (sizeof(struct btrfs_io_stripe) * (n)))
441
442 struct btrfs_balance_args;
443 struct btrfs_balance_progress;
444 struct btrfs_balance_control {
445 struct btrfs_balance_args data;
446 struct btrfs_balance_args meta;
447 struct btrfs_balance_args sys;
448
449 u64 flags;
450
451 struct btrfs_balance_progress stat;
452 };
453
454 /*
455 * Search for a given device by the set parameters
456 */
457 struct btrfs_dev_lookup_args {
458 u64 devid;
459 u8 *uuid;
460 u8 *fsid;
461 bool missing;
462 };
463
464 /* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */
465 #define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 }
466
467 #define BTRFS_DEV_LOOKUP_ARGS(name) \
468 struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT
469
470 enum btrfs_map_op {
471 BTRFS_MAP_READ,
472 BTRFS_MAP_WRITE,
473 BTRFS_MAP_DISCARD,
474 BTRFS_MAP_GET_READ_MIRRORS,
475 };
476
btrfs_op(struct bio * bio)477 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
478 {
479 switch (bio_op(bio)) {
480 case REQ_OP_DISCARD:
481 return BTRFS_MAP_DISCARD;
482 case REQ_OP_WRITE:
483 case REQ_OP_ZONE_APPEND:
484 return BTRFS_MAP_WRITE;
485 default:
486 WARN_ON_ONCE(1);
487 fallthrough;
488 case REQ_OP_READ:
489 return BTRFS_MAP_READ;
490 }
491 }
492
493 void btrfs_get_bioc(struct btrfs_io_context *bioc);
494 void btrfs_put_bioc(struct btrfs_io_context *bioc);
495 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
496 u64 logical, u64 *length,
497 struct btrfs_io_context **bioc_ret, int mirror_num);
498 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
499 u64 logical, u64 *length,
500 struct btrfs_io_context **bioc_ret);
501 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *map,
502 enum btrfs_map_op op, u64 logical,
503 struct btrfs_io_geometry *io_geom);
504 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
505 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
506 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
507 u64 type);
508 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
509 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
510 int mirror_num);
511 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
512 fmode_t flags, void *holder);
513 struct btrfs_device *btrfs_scan_one_device(const char *path,
514 fmode_t flags, void *holder);
515 int btrfs_forget_devices(const char *path);
516 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
517 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
518 void btrfs_assign_next_active_device(struct btrfs_device *device,
519 struct btrfs_device *this_dev);
520 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
521 u64 devid,
522 const char *devpath);
523 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
524 struct btrfs_dev_lookup_args *args,
525 const char *path);
526 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
527 const u64 *devid,
528 const u8 *uuid);
529 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args);
530 void btrfs_free_device(struct btrfs_device *device);
531 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
532 struct btrfs_dev_lookup_args *args,
533 struct block_device **bdev, fmode_t *mode);
534 void __exit btrfs_cleanup_fs_uuids(void);
535 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
536 int btrfs_grow_device(struct btrfs_trans_handle *trans,
537 struct btrfs_device *device, u64 new_size);
538 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
539 const struct btrfs_dev_lookup_args *args);
540 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
541 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
542 int btrfs_balance(struct btrfs_fs_info *fs_info,
543 struct btrfs_balance_control *bctl,
544 struct btrfs_ioctl_balance_args *bargs);
545 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
546 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
547 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
548 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
549 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
550 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
551 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
552 int btrfs_uuid_scan_kthread(void *data);
553 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset);
554 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
555 u64 *start, u64 *max_avail);
556 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
557 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
558 struct btrfs_ioctl_get_dev_stats *stats);
559 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
560 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
561 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
562 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
563 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
564 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
565 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
566 u64 logical, u64 len);
567 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
568 u64 logical);
569 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
570 struct btrfs_block_group *bg);
571 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
572 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
573 u64 logical, u64 length);
574 void btrfs_release_disk_super(struct btrfs_super_block *super);
575
btrfs_dev_stat_inc(struct btrfs_device * dev,int index)576 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
577 int index)
578 {
579 atomic_inc(dev->dev_stat_values + index);
580 /*
581 * This memory barrier orders stores updating statistics before stores
582 * updating dev_stats_ccnt.
583 *
584 * It pairs with smp_rmb() in btrfs_run_dev_stats().
585 */
586 smp_mb__before_atomic();
587 atomic_inc(&dev->dev_stats_ccnt);
588 }
589
btrfs_dev_stat_read(struct btrfs_device * dev,int index)590 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
591 int index)
592 {
593 return atomic_read(dev->dev_stat_values + index);
594 }
595
btrfs_dev_stat_read_and_reset(struct btrfs_device * dev,int index)596 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
597 int index)
598 {
599 int ret;
600
601 ret = atomic_xchg(dev->dev_stat_values + index, 0);
602 /*
603 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
604 * - RMW operations that have a return value are fully ordered;
605 *
606 * This implicit memory barriers is paired with the smp_rmb in
607 * btrfs_run_dev_stats
608 */
609 atomic_inc(&dev->dev_stats_ccnt);
610 return ret;
611 }
612
btrfs_dev_stat_set(struct btrfs_device * dev,int index,unsigned long val)613 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
614 int index, unsigned long val)
615 {
616 atomic_set(dev->dev_stat_values + index, val);
617 /*
618 * This memory barrier orders stores updating statistics before stores
619 * updating dev_stats_ccnt.
620 *
621 * It pairs with smp_rmb() in btrfs_run_dev_stats().
622 */
623 smp_mb__before_atomic();
624 atomic_inc(&dev->dev_stats_ccnt);
625 }
626
627 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
628
629 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
630 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
631 struct btrfs_device *failing_dev);
632 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
633 struct block_device *bdev,
634 const char *device_path);
635
636 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
637 int btrfs_bg_type_to_factor(u64 flags);
638 const char *btrfs_bg_type_to_raid_name(u64 flags);
639 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
640 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
641
642 #endif
643