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 /* When true means this device has pending chunk alloc in
58 * current transaction. Protected by chunk_mutex.
59 */
60 bool has_pending_chunks;
61
62 /* regular prio bios */
63 struct btrfs_pending_bios pending_bios;
64 /* sync bios */
65 struct btrfs_pending_bios pending_sync_bios;
66
67 struct block_device *bdev;
68
69 /* the mode sent to blkdev_get */
70 fmode_t mode;
71
72 unsigned long dev_state;
73 blk_status_t last_flush_error;
74 int flush_bio_sent;
75
76 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
77 seqcount_t data_seqcount;
78 #endif
79
80 /* the internal btrfs device id */
81 u64 devid;
82
83 /* size of the device in memory */
84 u64 total_bytes;
85
86 /* size of the device on disk */
87 u64 disk_total_bytes;
88
89 /* bytes used */
90 u64 bytes_used;
91
92 /* optimal io alignment for this device */
93 u32 io_align;
94
95 /* optimal io width for this device */
96 u32 io_width;
97 /* type and info about this device */
98 u64 type;
99
100 /* minimal io size for this device */
101 u32 sector_size;
102
103 /* physical drive uuid (or lvm uuid) */
104 u8 uuid[BTRFS_UUID_SIZE];
105
106 /*
107 * size of the device on the current transaction
108 *
109 * This variant is update when committing the transaction,
110 * and protected by device_list_mutex
111 */
112 u64 commit_total_bytes;
113
114 /* bytes used on the current transaction */
115 u64 commit_bytes_used;
116 /*
117 * used to manage the device which is resized
118 *
119 * It is protected by chunk_lock.
120 */
121 struct list_head resized_list;
122
123 /* for sending down flush barriers */
124 struct bio *flush_bio;
125 struct completion flush_wait;
126
127 /* per-device scrub information */
128 struct scrub_ctx *scrub_ctx;
129
130 struct btrfs_work work;
131 struct rcu_head rcu;
132
133 /* readahead state */
134 atomic_t reada_in_flight;
135 u64 reada_next;
136 struct reada_zone *reada_curr_zone;
137 struct radix_tree_root reada_zones;
138 struct radix_tree_root reada_extents;
139
140 /* disk I/O failure stats. For detailed description refer to
141 * enum btrfs_dev_stat_values in ioctl.h */
142 int dev_stats_valid;
143
144 /* Counter to record the change of device stats */
145 atomic_t dev_stats_ccnt;
146 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
147 };
148
149 /*
150 * If we read those variants at the context of their own lock, we needn't
151 * use the following helpers, reading them directly is safe.
152 */
153 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
154 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
155 static inline u64 \
156 btrfs_device_get_##name(const struct btrfs_device *dev) \
157 { \
158 u64 size; \
159 unsigned int seq; \
160 \
161 do { \
162 seq = read_seqcount_begin(&dev->data_seqcount); \
163 size = dev->name; \
164 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
165 return size; \
166 } \
167 \
168 static inline void \
169 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
170 { \
171 preempt_disable(); \
172 write_seqcount_begin(&dev->data_seqcount); \
173 dev->name = size; \
174 write_seqcount_end(&dev->data_seqcount); \
175 preempt_enable(); \
176 }
177 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
178 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
179 static inline u64 \
180 btrfs_device_get_##name(const struct btrfs_device *dev) \
181 { \
182 u64 size; \
183 \
184 preempt_disable(); \
185 size = dev->name; \
186 preempt_enable(); \
187 return size; \
188 } \
189 \
190 static inline void \
191 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
192 { \
193 preempt_disable(); \
194 dev->name = size; \
195 preempt_enable(); \
196 }
197 #else
198 #define BTRFS_DEVICE_GETSET_FUNCS(name) \
199 static inline u64 \
200 btrfs_device_get_##name(const struct btrfs_device *dev) \
201 { \
202 return dev->name; \
203 } \
204 \
205 static inline void \
206 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
207 { \
208 dev->name = size; \
209 }
210 #endif
211
212 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
213 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
214 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
215
216 struct btrfs_fs_devices {
217 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
218 struct list_head fs_list;
219
220 u64 num_devices;
221 u64 open_devices;
222 u64 rw_devices;
223 u64 missing_devices;
224 u64 total_rw_bytes;
225 u64 total_devices;
226 struct block_device *latest_bdev;
227
228 /* all of the devices in the FS, protected by a mutex
229 * so we can safely walk it to write out the supers without
230 * worrying about add/remove by the multi-device code.
231 * Scrubbing super can kick off supers writing by holding
232 * this mutex lock.
233 */
234 struct mutex device_list_mutex;
235 struct list_head devices;
236
237 struct list_head resized_devices;
238 /* devices not currently being allocated */
239 struct list_head alloc_list;
240
241 struct btrfs_fs_devices *seed;
242 int seeding;
243
244 int opened;
245
246 /* set when we find or add a device that doesn't have the
247 * nonrot flag set
248 */
249 int rotating;
250
251 struct btrfs_fs_info *fs_info;
252 /* sysfs kobjects */
253 struct kobject fsid_kobj;
254 struct kobject *device_dir_kobj;
255 struct completion kobj_unregister;
256 };
257
258 #define BTRFS_BIO_INLINE_CSUM_SIZE 64
259
260 /*
261 * we need the mirror number and stripe index to be passed around
262 * the call chain while we are processing end_io (especially errors).
263 * Really, what we need is a btrfs_bio structure that has this info
264 * and is properly sized with its stripe array, but we're not there
265 * quite yet. We have our own btrfs bioset, and all of the bios
266 * we allocate are actually btrfs_io_bios. We'll cram as much of
267 * struct btrfs_bio as we can into this over time.
268 */
269 typedef void (btrfs_io_bio_end_io_t) (struct btrfs_io_bio *bio, int err);
270 struct btrfs_io_bio {
271 unsigned int mirror_num;
272 unsigned int stripe_index;
273 u64 logical;
274 u8 *csum;
275 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
276 u8 *csum_allocated;
277 btrfs_io_bio_end_io_t *end_io;
278 struct bvec_iter iter;
279 /*
280 * This member must come last, bio_alloc_bioset will allocate enough
281 * bytes for entire btrfs_io_bio but relies on bio being last.
282 */
283 struct bio bio;
284 };
285
btrfs_io_bio(struct bio * bio)286 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
287 {
288 return container_of(bio, struct btrfs_io_bio, bio);
289 }
290
291 struct btrfs_bio_stripe {
292 struct btrfs_device *dev;
293 u64 physical;
294 u64 length; /* only used for discard mappings */
295 };
296
297 struct btrfs_bio;
298 typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
299
300 struct btrfs_bio {
301 refcount_t refs;
302 atomic_t stripes_pending;
303 struct btrfs_fs_info *fs_info;
304 u64 map_type; /* get from map_lookup->type */
305 bio_end_io_t *end_io;
306 struct bio *orig_bio;
307 void *private;
308 atomic_t error;
309 int max_errors;
310 int num_stripes;
311 int mirror_num;
312 int num_tgtdevs;
313 int *tgtdev_map;
314 /*
315 * logical block numbers for the start of each stripe
316 * The last one or two are p/q. These are sorted,
317 * so raid_map[0] is the start of our full stripe
318 */
319 u64 *raid_map;
320 struct btrfs_bio_stripe stripes[];
321 };
322
323 struct btrfs_device_info {
324 struct btrfs_device *dev;
325 u64 dev_offset;
326 u64 max_avail;
327 u64 total_avail;
328 };
329
330 struct btrfs_raid_attr {
331 int sub_stripes; /* sub_stripes info for map */
332 int dev_stripes; /* stripes per dev */
333 int devs_max; /* max devs to use */
334 int devs_min; /* min devs needed */
335 int tolerated_failures; /* max tolerated fail devs */
336 int devs_increment; /* ndevs has to be a multiple of this */
337 int ncopies; /* how many copies to data has */
338 int mindev_error; /* error code if min devs requisite is unmet */
339 const char raid_name[8]; /* name of the raid */
340 u64 bg_flag; /* block group flag of the raid */
341 };
342
343 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
344
345 struct map_lookup {
346 u64 type;
347 int io_align;
348 int io_width;
349 u64 stripe_len;
350 int num_stripes;
351 int sub_stripes;
352 int verified_stripes; /* For mount time dev extent verification */
353 struct btrfs_bio_stripe stripes[];
354 };
355
356 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
357 (sizeof(struct btrfs_bio_stripe) * (n)))
358
359 struct btrfs_balance_args;
360 struct btrfs_balance_progress;
361 struct btrfs_balance_control {
362 struct btrfs_balance_args data;
363 struct btrfs_balance_args meta;
364 struct btrfs_balance_args sys;
365
366 u64 flags;
367
368 struct btrfs_balance_progress stat;
369 };
370
371 enum btrfs_map_op {
372 BTRFS_MAP_READ,
373 BTRFS_MAP_WRITE,
374 BTRFS_MAP_DISCARD,
375 BTRFS_MAP_GET_READ_MIRRORS,
376 };
377
btrfs_op(struct bio * bio)378 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
379 {
380 switch (bio_op(bio)) {
381 case REQ_OP_DISCARD:
382 return BTRFS_MAP_DISCARD;
383 case REQ_OP_WRITE:
384 return BTRFS_MAP_WRITE;
385 default:
386 WARN_ON_ONCE(1);
387 case REQ_OP_READ:
388 return BTRFS_MAP_READ;
389 }
390 }
391
392 void btrfs_get_bbio(struct btrfs_bio *bbio);
393 void btrfs_put_bbio(struct btrfs_bio *bbio);
394 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
395 u64 logical, u64 *length,
396 struct btrfs_bio **bbio_ret, int mirror_num);
397 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
398 u64 logical, u64 *length,
399 struct btrfs_bio **bbio_ret);
400 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
401 u64 physical, u64 **logical, int *naddrs, int *stripe_len);
402 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
403 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
404 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
405 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
406 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
407 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
408 int mirror_num, int async_submit);
409 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
410 fmode_t flags, void *holder);
411 struct btrfs_device *btrfs_scan_one_device(const char *path,
412 fmode_t flags, void *holder);
413 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
414 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
415 void btrfs_assign_next_active_device(struct btrfs_device *device,
416 struct btrfs_device *this_dev);
417 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
418 const char *device_path,
419 struct btrfs_device **device);
420 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
421 const char *devpath,
422 struct btrfs_device **device);
423 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
424 const u64 *devid,
425 const u8 *uuid);
426 void btrfs_free_device(struct btrfs_device *device);
427 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
428 const char *device_path, u64 devid);
429 void __exit btrfs_cleanup_fs_uuids(void);
430 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
431 int btrfs_grow_device(struct btrfs_trans_handle *trans,
432 struct btrfs_device *device, u64 new_size);
433 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
434 u64 devid, u8 *uuid, u8 *fsid, bool seed);
435 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
436 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
437 int btrfs_balance(struct btrfs_fs_info *fs_info,
438 struct btrfs_balance_control *bctl,
439 struct btrfs_ioctl_balance_args *bargs);
440 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
441 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
442 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
443 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
444 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
445 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
446 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
447 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
448 struct btrfs_device *device, u64 num_bytes,
449 u64 search_start, u64 *start, u64 *max_avail);
450 int find_free_dev_extent(struct btrfs_trans_handle *trans,
451 struct btrfs_device *device, u64 num_bytes,
452 u64 *start, u64 *max_avail);
453 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
454 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
455 struct btrfs_ioctl_get_dev_stats *stats);
456 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
457 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
458 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
459 struct btrfs_fs_info *fs_info);
460 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
461 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
462 struct btrfs_device *srcdev);
463 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
464 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
465 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
466 u64 logical, u64 len);
467 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
468 u64 logical);
469 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
470 u64 chunk_offset, u64 chunk_size);
471 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
472
btrfs_dev_stat_inc(struct btrfs_device * dev,int index)473 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
474 int index)
475 {
476 atomic_inc(dev->dev_stat_values + index);
477 /*
478 * This memory barrier orders stores updating statistics before stores
479 * updating dev_stats_ccnt.
480 *
481 * It pairs with smp_rmb() in btrfs_run_dev_stats().
482 */
483 smp_mb__before_atomic();
484 atomic_inc(&dev->dev_stats_ccnt);
485 }
486
btrfs_dev_stat_read(struct btrfs_device * dev,int index)487 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
488 int index)
489 {
490 return atomic_read(dev->dev_stat_values + index);
491 }
492
btrfs_dev_stat_read_and_reset(struct btrfs_device * dev,int index)493 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
494 int index)
495 {
496 int ret;
497
498 ret = atomic_xchg(dev->dev_stat_values + index, 0);
499 /*
500 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
501 * - RMW operations that have a return value are fully ordered;
502 *
503 * This implicit memory barriers is paired with the smp_rmb in
504 * btrfs_run_dev_stats
505 */
506 atomic_inc(&dev->dev_stats_ccnt);
507 return ret;
508 }
509
btrfs_dev_stat_set(struct btrfs_device * dev,int index,unsigned long val)510 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
511 int index, unsigned long val)
512 {
513 atomic_set(dev->dev_stat_values + index, val);
514 /*
515 * This memory barrier orders stores updating statistics before stores
516 * updating dev_stats_ccnt.
517 *
518 * It pairs with smp_rmb() in btrfs_run_dev_stats().
519 */
520 smp_mb__before_atomic();
521 atomic_inc(&dev->dev_stats_ccnt);
522 }
523
btrfs_dev_stat_reset(struct btrfs_device * dev,int index)524 static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
525 int index)
526 {
527 btrfs_dev_stat_set(dev, index, 0);
528 }
529
530 /*
531 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
532 * can be used as index to access btrfs_raid_array[].
533 */
btrfs_bg_flags_to_raid_index(u64 flags)534 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
535 {
536 if (flags & BTRFS_BLOCK_GROUP_RAID10)
537 return BTRFS_RAID_RAID10;
538 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
539 return BTRFS_RAID_RAID1;
540 else if (flags & BTRFS_BLOCK_GROUP_DUP)
541 return BTRFS_RAID_DUP;
542 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
543 return BTRFS_RAID_RAID0;
544 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
545 return BTRFS_RAID_RAID5;
546 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
547 return BTRFS_RAID_RAID6;
548
549 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
550 }
551
552 const char *get_raid_name(enum btrfs_raid_types type);
553
554 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info);
555 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans);
556
557 struct list_head *btrfs_get_fs_uuids(void);
558 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
559 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
560 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
561 struct btrfs_device *failing_dev);
562
563 int btrfs_bg_type_to_factor(u64 flags);
564 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
565
566 #endif
567