xref: /linux/fs/btrfs/volumes.h (revision f6f39f7a0add4e7fd120a709545b57586a1d0393)
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 	u64 num_devices;
240 	u64 open_devices;
241 	u64 rw_devices;
242 	u64 missing_devices;
243 	u64 total_rw_bytes;
244 	u64 total_devices;
245 
246 	/* Highest generation number of seen devices */
247 	u64 latest_generation;
248 
249 	struct block_device *latest_bdev;
250 
251 	/* all of the devices in the FS, protected by a mutex
252 	 * so we can safely walk it to write out the supers without
253 	 * worrying about add/remove by the multi-device code.
254 	 * Scrubbing super can kick off supers writing by holding
255 	 * this mutex lock.
256 	 */
257 	struct mutex device_list_mutex;
258 
259 	/* List of all devices, protected by device_list_mutex */
260 	struct list_head devices;
261 
262 	/*
263 	 * Devices which can satisfy space allocation. Protected by
264 	 * chunk_mutex
265 	 */
266 	struct list_head alloc_list;
267 
268 	struct list_head seed_list;
269 	bool seeding;
270 
271 	int opened;
272 
273 	/* set when we find or add a device that doesn't have the
274 	 * nonrot flag set
275 	 */
276 	bool rotating;
277 
278 	struct btrfs_fs_info *fs_info;
279 	/* sysfs kobjects */
280 	struct kobject fsid_kobj;
281 	struct kobject *devices_kobj;
282 	struct kobject *devinfo_kobj;
283 	struct completion kobj_unregister;
284 
285 	enum btrfs_chunk_allocation_policy chunk_alloc_policy;
286 
287 	/* Policy used to read the mirrored stripes */
288 	enum btrfs_read_policy read_policy;
289 };
290 
291 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
292 
293 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
294 			- sizeof(struct btrfs_chunk))		\
295 			/ sizeof(struct btrfs_stripe) + 1)
296 
297 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
298 				- 2 * sizeof(struct btrfs_disk_key)	\
299 				- 2 * sizeof(struct btrfs_chunk))	\
300 				/ sizeof(struct btrfs_stripe) + 1)
301 
302 /*
303  * we need the mirror number and stripe index to be passed around
304  * the call chain while we are processing end_io (especially errors).
305  * Really, what we need is a btrfs_bio structure that has this info
306  * and is properly sized with its stripe array, but we're not there
307  * quite yet.  We have our own btrfs bioset, and all of the bios
308  * we allocate are actually btrfs_io_bios.  We'll cram as much of
309  * struct btrfs_bio as we can into this over time.
310  */
311 struct btrfs_io_bio {
312 	unsigned int mirror_num;
313 	struct btrfs_device *device;
314 	u64 logical;
315 	u8 *csum;
316 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
317 	struct bvec_iter iter;
318 	/*
319 	 * This member must come last, bio_alloc_bioset will allocate enough
320 	 * bytes for entire btrfs_io_bio but relies on bio being last.
321 	 */
322 	struct bio bio;
323 };
324 
325 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
326 {
327 	return container_of(bio, struct btrfs_io_bio, bio);
328 }
329 
330 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
331 {
332 	if (io_bio->csum != io_bio->csum_inline) {
333 		kfree(io_bio->csum);
334 		io_bio->csum = NULL;
335 	}
336 }
337 
338 struct btrfs_bio_stripe {
339 	struct btrfs_device *dev;
340 	u64 physical;
341 	u64 length; /* only used for discard mappings */
342 };
343 
344 struct btrfs_bio {
345 	refcount_t refs;
346 	atomic_t stripes_pending;
347 	struct btrfs_fs_info *fs_info;
348 	u64 map_type; /* get from map_lookup->type */
349 	bio_end_io_t *end_io;
350 	struct bio *orig_bio;
351 	void *private;
352 	atomic_t error;
353 	int max_errors;
354 	int num_stripes;
355 	int mirror_num;
356 	int num_tgtdevs;
357 	int *tgtdev_map;
358 	/*
359 	 * logical block numbers for the start of each stripe
360 	 * The last one or two are p/q.  These are sorted,
361 	 * so raid_map[0] is the start of our full stripe
362 	 */
363 	u64 *raid_map;
364 	struct btrfs_bio_stripe stripes[];
365 };
366 
367 struct btrfs_device_info {
368 	struct btrfs_device *dev;
369 	u64 dev_offset;
370 	u64 max_avail;
371 	u64 total_avail;
372 };
373 
374 struct btrfs_raid_attr {
375 	u8 sub_stripes;		/* sub_stripes info for map */
376 	u8 dev_stripes;		/* stripes per dev */
377 	u8 devs_max;		/* max devs to use */
378 	u8 devs_min;		/* min devs needed */
379 	u8 tolerated_failures;	/* max tolerated fail devs */
380 	u8 devs_increment;	/* ndevs has to be a multiple of this */
381 	u8 ncopies;		/* how many copies to data has */
382 	u8 nparity;		/* number of stripes worth of bytes to store
383 				 * parity information */
384 	u8 mindev_error;	/* error code if min devs requisite is unmet */
385 	const char raid_name[8]; /* name of the raid */
386 	u64 bg_flag;		/* block group flag of the raid */
387 };
388 
389 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
390 
391 struct map_lookup {
392 	u64 type;
393 	int io_align;
394 	int io_width;
395 	u64 stripe_len;
396 	int num_stripes;
397 	int sub_stripes;
398 	int verified_stripes; /* For mount time dev extent verification */
399 	struct btrfs_bio_stripe stripes[];
400 };
401 
402 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
403 			    (sizeof(struct btrfs_bio_stripe) * (n)))
404 
405 struct btrfs_balance_args;
406 struct btrfs_balance_progress;
407 struct btrfs_balance_control {
408 	struct btrfs_balance_args data;
409 	struct btrfs_balance_args meta;
410 	struct btrfs_balance_args sys;
411 
412 	u64 flags;
413 
414 	struct btrfs_balance_progress stat;
415 };
416 
417 enum btrfs_map_op {
418 	BTRFS_MAP_READ,
419 	BTRFS_MAP_WRITE,
420 	BTRFS_MAP_DISCARD,
421 	BTRFS_MAP_GET_READ_MIRRORS,
422 };
423 
424 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
425 {
426 	switch (bio_op(bio)) {
427 	case REQ_OP_DISCARD:
428 		return BTRFS_MAP_DISCARD;
429 	case REQ_OP_WRITE:
430 	case REQ_OP_ZONE_APPEND:
431 		return BTRFS_MAP_WRITE;
432 	default:
433 		WARN_ON_ONCE(1);
434 		fallthrough;
435 	case REQ_OP_READ:
436 		return BTRFS_MAP_READ;
437 	}
438 }
439 
440 void btrfs_get_bbio(struct btrfs_bio *bbio);
441 void btrfs_put_bbio(struct btrfs_bio *bbio);
442 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
443 		    u64 logical, u64 *length,
444 		    struct btrfs_bio **bbio_ret, int mirror_num);
445 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
446 		     u64 logical, u64 *length,
447 		     struct btrfs_bio **bbio_ret);
448 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *map,
449 			  enum btrfs_map_op op, u64 logical,
450 			  struct btrfs_io_geometry *io_geom);
451 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
452 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
453 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
454 					    u64 type);
455 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
456 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
457 			   int mirror_num);
458 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
459 		       fmode_t flags, void *holder);
460 struct btrfs_device *btrfs_scan_one_device(const char *path,
461 					   fmode_t flags, void *holder);
462 int btrfs_forget_devices(const char *path);
463 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
464 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
465 void btrfs_assign_next_active_device(struct btrfs_device *device,
466 				     struct btrfs_device *this_dev);
467 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
468 						  u64 devid,
469 						  const char *devpath);
470 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
471 					const u64 *devid,
472 					const u8 *uuid);
473 void btrfs_free_device(struct btrfs_device *device);
474 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
475 		    const char *device_path, u64 devid,
476 		    struct block_device **bdev, fmode_t *mode);
477 void __exit btrfs_cleanup_fs_uuids(void);
478 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
479 int btrfs_grow_device(struct btrfs_trans_handle *trans,
480 		      struct btrfs_device *device, u64 new_size);
481 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
482 				       u64 devid, u8 *uuid, u8 *fsid);
483 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
484 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
485 int btrfs_balance(struct btrfs_fs_info *fs_info,
486 		  struct btrfs_balance_control *bctl,
487 		  struct btrfs_ioctl_balance_args *bargs);
488 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
489 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
490 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
491 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
492 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
493 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
494 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
495 int btrfs_uuid_scan_kthread(void *data);
496 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
497 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
498 			 u64 *start, u64 *max_avail);
499 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
500 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
501 			struct btrfs_ioctl_get_dev_stats *stats);
502 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
503 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
504 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
505 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
506 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
507 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
508 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
509 			   u64 logical, u64 len);
510 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
511 				    u64 logical);
512 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
513 				     struct btrfs_block_group *bg);
514 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
515 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
516 				       u64 logical, u64 length);
517 void btrfs_release_disk_super(struct btrfs_super_block *super);
518 
519 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
520 				      int index)
521 {
522 	atomic_inc(dev->dev_stat_values + index);
523 	/*
524 	 * This memory barrier orders stores updating statistics before stores
525 	 * updating dev_stats_ccnt.
526 	 *
527 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
528 	 */
529 	smp_mb__before_atomic();
530 	atomic_inc(&dev->dev_stats_ccnt);
531 }
532 
533 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
534 				      int index)
535 {
536 	return atomic_read(dev->dev_stat_values + index);
537 }
538 
539 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
540 						int index)
541 {
542 	int ret;
543 
544 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
545 	/*
546 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
547 	 * - RMW operations that have a return value are fully ordered;
548 	 *
549 	 * This implicit memory barriers is paired with the smp_rmb in
550 	 * btrfs_run_dev_stats
551 	 */
552 	atomic_inc(&dev->dev_stats_ccnt);
553 	return ret;
554 }
555 
556 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
557 				      int index, unsigned long val)
558 {
559 	atomic_set(dev->dev_stat_values + index, val);
560 	/*
561 	 * This memory barrier orders stores updating statistics before stores
562 	 * updating dev_stats_ccnt.
563 	 *
564 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
565 	 */
566 	smp_mb__before_atomic();
567 	atomic_inc(&dev->dev_stats_ccnt);
568 }
569 
570 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
571 
572 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
573 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
574 					struct btrfs_device *failing_dev);
575 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
576 			       struct block_device *bdev,
577 			       const char *device_path);
578 
579 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
580 int btrfs_bg_type_to_factor(u64 flags);
581 const char *btrfs_bg_type_to_raid_name(u64 flags);
582 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
583 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
584 
585 #endif
586