xref: /linux/fs/btrfs/volumes.h (revision 3da3cc1b5f47115b16b5ffeeb4bf09ec331b0164)
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 
148 /*
149  * If we read those variants at the context of their own lock, we needn't
150  * use the following helpers, reading them directly is safe.
151  */
152 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
153 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
154 static inline u64							\
155 btrfs_device_get_##name(const struct btrfs_device *dev)			\
156 {									\
157 	u64 size;							\
158 	unsigned int seq;						\
159 									\
160 	do {								\
161 		seq = read_seqcount_begin(&dev->data_seqcount);		\
162 		size = dev->name;					\
163 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
164 	return size;							\
165 }									\
166 									\
167 static inline void							\
168 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
169 {									\
170 	preempt_disable();						\
171 	write_seqcount_begin(&dev->data_seqcount);			\
172 	dev->name = size;						\
173 	write_seqcount_end(&dev->data_seqcount);			\
174 	preempt_enable();						\
175 }
176 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
177 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
178 static inline u64							\
179 btrfs_device_get_##name(const struct btrfs_device *dev)			\
180 {									\
181 	u64 size;							\
182 									\
183 	preempt_disable();						\
184 	size = dev->name;						\
185 	preempt_enable();						\
186 	return size;							\
187 }									\
188 									\
189 static inline void							\
190 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
191 {									\
192 	preempt_disable();						\
193 	dev->name = size;						\
194 	preempt_enable();						\
195 }
196 #else
197 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
198 static inline u64							\
199 btrfs_device_get_##name(const struct btrfs_device *dev)			\
200 {									\
201 	return dev->name;						\
202 }									\
203 									\
204 static inline void							\
205 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
206 {									\
207 	dev->name = size;						\
208 }
209 #endif
210 
211 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
212 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
213 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
214 
215 enum btrfs_chunk_allocation_policy {
216 	BTRFS_CHUNK_ALLOC_REGULAR,
217 };
218 
219 /*
220  * Read policies for mirrored block group profiles, read picks the stripe based
221  * on these policies.
222  */
223 enum btrfs_read_policy {
224 	/* Use process PID to choose the stripe */
225 	BTRFS_READ_POLICY_PID,
226 	BTRFS_NR_READ_POLICY,
227 };
228 
229 struct btrfs_fs_devices {
230 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
231 	u8 metadata_uuid[BTRFS_FSID_SIZE];
232 	bool fsid_change;
233 	struct list_head fs_list;
234 
235 	u64 num_devices;
236 	u64 open_devices;
237 	u64 rw_devices;
238 	u64 missing_devices;
239 	u64 total_rw_bytes;
240 	u64 total_devices;
241 
242 	/* Highest generation number of seen devices */
243 	u64 latest_generation;
244 
245 	struct block_device *latest_bdev;
246 
247 	/* all of the devices in the FS, protected by a mutex
248 	 * so we can safely walk it to write out the supers without
249 	 * worrying about add/remove by the multi-device code.
250 	 * Scrubbing super can kick off supers writing by holding
251 	 * this mutex lock.
252 	 */
253 	struct mutex device_list_mutex;
254 
255 	/* List of all devices, protected by device_list_mutex */
256 	struct list_head devices;
257 
258 	/*
259 	 * Devices which can satisfy space allocation. Protected by
260 	 * chunk_mutex
261 	 */
262 	struct list_head alloc_list;
263 
264 	struct list_head seed_list;
265 	bool seeding;
266 
267 	int opened;
268 
269 	/* set when we find or add a device that doesn't have the
270 	 * nonrot flag set
271 	 */
272 	bool rotating;
273 
274 	struct btrfs_fs_info *fs_info;
275 	/* sysfs kobjects */
276 	struct kobject fsid_kobj;
277 	struct kobject *devices_kobj;
278 	struct kobject *devinfo_kobj;
279 	struct completion kobj_unregister;
280 
281 	enum btrfs_chunk_allocation_policy chunk_alloc_policy;
282 
283 	/* Policy used to read the mirrored stripes */
284 	enum btrfs_read_policy read_policy;
285 };
286 
287 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
288 
289 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
290 			- sizeof(struct btrfs_chunk))		\
291 			/ sizeof(struct btrfs_stripe) + 1)
292 
293 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
294 				- 2 * sizeof(struct btrfs_disk_key)	\
295 				- 2 * sizeof(struct btrfs_chunk))	\
296 				/ sizeof(struct btrfs_stripe) + 1)
297 
298 /*
299  * we need the mirror number and stripe index to be passed around
300  * the call chain while we are processing end_io (especially errors).
301  * Really, what we need is a btrfs_bio structure that has this info
302  * and is properly sized with its stripe array, but we're not there
303  * quite yet.  We have our own btrfs bioset, and all of the bios
304  * we allocate are actually btrfs_io_bios.  We'll cram as much of
305  * struct btrfs_bio as we can into this over time.
306  */
307 struct btrfs_io_bio {
308 	unsigned int mirror_num;
309 	struct btrfs_device *device;
310 	u64 logical;
311 	u8 *csum;
312 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
313 	struct bvec_iter iter;
314 	/*
315 	 * This member must come last, bio_alloc_bioset will allocate enough
316 	 * bytes for entire btrfs_io_bio but relies on bio being last.
317 	 */
318 	struct bio bio;
319 };
320 
321 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
322 {
323 	return container_of(bio, struct btrfs_io_bio, bio);
324 }
325 
326 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
327 {
328 	if (io_bio->csum != io_bio->csum_inline) {
329 		kfree(io_bio->csum);
330 		io_bio->csum = NULL;
331 	}
332 }
333 
334 struct btrfs_bio_stripe {
335 	struct btrfs_device *dev;
336 	u64 physical;
337 	u64 length; /* only used for discard mappings */
338 };
339 
340 struct btrfs_bio {
341 	refcount_t refs;
342 	atomic_t stripes_pending;
343 	struct btrfs_fs_info *fs_info;
344 	u64 map_type; /* get from map_lookup->type */
345 	bio_end_io_t *end_io;
346 	struct bio *orig_bio;
347 	void *private;
348 	atomic_t error;
349 	int max_errors;
350 	int num_stripes;
351 	int mirror_num;
352 	int num_tgtdevs;
353 	int *tgtdev_map;
354 	/*
355 	 * logical block numbers for the start of each stripe
356 	 * The last one or two are p/q.  These are sorted,
357 	 * so raid_map[0] is the start of our full stripe
358 	 */
359 	u64 *raid_map;
360 	struct btrfs_bio_stripe stripes[];
361 };
362 
363 struct btrfs_device_info {
364 	struct btrfs_device *dev;
365 	u64 dev_offset;
366 	u64 max_avail;
367 	u64 total_avail;
368 };
369 
370 struct btrfs_raid_attr {
371 	u8 sub_stripes;		/* sub_stripes info for map */
372 	u8 dev_stripes;		/* stripes per dev */
373 	u8 devs_max;		/* max devs to use */
374 	u8 devs_min;		/* min devs needed */
375 	u8 tolerated_failures;	/* max tolerated fail devs */
376 	u8 devs_increment;	/* ndevs has to be a multiple of this */
377 	u8 ncopies;		/* how many copies to data has */
378 	u8 nparity;		/* number of stripes worth of bytes to store
379 				 * parity information */
380 	u8 mindev_error;	/* error code if min devs requisite is unmet */
381 	const char raid_name[8]; /* name of the raid */
382 	u64 bg_flag;		/* block group flag of the raid */
383 };
384 
385 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
386 
387 struct map_lookup {
388 	u64 type;
389 	int io_align;
390 	int io_width;
391 	u64 stripe_len;
392 	int num_stripes;
393 	int sub_stripes;
394 	int verified_stripes; /* For mount time dev extent verification */
395 	struct btrfs_bio_stripe stripes[];
396 };
397 
398 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
399 			    (sizeof(struct btrfs_bio_stripe) * (n)))
400 
401 struct btrfs_balance_args;
402 struct btrfs_balance_progress;
403 struct btrfs_balance_control {
404 	struct btrfs_balance_args data;
405 	struct btrfs_balance_args meta;
406 	struct btrfs_balance_args sys;
407 
408 	u64 flags;
409 
410 	struct btrfs_balance_progress stat;
411 };
412 
413 enum btrfs_map_op {
414 	BTRFS_MAP_READ,
415 	BTRFS_MAP_WRITE,
416 	BTRFS_MAP_DISCARD,
417 	BTRFS_MAP_GET_READ_MIRRORS,
418 };
419 
420 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
421 {
422 	switch (bio_op(bio)) {
423 	case REQ_OP_DISCARD:
424 		return BTRFS_MAP_DISCARD;
425 	case REQ_OP_WRITE:
426 		return BTRFS_MAP_WRITE;
427 	default:
428 		WARN_ON_ONCE(1);
429 		fallthrough;
430 	case REQ_OP_READ:
431 		return BTRFS_MAP_READ;
432 	}
433 }
434 
435 void btrfs_get_bbio(struct btrfs_bio *bbio);
436 void btrfs_put_bbio(struct btrfs_bio *bbio);
437 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
438 		    u64 logical, u64 *length,
439 		    struct btrfs_bio **bbio_ret, int mirror_num);
440 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
441 		     u64 logical, u64 *length,
442 		     struct btrfs_bio **bbio_ret);
443 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
444 		u64 logical, u64 len, struct btrfs_io_geometry *io_geom);
445 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
446 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
447 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
448 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
449 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
450 			   int mirror_num);
451 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
452 		       fmode_t flags, void *holder);
453 struct btrfs_device *btrfs_scan_one_device(const char *path,
454 					   fmode_t flags, void *holder);
455 int btrfs_forget_devices(const char *path);
456 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
457 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
458 void btrfs_assign_next_active_device(struct btrfs_device *device,
459 				     struct btrfs_device *this_dev);
460 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
461 						  u64 devid,
462 						  const char *devpath);
463 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
464 					const u64 *devid,
465 					const u8 *uuid);
466 void btrfs_free_device(struct btrfs_device *device);
467 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
468 		    const char *device_path, u64 devid);
469 void __exit btrfs_cleanup_fs_uuids(void);
470 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
471 int btrfs_grow_device(struct btrfs_trans_handle *trans,
472 		      struct btrfs_device *device, u64 new_size);
473 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
474 				       u64 devid, u8 *uuid, u8 *fsid);
475 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
476 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
477 int btrfs_balance(struct btrfs_fs_info *fs_info,
478 		  struct btrfs_balance_control *bctl,
479 		  struct btrfs_ioctl_balance_args *bargs);
480 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
481 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
482 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
483 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
484 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
485 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
486 int btrfs_uuid_scan_kthread(void *data);
487 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
488 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
489 			 u64 *start, u64 *max_avail);
490 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
491 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
492 			struct btrfs_ioctl_get_dev_stats *stats);
493 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
494 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
495 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
496 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
497 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
498 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
499 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
500 			   u64 logical, u64 len);
501 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
502 				    u64 logical);
503 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
504 			     u64 chunk_offset, u64 chunk_size);
505 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
506 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
507 				       u64 logical, u64 length);
508 void btrfs_release_disk_super(struct btrfs_super_block *super);
509 
510 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
511 				      int index)
512 {
513 	atomic_inc(dev->dev_stat_values + index);
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 
524 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
525 				      int index)
526 {
527 	return atomic_read(dev->dev_stat_values + index);
528 }
529 
530 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
531 						int index)
532 {
533 	int ret;
534 
535 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
536 	/*
537 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
538 	 * - RMW operations that have a return value are fully ordered;
539 	 *
540 	 * This implicit memory barriers is paired with the smp_rmb in
541 	 * btrfs_run_dev_stats
542 	 */
543 	atomic_inc(&dev->dev_stats_ccnt);
544 	return ret;
545 }
546 
547 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
548 				      int index, unsigned long val)
549 {
550 	atomic_set(dev->dev_stat_values + index, val);
551 	/*
552 	 * This memory barrier orders stores updating statistics before stores
553 	 * updating dev_stats_ccnt.
554 	 *
555 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
556 	 */
557 	smp_mb__before_atomic();
558 	atomic_inc(&dev->dev_stats_ccnt);
559 }
560 
561 /*
562  * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
563  * can be used as index to access btrfs_raid_array[].
564  */
565 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
566 {
567 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
568 		return BTRFS_RAID_RAID10;
569 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
570 		return BTRFS_RAID_RAID1;
571 	else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
572 		return BTRFS_RAID_RAID1C3;
573 	else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
574 		return BTRFS_RAID_RAID1C4;
575 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
576 		return BTRFS_RAID_DUP;
577 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
578 		return BTRFS_RAID_RAID0;
579 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
580 		return BTRFS_RAID_RAID5;
581 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
582 		return BTRFS_RAID_RAID6;
583 
584 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
585 }
586 
587 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
588 
589 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
590 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
591 					struct btrfs_device *failing_dev);
592 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
593 			       struct block_device *bdev,
594 			       const char *device_path);
595 
596 int btrfs_bg_type_to_factor(u64 flags);
597 const char *btrfs_bg_type_to_raid_name(u64 flags);
598 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
599 
600 #endif
601