xref: /openbmc/linux/fs/btrfs/volumes.h (revision 5e012745)
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 struct btrfs_io_geometry {
27 	/* remaining bytes before crossing a stripe */
28 	u64 len;
29 	/* offset of logical address in chunk */
30 	u64 offset;
31 	/* length of single IO stripe */
32 	u64 stripe_len;
33 	/* number of stripe where address falls */
34 	u64 stripe_nr;
35 	/* offset of address in stripe */
36 	u64 stripe_offset;
37 	/* offset of raid56 stripe into the chunk */
38 	u64 raid56_stripe_offset;
39 };
40 
41 /*
42  * Use sequence counter to get consistent device stat data on
43  * 32-bit processors.
44  */
45 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
46 #include <linux/seqlock.h>
47 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
48 #define btrfs_device_data_ordered_init(device)	\
49 	seqcount_init(&device->data_seqcount)
50 #else
51 #define btrfs_device_data_ordered_init(device) do { } while (0)
52 #endif
53 
54 #define BTRFS_DEV_STATE_WRITEABLE	(0)
55 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
56 #define BTRFS_DEV_STATE_MISSING		(2)
57 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
58 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
59 
60 struct btrfs_device {
61 	struct list_head dev_list; /* device_list_mutex */
62 	struct list_head dev_alloc_list; /* chunk mutex */
63 	struct list_head post_commit_list; /* chunk mutex */
64 	struct btrfs_fs_devices *fs_devices;
65 	struct btrfs_fs_info *fs_info;
66 
67 	struct rcu_string *name;
68 
69 	u64 generation;
70 
71 	spinlock_t io_lock ____cacheline_aligned;
72 	int running_pending;
73 	/* regular prio bios */
74 	struct btrfs_pending_bios pending_bios;
75 	/* sync bios */
76 	struct btrfs_pending_bios pending_sync_bios;
77 
78 	struct block_device *bdev;
79 
80 	/* the mode sent to blkdev_get */
81 	fmode_t mode;
82 
83 	unsigned long dev_state;
84 	blk_status_t last_flush_error;
85 	int flush_bio_sent;
86 
87 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
88 	seqcount_t data_seqcount;
89 #endif
90 
91 	/* the internal btrfs device id */
92 	u64 devid;
93 
94 	/* size of the device in memory */
95 	u64 total_bytes;
96 
97 	/* size of the device on disk */
98 	u64 disk_total_bytes;
99 
100 	/* bytes used */
101 	u64 bytes_used;
102 
103 	/* optimal io alignment for this device */
104 	u32 io_align;
105 
106 	/* optimal io width for this device */
107 	u32 io_width;
108 	/* type and info about this device */
109 	u64 type;
110 
111 	/* minimal io size for this device */
112 	u32 sector_size;
113 
114 	/* physical drive uuid (or lvm uuid) */
115 	u8 uuid[BTRFS_UUID_SIZE];
116 
117 	/*
118 	 * size of the device on the current transaction
119 	 *
120 	 * This variant is update when committing the transaction,
121 	 * and protected by chunk mutex
122 	 */
123 	u64 commit_total_bytes;
124 
125 	/* bytes used on the current transaction */
126 	u64 commit_bytes_used;
127 
128 	/* for sending down flush barriers */
129 	struct bio *flush_bio;
130 	struct completion flush_wait;
131 
132 	/* per-device scrub information */
133 	struct scrub_ctx *scrub_ctx;
134 
135 	struct btrfs_work work;
136 
137 	/* readahead state */
138 	atomic_t reada_in_flight;
139 	u64 reada_next;
140 	struct reada_zone *reada_curr_zone;
141 	struct radix_tree_root reada_zones;
142 	struct radix_tree_root reada_extents;
143 
144 	/* disk I/O failure stats. For detailed description refer to
145 	 * enum btrfs_dev_stat_values in ioctl.h */
146 	int dev_stats_valid;
147 
148 	/* Counter to record the change of device stats */
149 	atomic_t dev_stats_ccnt;
150 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
151 
152 	struct extent_io_tree alloc_state;
153 };
154 
155 /*
156  * If we read those variants at the context of their own lock, we needn't
157  * use the following helpers, reading them directly is safe.
158  */
159 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
160 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
161 static inline u64							\
162 btrfs_device_get_##name(const struct btrfs_device *dev)			\
163 {									\
164 	u64 size;							\
165 	unsigned int seq;						\
166 									\
167 	do {								\
168 		seq = read_seqcount_begin(&dev->data_seqcount);		\
169 		size = dev->name;					\
170 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
171 	return size;							\
172 }									\
173 									\
174 static inline void							\
175 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
176 {									\
177 	preempt_disable();						\
178 	write_seqcount_begin(&dev->data_seqcount);			\
179 	dev->name = size;						\
180 	write_seqcount_end(&dev->data_seqcount);			\
181 	preempt_enable();						\
182 }
183 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
184 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
185 static inline u64							\
186 btrfs_device_get_##name(const struct btrfs_device *dev)			\
187 {									\
188 	u64 size;							\
189 									\
190 	preempt_disable();						\
191 	size = dev->name;						\
192 	preempt_enable();						\
193 	return size;							\
194 }									\
195 									\
196 static inline void							\
197 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
198 {									\
199 	preempt_disable();						\
200 	dev->name = size;						\
201 	preempt_enable();						\
202 }
203 #else
204 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
205 static inline u64							\
206 btrfs_device_get_##name(const struct btrfs_device *dev)			\
207 {									\
208 	return dev->name;						\
209 }									\
210 									\
211 static inline void							\
212 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
213 {									\
214 	dev->name = size;						\
215 }
216 #endif
217 
218 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
219 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
220 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
221 
222 struct btrfs_fs_devices {
223 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
224 	u8 metadata_uuid[BTRFS_FSID_SIZE];
225 	bool fsid_change;
226 	struct list_head fs_list;
227 
228 	u64 num_devices;
229 	u64 open_devices;
230 	u64 rw_devices;
231 	u64 missing_devices;
232 	u64 total_rw_bytes;
233 	u64 total_devices;
234 
235 	/* Highest generation number of seen devices */
236 	u64 latest_generation;
237 
238 	struct block_device *latest_bdev;
239 
240 	/* all of the devices in the FS, protected by a mutex
241 	 * so we can safely walk it to write out the supers without
242 	 * worrying about add/remove by the multi-device code.
243 	 * Scrubbing super can kick off supers writing by holding
244 	 * this mutex lock.
245 	 */
246 	struct mutex device_list_mutex;
247 
248 	/* List of all devices, protected by device_list_mutex */
249 	struct list_head devices;
250 
251 	/*
252 	 * Devices which can satisfy space allocation. Protected by
253 	 * chunk_mutex
254 	 */
255 	struct list_head alloc_list;
256 
257 	struct btrfs_fs_devices *seed;
258 	int seeding;
259 
260 	int opened;
261 
262 	/* set when we find or add a device that doesn't have the
263 	 * nonrot flag set
264 	 */
265 	int rotating;
266 
267 	struct btrfs_fs_info *fs_info;
268 	/* sysfs kobjects */
269 	struct kobject fsid_kobj;
270 	struct kobject *device_dir_kobj;
271 	struct completion kobj_unregister;
272 };
273 
274 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
275 
276 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
277 			- sizeof(struct btrfs_chunk))		\
278 			/ sizeof(struct btrfs_stripe) + 1)
279 
280 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
281 				- 2 * sizeof(struct btrfs_disk_key)	\
282 				- 2 * sizeof(struct btrfs_chunk))	\
283 				/ sizeof(struct btrfs_stripe) + 1)
284 
285 /*
286  * we need the mirror number and stripe index to be passed around
287  * the call chain while we are processing end_io (especially errors).
288  * Really, what we need is a btrfs_bio structure that has this info
289  * and is properly sized with its stripe array, but we're not there
290  * quite yet.  We have our own btrfs bioset, and all of the bios
291  * we allocate are actually btrfs_io_bios.  We'll cram as much of
292  * struct btrfs_bio as we can into this over time.
293  */
294 struct btrfs_io_bio {
295 	unsigned int mirror_num;
296 	unsigned int stripe_index;
297 	u64 logical;
298 	u8 *csum;
299 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
300 	struct bvec_iter iter;
301 	/*
302 	 * This member must come last, bio_alloc_bioset will allocate enough
303 	 * bytes for entire btrfs_io_bio but relies on bio being last.
304 	 */
305 	struct bio bio;
306 };
307 
308 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
309 {
310 	return container_of(bio, struct btrfs_io_bio, bio);
311 }
312 
313 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
314 {
315 	if (io_bio->csum != io_bio->csum_inline) {
316 		kfree(io_bio->csum);
317 		io_bio->csum = NULL;
318 	}
319 }
320 
321 struct btrfs_bio_stripe {
322 	struct btrfs_device *dev;
323 	u64 physical;
324 	u64 length; /* only used for discard mappings */
325 };
326 
327 struct btrfs_bio {
328 	refcount_t refs;
329 	atomic_t stripes_pending;
330 	struct btrfs_fs_info *fs_info;
331 	u64 map_type; /* get from map_lookup->type */
332 	bio_end_io_t *end_io;
333 	struct bio *orig_bio;
334 	unsigned long flags;
335 	void *private;
336 	atomic_t error;
337 	int max_errors;
338 	int num_stripes;
339 	int mirror_num;
340 	int num_tgtdevs;
341 	int *tgtdev_map;
342 	/*
343 	 * logical block numbers for the start of each stripe
344 	 * The last one or two are p/q.  These are sorted,
345 	 * so raid_map[0] is the start of our full stripe
346 	 */
347 	u64 *raid_map;
348 	struct btrfs_bio_stripe stripes[];
349 };
350 
351 struct btrfs_device_info {
352 	struct btrfs_device *dev;
353 	u64 dev_offset;
354 	u64 max_avail;
355 	u64 total_avail;
356 };
357 
358 struct btrfs_raid_attr {
359 	u8 sub_stripes;		/* sub_stripes info for map */
360 	u8 dev_stripes;		/* stripes per dev */
361 	u8 devs_max;		/* max devs to use */
362 	u8 devs_min;		/* min devs needed */
363 	u8 tolerated_failures;	/* max tolerated fail devs */
364 	u8 devs_increment;	/* ndevs has to be a multiple of this */
365 	u8 ncopies;		/* how many copies to data has */
366 	u8 nparity;		/* number of stripes worth of bytes to store
367 				 * parity information */
368 	u8 mindev_error;	/* error code if min devs requisite is unmet */
369 	const char raid_name[8]; /* name of the raid */
370 	u64 bg_flag;		/* block group flag of the raid */
371 };
372 
373 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
374 
375 struct map_lookup {
376 	u64 type;
377 	int io_align;
378 	int io_width;
379 	u64 stripe_len;
380 	int num_stripes;
381 	int sub_stripes;
382 	int verified_stripes; /* For mount time dev extent verification */
383 	struct btrfs_bio_stripe stripes[];
384 };
385 
386 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
387 			    (sizeof(struct btrfs_bio_stripe) * (n)))
388 
389 struct btrfs_balance_args;
390 struct btrfs_balance_progress;
391 struct btrfs_balance_control {
392 	struct btrfs_balance_args data;
393 	struct btrfs_balance_args meta;
394 	struct btrfs_balance_args sys;
395 
396 	u64 flags;
397 
398 	struct btrfs_balance_progress stat;
399 };
400 
401 enum btrfs_map_op {
402 	BTRFS_MAP_READ,
403 	BTRFS_MAP_WRITE,
404 	BTRFS_MAP_DISCARD,
405 	BTRFS_MAP_GET_READ_MIRRORS,
406 };
407 
408 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
409 {
410 	switch (bio_op(bio)) {
411 	case REQ_OP_DISCARD:
412 		return BTRFS_MAP_DISCARD;
413 	case REQ_OP_WRITE:
414 		return BTRFS_MAP_WRITE;
415 	default:
416 		WARN_ON_ONCE(1);
417 		/* fall through */
418 	case REQ_OP_READ:
419 		return BTRFS_MAP_READ;
420 	}
421 }
422 
423 void btrfs_get_bbio(struct btrfs_bio *bbio);
424 void btrfs_put_bbio(struct btrfs_bio *bbio);
425 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
426 		    u64 logical, u64 *length,
427 		    struct btrfs_bio **bbio_ret, int mirror_num);
428 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
429 		     u64 logical, u64 *length,
430 		     struct btrfs_bio **bbio_ret);
431 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
432 		u64 logical, u64 len, struct btrfs_io_geometry *io_geom);
433 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
434 		     u64 physical, u64 **logical, int *naddrs, int *stripe_len);
435 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
436 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
437 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
438 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
439 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
440 			   int mirror_num, int async_submit);
441 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
442 		       fmode_t flags, void *holder);
443 struct btrfs_device *btrfs_scan_one_device(const char *path,
444 					   fmode_t flags, void *holder);
445 int btrfs_forget_devices(const char *path);
446 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
447 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
448 void btrfs_assign_next_active_device(struct btrfs_device *device,
449 				     struct btrfs_device *this_dev);
450 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
451 						  u64 devid,
452 						  const char *devpath);
453 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
454 					const u64 *devid,
455 					const u8 *uuid);
456 void btrfs_free_device(struct btrfs_device *device);
457 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
458 		    const char *device_path, u64 devid);
459 void __exit btrfs_cleanup_fs_uuids(void);
460 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
461 int btrfs_grow_device(struct btrfs_trans_handle *trans,
462 		      struct btrfs_device *device, u64 new_size);
463 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
464 				       u64 devid, u8 *uuid, u8 *fsid, bool seed);
465 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
466 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
467 int btrfs_balance(struct btrfs_fs_info *fs_info,
468 		  struct btrfs_balance_control *bctl,
469 		  struct btrfs_ioctl_balance_args *bargs);
470 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
471 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
472 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
473 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
474 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
475 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
476 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
477 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
478 int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes,
479 			       u64 search_start, u64 *start, u64 *max_avail);
480 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
481 			 u64 *start, u64 *max_avail);
482 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
483 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
484 			struct btrfs_ioctl_get_dev_stats *stats);
485 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
486 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
487 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
488 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
489 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
490 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
491 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
492 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
493 			   u64 logical, u64 len);
494 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
495 				    u64 logical);
496 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
497 			     u64 chunk_offset, u64 chunk_size);
498 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
499 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
500 				       u64 logical, u64 length);
501 
502 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
503 				      int index)
504 {
505 	atomic_inc(dev->dev_stat_values + index);
506 	/*
507 	 * This memory barrier orders stores updating statistics before stores
508 	 * updating dev_stats_ccnt.
509 	 *
510 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
511 	 */
512 	smp_mb__before_atomic();
513 	atomic_inc(&dev->dev_stats_ccnt);
514 }
515 
516 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
517 				      int index)
518 {
519 	return atomic_read(dev->dev_stat_values + index);
520 }
521 
522 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
523 						int index)
524 {
525 	int ret;
526 
527 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
528 	/*
529 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
530 	 * - RMW operations that have a return value are fully ordered;
531 	 *
532 	 * This implicit memory barriers is paired with the smp_rmb in
533 	 * btrfs_run_dev_stats
534 	 */
535 	atomic_inc(&dev->dev_stats_ccnt);
536 	return ret;
537 }
538 
539 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
540 				      int index, unsigned long val)
541 {
542 	atomic_set(dev->dev_stat_values + index, val);
543 	/*
544 	 * This memory barrier orders stores updating statistics before stores
545 	 * updating dev_stats_ccnt.
546 	 *
547 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
548 	 */
549 	smp_mb__before_atomic();
550 	atomic_inc(&dev->dev_stats_ccnt);
551 }
552 
553 static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
554 					int index)
555 {
556 	btrfs_dev_stat_set(dev, index, 0);
557 }
558 
559 /*
560  * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
561  * can be used as index to access btrfs_raid_array[].
562  */
563 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
564 {
565 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
566 		return BTRFS_RAID_RAID10;
567 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
568 		return BTRFS_RAID_RAID1;
569 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
570 		return BTRFS_RAID_DUP;
571 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
572 		return BTRFS_RAID_RAID0;
573 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
574 		return BTRFS_RAID_RAID5;
575 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
576 		return BTRFS_RAID_RAID6;
577 
578 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
579 }
580 
581 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
582 
583 struct list_head *btrfs_get_fs_uuids(void);
584 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
585 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
586 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
587 					struct btrfs_device *failing_dev);
588 
589 int btrfs_bg_type_to_factor(u64 flags);
590 const char *btrfs_bg_type_to_raid_name(u64 flags);
591 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
592 
593 #endif
594