xref: /openbmc/linux/fs/btrfs/volumes.h (revision ae213c44)
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 list_head post_commit_list; /* chunk mutex */
49 	struct btrfs_fs_devices *fs_devices;
50 	struct btrfs_fs_info *fs_info;
51 
52 	struct rcu_string *name;
53 
54 	u64 generation;
55 
56 	spinlock_t io_lock ____cacheline_aligned;
57 	int running_pending;
58 	/* regular prio bios */
59 	struct btrfs_pending_bios pending_bios;
60 	/* sync bios */
61 	struct btrfs_pending_bios pending_sync_bios;
62 
63 	struct block_device *bdev;
64 
65 	/* the mode sent to blkdev_get */
66 	fmode_t mode;
67 
68 	unsigned long dev_state;
69 	blk_status_t last_flush_error;
70 	int flush_bio_sent;
71 
72 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
73 	seqcount_t data_seqcount;
74 #endif
75 
76 	/* the internal btrfs device id */
77 	u64 devid;
78 
79 	/* size of the device in memory */
80 	u64 total_bytes;
81 
82 	/* size of the device on disk */
83 	u64 disk_total_bytes;
84 
85 	/* bytes used */
86 	u64 bytes_used;
87 
88 	/* optimal io alignment for this device */
89 	u32 io_align;
90 
91 	/* optimal io width for this device */
92 	u32 io_width;
93 	/* type and info about this device */
94 	u64 type;
95 
96 	/* minimal io size for this device */
97 	u32 sector_size;
98 
99 	/* physical drive uuid (or lvm uuid) */
100 	u8 uuid[BTRFS_UUID_SIZE];
101 
102 	/*
103 	 * size of the device on the current transaction
104 	 *
105 	 * This variant is update when committing the transaction,
106 	 * and protected by chunk mutex
107 	 */
108 	u64 commit_total_bytes;
109 
110 	/* bytes used on the current transaction */
111 	u64 commit_bytes_used;
112 
113 	/* for sending down flush barriers */
114 	struct bio *flush_bio;
115 	struct completion flush_wait;
116 
117 	/* per-device scrub information */
118 	struct scrub_ctx *scrub_ctx;
119 
120 	struct btrfs_work work;
121 
122 	/* readahead state */
123 	atomic_t reada_in_flight;
124 	u64 reada_next;
125 	struct reada_zone *reada_curr_zone;
126 	struct radix_tree_root reada_zones;
127 	struct radix_tree_root reada_extents;
128 
129 	/* disk I/O failure stats. For detailed description refer to
130 	 * enum btrfs_dev_stat_values in ioctl.h */
131 	int dev_stats_valid;
132 
133 	/* Counter to record the change of device stats */
134 	atomic_t dev_stats_ccnt;
135 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
136 
137 	struct extent_io_tree alloc_state;
138 };
139 
140 /*
141  * If we read those variants at the context of their own lock, we needn't
142  * use the following helpers, reading them directly is safe.
143  */
144 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
145 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
146 static inline u64							\
147 btrfs_device_get_##name(const struct btrfs_device *dev)			\
148 {									\
149 	u64 size;							\
150 	unsigned int seq;						\
151 									\
152 	do {								\
153 		seq = read_seqcount_begin(&dev->data_seqcount);		\
154 		size = dev->name;					\
155 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
156 	return size;							\
157 }									\
158 									\
159 static inline void							\
160 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
161 {									\
162 	preempt_disable();						\
163 	write_seqcount_begin(&dev->data_seqcount);			\
164 	dev->name = size;						\
165 	write_seqcount_end(&dev->data_seqcount);			\
166 	preempt_enable();						\
167 }
168 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
169 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
170 static inline u64							\
171 btrfs_device_get_##name(const struct btrfs_device *dev)			\
172 {									\
173 	u64 size;							\
174 									\
175 	preempt_disable();						\
176 	size = dev->name;						\
177 	preempt_enable();						\
178 	return size;							\
179 }									\
180 									\
181 static inline void							\
182 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
183 {									\
184 	preempt_disable();						\
185 	dev->name = size;						\
186 	preempt_enable();						\
187 }
188 #else
189 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
190 static inline u64							\
191 btrfs_device_get_##name(const struct btrfs_device *dev)			\
192 {									\
193 	return dev->name;						\
194 }									\
195 									\
196 static inline void							\
197 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
198 {									\
199 	dev->name = size;						\
200 }
201 #endif
202 
203 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
204 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
205 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
206 
207 struct btrfs_fs_devices {
208 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
209 	u8 metadata_uuid[BTRFS_FSID_SIZE];
210 	bool fsid_change;
211 	struct list_head fs_list;
212 
213 	u64 num_devices;
214 	u64 open_devices;
215 	u64 rw_devices;
216 	u64 missing_devices;
217 	u64 total_rw_bytes;
218 	u64 total_devices;
219 
220 	/* Highest generation number of seen devices */
221 	u64 latest_generation;
222 
223 	struct block_device *latest_bdev;
224 
225 	/* all of the devices in the FS, protected by a mutex
226 	 * so we can safely walk it to write out the supers without
227 	 * worrying about add/remove by the multi-device code.
228 	 * Scrubbing super can kick off supers writing by holding
229 	 * this mutex lock.
230 	 */
231 	struct mutex device_list_mutex;
232 	struct list_head devices;
233 
234 	/* devices not currently being allocated */
235 	struct list_head alloc_list;
236 
237 	struct btrfs_fs_devices *seed;
238 	int seeding;
239 
240 	int opened;
241 
242 	/* set when we find or add a device that doesn't have the
243 	 * nonrot flag set
244 	 */
245 	int rotating;
246 
247 	struct btrfs_fs_info *fs_info;
248 	/* sysfs kobjects */
249 	struct kobject fsid_kobj;
250 	struct kobject *device_dir_kobj;
251 	struct completion kobj_unregister;
252 };
253 
254 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
255 
256 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
257 			- sizeof(struct btrfs_chunk))		\
258 			/ sizeof(struct btrfs_stripe) + 1)
259 
260 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
261 				- 2 * sizeof(struct btrfs_disk_key)	\
262 				- 2 * sizeof(struct btrfs_chunk))	\
263 				/ sizeof(struct btrfs_stripe) + 1)
264 
265 /*
266  * we need the mirror number and stripe index to be passed around
267  * the call chain while we are processing end_io (especially errors).
268  * Really, what we need is a btrfs_bio structure that has this info
269  * and is properly sized with its stripe array, but we're not there
270  * quite yet.  We have our own btrfs bioset, and all of the bios
271  * we allocate are actually btrfs_io_bios.  We'll cram as much of
272  * struct btrfs_bio as we can into this over time.
273  */
274 struct btrfs_io_bio {
275 	unsigned int mirror_num;
276 	unsigned int stripe_index;
277 	u64 logical;
278 	u8 *csum;
279 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
280 	struct bvec_iter iter;
281 	/*
282 	 * This member must come last, bio_alloc_bioset will allocate enough
283 	 * bytes for entire btrfs_io_bio but relies on bio being last.
284 	 */
285 	struct bio bio;
286 };
287 
288 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
289 {
290 	return container_of(bio, struct btrfs_io_bio, bio);
291 }
292 
293 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
294 {
295 	if (io_bio->csum != io_bio->csum_inline) {
296 		kfree(io_bio->csum);
297 		io_bio->csum = NULL;
298 	}
299 }
300 
301 struct btrfs_bio_stripe {
302 	struct btrfs_device *dev;
303 	u64 physical;
304 	u64 length; /* only used for discard mappings */
305 };
306 
307 struct btrfs_bio {
308 	refcount_t refs;
309 	atomic_t stripes_pending;
310 	struct btrfs_fs_info *fs_info;
311 	u64 map_type; /* get from map_lookup->type */
312 	bio_end_io_t *end_io;
313 	struct bio *orig_bio;
314 	unsigned long flags;
315 	void *private;
316 	atomic_t error;
317 	int max_errors;
318 	int num_stripes;
319 	int mirror_num;
320 	int num_tgtdevs;
321 	int *tgtdev_map;
322 	/*
323 	 * logical block numbers for the start of each stripe
324 	 * The last one or two are p/q.  These are sorted,
325 	 * so raid_map[0] is the start of our full stripe
326 	 */
327 	u64 *raid_map;
328 	struct btrfs_bio_stripe stripes[];
329 };
330 
331 struct btrfs_device_info {
332 	struct btrfs_device *dev;
333 	u64 dev_offset;
334 	u64 max_avail;
335 	u64 total_avail;
336 };
337 
338 struct btrfs_raid_attr {
339 	int sub_stripes;	/* sub_stripes info for map */
340 	int dev_stripes;	/* stripes per dev */
341 	int devs_max;		/* max devs to use */
342 	int devs_min;		/* min devs needed */
343 	int tolerated_failures; /* max tolerated fail devs */
344 	int devs_increment;	/* ndevs has to be a multiple of this */
345 	int ncopies;		/* how many copies to data has */
346 	int nparity;		/* number of stripes worth of bytes to store
347 				 * parity information */
348 	int mindev_error;	/* error code if min devs requisite is unmet */
349 	const char raid_name[8]; /* name of the raid */
350 	u64 bg_flag;		/* block group flag of the raid */
351 };
352 
353 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
354 
355 struct map_lookup {
356 	u64 type;
357 	int io_align;
358 	int io_width;
359 	u64 stripe_len;
360 	int num_stripes;
361 	int sub_stripes;
362 	int verified_stripes; /* For mount time dev extent verification */
363 	struct btrfs_bio_stripe stripes[];
364 };
365 
366 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
367 			    (sizeof(struct btrfs_bio_stripe) * (n)))
368 
369 struct btrfs_balance_args;
370 struct btrfs_balance_progress;
371 struct btrfs_balance_control {
372 	struct btrfs_balance_args data;
373 	struct btrfs_balance_args meta;
374 	struct btrfs_balance_args sys;
375 
376 	u64 flags;
377 
378 	struct btrfs_balance_progress stat;
379 };
380 
381 enum btrfs_map_op {
382 	BTRFS_MAP_READ,
383 	BTRFS_MAP_WRITE,
384 	BTRFS_MAP_DISCARD,
385 	BTRFS_MAP_GET_READ_MIRRORS,
386 };
387 
388 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
389 {
390 	switch (bio_op(bio)) {
391 	case REQ_OP_DISCARD:
392 		return BTRFS_MAP_DISCARD;
393 	case REQ_OP_WRITE:
394 		return BTRFS_MAP_WRITE;
395 	default:
396 		WARN_ON_ONCE(1);
397 		/* fall through */
398 	case REQ_OP_READ:
399 		return BTRFS_MAP_READ;
400 	}
401 }
402 
403 void btrfs_get_bbio(struct btrfs_bio *bbio);
404 void btrfs_put_bbio(struct btrfs_bio *bbio);
405 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
406 		    u64 logical, u64 *length,
407 		    struct btrfs_bio **bbio_ret, int mirror_num);
408 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
409 		     u64 logical, u64 *length,
410 		     struct btrfs_bio **bbio_ret);
411 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
412 		     u64 physical, u64 **logical, int *naddrs, int *stripe_len);
413 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
414 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
415 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
416 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
417 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
418 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
419 			   int mirror_num, int async_submit);
420 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
421 		       fmode_t flags, void *holder);
422 struct btrfs_device *btrfs_scan_one_device(const char *path,
423 					   fmode_t flags, void *holder);
424 int btrfs_forget_devices(const char *path);
425 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
426 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
427 void btrfs_assign_next_active_device(struct btrfs_device *device,
428 				     struct btrfs_device *this_dev);
429 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
430 						  u64 devid,
431 						  const char *devpath);
432 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
433 					const u64 *devid,
434 					const u8 *uuid);
435 void btrfs_free_device(struct btrfs_device *device);
436 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
437 		    const char *device_path, u64 devid);
438 void __exit btrfs_cleanup_fs_uuids(void);
439 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
440 int btrfs_grow_device(struct btrfs_trans_handle *trans,
441 		      struct btrfs_device *device, u64 new_size);
442 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
443 				       u64 devid, u8 *uuid, u8 *fsid, bool seed);
444 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
445 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
446 int btrfs_balance(struct btrfs_fs_info *fs_info,
447 		  struct btrfs_balance_control *bctl,
448 		  struct btrfs_ioctl_balance_args *bargs);
449 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
450 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
451 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
452 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
453 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
454 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
455 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
456 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
457 int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes,
458 			       u64 search_start, u64 *start, u64 *max_avail);
459 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
460 			 u64 *start, u64 *max_avail);
461 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
462 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
463 			struct btrfs_ioctl_get_dev_stats *stats);
464 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
465 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
466 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
467 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
468 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
469 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
470 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
471 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
472 			   u64 logical, u64 len);
473 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
474 				    u64 logical);
475 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
476 			     u64 chunk_offset, u64 chunk_size);
477 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
478 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
479 				       u64 logical, u64 length);
480 
481 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
482 				      int index)
483 {
484 	atomic_inc(dev->dev_stat_values + index);
485 	/*
486 	 * This memory barrier orders stores updating statistics before stores
487 	 * updating dev_stats_ccnt.
488 	 *
489 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
490 	 */
491 	smp_mb__before_atomic();
492 	atomic_inc(&dev->dev_stats_ccnt);
493 }
494 
495 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
496 				      int index)
497 {
498 	return atomic_read(dev->dev_stat_values + index);
499 }
500 
501 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
502 						int index)
503 {
504 	int ret;
505 
506 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
507 	/*
508 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
509 	 * - RMW operations that have a return value are fully ordered;
510 	 *
511 	 * This implicit memory barriers is paired with the smp_rmb in
512 	 * btrfs_run_dev_stats
513 	 */
514 	atomic_inc(&dev->dev_stats_ccnt);
515 	return ret;
516 }
517 
518 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
519 				      int index, unsigned long val)
520 {
521 	atomic_set(dev->dev_stat_values + index, val);
522 	/*
523 	 * This memory barrier orders stores updating statistics before stores
524 	 * updating dev_stats_ccnt.
525 	 *
526 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
527 	 */
528 	smp_mb__before_atomic();
529 	atomic_inc(&dev->dev_stats_ccnt);
530 }
531 
532 static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
533 					int index)
534 {
535 	btrfs_dev_stat_set(dev, index, 0);
536 }
537 
538 /*
539  * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
540  * can be used as index to access btrfs_raid_array[].
541  */
542 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
543 {
544 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
545 		return BTRFS_RAID_RAID10;
546 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
547 		return BTRFS_RAID_RAID1;
548 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
549 		return BTRFS_RAID_DUP;
550 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
551 		return BTRFS_RAID_RAID0;
552 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
553 		return BTRFS_RAID_RAID5;
554 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
555 		return BTRFS_RAID_RAID6;
556 
557 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
558 }
559 
560 const char *get_raid_name(enum btrfs_raid_types type);
561 
562 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
563 
564 struct list_head *btrfs_get_fs_uuids(void);
565 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
566 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
567 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
568 					struct btrfs_device *failing_dev);
569 
570 int btrfs_bg_type_to_factor(u64 flags);
571 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
572 
573 #endif
574