xref: /openbmc/linux/fs/btrfs/volumes.h (revision a44e4f3a)
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 
86 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
87 	seqcount_t data_seqcount;
88 #endif
89 
90 	/* the internal btrfs device id */
91 	u64 devid;
92 
93 	/* size of the device in memory */
94 	u64 total_bytes;
95 
96 	/* size of the device on disk */
97 	u64 disk_total_bytes;
98 
99 	/* bytes used */
100 	u64 bytes_used;
101 
102 	/* optimal io alignment for this device */
103 	u32 io_align;
104 
105 	/* optimal io width for this device */
106 	u32 io_width;
107 	/* type and info about this device */
108 	u64 type;
109 
110 	/* minimal io size for this device */
111 	u32 sector_size;
112 
113 	/* physical drive uuid (or lvm uuid) */
114 	u8 uuid[BTRFS_UUID_SIZE];
115 
116 	/*
117 	 * size of the device on the current transaction
118 	 *
119 	 * This variant is update when committing the transaction,
120 	 * and protected by chunk mutex
121 	 */
122 	u64 commit_total_bytes;
123 
124 	/* bytes used on the current transaction */
125 	u64 commit_bytes_used;
126 
127 	/* for sending down flush barriers */
128 	struct bio *flush_bio;
129 	struct completion flush_wait;
130 
131 	/* per-device scrub information */
132 	struct scrub_ctx *scrub_ctx;
133 
134 	struct btrfs_work work;
135 
136 	/* readahead state */
137 	atomic_t reada_in_flight;
138 	u64 reada_next;
139 	struct reada_zone *reada_curr_zone;
140 	struct radix_tree_root reada_zones;
141 	struct radix_tree_root reada_extents;
142 
143 	/* disk I/O failure stats. For detailed description refer to
144 	 * enum btrfs_dev_stat_values in ioctl.h */
145 	int dev_stats_valid;
146 
147 	/* Counter to record the change of device stats */
148 	atomic_t dev_stats_ccnt;
149 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
150 
151 	struct extent_io_tree alloc_state;
152 };
153 
154 /*
155  * If we read those variants at the context of their own lock, we needn't
156  * use the following helpers, reading them directly is safe.
157  */
158 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
159 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
160 static inline u64							\
161 btrfs_device_get_##name(const struct btrfs_device *dev)			\
162 {									\
163 	u64 size;							\
164 	unsigned int seq;						\
165 									\
166 	do {								\
167 		seq = read_seqcount_begin(&dev->data_seqcount);		\
168 		size = dev->name;					\
169 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
170 	return size;							\
171 }									\
172 									\
173 static inline void							\
174 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
175 {									\
176 	preempt_disable();						\
177 	write_seqcount_begin(&dev->data_seqcount);			\
178 	dev->name = size;						\
179 	write_seqcount_end(&dev->data_seqcount);			\
180 	preempt_enable();						\
181 }
182 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
183 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
184 static inline u64							\
185 btrfs_device_get_##name(const struct btrfs_device *dev)			\
186 {									\
187 	u64 size;							\
188 									\
189 	preempt_disable();						\
190 	size = dev->name;						\
191 	preempt_enable();						\
192 	return size;							\
193 }									\
194 									\
195 static inline void							\
196 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
197 {									\
198 	preempt_disable();						\
199 	dev->name = size;						\
200 	preempt_enable();						\
201 }
202 #else
203 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
204 static inline u64							\
205 btrfs_device_get_##name(const struct btrfs_device *dev)			\
206 {									\
207 	return dev->name;						\
208 }									\
209 									\
210 static inline void							\
211 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
212 {									\
213 	dev->name = size;						\
214 }
215 #endif
216 
217 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
218 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
219 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
220 
221 struct btrfs_fs_devices {
222 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
223 	u8 metadata_uuid[BTRFS_FSID_SIZE];
224 	bool fsid_change;
225 	struct list_head fs_list;
226 
227 	u64 num_devices;
228 	u64 open_devices;
229 	u64 rw_devices;
230 	u64 missing_devices;
231 	u64 total_rw_bytes;
232 	u64 total_devices;
233 
234 	/* Highest generation number of seen devices */
235 	u64 latest_generation;
236 
237 	struct block_device *latest_bdev;
238 
239 	/* all of the devices in the FS, protected by a mutex
240 	 * so we can safely walk it to write out the supers without
241 	 * worrying about add/remove by the multi-device code.
242 	 * Scrubbing super can kick off supers writing by holding
243 	 * this mutex lock.
244 	 */
245 	struct mutex device_list_mutex;
246 
247 	/* List of all devices, protected by device_list_mutex */
248 	struct list_head devices;
249 
250 	/*
251 	 * Devices which can satisfy space allocation. Protected by
252 	 * chunk_mutex
253 	 */
254 	struct list_head alloc_list;
255 
256 	struct btrfs_fs_devices *seed;
257 	int seeding;
258 
259 	int opened;
260 
261 	/* set when we find or add a device that doesn't have the
262 	 * nonrot flag set
263 	 */
264 	int rotating;
265 
266 	struct btrfs_fs_info *fs_info;
267 	/* sysfs kobjects */
268 	struct kobject fsid_kobj;
269 	struct kobject *device_dir_kobj;
270 	struct completion kobj_unregister;
271 };
272 
273 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
274 
275 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
276 			- sizeof(struct btrfs_chunk))		\
277 			/ sizeof(struct btrfs_stripe) + 1)
278 
279 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
280 				- 2 * sizeof(struct btrfs_disk_key)	\
281 				- 2 * sizeof(struct btrfs_chunk))	\
282 				/ sizeof(struct btrfs_stripe) + 1)
283 
284 /*
285  * we need the mirror number and stripe index to be passed around
286  * the call chain while we are processing end_io (especially errors).
287  * Really, what we need is a btrfs_bio structure that has this info
288  * and is properly sized with its stripe array, but we're not there
289  * quite yet.  We have our own btrfs bioset, and all of the bios
290  * we allocate are actually btrfs_io_bios.  We'll cram as much of
291  * struct btrfs_bio as we can into this over time.
292  */
293 struct btrfs_io_bio {
294 	unsigned int mirror_num;
295 	unsigned int stripe_index;
296 	u64 logical;
297 	u8 *csum;
298 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
299 	struct bvec_iter iter;
300 	/*
301 	 * This member must come last, bio_alloc_bioset will allocate enough
302 	 * bytes for entire btrfs_io_bio but relies on bio being last.
303 	 */
304 	struct bio bio;
305 };
306 
307 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
308 {
309 	return container_of(bio, struct btrfs_io_bio, bio);
310 }
311 
312 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
313 {
314 	if (io_bio->csum != io_bio->csum_inline) {
315 		kfree(io_bio->csum);
316 		io_bio->csum = NULL;
317 	}
318 }
319 
320 struct btrfs_bio_stripe {
321 	struct btrfs_device *dev;
322 	u64 physical;
323 	u64 length; /* only used for discard mappings */
324 };
325 
326 struct btrfs_bio {
327 	refcount_t refs;
328 	atomic_t stripes_pending;
329 	struct btrfs_fs_info *fs_info;
330 	u64 map_type; /* get from map_lookup->type */
331 	bio_end_io_t *end_io;
332 	struct bio *orig_bio;
333 	unsigned long flags;
334 	void *private;
335 	atomic_t error;
336 	int max_errors;
337 	int num_stripes;
338 	int mirror_num;
339 	int num_tgtdevs;
340 	int *tgtdev_map;
341 	/*
342 	 * logical block numbers for the start of each stripe
343 	 * The last one or two are p/q.  These are sorted,
344 	 * so raid_map[0] is the start of our full stripe
345 	 */
346 	u64 *raid_map;
347 	struct btrfs_bio_stripe stripes[];
348 };
349 
350 struct btrfs_device_info {
351 	struct btrfs_device *dev;
352 	u64 dev_offset;
353 	u64 max_avail;
354 	u64 total_avail;
355 };
356 
357 struct btrfs_raid_attr {
358 	u8 sub_stripes;		/* sub_stripes info for map */
359 	u8 dev_stripes;		/* stripes per dev */
360 	u8 devs_max;		/* max devs to use */
361 	u8 devs_min;		/* min devs needed */
362 	u8 tolerated_failures;	/* max tolerated fail devs */
363 	u8 devs_increment;	/* ndevs has to be a multiple of this */
364 	u8 ncopies;		/* how many copies to data has */
365 	u8 nparity;		/* number of stripes worth of bytes to store
366 				 * parity information */
367 	u8 mindev_error;	/* error code if min devs requisite is unmet */
368 	const char raid_name[8]; /* name of the raid */
369 	u64 bg_flag;		/* block group flag of the raid */
370 };
371 
372 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
373 
374 struct map_lookup {
375 	u64 type;
376 	int io_align;
377 	int io_width;
378 	u64 stripe_len;
379 	int num_stripes;
380 	int sub_stripes;
381 	int verified_stripes; /* For mount time dev extent verification */
382 	struct btrfs_bio_stripe stripes[];
383 };
384 
385 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
386 			    (sizeof(struct btrfs_bio_stripe) * (n)))
387 
388 struct btrfs_balance_args;
389 struct btrfs_balance_progress;
390 struct btrfs_balance_control {
391 	struct btrfs_balance_args data;
392 	struct btrfs_balance_args meta;
393 	struct btrfs_balance_args sys;
394 
395 	u64 flags;
396 
397 	struct btrfs_balance_progress stat;
398 };
399 
400 enum btrfs_map_op {
401 	BTRFS_MAP_READ,
402 	BTRFS_MAP_WRITE,
403 	BTRFS_MAP_DISCARD,
404 	BTRFS_MAP_GET_READ_MIRRORS,
405 };
406 
407 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
408 {
409 	switch (bio_op(bio)) {
410 	case REQ_OP_DISCARD:
411 		return BTRFS_MAP_DISCARD;
412 	case REQ_OP_WRITE:
413 		return BTRFS_MAP_WRITE;
414 	default:
415 		WARN_ON_ONCE(1);
416 		/* fall through */
417 	case REQ_OP_READ:
418 		return BTRFS_MAP_READ;
419 	}
420 }
421 
422 void btrfs_get_bbio(struct btrfs_bio *bbio);
423 void btrfs_put_bbio(struct btrfs_bio *bbio);
424 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
425 		    u64 logical, u64 *length,
426 		    struct btrfs_bio **bbio_ret, int mirror_num);
427 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
428 		     u64 logical, u64 *length,
429 		     struct btrfs_bio **bbio_ret);
430 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
431 		u64 logical, u64 len, struct btrfs_io_geometry *io_geom);
432 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
433 		     u64 physical, u64 **logical, int *naddrs, int *stripe_len);
434 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
435 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
436 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
437 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
438 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
439 			   int mirror_num, int async_submit);
440 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
441 		       fmode_t flags, void *holder);
442 struct btrfs_device *btrfs_scan_one_device(const char *path,
443 					   fmode_t flags, void *holder);
444 int btrfs_forget_devices(const char *path);
445 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
446 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
447 void btrfs_assign_next_active_device(struct btrfs_device *device,
448 				     struct btrfs_device *this_dev);
449 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
450 						  u64 devid,
451 						  const char *devpath);
452 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
453 					const u64 *devid,
454 					const u8 *uuid);
455 void btrfs_free_device(struct btrfs_device *device);
456 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
457 		    const char *device_path, u64 devid);
458 void __exit btrfs_cleanup_fs_uuids(void);
459 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
460 int btrfs_grow_device(struct btrfs_trans_handle *trans,
461 		      struct btrfs_device *device, u64 new_size);
462 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
463 				       u64 devid, u8 *uuid, u8 *fsid, bool seed);
464 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
465 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
466 int btrfs_balance(struct btrfs_fs_info *fs_info,
467 		  struct btrfs_balance_control *bctl,
468 		  struct btrfs_ioctl_balance_args *bargs);
469 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
470 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
471 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
472 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
473 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
474 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
475 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
476 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
477 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
478 			 u64 *start, u64 *max_avail);
479 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
480 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
481 			struct btrfs_ioctl_get_dev_stats *stats);
482 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
483 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
484 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
485 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
486 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
487 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
488 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
489 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
490 			   u64 logical, u64 len);
491 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
492 				    u64 logical);
493 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
494 			     u64 chunk_offset, u64 chunk_size);
495 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
496 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
497 				       u64 logical, u64 length);
498 
499 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
500 				      int index)
501 {
502 	atomic_inc(dev->dev_stat_values + index);
503 	/*
504 	 * This memory barrier orders stores updating statistics before stores
505 	 * updating dev_stats_ccnt.
506 	 *
507 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
508 	 */
509 	smp_mb__before_atomic();
510 	atomic_inc(&dev->dev_stats_ccnt);
511 }
512 
513 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
514 				      int index)
515 {
516 	return atomic_read(dev->dev_stat_values + index);
517 }
518 
519 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
520 						int index)
521 {
522 	int ret;
523 
524 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
525 	/*
526 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
527 	 * - RMW operations that have a return value are fully ordered;
528 	 *
529 	 * This implicit memory barriers is paired with the smp_rmb in
530 	 * btrfs_run_dev_stats
531 	 */
532 	atomic_inc(&dev->dev_stats_ccnt);
533 	return ret;
534 }
535 
536 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
537 				      int index, unsigned long val)
538 {
539 	atomic_set(dev->dev_stat_values + index, val);
540 	/*
541 	 * This memory barrier orders stores updating statistics before stores
542 	 * updating dev_stats_ccnt.
543 	 *
544 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
545 	 */
546 	smp_mb__before_atomic();
547 	atomic_inc(&dev->dev_stats_ccnt);
548 }
549 
550 /*
551  * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
552  * can be used as index to access btrfs_raid_array[].
553  */
554 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
555 {
556 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
557 		return BTRFS_RAID_RAID10;
558 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
559 		return BTRFS_RAID_RAID1;
560 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
561 		return BTRFS_RAID_DUP;
562 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
563 		return BTRFS_RAID_RAID0;
564 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
565 		return BTRFS_RAID_RAID5;
566 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
567 		return BTRFS_RAID_RAID6;
568 
569 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
570 }
571 
572 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
573 
574 struct list_head *btrfs_get_fs_uuids(void);
575 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
576 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
577 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
578 					struct btrfs_device *failing_dev);
579 
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 
584 #endif
585