xref: /openbmc/linux/fs/btrfs/volumes.h (revision 7f877908)
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 
22 struct btrfs_io_geometry {
23 	/* remaining bytes before crossing a stripe */
24 	u64 len;
25 	/* offset of logical address in chunk */
26 	u64 offset;
27 	/* length of single IO stripe */
28 	u64 stripe_len;
29 	/* number of stripe where address falls */
30 	u64 stripe_nr;
31 	/* offset of address in stripe */
32 	u64 stripe_offset;
33 	/* offset of raid56 stripe into the chunk */
34 	u64 raid56_stripe_offset;
35 };
36 
37 /*
38  * Use sequence counter to get consistent device stat data on
39  * 32-bit processors.
40  */
41 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
42 #include <linux/seqlock.h>
43 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
44 #define btrfs_device_data_ordered_init(device)	\
45 	seqcount_init(&device->data_seqcount)
46 #else
47 #define btrfs_device_data_ordered_init(device) do { } while (0)
48 #endif
49 
50 #define BTRFS_DEV_STATE_WRITEABLE	(0)
51 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
52 #define BTRFS_DEV_STATE_MISSING		(2)
53 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
54 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
55 
56 struct btrfs_device {
57 	struct list_head dev_list; /* device_list_mutex */
58 	struct list_head dev_alloc_list; /* chunk mutex */
59 	struct list_head post_commit_list; /* chunk mutex */
60 	struct btrfs_fs_devices *fs_devices;
61 	struct btrfs_fs_info *fs_info;
62 
63 	struct rcu_string *name;
64 
65 	u64 generation;
66 
67 	struct block_device *bdev;
68 
69 	/* the mode sent to blkdev_get */
70 	fmode_t mode;
71 
72 	unsigned long dev_state;
73 	blk_status_t last_flush_error;
74 
75 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
76 	seqcount_t data_seqcount;
77 #endif
78 
79 	/* the internal btrfs device id */
80 	u64 devid;
81 
82 	/* size of the device in memory */
83 	u64 total_bytes;
84 
85 	/* size of the device on disk */
86 	u64 disk_total_bytes;
87 
88 	/* bytes used */
89 	u64 bytes_used;
90 
91 	/* optimal io alignment for this device */
92 	u32 io_align;
93 
94 	/* optimal io width for this device */
95 	u32 io_width;
96 	/* type and info about this device */
97 	u64 type;
98 
99 	/* minimal io size for this device */
100 	u32 sector_size;
101 
102 	/* physical drive uuid (or lvm uuid) */
103 	u8 uuid[BTRFS_UUID_SIZE];
104 
105 	/*
106 	 * size of the device on the current transaction
107 	 *
108 	 * This variant is update when committing the transaction,
109 	 * and protected by chunk mutex
110 	 */
111 	u64 commit_total_bytes;
112 
113 	/* bytes used on the current transaction */
114 	u64 commit_bytes_used;
115 
116 	/* for sending down flush barriers */
117 	struct bio *flush_bio;
118 	struct completion flush_wait;
119 
120 	/* per-device scrub information */
121 	struct scrub_ctx *scrub_ctx;
122 
123 	/* readahead state */
124 	atomic_t reada_in_flight;
125 	u64 reada_next;
126 	struct reada_zone *reada_curr_zone;
127 	struct radix_tree_root reada_zones;
128 	struct radix_tree_root reada_extents;
129 
130 	/* disk I/O failure stats. For detailed description refer to
131 	 * enum btrfs_dev_stat_values in ioctl.h */
132 	int dev_stats_valid;
133 
134 	/* Counter to record the change of device stats */
135 	atomic_t dev_stats_ccnt;
136 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
137 
138 	struct extent_io_tree alloc_state;
139 
140 	struct completion kobj_unregister;
141 	/* For sysfs/FSID/devinfo/devid/ */
142 	struct kobject devid_kobj;
143 };
144 
145 /*
146  * If we read those variants at the context of their own lock, we needn't
147  * use the following helpers, reading them directly is safe.
148  */
149 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
150 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
151 static inline u64							\
152 btrfs_device_get_##name(const struct btrfs_device *dev)			\
153 {									\
154 	u64 size;							\
155 	unsigned int seq;						\
156 									\
157 	do {								\
158 		seq = read_seqcount_begin(&dev->data_seqcount);		\
159 		size = dev->name;					\
160 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
161 	return size;							\
162 }									\
163 									\
164 static inline void							\
165 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
166 {									\
167 	preempt_disable();						\
168 	write_seqcount_begin(&dev->data_seqcount);			\
169 	dev->name = size;						\
170 	write_seqcount_end(&dev->data_seqcount);			\
171 	preempt_enable();						\
172 }
173 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
174 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
175 static inline u64							\
176 btrfs_device_get_##name(const struct btrfs_device *dev)			\
177 {									\
178 	u64 size;							\
179 									\
180 	preempt_disable();						\
181 	size = dev->name;						\
182 	preempt_enable();						\
183 	return size;							\
184 }									\
185 									\
186 static inline void							\
187 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
188 {									\
189 	preempt_disable();						\
190 	dev->name = size;						\
191 	preempt_enable();						\
192 }
193 #else
194 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
195 static inline u64							\
196 btrfs_device_get_##name(const struct btrfs_device *dev)			\
197 {									\
198 	return dev->name;						\
199 }									\
200 									\
201 static inline void							\
202 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
203 {									\
204 	dev->name = size;						\
205 }
206 #endif
207 
208 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
209 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
210 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
211 
212 struct btrfs_fs_devices {
213 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
214 	u8 metadata_uuid[BTRFS_FSID_SIZE];
215 	bool fsid_change;
216 	struct list_head fs_list;
217 
218 	u64 num_devices;
219 	u64 open_devices;
220 	u64 rw_devices;
221 	u64 missing_devices;
222 	u64 total_rw_bytes;
223 	u64 total_devices;
224 
225 	/* Highest generation number of seen devices */
226 	u64 latest_generation;
227 
228 	struct block_device *latest_bdev;
229 
230 	/* all of the devices in the FS, protected by a mutex
231 	 * so we can safely walk it to write out the supers without
232 	 * worrying about add/remove by the multi-device code.
233 	 * Scrubbing super can kick off supers writing by holding
234 	 * this mutex lock.
235 	 */
236 	struct mutex device_list_mutex;
237 
238 	/* List of all devices, protected by device_list_mutex */
239 	struct list_head devices;
240 
241 	/*
242 	 * Devices which can satisfy space allocation. Protected by
243 	 * chunk_mutex
244 	 */
245 	struct list_head alloc_list;
246 
247 	struct btrfs_fs_devices *seed;
248 	bool seeding;
249 
250 	int opened;
251 
252 	/* set when we find or add a device that doesn't have the
253 	 * nonrot flag set
254 	 */
255 	bool rotating;
256 
257 	struct btrfs_fs_info *fs_info;
258 	/* sysfs kobjects */
259 	struct kobject fsid_kobj;
260 	struct kobject *devices_kobj;
261 	struct completion kobj_unregister;
262 };
263 
264 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
265 
266 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
267 			- sizeof(struct btrfs_chunk))		\
268 			/ sizeof(struct btrfs_stripe) + 1)
269 
270 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
271 				- 2 * sizeof(struct btrfs_disk_key)	\
272 				- 2 * sizeof(struct btrfs_chunk))	\
273 				/ sizeof(struct btrfs_stripe) + 1)
274 
275 /*
276  * we need the mirror number and stripe index to be passed around
277  * the call chain while we are processing end_io (especially errors).
278  * Really, what we need is a btrfs_bio structure that has this info
279  * and is properly sized with its stripe array, but we're not there
280  * quite yet.  We have our own btrfs bioset, and all of the bios
281  * we allocate are actually btrfs_io_bios.  We'll cram as much of
282  * struct btrfs_bio as we can into this over time.
283  */
284 struct btrfs_io_bio {
285 	unsigned int mirror_num;
286 	unsigned int stripe_index;
287 	u64 logical;
288 	u8 *csum;
289 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
290 	struct bvec_iter iter;
291 	/*
292 	 * This member must come last, bio_alloc_bioset will allocate enough
293 	 * bytes for entire btrfs_io_bio but relies on bio being last.
294 	 */
295 	struct bio bio;
296 };
297 
298 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
299 {
300 	return container_of(bio, struct btrfs_io_bio, bio);
301 }
302 
303 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
304 {
305 	if (io_bio->csum != io_bio->csum_inline) {
306 		kfree(io_bio->csum);
307 		io_bio->csum = NULL;
308 	}
309 }
310 
311 struct btrfs_bio_stripe {
312 	struct btrfs_device *dev;
313 	u64 physical;
314 	u64 length; /* only used for discard mappings */
315 };
316 
317 struct btrfs_bio {
318 	refcount_t refs;
319 	atomic_t stripes_pending;
320 	struct btrfs_fs_info *fs_info;
321 	u64 map_type; /* get from map_lookup->type */
322 	bio_end_io_t *end_io;
323 	struct bio *orig_bio;
324 	void *private;
325 	atomic_t error;
326 	int max_errors;
327 	int num_stripes;
328 	int mirror_num;
329 	int num_tgtdevs;
330 	int *tgtdev_map;
331 	/*
332 	 * logical block numbers for the start of each stripe
333 	 * The last one or two are p/q.  These are sorted,
334 	 * so raid_map[0] is the start of our full stripe
335 	 */
336 	u64 *raid_map;
337 	struct btrfs_bio_stripe stripes[];
338 };
339 
340 struct btrfs_device_info {
341 	struct btrfs_device *dev;
342 	u64 dev_offset;
343 	u64 max_avail;
344 	u64 total_avail;
345 };
346 
347 struct btrfs_raid_attr {
348 	u8 sub_stripes;		/* sub_stripes info for map */
349 	u8 dev_stripes;		/* stripes per dev */
350 	u8 devs_max;		/* max devs to use */
351 	u8 devs_min;		/* min devs needed */
352 	u8 tolerated_failures;	/* max tolerated fail devs */
353 	u8 devs_increment;	/* ndevs has to be a multiple of this */
354 	u8 ncopies;		/* how many copies to data has */
355 	u8 nparity;		/* number of stripes worth of bytes to store
356 				 * parity information */
357 	u8 mindev_error;	/* error code if min devs requisite is unmet */
358 	const char raid_name[8]; /* name of the raid */
359 	u64 bg_flag;		/* block group flag of the raid */
360 };
361 
362 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
363 
364 struct map_lookup {
365 	u64 type;
366 	int io_align;
367 	int io_width;
368 	u64 stripe_len;
369 	int num_stripes;
370 	int sub_stripes;
371 	int verified_stripes; /* For mount time dev extent verification */
372 	struct btrfs_bio_stripe stripes[];
373 };
374 
375 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
376 			    (sizeof(struct btrfs_bio_stripe) * (n)))
377 
378 struct btrfs_balance_args;
379 struct btrfs_balance_progress;
380 struct btrfs_balance_control {
381 	struct btrfs_balance_args data;
382 	struct btrfs_balance_args meta;
383 	struct btrfs_balance_args sys;
384 
385 	u64 flags;
386 
387 	struct btrfs_balance_progress stat;
388 };
389 
390 enum btrfs_map_op {
391 	BTRFS_MAP_READ,
392 	BTRFS_MAP_WRITE,
393 	BTRFS_MAP_DISCARD,
394 	BTRFS_MAP_GET_READ_MIRRORS,
395 };
396 
397 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
398 {
399 	switch (bio_op(bio)) {
400 	case REQ_OP_DISCARD:
401 		return BTRFS_MAP_DISCARD;
402 	case REQ_OP_WRITE:
403 		return BTRFS_MAP_WRITE;
404 	default:
405 		WARN_ON_ONCE(1);
406 		/* fall through */
407 	case REQ_OP_READ:
408 		return BTRFS_MAP_READ;
409 	}
410 }
411 
412 void btrfs_get_bbio(struct btrfs_bio *bbio);
413 void btrfs_put_bbio(struct btrfs_bio *bbio);
414 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
415 		    u64 logical, u64 *length,
416 		    struct btrfs_bio **bbio_ret, int mirror_num);
417 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
418 		     u64 logical, u64 *length,
419 		     struct btrfs_bio **bbio_ret);
420 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
421 		u64 logical, u64 len, struct btrfs_io_geometry *io_geom);
422 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
423 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
424 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
425 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
426 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
427 			   int mirror_num);
428 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
429 		       fmode_t flags, void *holder);
430 struct btrfs_device *btrfs_scan_one_device(const char *path,
431 					   fmode_t flags, void *holder);
432 int btrfs_forget_devices(const char *path);
433 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
434 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
435 void btrfs_assign_next_active_device(struct btrfs_device *device,
436 				     struct btrfs_device *this_dev);
437 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
438 						  u64 devid,
439 						  const char *devpath);
440 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
441 					const u64 *devid,
442 					const u8 *uuid);
443 void btrfs_free_device(struct btrfs_device *device);
444 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
445 		    const char *device_path, u64 devid);
446 void __exit btrfs_cleanup_fs_uuids(void);
447 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
448 int btrfs_grow_device(struct btrfs_trans_handle *trans,
449 		      struct btrfs_device *device, u64 new_size);
450 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
451 				       u64 devid, u8 *uuid, u8 *fsid, bool seed);
452 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
453 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
454 int btrfs_balance(struct btrfs_fs_info *fs_info,
455 		  struct btrfs_balance_control *bctl,
456 		  struct btrfs_ioctl_balance_args *bargs);
457 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
458 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
459 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
460 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
461 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
462 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
463 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
464 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
465 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
466 			 u64 *start, u64 *max_avail);
467 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
468 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
469 			struct btrfs_ioctl_get_dev_stats *stats);
470 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
471 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
472 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
473 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
474 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
475 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
476 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
477 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
478 			   u64 logical, u64 len);
479 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
480 				    u64 logical);
481 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
482 			     u64 chunk_offset, u64 chunk_size);
483 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
484 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
485 				       u64 logical, u64 length);
486 
487 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
488 				      int index)
489 {
490 	atomic_inc(dev->dev_stat_values + index);
491 	/*
492 	 * This memory barrier orders stores updating statistics before stores
493 	 * updating dev_stats_ccnt.
494 	 *
495 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
496 	 */
497 	smp_mb__before_atomic();
498 	atomic_inc(&dev->dev_stats_ccnt);
499 }
500 
501 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
502 				      int index)
503 {
504 	return atomic_read(dev->dev_stat_values + index);
505 }
506 
507 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
508 						int index)
509 {
510 	int ret;
511 
512 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
513 	/*
514 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
515 	 * - RMW operations that have a return value are fully ordered;
516 	 *
517 	 * This implicit memory barriers is paired with the smp_rmb in
518 	 * btrfs_run_dev_stats
519 	 */
520 	atomic_inc(&dev->dev_stats_ccnt);
521 	return ret;
522 }
523 
524 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
525 				      int index, unsigned long val)
526 {
527 	atomic_set(dev->dev_stat_values + index, val);
528 	/*
529 	 * This memory barrier orders stores updating statistics before stores
530 	 * updating dev_stats_ccnt.
531 	 *
532 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
533 	 */
534 	smp_mb__before_atomic();
535 	atomic_inc(&dev->dev_stats_ccnt);
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_RAID1C3)
549 		return BTRFS_RAID_RAID1C3;
550 	else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
551 		return BTRFS_RAID_RAID1C4;
552 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
553 		return BTRFS_RAID_DUP;
554 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
555 		return BTRFS_RAID_RAID0;
556 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
557 		return BTRFS_RAID_RAID5;
558 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
559 		return BTRFS_RAID_RAID6;
560 
561 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
562 }
563 
564 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
565 
566 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
567 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
568 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
569 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
570 					struct btrfs_device *failing_dev);
571 
572 int btrfs_bg_type_to_factor(u64 flags);
573 const char *btrfs_bg_type_to_raid_name(u64 flags);
574 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
575 
576 #endif
577