xref: /openbmc/linux/fs/btrfs/volumes.h (revision 4eb5928d)
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 
54 struct btrfs_device {
55 	struct list_head dev_list; /* device_list_mutex */
56 	struct list_head dev_alloc_list; /* chunk mutex */
57 	struct list_head post_commit_list; /* chunk mutex */
58 	struct btrfs_fs_devices *fs_devices;
59 	struct btrfs_fs_info *fs_info;
60 
61 	struct rcu_string *name;
62 
63 	u64 generation;
64 
65 	struct block_device *bdev;
66 
67 	/* the mode sent to blkdev_get */
68 	fmode_t mode;
69 
70 	unsigned long dev_state;
71 	blk_status_t last_flush_error;
72 
73 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
74 	seqcount_t data_seqcount;
75 #endif
76 
77 	/* the internal btrfs device id */
78 	u64 devid;
79 
80 	/* size of the device in memory */
81 	u64 total_bytes;
82 
83 	/* size of the device on disk */
84 	u64 disk_total_bytes;
85 
86 	/* bytes used */
87 	u64 bytes_used;
88 
89 	/* optimal io alignment for this device */
90 	u32 io_align;
91 
92 	/* optimal io width for this device */
93 	u32 io_width;
94 	/* type and info about this device */
95 	u64 type;
96 
97 	/* minimal io size for this device */
98 	u32 sector_size;
99 
100 	/* physical drive uuid (or lvm uuid) */
101 	u8 uuid[BTRFS_UUID_SIZE];
102 
103 	/*
104 	 * size of the device on the current transaction
105 	 *
106 	 * This variant is update when committing the transaction,
107 	 * and protected by chunk mutex
108 	 */
109 	u64 commit_total_bytes;
110 
111 	/* bytes used on the current transaction */
112 	u64 commit_bytes_used;
113 
114 	/* for sending down flush barriers */
115 	struct bio *flush_bio;
116 	struct completion flush_wait;
117 
118 	/* per-device scrub information */
119 	struct scrub_ctx *scrub_ctx;
120 
121 	/* readahead state */
122 	atomic_t reada_in_flight;
123 	u64 reada_next;
124 	struct reada_zone *reada_curr_zone;
125 	struct radix_tree_root reada_zones;
126 	struct radix_tree_root reada_extents;
127 
128 	/* disk I/O failure stats. For detailed description refer to
129 	 * enum btrfs_dev_stat_values in ioctl.h */
130 	int dev_stats_valid;
131 
132 	/* Counter to record the change of device stats */
133 	atomic_t dev_stats_ccnt;
134 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
135 
136 	struct extent_io_tree alloc_state;
137 
138 	struct completion kobj_unregister;
139 	/* For sysfs/FSID/devinfo/devid/ */
140 	struct kobject devid_kobj;
141 };
142 
143 /*
144  * If we read those variants at the context of their own lock, we needn't
145  * use the following helpers, reading them directly is safe.
146  */
147 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
148 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
149 static inline u64							\
150 btrfs_device_get_##name(const struct btrfs_device *dev)			\
151 {									\
152 	u64 size;							\
153 	unsigned int seq;						\
154 									\
155 	do {								\
156 		seq = read_seqcount_begin(&dev->data_seqcount);		\
157 		size = dev->name;					\
158 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
159 	return size;							\
160 }									\
161 									\
162 static inline void							\
163 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
164 {									\
165 	preempt_disable();						\
166 	write_seqcount_begin(&dev->data_seqcount);			\
167 	dev->name = size;						\
168 	write_seqcount_end(&dev->data_seqcount);			\
169 	preempt_enable();						\
170 }
171 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
172 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
173 static inline u64							\
174 btrfs_device_get_##name(const struct btrfs_device *dev)			\
175 {									\
176 	u64 size;							\
177 									\
178 	preempt_disable();						\
179 	size = dev->name;						\
180 	preempt_enable();						\
181 	return size;							\
182 }									\
183 									\
184 static inline void							\
185 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
186 {									\
187 	preempt_disable();						\
188 	dev->name = size;						\
189 	preempt_enable();						\
190 }
191 #else
192 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
193 static inline u64							\
194 btrfs_device_get_##name(const struct btrfs_device *dev)			\
195 {									\
196 	return dev->name;						\
197 }									\
198 									\
199 static inline void							\
200 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
201 {									\
202 	dev->name = size;						\
203 }
204 #endif
205 
206 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
207 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
208 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
209 
210 enum btrfs_chunk_allocation_policy {
211 	BTRFS_CHUNK_ALLOC_REGULAR,
212 };
213 
214 struct btrfs_fs_devices {
215 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
216 	u8 metadata_uuid[BTRFS_FSID_SIZE];
217 	bool fsid_change;
218 	struct list_head fs_list;
219 
220 	u64 num_devices;
221 	u64 open_devices;
222 	u64 rw_devices;
223 	u64 missing_devices;
224 	u64 total_rw_bytes;
225 	u64 total_devices;
226 
227 	/* Highest generation number of seen devices */
228 	u64 latest_generation;
229 
230 	struct block_device *latest_bdev;
231 
232 	/* all of the devices in the FS, protected by a mutex
233 	 * so we can safely walk it to write out the supers without
234 	 * worrying about add/remove by the multi-device code.
235 	 * Scrubbing super can kick off supers writing by holding
236 	 * this mutex lock.
237 	 */
238 	struct mutex device_list_mutex;
239 
240 	/* List of all devices, protected by device_list_mutex */
241 	struct list_head devices;
242 
243 	/*
244 	 * Devices which can satisfy space allocation. Protected by
245 	 * chunk_mutex
246 	 */
247 	struct list_head alloc_list;
248 
249 	struct btrfs_fs_devices *seed;
250 	bool seeding;
251 
252 	int opened;
253 
254 	/* set when we find or add a device that doesn't have the
255 	 * nonrot flag set
256 	 */
257 	bool rotating;
258 
259 	struct btrfs_fs_info *fs_info;
260 	/* sysfs kobjects */
261 	struct kobject fsid_kobj;
262 	struct kobject *devices_kobj;
263 	struct kobject *devinfo_kobj;
264 	struct completion kobj_unregister;
265 
266 	enum btrfs_chunk_allocation_policy chunk_alloc_policy;
267 };
268 
269 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
270 
271 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
272 			- sizeof(struct btrfs_chunk))		\
273 			/ sizeof(struct btrfs_stripe) + 1)
274 
275 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
276 				- 2 * sizeof(struct btrfs_disk_key)	\
277 				- 2 * sizeof(struct btrfs_chunk))	\
278 				/ sizeof(struct btrfs_stripe) + 1)
279 
280 /*
281  * we need the mirror number and stripe index to be passed around
282  * the call chain while we are processing end_io (especially errors).
283  * Really, what we need is a btrfs_bio structure that has this info
284  * and is properly sized with its stripe array, but we're not there
285  * quite yet.  We have our own btrfs bioset, and all of the bios
286  * we allocate are actually btrfs_io_bios.  We'll cram as much of
287  * struct btrfs_bio as we can into this over time.
288  */
289 struct btrfs_io_bio {
290 	unsigned int mirror_num;
291 	struct btrfs_device *device;
292 	u64 logical;
293 	u8 *csum;
294 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
295 	struct bvec_iter iter;
296 	/*
297 	 * This member must come last, bio_alloc_bioset will allocate enough
298 	 * bytes for entire btrfs_io_bio but relies on bio being last.
299 	 */
300 	struct bio bio;
301 };
302 
303 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
304 {
305 	return container_of(bio, struct btrfs_io_bio, bio);
306 }
307 
308 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
309 {
310 	if (io_bio->csum != io_bio->csum_inline) {
311 		kfree(io_bio->csum);
312 		io_bio->csum = NULL;
313 	}
314 }
315 
316 struct btrfs_bio_stripe {
317 	struct btrfs_device *dev;
318 	u64 physical;
319 	u64 length; /* only used for discard mappings */
320 };
321 
322 struct btrfs_bio {
323 	refcount_t refs;
324 	atomic_t stripes_pending;
325 	struct btrfs_fs_info *fs_info;
326 	u64 map_type; /* get from map_lookup->type */
327 	bio_end_io_t *end_io;
328 	struct bio *orig_bio;
329 	void *private;
330 	atomic_t error;
331 	int max_errors;
332 	int num_stripes;
333 	int mirror_num;
334 	int num_tgtdevs;
335 	int *tgtdev_map;
336 	/*
337 	 * logical block numbers for the start of each stripe
338 	 * The last one or two are p/q.  These are sorted,
339 	 * so raid_map[0] is the start of our full stripe
340 	 */
341 	u64 *raid_map;
342 	struct btrfs_bio_stripe stripes[];
343 };
344 
345 struct btrfs_device_info {
346 	struct btrfs_device *dev;
347 	u64 dev_offset;
348 	u64 max_avail;
349 	u64 total_avail;
350 };
351 
352 struct btrfs_raid_attr {
353 	u8 sub_stripes;		/* sub_stripes info for map */
354 	u8 dev_stripes;		/* stripes per dev */
355 	u8 devs_max;		/* max devs to use */
356 	u8 devs_min;		/* min devs needed */
357 	u8 tolerated_failures;	/* max tolerated fail devs */
358 	u8 devs_increment;	/* ndevs has to be a multiple of this */
359 	u8 ncopies;		/* how many copies to data has */
360 	u8 nparity;		/* number of stripes worth of bytes to store
361 				 * parity information */
362 	u8 mindev_error;	/* error code if min devs requisite is unmet */
363 	const char raid_name[8]; /* name of the raid */
364 	u64 bg_flag;		/* block group flag of the raid */
365 };
366 
367 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
368 
369 struct map_lookup {
370 	u64 type;
371 	int io_align;
372 	int io_width;
373 	u64 stripe_len;
374 	int num_stripes;
375 	int sub_stripes;
376 	int verified_stripes; /* For mount time dev extent verification */
377 	struct btrfs_bio_stripe stripes[];
378 };
379 
380 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
381 			    (sizeof(struct btrfs_bio_stripe) * (n)))
382 
383 struct btrfs_balance_args;
384 struct btrfs_balance_progress;
385 struct btrfs_balance_control {
386 	struct btrfs_balance_args data;
387 	struct btrfs_balance_args meta;
388 	struct btrfs_balance_args sys;
389 
390 	u64 flags;
391 
392 	struct btrfs_balance_progress stat;
393 };
394 
395 enum btrfs_map_op {
396 	BTRFS_MAP_READ,
397 	BTRFS_MAP_WRITE,
398 	BTRFS_MAP_DISCARD,
399 	BTRFS_MAP_GET_READ_MIRRORS,
400 };
401 
402 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
403 {
404 	switch (bio_op(bio)) {
405 	case REQ_OP_DISCARD:
406 		return BTRFS_MAP_DISCARD;
407 	case REQ_OP_WRITE:
408 		return BTRFS_MAP_WRITE;
409 	default:
410 		WARN_ON_ONCE(1);
411 		fallthrough;
412 	case REQ_OP_READ:
413 		return BTRFS_MAP_READ;
414 	}
415 }
416 
417 void btrfs_get_bbio(struct btrfs_bio *bbio);
418 void btrfs_put_bbio(struct btrfs_bio *bbio);
419 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
420 		    u64 logical, u64 *length,
421 		    struct btrfs_bio **bbio_ret, int mirror_num);
422 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
423 		     u64 logical, u64 *length,
424 		     struct btrfs_bio **bbio_ret);
425 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
426 		u64 logical, u64 len, struct btrfs_io_geometry *io_geom);
427 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
428 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
429 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type);
430 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
431 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
432 			   int mirror_num);
433 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
434 		       fmode_t flags, void *holder);
435 struct btrfs_device *btrfs_scan_one_device(const char *path,
436 					   fmode_t flags, void *holder);
437 int btrfs_forget_devices(const char *path);
438 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
439 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
440 void btrfs_assign_next_active_device(struct btrfs_device *device,
441 				     struct btrfs_device *this_dev);
442 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
443 						  u64 devid,
444 						  const char *devpath);
445 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
446 					const u64 *devid,
447 					const u8 *uuid);
448 void btrfs_free_device(struct btrfs_device *device);
449 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
450 		    const char *device_path, u64 devid);
451 void __exit btrfs_cleanup_fs_uuids(void);
452 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
453 int btrfs_grow_device(struct btrfs_trans_handle *trans,
454 		      struct btrfs_device *device, u64 new_size);
455 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
456 				       u64 devid, u8 *uuid, u8 *fsid, bool seed);
457 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
458 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
459 int btrfs_balance(struct btrfs_fs_info *fs_info,
460 		  struct btrfs_balance_control *bctl,
461 		  struct btrfs_ioctl_balance_args *bargs);
462 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
463 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
464 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
465 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
466 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
467 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
468 int btrfs_uuid_scan_kthread(void *data);
469 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
470 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
471 			 u64 *start, u64 *max_avail);
472 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
473 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
474 			struct btrfs_ioctl_get_dev_stats *stats);
475 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
476 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
477 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
478 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
479 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
480 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
481 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
482 			   u64 logical, u64 len);
483 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
484 				    u64 logical);
485 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
486 			     u64 chunk_offset, u64 chunk_size);
487 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
488 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
489 				       u64 logical, u64 length);
490 void btrfs_release_disk_super(struct btrfs_super_block *super);
491 
492 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
493 				      int index)
494 {
495 	atomic_inc(dev->dev_stat_values + index);
496 	/*
497 	 * This memory barrier orders stores updating statistics before stores
498 	 * updating dev_stats_ccnt.
499 	 *
500 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
501 	 */
502 	smp_mb__before_atomic();
503 	atomic_inc(&dev->dev_stats_ccnt);
504 }
505 
506 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
507 				      int index)
508 {
509 	return atomic_read(dev->dev_stat_values + index);
510 }
511 
512 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
513 						int index)
514 {
515 	int ret;
516 
517 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
518 	/*
519 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
520 	 * - RMW operations that have a return value are fully ordered;
521 	 *
522 	 * This implicit memory barriers is paired with the smp_rmb in
523 	 * btrfs_run_dev_stats
524 	 */
525 	atomic_inc(&dev->dev_stats_ccnt);
526 	return ret;
527 }
528 
529 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
530 				      int index, unsigned long val)
531 {
532 	atomic_set(dev->dev_stat_values + index, val);
533 	/*
534 	 * This memory barrier orders stores updating statistics before stores
535 	 * updating dev_stats_ccnt.
536 	 *
537 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
538 	 */
539 	smp_mb__before_atomic();
540 	atomic_inc(&dev->dev_stats_ccnt);
541 }
542 
543 /*
544  * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
545  * can be used as index to access btrfs_raid_array[].
546  */
547 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
548 {
549 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
550 		return BTRFS_RAID_RAID10;
551 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
552 		return BTRFS_RAID_RAID1;
553 	else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
554 		return BTRFS_RAID_RAID1C3;
555 	else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
556 		return BTRFS_RAID_RAID1C4;
557 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
558 		return BTRFS_RAID_DUP;
559 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
560 		return BTRFS_RAID_RAID0;
561 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
562 		return BTRFS_RAID_RAID5;
563 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
564 		return BTRFS_RAID_RAID6;
565 
566 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
567 }
568 
569 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
570 
571 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
572 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
573 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
574 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
575 					struct btrfs_device *failing_dev);
576 
577 int btrfs_bg_type_to_factor(u64 flags);
578 const char *btrfs_bg_type_to_raid_name(u64 flags);
579 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
580 
581 #endif
582