xref: /openbmc/linux/fs/btrfs/volumes.h (revision 2359ccdd)
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 extern struct mutex uuid_mutex;
15 
16 #define BTRFS_STRIPE_LEN	SZ_64K
17 
18 struct buffer_head;
19 struct btrfs_pending_bios {
20 	struct bio *head;
21 	struct bio *tail;
22 };
23 
24 /*
25  * Use sequence counter to get consistent device stat data on
26  * 32-bit processors.
27  */
28 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
29 #include <linux/seqlock.h>
30 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
31 #define btrfs_device_data_ordered_init(device)	\
32 	seqcount_init(&device->data_seqcount)
33 #else
34 #define btrfs_device_data_ordered_init(device) do { } while (0)
35 #endif
36 
37 #define BTRFS_DEV_STATE_WRITEABLE	(0)
38 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
39 #define BTRFS_DEV_STATE_MISSING		(2)
40 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
41 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
42 
43 struct btrfs_device {
44 	struct list_head dev_list;
45 	struct list_head dev_alloc_list;
46 	struct btrfs_fs_devices *fs_devices;
47 	struct btrfs_fs_info *fs_info;
48 
49 	struct rcu_string *name;
50 
51 	u64 generation;
52 
53 	spinlock_t io_lock ____cacheline_aligned;
54 	int running_pending;
55 	/* regular prio bios */
56 	struct btrfs_pending_bios pending_bios;
57 	/* sync bios */
58 	struct btrfs_pending_bios pending_sync_bios;
59 
60 	struct block_device *bdev;
61 
62 	/* the mode sent to blkdev_get */
63 	fmode_t mode;
64 
65 	unsigned long dev_state;
66 	blk_status_t last_flush_error;
67 	int flush_bio_sent;
68 
69 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
70 	seqcount_t data_seqcount;
71 #endif
72 
73 	/* the internal btrfs device id */
74 	u64 devid;
75 
76 	/* size of the device in memory */
77 	u64 total_bytes;
78 
79 	/* size of the device on disk */
80 	u64 disk_total_bytes;
81 
82 	/* bytes used */
83 	u64 bytes_used;
84 
85 	/* optimal io alignment for this device */
86 	u32 io_align;
87 
88 	/* optimal io width for this device */
89 	u32 io_width;
90 	/* type and info about this device */
91 	u64 type;
92 
93 	/* minimal io size for this device */
94 	u32 sector_size;
95 
96 	/* physical drive uuid (or lvm uuid) */
97 	u8 uuid[BTRFS_UUID_SIZE];
98 
99 	/*
100 	 * size of the device on the current transaction
101 	 *
102 	 * This variant is update when committing the transaction,
103 	 * and protected by device_list_mutex
104 	 */
105 	u64 commit_total_bytes;
106 
107 	/* bytes used on the current transaction */
108 	u64 commit_bytes_used;
109 	/*
110 	 * used to manage the device which is resized
111 	 *
112 	 * It is protected by chunk_lock.
113 	 */
114 	struct list_head resized_list;
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 	struct btrfs_work work;
124 	struct rcu_head rcu;
125 
126 	/* readahead state */
127 	atomic_t reada_in_flight;
128 	u64 reada_next;
129 	struct reada_zone *reada_curr_zone;
130 	struct radix_tree_root reada_zones;
131 	struct radix_tree_root reada_extents;
132 
133 	/* disk I/O failure stats. For detailed description refer to
134 	 * enum btrfs_dev_stat_values in ioctl.h */
135 	int dev_stats_valid;
136 
137 	/* Counter to record the change of device stats */
138 	atomic_t dev_stats_ccnt;
139 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
140 };
141 
142 /*
143  * If we read those variants at the context of their own lock, we needn't
144  * use the following helpers, reading them directly is safe.
145  */
146 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
147 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
148 static inline u64							\
149 btrfs_device_get_##name(const struct btrfs_device *dev)			\
150 {									\
151 	u64 size;							\
152 	unsigned int seq;						\
153 									\
154 	do {								\
155 		seq = read_seqcount_begin(&dev->data_seqcount);		\
156 		size = dev->name;					\
157 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
158 	return size;							\
159 }									\
160 									\
161 static inline void							\
162 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
163 {									\
164 	preempt_disable();						\
165 	write_seqcount_begin(&dev->data_seqcount);			\
166 	dev->name = size;						\
167 	write_seqcount_end(&dev->data_seqcount);			\
168 	preempt_enable();						\
169 }
170 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
171 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
172 static inline u64							\
173 btrfs_device_get_##name(const struct btrfs_device *dev)			\
174 {									\
175 	u64 size;							\
176 									\
177 	preempt_disable();						\
178 	size = dev->name;						\
179 	preempt_enable();						\
180 	return size;							\
181 }									\
182 									\
183 static inline void							\
184 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
185 {									\
186 	preempt_disable();						\
187 	dev->name = size;						\
188 	preempt_enable();						\
189 }
190 #else
191 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
192 static inline u64							\
193 btrfs_device_get_##name(const struct btrfs_device *dev)			\
194 {									\
195 	return dev->name;						\
196 }									\
197 									\
198 static inline void							\
199 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
200 {									\
201 	dev->name = size;						\
202 }
203 #endif
204 
205 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
206 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
207 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
208 
209 struct btrfs_fs_devices {
210 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
211 
212 	u64 num_devices;
213 	u64 open_devices;
214 	u64 rw_devices;
215 	u64 missing_devices;
216 	u64 total_rw_bytes;
217 	u64 total_devices;
218 	struct block_device *latest_bdev;
219 
220 	/* all of the devices in the FS, protected by a mutex
221 	 * so we can safely walk it to write out the supers without
222 	 * worrying about add/remove by the multi-device code.
223 	 * Scrubbing super can kick off supers writing by holding
224 	 * this mutex lock.
225 	 */
226 	struct mutex device_list_mutex;
227 	struct list_head devices;
228 
229 	struct list_head resized_devices;
230 	/* devices not currently being allocated */
231 	struct list_head alloc_list;
232 	struct list_head list;
233 
234 	struct btrfs_fs_devices *seed;
235 	int seeding;
236 
237 	int opened;
238 
239 	/* set when we find or add a device that doesn't have the
240 	 * nonrot flag set
241 	 */
242 	int rotating;
243 
244 	struct btrfs_fs_info *fs_info;
245 	/* sysfs kobjects */
246 	struct kobject fsid_kobj;
247 	struct kobject *device_dir_kobj;
248 	struct completion kobj_unregister;
249 };
250 
251 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
252 
253 /*
254  * we need the mirror number and stripe index to be passed around
255  * the call chain while we are processing end_io (especially errors).
256  * Really, what we need is a btrfs_bio structure that has this info
257  * and is properly sized with its stripe array, but we're not there
258  * quite yet.  We have our own btrfs bioset, and all of the bios
259  * we allocate are actually btrfs_io_bios.  We'll cram as much of
260  * struct btrfs_bio as we can into this over time.
261  */
262 typedef void (btrfs_io_bio_end_io_t) (struct btrfs_io_bio *bio, int err);
263 struct btrfs_io_bio {
264 	unsigned int mirror_num;
265 	unsigned int stripe_index;
266 	u64 logical;
267 	u8 *csum;
268 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
269 	u8 *csum_allocated;
270 	btrfs_io_bio_end_io_t *end_io;
271 	struct bvec_iter iter;
272 	/*
273 	 * This member must come last, bio_alloc_bioset will allocate enough
274 	 * bytes for entire btrfs_io_bio but relies on bio being last.
275 	 */
276 	struct bio bio;
277 };
278 
279 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
280 {
281 	return container_of(bio, struct btrfs_io_bio, bio);
282 }
283 
284 struct btrfs_bio_stripe {
285 	struct btrfs_device *dev;
286 	u64 physical;
287 	u64 length; /* only used for discard mappings */
288 };
289 
290 struct btrfs_bio;
291 typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
292 
293 struct btrfs_bio {
294 	refcount_t refs;
295 	atomic_t stripes_pending;
296 	struct btrfs_fs_info *fs_info;
297 	u64 map_type; /* get from map_lookup->type */
298 	bio_end_io_t *end_io;
299 	struct bio *orig_bio;
300 	unsigned long flags;
301 	void *private;
302 	atomic_t error;
303 	int max_errors;
304 	int num_stripes;
305 	int mirror_num;
306 	int num_tgtdevs;
307 	int *tgtdev_map;
308 	/*
309 	 * logical block numbers for the start of each stripe
310 	 * The last one or two are p/q.  These are sorted,
311 	 * so raid_map[0] is the start of our full stripe
312 	 */
313 	u64 *raid_map;
314 	struct btrfs_bio_stripe stripes[];
315 };
316 
317 struct btrfs_device_info {
318 	struct btrfs_device *dev;
319 	u64 dev_offset;
320 	u64 max_avail;
321 	u64 total_avail;
322 };
323 
324 struct btrfs_raid_attr {
325 	int sub_stripes;	/* sub_stripes info for map */
326 	int dev_stripes;	/* stripes per dev */
327 	int devs_max;		/* max devs to use */
328 	int devs_min;		/* min devs needed */
329 	int tolerated_failures; /* max tolerated fail devs */
330 	int devs_increment;	/* ndevs has to be a multiple of this */
331 	int ncopies;		/* how many copies to data has */
332 };
333 
334 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
335 extern const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES];
336 extern const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES];
337 
338 struct map_lookup {
339 	u64 type;
340 	int io_align;
341 	int io_width;
342 	u64 stripe_len;
343 	int num_stripes;
344 	int sub_stripes;
345 	struct btrfs_bio_stripe stripes[];
346 };
347 
348 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
349 			    (sizeof(struct btrfs_bio_stripe) * (n)))
350 
351 struct btrfs_balance_args;
352 struct btrfs_balance_progress;
353 struct btrfs_balance_control {
354 	struct btrfs_fs_info *fs_info;
355 
356 	struct btrfs_balance_args data;
357 	struct btrfs_balance_args meta;
358 	struct btrfs_balance_args sys;
359 
360 	u64 flags;
361 
362 	struct btrfs_balance_progress stat;
363 };
364 
365 enum btrfs_map_op {
366 	BTRFS_MAP_READ,
367 	BTRFS_MAP_WRITE,
368 	BTRFS_MAP_DISCARD,
369 	BTRFS_MAP_GET_READ_MIRRORS,
370 };
371 
372 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
373 {
374 	switch (bio_op(bio)) {
375 	case REQ_OP_DISCARD:
376 		return BTRFS_MAP_DISCARD;
377 	case REQ_OP_WRITE:
378 		return BTRFS_MAP_WRITE;
379 	default:
380 		WARN_ON_ONCE(1);
381 	case REQ_OP_READ:
382 		return BTRFS_MAP_READ;
383 	}
384 }
385 
386 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
387 				   u64 end, u64 *length);
388 void btrfs_get_bbio(struct btrfs_bio *bbio);
389 void btrfs_put_bbio(struct btrfs_bio *bbio);
390 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
391 		    u64 logical, u64 *length,
392 		    struct btrfs_bio **bbio_ret, int mirror_num);
393 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
394 		     u64 logical, u64 *length,
395 		     struct btrfs_bio **bbio_ret);
396 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
397 		     u64 chunk_start, u64 physical, u64 devid,
398 		     u64 **logical, int *naddrs, int *stripe_len);
399 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
400 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
401 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
402 		      struct btrfs_fs_info *fs_info, u64 type);
403 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
404 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
405 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
406 			   int mirror_num, int async_submit);
407 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
408 		       fmode_t flags, void *holder);
409 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
410 			  struct btrfs_fs_devices **fs_devices_ret);
411 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
412 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
413 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
414 		struct btrfs_device *device, struct btrfs_device *this_dev);
415 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
416 					 const char *device_path,
417 					 struct btrfs_device **device);
418 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
419 					 const char *devpath,
420 					 struct btrfs_device **device);
421 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
422 					const u64 *devid,
423 					const u8 *uuid);
424 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
425 		    const char *device_path, u64 devid);
426 void __exit btrfs_cleanup_fs_uuids(void);
427 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
428 int btrfs_grow_device(struct btrfs_trans_handle *trans,
429 		      struct btrfs_device *device, u64 new_size);
430 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
431 				       u8 *uuid, u8 *fsid);
432 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
433 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
434 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
435 				  const char *device_path,
436 				  struct btrfs_device *srcdev,
437 				  struct btrfs_device **device_out);
438 int btrfs_balance(struct btrfs_balance_control *bctl,
439 		  struct btrfs_ioctl_balance_args *bargs);
440 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
441 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
442 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
443 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
444 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
445 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
446 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
447 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
448 			 struct btrfs_device *device, u64 num_bytes,
449 			 u64 search_start, u64 *start, u64 *max_avail);
450 int find_free_dev_extent(struct btrfs_trans_handle *trans,
451 			 struct btrfs_device *device, u64 num_bytes,
452 			 u64 *start, u64 *max_avail);
453 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
454 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
455 			struct btrfs_ioctl_get_dev_stats *stats);
456 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
457 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
458 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
459 			struct btrfs_fs_info *fs_info);
460 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
461 					struct btrfs_device *srcdev);
462 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
463 				      struct btrfs_device *srcdev);
464 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
465 				      struct btrfs_device *tgtdev);
466 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
467 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
468 			   u64 logical, u64 len);
469 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
470 				    u64 logical);
471 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
472 				struct btrfs_fs_info *fs_info,
473 				u64 chunk_offset, u64 chunk_size);
474 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
475 		       struct btrfs_fs_info *fs_info, u64 chunk_offset);
476 
477 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
478 				      int index)
479 {
480 	atomic_inc(dev->dev_stat_values + index);
481 	/*
482 	 * This memory barrier orders stores updating statistics before stores
483 	 * updating dev_stats_ccnt.
484 	 *
485 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
486 	 */
487 	smp_mb__before_atomic();
488 	atomic_inc(&dev->dev_stats_ccnt);
489 }
490 
491 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
492 				      int index)
493 {
494 	return atomic_read(dev->dev_stat_values + index);
495 }
496 
497 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
498 						int index)
499 {
500 	int ret;
501 
502 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
503 	/*
504 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
505 	 * - RMW operations that have a return value are fully ordered;
506 	 *
507 	 * This implicit memory barriers is paired with the smp_rmb in
508 	 * btrfs_run_dev_stats
509 	 */
510 	atomic_inc(&dev->dev_stats_ccnt);
511 	return ret;
512 }
513 
514 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
515 				      int index, unsigned long val)
516 {
517 	atomic_set(dev->dev_stat_values + index, val);
518 	/*
519 	 * This memory barrier orders stores updating statistics before stores
520 	 * updating dev_stats_ccnt.
521 	 *
522 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
523 	 */
524 	smp_mb__before_atomic();
525 	atomic_inc(&dev->dev_stats_ccnt);
526 }
527 
528 static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
529 					int index)
530 {
531 	btrfs_dev_stat_set(dev, index, 0);
532 }
533 
534 /*
535  * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
536  * can be used as index to access btrfs_raid_array[].
537  */
538 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
539 {
540 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
541 		return BTRFS_RAID_RAID10;
542 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
543 		return BTRFS_RAID_RAID1;
544 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
545 		return BTRFS_RAID_DUP;
546 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
547 		return BTRFS_RAID_RAID0;
548 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
549 		return BTRFS_RAID_RAID5;
550 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
551 		return BTRFS_RAID_RAID6;
552 
553 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
554 }
555 
556 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info);
557 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans);
558 
559 struct list_head *btrfs_get_fs_uuids(void);
560 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
561 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
562 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
563 					struct btrfs_device *failing_dev);
564 
565 #endif
566