xref: /openbmc/linux/fs/btrfs/volumes.h (revision b755c25f)
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/sort.h>
10 #include <linux/btrfs.h>
11 #include "async-thread.h"
12 #include "messages.h"
13 #include "tree-checker.h"
14 #include "rcu-string.h"
15 
16 #define BTRFS_MAX_DATA_CHUNK_SIZE	(10ULL * SZ_1G)
17 
18 extern struct mutex uuid_mutex;
19 
20 #define BTRFS_STRIPE_LEN		SZ_64K
21 #define BTRFS_STRIPE_LEN_SHIFT		(16)
22 #define BTRFS_STRIPE_LEN_MASK		(BTRFS_STRIPE_LEN - 1)
23 
24 static_assert(const_ilog2(BTRFS_STRIPE_LEN) == BTRFS_STRIPE_LEN_SHIFT);
25 
26 /* Used by sanity check for btrfs_raid_types. */
27 #define const_ffs(n) (__builtin_ctzll(n) + 1)
28 
29 /*
30  * The conversion from BTRFS_BLOCK_GROUP_* bits to btrfs_raid_type requires
31  * RAID0 always to be the lowest profile bit.
32  * Although it's part of on-disk format and should never change, do extra
33  * compile-time sanity checks.
34  */
35 static_assert(const_ffs(BTRFS_BLOCK_GROUP_RAID0) <
36 	      const_ffs(BTRFS_BLOCK_GROUP_PROFILE_MASK & ~BTRFS_BLOCK_GROUP_RAID0));
37 static_assert(const_ilog2(BTRFS_BLOCK_GROUP_RAID0) >
38 	      ilog2(BTRFS_BLOCK_GROUP_TYPE_MASK));
39 
40 /* ilog2() can handle both constants and variables */
41 #define BTRFS_BG_FLAG_TO_INDEX(profile)					\
42 	ilog2((profile) >> (ilog2(BTRFS_BLOCK_GROUP_RAID0) - 1))
43 
44 enum btrfs_raid_types {
45 	/* SINGLE is the special one as it doesn't have on-disk bit. */
46 	BTRFS_RAID_SINGLE  = 0,
47 
48 	BTRFS_RAID_RAID0   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID0),
49 	BTRFS_RAID_RAID1   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1),
50 	BTRFS_RAID_DUP	   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_DUP),
51 	BTRFS_RAID_RAID10  = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID10),
52 	BTRFS_RAID_RAID5   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID5),
53 	BTRFS_RAID_RAID6   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID6),
54 	BTRFS_RAID_RAID1C3 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C3),
55 	BTRFS_RAID_RAID1C4 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C4),
56 
57 	BTRFS_NR_RAID_TYPES
58 };
59 
60 /*
61  * Use sequence counter to get consistent device stat data on
62  * 32-bit processors.
63  */
64 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
65 #include <linux/seqlock.h>
66 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
67 #define btrfs_device_data_ordered_init(device)	\
68 	seqcount_init(&device->data_seqcount)
69 #else
70 #define btrfs_device_data_ordered_init(device) do { } while (0)
71 #endif
72 
73 #define BTRFS_DEV_STATE_WRITEABLE	(0)
74 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
75 #define BTRFS_DEV_STATE_MISSING		(2)
76 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
77 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
78 #define BTRFS_DEV_STATE_NO_READA	(5)
79 
80 struct btrfs_zoned_device_info;
81 
82 struct btrfs_device {
83 	struct list_head dev_list; /* device_list_mutex */
84 	struct list_head dev_alloc_list; /* chunk mutex */
85 	struct list_head post_commit_list; /* chunk mutex */
86 	struct btrfs_fs_devices *fs_devices;
87 	struct btrfs_fs_info *fs_info;
88 
89 	struct rcu_string __rcu *name;
90 
91 	u64 generation;
92 
93 	struct block_device *bdev;
94 
95 	struct btrfs_zoned_device_info *zone_info;
96 
97 	/* block device holder for blkdev_get/put */
98 	void *holder;
99 
100 	/*
101 	 * Device's major-minor number. Must be set even if the device is not
102 	 * opened (bdev == NULL), unless the device is missing.
103 	 */
104 	dev_t devt;
105 	unsigned long dev_state;
106 	blk_status_t last_flush_error;
107 
108 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
109 	seqcount_t data_seqcount;
110 #endif
111 
112 	/* the internal btrfs device id */
113 	u64 devid;
114 
115 	/* size of the device in memory */
116 	u64 total_bytes;
117 
118 	/* size of the device on disk */
119 	u64 disk_total_bytes;
120 
121 	/* bytes used */
122 	u64 bytes_used;
123 
124 	/* optimal io alignment for this device */
125 	u32 io_align;
126 
127 	/* optimal io width for this device */
128 	u32 io_width;
129 	/* type and info about this device */
130 	u64 type;
131 
132 	/* minimal io size for this device */
133 	u32 sector_size;
134 
135 	/* physical drive uuid (or lvm uuid) */
136 	u8 uuid[BTRFS_UUID_SIZE];
137 
138 	/*
139 	 * size of the device on the current transaction
140 	 *
141 	 * This variant is update when committing the transaction,
142 	 * and protected by chunk mutex
143 	 */
144 	u64 commit_total_bytes;
145 
146 	/* bytes used on the current transaction */
147 	u64 commit_bytes_used;
148 
149 	/* Bio used for flushing device barriers */
150 	struct bio flush_bio;
151 	struct completion flush_wait;
152 
153 	/* per-device scrub information */
154 	struct scrub_ctx *scrub_ctx;
155 
156 	/* disk I/O failure stats. For detailed description refer to
157 	 * enum btrfs_dev_stat_values in ioctl.h */
158 	int dev_stats_valid;
159 
160 	/* Counter to record the change of device stats */
161 	atomic_t dev_stats_ccnt;
162 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
163 
164 	struct extent_io_tree alloc_state;
165 
166 	struct completion kobj_unregister;
167 	/* For sysfs/FSID/devinfo/devid/ */
168 	struct kobject devid_kobj;
169 
170 	/* Bandwidth limit for scrub, in bytes */
171 	u64 scrub_speed_max;
172 };
173 
174 /*
175  * Block group or device which contains an active swapfile. Used for preventing
176  * unsafe operations while a swapfile is active.
177  *
178  * These are sorted on (ptr, inode) (note that a block group or device can
179  * contain more than one swapfile). We compare the pointer values because we
180  * don't actually care what the object is, we just need a quick check whether
181  * the object exists in the rbtree.
182  */
183 struct btrfs_swapfile_pin {
184 	struct rb_node node;
185 	void *ptr;
186 	struct inode *inode;
187 	/*
188 	 * If true, ptr points to a struct btrfs_block_group. Otherwise, ptr
189 	 * points to a struct btrfs_device.
190 	 */
191 	bool is_block_group;
192 	/*
193 	 * Only used when 'is_block_group' is true and it is the number of
194 	 * extents used by a swapfile for this block group ('ptr' field).
195 	 */
196 	int bg_extent_count;
197 };
198 
199 /*
200  * If we read those variants at the context of their own lock, we needn't
201  * use the following helpers, reading them directly is safe.
202  */
203 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
204 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
205 static inline u64							\
206 btrfs_device_get_##name(const struct btrfs_device *dev)			\
207 {									\
208 	u64 size;							\
209 	unsigned int seq;						\
210 									\
211 	do {								\
212 		seq = read_seqcount_begin(&dev->data_seqcount);		\
213 		size = dev->name;					\
214 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
215 	return size;							\
216 }									\
217 									\
218 static inline void							\
219 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
220 {									\
221 	preempt_disable();						\
222 	write_seqcount_begin(&dev->data_seqcount);			\
223 	dev->name = size;						\
224 	write_seqcount_end(&dev->data_seqcount);			\
225 	preempt_enable();						\
226 }
227 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
228 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
229 static inline u64							\
230 btrfs_device_get_##name(const struct btrfs_device *dev)			\
231 {									\
232 	u64 size;							\
233 									\
234 	preempt_disable();						\
235 	size = dev->name;						\
236 	preempt_enable();						\
237 	return size;							\
238 }									\
239 									\
240 static inline void							\
241 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
242 {									\
243 	preempt_disable();						\
244 	dev->name = size;						\
245 	preempt_enable();						\
246 }
247 #else
248 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
249 static inline u64							\
250 btrfs_device_get_##name(const struct btrfs_device *dev)			\
251 {									\
252 	return dev->name;						\
253 }									\
254 									\
255 static inline void							\
256 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
257 {									\
258 	dev->name = size;						\
259 }
260 #endif
261 
262 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
263 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
264 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
265 
266 enum btrfs_chunk_allocation_policy {
267 	BTRFS_CHUNK_ALLOC_REGULAR,
268 	BTRFS_CHUNK_ALLOC_ZONED,
269 };
270 
271 /*
272  * Read policies for mirrored block group profiles, read picks the stripe based
273  * on these policies.
274  */
275 enum btrfs_read_policy {
276 	/* Use process PID to choose the stripe */
277 	BTRFS_READ_POLICY_PID,
278 	BTRFS_NR_READ_POLICY,
279 };
280 
281 struct btrfs_fs_devices {
282 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
283 
284 	/*
285 	 * UUID written into the btree blocks:
286 	 *
287 	 * - If metadata_uuid != fsid then super block must have
288 	 *   BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag set.
289 	 *
290 	 * - Following shall be true at all times:
291 	 *   - metadata_uuid == btrfs_header::fsid
292 	 *   - metadata_uuid == btrfs_dev_item::fsid
293 	 */
294 	u8 metadata_uuid[BTRFS_FSID_SIZE];
295 
296 	struct list_head fs_list;
297 
298 	/*
299 	 * Number of devices under this fsid including missing and
300 	 * replace-target device and excludes seed devices.
301 	 */
302 	u64 num_devices;
303 
304 	/*
305 	 * The number of devices that successfully opened, including
306 	 * replace-target, excludes seed devices.
307 	 */
308 	u64 open_devices;
309 
310 	/* The number of devices that are under the chunk allocation list. */
311 	u64 rw_devices;
312 
313 	/* Count of missing devices under this fsid excluding seed device. */
314 	u64 missing_devices;
315 	u64 total_rw_bytes;
316 
317 	/*
318 	 * Count of devices from btrfs_super_block::num_devices for this fsid,
319 	 * which includes the seed device, excludes the transient replace-target
320 	 * device.
321 	 */
322 	u64 total_devices;
323 
324 	/* Highest generation number of seen devices */
325 	u64 latest_generation;
326 
327 	/*
328 	 * The mount device or a device with highest generation after removal
329 	 * or replace.
330 	 */
331 	struct btrfs_device *latest_dev;
332 
333 	/*
334 	 * All of the devices in the filesystem, protected by a mutex so we can
335 	 * safely walk it to write out the super blocks without worrying about
336 	 * adding/removing by the multi-device code. Scrubbing super block can
337 	 * kick off supers writing by holding this mutex lock.
338 	 */
339 	struct mutex device_list_mutex;
340 
341 	/* List of all devices, protected by device_list_mutex */
342 	struct list_head devices;
343 
344 	/* Devices which can satisfy space allocation. Protected by * chunk_mutex. */
345 	struct list_head alloc_list;
346 
347 	struct list_head seed_list;
348 
349 	/* Count fs-devices opened. */
350 	int opened;
351 
352 	/* Set when we find or add a device that doesn't have the nonrot flag set. */
353 	bool rotating;
354 	/* Devices support TRIM/discard commands. */
355 	bool discardable;
356 	bool fsid_change;
357 	/* The filesystem is a seed filesystem. */
358 	bool seeding;
359 
360 	struct btrfs_fs_info *fs_info;
361 	/* sysfs kobjects */
362 	struct kobject fsid_kobj;
363 	struct kobject *devices_kobj;
364 	struct kobject *devinfo_kobj;
365 	struct completion kobj_unregister;
366 
367 	enum btrfs_chunk_allocation_policy chunk_alloc_policy;
368 
369 	/* Policy used to read the mirrored stripes. */
370 	enum btrfs_read_policy read_policy;
371 };
372 
373 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
374 			- sizeof(struct btrfs_chunk))		\
375 			/ sizeof(struct btrfs_stripe) + 1)
376 
377 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
378 				- 2 * sizeof(struct btrfs_disk_key)	\
379 				- 2 * sizeof(struct btrfs_chunk))	\
380 				/ sizeof(struct btrfs_stripe) + 1)
381 
382 struct btrfs_io_stripe {
383 	struct btrfs_device *dev;
384 	union {
385 		/* Block mapping */
386 		u64 physical;
387 		/* For the endio handler */
388 		struct btrfs_io_context *bioc;
389 	};
390 };
391 
392 struct btrfs_discard_stripe {
393 	struct btrfs_device *dev;
394 	u64 physical;
395 	u64 length;
396 };
397 
398 /*
399  * Context for IO subsmission for device stripe.
400  *
401  * - Track the unfinished mirrors for mirror based profiles
402  *   Mirror based profiles are SINGLE/DUP/RAID1/RAID10.
403  *
404  * - Contain the logical -> physical mapping info
405  *   Used by submit_stripe_bio() for mapping logical bio
406  *   into physical device address.
407  *
408  * - Contain device replace info
409  *   Used by handle_ops_on_dev_replace() to copy logical bios
410  *   into the new device.
411  *
412  * - Contain RAID56 full stripe logical bytenrs
413  */
414 struct btrfs_io_context {
415 	refcount_t refs;
416 	struct btrfs_fs_info *fs_info;
417 	u64 map_type; /* get from map_lookup->type */
418 	struct bio *orig_bio;
419 	atomic_t error;
420 	u16 max_errors;
421 
422 	/*
423 	 * The total number of stripes, including the extra duplicated
424 	 * stripe for replace.
425 	 */
426 	u16 num_stripes;
427 
428 	/*
429 	 * The mirror_num of this bioc.
430 	 *
431 	 * This is for reads which use 0 as mirror_num, thus we should return a
432 	 * valid mirror_num (>0) for the reader.
433 	 */
434 	u16 mirror_num;
435 
436 	/*
437 	 * The following two members are for dev-replace case only.
438 	 *
439 	 * @replace_nr_stripes:	Number of duplicated stripes which need to be
440 	 *			written to replace target.
441 	 *			Should be <= 2 (2 for DUP, otherwise <= 1).
442 	 * @replace_stripe_src:	The array indicates where the duplicated stripes
443 	 *			are from.
444 	 *
445 	 * The @replace_stripe_src[] array is mostly for RAID56 cases.
446 	 * As non-RAID56 stripes share the same contents of the mapped range,
447 	 * thus no need to bother where the duplicated ones are from.
448 	 *
449 	 * But for RAID56 case, all stripes contain different contents, thus
450 	 * we need a way to know the mapping.
451 	 *
452 	 * There is an example for the two members, using a RAID5 write:
453 	 *
454 	 *   num_stripes:	4 (3 + 1 duplicated write)
455 	 *   stripes[0]:	dev = devid 1, physical = X
456 	 *   stripes[1]:	dev = devid 2, physical = Y
457 	 *   stripes[2]:	dev = devid 3, physical = Z
458 	 *   stripes[3]:	dev = devid 0, physical = Y
459 	 *
460 	 * replace_nr_stripes = 1
461 	 * replace_stripe_src = 1	<- Means stripes[1] is involved in replace.
462 	 *				   The duplicated stripe index would be
463 	 *				   (@num_stripes - 1).
464 	 *
465 	 * Note, that we can still have cases replace_nr_stripes = 2 for DUP.
466 	 * In that case, all stripes share the same content, thus we don't
467 	 * need to bother @replace_stripe_src value at all.
468 	 */
469 	u16 replace_nr_stripes;
470 	s16 replace_stripe_src;
471 	/*
472 	 * Logical bytenr of the full stripe start, only for RAID56 cases.
473 	 *
474 	 * When this value is set to other than (u64)-1, the stripes[] should
475 	 * follow this pattern:
476 	 *
477 	 * (real_stripes = num_stripes - replace_nr_stripes)
478 	 * (data_stripes = (is_raid6) ? (real_stripes - 2) : (real_stripes - 1))
479 	 *
480 	 * stripes[0]:			The first data stripe
481 	 * stripes[1]:			The second data stripe
482 	 * ...
483 	 * stripes[data_stripes - 1]:	The last data stripe
484 	 * stripes[data_stripes]:	The P stripe
485 	 * stripes[data_stripes + 1]:	The Q stripe (only for RAID6).
486 	 */
487 	u64 full_stripe_logical;
488 	struct btrfs_io_stripe stripes[];
489 };
490 
491 struct btrfs_device_info {
492 	struct btrfs_device *dev;
493 	u64 dev_offset;
494 	u64 max_avail;
495 	u64 total_avail;
496 };
497 
498 struct btrfs_raid_attr {
499 	u8 sub_stripes;		/* sub_stripes info for map */
500 	u8 dev_stripes;		/* stripes per dev */
501 	u8 devs_max;		/* max devs to use */
502 	u8 devs_min;		/* min devs needed */
503 	u8 tolerated_failures;	/* max tolerated fail devs */
504 	u8 devs_increment;	/* ndevs has to be a multiple of this */
505 	u8 ncopies;		/* how many copies to data has */
506 	u8 nparity;		/* number of stripes worth of bytes to store
507 				 * parity information */
508 	u8 mindev_error;	/* error code if min devs requisite is unmet */
509 	const char raid_name[8]; /* name of the raid */
510 	u64 bg_flag;		/* block group flag of the raid */
511 };
512 
513 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
514 
515 struct map_lookup {
516 	u64 type;
517 	int io_align;
518 	int io_width;
519 	int num_stripes;
520 	int sub_stripes;
521 	int verified_stripes; /* For mount time dev extent verification */
522 	struct btrfs_io_stripe stripes[];
523 };
524 
525 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
526 			    (sizeof(struct btrfs_io_stripe) * (n)))
527 
528 struct btrfs_balance_args;
529 struct btrfs_balance_progress;
530 struct btrfs_balance_control {
531 	struct btrfs_balance_args data;
532 	struct btrfs_balance_args meta;
533 	struct btrfs_balance_args sys;
534 
535 	u64 flags;
536 
537 	struct btrfs_balance_progress stat;
538 };
539 
540 /*
541  * Search for a given device by the set parameters
542  */
543 struct btrfs_dev_lookup_args {
544 	u64 devid;
545 	u8 *uuid;
546 	u8 *fsid;
547 	bool missing;
548 };
549 
550 /* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */
551 #define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 }
552 
553 #define BTRFS_DEV_LOOKUP_ARGS(name) \
554 	struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT
555 
556 enum btrfs_map_op {
557 	BTRFS_MAP_READ,
558 	BTRFS_MAP_WRITE,
559 	BTRFS_MAP_GET_READ_MIRRORS,
560 };
561 
562 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
563 {
564 	switch (bio_op(bio)) {
565 	case REQ_OP_WRITE:
566 	case REQ_OP_ZONE_APPEND:
567 		return BTRFS_MAP_WRITE;
568 	default:
569 		WARN_ON_ONCE(1);
570 		fallthrough;
571 	case REQ_OP_READ:
572 		return BTRFS_MAP_READ;
573 	}
574 }
575 
576 static inline unsigned long btrfs_chunk_item_size(int num_stripes)
577 {
578 	ASSERT(num_stripes);
579 	return sizeof(struct btrfs_chunk) +
580 		sizeof(struct btrfs_stripe) * (num_stripes - 1);
581 }
582 
583 /*
584  * Do the type safe converstion from stripe_nr to offset inside the chunk.
585  *
586  * @stripe_nr is u32, with left shift it can overflow u32 for chunks larger
587  * than 4G.  This does the proper type cast to avoid overflow.
588  */
589 static inline u64 btrfs_stripe_nr_to_offset(u32 stripe_nr)
590 {
591 	return (u64)stripe_nr << BTRFS_STRIPE_LEN_SHIFT;
592 }
593 
594 void btrfs_get_bioc(struct btrfs_io_context *bioc);
595 void btrfs_put_bioc(struct btrfs_io_context *bioc);
596 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
597 		    u64 logical, u64 *length,
598 		    struct btrfs_io_context **bioc_ret,
599 		    struct btrfs_io_stripe *smap, int *mirror_num_ret,
600 		    int need_raid_map);
601 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
602 			   struct btrfs_io_stripe *smap, u64 logical,
603 			   u32 length, int mirror_num);
604 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
605 					       u64 logical, u64 *length_ret,
606 					       u32 *num_stripes);
607 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
608 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
609 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
610 					    u64 type);
611 void btrfs_mapping_tree_free(struct extent_map_tree *tree);
612 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
613 		       blk_mode_t flags, void *holder);
614 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags);
615 int btrfs_forget_devices(dev_t devt);
616 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
617 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
618 void btrfs_assign_next_active_device(struct btrfs_device *device,
619 				     struct btrfs_device *this_dev);
620 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
621 						  u64 devid,
622 						  const char *devpath);
623 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
624 				 struct btrfs_dev_lookup_args *args,
625 				 const char *path);
626 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
627 					const u64 *devid, const u8 *uuid,
628 					const char *path);
629 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args);
630 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
631 		    struct btrfs_dev_lookup_args *args,
632 		    struct block_device **bdev, void **holder);
633 void __exit btrfs_cleanup_fs_uuids(void);
634 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
635 int btrfs_grow_device(struct btrfs_trans_handle *trans,
636 		      struct btrfs_device *device, u64 new_size);
637 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
638 				       const struct btrfs_dev_lookup_args *args);
639 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
640 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
641 int btrfs_balance(struct btrfs_fs_info *fs_info,
642 		  struct btrfs_balance_control *bctl,
643 		  struct btrfs_ioctl_balance_args *bargs);
644 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
645 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
646 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
647 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
648 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
649 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
650 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
651 int btrfs_uuid_scan_kthread(void *data);
652 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset);
653 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
654 			 u64 *start, u64 *max_avail);
655 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
656 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
657 			struct btrfs_ioctl_get_dev_stats *stats);
658 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
659 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
660 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
661 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
662 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
663 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
664 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
665 			   u64 logical, u64 len);
666 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
667 				    u64 logical);
668 u64 btrfs_calc_stripe_length(const struct extent_map *em);
669 int btrfs_nr_parity_stripes(u64 type);
670 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
671 				     struct btrfs_block_group *bg);
672 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
673 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
674 				       u64 logical, u64 length);
675 void btrfs_release_disk_super(struct btrfs_super_block *super);
676 
677 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
678 				      int index)
679 {
680 	atomic_inc(dev->dev_stat_values + index);
681 	/*
682 	 * This memory barrier orders stores updating statistics before stores
683 	 * updating dev_stats_ccnt.
684 	 *
685 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
686 	 */
687 	smp_mb__before_atomic();
688 	atomic_inc(&dev->dev_stats_ccnt);
689 }
690 
691 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
692 				      int index)
693 {
694 	return atomic_read(dev->dev_stat_values + index);
695 }
696 
697 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
698 						int index)
699 {
700 	int ret;
701 
702 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
703 	/*
704 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
705 	 * - RMW operations that have a return value are fully ordered;
706 	 *
707 	 * This implicit memory barriers is paired with the smp_rmb in
708 	 * btrfs_run_dev_stats
709 	 */
710 	atomic_inc(&dev->dev_stats_ccnt);
711 	return ret;
712 }
713 
714 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
715 				      int index, unsigned long val)
716 {
717 	atomic_set(dev->dev_stat_values + index, val);
718 	/*
719 	 * This memory barrier orders stores updating statistics before stores
720 	 * updating dev_stats_ccnt.
721 	 *
722 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
723 	 */
724 	smp_mb__before_atomic();
725 	atomic_inc(&dev->dev_stats_ccnt);
726 }
727 
728 static inline const char *btrfs_dev_name(const struct btrfs_device *device)
729 {
730 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
731 		return "<missing disk>";
732 	else
733 		return rcu_str_deref(device->name);
734 }
735 
736 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
737 
738 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
739 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
740 					struct btrfs_device *failing_dev);
741 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
742 			       struct block_device *bdev,
743 			       const char *device_path);
744 
745 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
746 int btrfs_bg_type_to_factor(u64 flags);
747 const char *btrfs_bg_type_to_raid_name(u64 flags);
748 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
749 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
750 
751 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr);
752 
753 #endif
754