xref: /linux/fs/btrfs/volumes.h (revision 07eebd934c9cb9f12f589ea5b826fa7ca056cb4d)
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/blk_types.h>
10 #include <linux/blkdev.h>
11 #include <linux/sizes.h>
12 #include <linux/atomic.h>
13 #include <linux/sort.h>
14 #include <linux/list.h>
15 #include <linux/mutex.h>
16 #include <linux/log2.h>
17 #include <linux/kobject.h>
18 #include <linux/refcount.h>
19 #include <linux/completion.h>
20 #include <linux/rbtree.h>
21 #include <uapi/linux/btrfs.h>
22 #include <uapi/linux/btrfs_tree.h>
23 #include "messages.h"
24 #include "extent-io-tree.h"
25 
26 struct block_device;
27 struct bdev_handle;
28 struct btrfs_fs_info;
29 struct btrfs_block_group;
30 struct btrfs_trans_handle;
31 struct btrfs_transaction;
32 struct btrfs_zoned_device_info;
33 
34 #define BTRFS_MAX_DATA_CHUNK_SIZE	(10ULL * SZ_1G)
35 
36 /*
37  * Arbitrary maximum size of one discard request to limit potentially long time
38  * spent in blkdev_issue_discard().
39  */
40 #define BTRFS_MAX_DISCARD_CHUNK_SIZE	(SZ_1G)
41 
42 extern struct mutex uuid_mutex;
43 
44 #define BTRFS_STRIPE_LEN		SZ_64K
45 #define BTRFS_STRIPE_LEN_SHIFT		(16)
46 #define BTRFS_STRIPE_LEN_MASK		(BTRFS_STRIPE_LEN - 1)
47 
48 static_assert(ilog2(BTRFS_STRIPE_LEN) == BTRFS_STRIPE_LEN_SHIFT);
49 
50 /* Used by sanity check for btrfs_raid_types. */
51 #define const_ffs(n) (__builtin_ctzll(n) + 1)
52 
53 /*
54  * The conversion from BTRFS_BLOCK_GROUP_* bits to btrfs_raid_type requires
55  * RAID0 always to be the lowest profile bit.
56  * Although it's part of on-disk format and should never change, do extra
57  * compile-time sanity checks.
58  */
59 static_assert(const_ffs(BTRFS_BLOCK_GROUP_RAID0) <
60 	      const_ffs(BTRFS_BLOCK_GROUP_PROFILE_MASK & ~BTRFS_BLOCK_GROUP_RAID0));
61 static_assert(ilog2(BTRFS_BLOCK_GROUP_RAID0) > ilog2(BTRFS_BLOCK_GROUP_TYPE_MASK));
62 
63 /* ilog2() can handle both constants and variables */
64 #define BTRFS_BG_FLAG_TO_INDEX(profile)					\
65 	ilog2((profile) >> (ilog2(BTRFS_BLOCK_GROUP_RAID0) - 1))
66 
67 enum btrfs_raid_types {
68 	/* SINGLE is the special one as it doesn't have on-disk bit. */
69 	BTRFS_RAID_SINGLE  = 0,
70 
71 	BTRFS_RAID_RAID0   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID0),
72 	BTRFS_RAID_RAID1   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1),
73 	BTRFS_RAID_DUP	   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_DUP),
74 	BTRFS_RAID_RAID10  = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID10),
75 	BTRFS_RAID_RAID5   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID5),
76 	BTRFS_RAID_RAID6   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID6),
77 	BTRFS_RAID_RAID1C3 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C3),
78 	BTRFS_RAID_RAID1C4 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C4),
79 
80 	BTRFS_NR_RAID_TYPES
81 };
82 
83 /*
84  * Use sequence counter to get consistent device stat data on
85  * 32-bit processors.
86  */
87 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
88 #include <linux/seqlock.h>
89 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
90 #define btrfs_device_data_ordered_init(device)	\
91 	seqcount_init(&device->data_seqcount)
92 #else
93 #define btrfs_device_data_ordered_init(device) do { } while (0)
94 #endif
95 
96 #define BTRFS_DEV_STATE_WRITEABLE	(0)
97 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
98 #define BTRFS_DEV_STATE_MISSING		(2)
99 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
100 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
101 #define BTRFS_DEV_STATE_NO_READA	(5)
102 
103 /* Set when the device item is found in chunk tree, used to catch unexpected registered device. */
104 #define BTRFS_DEV_STATE_ITEM_FOUND	(7)
105 
106 /* Special value encoding failure to write primary super block. */
107 #define BTRFS_SUPER_PRIMARY_WRITE_ERROR		(INT_MAX / 2)
108 
109 struct btrfs_fs_devices;
110 
111 struct btrfs_device {
112 	struct list_head dev_list; /* device_list_mutex */
113 	struct list_head dev_alloc_list; /* chunk mutex */
114 	struct list_head post_commit_list; /* chunk mutex */
115 	struct btrfs_fs_devices *fs_devices;
116 	struct btrfs_fs_info *fs_info;
117 
118 	/* Device path or NULL if missing. */
119 	const char __rcu *name;
120 
121 	u64 generation;
122 
123 	struct file *bdev_file;
124 	struct block_device *bdev;
125 
126 	struct btrfs_zoned_device_info *zone_info;
127 
128 	/*
129 	 * Device's major-minor number. Must be set even if the device is not
130 	 * opened (bdev == NULL), unless the device is missing.
131 	 */
132 	dev_t devt;
133 	unsigned long dev_state;
134 	blk_status_t last_flush_error;
135 
136 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
137 	seqcount_t data_seqcount;
138 #endif
139 
140 	/* the internal btrfs device id */
141 	u64 devid;
142 
143 	/* size of the device in memory */
144 	u64 total_bytes;
145 
146 	/* size of the device on disk */
147 	u64 disk_total_bytes;
148 
149 	/* bytes used */
150 	u64 bytes_used;
151 
152 	/* optimal io alignment for this device */
153 	u32 io_align;
154 
155 	/* optimal io width for this device */
156 	u32 io_width;
157 	/* type and info about this device */
158 	u64 type;
159 
160 	/*
161 	 * Counter of super block write errors, values larger than
162 	 * BTRFS_SUPER_PRIMARY_WRITE_ERROR encode primary super block write failure.
163 	 */
164 	atomic_t sb_write_errors;
165 
166 	/* minimal io size for this device */
167 	u32 sector_size;
168 
169 	/* physical drive uuid (or lvm uuid) */
170 	u8 uuid[BTRFS_UUID_SIZE];
171 
172 	/*
173 	 * size of the device on the current transaction
174 	 *
175 	 * This variant is update when committing the transaction,
176 	 * and protected by chunk mutex
177 	 */
178 	u64 commit_total_bytes;
179 
180 	/* bytes used on the current transaction */
181 	u64 commit_bytes_used;
182 
183 	/* Bio used for flushing device barriers */
184 	struct bio flush_bio;
185 	struct completion flush_wait;
186 
187 	/* per-device scrub information */
188 	struct scrub_ctx *scrub_ctx;
189 
190 	/* disk I/O failure stats. For detailed description refer to
191 	 * enum btrfs_dev_stat_values in ioctl.h */
192 	int dev_stats_valid;
193 
194 	/* Counter to record the change of device stats */
195 	atomic_t dev_stats_ccnt;
196 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
197 
198 	struct extent_io_tree alloc_state;
199 
200 	struct completion kobj_unregister;
201 	/* For sysfs/FSID/devinfo/devid/ */
202 	struct kobject devid_kobj;
203 
204 	/* Bandwidth limit for scrub, in bytes */
205 	u64 scrub_speed_max;
206 };
207 
208 /*
209  * Block group or device which contains an active swapfile. Used for preventing
210  * unsafe operations while a swapfile is active.
211  *
212  * These are sorted on (ptr, inode) (note that a block group or device can
213  * contain more than one swapfile). We compare the pointer values because we
214  * don't actually care what the object is, we just need a quick check whether
215  * the object exists in the rbtree.
216  */
217 struct btrfs_swapfile_pin {
218 	struct rb_node node;
219 	void *ptr;
220 	struct inode *inode;
221 	/*
222 	 * If true, ptr points to a struct btrfs_block_group. Otherwise, ptr
223 	 * points to a struct btrfs_device.
224 	 */
225 	bool is_block_group;
226 	/*
227 	 * Only used when 'is_block_group' is true and it is the number of
228 	 * extents used by a swapfile for this block group ('ptr' field).
229 	 */
230 	int bg_extent_count;
231 };
232 
233 /*
234  * If we read those variants at the context of their own lock, we needn't
235  * use the following helpers, reading them directly is safe.
236  */
237 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
238 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
239 static inline u64							\
240 btrfs_device_get_##name(const struct btrfs_device *dev)			\
241 {									\
242 	u64 size;							\
243 	unsigned int seq;						\
244 									\
245 	do {								\
246 		seq = read_seqcount_begin(&dev->data_seqcount);		\
247 		size = dev->name;					\
248 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
249 	return size;							\
250 }									\
251 									\
252 static inline void							\
253 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
254 {									\
255 	preempt_disable();						\
256 	write_seqcount_begin(&dev->data_seqcount);			\
257 	dev->name = size;						\
258 	write_seqcount_end(&dev->data_seqcount);			\
259 	preempt_enable();						\
260 }
261 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
262 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
263 static inline u64							\
264 btrfs_device_get_##name(const struct btrfs_device *dev)			\
265 {									\
266 	u64 size;							\
267 									\
268 	preempt_disable();						\
269 	size = dev->name;						\
270 	preempt_enable();						\
271 	return size;							\
272 }									\
273 									\
274 static inline void							\
275 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
276 {									\
277 	preempt_disable();						\
278 	dev->name = size;						\
279 	preempt_enable();						\
280 }
281 #else
282 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
283 static inline u64							\
284 btrfs_device_get_##name(const struct btrfs_device *dev)			\
285 {									\
286 	return dev->name;						\
287 }									\
288 									\
289 static inline void							\
290 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
291 {									\
292 	dev->name = size;						\
293 }
294 #endif
295 
296 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
297 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
298 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
299 
300 enum btrfs_chunk_allocation_policy {
301 	BTRFS_CHUNK_ALLOC_REGULAR,
302 	BTRFS_CHUNK_ALLOC_ZONED,
303 };
304 
305 #define BTRFS_DEFAULT_RR_MIN_CONTIG_READ	(SZ_256K)
306 /* Keep in sync with raid_attr table, current maximum is RAID1C4. */
307 #define BTRFS_RAID1_MAX_MIRRORS			(4)
308 /*
309  * Read policies for mirrored block group profiles, read picks the stripe based
310  * on these policies.
311  */
312 enum btrfs_read_policy {
313 	/* Use process PID to choose the stripe */
314 	BTRFS_READ_POLICY_PID,
315 #ifdef CONFIG_BTRFS_EXPERIMENTAL
316 	/* Balancing RAID1 reads across all striped devices (round-robin). */
317 	BTRFS_READ_POLICY_RR,
318 	/* Read from a specific device. */
319 	BTRFS_READ_POLICY_DEVID,
320 #endif
321 	BTRFS_NR_READ_POLICY,
322 };
323 
324 #ifdef CONFIG_BTRFS_EXPERIMENTAL
325 /*
326  * Checksum mode - offload it to workqueues or do it synchronously in
327  * btrfs_submit_chunk().
328  */
329 enum btrfs_offload_csum_mode {
330 	/*
331 	 * Choose offloading checksum or do it synchronously automatically.
332 	 * Do it synchronously if the checksum is fast, or offload to workqueues
333 	 * otherwise.
334 	 */
335 	BTRFS_OFFLOAD_CSUM_AUTO,
336 	/* Always offload checksum to workqueues. */
337 	BTRFS_OFFLOAD_CSUM_FORCE_ON,
338 	/* Never offload checksum to workqueues. */
339 	BTRFS_OFFLOAD_CSUM_FORCE_OFF,
340 };
341 #endif
342 
343 struct btrfs_fs_devices {
344 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
345 
346 	/*
347 	 * UUID written into the btree blocks:
348 	 *
349 	 * - If metadata_uuid != fsid then super block must have
350 	 *   BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag set.
351 	 *
352 	 * - Following shall be true at all times:
353 	 *   - metadata_uuid == btrfs_header::fsid
354 	 *   - metadata_uuid == btrfs_dev_item::fsid
355 	 *
356 	 * - Relations between fsid and metadata_uuid in sb and fs_devices:
357 	 *   - Normal:
358 	 *       fs_devices->fsid == fs_devices->metadata_uuid == sb->fsid
359 	 *       sb->metadata_uuid == 0
360 	 *
361 	 *   - When the BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag is set:
362 	 *       fs_devices->fsid == sb->fsid
363 	 *       fs_devices->metadata_uuid == sb->metadata_uuid
364 	 *
365 	 *   - When in-memory fs_devices->temp_fsid is true
366 	 *	 fs_devices->fsid = random
367 	 *	 fs_devices->metadata_uuid == sb->fsid
368 	 */
369 	u8 metadata_uuid[BTRFS_FSID_SIZE];
370 
371 	struct list_head fs_list;
372 
373 	/*
374 	 * Number of devices under this fsid including missing and
375 	 * replace-target device and excludes seed devices.
376 	 */
377 	u64 num_devices;
378 
379 	/*
380 	 * The number of devices that successfully opened, including
381 	 * replace-target, excludes seed devices.
382 	 */
383 	u64 open_devices;
384 
385 	/* The number of devices that are under the chunk allocation list. */
386 	u64 rw_devices;
387 
388 	/* Count of missing devices under this fsid excluding seed device. */
389 	u64 missing_devices;
390 	u64 total_rw_bytes;
391 
392 	/*
393 	 * Count of devices from btrfs_super_block::num_devices for this fsid,
394 	 * which includes the seed device, excludes the transient replace-target
395 	 * device.
396 	 */
397 	u64 total_devices;
398 
399 	/* Highest generation number of seen devices */
400 	u64 latest_generation;
401 
402 	/*
403 	 * The mount device or a device with highest generation after removal
404 	 * or replace.
405 	 */
406 	struct btrfs_device *latest_dev;
407 
408 	/*
409 	 * All of the devices in the filesystem, protected by a mutex so we can
410 	 * safely walk it to write out the super blocks without worrying about
411 	 * adding/removing by the multi-device code. Scrubbing super block can
412 	 * kick off supers writing by holding this mutex lock.
413 	 */
414 	struct mutex device_list_mutex;
415 
416 	/* List of all devices, protected by device_list_mutex */
417 	struct list_head devices;
418 
419 	/* Devices which can satisfy space allocation. Protected by * chunk_mutex. */
420 	struct list_head alloc_list;
421 
422 	struct list_head seed_list;
423 
424 	/* Count fs-devices opened. */
425 	int opened;
426 
427 	/*
428 	 * Counter of the processes that are holding this fs_devices but not
429 	 * yet opened.
430 	 * This is for mounting handling, as we can only open the fs_devices
431 	 * after a super block is created.  But we cannot take uuid_mutex
432 	 * during sget_fc(), thus we have to hold the fs_devices (meaning it
433 	 * cannot be released) until a super block is returned.
434 	 */
435 	int holding;
436 
437 	/* Set when we find or add a device that doesn't have the nonrot flag set. */
438 	bool rotating;
439 	/* Devices support TRIM/discard commands. */
440 	bool discardable;
441 	/* The filesystem is a seed filesystem. */
442 	bool seeding;
443 	/* The mount needs to use a randomly generated fsid. */
444 	bool temp_fsid;
445 	/* Enable/disable the filesystem stats tracking. */
446 	bool collect_fs_stats;
447 
448 	struct btrfs_fs_info *fs_info;
449 	/* sysfs kobjects */
450 	struct kobject fsid_kobj;
451 	struct kobject *devices_kobj;
452 	struct kobject *devinfo_kobj;
453 	struct completion kobj_unregister;
454 
455 	enum btrfs_chunk_allocation_policy chunk_alloc_policy;
456 
457 	/* Policy used to read the mirrored stripes. */
458 	enum btrfs_read_policy read_policy;
459 
460 #ifdef CONFIG_BTRFS_EXPERIMENTAL
461 	/*
462 	 * Minimum contiguous reads before switching to next device, the unit
463 	 * is one block/sectorsize.
464 	 */
465 	u32 rr_min_contig_read;
466 
467 	/* Device to be used for reading in case of RAID1. */
468 	u64 read_devid;
469 
470 	/* Checksum mode - offload it or do it synchronously. */
471 	enum btrfs_offload_csum_mode offload_csum_mode;
472 #endif
473 };
474 
475 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
476 			- sizeof(struct btrfs_chunk))		\
477 			/ sizeof(struct btrfs_stripe) + 1)
478 
479 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
480 				- 2 * sizeof(struct btrfs_disk_key)	\
481 				- 2 * sizeof(struct btrfs_chunk))	\
482 				/ sizeof(struct btrfs_stripe) + 1)
483 
484 struct btrfs_io_stripe {
485 	struct btrfs_device *dev;
486 	/* Block mapping. */
487 	u64 physical;
488 	bool rst_search_commit_root;
489 	/* For the endio handler. */
490 	struct btrfs_io_context *bioc;
491 };
492 
493 struct btrfs_discard_stripe {
494 	struct btrfs_device *dev;
495 	u64 physical;
496 	u64 length;
497 };
498 
499 /*
500  * Context for IO submission for device stripe.
501  *
502  * - Track the unfinished mirrors for mirror based profiles
503  *   Mirror based profiles are SINGLE/DUP/RAID1/RAID10.
504  *
505  * - Contain the logical -> physical mapping info
506  *   Used by submit_stripe_bio() for mapping logical bio
507  *   into physical device address.
508  *
509  * - Contain device replace info
510  *   Used by handle_ops_on_dev_replace() to copy logical bios
511  *   into the new device.
512  *
513  * - Contain RAID56 full stripe logical bytenrs
514  */
515 struct btrfs_io_context {
516 	refcount_t refs;
517 	struct btrfs_fs_info *fs_info;
518 	/* Taken from struct btrfs_chunk_map::type. */
519 	u64 map_type;
520 	struct bio *orig_bio;
521 	atomic_t error;
522 	u16 max_errors;
523 	bool use_rst;
524 
525 	u64 logical;
526 	u64 size;
527 	/* Raid stripe tree ordered entry. */
528 	struct list_head rst_ordered_entry;
529 
530 	/*
531 	 * The total number of stripes, including the extra duplicated
532 	 * stripe for replace.
533 	 */
534 	u16 num_stripes;
535 
536 	/*
537 	 * The mirror_num of this bioc.
538 	 *
539 	 * This is for reads which use 0 as mirror_num, thus we should return a
540 	 * valid mirror_num (>0) for the reader.
541 	 */
542 	u16 mirror_num;
543 
544 	/*
545 	 * The following two members are for dev-replace case only.
546 	 *
547 	 * @replace_nr_stripes:	Number of duplicated stripes which need to be
548 	 *			written to replace target.
549 	 *			Should be <= 2 (2 for DUP, otherwise <= 1).
550 	 * @replace_stripe_src:	The array indicates where the duplicated stripes
551 	 *			are from.
552 	 *
553 	 * The @replace_stripe_src[] array is mostly for RAID56 cases.
554 	 * As non-RAID56 stripes share the same contents of the mapped range,
555 	 * thus no need to bother where the duplicated ones are from.
556 	 *
557 	 * But for RAID56 case, all stripes contain different contents, thus
558 	 * we need a way to know the mapping.
559 	 *
560 	 * There is an example for the two members, using a RAID5 write:
561 	 *
562 	 *   num_stripes:	4 (3 + 1 duplicated write)
563 	 *   stripes[0]:	dev = devid 1, physical = X
564 	 *   stripes[1]:	dev = devid 2, physical = Y
565 	 *   stripes[2]:	dev = devid 3, physical = Z
566 	 *   stripes[3]:	dev = devid 0, physical = Y
567 	 *
568 	 * replace_nr_stripes = 1
569 	 * replace_stripe_src = 1	<- Means stripes[1] is involved in replace.
570 	 *				   The duplicated stripe index would be
571 	 *				   (@num_stripes - 1).
572 	 *
573 	 * Note, that we can still have cases replace_nr_stripes = 2 for DUP.
574 	 * In that case, all stripes share the same content, thus we don't
575 	 * need to bother @replace_stripe_src value at all.
576 	 */
577 	u16 replace_nr_stripes;
578 	s16 replace_stripe_src;
579 	/*
580 	 * Logical bytenr of the full stripe start, only for RAID56 cases.
581 	 *
582 	 * When this value is set to other than (u64)-1, the stripes[] should
583 	 * follow this pattern:
584 	 *
585 	 * (real_stripes = num_stripes - replace_nr_stripes)
586 	 * (data_stripes = (is_raid6) ? (real_stripes - 2) : (real_stripes - 1))
587 	 *
588 	 * stripes[0]:			The first data stripe
589 	 * stripes[1]:			The second data stripe
590 	 * ...
591 	 * stripes[data_stripes - 1]:	The last data stripe
592 	 * stripes[data_stripes]:	The P stripe
593 	 * stripes[data_stripes + 1]:	The Q stripe (only for RAID6).
594 	 */
595 	u64 full_stripe_logical;
596 	struct btrfs_io_stripe stripes[];
597 };
598 
599 struct btrfs_device_info {
600 	struct btrfs_device *dev;
601 	u64 dev_offset;
602 	u64 max_avail;
603 	u64 total_avail;
604 };
605 
606 struct btrfs_raid_attr {
607 	u8 sub_stripes;		/* sub_stripes info for map */
608 	u8 dev_stripes;		/* stripes per dev */
609 	u8 devs_max;		/* max devs to use */
610 	u8 devs_min;		/* min devs needed */
611 	u8 tolerated_failures;	/* max tolerated fail devs */
612 	u8 devs_increment;	/* ndevs has to be a multiple of this */
613 	u8 ncopies;		/* how many copies to data has */
614 	u8 nparity;		/* number of stripes worth of bytes to store
615 				 * parity information */
616 	u8 mindev_error;	/* error code if min devs requisite is unmet */
617 	const char raid_name[8]; /* name of the raid */
618 	u64 bg_flag;		/* block group flag of the raid */
619 };
620 
621 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
622 
623 struct btrfs_chunk_map {
624 	struct rb_node rb_node;
625 	/* For mount time dev extent verification. */
626 	int verified_stripes;
627 	refcount_t refs;
628 	u64 start;
629 	u64 chunk_len;
630 	u64 stripe_size;
631 	u64 type;
632 	int io_align;
633 	int io_width;
634 	int num_stripes;
635 	int sub_stripes;
636 	struct btrfs_io_stripe stripes[];
637 };
638 
639 #define btrfs_chunk_map_size(n) (sizeof(struct btrfs_chunk_map) + \
640 				 (sizeof(struct btrfs_io_stripe) * (n)))
641 
btrfs_free_chunk_map(struct btrfs_chunk_map * map)642 static inline void btrfs_free_chunk_map(struct btrfs_chunk_map *map)
643 {
644 	if (map && refcount_dec_and_test(&map->refs)) {
645 		ASSERT(RB_EMPTY_NODE(&map->rb_node));
646 		kfree(map);
647 	}
648 }
649 
650 struct btrfs_balance_control {
651 	struct btrfs_balance_args data;
652 	struct btrfs_balance_args meta;
653 	struct btrfs_balance_args sys;
654 
655 	u64 flags;
656 
657 	struct btrfs_balance_progress stat;
658 };
659 
660 /*
661  * Search for a given device by the set parameters
662  */
663 struct btrfs_dev_lookup_args {
664 	u64 devid;
665 	u8 *uuid;
666 	u8 *fsid;
667 	/*
668 	 * If devt is specified, all other members will be ignored as it is
669 	 * enough to uniquely locate a device.
670 	 */
671 	dev_t devt;
672 	bool missing;
673 };
674 
675 /* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */
676 #define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 }
677 
678 #define BTRFS_DEV_LOOKUP_ARGS(name) \
679 	struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT
680 
681 enum btrfs_map_op {
682 	BTRFS_MAP_READ,
683 	BTRFS_MAP_WRITE,
684 	BTRFS_MAP_GET_READ_MIRRORS,
685 };
686 
btrfs_op(const struct bio * bio)687 static inline enum btrfs_map_op btrfs_op(const struct bio *bio)
688 {
689 	switch (bio_op(bio)) {
690 	case REQ_OP_WRITE:
691 	case REQ_OP_ZONE_APPEND:
692 		return BTRFS_MAP_WRITE;
693 	default:
694 		WARN_ON_ONCE(1);
695 		fallthrough;
696 	case REQ_OP_READ:
697 		return BTRFS_MAP_READ;
698 	}
699 }
700 
btrfs_chunk_item_size(int num_stripes)701 static inline unsigned long btrfs_chunk_item_size(int num_stripes)
702 {
703 	ASSERT(num_stripes);
704 	return sizeof(struct btrfs_chunk) +
705 		sizeof(struct btrfs_stripe) * (num_stripes - 1);
706 }
707 
708 /*
709  * Do the type safe conversion from stripe_nr to offset inside the chunk.
710  *
711  * @stripe_nr is u32, with left shift it can overflow u32 for chunks larger
712  * than 4G.  This does the proper type cast to avoid overflow.
713  */
btrfs_stripe_nr_to_offset(u32 stripe_nr)714 static inline u64 btrfs_stripe_nr_to_offset(u32 stripe_nr)
715 {
716 	return (u64)stripe_nr << BTRFS_STRIPE_LEN_SHIFT;
717 }
718 
719 void btrfs_get_bioc(struct btrfs_io_context *bioc);
720 void btrfs_put_bioc(struct btrfs_io_context *bioc);
721 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
722 		    u64 logical, u64 *length,
723 		    struct btrfs_io_context **bioc_ret,
724 		    struct btrfs_io_stripe *smap, int *mirror_num_ret);
725 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
726 			   struct btrfs_io_stripe *smap, u64 logical,
727 			   u32 length, int mirror_num);
728 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
729 					       u64 logical, u64 *length_ret,
730 					       u32 *num_stripes);
731 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
732 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
733 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
734 					     struct btrfs_space_info *space_info,
735 					     u64 type);
736 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info);
737 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
738 		       blk_mode_t flags, void *holder);
739 struct btrfs_device *btrfs_scan_one_device(const char *path, bool mount_arg_dev);
740 int btrfs_forget_devices(dev_t devt);
741 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
742 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
743 void btrfs_assign_next_active_device(struct btrfs_device *device,
744 				     struct btrfs_device *this_dev);
745 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
746 						  u64 devid,
747 						  const char *devpath);
748 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
749 				 struct btrfs_dev_lookup_args *args,
750 				 const char *path);
751 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
752 					const u64 *devid, const u8 *uuid,
753 					const char *path);
754 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args);
755 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
756 		    struct btrfs_dev_lookup_args *args,
757 		    struct file **bdev_file);
758 void __exit btrfs_cleanup_fs_uuids(void);
759 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
760 int btrfs_grow_device(struct btrfs_trans_handle *trans,
761 		      struct btrfs_device *device, u64 new_size);
762 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
763 				       const struct btrfs_dev_lookup_args *args);
764 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
765 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
766 int btrfs_balance(struct btrfs_fs_info *fs_info,
767 		  struct btrfs_balance_control *bctl,
768 		  struct btrfs_ioctl_balance_args *bargs);
769 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
770 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
771 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
772 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
773 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset,
774 			 bool verbose);
775 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
776 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset);
777 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
778 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
779 			struct btrfs_ioctl_get_dev_stats *stats);
780 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
781 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
782 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
783 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
784 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
785 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
786 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
787 				    u64 logical);
788 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map);
789 int btrfs_nr_parity_stripes(u64 type);
790 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
791 				     struct btrfs_block_group *bg);
792 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
793 
794 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
795 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp);
796 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
797 #endif
798 
799 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
800 					     u64 logical, u64 length);
801 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
802 						    u64 logical, u64 length);
803 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
804 					    u64 logical, u64 length);
805 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
806 struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
807 						int copy_num, bool drop_cache);
808 void btrfs_release_disk_super(struct btrfs_super_block *super);
809 
btrfs_dev_stat_inc(struct btrfs_device * dev,int index)810 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
811 				      int index)
812 {
813 	atomic_inc(dev->dev_stat_values + index);
814 	/*
815 	 * This memory barrier orders stores updating statistics before stores
816 	 * updating dev_stats_ccnt.
817 	 *
818 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
819 	 */
820 	smp_mb__before_atomic();
821 	atomic_inc(&dev->dev_stats_ccnt);
822 }
823 
btrfs_dev_stat_read(struct btrfs_device * dev,int index)824 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
825 				      int index)
826 {
827 	return atomic_read(dev->dev_stat_values + index);
828 }
829 
btrfs_dev_stat_read_and_reset(struct btrfs_device * dev,int index)830 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
831 						int index)
832 {
833 	int ret;
834 
835 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
836 	/*
837 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
838 	 * - RMW operations that have a return value are fully ordered;
839 	 *
840 	 * This implicit memory barriers is paired with the smp_rmb in
841 	 * btrfs_run_dev_stats
842 	 */
843 	atomic_inc(&dev->dev_stats_ccnt);
844 	return ret;
845 }
846 
btrfs_dev_stat_set(struct btrfs_device * dev,int index,unsigned long val)847 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
848 				      int index, unsigned long val)
849 {
850 	atomic_set(dev->dev_stat_values + index, val);
851 	/*
852 	 * This memory barrier orders stores updating statistics before stores
853 	 * updating dev_stats_ccnt.
854 	 *
855 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
856 	 */
857 	smp_mb__before_atomic();
858 	atomic_inc(&dev->dev_stats_ccnt);
859 }
860 
btrfs_dev_name(const struct btrfs_device * device)861 static inline const char *btrfs_dev_name(const struct btrfs_device *device)
862 {
863 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
864 		return "<missing disk>";
865 	else
866 		return rcu_dereference(device->name);
867 }
868 
btrfs_warn_unknown_chunk_allocation(enum btrfs_chunk_allocation_policy pol)869 static inline void btrfs_warn_unknown_chunk_allocation(enum btrfs_chunk_allocation_policy pol)
870 {
871 	WARN_ONCE(1, "unknown allocation policy %d, fallback to regular", pol);
872 }
873 
btrfs_fs_devices_inc_holding(struct btrfs_fs_devices * fs_devices)874 static inline void btrfs_fs_devices_inc_holding(struct btrfs_fs_devices *fs_devices)
875 {
876 	lockdep_assert_held(&uuid_mutex);
877 	ASSERT(fs_devices->holding >= 0);
878 	fs_devices->holding++;
879 }
880 
btrfs_fs_devices_dec_holding(struct btrfs_fs_devices * fs_devices)881 static inline void btrfs_fs_devices_dec_holding(struct btrfs_fs_devices *fs_devices)
882 {
883 	lockdep_assert_held(&uuid_mutex);
884 	ASSERT(fs_devices->holding > 0);
885 	fs_devices->holding--;
886 }
887 
888 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
889 
890 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
891 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
892 					struct btrfs_device *failing_dev);
893 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device);
894 
895 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
896 int btrfs_bg_type_to_factor(u64 flags);
897 const char *btrfs_bg_type_to_raid_name(u64 flags);
898 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
899 bool btrfs_verify_dev_items(const struct btrfs_fs_info *fs_info);
900 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
901 
902 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr);
903 const u8 *btrfs_sb_fsid_ptr(const struct btrfs_super_block *sb);
904 
905 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
906 struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
907 						u64 logical, u16 total_stripes);
908 #endif
909 
910 #endif
911