xref: /linux/fs/btrfs/volumes.h (revision f92b71ffca8c7e45e194aecc85e31bd11582f4d2)
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  * Arbitratry 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(const_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(const_ilog2(BTRFS_BLOCK_GROUP_RAID0) >
62 	      ilog2(BTRFS_BLOCK_GROUP_TYPE_MASK));
63 
64 /* ilog2() can handle both constants and variables */
65 #define BTRFS_BG_FLAG_TO_INDEX(profile)					\
66 	ilog2((profile) >> (ilog2(BTRFS_BLOCK_GROUP_RAID0) - 1))
67 
68 enum btrfs_raid_types {
69 	/* SINGLE is the special one as it doesn't have on-disk bit. */
70 	BTRFS_RAID_SINGLE  = 0,
71 
72 	BTRFS_RAID_RAID0   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID0),
73 	BTRFS_RAID_RAID1   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1),
74 	BTRFS_RAID_DUP	   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_DUP),
75 	BTRFS_RAID_RAID10  = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID10),
76 	BTRFS_RAID_RAID5   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID5),
77 	BTRFS_RAID_RAID6   = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID6),
78 	BTRFS_RAID_RAID1C3 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C3),
79 	BTRFS_RAID_RAID1C4 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C4),
80 
81 	BTRFS_NR_RAID_TYPES
82 };
83 
84 /*
85  * Use sequence counter to get consistent device stat data on
86  * 32-bit processors.
87  */
88 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
89 #include <linux/seqlock.h>
90 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
91 #define btrfs_device_data_ordered_init(device)	\
92 	seqcount_init(&device->data_seqcount)
93 #else
94 #define btrfs_device_data_ordered_init(device) do { } while (0)
95 #endif
96 
97 #define BTRFS_DEV_STATE_WRITEABLE	(0)
98 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
99 #define BTRFS_DEV_STATE_MISSING		(2)
100 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
101 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
102 #define BTRFS_DEV_STATE_NO_READA	(5)
103 
104 /* Special value encoding failure to write primary super block. */
105 #define BTRFS_SUPER_PRIMARY_WRITE_ERROR		(INT_MAX / 2)
106 
107 struct btrfs_fs_devices;
108 
109 struct btrfs_device {
110 	struct list_head dev_list; /* device_list_mutex */
111 	struct list_head dev_alloc_list; /* chunk mutex */
112 	struct list_head post_commit_list; /* chunk mutex */
113 	struct btrfs_fs_devices *fs_devices;
114 	struct btrfs_fs_info *fs_info;
115 
116 	/* Device path or NULL if missing. */
117 	const char __rcu *name;
118 
119 	u64 generation;
120 
121 	struct file *bdev_file;
122 	struct block_device *bdev;
123 
124 	struct btrfs_zoned_device_info *zone_info;
125 
126 	/*
127 	 * Device's major-minor number. Must be set even if the device is not
128 	 * opened (bdev == NULL), unless the device is missing.
129 	 */
130 	dev_t devt;
131 	unsigned long dev_state;
132 	blk_status_t last_flush_error;
133 
134 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
135 	seqcount_t data_seqcount;
136 #endif
137 
138 	/* the internal btrfs device id */
139 	u64 devid;
140 
141 	/* size of the device in memory */
142 	u64 total_bytes;
143 
144 	/* size of the device on disk */
145 	u64 disk_total_bytes;
146 
147 	/* bytes used */
148 	u64 bytes_used;
149 
150 	/* optimal io alignment for this device */
151 	u32 io_align;
152 
153 	/* optimal io width for this device */
154 	u32 io_width;
155 	/* type and info about this device */
156 	u64 type;
157 
158 	/*
159 	 * Counter of super block write errors, values larger than
160 	 * BTRFS_SUPER_PRIMARY_WRITE_ERROR encode primary super block write failure.
161 	 */
162 	atomic_t sb_write_errors;
163 
164 	/* minimal io size for this device */
165 	u32 sector_size;
166 
167 	/* physical drive uuid (or lvm uuid) */
168 	u8 uuid[BTRFS_UUID_SIZE];
169 
170 	/*
171 	 * size of the device on the current transaction
172 	 *
173 	 * This variant is update when committing the transaction,
174 	 * and protected by chunk mutex
175 	 */
176 	u64 commit_total_bytes;
177 
178 	/* bytes used on the current transaction */
179 	u64 commit_bytes_used;
180 
181 	/* Bio used for flushing device barriers */
182 	struct bio flush_bio;
183 	struct completion flush_wait;
184 
185 	/* per-device scrub information */
186 	struct scrub_ctx *scrub_ctx;
187 
188 	/* disk I/O failure stats. For detailed description refer to
189 	 * enum btrfs_dev_stat_values in ioctl.h */
190 	int dev_stats_valid;
191 
192 	/* Counter to record the change of device stats */
193 	atomic_t dev_stats_ccnt;
194 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
195 
196 	struct extent_io_tree alloc_state;
197 
198 	struct completion kobj_unregister;
199 	/* For sysfs/FSID/devinfo/devid/ */
200 	struct kobject devid_kobj;
201 
202 	/* Bandwidth limit for scrub, in bytes */
203 	u64 scrub_speed_max;
204 };
205 
206 /*
207  * Block group or device which contains an active swapfile. Used for preventing
208  * unsafe operations while a swapfile is active.
209  *
210  * These are sorted on (ptr, inode) (note that a block group or device can
211  * contain more than one swapfile). We compare the pointer values because we
212  * don't actually care what the object is, we just need a quick check whether
213  * the object exists in the rbtree.
214  */
215 struct btrfs_swapfile_pin {
216 	struct rb_node node;
217 	void *ptr;
218 	struct inode *inode;
219 	/*
220 	 * If true, ptr points to a struct btrfs_block_group. Otherwise, ptr
221 	 * points to a struct btrfs_device.
222 	 */
223 	bool is_block_group;
224 	/*
225 	 * Only used when 'is_block_group' is true and it is the number of
226 	 * extents used by a swapfile for this block group ('ptr' field).
227 	 */
228 	int bg_extent_count;
229 };
230 
231 /*
232  * If we read those variants at the context of their own lock, we needn't
233  * use the following helpers, reading them directly is safe.
234  */
235 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
236 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
237 static inline u64							\
238 btrfs_device_get_##name(const struct btrfs_device *dev)			\
239 {									\
240 	u64 size;							\
241 	unsigned int seq;						\
242 									\
243 	do {								\
244 		seq = read_seqcount_begin(&dev->data_seqcount);		\
245 		size = dev->name;					\
246 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
247 	return size;							\
248 }									\
249 									\
250 static inline void							\
251 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
252 {									\
253 	preempt_disable();						\
254 	write_seqcount_begin(&dev->data_seqcount);			\
255 	dev->name = size;						\
256 	write_seqcount_end(&dev->data_seqcount);			\
257 	preempt_enable();						\
258 }
259 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
260 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
261 static inline u64							\
262 btrfs_device_get_##name(const struct btrfs_device *dev)			\
263 {									\
264 	u64 size;							\
265 									\
266 	preempt_disable();						\
267 	size = dev->name;						\
268 	preempt_enable();						\
269 	return size;							\
270 }									\
271 									\
272 static inline void							\
273 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
274 {									\
275 	preempt_disable();						\
276 	dev->name = size;						\
277 	preempt_enable();						\
278 }
279 #else
280 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
281 static inline u64							\
282 btrfs_device_get_##name(const struct btrfs_device *dev)			\
283 {									\
284 	return dev->name;						\
285 }									\
286 									\
287 static inline void							\
288 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
289 {									\
290 	dev->name = size;						\
291 }
292 #endif
293 
294 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
295 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
296 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
297 
298 enum btrfs_chunk_allocation_policy {
299 	BTRFS_CHUNK_ALLOC_REGULAR,
300 	BTRFS_CHUNK_ALLOC_ZONED,
301 };
302 
303 #define BTRFS_DEFAULT_RR_MIN_CONTIG_READ	(SZ_256K)
304 /* Keep in sync with raid_attr table, current maximum is RAID1C4. */
305 #define BTRFS_RAID1_MAX_MIRRORS			(4)
306 /*
307  * Read policies for mirrored block group profiles, read picks the stripe based
308  * on these policies.
309  */
310 enum btrfs_read_policy {
311 	/* Use process PID to choose the stripe */
312 	BTRFS_READ_POLICY_PID,
313 #ifdef CONFIG_BTRFS_EXPERIMENTAL
314 	/* Balancing RAID1 reads across all striped devices (round-robin). */
315 	BTRFS_READ_POLICY_RR,
316 	/* Read from a specific device. */
317 	BTRFS_READ_POLICY_DEVID,
318 #endif
319 	BTRFS_NR_READ_POLICY,
320 };
321 
322 #ifdef CONFIG_BTRFS_EXPERIMENTAL
323 /*
324  * Checksum mode - offload it to workqueues or do it synchronously in
325  * btrfs_submit_chunk().
326  */
327 enum btrfs_offload_csum_mode {
328 	/*
329 	 * Choose offloading checksum or do it synchronously automatically.
330 	 * Do it synchronously if the checksum is fast, or offload to workqueues
331 	 * otherwise.
332 	 */
333 	BTRFS_OFFLOAD_CSUM_AUTO,
334 	/* Always offload checksum to workqueues. */
335 	BTRFS_OFFLOAD_CSUM_FORCE_ON,
336 	/* Never offload checksum to workqueues. */
337 	BTRFS_OFFLOAD_CSUM_FORCE_OFF,
338 };
339 #endif
340 
341 struct btrfs_fs_devices {
342 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
343 
344 	/*
345 	 * UUID written into the btree blocks:
346 	 *
347 	 * - If metadata_uuid != fsid then super block must have
348 	 *   BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag set.
349 	 *
350 	 * - Following shall be true at all times:
351 	 *   - metadata_uuid == btrfs_header::fsid
352 	 *   - metadata_uuid == btrfs_dev_item::fsid
353 	 *
354 	 * - Relations between fsid and metadata_uuid in sb and fs_devices:
355 	 *   - Normal:
356 	 *       fs_devices->fsid == fs_devices->metadata_uuid == sb->fsid
357 	 *       sb->metadata_uuid == 0
358 	 *
359 	 *   - When the BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag is set:
360 	 *       fs_devices->fsid == sb->fsid
361 	 *       fs_devices->metadata_uuid == sb->metadata_uuid
362 	 *
363 	 *   - When in-memory fs_devices->temp_fsid is true
364 	 *	 fs_devices->fsid = random
365 	 *	 fs_devices->metadata_uuid == sb->fsid
366 	 */
367 	u8 metadata_uuid[BTRFS_FSID_SIZE];
368 
369 	struct list_head fs_list;
370 
371 	/*
372 	 * Number of devices under this fsid including missing and
373 	 * replace-target device and excludes seed devices.
374 	 */
375 	u64 num_devices;
376 
377 	/*
378 	 * The number of devices that successfully opened, including
379 	 * replace-target, excludes seed devices.
380 	 */
381 	u64 open_devices;
382 
383 	/* The number of devices that are under the chunk allocation list. */
384 	u64 rw_devices;
385 
386 	/* Count of missing devices under this fsid excluding seed device. */
387 	u64 missing_devices;
388 	u64 total_rw_bytes;
389 
390 	/*
391 	 * Count of devices from btrfs_super_block::num_devices for this fsid,
392 	 * which includes the seed device, excludes the transient replace-target
393 	 * device.
394 	 */
395 	u64 total_devices;
396 
397 	/* Highest generation number of seen devices */
398 	u64 latest_generation;
399 
400 	/*
401 	 * The mount device or a device with highest generation after removal
402 	 * or replace.
403 	 */
404 	struct btrfs_device *latest_dev;
405 
406 	/*
407 	 * All of the devices in the filesystem, protected by a mutex so we can
408 	 * safely walk it to write out the super blocks without worrying about
409 	 * adding/removing by the multi-device code. Scrubbing super block can
410 	 * kick off supers writing by holding this mutex lock.
411 	 */
412 	struct mutex device_list_mutex;
413 
414 	/* List of all devices, protected by device_list_mutex */
415 	struct list_head devices;
416 
417 	/* Devices which can satisfy space allocation. Protected by * chunk_mutex. */
418 	struct list_head alloc_list;
419 
420 	struct list_head seed_list;
421 
422 	/* Count fs-devices opened. */
423 	int opened;
424 
425 	/*
426 	 * Counter of the processes that are holding this fs_devices but not
427 	 * yet opened.
428 	 * This is for mounting handling, as we can only open the fs_devices
429 	 * after a super block is created.  But we cannot take uuid_mutex
430 	 * during sget_fc(), thus we have to hold the fs_devices (meaning it
431 	 * cannot be released) until a super block is returned.
432 	 */
433 	int holding;
434 
435 	/* Set when we find or add a device that doesn't have the nonrot flag set. */
436 	bool rotating;
437 	/* Devices support TRIM/discard commands. */
438 	bool discardable;
439 	/* The filesystem is a seed filesystem. */
440 	bool seeding;
441 	/* The mount needs to use a randomly generated fsid. */
442 	bool temp_fsid;
443 	/* Enable/disable the filesystem stats tracking. */
444 	bool collect_fs_stats;
445 
446 	struct btrfs_fs_info *fs_info;
447 	/* sysfs kobjects */
448 	struct kobject fsid_kobj;
449 	struct kobject *devices_kobj;
450 	struct kobject *devinfo_kobj;
451 	struct completion kobj_unregister;
452 
453 	enum btrfs_chunk_allocation_policy chunk_alloc_policy;
454 
455 	/* Policy used to read the mirrored stripes. */
456 	enum btrfs_read_policy read_policy;
457 
458 #ifdef CONFIG_BTRFS_EXPERIMENTAL
459 	/*
460 	 * Minimum contiguous reads before switching to next device, the unit
461 	 * is one block/sectorsize.
462 	 */
463 	u32 rr_min_contig_read;
464 
465 	/* Device to be used for reading in case of RAID1. */
466 	u64 read_devid;
467 
468 	/* Checksum mode - offload it or do it synchronously. */
469 	enum btrfs_offload_csum_mode offload_csum_mode;
470 #endif
471 };
472 
473 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)	\
474 			- sizeof(struct btrfs_chunk))		\
475 			/ sizeof(struct btrfs_stripe) + 1)
476 
477 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
478 				- 2 * sizeof(struct btrfs_disk_key)	\
479 				- 2 * sizeof(struct btrfs_chunk))	\
480 				/ sizeof(struct btrfs_stripe) + 1)
481 
482 struct btrfs_io_stripe {
483 	struct btrfs_device *dev;
484 	/* Block mapping. */
485 	u64 physical;
486 	bool rst_search_commit_root;
487 	/* For the endio handler. */
488 	struct btrfs_io_context *bioc;
489 };
490 
491 struct btrfs_discard_stripe {
492 	struct btrfs_device *dev;
493 	u64 physical;
494 	u64 length;
495 };
496 
497 /*
498  * Context for IO subsmission for device stripe.
499  *
500  * - Track the unfinished mirrors for mirror based profiles
501  *   Mirror based profiles are SINGLE/DUP/RAID1/RAID10.
502  *
503  * - Contain the logical -> physical mapping info
504  *   Used by submit_stripe_bio() for mapping logical bio
505  *   into physical device address.
506  *
507  * - Contain device replace info
508  *   Used by handle_ops_on_dev_replace() to copy logical bios
509  *   into the new device.
510  *
511  * - Contain RAID56 full stripe logical bytenrs
512  */
513 struct btrfs_io_context {
514 	refcount_t refs;
515 	struct btrfs_fs_info *fs_info;
516 	/* Taken from struct btrfs_chunk_map::type. */
517 	u64 map_type;
518 	struct bio *orig_bio;
519 	atomic_t error;
520 	u16 max_errors;
521 	bool use_rst;
522 
523 	u64 logical;
524 	u64 size;
525 	/* Raid stripe tree ordered entry. */
526 	struct list_head rst_ordered_entry;
527 
528 	/*
529 	 * The total number of stripes, including the extra duplicated
530 	 * stripe for replace.
531 	 */
532 	u16 num_stripes;
533 
534 	/*
535 	 * The mirror_num of this bioc.
536 	 *
537 	 * This is for reads which use 0 as mirror_num, thus we should return a
538 	 * valid mirror_num (>0) for the reader.
539 	 */
540 	u16 mirror_num;
541 
542 	/*
543 	 * The following two members are for dev-replace case only.
544 	 *
545 	 * @replace_nr_stripes:	Number of duplicated stripes which need to be
546 	 *			written to replace target.
547 	 *			Should be <= 2 (2 for DUP, otherwise <= 1).
548 	 * @replace_stripe_src:	The array indicates where the duplicated stripes
549 	 *			are from.
550 	 *
551 	 * The @replace_stripe_src[] array is mostly for RAID56 cases.
552 	 * As non-RAID56 stripes share the same contents of the mapped range,
553 	 * thus no need to bother where the duplicated ones are from.
554 	 *
555 	 * But for RAID56 case, all stripes contain different contents, thus
556 	 * we need a way to know the mapping.
557 	 *
558 	 * There is an example for the two members, using a RAID5 write:
559 	 *
560 	 *   num_stripes:	4 (3 + 1 duplicated write)
561 	 *   stripes[0]:	dev = devid 1, physical = X
562 	 *   stripes[1]:	dev = devid 2, physical = Y
563 	 *   stripes[2]:	dev = devid 3, physical = Z
564 	 *   stripes[3]:	dev = devid 0, physical = Y
565 	 *
566 	 * replace_nr_stripes = 1
567 	 * replace_stripe_src = 1	<- Means stripes[1] is involved in replace.
568 	 *				   The duplicated stripe index would be
569 	 *				   (@num_stripes - 1).
570 	 *
571 	 * Note, that we can still have cases replace_nr_stripes = 2 for DUP.
572 	 * In that case, all stripes share the same content, thus we don't
573 	 * need to bother @replace_stripe_src value at all.
574 	 */
575 	u16 replace_nr_stripes;
576 	s16 replace_stripe_src;
577 	/*
578 	 * Logical bytenr of the full stripe start, only for RAID56 cases.
579 	 *
580 	 * When this value is set to other than (u64)-1, the stripes[] should
581 	 * follow this pattern:
582 	 *
583 	 * (real_stripes = num_stripes - replace_nr_stripes)
584 	 * (data_stripes = (is_raid6) ? (real_stripes - 2) : (real_stripes - 1))
585 	 *
586 	 * stripes[0]:			The first data stripe
587 	 * stripes[1]:			The second data stripe
588 	 * ...
589 	 * stripes[data_stripes - 1]:	The last data stripe
590 	 * stripes[data_stripes]:	The P stripe
591 	 * stripes[data_stripes + 1]:	The Q stripe (only for RAID6).
592 	 */
593 	u64 full_stripe_logical;
594 	struct btrfs_io_stripe stripes[];
595 };
596 
597 struct btrfs_device_info {
598 	struct btrfs_device *dev;
599 	u64 dev_offset;
600 	u64 max_avail;
601 	u64 total_avail;
602 };
603 
604 struct btrfs_raid_attr {
605 	u8 sub_stripes;		/* sub_stripes info for map */
606 	u8 dev_stripes;		/* stripes per dev */
607 	u8 devs_max;		/* max devs to use */
608 	u8 devs_min;		/* min devs needed */
609 	u8 tolerated_failures;	/* max tolerated fail devs */
610 	u8 devs_increment;	/* ndevs has to be a multiple of this */
611 	u8 ncopies;		/* how many copies to data has */
612 	u8 nparity;		/* number of stripes worth of bytes to store
613 				 * parity information */
614 	u8 mindev_error;	/* error code if min devs requisite is unmet */
615 	const char raid_name[8]; /* name of the raid */
616 	u64 bg_flag;		/* block group flag of the raid */
617 };
618 
619 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
620 
621 struct btrfs_chunk_map {
622 	struct rb_node rb_node;
623 	/* For mount time dev extent verification. */
624 	int verified_stripes;
625 	refcount_t refs;
626 	u64 start;
627 	u64 chunk_len;
628 	u64 stripe_size;
629 	u64 type;
630 	int io_align;
631 	int io_width;
632 	int num_stripes;
633 	int sub_stripes;
634 	struct btrfs_io_stripe stripes[];
635 };
636 
637 #define btrfs_chunk_map_size(n) (sizeof(struct btrfs_chunk_map) + \
638 				 (sizeof(struct btrfs_io_stripe) * (n)))
639 
btrfs_free_chunk_map(struct btrfs_chunk_map * map)640 static inline void btrfs_free_chunk_map(struct btrfs_chunk_map *map)
641 {
642 	if (map && refcount_dec_and_test(&map->refs)) {
643 		ASSERT(RB_EMPTY_NODE(&map->rb_node));
644 		kfree(map);
645 	}
646 }
647 
648 struct btrfs_balance_control {
649 	struct btrfs_balance_args data;
650 	struct btrfs_balance_args meta;
651 	struct btrfs_balance_args sys;
652 
653 	u64 flags;
654 
655 	struct btrfs_balance_progress stat;
656 };
657 
658 /*
659  * Search for a given device by the set parameters
660  */
661 struct btrfs_dev_lookup_args {
662 	u64 devid;
663 	u8 *uuid;
664 	u8 *fsid;
665 	bool missing;
666 };
667 
668 /* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */
669 #define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 }
670 
671 #define BTRFS_DEV_LOOKUP_ARGS(name) \
672 	struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT
673 
674 enum btrfs_map_op {
675 	BTRFS_MAP_READ,
676 	BTRFS_MAP_WRITE,
677 	BTRFS_MAP_GET_READ_MIRRORS,
678 };
679 
btrfs_op(const struct bio * bio)680 static inline enum btrfs_map_op btrfs_op(const struct bio *bio)
681 {
682 	switch (bio_op(bio)) {
683 	case REQ_OP_WRITE:
684 	case REQ_OP_ZONE_APPEND:
685 		return BTRFS_MAP_WRITE;
686 	default:
687 		WARN_ON_ONCE(1);
688 		fallthrough;
689 	case REQ_OP_READ:
690 		return BTRFS_MAP_READ;
691 	}
692 }
693 
btrfs_chunk_item_size(int num_stripes)694 static inline unsigned long btrfs_chunk_item_size(int num_stripes)
695 {
696 	ASSERT(num_stripes);
697 	return sizeof(struct btrfs_chunk) +
698 		sizeof(struct btrfs_stripe) * (num_stripes - 1);
699 }
700 
701 /*
702  * Do the type safe conversion from stripe_nr to offset inside the chunk.
703  *
704  * @stripe_nr is u32, with left shift it can overflow u32 for chunks larger
705  * than 4G.  This does the proper type cast to avoid overflow.
706  */
btrfs_stripe_nr_to_offset(u32 stripe_nr)707 static inline u64 btrfs_stripe_nr_to_offset(u32 stripe_nr)
708 {
709 	return (u64)stripe_nr << BTRFS_STRIPE_LEN_SHIFT;
710 }
711 
712 void btrfs_get_bioc(struct btrfs_io_context *bioc);
713 void btrfs_put_bioc(struct btrfs_io_context *bioc);
714 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
715 		    u64 logical, u64 *length,
716 		    struct btrfs_io_context **bioc_ret,
717 		    struct btrfs_io_stripe *smap, int *mirror_num_ret);
718 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
719 			   struct btrfs_io_stripe *smap, u64 logical,
720 			   u32 length, int mirror_num);
721 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
722 					       u64 logical, u64 *length_ret,
723 					       u32 *num_stripes);
724 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
725 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
726 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
727 					     struct btrfs_space_info *space_info,
728 					     u64 type);
729 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info);
730 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
731 		       blk_mode_t flags, void *holder);
732 struct btrfs_device *btrfs_scan_one_device(const char *path, bool mount_arg_dev);
733 int btrfs_forget_devices(dev_t devt);
734 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
735 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
736 void btrfs_assign_next_active_device(struct btrfs_device *device,
737 				     struct btrfs_device *this_dev);
738 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
739 						  u64 devid,
740 						  const char *devpath);
741 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
742 				 struct btrfs_dev_lookup_args *args,
743 				 const char *path);
744 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
745 					const u64 *devid, const u8 *uuid,
746 					const char *path);
747 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args);
748 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
749 		    struct btrfs_dev_lookup_args *args,
750 		    struct file **bdev_file);
751 void __exit btrfs_cleanup_fs_uuids(void);
752 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
753 int btrfs_grow_device(struct btrfs_trans_handle *trans,
754 		      struct btrfs_device *device, u64 new_size);
755 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
756 				       const struct btrfs_dev_lookup_args *args);
757 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
758 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
759 int btrfs_balance(struct btrfs_fs_info *fs_info,
760 		  struct btrfs_balance_control *bctl,
761 		  struct btrfs_ioctl_balance_args *bargs);
762 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
763 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
764 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
765 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
766 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset,
767 			 bool verbose);
768 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
769 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset);
770 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
771 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
772 			struct btrfs_ioctl_get_dev_stats *stats);
773 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
774 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
775 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
776 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
777 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
778 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
779 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
780 				    u64 logical);
781 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map);
782 int btrfs_nr_parity_stripes(u64 type);
783 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
784 				     struct btrfs_block_group *bg);
785 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
786 
787 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
788 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp);
789 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
790 #endif
791 
792 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
793 					     u64 logical, u64 length);
794 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
795 						    u64 logical, u64 length);
796 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
797 					    u64 logical, u64 length);
798 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
799 struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
800 						int copy_num, bool drop_cache);
801 void btrfs_release_disk_super(struct btrfs_super_block *super);
802 
btrfs_dev_stat_inc(struct btrfs_device * dev,int index)803 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
804 				      int index)
805 {
806 	atomic_inc(dev->dev_stat_values + index);
807 	/*
808 	 * This memory barrier orders stores updating statistics before stores
809 	 * updating dev_stats_ccnt.
810 	 *
811 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
812 	 */
813 	smp_mb__before_atomic();
814 	atomic_inc(&dev->dev_stats_ccnt);
815 }
816 
btrfs_dev_stat_read(struct btrfs_device * dev,int index)817 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
818 				      int index)
819 {
820 	return atomic_read(dev->dev_stat_values + index);
821 }
822 
btrfs_dev_stat_read_and_reset(struct btrfs_device * dev,int index)823 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
824 						int index)
825 {
826 	int ret;
827 
828 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
829 	/*
830 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
831 	 * - RMW operations that have a return value are fully ordered;
832 	 *
833 	 * This implicit memory barriers is paired with the smp_rmb in
834 	 * btrfs_run_dev_stats
835 	 */
836 	atomic_inc(&dev->dev_stats_ccnt);
837 	return ret;
838 }
839 
btrfs_dev_stat_set(struct btrfs_device * dev,int index,unsigned long val)840 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
841 				      int index, unsigned long val)
842 {
843 	atomic_set(dev->dev_stat_values + index, val);
844 	/*
845 	 * This memory barrier orders stores updating statistics before stores
846 	 * updating dev_stats_ccnt.
847 	 *
848 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
849 	 */
850 	smp_mb__before_atomic();
851 	atomic_inc(&dev->dev_stats_ccnt);
852 }
853 
btrfs_dev_name(const struct btrfs_device * device)854 static inline const char *btrfs_dev_name(const struct btrfs_device *device)
855 {
856 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
857 		return "<missing disk>";
858 	else
859 		return rcu_dereference(device->name);
860 }
861 
btrfs_warn_unknown_chunk_allocation(enum btrfs_chunk_allocation_policy pol)862 static inline void btrfs_warn_unknown_chunk_allocation(enum btrfs_chunk_allocation_policy pol)
863 {
864 	WARN_ONCE(1, "unknown allocation policy %d, fallback to regular", pol);
865 }
866 
btrfs_fs_devices_inc_holding(struct btrfs_fs_devices * fs_devices)867 static inline void btrfs_fs_devices_inc_holding(struct btrfs_fs_devices *fs_devices)
868 {
869 	lockdep_assert_held(&uuid_mutex);
870 	ASSERT(fs_devices->holding >= 0);
871 	fs_devices->holding++;
872 }
873 
btrfs_fs_devices_dec_holding(struct btrfs_fs_devices * fs_devices)874 static inline void btrfs_fs_devices_dec_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 
881 void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
882 
883 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
884 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
885 					struct btrfs_device *failing_dev);
886 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device);
887 
888 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
889 int btrfs_bg_type_to_factor(u64 flags);
890 const char *btrfs_bg_type_to_raid_name(u64 flags);
891 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
892 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
893 
894 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr);
895 const u8 *btrfs_sb_fsid_ptr(const struct btrfs_super_block *sb);
896 
897 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
898 struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
899 						u64 logical, u16 total_stripes);
900 #endif
901 
902 #endif
903