Loading...
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
18extern 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
24static_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 */
35static_assert(const_ffs(BTRFS_BLOCK_GROUP_RAID0) <
36 const_ffs(BTRFS_BLOCK_GROUP_PROFILE_MASK & ~BTRFS_BLOCK_GROUP_RAID0));
37static_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
44enum 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
80struct btrfs_zoned_device_info;
81
82struct 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 bdev_handle *bdev_handle;
94 struct block_device *bdev;
95
96 struct btrfs_zoned_device_info *zone_info;
97
98 /*
99 * Device's major-minor number. Must be set even if the device is not
100 * opened (bdev == NULL), unless the device is missing.
101 */
102 dev_t devt;
103 unsigned long dev_state;
104 blk_status_t last_flush_error;
105
106#ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
107 seqcount_t data_seqcount;
108#endif
109
110 /* the internal btrfs device id */
111 u64 devid;
112
113 /* size of the device in memory */
114 u64 total_bytes;
115
116 /* size of the device on disk */
117 u64 disk_total_bytes;
118
119 /* bytes used */
120 u64 bytes_used;
121
122 /* optimal io alignment for this device */
123 u32 io_align;
124
125 /* optimal io width for this device */
126 u32 io_width;
127 /* type and info about this device */
128 u64 type;
129
130 /* minimal io size for this device */
131 u32 sector_size;
132
133 /* physical drive uuid (or lvm uuid) */
134 u8 uuid[BTRFS_UUID_SIZE];
135
136 /*
137 * size of the device on the current transaction
138 *
139 * This variant is update when committing the transaction,
140 * and protected by chunk mutex
141 */
142 u64 commit_total_bytes;
143
144 /* bytes used on the current transaction */
145 u64 commit_bytes_used;
146
147 /* Bio used for flushing device barriers */
148 struct bio flush_bio;
149 struct completion flush_wait;
150
151 /* per-device scrub information */
152 struct scrub_ctx *scrub_ctx;
153
154 /* disk I/O failure stats. For detailed description refer to
155 * enum btrfs_dev_stat_values in ioctl.h */
156 int dev_stats_valid;
157
158 /* Counter to record the change of device stats */
159 atomic_t dev_stats_ccnt;
160 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
161
162 struct extent_io_tree alloc_state;
163
164 struct completion kobj_unregister;
165 /* For sysfs/FSID/devinfo/devid/ */
166 struct kobject devid_kobj;
167
168 /* Bandwidth limit for scrub, in bytes */
169 u64 scrub_speed_max;
170};
171
172/*
173 * Block group or device which contains an active swapfile. Used for preventing
174 * unsafe operations while a swapfile is active.
175 *
176 * These are sorted on (ptr, inode) (note that a block group or device can
177 * contain more than one swapfile). We compare the pointer values because we
178 * don't actually care what the object is, we just need a quick check whether
179 * the object exists in the rbtree.
180 */
181struct btrfs_swapfile_pin {
182 struct rb_node node;
183 void *ptr;
184 struct inode *inode;
185 /*
186 * If true, ptr points to a struct btrfs_block_group. Otherwise, ptr
187 * points to a struct btrfs_device.
188 */
189 bool is_block_group;
190 /*
191 * Only used when 'is_block_group' is true and it is the number of
192 * extents used by a swapfile for this block group ('ptr' field).
193 */
194 int bg_extent_count;
195};
196
197/*
198 * If we read those variants at the context of their own lock, we needn't
199 * use the following helpers, reading them directly is safe.
200 */
201#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
202#define BTRFS_DEVICE_GETSET_FUNCS(name) \
203static inline u64 \
204btrfs_device_get_##name(const struct btrfs_device *dev) \
205{ \
206 u64 size; \
207 unsigned int seq; \
208 \
209 do { \
210 seq = read_seqcount_begin(&dev->data_seqcount); \
211 size = dev->name; \
212 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
213 return size; \
214} \
215 \
216static inline void \
217btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
218{ \
219 preempt_disable(); \
220 write_seqcount_begin(&dev->data_seqcount); \
221 dev->name = size; \
222 write_seqcount_end(&dev->data_seqcount); \
223 preempt_enable(); \
224}
225#elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
226#define BTRFS_DEVICE_GETSET_FUNCS(name) \
227static inline u64 \
228btrfs_device_get_##name(const struct btrfs_device *dev) \
229{ \
230 u64 size; \
231 \
232 preempt_disable(); \
233 size = dev->name; \
234 preempt_enable(); \
235 return size; \
236} \
237 \
238static inline void \
239btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
240{ \
241 preempt_disable(); \
242 dev->name = size; \
243 preempt_enable(); \
244}
245#else
246#define BTRFS_DEVICE_GETSET_FUNCS(name) \
247static inline u64 \
248btrfs_device_get_##name(const struct btrfs_device *dev) \
249{ \
250 return dev->name; \
251} \
252 \
253static inline void \
254btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
255{ \
256 dev->name = size; \
257}
258#endif
259
260BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
261BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
262BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
263
264enum btrfs_chunk_allocation_policy {
265 BTRFS_CHUNK_ALLOC_REGULAR,
266 BTRFS_CHUNK_ALLOC_ZONED,
267};
268
269/*
270 * Read policies for mirrored block group profiles, read picks the stripe based
271 * on these policies.
272 */
273enum btrfs_read_policy {
274 /* Use process PID to choose the stripe */
275 BTRFS_READ_POLICY_PID,
276 BTRFS_NR_READ_POLICY,
277};
278
279struct btrfs_fs_devices {
280 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
281
282 /*
283 * UUID written into the btree blocks:
284 *
285 * - If metadata_uuid != fsid then super block must have
286 * BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag set.
287 *
288 * - Following shall be true at all times:
289 * - metadata_uuid == btrfs_header::fsid
290 * - metadata_uuid == btrfs_dev_item::fsid
291 *
292 * - Relations between fsid and metadata_uuid in sb and fs_devices:
293 * - Normal:
294 * fs_devices->fsid == fs_devices->metadata_uuid == sb->fsid
295 * sb->metadata_uuid == 0
296 *
297 * - When the BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag is set:
298 * fs_devices->fsid == sb->fsid
299 * fs_devices->metadata_uuid == sb->metadata_uuid
300 *
301 * - When in-memory fs_devices->temp_fsid is true
302 * fs_devices->fsid = random
303 * fs_devices->metadata_uuid == sb->fsid
304 */
305 u8 metadata_uuid[BTRFS_FSID_SIZE];
306
307 struct list_head fs_list;
308
309 /*
310 * Number of devices under this fsid including missing and
311 * replace-target device and excludes seed devices.
312 */
313 u64 num_devices;
314
315 /*
316 * The number of devices that successfully opened, including
317 * replace-target, excludes seed devices.
318 */
319 u64 open_devices;
320
321 /* The number of devices that are under the chunk allocation list. */
322 u64 rw_devices;
323
324 /* Count of missing devices under this fsid excluding seed device. */
325 u64 missing_devices;
326 u64 total_rw_bytes;
327
328 /*
329 * Count of devices from btrfs_super_block::num_devices for this fsid,
330 * which includes the seed device, excludes the transient replace-target
331 * device.
332 */
333 u64 total_devices;
334
335 /* Highest generation number of seen devices */
336 u64 latest_generation;
337
338 /*
339 * The mount device or a device with highest generation after removal
340 * or replace.
341 */
342 struct btrfs_device *latest_dev;
343
344 /*
345 * All of the devices in the filesystem, protected by a mutex so we can
346 * safely walk it to write out the super blocks without worrying about
347 * adding/removing by the multi-device code. Scrubbing super block can
348 * kick off supers writing by holding this mutex lock.
349 */
350 struct mutex device_list_mutex;
351
352 /* List of all devices, protected by device_list_mutex */
353 struct list_head devices;
354
355 /* Devices which can satisfy space allocation. Protected by * chunk_mutex. */
356 struct list_head alloc_list;
357
358 struct list_head seed_list;
359
360 /* Count fs-devices opened. */
361 int opened;
362
363 /* Set when we find or add a device that doesn't have the nonrot flag set. */
364 bool rotating;
365 /* Devices support TRIM/discard commands. */
366 bool discardable;
367 /* The filesystem is a seed filesystem. */
368 bool seeding;
369 /* The mount needs to use a randomly generated fsid. */
370 bool temp_fsid;
371
372 struct btrfs_fs_info *fs_info;
373 /* sysfs kobjects */
374 struct kobject fsid_kobj;
375 struct kobject *devices_kobj;
376 struct kobject *devinfo_kobj;
377 struct completion kobj_unregister;
378
379 enum btrfs_chunk_allocation_policy chunk_alloc_policy;
380
381 /* Policy used to read the mirrored stripes. */
382 enum btrfs_read_policy read_policy;
383};
384
385#define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
386 - sizeof(struct btrfs_chunk)) \
387 / sizeof(struct btrfs_stripe) + 1)
388
389#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
390 - 2 * sizeof(struct btrfs_disk_key) \
391 - 2 * sizeof(struct btrfs_chunk)) \
392 / sizeof(struct btrfs_stripe) + 1)
393
394struct btrfs_io_stripe {
395 struct btrfs_device *dev;
396 /* Block mapping. */
397 u64 physical;
398 u64 length;
399 bool is_scrub;
400 /* For the endio handler. */
401 struct btrfs_io_context *bioc;
402};
403
404struct btrfs_discard_stripe {
405 struct btrfs_device *dev;
406 u64 physical;
407 u64 length;
408};
409
410/*
411 * Context for IO subsmission for device stripe.
412 *
413 * - Track the unfinished mirrors for mirror based profiles
414 * Mirror based profiles are SINGLE/DUP/RAID1/RAID10.
415 *
416 * - Contain the logical -> physical mapping info
417 * Used by submit_stripe_bio() for mapping logical bio
418 * into physical device address.
419 *
420 * - Contain device replace info
421 * Used by handle_ops_on_dev_replace() to copy logical bios
422 * into the new device.
423 *
424 * - Contain RAID56 full stripe logical bytenrs
425 */
426struct btrfs_io_context {
427 refcount_t refs;
428 struct btrfs_fs_info *fs_info;
429 /* Taken from struct btrfs_chunk_map::type. */
430 u64 map_type;
431 struct bio *orig_bio;
432 atomic_t error;
433 u16 max_errors;
434
435 u64 logical;
436 u64 size;
437 /* Raid stripe tree ordered entry. */
438 struct list_head rst_ordered_entry;
439
440 /*
441 * The total number of stripes, including the extra duplicated
442 * stripe for replace.
443 */
444 u16 num_stripes;
445
446 /*
447 * The mirror_num of this bioc.
448 *
449 * This is for reads which use 0 as mirror_num, thus we should return a
450 * valid mirror_num (>0) for the reader.
451 */
452 u16 mirror_num;
453
454 /*
455 * The following two members are for dev-replace case only.
456 *
457 * @replace_nr_stripes: Number of duplicated stripes which need to be
458 * written to replace target.
459 * Should be <= 2 (2 for DUP, otherwise <= 1).
460 * @replace_stripe_src: The array indicates where the duplicated stripes
461 * are from.
462 *
463 * The @replace_stripe_src[] array is mostly for RAID56 cases.
464 * As non-RAID56 stripes share the same contents of the mapped range,
465 * thus no need to bother where the duplicated ones are from.
466 *
467 * But for RAID56 case, all stripes contain different contents, thus
468 * we need a way to know the mapping.
469 *
470 * There is an example for the two members, using a RAID5 write:
471 *
472 * num_stripes: 4 (3 + 1 duplicated write)
473 * stripes[0]: dev = devid 1, physical = X
474 * stripes[1]: dev = devid 2, physical = Y
475 * stripes[2]: dev = devid 3, physical = Z
476 * stripes[3]: dev = devid 0, physical = Y
477 *
478 * replace_nr_stripes = 1
479 * replace_stripe_src = 1 <- Means stripes[1] is involved in replace.
480 * The duplicated stripe index would be
481 * (@num_stripes - 1).
482 *
483 * Note, that we can still have cases replace_nr_stripes = 2 for DUP.
484 * In that case, all stripes share the same content, thus we don't
485 * need to bother @replace_stripe_src value at all.
486 */
487 u16 replace_nr_stripes;
488 s16 replace_stripe_src;
489 /*
490 * Logical bytenr of the full stripe start, only for RAID56 cases.
491 *
492 * When this value is set to other than (u64)-1, the stripes[] should
493 * follow this pattern:
494 *
495 * (real_stripes = num_stripes - replace_nr_stripes)
496 * (data_stripes = (is_raid6) ? (real_stripes - 2) : (real_stripes - 1))
497 *
498 * stripes[0]: The first data stripe
499 * stripes[1]: The second data stripe
500 * ...
501 * stripes[data_stripes - 1]: The last data stripe
502 * stripes[data_stripes]: The P stripe
503 * stripes[data_stripes + 1]: The Q stripe (only for RAID6).
504 */
505 u64 full_stripe_logical;
506 struct btrfs_io_stripe stripes[];
507};
508
509struct btrfs_device_info {
510 struct btrfs_device *dev;
511 u64 dev_offset;
512 u64 max_avail;
513 u64 total_avail;
514};
515
516struct btrfs_raid_attr {
517 u8 sub_stripes; /* sub_stripes info for map */
518 u8 dev_stripes; /* stripes per dev */
519 u8 devs_max; /* max devs to use */
520 u8 devs_min; /* min devs needed */
521 u8 tolerated_failures; /* max tolerated fail devs */
522 u8 devs_increment; /* ndevs has to be a multiple of this */
523 u8 ncopies; /* how many copies to data has */
524 u8 nparity; /* number of stripes worth of bytes to store
525 * parity information */
526 u8 mindev_error; /* error code if min devs requisite is unmet */
527 const char raid_name[8]; /* name of the raid */
528 u64 bg_flag; /* block group flag of the raid */
529};
530
531extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
532
533struct btrfs_chunk_map {
534 struct rb_node rb_node;
535 /* For mount time dev extent verification. */
536 int verified_stripes;
537 refcount_t refs;
538 u64 start;
539 u64 chunk_len;
540 u64 stripe_size;
541 u64 type;
542 int io_align;
543 int io_width;
544 int num_stripes;
545 int sub_stripes;
546 struct btrfs_io_stripe stripes[];
547};
548
549#define btrfs_chunk_map_size(n) (sizeof(struct btrfs_chunk_map) + \
550 (sizeof(struct btrfs_io_stripe) * (n)))
551
552static inline void btrfs_free_chunk_map(struct btrfs_chunk_map *map)
553{
554 if (map && refcount_dec_and_test(&map->refs)) {
555 ASSERT(RB_EMPTY_NODE(&map->rb_node));
556 kfree(map);
557 }
558}
559
560struct btrfs_balance_args;
561struct btrfs_balance_progress;
562struct btrfs_balance_control {
563 struct btrfs_balance_args data;
564 struct btrfs_balance_args meta;
565 struct btrfs_balance_args sys;
566
567 u64 flags;
568
569 struct btrfs_balance_progress stat;
570};
571
572/*
573 * Search for a given device by the set parameters
574 */
575struct btrfs_dev_lookup_args {
576 u64 devid;
577 u8 *uuid;
578 u8 *fsid;
579 bool missing;
580};
581
582/* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */
583#define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 }
584
585#define BTRFS_DEV_LOOKUP_ARGS(name) \
586 struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT
587
588enum btrfs_map_op {
589 BTRFS_MAP_READ,
590 BTRFS_MAP_WRITE,
591 BTRFS_MAP_GET_READ_MIRRORS,
592};
593
594static inline enum btrfs_map_op btrfs_op(struct bio *bio)
595{
596 switch (bio_op(bio)) {
597 case REQ_OP_WRITE:
598 case REQ_OP_ZONE_APPEND:
599 return BTRFS_MAP_WRITE;
600 default:
601 WARN_ON_ONCE(1);
602 fallthrough;
603 case REQ_OP_READ:
604 return BTRFS_MAP_READ;
605 }
606}
607
608static inline unsigned long btrfs_chunk_item_size(int num_stripes)
609{
610 ASSERT(num_stripes);
611 return sizeof(struct btrfs_chunk) +
612 sizeof(struct btrfs_stripe) * (num_stripes - 1);
613}
614
615/*
616 * Do the type safe conversion from stripe_nr to offset inside the chunk.
617 *
618 * @stripe_nr is u32, with left shift it can overflow u32 for chunks larger
619 * than 4G. This does the proper type cast to avoid overflow.
620 */
621static inline u64 btrfs_stripe_nr_to_offset(u32 stripe_nr)
622{
623 return (u64)stripe_nr << BTRFS_STRIPE_LEN_SHIFT;
624}
625
626void btrfs_get_bioc(struct btrfs_io_context *bioc);
627void btrfs_put_bioc(struct btrfs_io_context *bioc);
628int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
629 u64 logical, u64 *length,
630 struct btrfs_io_context **bioc_ret,
631 struct btrfs_io_stripe *smap, int *mirror_num_ret);
632int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
633 struct btrfs_io_stripe *smap, u64 logical,
634 u32 length, int mirror_num);
635struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
636 u64 logical, u64 *length_ret,
637 u32 *num_stripes);
638int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
639int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
640struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
641 u64 type);
642void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info);
643int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
644 blk_mode_t flags, void *holder);
645struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
646 bool mount_arg_dev);
647int btrfs_forget_devices(dev_t devt);
648void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
649void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
650void btrfs_assign_next_active_device(struct btrfs_device *device,
651 struct btrfs_device *this_dev);
652struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
653 u64 devid,
654 const char *devpath);
655int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
656 struct btrfs_dev_lookup_args *args,
657 const char *path);
658struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
659 const u64 *devid, const u8 *uuid,
660 const char *path);
661void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args);
662int btrfs_rm_device(struct btrfs_fs_info *fs_info,
663 struct btrfs_dev_lookup_args *args,
664 struct bdev_handle **bdev_handle);
665void __exit btrfs_cleanup_fs_uuids(void);
666int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
667int btrfs_grow_device(struct btrfs_trans_handle *trans,
668 struct btrfs_device *device, u64 new_size);
669struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
670 const struct btrfs_dev_lookup_args *args);
671int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
672int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
673int btrfs_balance(struct btrfs_fs_info *fs_info,
674 struct btrfs_balance_control *bctl,
675 struct btrfs_ioctl_balance_args *bargs);
676void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
677int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
678int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
679int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
680int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
681int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
682int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
683int btrfs_uuid_scan_kthread(void *data);
684bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset);
685void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
686int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
687 struct btrfs_ioctl_get_dev_stats *stats);
688int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
689int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
690int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
691void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
692void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
693void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
694int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
695 u64 logical, u64 len);
696unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
697 u64 logical);
698u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map);
699int btrfs_nr_parity_stripes(u64 type);
700int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
701 struct btrfs_block_group *bg);
702int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
703
704#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
705struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp);
706int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
707#endif
708
709struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp);
710struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
711 u64 logical, u64 length);
712struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
713 u64 logical, u64 length);
714struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
715 u64 logical, u64 length);
716void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
717void btrfs_release_disk_super(struct btrfs_super_block *super);
718
719static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
720 int index)
721{
722 atomic_inc(dev->dev_stat_values + index);
723 /*
724 * This memory barrier orders stores updating statistics before stores
725 * updating dev_stats_ccnt.
726 *
727 * It pairs with smp_rmb() in btrfs_run_dev_stats().
728 */
729 smp_mb__before_atomic();
730 atomic_inc(&dev->dev_stats_ccnt);
731}
732
733static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
734 int index)
735{
736 return atomic_read(dev->dev_stat_values + index);
737}
738
739static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
740 int index)
741{
742 int ret;
743
744 ret = atomic_xchg(dev->dev_stat_values + index, 0);
745 /*
746 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
747 * - RMW operations that have a return value are fully ordered;
748 *
749 * This implicit memory barriers is paired with the smp_rmb in
750 * btrfs_run_dev_stats
751 */
752 atomic_inc(&dev->dev_stats_ccnt);
753 return ret;
754}
755
756static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
757 int index, unsigned long val)
758{
759 atomic_set(dev->dev_stat_values + index, val);
760 /*
761 * This memory barrier orders stores updating statistics before stores
762 * updating dev_stats_ccnt.
763 *
764 * It pairs with smp_rmb() in btrfs_run_dev_stats().
765 */
766 smp_mb__before_atomic();
767 atomic_inc(&dev->dev_stats_ccnt);
768}
769
770static inline const char *btrfs_dev_name(const struct btrfs_device *device)
771{
772 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
773 return "<missing disk>";
774 else
775 return rcu_str_deref(device->name);
776}
777
778void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
779
780struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
781bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
782 struct btrfs_device *failing_dev);
783void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
784 struct block_device *bdev,
785 const char *device_path);
786
787enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
788int btrfs_bg_type_to_factor(u64 flags);
789const char *btrfs_bg_type_to_raid_name(u64 flags);
790int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
791bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
792
793bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr);
794u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb);
795
796#endif
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/sizes.h>
11#include <linux/atomic.h>
12#include <linux/sort.h>
13#include <linux/list.h>
14#include <linux/mutex.h>
15#include <linux/log2.h>
16#include <linux/kobject.h>
17#include <linux/refcount.h>
18#include <linux/completion.h>
19#include <linux/rbtree.h>
20#include <uapi/linux/btrfs.h>
21#include "messages.h"
22#include "rcu-string.h"
23
24struct block_device;
25struct bdev_handle;
26struct btrfs_fs_info;
27struct btrfs_block_group;
28struct btrfs_trans_handle;
29struct btrfs_zoned_device_info;
30
31#define BTRFS_MAX_DATA_CHUNK_SIZE (10ULL * SZ_1G)
32
33extern struct mutex uuid_mutex;
34
35#define BTRFS_STRIPE_LEN SZ_64K
36#define BTRFS_STRIPE_LEN_SHIFT (16)
37#define BTRFS_STRIPE_LEN_MASK (BTRFS_STRIPE_LEN - 1)
38
39static_assert(const_ilog2(BTRFS_STRIPE_LEN) == BTRFS_STRIPE_LEN_SHIFT);
40
41/* Used by sanity check for btrfs_raid_types. */
42#define const_ffs(n) (__builtin_ctzll(n) + 1)
43
44/*
45 * The conversion from BTRFS_BLOCK_GROUP_* bits to btrfs_raid_type requires
46 * RAID0 always to be the lowest profile bit.
47 * Although it's part of on-disk format and should never change, do extra
48 * compile-time sanity checks.
49 */
50static_assert(const_ffs(BTRFS_BLOCK_GROUP_RAID0) <
51 const_ffs(BTRFS_BLOCK_GROUP_PROFILE_MASK & ~BTRFS_BLOCK_GROUP_RAID0));
52static_assert(const_ilog2(BTRFS_BLOCK_GROUP_RAID0) >
53 ilog2(BTRFS_BLOCK_GROUP_TYPE_MASK));
54
55/* ilog2() can handle both constants and variables */
56#define BTRFS_BG_FLAG_TO_INDEX(profile) \
57 ilog2((profile) >> (ilog2(BTRFS_BLOCK_GROUP_RAID0) - 1))
58
59enum btrfs_raid_types {
60 /* SINGLE is the special one as it doesn't have on-disk bit. */
61 BTRFS_RAID_SINGLE = 0,
62
63 BTRFS_RAID_RAID0 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID0),
64 BTRFS_RAID_RAID1 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1),
65 BTRFS_RAID_DUP = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_DUP),
66 BTRFS_RAID_RAID10 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID10),
67 BTRFS_RAID_RAID5 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID5),
68 BTRFS_RAID_RAID6 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID6),
69 BTRFS_RAID_RAID1C3 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C3),
70 BTRFS_RAID_RAID1C4 = BTRFS_BG_FLAG_TO_INDEX(BTRFS_BLOCK_GROUP_RAID1C4),
71
72 BTRFS_NR_RAID_TYPES
73};
74
75/*
76 * Use sequence counter to get consistent device stat data on
77 * 32-bit processors.
78 */
79#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
80#include <linux/seqlock.h>
81#define __BTRFS_NEED_DEVICE_DATA_ORDERED
82#define btrfs_device_data_ordered_init(device) \
83 seqcount_init(&device->data_seqcount)
84#else
85#define btrfs_device_data_ordered_init(device) do { } while (0)
86#endif
87
88#define BTRFS_DEV_STATE_WRITEABLE (0)
89#define BTRFS_DEV_STATE_IN_FS_METADATA (1)
90#define BTRFS_DEV_STATE_MISSING (2)
91#define BTRFS_DEV_STATE_REPLACE_TGT (3)
92#define BTRFS_DEV_STATE_FLUSH_SENT (4)
93#define BTRFS_DEV_STATE_NO_READA (5)
94
95struct btrfs_fs_devices;
96
97struct btrfs_device {
98 struct list_head dev_list; /* device_list_mutex */
99 struct list_head dev_alloc_list; /* chunk mutex */
100 struct list_head post_commit_list; /* chunk mutex */
101 struct btrfs_fs_devices *fs_devices;
102 struct btrfs_fs_info *fs_info;
103
104 struct rcu_string __rcu *name;
105
106 u64 generation;
107
108 struct file *bdev_file;
109 struct block_device *bdev;
110
111 struct btrfs_zoned_device_info *zone_info;
112
113 /*
114 * Device's major-minor number. Must be set even if the device is not
115 * opened (bdev == NULL), unless the device is missing.
116 */
117 dev_t devt;
118 unsigned long dev_state;
119 blk_status_t last_flush_error;
120
121#ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
122 seqcount_t data_seqcount;
123#endif
124
125 /* the internal btrfs device id */
126 u64 devid;
127
128 /* size of the device in memory */
129 u64 total_bytes;
130
131 /* size of the device on disk */
132 u64 disk_total_bytes;
133
134 /* bytes used */
135 u64 bytes_used;
136
137 /* optimal io alignment for this device */
138 u32 io_align;
139
140 /* optimal io width for this device */
141 u32 io_width;
142 /* type and info about this device */
143 u64 type;
144
145 /* minimal io size for this device */
146 u32 sector_size;
147
148 /* physical drive uuid (or lvm uuid) */
149 u8 uuid[BTRFS_UUID_SIZE];
150
151 /*
152 * size of the device on the current transaction
153 *
154 * This variant is update when committing the transaction,
155 * and protected by chunk mutex
156 */
157 u64 commit_total_bytes;
158
159 /* bytes used on the current transaction */
160 u64 commit_bytes_used;
161
162 /* Bio used for flushing device barriers */
163 struct bio flush_bio;
164 struct completion flush_wait;
165
166 /* per-device scrub information */
167 struct scrub_ctx *scrub_ctx;
168
169 /* disk I/O failure stats. For detailed description refer to
170 * enum btrfs_dev_stat_values in ioctl.h */
171 int dev_stats_valid;
172
173 /* Counter to record the change of device stats */
174 atomic_t dev_stats_ccnt;
175 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
176
177 struct extent_io_tree alloc_state;
178
179 struct completion kobj_unregister;
180 /* For sysfs/FSID/devinfo/devid/ */
181 struct kobject devid_kobj;
182
183 /* Bandwidth limit for scrub, in bytes */
184 u64 scrub_speed_max;
185};
186
187/*
188 * Block group or device which contains an active swapfile. Used for preventing
189 * unsafe operations while a swapfile is active.
190 *
191 * These are sorted on (ptr, inode) (note that a block group or device can
192 * contain more than one swapfile). We compare the pointer values because we
193 * don't actually care what the object is, we just need a quick check whether
194 * the object exists in the rbtree.
195 */
196struct btrfs_swapfile_pin {
197 struct rb_node node;
198 void *ptr;
199 struct inode *inode;
200 /*
201 * If true, ptr points to a struct btrfs_block_group. Otherwise, ptr
202 * points to a struct btrfs_device.
203 */
204 bool is_block_group;
205 /*
206 * Only used when 'is_block_group' is true and it is the number of
207 * extents used by a swapfile for this block group ('ptr' field).
208 */
209 int bg_extent_count;
210};
211
212/*
213 * If we read those variants at the context of their own lock, we needn't
214 * use the following helpers, reading them directly is safe.
215 */
216#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
217#define BTRFS_DEVICE_GETSET_FUNCS(name) \
218static inline u64 \
219btrfs_device_get_##name(const struct btrfs_device *dev) \
220{ \
221 u64 size; \
222 unsigned int seq; \
223 \
224 do { \
225 seq = read_seqcount_begin(&dev->data_seqcount); \
226 size = dev->name; \
227 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
228 return size; \
229} \
230 \
231static inline void \
232btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
233{ \
234 preempt_disable(); \
235 write_seqcount_begin(&dev->data_seqcount); \
236 dev->name = size; \
237 write_seqcount_end(&dev->data_seqcount); \
238 preempt_enable(); \
239}
240#elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
241#define BTRFS_DEVICE_GETSET_FUNCS(name) \
242static inline u64 \
243btrfs_device_get_##name(const struct btrfs_device *dev) \
244{ \
245 u64 size; \
246 \
247 preempt_disable(); \
248 size = dev->name; \
249 preempt_enable(); \
250 return size; \
251} \
252 \
253static inline void \
254btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
255{ \
256 preempt_disable(); \
257 dev->name = size; \
258 preempt_enable(); \
259}
260#else
261#define BTRFS_DEVICE_GETSET_FUNCS(name) \
262static inline u64 \
263btrfs_device_get_##name(const struct btrfs_device *dev) \
264{ \
265 return dev->name; \
266} \
267 \
268static inline void \
269btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
270{ \
271 dev->name = size; \
272}
273#endif
274
275BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
276BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
277BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
278
279enum btrfs_chunk_allocation_policy {
280 BTRFS_CHUNK_ALLOC_REGULAR,
281 BTRFS_CHUNK_ALLOC_ZONED,
282};
283
284/*
285 * Read policies for mirrored block group profiles, read picks the stripe based
286 * on these policies.
287 */
288enum btrfs_read_policy {
289 /* Use process PID to choose the stripe */
290 BTRFS_READ_POLICY_PID,
291 BTRFS_NR_READ_POLICY,
292};
293
294#ifdef CONFIG_BTRFS_DEBUG
295/*
296 * Checksum mode - offload it to workqueues or do it synchronously in
297 * btrfs_submit_chunk().
298 */
299enum btrfs_offload_csum_mode {
300 /*
301 * Choose offloading checksum or do it synchronously automatically.
302 * Do it synchronously if the checksum is fast, or offload to workqueues
303 * otherwise.
304 */
305 BTRFS_OFFLOAD_CSUM_AUTO,
306 /* Always offload checksum to workqueues. */
307 BTRFS_OFFLOAD_CSUM_FORCE_ON,
308 /* Never offload checksum to workqueues. */
309 BTRFS_OFFLOAD_CSUM_FORCE_OFF,
310};
311#endif
312
313struct btrfs_fs_devices {
314 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
315
316 /*
317 * UUID written into the btree blocks:
318 *
319 * - If metadata_uuid != fsid then super block must have
320 * BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag set.
321 *
322 * - Following shall be true at all times:
323 * - metadata_uuid == btrfs_header::fsid
324 * - metadata_uuid == btrfs_dev_item::fsid
325 *
326 * - Relations between fsid and metadata_uuid in sb and fs_devices:
327 * - Normal:
328 * fs_devices->fsid == fs_devices->metadata_uuid == sb->fsid
329 * sb->metadata_uuid == 0
330 *
331 * - When the BTRFS_FEATURE_INCOMPAT_METADATA_UUID flag is set:
332 * fs_devices->fsid == sb->fsid
333 * fs_devices->metadata_uuid == sb->metadata_uuid
334 *
335 * - When in-memory fs_devices->temp_fsid is true
336 * fs_devices->fsid = random
337 * fs_devices->metadata_uuid == sb->fsid
338 */
339 u8 metadata_uuid[BTRFS_FSID_SIZE];
340
341 struct list_head fs_list;
342
343 /*
344 * Number of devices under this fsid including missing and
345 * replace-target device and excludes seed devices.
346 */
347 u64 num_devices;
348
349 /*
350 * The number of devices that successfully opened, including
351 * replace-target, excludes seed devices.
352 */
353 u64 open_devices;
354
355 /* The number of devices that are under the chunk allocation list. */
356 u64 rw_devices;
357
358 /* Count of missing devices under this fsid excluding seed device. */
359 u64 missing_devices;
360 u64 total_rw_bytes;
361
362 /*
363 * Count of devices from btrfs_super_block::num_devices for this fsid,
364 * which includes the seed device, excludes the transient replace-target
365 * device.
366 */
367 u64 total_devices;
368
369 /* Highest generation number of seen devices */
370 u64 latest_generation;
371
372 /*
373 * The mount device or a device with highest generation after removal
374 * or replace.
375 */
376 struct btrfs_device *latest_dev;
377
378 /*
379 * All of the devices in the filesystem, protected by a mutex so we can
380 * safely walk it to write out the super blocks without worrying about
381 * adding/removing by the multi-device code. Scrubbing super block can
382 * kick off supers writing by holding this mutex lock.
383 */
384 struct mutex device_list_mutex;
385
386 /* List of all devices, protected by device_list_mutex */
387 struct list_head devices;
388
389 /* Devices which can satisfy space allocation. Protected by * chunk_mutex. */
390 struct list_head alloc_list;
391
392 struct list_head seed_list;
393
394 /* Count fs-devices opened. */
395 int opened;
396
397 /* Set when we find or add a device that doesn't have the nonrot flag set. */
398 bool rotating;
399 /* Devices support TRIM/discard commands. */
400 bool discardable;
401 /* The filesystem is a seed filesystem. */
402 bool seeding;
403 /* The mount needs to use a randomly generated fsid. */
404 bool temp_fsid;
405
406 struct btrfs_fs_info *fs_info;
407 /* sysfs kobjects */
408 struct kobject fsid_kobj;
409 struct kobject *devices_kobj;
410 struct kobject *devinfo_kobj;
411 struct completion kobj_unregister;
412
413 enum btrfs_chunk_allocation_policy chunk_alloc_policy;
414
415 /* Policy used to read the mirrored stripes. */
416 enum btrfs_read_policy read_policy;
417
418#ifdef CONFIG_BTRFS_DEBUG
419 /* Checksum mode - offload it or do it synchronously. */
420 enum btrfs_offload_csum_mode offload_csum_mode;
421#endif
422};
423
424#define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
425 - sizeof(struct btrfs_chunk)) \
426 / sizeof(struct btrfs_stripe) + 1)
427
428#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
429 - 2 * sizeof(struct btrfs_disk_key) \
430 - 2 * sizeof(struct btrfs_chunk)) \
431 / sizeof(struct btrfs_stripe) + 1)
432
433struct btrfs_io_stripe {
434 struct btrfs_device *dev;
435 /* Block mapping. */
436 u64 physical;
437 u64 length;
438 bool is_scrub;
439 /* For the endio handler. */
440 struct btrfs_io_context *bioc;
441};
442
443struct btrfs_discard_stripe {
444 struct btrfs_device *dev;
445 u64 physical;
446 u64 length;
447};
448
449/*
450 * Context for IO subsmission for device stripe.
451 *
452 * - Track the unfinished mirrors for mirror based profiles
453 * Mirror based profiles are SINGLE/DUP/RAID1/RAID10.
454 *
455 * - Contain the logical -> physical mapping info
456 * Used by submit_stripe_bio() for mapping logical bio
457 * into physical device address.
458 *
459 * - Contain device replace info
460 * Used by handle_ops_on_dev_replace() to copy logical bios
461 * into the new device.
462 *
463 * - Contain RAID56 full stripe logical bytenrs
464 */
465struct btrfs_io_context {
466 refcount_t refs;
467 struct btrfs_fs_info *fs_info;
468 /* Taken from struct btrfs_chunk_map::type. */
469 u64 map_type;
470 struct bio *orig_bio;
471 atomic_t error;
472 u16 max_errors;
473
474 u64 logical;
475 u64 size;
476 /* Raid stripe tree ordered entry. */
477 struct list_head rst_ordered_entry;
478
479 /*
480 * The total number of stripes, including the extra duplicated
481 * stripe for replace.
482 */
483 u16 num_stripes;
484
485 /*
486 * The mirror_num of this bioc.
487 *
488 * This is for reads which use 0 as mirror_num, thus we should return a
489 * valid mirror_num (>0) for the reader.
490 */
491 u16 mirror_num;
492
493 /*
494 * The following two members are for dev-replace case only.
495 *
496 * @replace_nr_stripes: Number of duplicated stripes which need to be
497 * written to replace target.
498 * Should be <= 2 (2 for DUP, otherwise <= 1).
499 * @replace_stripe_src: The array indicates where the duplicated stripes
500 * are from.
501 *
502 * The @replace_stripe_src[] array is mostly for RAID56 cases.
503 * As non-RAID56 stripes share the same contents of the mapped range,
504 * thus no need to bother where the duplicated ones are from.
505 *
506 * But for RAID56 case, all stripes contain different contents, thus
507 * we need a way to know the mapping.
508 *
509 * There is an example for the two members, using a RAID5 write:
510 *
511 * num_stripes: 4 (3 + 1 duplicated write)
512 * stripes[0]: dev = devid 1, physical = X
513 * stripes[1]: dev = devid 2, physical = Y
514 * stripes[2]: dev = devid 3, physical = Z
515 * stripes[3]: dev = devid 0, physical = Y
516 *
517 * replace_nr_stripes = 1
518 * replace_stripe_src = 1 <- Means stripes[1] is involved in replace.
519 * The duplicated stripe index would be
520 * (@num_stripes - 1).
521 *
522 * Note, that we can still have cases replace_nr_stripes = 2 for DUP.
523 * In that case, all stripes share the same content, thus we don't
524 * need to bother @replace_stripe_src value at all.
525 */
526 u16 replace_nr_stripes;
527 s16 replace_stripe_src;
528 /*
529 * Logical bytenr of the full stripe start, only for RAID56 cases.
530 *
531 * When this value is set to other than (u64)-1, the stripes[] should
532 * follow this pattern:
533 *
534 * (real_stripes = num_stripes - replace_nr_stripes)
535 * (data_stripes = (is_raid6) ? (real_stripes - 2) : (real_stripes - 1))
536 *
537 * stripes[0]: The first data stripe
538 * stripes[1]: The second data stripe
539 * ...
540 * stripes[data_stripes - 1]: The last data stripe
541 * stripes[data_stripes]: The P stripe
542 * stripes[data_stripes + 1]: The Q stripe (only for RAID6).
543 */
544 u64 full_stripe_logical;
545 struct btrfs_io_stripe stripes[];
546};
547
548struct btrfs_device_info {
549 struct btrfs_device *dev;
550 u64 dev_offset;
551 u64 max_avail;
552 u64 total_avail;
553};
554
555struct btrfs_raid_attr {
556 u8 sub_stripes; /* sub_stripes info for map */
557 u8 dev_stripes; /* stripes per dev */
558 u8 devs_max; /* max devs to use */
559 u8 devs_min; /* min devs needed */
560 u8 tolerated_failures; /* max tolerated fail devs */
561 u8 devs_increment; /* ndevs has to be a multiple of this */
562 u8 ncopies; /* how many copies to data has */
563 u8 nparity; /* number of stripes worth of bytes to store
564 * parity information */
565 u8 mindev_error; /* error code if min devs requisite is unmet */
566 const char raid_name[8]; /* name of the raid */
567 u64 bg_flag; /* block group flag of the raid */
568};
569
570extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
571
572struct btrfs_chunk_map {
573 struct rb_node rb_node;
574 /* For mount time dev extent verification. */
575 int verified_stripes;
576 refcount_t refs;
577 u64 start;
578 u64 chunk_len;
579 u64 stripe_size;
580 u64 type;
581 int io_align;
582 int io_width;
583 int num_stripes;
584 int sub_stripes;
585 struct btrfs_io_stripe stripes[];
586};
587
588#define btrfs_chunk_map_size(n) (sizeof(struct btrfs_chunk_map) + \
589 (sizeof(struct btrfs_io_stripe) * (n)))
590
591static inline void btrfs_free_chunk_map(struct btrfs_chunk_map *map)
592{
593 if (map && refcount_dec_and_test(&map->refs)) {
594 ASSERT(RB_EMPTY_NODE(&map->rb_node));
595 kfree(map);
596 }
597}
598
599struct btrfs_balance_control {
600 struct btrfs_balance_args data;
601 struct btrfs_balance_args meta;
602 struct btrfs_balance_args sys;
603
604 u64 flags;
605
606 struct btrfs_balance_progress stat;
607};
608
609/*
610 * Search for a given device by the set parameters
611 */
612struct btrfs_dev_lookup_args {
613 u64 devid;
614 u8 *uuid;
615 u8 *fsid;
616 bool missing;
617};
618
619/* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */
620#define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 }
621
622#define BTRFS_DEV_LOOKUP_ARGS(name) \
623 struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT
624
625enum btrfs_map_op {
626 BTRFS_MAP_READ,
627 BTRFS_MAP_WRITE,
628 BTRFS_MAP_GET_READ_MIRRORS,
629};
630
631static inline enum btrfs_map_op btrfs_op(struct bio *bio)
632{
633 switch (bio_op(bio)) {
634 case REQ_OP_WRITE:
635 case REQ_OP_ZONE_APPEND:
636 return BTRFS_MAP_WRITE;
637 default:
638 WARN_ON_ONCE(1);
639 fallthrough;
640 case REQ_OP_READ:
641 return BTRFS_MAP_READ;
642 }
643}
644
645static inline unsigned long btrfs_chunk_item_size(int num_stripes)
646{
647 ASSERT(num_stripes);
648 return sizeof(struct btrfs_chunk) +
649 sizeof(struct btrfs_stripe) * (num_stripes - 1);
650}
651
652/*
653 * Do the type safe conversion from stripe_nr to offset inside the chunk.
654 *
655 * @stripe_nr is u32, with left shift it can overflow u32 for chunks larger
656 * than 4G. This does the proper type cast to avoid overflow.
657 */
658static inline u64 btrfs_stripe_nr_to_offset(u32 stripe_nr)
659{
660 return (u64)stripe_nr << BTRFS_STRIPE_LEN_SHIFT;
661}
662
663void btrfs_get_bioc(struct btrfs_io_context *bioc);
664void btrfs_put_bioc(struct btrfs_io_context *bioc);
665int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
666 u64 logical, u64 *length,
667 struct btrfs_io_context **bioc_ret,
668 struct btrfs_io_stripe *smap, int *mirror_num_ret);
669int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
670 struct btrfs_io_stripe *smap, u64 logical,
671 u32 length, int mirror_num);
672struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
673 u64 logical, u64 *length_ret,
674 u32 *num_stripes);
675int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
676int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
677struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
678 u64 type);
679void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info);
680int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
681 blk_mode_t flags, void *holder);
682struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
683 bool mount_arg_dev);
684int btrfs_forget_devices(dev_t devt);
685void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
686void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
687void btrfs_assign_next_active_device(struct btrfs_device *device,
688 struct btrfs_device *this_dev);
689struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
690 u64 devid,
691 const char *devpath);
692int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
693 struct btrfs_dev_lookup_args *args,
694 const char *path);
695struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
696 const u64 *devid, const u8 *uuid,
697 const char *path);
698void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args);
699int btrfs_rm_device(struct btrfs_fs_info *fs_info,
700 struct btrfs_dev_lookup_args *args,
701 struct file **bdev_file);
702void __exit btrfs_cleanup_fs_uuids(void);
703int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
704int btrfs_grow_device(struct btrfs_trans_handle *trans,
705 struct btrfs_device *device, u64 new_size);
706struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
707 const struct btrfs_dev_lookup_args *args);
708int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
709int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
710int btrfs_balance(struct btrfs_fs_info *fs_info,
711 struct btrfs_balance_control *bctl,
712 struct btrfs_ioctl_balance_args *bargs);
713void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
714int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
715int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
716int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
717int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
718int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
719int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
720int btrfs_uuid_scan_kthread(void *data);
721bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset);
722void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
723int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
724 struct btrfs_ioctl_get_dev_stats *stats);
725int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
726int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
727int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
728void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
729void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
730void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
731int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
732 u64 logical, u64 len);
733unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
734 u64 logical);
735u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map);
736int btrfs_nr_parity_stripes(u64 type);
737int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
738 struct btrfs_block_group *bg);
739int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
740
741#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
742struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp);
743int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
744#endif
745
746struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp);
747struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
748 u64 logical, u64 length);
749struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
750 u64 logical, u64 length);
751struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
752 u64 logical, u64 length);
753void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map);
754void btrfs_release_disk_super(struct btrfs_super_block *super);
755
756static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
757 int index)
758{
759 atomic_inc(dev->dev_stat_values + index);
760 /*
761 * This memory barrier orders stores updating statistics before stores
762 * updating dev_stats_ccnt.
763 *
764 * It pairs with smp_rmb() in btrfs_run_dev_stats().
765 */
766 smp_mb__before_atomic();
767 atomic_inc(&dev->dev_stats_ccnt);
768}
769
770static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
771 int index)
772{
773 return atomic_read(dev->dev_stat_values + index);
774}
775
776static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
777 int index)
778{
779 int ret;
780
781 ret = atomic_xchg(dev->dev_stat_values + index, 0);
782 /*
783 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
784 * - RMW operations that have a return value are fully ordered;
785 *
786 * This implicit memory barriers is paired with the smp_rmb in
787 * btrfs_run_dev_stats
788 */
789 atomic_inc(&dev->dev_stats_ccnt);
790 return ret;
791}
792
793static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
794 int index, unsigned long val)
795{
796 atomic_set(dev->dev_stat_values + index, val);
797 /*
798 * This memory barrier orders stores updating statistics before stores
799 * updating dev_stats_ccnt.
800 *
801 * It pairs with smp_rmb() in btrfs_run_dev_stats().
802 */
803 smp_mb__before_atomic();
804 atomic_inc(&dev->dev_stats_ccnt);
805}
806
807static inline const char *btrfs_dev_name(const struct btrfs_device *device)
808{
809 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
810 return "<missing disk>";
811 else
812 return rcu_str_deref(device->name);
813}
814
815void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
816
817struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
818bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
819 struct btrfs_device *failing_dev);
820void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device);
821
822enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags);
823int btrfs_bg_type_to_factor(u64 flags);
824const char *btrfs_bg_type_to_raid_name(u64 flags);
825int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
826bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
827
828bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr);
829u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb);
830
831#endif