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/bio.h>
10#include <linux/sort.h>
11#include <linux/btrfs.h>
12#include "async-thread.h"
13
14extern struct mutex uuid_mutex;
15
16#define BTRFS_STRIPE_LEN SZ_64K
17
18struct buffer_head;
19struct btrfs_pending_bios {
20 struct bio *head;
21 struct bio *tail;
22};
23
24/*
25 * Use sequence counter to get consistent device stat data on
26 * 32-bit processors.
27 */
28#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
29#include <linux/seqlock.h>
30#define __BTRFS_NEED_DEVICE_DATA_ORDERED
31#define btrfs_device_data_ordered_init(device) \
32 seqcount_init(&device->data_seqcount)
33#else
34#define btrfs_device_data_ordered_init(device) do { } while (0)
35#endif
36
37#define BTRFS_DEV_STATE_WRITEABLE (0)
38#define BTRFS_DEV_STATE_IN_FS_METADATA (1)
39#define BTRFS_DEV_STATE_MISSING (2)
40#define BTRFS_DEV_STATE_REPLACE_TGT (3)
41#define BTRFS_DEV_STATE_FLUSH_SENT (4)
42
43struct btrfs_device {
44 struct list_head dev_list;
45 struct list_head dev_alloc_list;
46 struct btrfs_fs_devices *fs_devices;
47 struct btrfs_fs_info *fs_info;
48
49 struct rcu_string *name;
50
51 u64 generation;
52
53 spinlock_t io_lock ____cacheline_aligned;
54 int running_pending;
55 /* regular prio bios */
56 struct btrfs_pending_bios pending_bios;
57 /* sync bios */
58 struct btrfs_pending_bios pending_sync_bios;
59
60 struct block_device *bdev;
61
62 /* the mode sent to blkdev_get */
63 fmode_t mode;
64
65 unsigned long dev_state;
66 blk_status_t last_flush_error;
67 int flush_bio_sent;
68
69#ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
70 seqcount_t data_seqcount;
71#endif
72
73 /* the internal btrfs device id */
74 u64 devid;
75
76 /* size of the device in memory */
77 u64 total_bytes;
78
79 /* size of the device on disk */
80 u64 disk_total_bytes;
81
82 /* bytes used */
83 u64 bytes_used;
84
85 /* optimal io alignment for this device */
86 u32 io_align;
87
88 /* optimal io width for this device */
89 u32 io_width;
90 /* type and info about this device */
91 u64 type;
92
93 /* minimal io size for this device */
94 u32 sector_size;
95
96 /* physical drive uuid (or lvm uuid) */
97 u8 uuid[BTRFS_UUID_SIZE];
98
99 /*
100 * size of the device on the current transaction
101 *
102 * This variant is update when committing the transaction,
103 * and protected by device_list_mutex
104 */
105 u64 commit_total_bytes;
106
107 /* bytes used on the current transaction */
108 u64 commit_bytes_used;
109 /*
110 * used to manage the device which is resized
111 *
112 * It is protected by chunk_lock.
113 */
114 struct list_head resized_list;
115
116 /* for sending down flush barriers */
117 struct bio *flush_bio;
118 struct completion flush_wait;
119
120 /* per-device scrub information */
121 struct scrub_ctx *scrub_ctx;
122
123 struct btrfs_work work;
124 struct rcu_head rcu;
125
126 /* readahead state */
127 atomic_t reada_in_flight;
128 u64 reada_next;
129 struct reada_zone *reada_curr_zone;
130 struct radix_tree_root reada_zones;
131 struct radix_tree_root reada_extents;
132
133 /* disk I/O failure stats. For detailed description refer to
134 * enum btrfs_dev_stat_values in ioctl.h */
135 int dev_stats_valid;
136
137 /* Counter to record the change of device stats */
138 atomic_t dev_stats_ccnt;
139 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
140};
141
142/*
143 * If we read those variants at the context of their own lock, we needn't
144 * use the following helpers, reading them directly is safe.
145 */
146#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
147#define BTRFS_DEVICE_GETSET_FUNCS(name) \
148static inline u64 \
149btrfs_device_get_##name(const struct btrfs_device *dev) \
150{ \
151 u64 size; \
152 unsigned int seq; \
153 \
154 do { \
155 seq = read_seqcount_begin(&dev->data_seqcount); \
156 size = dev->name; \
157 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
158 return size; \
159} \
160 \
161static inline void \
162btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
163{ \
164 preempt_disable(); \
165 write_seqcount_begin(&dev->data_seqcount); \
166 dev->name = size; \
167 write_seqcount_end(&dev->data_seqcount); \
168 preempt_enable(); \
169}
170#elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
171#define BTRFS_DEVICE_GETSET_FUNCS(name) \
172static inline u64 \
173btrfs_device_get_##name(const struct btrfs_device *dev) \
174{ \
175 u64 size; \
176 \
177 preempt_disable(); \
178 size = dev->name; \
179 preempt_enable(); \
180 return size; \
181} \
182 \
183static inline void \
184btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
185{ \
186 preempt_disable(); \
187 dev->name = size; \
188 preempt_enable(); \
189}
190#else
191#define BTRFS_DEVICE_GETSET_FUNCS(name) \
192static inline u64 \
193btrfs_device_get_##name(const struct btrfs_device *dev) \
194{ \
195 return dev->name; \
196} \
197 \
198static inline void \
199btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
200{ \
201 dev->name = size; \
202}
203#endif
204
205BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
206BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
207BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
208
209struct btrfs_fs_devices {
210 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
211
212 u64 num_devices;
213 u64 open_devices;
214 u64 rw_devices;
215 u64 missing_devices;
216 u64 total_rw_bytes;
217 u64 total_devices;
218 struct block_device *latest_bdev;
219
220 /* all of the devices in the FS, protected by a mutex
221 * so we can safely walk it to write out the supers without
222 * worrying about add/remove by the multi-device code.
223 * Scrubbing super can kick off supers writing by holding
224 * this mutex lock.
225 */
226 struct mutex device_list_mutex;
227 struct list_head devices;
228
229 struct list_head resized_devices;
230 /* devices not currently being allocated */
231 struct list_head alloc_list;
232 struct list_head list;
233
234 struct btrfs_fs_devices *seed;
235 int seeding;
236
237 int opened;
238
239 /* set when we find or add a device that doesn't have the
240 * nonrot flag set
241 */
242 int rotating;
243
244 struct btrfs_fs_info *fs_info;
245 /* sysfs kobjects */
246 struct kobject fsid_kobj;
247 struct kobject *device_dir_kobj;
248 struct completion kobj_unregister;
249};
250
251#define BTRFS_BIO_INLINE_CSUM_SIZE 64
252
253/*
254 * we need the mirror number and stripe index to be passed around
255 * the call chain while we are processing end_io (especially errors).
256 * Really, what we need is a btrfs_bio structure that has this info
257 * and is properly sized with its stripe array, but we're not there
258 * quite yet. We have our own btrfs bioset, and all of the bios
259 * we allocate are actually btrfs_io_bios. We'll cram as much of
260 * struct btrfs_bio as we can into this over time.
261 */
262typedef void (btrfs_io_bio_end_io_t) (struct btrfs_io_bio *bio, int err);
263struct btrfs_io_bio {
264 unsigned int mirror_num;
265 unsigned int stripe_index;
266 u64 logical;
267 u8 *csum;
268 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
269 u8 *csum_allocated;
270 btrfs_io_bio_end_io_t *end_io;
271 struct bvec_iter iter;
272 /*
273 * This member must come last, bio_alloc_bioset will allocate enough
274 * bytes for entire btrfs_io_bio but relies on bio being last.
275 */
276 struct bio bio;
277};
278
279static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
280{
281 return container_of(bio, struct btrfs_io_bio, bio);
282}
283
284struct btrfs_bio_stripe {
285 struct btrfs_device *dev;
286 u64 physical;
287 u64 length; /* only used for discard mappings */
288};
289
290struct btrfs_bio;
291typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
292
293struct btrfs_bio {
294 refcount_t refs;
295 atomic_t stripes_pending;
296 struct btrfs_fs_info *fs_info;
297 u64 map_type; /* get from map_lookup->type */
298 bio_end_io_t *end_io;
299 struct bio *orig_bio;
300 unsigned long flags;
301 void *private;
302 atomic_t error;
303 int max_errors;
304 int num_stripes;
305 int mirror_num;
306 int num_tgtdevs;
307 int *tgtdev_map;
308 /*
309 * logical block numbers for the start of each stripe
310 * The last one or two are p/q. These are sorted,
311 * so raid_map[0] is the start of our full stripe
312 */
313 u64 *raid_map;
314 struct btrfs_bio_stripe stripes[];
315};
316
317struct btrfs_device_info {
318 struct btrfs_device *dev;
319 u64 dev_offset;
320 u64 max_avail;
321 u64 total_avail;
322};
323
324struct btrfs_raid_attr {
325 int sub_stripes; /* sub_stripes info for map */
326 int dev_stripes; /* stripes per dev */
327 int devs_max; /* max devs to use */
328 int devs_min; /* min devs needed */
329 int tolerated_failures; /* max tolerated fail devs */
330 int devs_increment; /* ndevs has to be a multiple of this */
331 int ncopies; /* how many copies to data has */
332};
333
334extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
335extern const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES];
336extern const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES];
337
338struct map_lookup {
339 u64 type;
340 int io_align;
341 int io_width;
342 u64 stripe_len;
343 int num_stripes;
344 int sub_stripes;
345 struct btrfs_bio_stripe stripes[];
346};
347
348#define map_lookup_size(n) (sizeof(struct map_lookup) + \
349 (sizeof(struct btrfs_bio_stripe) * (n)))
350
351struct btrfs_balance_args;
352struct btrfs_balance_progress;
353struct btrfs_balance_control {
354 struct btrfs_fs_info *fs_info;
355
356 struct btrfs_balance_args data;
357 struct btrfs_balance_args meta;
358 struct btrfs_balance_args sys;
359
360 u64 flags;
361
362 struct btrfs_balance_progress stat;
363};
364
365enum btrfs_map_op {
366 BTRFS_MAP_READ,
367 BTRFS_MAP_WRITE,
368 BTRFS_MAP_DISCARD,
369 BTRFS_MAP_GET_READ_MIRRORS,
370};
371
372static inline enum btrfs_map_op btrfs_op(struct bio *bio)
373{
374 switch (bio_op(bio)) {
375 case REQ_OP_DISCARD:
376 return BTRFS_MAP_DISCARD;
377 case REQ_OP_WRITE:
378 return BTRFS_MAP_WRITE;
379 default:
380 WARN_ON_ONCE(1);
381 case REQ_OP_READ:
382 return BTRFS_MAP_READ;
383 }
384}
385
386int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
387 u64 end, u64 *length);
388void btrfs_get_bbio(struct btrfs_bio *bbio);
389void btrfs_put_bbio(struct btrfs_bio *bbio);
390int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
391 u64 logical, u64 *length,
392 struct btrfs_bio **bbio_ret, int mirror_num);
393int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
394 u64 logical, u64 *length,
395 struct btrfs_bio **bbio_ret);
396int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
397 u64 chunk_start, u64 physical, u64 devid,
398 u64 **logical, int *naddrs, int *stripe_len);
399int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
400int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
401int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
402 struct btrfs_fs_info *fs_info, u64 type);
403void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
404void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
405blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
406 int mirror_num, int async_submit);
407int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
408 fmode_t flags, void *holder);
409int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
410 struct btrfs_fs_devices **fs_devices_ret);
411int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
412void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step);
413void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
414 struct btrfs_device *device, struct btrfs_device *this_dev);
415int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
416 const char *device_path,
417 struct btrfs_device **device);
418int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
419 const char *devpath,
420 struct btrfs_device **device);
421struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
422 const u64 *devid,
423 const u8 *uuid);
424int btrfs_rm_device(struct btrfs_fs_info *fs_info,
425 const char *device_path, u64 devid);
426void __exit btrfs_cleanup_fs_uuids(void);
427int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
428int btrfs_grow_device(struct btrfs_trans_handle *trans,
429 struct btrfs_device *device, u64 new_size);
430struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
431 u8 *uuid, u8 *fsid);
432int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
433int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
434int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
435 const char *device_path,
436 struct btrfs_device *srcdev,
437 struct btrfs_device **device_out);
438int btrfs_balance(struct btrfs_balance_control *bctl,
439 struct btrfs_ioctl_balance_args *bargs);
440int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
441int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
442int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
443int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
444int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
445int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
446int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
447int find_free_dev_extent_start(struct btrfs_transaction *transaction,
448 struct btrfs_device *device, u64 num_bytes,
449 u64 search_start, u64 *start, u64 *max_avail);
450int find_free_dev_extent(struct btrfs_trans_handle *trans,
451 struct btrfs_device *device, u64 num_bytes,
452 u64 *start, u64 *max_avail);
453void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
454int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
455 struct btrfs_ioctl_get_dev_stats *stats);
456void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
457int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
458int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
459 struct btrfs_fs_info *fs_info);
460void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
461 struct btrfs_device *srcdev);
462void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
463 struct btrfs_device *srcdev);
464void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
465 struct btrfs_device *tgtdev);
466void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
467int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
468 u64 logical, u64 len);
469unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
470 u64 logical);
471int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
472 struct btrfs_fs_info *fs_info,
473 u64 chunk_offset, u64 chunk_size);
474int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
475 struct btrfs_fs_info *fs_info, u64 chunk_offset);
476
477static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
478 int index)
479{
480 atomic_inc(dev->dev_stat_values + index);
481 /*
482 * This memory barrier orders stores updating statistics before stores
483 * updating dev_stats_ccnt.
484 *
485 * It pairs with smp_rmb() in btrfs_run_dev_stats().
486 */
487 smp_mb__before_atomic();
488 atomic_inc(&dev->dev_stats_ccnt);
489}
490
491static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
492 int index)
493{
494 return atomic_read(dev->dev_stat_values + index);
495}
496
497static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
498 int index)
499{
500 int ret;
501
502 ret = atomic_xchg(dev->dev_stat_values + index, 0);
503 /*
504 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
505 * - RMW operations that have a return value are fully ordered;
506 *
507 * This implicit memory barriers is paired with the smp_rmb in
508 * btrfs_run_dev_stats
509 */
510 atomic_inc(&dev->dev_stats_ccnt);
511 return ret;
512}
513
514static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
515 int index, unsigned long val)
516{
517 atomic_set(dev->dev_stat_values + index, val);
518 /*
519 * This memory barrier orders stores updating statistics before stores
520 * updating dev_stats_ccnt.
521 *
522 * It pairs with smp_rmb() in btrfs_run_dev_stats().
523 */
524 smp_mb__before_atomic();
525 atomic_inc(&dev->dev_stats_ccnt);
526}
527
528static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
529 int index)
530{
531 btrfs_dev_stat_set(dev, index, 0);
532}
533
534/*
535 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
536 * can be used as index to access btrfs_raid_array[].
537 */
538static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
539{
540 if (flags & BTRFS_BLOCK_GROUP_RAID10)
541 return BTRFS_RAID_RAID10;
542 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
543 return BTRFS_RAID_RAID1;
544 else if (flags & BTRFS_BLOCK_GROUP_DUP)
545 return BTRFS_RAID_DUP;
546 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
547 return BTRFS_RAID_RAID0;
548 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
549 return BTRFS_RAID_RAID5;
550 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
551 return BTRFS_RAID_RAID6;
552
553 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
554}
555
556void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info);
557void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans);
558
559struct list_head *btrfs_get_fs_uuids(void);
560void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
561void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
562bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
563 struct btrfs_device *failing_dev);
564
565#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/bio.h>
10#include <linux/sort.h>
11#include <linux/btrfs.h>
12#include "async-thread.h"
13
14#define BTRFS_MAX_DATA_CHUNK_SIZE (10ULL * SZ_1G)
15
16extern struct mutex uuid_mutex;
17
18#define BTRFS_STRIPE_LEN SZ_64K
19
20struct btrfs_io_geometry {
21 /* remaining bytes before crossing a stripe */
22 u64 len;
23 /* offset of logical address in chunk */
24 u64 offset;
25 /* length of single IO stripe */
26 u64 stripe_len;
27 /* number of stripe where address falls */
28 u64 stripe_nr;
29 /* offset of address in stripe */
30 u64 stripe_offset;
31 /* offset of raid56 stripe into the chunk */
32 u64 raid56_stripe_offset;
33};
34
35/*
36 * Use sequence counter to get consistent device stat data on
37 * 32-bit processors.
38 */
39#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
40#include <linux/seqlock.h>
41#define __BTRFS_NEED_DEVICE_DATA_ORDERED
42#define btrfs_device_data_ordered_init(device) \
43 seqcount_init(&device->data_seqcount)
44#else
45#define btrfs_device_data_ordered_init(device) do { } while (0)
46#endif
47
48#define BTRFS_DEV_STATE_WRITEABLE (0)
49#define BTRFS_DEV_STATE_IN_FS_METADATA (1)
50#define BTRFS_DEV_STATE_MISSING (2)
51#define BTRFS_DEV_STATE_REPLACE_TGT (3)
52#define BTRFS_DEV_STATE_FLUSH_SENT (4)
53#define BTRFS_DEV_STATE_NO_READA (5)
54
55struct btrfs_zoned_device_info;
56
57struct btrfs_device {
58 struct list_head dev_list; /* device_list_mutex */
59 struct list_head dev_alloc_list; /* chunk mutex */
60 struct list_head post_commit_list; /* chunk mutex */
61 struct btrfs_fs_devices *fs_devices;
62 struct btrfs_fs_info *fs_info;
63
64 struct rcu_string __rcu *name;
65
66 u64 generation;
67
68 struct block_device *bdev;
69
70 struct btrfs_zoned_device_info *zone_info;
71
72 /* the mode sent to blkdev_get */
73 fmode_t mode;
74
75 unsigned long dev_state;
76 blk_status_t last_flush_error;
77
78#ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
79 seqcount_t data_seqcount;
80#endif
81
82 /* the internal btrfs device id */
83 u64 devid;
84
85 /* size of the device in memory */
86 u64 total_bytes;
87
88 /* size of the device on disk */
89 u64 disk_total_bytes;
90
91 /* bytes used */
92 u64 bytes_used;
93
94 /* optimal io alignment for this device */
95 u32 io_align;
96
97 /* optimal io width for this device */
98 u32 io_width;
99 /* type and info about this device */
100 u64 type;
101
102 /* minimal io size for this device */
103 u32 sector_size;
104
105 /* physical drive uuid (or lvm uuid) */
106 u8 uuid[BTRFS_UUID_SIZE];
107
108 /*
109 * size of the device on the current transaction
110 *
111 * This variant is update when committing the transaction,
112 * and protected by chunk mutex
113 */
114 u64 commit_total_bytes;
115
116 /* bytes used on the current transaction */
117 u64 commit_bytes_used;
118
119 /* for sending down flush barriers */
120 struct bio *flush_bio;
121 struct completion flush_wait;
122
123 /* per-device scrub information */
124 struct scrub_ctx *scrub_ctx;
125
126 /* readahead state */
127 atomic_t reada_in_flight;
128 u64 reada_next;
129 struct reada_zone *reada_curr_zone;
130 struct radix_tree_root reada_zones;
131 struct radix_tree_root reada_extents;
132
133 /* disk I/O failure stats. For detailed description refer to
134 * enum btrfs_dev_stat_values in ioctl.h */
135 int dev_stats_valid;
136
137 /* Counter to record the change of device stats */
138 atomic_t dev_stats_ccnt;
139 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
140
141 struct extent_io_tree alloc_state;
142
143 struct completion kobj_unregister;
144 /* For sysfs/FSID/devinfo/devid/ */
145 struct kobject devid_kobj;
146
147 /* Bandwidth limit for scrub, in bytes */
148 u64 scrub_speed_max;
149};
150
151/*
152 * If we read those variants at the context of their own lock, we needn't
153 * use the following helpers, reading them directly is safe.
154 */
155#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
156#define BTRFS_DEVICE_GETSET_FUNCS(name) \
157static inline u64 \
158btrfs_device_get_##name(const struct btrfs_device *dev) \
159{ \
160 u64 size; \
161 unsigned int seq; \
162 \
163 do { \
164 seq = read_seqcount_begin(&dev->data_seqcount); \
165 size = dev->name; \
166 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \
167 return size; \
168} \
169 \
170static inline void \
171btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
172{ \
173 preempt_disable(); \
174 write_seqcount_begin(&dev->data_seqcount); \
175 dev->name = size; \
176 write_seqcount_end(&dev->data_seqcount); \
177 preempt_enable(); \
178}
179#elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
180#define BTRFS_DEVICE_GETSET_FUNCS(name) \
181static inline u64 \
182btrfs_device_get_##name(const struct btrfs_device *dev) \
183{ \
184 u64 size; \
185 \
186 preempt_disable(); \
187 size = dev->name; \
188 preempt_enable(); \
189 return size; \
190} \
191 \
192static inline void \
193btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
194{ \
195 preempt_disable(); \
196 dev->name = size; \
197 preempt_enable(); \
198}
199#else
200#define BTRFS_DEVICE_GETSET_FUNCS(name) \
201static inline u64 \
202btrfs_device_get_##name(const struct btrfs_device *dev) \
203{ \
204 return dev->name; \
205} \
206 \
207static inline void \
208btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \
209{ \
210 dev->name = size; \
211}
212#endif
213
214BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
215BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
216BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
217
218enum btrfs_chunk_allocation_policy {
219 BTRFS_CHUNK_ALLOC_REGULAR,
220 BTRFS_CHUNK_ALLOC_ZONED,
221};
222
223/*
224 * Read policies for mirrored block group profiles, read picks the stripe based
225 * on these policies.
226 */
227enum btrfs_read_policy {
228 /* Use process PID to choose the stripe */
229 BTRFS_READ_POLICY_PID,
230 BTRFS_NR_READ_POLICY,
231};
232
233struct btrfs_fs_devices {
234 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
235 u8 metadata_uuid[BTRFS_FSID_SIZE];
236 bool fsid_change;
237 struct list_head fs_list;
238
239 u64 num_devices;
240 u64 open_devices;
241 u64 rw_devices;
242 u64 missing_devices;
243 u64 total_rw_bytes;
244 u64 total_devices;
245
246 /* Highest generation number of seen devices */
247 u64 latest_generation;
248
249 struct block_device *latest_bdev;
250
251 /* all of the devices in the FS, protected by a mutex
252 * so we can safely walk it to write out the supers without
253 * worrying about add/remove by the multi-device code.
254 * Scrubbing super can kick off supers writing by holding
255 * this mutex lock.
256 */
257 struct mutex device_list_mutex;
258
259 /* List of all devices, protected by device_list_mutex */
260 struct list_head devices;
261
262 /*
263 * Devices which can satisfy space allocation. Protected by
264 * chunk_mutex
265 */
266 struct list_head alloc_list;
267
268 struct list_head seed_list;
269 bool seeding;
270
271 int opened;
272
273 /* set when we find or add a device that doesn't have the
274 * nonrot flag set
275 */
276 bool rotating;
277
278 struct btrfs_fs_info *fs_info;
279 /* sysfs kobjects */
280 struct kobject fsid_kobj;
281 struct kobject *devices_kobj;
282 struct kobject *devinfo_kobj;
283 struct completion kobj_unregister;
284
285 enum btrfs_chunk_allocation_policy chunk_alloc_policy;
286
287 /* Policy used to read the mirrored stripes */
288 enum btrfs_read_policy read_policy;
289};
290
291#define BTRFS_BIO_INLINE_CSUM_SIZE 64
292
293#define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
294 - sizeof(struct btrfs_chunk)) \
295 / sizeof(struct btrfs_stripe) + 1)
296
297#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
298 - 2 * sizeof(struct btrfs_disk_key) \
299 - 2 * sizeof(struct btrfs_chunk)) \
300 / sizeof(struct btrfs_stripe) + 1)
301
302/*
303 * we need the mirror number and stripe index to be passed around
304 * the call chain while we are processing end_io (especially errors).
305 * Really, what we need is a btrfs_bio structure that has this info
306 * and is properly sized with its stripe array, but we're not there
307 * quite yet. We have our own btrfs bioset, and all of the bios
308 * we allocate are actually btrfs_io_bios. We'll cram as much of
309 * struct btrfs_bio as we can into this over time.
310 */
311struct btrfs_io_bio {
312 unsigned int mirror_num;
313 struct btrfs_device *device;
314 u64 logical;
315 u8 *csum;
316 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
317 struct bvec_iter iter;
318 /*
319 * This member must come last, bio_alloc_bioset will allocate enough
320 * bytes for entire btrfs_io_bio but relies on bio being last.
321 */
322 struct bio bio;
323};
324
325static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
326{
327 return container_of(bio, struct btrfs_io_bio, bio);
328}
329
330static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio)
331{
332 if (io_bio->csum != io_bio->csum_inline) {
333 kfree(io_bio->csum);
334 io_bio->csum = NULL;
335 }
336}
337
338struct btrfs_bio_stripe {
339 struct btrfs_device *dev;
340 u64 physical;
341 u64 length; /* only used for discard mappings */
342};
343
344struct btrfs_bio {
345 refcount_t refs;
346 atomic_t stripes_pending;
347 struct btrfs_fs_info *fs_info;
348 u64 map_type; /* get from map_lookup->type */
349 bio_end_io_t *end_io;
350 struct bio *orig_bio;
351 void *private;
352 atomic_t error;
353 int max_errors;
354 int num_stripes;
355 int mirror_num;
356 int num_tgtdevs;
357 int *tgtdev_map;
358 /*
359 * logical block numbers for the start of each stripe
360 * The last one or two are p/q. These are sorted,
361 * so raid_map[0] is the start of our full stripe
362 */
363 u64 *raid_map;
364 struct btrfs_bio_stripe stripes[];
365};
366
367struct btrfs_device_info {
368 struct btrfs_device *dev;
369 u64 dev_offset;
370 u64 max_avail;
371 u64 total_avail;
372};
373
374struct btrfs_raid_attr {
375 u8 sub_stripes; /* sub_stripes info for map */
376 u8 dev_stripes; /* stripes per dev */
377 u8 devs_max; /* max devs to use */
378 u8 devs_min; /* min devs needed */
379 u8 tolerated_failures; /* max tolerated fail devs */
380 u8 devs_increment; /* ndevs has to be a multiple of this */
381 u8 ncopies; /* how many copies to data has */
382 u8 nparity; /* number of stripes worth of bytes to store
383 * parity information */
384 u8 mindev_error; /* error code if min devs requisite is unmet */
385 const char raid_name[8]; /* name of the raid */
386 u64 bg_flag; /* block group flag of the raid */
387};
388
389extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
390
391struct map_lookup {
392 u64 type;
393 int io_align;
394 int io_width;
395 u64 stripe_len;
396 int num_stripes;
397 int sub_stripes;
398 int verified_stripes; /* For mount time dev extent verification */
399 struct btrfs_bio_stripe stripes[];
400};
401
402#define map_lookup_size(n) (sizeof(struct map_lookup) + \
403 (sizeof(struct btrfs_bio_stripe) * (n)))
404
405struct btrfs_balance_args;
406struct btrfs_balance_progress;
407struct btrfs_balance_control {
408 struct btrfs_balance_args data;
409 struct btrfs_balance_args meta;
410 struct btrfs_balance_args sys;
411
412 u64 flags;
413
414 struct btrfs_balance_progress stat;
415};
416
417enum btrfs_map_op {
418 BTRFS_MAP_READ,
419 BTRFS_MAP_WRITE,
420 BTRFS_MAP_DISCARD,
421 BTRFS_MAP_GET_READ_MIRRORS,
422};
423
424static inline enum btrfs_map_op btrfs_op(struct bio *bio)
425{
426 switch (bio_op(bio)) {
427 case REQ_OP_DISCARD:
428 return BTRFS_MAP_DISCARD;
429 case REQ_OP_WRITE:
430 case REQ_OP_ZONE_APPEND:
431 return BTRFS_MAP_WRITE;
432 default:
433 WARN_ON_ONCE(1);
434 fallthrough;
435 case REQ_OP_READ:
436 return BTRFS_MAP_READ;
437 }
438}
439
440void btrfs_get_bbio(struct btrfs_bio *bbio);
441void btrfs_put_bbio(struct btrfs_bio *bbio);
442int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
443 u64 logical, u64 *length,
444 struct btrfs_bio **bbio_ret, int mirror_num);
445int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
446 u64 logical, u64 *length,
447 struct btrfs_bio **bbio_ret);
448int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *map,
449 enum btrfs_map_op op, u64 logical,
450 struct btrfs_io_geometry *io_geom);
451int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
452int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
453struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
454 u64 type);
455void btrfs_mapping_tree_free(struct extent_map_tree *tree);
456blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
457 int mirror_num);
458int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
459 fmode_t flags, void *holder);
460struct btrfs_device *btrfs_scan_one_device(const char *path,
461 fmode_t flags, void *holder);
462int btrfs_forget_devices(const char *path);
463void btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
464void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices);
465void btrfs_assign_next_active_device(struct btrfs_device *device,
466 struct btrfs_device *this_dev);
467struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info,
468 u64 devid,
469 const char *devpath);
470struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
471 const u64 *devid,
472 const u8 *uuid);
473void btrfs_free_device(struct btrfs_device *device);
474int btrfs_rm_device(struct btrfs_fs_info *fs_info,
475 const char *device_path, u64 devid,
476 struct block_device **bdev, fmode_t *mode);
477void __exit btrfs_cleanup_fs_uuids(void);
478int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
479int btrfs_grow_device(struct btrfs_trans_handle *trans,
480 struct btrfs_device *device, u64 new_size);
481struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
482 u64 devid, u8 *uuid, u8 *fsid);
483int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
484int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
485int btrfs_balance(struct btrfs_fs_info *fs_info,
486 struct btrfs_balance_control *bctl,
487 struct btrfs_ioctl_balance_args *bargs);
488void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
489int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
490int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
491int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
492int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
493int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
494int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
495int btrfs_uuid_scan_kthread(void *data);
496int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
497int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
498 u64 *start, u64 *max_avail);
499void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
500int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
501 struct btrfs_ioctl_get_dev_stats *stats);
502void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
503int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
504int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
505void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev);
506void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev);
507void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev);
508int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
509 u64 logical, u64 len);
510unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
511 u64 logical);
512int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
513 u64 chunk_offset, u64 chunk_size);
514int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
515 struct btrfs_block_group *bg);
516int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset);
517struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
518 u64 logical, u64 length);
519void btrfs_release_disk_super(struct btrfs_super_block *super);
520
521static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
522 int index)
523{
524 atomic_inc(dev->dev_stat_values + index);
525 /*
526 * This memory barrier orders stores updating statistics before stores
527 * updating dev_stats_ccnt.
528 *
529 * It pairs with smp_rmb() in btrfs_run_dev_stats().
530 */
531 smp_mb__before_atomic();
532 atomic_inc(&dev->dev_stats_ccnt);
533}
534
535static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
536 int index)
537{
538 return atomic_read(dev->dev_stat_values + index);
539}
540
541static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
542 int index)
543{
544 int ret;
545
546 ret = atomic_xchg(dev->dev_stat_values + index, 0);
547 /*
548 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
549 * - RMW operations that have a return value are fully ordered;
550 *
551 * This implicit memory barriers is paired with the smp_rmb in
552 * btrfs_run_dev_stats
553 */
554 atomic_inc(&dev->dev_stats_ccnt);
555 return ret;
556}
557
558static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
559 int index, unsigned long val)
560{
561 atomic_set(dev->dev_stat_values + index, val);
562 /*
563 * This memory barrier orders stores updating statistics before stores
564 * updating dev_stats_ccnt.
565 *
566 * It pairs with smp_rmb() in btrfs_run_dev_stats().
567 */
568 smp_mb__before_atomic();
569 atomic_inc(&dev->dev_stats_ccnt);
570}
571
572/*
573 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
574 * can be used as index to access btrfs_raid_array[].
575 */
576static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags)
577{
578 if (flags & BTRFS_BLOCK_GROUP_RAID10)
579 return BTRFS_RAID_RAID10;
580 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
581 return BTRFS_RAID_RAID1;
582 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
583 return BTRFS_RAID_RAID1C3;
584 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
585 return BTRFS_RAID_RAID1C4;
586 else if (flags & BTRFS_BLOCK_GROUP_DUP)
587 return BTRFS_RAID_DUP;
588 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
589 return BTRFS_RAID_RAID0;
590 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
591 return BTRFS_RAID_RAID5;
592 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
593 return BTRFS_RAID_RAID6;
594
595 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
596}
597
598void btrfs_commit_device_sizes(struct btrfs_transaction *trans);
599
600struct list_head * __attribute_const__ btrfs_get_fs_uuids(void);
601bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
602 struct btrfs_device *failing_dev);
603void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
604 struct block_device *bdev,
605 const char *device_path);
606
607int btrfs_bg_type_to_factor(u64 flags);
608const char *btrfs_bg_type_to_raid_name(u64 flags);
609int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
610int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical);
611
612#endif