xref: /linux/fs/btrfs/volumes.h (revision 9a379e77033f02c4a071891afdf0f0a01eff8ccb)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #ifndef __BTRFS_VOLUMES_
20 #define __BTRFS_VOLUMES_
21 
22 #include <linux/bio.h>
23 #include <linux/sort.h>
24 #include <linux/btrfs.h>
25 #include "async-thread.h"
26 
27 extern struct mutex uuid_mutex;
28 
29 #define BTRFS_STRIPE_LEN	SZ_64K
30 
31 struct buffer_head;
32 struct btrfs_pending_bios {
33 	struct bio *head;
34 	struct bio *tail;
35 };
36 
37 /*
38  * Use sequence counter to get consistent device stat data on
39  * 32-bit processors.
40  */
41 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
42 #include <linux/seqlock.h>
43 #define __BTRFS_NEED_DEVICE_DATA_ORDERED
44 #define btrfs_device_data_ordered_init(device)	\
45 	seqcount_init(&device->data_seqcount)
46 #else
47 #define btrfs_device_data_ordered_init(device) do { } while (0)
48 #endif
49 
50 #define BTRFS_DEV_STATE_WRITEABLE	(0)
51 #define BTRFS_DEV_STATE_IN_FS_METADATA	(1)
52 #define BTRFS_DEV_STATE_MISSING		(2)
53 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
54 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
55 
56 struct btrfs_device {
57 	struct list_head dev_list;
58 	struct list_head dev_alloc_list;
59 	struct btrfs_fs_devices *fs_devices;
60 	struct btrfs_fs_info *fs_info;
61 
62 	struct rcu_string *name;
63 
64 	u64 generation;
65 
66 	spinlock_t io_lock ____cacheline_aligned;
67 	int running_pending;
68 	/* regular prio bios */
69 	struct btrfs_pending_bios pending_bios;
70 	/* sync bios */
71 	struct btrfs_pending_bios pending_sync_bios;
72 
73 	struct block_device *bdev;
74 
75 	/* the mode sent to blkdev_get */
76 	fmode_t mode;
77 
78 	unsigned long dev_state;
79 	blk_status_t last_flush_error;
80 	int flush_bio_sent;
81 
82 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
83 	seqcount_t data_seqcount;
84 #endif
85 
86 	/* the internal btrfs device id */
87 	u64 devid;
88 
89 	/* size of the device in memory */
90 	u64 total_bytes;
91 
92 	/* size of the device on disk */
93 	u64 disk_total_bytes;
94 
95 	/* bytes used */
96 	u64 bytes_used;
97 
98 	/* optimal io alignment for this device */
99 	u32 io_align;
100 
101 	/* optimal io width for this device */
102 	u32 io_width;
103 	/* type and info about this device */
104 	u64 type;
105 
106 	/* minimal io size for this device */
107 	u32 sector_size;
108 
109 	/* physical drive uuid (or lvm uuid) */
110 	u8 uuid[BTRFS_UUID_SIZE];
111 
112 	/*
113 	 * size of the device on the current transaction
114 	 *
115 	 * This variant is update when committing the transaction,
116 	 * and protected by device_list_mutex
117 	 */
118 	u64 commit_total_bytes;
119 
120 	/* bytes used on the current transaction */
121 	u64 commit_bytes_used;
122 	/*
123 	 * used to manage the device which is resized
124 	 *
125 	 * It is protected by chunk_lock.
126 	 */
127 	struct list_head resized_list;
128 
129 	/* for sending down flush barriers */
130 	struct bio *flush_bio;
131 	struct completion flush_wait;
132 
133 	/* per-device scrub information */
134 	struct scrub_ctx *scrub_ctx;
135 
136 	struct btrfs_work work;
137 	struct rcu_head rcu;
138 
139 	/* readahead state */
140 	atomic_t reada_in_flight;
141 	u64 reada_next;
142 	struct reada_zone *reada_curr_zone;
143 	struct radix_tree_root reada_zones;
144 	struct radix_tree_root reada_extents;
145 
146 	/* disk I/O failure stats. For detailed description refer to
147 	 * enum btrfs_dev_stat_values in ioctl.h */
148 	int dev_stats_valid;
149 
150 	/* Counter to record the change of device stats */
151 	atomic_t dev_stats_ccnt;
152 	atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
153 };
154 
155 /*
156  * If we read those variants at the context of their own lock, we needn't
157  * use the following helpers, reading them directly is safe.
158  */
159 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
160 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
161 static inline u64							\
162 btrfs_device_get_##name(const struct btrfs_device *dev)			\
163 {									\
164 	u64 size;							\
165 	unsigned int seq;						\
166 									\
167 	do {								\
168 		seq = read_seqcount_begin(&dev->data_seqcount);		\
169 		size = dev->name;					\
170 	} while (read_seqcount_retry(&dev->data_seqcount, seq));	\
171 	return size;							\
172 }									\
173 									\
174 static inline void							\
175 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
176 {									\
177 	preempt_disable();						\
178 	write_seqcount_begin(&dev->data_seqcount);			\
179 	dev->name = size;						\
180 	write_seqcount_end(&dev->data_seqcount);			\
181 	preempt_enable();						\
182 }
183 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPT)
184 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
185 static inline u64							\
186 btrfs_device_get_##name(const struct btrfs_device *dev)			\
187 {									\
188 	u64 size;							\
189 									\
190 	preempt_disable();						\
191 	size = dev->name;						\
192 	preempt_enable();						\
193 	return size;							\
194 }									\
195 									\
196 static inline void							\
197 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
198 {									\
199 	preempt_disable();						\
200 	dev->name = size;						\
201 	preempt_enable();						\
202 }
203 #else
204 #define BTRFS_DEVICE_GETSET_FUNCS(name)					\
205 static inline u64							\
206 btrfs_device_get_##name(const struct btrfs_device *dev)			\
207 {									\
208 	return dev->name;						\
209 }									\
210 									\
211 static inline void							\
212 btrfs_device_set_##name(struct btrfs_device *dev, u64 size)		\
213 {									\
214 	dev->name = size;						\
215 }
216 #endif
217 
218 BTRFS_DEVICE_GETSET_FUNCS(total_bytes);
219 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes);
220 BTRFS_DEVICE_GETSET_FUNCS(bytes_used);
221 
222 struct btrfs_fs_devices {
223 	u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
224 
225 	u64 num_devices;
226 	u64 open_devices;
227 	u64 rw_devices;
228 	u64 missing_devices;
229 	u64 total_rw_bytes;
230 	u64 total_devices;
231 	struct block_device *latest_bdev;
232 
233 	/* all of the devices in the FS, protected by a mutex
234 	 * so we can safely walk it to write out the supers without
235 	 * worrying about add/remove by the multi-device code.
236 	 * Scrubbing super can kick off supers writing by holding
237 	 * this mutex lock.
238 	 */
239 	struct mutex device_list_mutex;
240 	struct list_head devices;
241 
242 	struct list_head resized_devices;
243 	/* devices not currently being allocated */
244 	struct list_head alloc_list;
245 	struct list_head list;
246 
247 	struct btrfs_fs_devices *seed;
248 	int seeding;
249 
250 	int opened;
251 
252 	/* set when we find or add a device that doesn't have the
253 	 * nonrot flag set
254 	 */
255 	int rotating;
256 
257 	struct btrfs_fs_info *fs_info;
258 	/* sysfs kobjects */
259 	struct kobject fsid_kobj;
260 	struct kobject *device_dir_kobj;
261 	struct completion kobj_unregister;
262 };
263 
264 #define BTRFS_BIO_INLINE_CSUM_SIZE	64
265 
266 /*
267  * we need the mirror number and stripe index to be passed around
268  * the call chain while we are processing end_io (especially errors).
269  * Really, what we need is a btrfs_bio structure that has this info
270  * and is properly sized with its stripe array, but we're not there
271  * quite yet.  We have our own btrfs bioset, and all of the bios
272  * we allocate are actually btrfs_io_bios.  We'll cram as much of
273  * struct btrfs_bio as we can into this over time.
274  */
275 typedef void (btrfs_io_bio_end_io_t) (struct btrfs_io_bio *bio, int err);
276 struct btrfs_io_bio {
277 	unsigned int mirror_num;
278 	unsigned int stripe_index;
279 	u64 logical;
280 	u8 *csum;
281 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
282 	u8 *csum_allocated;
283 	btrfs_io_bio_end_io_t *end_io;
284 	struct bvec_iter iter;
285 	/*
286 	 * This member must come last, bio_alloc_bioset will allocate enough
287 	 * bytes for entire btrfs_io_bio but relies on bio being last.
288 	 */
289 	struct bio bio;
290 };
291 
292 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
293 {
294 	return container_of(bio, struct btrfs_io_bio, bio);
295 }
296 
297 struct btrfs_bio_stripe {
298 	struct btrfs_device *dev;
299 	u64 physical;
300 	u64 length; /* only used for discard mappings */
301 };
302 
303 struct btrfs_bio;
304 typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
305 
306 struct btrfs_bio {
307 	refcount_t refs;
308 	atomic_t stripes_pending;
309 	struct btrfs_fs_info *fs_info;
310 	u64 map_type; /* get from map_lookup->type */
311 	bio_end_io_t *end_io;
312 	struct bio *orig_bio;
313 	unsigned long flags;
314 	void *private;
315 	atomic_t error;
316 	int max_errors;
317 	int num_stripes;
318 	int mirror_num;
319 	int num_tgtdevs;
320 	int *tgtdev_map;
321 	/*
322 	 * logical block numbers for the start of each stripe
323 	 * The last one or two are p/q.  These are sorted,
324 	 * so raid_map[0] is the start of our full stripe
325 	 */
326 	u64 *raid_map;
327 	struct btrfs_bio_stripe stripes[];
328 };
329 
330 struct btrfs_device_info {
331 	struct btrfs_device *dev;
332 	u64 dev_offset;
333 	u64 max_avail;
334 	u64 total_avail;
335 };
336 
337 struct btrfs_raid_attr {
338 	int sub_stripes;	/* sub_stripes info for map */
339 	int dev_stripes;	/* stripes per dev */
340 	int devs_max;		/* max devs to use */
341 	int devs_min;		/* min devs needed */
342 	int tolerated_failures; /* max tolerated fail devs */
343 	int devs_increment;	/* ndevs has to be a multiple of this */
344 	int ncopies;		/* how many copies to data has */
345 };
346 
347 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES];
348 extern const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES];
349 extern const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES];
350 
351 struct map_lookup {
352 	u64 type;
353 	int io_align;
354 	int io_width;
355 	u64 stripe_len;
356 	int num_stripes;
357 	int sub_stripes;
358 	struct btrfs_bio_stripe stripes[];
359 };
360 
361 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
362 			    (sizeof(struct btrfs_bio_stripe) * (n)))
363 
364 struct btrfs_balance_args;
365 struct btrfs_balance_progress;
366 struct btrfs_balance_control {
367 	struct btrfs_fs_info *fs_info;
368 
369 	struct btrfs_balance_args data;
370 	struct btrfs_balance_args meta;
371 	struct btrfs_balance_args sys;
372 
373 	u64 flags;
374 
375 	struct btrfs_balance_progress stat;
376 };
377 
378 enum btrfs_map_op {
379 	BTRFS_MAP_READ,
380 	BTRFS_MAP_WRITE,
381 	BTRFS_MAP_DISCARD,
382 	BTRFS_MAP_GET_READ_MIRRORS,
383 };
384 
385 static inline enum btrfs_map_op btrfs_op(struct bio *bio)
386 {
387 	switch (bio_op(bio)) {
388 	case REQ_OP_DISCARD:
389 		return BTRFS_MAP_DISCARD;
390 	case REQ_OP_WRITE:
391 		return BTRFS_MAP_WRITE;
392 	default:
393 		WARN_ON_ONCE(1);
394 	case REQ_OP_READ:
395 		return BTRFS_MAP_READ;
396 	}
397 }
398 
399 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
400 				   u64 end, u64 *length);
401 void btrfs_get_bbio(struct btrfs_bio *bbio);
402 void btrfs_put_bbio(struct btrfs_bio *bbio);
403 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
404 		    u64 logical, u64 *length,
405 		    struct btrfs_bio **bbio_ret, int mirror_num);
406 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
407 		     u64 logical, u64 *length,
408 		     struct btrfs_bio **bbio_ret);
409 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
410 		     u64 chunk_start, u64 physical, u64 devid,
411 		     u64 **logical, int *naddrs, int *stripe_len);
412 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info);
413 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info);
414 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
415 		      struct btrfs_fs_info *fs_info, u64 type);
416 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
417 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
418 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
419 			   int mirror_num, int async_submit);
420 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
421 		       fmode_t flags, void *holder);
422 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
423 			  struct btrfs_fs_devices **fs_devices_ret);
424 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
425 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step);
426 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
427 		struct btrfs_device *device, struct btrfs_device *this_dev);
428 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
429 					 const char *device_path,
430 					 struct btrfs_device **device);
431 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
432 					 const char *devpath,
433 					 struct btrfs_device **device);
434 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
435 					const u64 *devid,
436 					const u8 *uuid);
437 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
438 		    const char *device_path, u64 devid);
439 void btrfs_cleanup_fs_uuids(void);
440 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
441 int btrfs_grow_device(struct btrfs_trans_handle *trans,
442 		      struct btrfs_device *device, u64 new_size);
443 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
444 				       u8 *uuid, u8 *fsid);
445 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
446 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path);
447 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
448 				  const char *device_path,
449 				  struct btrfs_device *srcdev,
450 				  struct btrfs_device **device_out);
451 int btrfs_balance(struct btrfs_balance_control *bctl,
452 		  struct btrfs_ioctl_balance_args *bargs);
453 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
454 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
455 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
456 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
457 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
458 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info);
459 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset);
460 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
461 			 struct btrfs_device *device, u64 num_bytes,
462 			 u64 search_start, u64 *start, u64 *max_avail);
463 int find_free_dev_extent(struct btrfs_trans_handle *trans,
464 			 struct btrfs_device *device, u64 num_bytes,
465 			 u64 *start, u64 *max_avail);
466 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
467 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
468 			struct btrfs_ioctl_get_dev_stats *stats);
469 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
470 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
471 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
472 			struct btrfs_fs_info *fs_info);
473 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
474 					struct btrfs_device *srcdev);
475 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
476 				      struct btrfs_device *srcdev);
477 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
478 				      struct btrfs_device *tgtdev);
479 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
480 					      struct btrfs_device *tgtdev);
481 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path);
482 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
483 			   u64 logical, u64 len);
484 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
485 				    u64 logical);
486 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
487 				struct btrfs_fs_info *fs_info,
488 				u64 chunk_offset, u64 chunk_size);
489 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
490 		       struct btrfs_fs_info *fs_info, u64 chunk_offset);
491 
492 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
493 				      int index)
494 {
495 	atomic_inc(dev->dev_stat_values + index);
496 	/*
497 	 * This memory barrier orders stores updating statistics before stores
498 	 * updating dev_stats_ccnt.
499 	 *
500 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
501 	 */
502 	smp_mb__before_atomic();
503 	atomic_inc(&dev->dev_stats_ccnt);
504 }
505 
506 static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
507 				      int index)
508 {
509 	return atomic_read(dev->dev_stat_values + index);
510 }
511 
512 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
513 						int index)
514 {
515 	int ret;
516 
517 	ret = atomic_xchg(dev->dev_stat_values + index, 0);
518 	/*
519 	 * atomic_xchg implies a full memory barriers as per atomic_t.txt:
520 	 * - RMW operations that have a return value are fully ordered;
521 	 *
522 	 * This implicit memory barriers is paired with the smp_rmb in
523 	 * btrfs_run_dev_stats
524 	 */
525 	atomic_inc(&dev->dev_stats_ccnt);
526 	return ret;
527 }
528 
529 static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
530 				      int index, unsigned long val)
531 {
532 	atomic_set(dev->dev_stat_values + index, val);
533 	/*
534 	 * This memory barrier orders stores updating statistics before stores
535 	 * updating dev_stats_ccnt.
536 	 *
537 	 * It pairs with smp_rmb() in btrfs_run_dev_stats().
538 	 */
539 	smp_mb__before_atomic();
540 	atomic_inc(&dev->dev_stats_ccnt);
541 }
542 
543 static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
544 					int index)
545 {
546 	btrfs_dev_stat_set(dev, index, 0);
547 }
548 
549 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info);
550 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
551 					struct btrfs_transaction *transaction);
552 
553 struct list_head *btrfs_get_fs_uuids(void);
554 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
555 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
556 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
557 					struct btrfs_device *failing_dev);
558 
559 #endif
560