blob: f6247e2a47f7b643d8c88f2ee4c5b53f5c79f043 [file] [log] [blame]
Chris Mason0b86a832008-03-24 15:01:56 -04001/*
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_
Chris Mason8790d502008-04-03 16:29:03 -040021
Chris Masoncea9e442008-04-09 16:28:12 -040022#include <linux/bio.h>
Miao Xieb2117a32011-01-05 10:07:28 +000023#include <linux/sort.h>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000024#include <linux/btrfs.h>
Chris Mason8b712842008-06-11 16:50:36 -040025#include "async-thread.h"
Chris Masoncea9e442008-04-09 16:28:12 -040026
Miao Xieb2117a32011-01-05 10:07:28 +000027#define BTRFS_STRIPE_LEN (64 * 1024)
28
Chris Masonf2984462008-04-10 16:19:33 -040029struct buffer_head;
Chris Masonffbd5172009-04-20 15:50:09 -040030struct btrfs_pending_bios {
31 struct bio *head;
32 struct bio *tail;
33};
34
Chris Mason0b86a832008-03-24 15:01:56 -040035struct btrfs_device {
36 struct list_head dev_list;
Chris Masonb3075712008-04-22 09:22:07 -040037 struct list_head dev_alloc_list;
Yan Zheng2b820322008-11-17 21:11:30 -050038 struct btrfs_fs_devices *fs_devices;
Chris Mason0b86a832008-03-24 15:01:56 -040039 struct btrfs_root *dev_root;
Chris Masonffbd5172009-04-20 15:50:09 -040040
41 /* regular prio bios */
42 struct btrfs_pending_bios pending_bios;
43 /* WRITE_SYNC bios */
44 struct btrfs_pending_bios pending_sync_bios;
45
Chris Mason8b712842008-06-11 16:50:36 -040046 int running_pending;
Chris Masondfe25022008-05-13 13:46:40 -040047 u64 generation;
Chris Masonb3075712008-04-22 09:22:07 -040048
Yan Zheng2b820322008-11-17 21:11:30 -050049 int writeable;
Chris Masondfe25022008-05-13 13:46:40 -040050 int in_fs_metadata;
Chris Masoncd02dca2010-12-13 14:56:23 -050051 int missing;
Josef Bacikd5e20032011-08-04 14:52:27 +000052 int can_discard;
Stefan Behrens63a212a2012-11-05 18:29:28 +010053 int is_tgtdev_for_dev_replace;
Chris Masonb3075712008-04-22 09:22:07 -040054
Chris Mason8790d502008-04-03 16:29:03 -040055 spinlock_t io_lock;
Chris Mason0b86a832008-03-24 15:01:56 -040056
57 struct block_device *bdev;
58
Tejun Heod4d77622010-11-13 11:55:18 +010059 /* the mode sent to blkdev_get */
Chris Mason15916de2008-11-19 21:17:22 -050060 fmode_t mode;
61
Josef Bacik606686e2012-06-04 14:03:51 -040062 struct rcu_string *name;
Chris Mason8a4b83c2008-03-24 15:02:07 -040063
Chris Mason0b86a832008-03-24 15:01:56 -040064 /* the internal btrfs device id */
65 u64 devid;
66
67 /* size of the device */
68 u64 total_bytes;
69
Chris Balld6397ba2009-04-27 07:29:03 -040070 /* size of the disk */
71 u64 disk_total_bytes;
72
Chris Mason0b86a832008-03-24 15:01:56 -040073 /* bytes used */
74 u64 bytes_used;
75
76 /* optimal io alignment for this device */
77 u32 io_align;
78
79 /* optimal io width for this device */
80 u32 io_width;
81
82 /* minimal io size for this device */
83 u32 sector_size;
84
Chris Mason0b86a832008-03-24 15:01:56 -040085 /* type and info about this device */
86 u64 type;
87
Chris Mason0b86a832008-03-24 15:01:56 -040088 /* physical drive uuid (or lvm uuid) */
Chris Masone17cade2008-04-15 15:41:47 -040089 u8 uuid[BTRFS_UUID_SIZE];
Chris Mason8b712842008-06-11 16:50:36 -040090
Arne Jansena2de7332011-03-08 14:14:00 +010091 /* per-device scrub information */
Stefan Behrensd9d181c2012-11-02 09:58:09 +010092 struct scrub_ctx *scrub_device;
Arne Jansena2de7332011-03-08 14:14:00 +010093
Chris Mason8b712842008-06-11 16:50:36 -040094 struct btrfs_work work;
Xiao Guangrong1f781602011-04-20 10:09:16 +000095 struct rcu_head rcu;
96 struct work_struct rcu_work;
Arne Jansen90519d62011-05-23 14:30:00 +020097
98 /* readahead state */
99 spinlock_t reada_lock;
100 atomic_t reada_in_flight;
101 u64 reada_next;
102 struct reada_zone *reada_curr_zone;
103 struct radix_tree_root reada_zones;
104 struct radix_tree_root reada_extents;
Chris Mason387125f2011-11-18 15:07:51 -0500105
106 /* for sending down flush barriers */
107 struct bio *flush_bio;
108 struct completion flush_wait;
109 int nobarriers;
110
Stefan Behrens442a4f62012-05-25 16:06:08 +0200111 /* disk I/O failure stats. For detailed description refer to
112 * enum btrfs_dev_stat_values in ioctl.h */
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200113 int dev_stats_valid;
Stefan Behrens442a4f62012-05-25 16:06:08 +0200114 int dev_stats_dirty; /* counters need to be written to disk */
115 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX];
Chris Mason0b86a832008-03-24 15:01:56 -0400116};
117
Chris Mason8a4b83c2008-03-24 15:02:07 -0400118struct btrfs_fs_devices {
119 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */
120
Wu Fengguangd4a78942009-04-02 16:46:06 -0400121 /* the device with this id has the most recent copy of the super */
Chris Mason8a4b83c2008-03-24 15:02:07 -0400122 u64 latest_devid;
123 u64 latest_trans;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400124 u64 num_devices;
Chris Masona0af4692008-05-13 16:03:06 -0400125 u64 open_devices;
Yan Zheng2b820322008-11-17 21:11:30 -0500126 u64 rw_devices;
Chris Masoncd02dca2010-12-13 14:56:23 -0500127 u64 missing_devices;
Yan Zheng2b820322008-11-17 21:11:30 -0500128 u64 total_rw_bytes;
Josef Bacikd5e20032011-08-04 14:52:27 +0000129 u64 num_can_discard;
Josef Bacik02db0842012-06-21 16:03:58 -0400130 u64 total_devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400131 struct block_device *latest_bdev;
Chris Masone5e9a522009-06-10 15:17:02 -0400132
133 /* all of the devices in the FS, protected by a mutex
134 * so we can safely walk it to write out the supers without
135 * worrying about add/remove by the multi-device code
136 */
137 struct mutex device_list_mutex;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400138 struct list_head devices;
Chris Masonb3075712008-04-22 09:22:07 -0400139
140 /* devices not currently being allocated */
141 struct list_head alloc_list;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400142 struct list_head list;
Yan Zheng2b820322008-11-17 21:11:30 -0500143
144 struct btrfs_fs_devices *seed;
145 int seeding;
Yan Zheng2b820322008-11-17 21:11:30 -0500146
147 int opened;
Chris Masonc2898112009-06-10 09:51:32 -0400148
149 /* set when we find or add a device that doesn't have the
150 * nonrot flag set
151 */
152 int rotating;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400153};
154
Chris Mason9be33952013-05-17 18:30:14 -0400155/*
156 * we need the mirror number and stripe index to be passed around
157 * the call chain while we are processing end_io (especially errors).
158 * Really, what we need is a btrfs_bio structure that has this info
159 * and is properly sized with its stripe array, but we're not there
160 * quite yet. We have our own btrfs bioset, and all of the bios
161 * we allocate are actually btrfs_io_bios. We'll cram as much of
162 * struct btrfs_bio as we can into this over time.
163 */
164struct btrfs_io_bio {
165 unsigned long mirror_num;
166 unsigned long stripe_index;
167 struct bio bio;
168};
169
170static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
171{
172 return container_of(bio, struct btrfs_io_bio, bio);
173}
174
Chris Masoncea9e442008-04-09 16:28:12 -0400175struct btrfs_bio_stripe {
176 struct btrfs_device *dev;
177 u64 physical;
Li Dongyangfce3bb92011-03-24 10:24:26 +0000178 u64 length; /* only used for discard mappings */
Chris Masoncea9e442008-04-09 16:28:12 -0400179};
180
Jan Schmidta1d3c472011-08-04 17:15:33 +0200181struct btrfs_bio;
182typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
183
184struct btrfs_bio {
Chris Masoncea9e442008-04-09 16:28:12 -0400185 atomic_t stripes_pending;
186 bio_end_io_t *end_io;
Chris Mason7d2b4da2008-08-05 10:13:57 -0400187 struct bio *orig_bio;
Chris Masoncea9e442008-04-09 16:28:12 -0400188 void *private;
Chris Masona236aed2008-04-29 09:38:00 -0400189 atomic_t error;
190 int max_errors;
Chris Masoncea9e442008-04-09 16:28:12 -0400191 int num_stripes;
Jan Schmidta1d3c472011-08-04 17:15:33 +0200192 int mirror_num;
Chris Masoncea9e442008-04-09 16:28:12 -0400193 struct btrfs_bio_stripe stripes[];
194};
195
Miao Xieb2117a32011-01-05 10:07:28 +0000196struct btrfs_device_info {
197 struct btrfs_device *dev;
198 u64 dev_offset;
199 u64 max_avail;
Arne Jansen73c5de02011-04-12 12:07:57 +0200200 u64 total_avail;
Miao Xieb2117a32011-01-05 10:07:28 +0000201};
202
Liu Bo31e50222012-11-21 14:18:10 +0000203struct btrfs_raid_attr {
204 int sub_stripes; /* sub_stripes info for map */
205 int dev_stripes; /* stripes per dev */
206 int devs_max; /* max devs to use */
207 int devs_min; /* min devs needed */
208 int devs_increment; /* ndevs has to be a multiple of this */
209 int ncopies; /* how many copies to data has */
210};
211
liubo1abe9b82011-03-24 11:18:59 +0000212struct map_lookup {
213 u64 type;
214 int io_align;
215 int io_width;
216 int stripe_len;
217 int sector_size;
218 int num_stripes;
219 int sub_stripes;
220 struct btrfs_bio_stripe stripes[];
221};
222
Arne Jansena2de7332011-03-08 14:14:00 +0100223#define map_lookup_size(n) (sizeof(struct map_lookup) + \
224 (sizeof(struct btrfs_bio_stripe) * (n)))
225
Ilya Dryomovf43ffb62012-01-16 22:04:47 +0200226/*
227 * Restriper's general type filter
228 */
229#define BTRFS_BALANCE_DATA (1ULL << 0)
230#define BTRFS_BALANCE_SYSTEM (1ULL << 1)
231#define BTRFS_BALANCE_METADATA (1ULL << 2)
232
233#define BTRFS_BALANCE_TYPE_MASK (BTRFS_BALANCE_DATA | \
234 BTRFS_BALANCE_SYSTEM | \
235 BTRFS_BALANCE_METADATA)
Ilya Dryomove4d8ec02012-01-16 22:04:48 +0200236
237#define BTRFS_BALANCE_FORCE (1ULL << 3)
Ilya Dryomov59641012012-01-16 22:04:48 +0200238#define BTRFS_BALANCE_RESUME (1ULL << 4)
Ilya Dryomove4d8ec02012-01-16 22:04:48 +0200239
Ilya Dryomoved25e9b2012-01-16 22:04:47 +0200240/*
241 * Balance filters
242 */
243#define BTRFS_BALANCE_ARGS_PROFILES (1ULL << 0)
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +0200244#define BTRFS_BALANCE_ARGS_USAGE (1ULL << 1)
Ilya Dryomov409d4042012-01-16 22:04:47 +0200245#define BTRFS_BALANCE_ARGS_DEVID (1ULL << 2)
Ilya Dryomov94e60d52012-01-16 22:04:48 +0200246#define BTRFS_BALANCE_ARGS_DRANGE (1ULL << 3)
Ilya Dryomovea671762012-01-16 22:04:48 +0200247#define BTRFS_BALANCE_ARGS_VRANGE (1ULL << 4)
Ilya Dryomovf43ffb62012-01-16 22:04:47 +0200248
Ilya Dryomovcfa4c962012-01-16 22:04:48 +0200249/*
250 * Profile changing flags. When SOFT is set we won't relocate chunk if
251 * it already has the target profile (even though it may be
252 * half-filled).
253 */
Ilya Dryomove4d8ec02012-01-16 22:04:48 +0200254#define BTRFS_BALANCE_ARGS_CONVERT (1ULL << 8)
Ilya Dryomovcfa4c962012-01-16 22:04:48 +0200255#define BTRFS_BALANCE_ARGS_SOFT (1ULL << 9)
Ilya Dryomove4d8ec02012-01-16 22:04:48 +0200256
Ilya Dryomovc9e9f972012-01-16 22:04:47 +0200257struct btrfs_balance_args;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +0200258struct btrfs_balance_progress;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +0200259struct btrfs_balance_control {
260 struct btrfs_fs_info *fs_info;
261
262 struct btrfs_balance_args data;
263 struct btrfs_balance_args meta;
264 struct btrfs_balance_args sys;
265
266 u64 flags;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +0200267
268 struct btrfs_balance_progress stat;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +0200269};
270
Miao Xie6d07bce2011-01-05 10:07:31 +0000271int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
272 u64 end, u64 *length);
273
Jan Schmidta1d3c472011-08-04 17:15:33 +0200274#define btrfs_bio_size(n) (sizeof(struct btrfs_bio) + \
Chris Masoncea9e442008-04-09 16:28:12 -0400275 (sizeof(struct btrfs_bio_stripe) * (n)))
276
Stefan Behrens3ec706c2012-11-05 15:46:42 +0100277int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
Chris Masoncea9e442008-04-09 16:28:12 -0400278 u64 logical, u64 *length,
Jan Schmidta1d3c472011-08-04 17:15:33 +0200279 struct btrfs_bio **bbio_ret, int mirror_num);
Yan Zhenga512bbf2008-12-08 16:46:26 -0500280int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
281 u64 chunk_start, u64 physical, u64 devid,
282 u64 **logical, int *naddrs, int *stripe_len);
Yan Zhenge4404d62008-12-12 10:03:26 -0500283int btrfs_read_sys_array(struct btrfs_root *root);
Chris Mason0b86a832008-03-24 15:01:56 -0400284int btrfs_read_chunk_tree(struct btrfs_root *root);
285int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
Yan Zheng2b820322008-11-17 21:11:30 -0500286 struct btrfs_root *extent_root, u64 type);
Chris Mason0b86a832008-03-24 15:01:56 -0400287void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
288void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
Chris Masonf1885912008-04-09 16:28:12 -0400289int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
Chris Mason8b712842008-06-11 16:50:36 -0400290 int mirror_num, int async_submit);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400291int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
Christoph Hellwig97288f22008-12-02 06:36:09 -0500292 fmode_t flags, void *holder);
293int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
Chris Mason8a4b83c2008-03-24 15:02:07 -0400294 struct btrfs_fs_devices **fs_devices_ret);
295int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
Stefan Behrens8dabb742012-11-06 13:15:27 +0100296void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
297 struct btrfs_fs_devices *fs_devices, int step);
Stefan Behrens7ba15b72012-11-05 14:42:30 +0100298int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
299 char *device_path,
300 struct btrfs_device **device);
Chris Masona061fc82008-05-07 11:43:44 -0400301int btrfs_rm_device(struct btrfs_root *root, char *device_path);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100302void btrfs_cleanup_fs_uuids(void);
Stefan Behrens5d964052012-11-05 14:59:07 +0100303int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len);
Chris Mason8f18cf12008-04-25 16:53:30 -0400304int btrfs_grow_device(struct btrfs_trans_handle *trans,
305 struct btrfs_device *device, u64 new_size);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +0100306struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
Yan Zheng2b820322008-11-17 21:11:30 -0500307 u8 *uuid, u8 *fsid);
Chris Mason8f18cf12008-04-25 16:53:30 -0400308int btrfs_shrink_device(struct btrfs_device *device, u64 new_size);
Chris Mason788f20e2008-04-28 15:29:42 -0400309int btrfs_init_new_device(struct btrfs_root *root, char *path);
Stefan Behrense93c89c2012-11-05 17:33:06 +0100310int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
311 struct btrfs_device **device_out);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +0200312int btrfs_balance(struct btrfs_balance_control *bctl,
313 struct btrfs_ioctl_balance_args *bargs);
Ilya Dryomov2b6ba622012-06-22 12:24:13 -0600314int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
Ilya Dryomov68310a52012-06-22 12:24:12 -0600315int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
Ilya Dryomov837d5b62012-01-16 22:04:49 +0200316int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
Ilya Dryomova7e99c62012-01-16 22:04:49 +0200317int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
Yan Zheng2b820322008-11-17 21:11:30 -0500318int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset);
Li Zefan125ccb02011-12-08 15:07:24 +0800319int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
Josef Bacikba1bf482009-09-11 16:11:19 -0400320 u64 *start, u64 *max_avail);
Stefan Behrens442a4f62012-05-25 16:06:08 +0200321void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index);
Stefan Behrensc11d2c22012-05-25 16:06:09 +0200322int btrfs_get_dev_stats(struct btrfs_root *root,
David Sterbab27f7c02012-06-22 06:30:39 -0600323 struct btrfs_ioctl_get_dev_stats *stats);
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200324int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
325int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
326 struct btrfs_fs_info *fs_info);
Stefan Behrense93c89c2012-11-05 17:33:06 +0100327void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
328 struct btrfs_device *srcdev);
329void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
330 struct btrfs_device *tgtdev);
331void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
332 struct btrfs_device *tgtdev);
Stefan Behrensa8a6dab2012-11-05 15:50:14 +0100333int btrfs_scratch_superblock(struct btrfs_device *device);
David Woodhouse53b381b2013-01-29 18:40:14 -0500334int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
335 u64 logical, u64 len, int mirror_num);
336unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
337 struct btrfs_mapping_tree *map_tree,
338 u64 logical);
Stefan Behrens442a4f62012-05-25 16:06:08 +0200339static inline void btrfs_dev_stat_inc(struct btrfs_device *dev,
340 int index)
341{
342 atomic_inc(dev->dev_stat_values + index);
343 dev->dev_stats_dirty = 1;
344}
345
346static inline int btrfs_dev_stat_read(struct btrfs_device *dev,
347 int index)
348{
349 return atomic_read(dev->dev_stat_values + index);
350}
351
352static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev,
353 int index)
354{
355 int ret;
356
357 ret = atomic_xchg(dev->dev_stat_values + index, 0);
358 dev->dev_stats_dirty = 1;
359 return ret;
360}
361
362static inline void btrfs_dev_stat_set(struct btrfs_device *dev,
363 int index, unsigned long val)
364{
365 atomic_set(dev->dev_stat_values + index, val);
366 dev->dev_stats_dirty = 1;
367}
368
369static inline void btrfs_dev_stat_reset(struct btrfs_device *dev,
370 int index)
371{
372 btrfs_dev_stat_set(dev, index, 0);
373}
Chris Mason0b86a832008-03-24 15:01:56 -0400374#endif