]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - include/linux/raid/md_k.h
[PATCH] md: improve locking on 'safemode' and move superblock writes
[mirror_ubuntu-bionic-kernel.git] / include / linux / raid / md_k.h
CommitLineData
1da177e4
LT
1/*
2 md_k.h : kernel internal structure of the Linux MD driver
3 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
4
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2, or (at your option)
8 any later version.
9
10 You should have received a copy of the GNU General Public License
11 (for example /usr/src/linux/COPYING); if not, write to the Free
12 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
13*/
14
15#ifndef _MD_K_H
16#define _MD_K_H
17
06d91a5f
N
18/* and dm-bio-list.h is not under include/linux because.... ??? */
19#include "../../../drivers/md/dm-bio-list.h"
20
1da177e4
LT
21#define MD_RESERVED 0UL
22#define LINEAR 1UL
23#define RAID0 2UL
24#define RAID1 3UL
25#define RAID5 4UL
26#define TRANSLUCENT 5UL
27#define HSM 6UL
28#define MULTIPATH 7UL
29#define RAID6 8UL
30#define RAID10 9UL
31#define FAULTY 10UL
32#define MAX_PERSONALITY 11UL
33
34#define LEVEL_MULTIPATH (-4)
35#define LEVEL_LINEAR (-1)
36#define LEVEL_FAULTY (-5)
37
38#define MaxSector (~(sector_t)0)
39#define MD_THREAD_NAME_MAX 14
40
41static inline int pers_to_level (int pers)
42{
43 switch (pers) {
44 case FAULTY: return LEVEL_FAULTY;
45 case MULTIPATH: return LEVEL_MULTIPATH;
46 case HSM: return -3;
47 case TRANSLUCENT: return -2;
48 case LINEAR: return LEVEL_LINEAR;
49 case RAID0: return 0;
50 case RAID1: return 1;
51 case RAID5: return 5;
52 case RAID6: return 6;
53 case RAID10: return 10;
54 }
55 BUG();
56 return MD_RESERVED;
57}
58
59static inline int level_to_pers (int level)
60{
61 switch (level) {
62 case LEVEL_FAULTY: return FAULTY;
63 case LEVEL_MULTIPATH: return MULTIPATH;
64 case -3: return HSM;
65 case -2: return TRANSLUCENT;
66 case LEVEL_LINEAR: return LINEAR;
67 case 0: return RAID0;
68 case 1: return RAID1;
69 case 4:
70 case 5: return RAID5;
71 case 6: return RAID6;
72 case 10: return RAID10;
73 }
74 return MD_RESERVED;
75}
76
77typedef struct mddev_s mddev_t;
78typedef struct mdk_rdev_s mdk_rdev_t;
79
80#define MAX_MD_DEVS 256 /* Max number of md dev */
81
82/*
83 * options passed in raidrun:
84 */
85
86#define MAX_CHUNK_SIZE (4096*1024)
87
88/*
89 * default readahead
90 */
91
92static inline int disk_faulty(mdp_disk_t * d)
93{
94 return d->state & (1 << MD_DISK_FAULTY);
95}
96
97static inline int disk_active(mdp_disk_t * d)
98{
99 return d->state & (1 << MD_DISK_ACTIVE);
100}
101
102static inline int disk_sync(mdp_disk_t * d)
103{
104 return d->state & (1 << MD_DISK_SYNC);
105}
106
107static inline int disk_spare(mdp_disk_t * d)
108{
109 return !disk_sync(d) && !disk_active(d) && !disk_faulty(d);
110}
111
112static inline int disk_removed(mdp_disk_t * d)
113{
114 return d->state & (1 << MD_DISK_REMOVED);
115}
116
117static inline void mark_disk_faulty(mdp_disk_t * d)
118{
119 d->state |= (1 << MD_DISK_FAULTY);
120}
121
122static inline void mark_disk_active(mdp_disk_t * d)
123{
124 d->state |= (1 << MD_DISK_ACTIVE);
125}
126
127static inline void mark_disk_sync(mdp_disk_t * d)
128{
129 d->state |= (1 << MD_DISK_SYNC);
130}
131
132static inline void mark_disk_spare(mdp_disk_t * d)
133{
134 d->state = 0;
135}
136
137static inline void mark_disk_removed(mdp_disk_t * d)
138{
139 d->state = (1 << MD_DISK_FAULTY) | (1 << MD_DISK_REMOVED);
140}
141
142static inline void mark_disk_inactive(mdp_disk_t * d)
143{
144 d->state &= ~(1 << MD_DISK_ACTIVE);
145}
146
147static inline void mark_disk_nonsync(mdp_disk_t * d)
148{
149 d->state &= ~(1 << MD_DISK_SYNC);
150}
151
152/*
153 * MD's 'extended' device
154 */
155struct mdk_rdev_s
156{
157 struct list_head same_set; /* RAID devices within the same set */
158
159 sector_t size; /* Device size (in blocks) */
160 mddev_t *mddev; /* RAID array if running */
161 unsigned long last_events; /* IO event timestamp */
162
163 struct block_device *bdev; /* block device handle */
164
165 struct page *sb_page;
166 int sb_loaded;
167 sector_t data_offset; /* start of data in array */
168 sector_t sb_offset;
169 int preferred_minor; /* autorun support */
170
171 /* A device can be in one of three states based on two flags:
172 * Not working: faulty==1 in_sync==0
173 * Fully working: faulty==0 in_sync==1
174 * Working, but not
175 * in sync with array
176 * faulty==0 in_sync==0
177 *
178 * It can never have faulty==1, in_sync==1
179 * This reduces the burden of testing multiple flags in many cases
180 */
181 int faulty; /* if faulty do not issue IO requests */
182 int in_sync; /* device is a full member of the array */
183
184 int desc_nr; /* descriptor index in the superblock */
185 int raid_disk; /* role of device in array */
186
187 atomic_t nr_pending; /* number of pending requests.
188 * only maintained for arrays that
189 * support hot removal
190 */
191};
192
193typedef struct mdk_personality_s mdk_personality_t;
194
195struct mddev_s
196{
197 void *private;
198 mdk_personality_t *pers;
199 dev_t unit;
200 int md_minor;
201 struct list_head disks;
202 int sb_dirty;
203 int ro;
204
205 struct gendisk *gendisk;
206
207 /* Superblock information */
208 int major_version,
209 minor_version,
210 patch_version;
211 int persistent;
212 int chunk_size;
213 time_t ctime, utime;
214 int level, layout;
215 int raid_disks;
216 int max_disks;
217 sector_t size; /* used size of component devices */
218 sector_t array_size; /* exported array size */
219 __u64 events;
220
221 char uuid[16];
222
223 struct mdk_thread_s *thread; /* management thread */
224 struct mdk_thread_s *sync_thread; /* doing resync or reconstruct */
225 sector_t curr_resync; /* blocks scheduled */
226 unsigned long resync_mark; /* a recent timestamp */
227 sector_t resync_mark_cnt;/* blocks written at resync_mark */
228
229 sector_t resync_max_sectors; /* may be set by personality */
230 /* recovery/resync flags
231 * NEEDED: we might need to start a resync/recover
232 * RUNNING: a thread is running, or about to be started
233 * SYNC: actually doing a resync, not a recovery
234 * ERR: and IO error was detected - abort the resync/recovery
235 * INTR: someone requested a (clean) early abort.
236 * DONE: thread is done and is waiting to be reaped
237 */
238#define MD_RECOVERY_RUNNING 0
239#define MD_RECOVERY_SYNC 1
240#define MD_RECOVERY_ERR 2
241#define MD_RECOVERY_INTR 3
242#define MD_RECOVERY_DONE 4
243#define MD_RECOVERY_NEEDED 5
244 unsigned long recovery;
245
246 int in_sync; /* know to not need resync */
247 struct semaphore reconfig_sem;
248 atomic_t active;
249
250 int changed; /* true if we might need to reread partition info */
251 int degraded; /* whether md should consider
252 * adding a spare
253 */
254
255 atomic_t recovery_active; /* blocks scheduled, but not written */
256 wait_queue_head_t recovery_wait;
257 sector_t recovery_cp;
06d91a5f
N
258
259 spinlock_t write_lock;
260 struct bio_list write_list;
261
1da177e4
LT
262 unsigned int safemode; /* if set, update "clean" superblock
263 * when no writes pending.
264 */
265 unsigned int safemode_delay;
266 struct timer_list safemode_timer;
267 atomic_t writes_pending;
268 request_queue_t *queue; /* for plugging ... */
269
270 struct list_head all_mddevs;
271};
272
273
274static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
275{
276 int faulty = rdev->faulty;
277 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
278 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
279}
280
281static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
282{
283 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
284}
285
286struct mdk_personality_s
287{
288 char *name;
289 struct module *owner;
290 int (*make_request)(request_queue_t *q, struct bio *bio);
291 int (*run)(mddev_t *mddev);
292 int (*stop)(mddev_t *mddev);
293 void (*status)(struct seq_file *seq, mddev_t *mddev);
294 /* error_handler must set ->faulty and clear ->in_sync
295 * if appropriate, and should abort recovery if needed
296 */
297 void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
298 int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
299 int (*hot_remove_disk) (mddev_t *mddev, int number);
300 int (*spare_active) (mddev_t *mddev);
301 int (*sync_request)(mddev_t *mddev, sector_t sector_nr, int go_faster);
302 int (*resize) (mddev_t *mddev, sector_t sectors);
303 int (*reshape) (mddev_t *mddev, int raid_disks);
304 int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
305};
306
307
308static inline char * mdname (mddev_t * mddev)
309{
310 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
311}
312
313extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr);
314
315/*
316 * iterates through some rdev ringlist. It's safe to remove the
317 * current 'rdev'. Dont touch 'tmp' though.
318 */
319#define ITERATE_RDEV_GENERIC(head,rdev,tmp) \
320 \
321 for ((tmp) = (head).next; \
322 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)), \
323 (tmp) = (tmp)->next, (tmp)->prev != &(head) \
324 ; )
325/*
326 * iterates through the 'same array disks' ringlist
327 */
328#define ITERATE_RDEV(mddev,rdev,tmp) \
329 ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
330
331/*
332 * Iterates through 'pending RAID disks'
333 */
334#define ITERATE_RDEV_PENDING(rdev,tmp) \
335 ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
336
337typedef struct mdk_thread_s {
338 void (*run) (mddev_t *mddev);
339 mddev_t *mddev;
340 wait_queue_head_t wqueue;
341 unsigned long flags;
342 struct completion *event;
343 struct task_struct *tsk;
344 const char *name;
345} mdk_thread_t;
346
347#define THREAD_WAKEUP 0
348
349#define __wait_event_lock_irq(wq, condition, lock, cmd) \
350do { \
351 wait_queue_t __wait; \
352 init_waitqueue_entry(&__wait, current); \
353 \
354 add_wait_queue(&wq, &__wait); \
355 for (;;) { \
356 set_current_state(TASK_UNINTERRUPTIBLE); \
357 if (condition) \
358 break; \
359 spin_unlock_irq(&lock); \
360 cmd; \
361 schedule(); \
362 spin_lock_irq(&lock); \
363 } \
364 current->state = TASK_RUNNING; \
365 remove_wait_queue(&wq, &__wait); \
366} while (0)
367
368#define wait_event_lock_irq(wq, condition, lock, cmd) \
369do { \
370 if (condition) \
371 break; \
372 __wait_event_lock_irq(wq, condition, lock, cmd); \
373} while (0)
374
375#endif
376