| 1 |
1275 |
phoenix |
/*
|
| 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 |
|
|
|
| 18 |
|
|
#define MD_RESERVED 0UL
|
| 19 |
|
|
#define LINEAR 1UL
|
| 20 |
|
|
#define RAID0 2UL
|
| 21 |
|
|
#define RAID1 3UL
|
| 22 |
|
|
#define RAID5 4UL
|
| 23 |
|
|
#define TRANSLUCENT 5UL
|
| 24 |
|
|
#define HSM 6UL
|
| 25 |
|
|
#define MULTIPATH 7UL
|
| 26 |
|
|
#define MAX_PERSONALITY 8UL
|
| 27 |
|
|
|
| 28 |
|
|
static inline int pers_to_level (int pers)
|
| 29 |
|
|
{
|
| 30 |
|
|
switch (pers) {
|
| 31 |
|
|
case MULTIPATH: return -4;
|
| 32 |
|
|
case HSM: return -3;
|
| 33 |
|
|
case TRANSLUCENT: return -2;
|
| 34 |
|
|
case LINEAR: return -1;
|
| 35 |
|
|
case RAID0: return 0;
|
| 36 |
|
|
case RAID1: return 1;
|
| 37 |
|
|
case RAID5: return 5;
|
| 38 |
|
|
}
|
| 39 |
|
|
BUG();
|
| 40 |
|
|
return MD_RESERVED;
|
| 41 |
|
|
}
|
| 42 |
|
|
|
| 43 |
|
|
static inline int level_to_pers (int level)
|
| 44 |
|
|
{
|
| 45 |
|
|
switch (level) {
|
| 46 |
|
|
case -4: return MULTIPATH;
|
| 47 |
|
|
case -3: return HSM;
|
| 48 |
|
|
case -2: return TRANSLUCENT;
|
| 49 |
|
|
case -1: return LINEAR;
|
| 50 |
|
|
case 0: return RAID0;
|
| 51 |
|
|
case 1: return RAID1;
|
| 52 |
|
|
case 4:
|
| 53 |
|
|
case 5: return RAID5;
|
| 54 |
|
|
}
|
| 55 |
|
|
return MD_RESERVED;
|
| 56 |
|
|
}
|
| 57 |
|
|
|
| 58 |
|
|
typedef struct mddev_s mddev_t;
|
| 59 |
|
|
typedef struct mdk_rdev_s mdk_rdev_t;
|
| 60 |
|
|
|
| 61 |
|
|
#if (MINORBITS != 8)
|
| 62 |
|
|
#error MD does not handle bigger kdev yet
|
| 63 |
|
|
#endif
|
| 64 |
|
|
|
| 65 |
|
|
#define MAX_MD_DEVS (1<<MINORBITS) /* Max number of md dev */
|
| 66 |
|
|
|
| 67 |
|
|
/*
|
| 68 |
|
|
* Maps a kdev to an mddev/subdev. How 'data' is handled is up to
|
| 69 |
|
|
* the personality. (eg. HSM uses this to identify individual LVs)
|
| 70 |
|
|
*/
|
| 71 |
|
|
typedef struct dev_mapping_s {
|
| 72 |
|
|
mddev_t *mddev;
|
| 73 |
|
|
void *data;
|
| 74 |
|
|
} dev_mapping_t;
|
| 75 |
|
|
|
| 76 |
|
|
extern dev_mapping_t mddev_map [MAX_MD_DEVS];
|
| 77 |
|
|
|
| 78 |
|
|
static inline mddev_t * kdev_to_mddev (kdev_t dev)
|
| 79 |
|
|
{
|
| 80 |
|
|
if (MAJOR(dev) != MD_MAJOR)
|
| 81 |
|
|
BUG();
|
| 82 |
|
|
return mddev_map[MINOR(dev)].mddev;
|
| 83 |
|
|
}
|
| 84 |
|
|
|
| 85 |
|
|
/*
|
| 86 |
|
|
* options passed in raidrun:
|
| 87 |
|
|
*/
|
| 88 |
|
|
|
| 89 |
|
|
#define MAX_CHUNK_SIZE (4096*1024)
|
| 90 |
|
|
|
| 91 |
|
|
/*
|
| 92 |
|
|
* default readahead
|
| 93 |
|
|
*/
|
| 94 |
|
|
#define MD_READAHEAD vm_max_readahead
|
| 95 |
|
|
|
| 96 |
|
|
static inline int disk_faulty(mdp_disk_t * d)
|
| 97 |
|
|
{
|
| 98 |
|
|
return d->state & (1 << MD_DISK_FAULTY);
|
| 99 |
|
|
}
|
| 100 |
|
|
|
| 101 |
|
|
static inline int disk_active(mdp_disk_t * d)
|
| 102 |
|
|
{
|
| 103 |
|
|
return d->state & (1 << MD_DISK_ACTIVE);
|
| 104 |
|
|
}
|
| 105 |
|
|
|
| 106 |
|
|
static inline int disk_sync(mdp_disk_t * d)
|
| 107 |
|
|
{
|
| 108 |
|
|
return d->state & (1 << MD_DISK_SYNC);
|
| 109 |
|
|
}
|
| 110 |
|
|
|
| 111 |
|
|
static inline int disk_spare(mdp_disk_t * d)
|
| 112 |
|
|
{
|
| 113 |
|
|
return !disk_sync(d) && !disk_active(d) && !disk_faulty(d);
|
| 114 |
|
|
}
|
| 115 |
|
|
|
| 116 |
|
|
static inline int disk_removed(mdp_disk_t * d)
|
| 117 |
|
|
{
|
| 118 |
|
|
return d->state & (1 << MD_DISK_REMOVED);
|
| 119 |
|
|
}
|
| 120 |
|
|
|
| 121 |
|
|
static inline void mark_disk_faulty(mdp_disk_t * d)
|
| 122 |
|
|
{
|
| 123 |
|
|
d->state |= (1 << MD_DISK_FAULTY);
|
| 124 |
|
|
}
|
| 125 |
|
|
|
| 126 |
|
|
static inline void mark_disk_active(mdp_disk_t * d)
|
| 127 |
|
|
{
|
| 128 |
|
|
d->state |= (1 << MD_DISK_ACTIVE);
|
| 129 |
|
|
}
|
| 130 |
|
|
|
| 131 |
|
|
static inline void mark_disk_sync(mdp_disk_t * d)
|
| 132 |
|
|
{
|
| 133 |
|
|
d->state |= (1 << MD_DISK_SYNC);
|
| 134 |
|
|
}
|
| 135 |
|
|
|
| 136 |
|
|
static inline void mark_disk_spare(mdp_disk_t * d)
|
| 137 |
|
|
{
|
| 138 |
|
|
d->state = 0;
|
| 139 |
|
|
}
|
| 140 |
|
|
|
| 141 |
|
|
static inline void mark_disk_removed(mdp_disk_t * d)
|
| 142 |
|
|
{
|
| 143 |
|
|
d->state = (1 << MD_DISK_FAULTY) | (1 << MD_DISK_REMOVED);
|
| 144 |
|
|
}
|
| 145 |
|
|
|
| 146 |
|
|
static inline void mark_disk_inactive(mdp_disk_t * d)
|
| 147 |
|
|
{
|
| 148 |
|
|
d->state &= ~(1 << MD_DISK_ACTIVE);
|
| 149 |
|
|
}
|
| 150 |
|
|
|
| 151 |
|
|
static inline void mark_disk_nonsync(mdp_disk_t * d)
|
| 152 |
|
|
{
|
| 153 |
|
|
d->state &= ~(1 << MD_DISK_SYNC);
|
| 154 |
|
|
}
|
| 155 |
|
|
|
| 156 |
|
|
/*
|
| 157 |
|
|
* MD's 'extended' device
|
| 158 |
|
|
*/
|
| 159 |
|
|
struct mdk_rdev_s
|
| 160 |
|
|
{
|
| 161 |
|
|
struct md_list_head same_set; /* RAID devices within the same set */
|
| 162 |
|
|
struct md_list_head all; /* all RAID devices */
|
| 163 |
|
|
struct md_list_head pending; /* undetected RAID devices */
|
| 164 |
|
|
|
| 165 |
|
|
kdev_t dev; /* Device number */
|
| 166 |
|
|
kdev_t old_dev; /* "" when it was last imported */
|
| 167 |
|
|
unsigned long size; /* Device size (in blocks) */
|
| 168 |
|
|
mddev_t *mddev; /* RAID array if running */
|
| 169 |
|
|
unsigned long last_events; /* IO event timestamp */
|
| 170 |
|
|
|
| 171 |
|
|
struct block_device *bdev; /* block device handle */
|
| 172 |
|
|
|
| 173 |
|
|
mdp_super_t *sb;
|
| 174 |
|
|
struct page *sb_page;
|
| 175 |
|
|
unsigned long sb_offset;
|
| 176 |
|
|
|
| 177 |
|
|
int alias_device; /* device alias to the same disk */
|
| 178 |
|
|
int faulty; /* if faulty do not issue IO requests */
|
| 179 |
|
|
int desc_nr; /* descriptor index in the superblock */
|
| 180 |
|
|
};
|
| 181 |
|
|
|
| 182 |
|
|
|
| 183 |
|
|
/*
|
| 184 |
|
|
* disk operations in a working array:
|
| 185 |
|
|
*/
|
| 186 |
|
|
#define DISKOP_SPARE_INACTIVE 0
|
| 187 |
|
|
#define DISKOP_SPARE_WRITE 1
|
| 188 |
|
|
#define DISKOP_SPARE_ACTIVE 2
|
| 189 |
|
|
#define DISKOP_HOT_REMOVE_DISK 3
|
| 190 |
|
|
#define DISKOP_HOT_ADD_DISK 4
|
| 191 |
|
|
|
| 192 |
|
|
typedef struct mdk_personality_s mdk_personality_t;
|
| 193 |
|
|
|
| 194 |
|
|
struct mddev_s
|
| 195 |
|
|
{
|
| 196 |
|
|
void *private;
|
| 197 |
|
|
mdk_personality_t *pers;
|
| 198 |
|
|
int __minor;
|
| 199 |
|
|
mdp_super_t *sb;
|
| 200 |
|
|
int nb_dev;
|
| 201 |
|
|
struct md_list_head disks;
|
| 202 |
|
|
int sb_dirty;
|
| 203 |
|
|
mdu_param_t param;
|
| 204 |
|
|
int ro;
|
| 205 |
|
|
unsigned long curr_resync; /* blocks scheduled */
|
| 206 |
|
|
unsigned long resync_mark; /* a recent timestamp */
|
| 207 |
|
|
unsigned long resync_mark_cnt;/* blocks written at resync_mark */
|
| 208 |
|
|
char *name;
|
| 209 |
|
|
int recovery_running;
|
| 210 |
|
|
struct semaphore reconfig_sem;
|
| 211 |
|
|
struct semaphore recovery_sem;
|
| 212 |
|
|
struct semaphore resync_sem;
|
| 213 |
|
|
atomic_t active;
|
| 214 |
|
|
|
| 215 |
|
|
atomic_t recovery_active; /* blocks scheduled, but not written */
|
| 216 |
|
|
md_wait_queue_head_t recovery_wait;
|
| 217 |
|
|
|
| 218 |
|
|
struct md_list_head all_mddevs;
|
| 219 |
|
|
};
|
| 220 |
|
|
|
| 221 |
|
|
struct mdk_personality_s
|
| 222 |
|
|
{
|
| 223 |
|
|
char *name;
|
| 224 |
|
|
int (*make_request)(mddev_t *mddev, int rw, struct buffer_head * bh);
|
| 225 |
|
|
int (*run)(mddev_t *mddev);
|
| 226 |
|
|
int (*stop)(mddev_t *mddev);
|
| 227 |
|
|
void (*status)(struct seq_file *seq, mddev_t *mddev);
|
| 228 |
|
|
int (*error_handler)(mddev_t *mddev, kdev_t dev);
|
| 229 |
|
|
|
| 230 |
|
|
/*
|
| 231 |
|
|
* Some personalities (RAID-1, RAID-5) can have disks hot-added and
|
| 232 |
|
|
* hot-removed. Hot removal is different from failure. (failure marks
|
| 233 |
|
|
* a disk inactive, but the disk is still part of the array) The interface
|
| 234 |
|
|
* to such operations is the 'pers->diskop()' function, can be NULL.
|
| 235 |
|
|
*
|
| 236 |
|
|
* the diskop function can change the pointer pointing to the incoming
|
| 237 |
|
|
* descriptor, but must do so very carefully. (currently only
|
| 238 |
|
|
* SPARE_ACTIVE expects such a change)
|
| 239 |
|
|
*/
|
| 240 |
|
|
int (*diskop) (mddev_t *mddev, mdp_disk_t **descriptor, int state);
|
| 241 |
|
|
|
| 242 |
|
|
int (*stop_resync)(mddev_t *mddev);
|
| 243 |
|
|
int (*restart_resync)(mddev_t *mddev);
|
| 244 |
|
|
int (*sync_request)(mddev_t *mddev, unsigned long block_nr);
|
| 245 |
|
|
};
|
| 246 |
|
|
|
| 247 |
|
|
|
| 248 |
|
|
/*
|
| 249 |
|
|
* Currently we index md_array directly, based on the minor
|
| 250 |
|
|
* number. This will have to change to dynamic allocation
|
| 251 |
|
|
* once we start supporting partitioning of md devices.
|
| 252 |
|
|
*/
|
| 253 |
|
|
static inline int mdidx (mddev_t * mddev)
|
| 254 |
|
|
{
|
| 255 |
|
|
return mddev->__minor;
|
| 256 |
|
|
}
|
| 257 |
|
|
|
| 258 |
|
|
static inline kdev_t mddev_to_kdev(mddev_t * mddev)
|
| 259 |
|
|
{
|
| 260 |
|
|
return MKDEV(MD_MAJOR, mdidx(mddev));
|
| 261 |
|
|
}
|
| 262 |
|
|
|
| 263 |
|
|
extern mdk_rdev_t * find_rdev(mddev_t * mddev, kdev_t dev);
|
| 264 |
|
|
extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr);
|
| 265 |
|
|
extern mdp_disk_t *get_spare(mddev_t *mddev);
|
| 266 |
|
|
|
| 267 |
|
|
/*
|
| 268 |
|
|
* iterates through some rdev ringlist. It's safe to remove the
|
| 269 |
|
|
* current 'rdev'. Dont touch 'tmp' though.
|
| 270 |
|
|
*/
|
| 271 |
|
|
#define ITERATE_RDEV_GENERIC(head,field,rdev,tmp) \
|
| 272 |
|
|
\
|
| 273 |
|
|
for (tmp = head.next; \
|
| 274 |
|
|
rdev = md_list_entry(tmp, mdk_rdev_t, field), \
|
| 275 |
|
|
tmp = tmp->next, tmp->prev != &head \
|
| 276 |
|
|
; )
|
| 277 |
|
|
/*
|
| 278 |
|
|
* iterates through the 'same array disks' ringlist
|
| 279 |
|
|
*/
|
| 280 |
|
|
#define ITERATE_RDEV(mddev,rdev,tmp) \
|
| 281 |
|
|
ITERATE_RDEV_GENERIC((mddev)->disks,same_set,rdev,tmp)
|
| 282 |
|
|
|
| 283 |
|
|
/*
|
| 284 |
|
|
* Same as above, but assumes that the device has rdev->desc_nr numbered
|
| 285 |
|
|
* from 0 to mddev->nb_dev, and iterates through rdevs in ascending order.
|
| 286 |
|
|
*/
|
| 287 |
|
|
#define ITERATE_RDEV_ORDERED(mddev,rdev,i) \
|
| 288 |
|
|
for (i = 0; rdev = find_rdev_nr(mddev, i), i < mddev->nb_dev; i++)
|
| 289 |
|
|
|
| 290 |
|
|
|
| 291 |
|
|
/*
|
| 292 |
|
|
* Iterates through all 'RAID managed disks'
|
| 293 |
|
|
*/
|
| 294 |
|
|
#define ITERATE_RDEV_ALL(rdev,tmp) \
|
| 295 |
|
|
ITERATE_RDEV_GENERIC(all_raid_disks,all,rdev,tmp)
|
| 296 |
|
|
|
| 297 |
|
|
/*
|
| 298 |
|
|
* Iterates through 'pending RAID disks'
|
| 299 |
|
|
*/
|
| 300 |
|
|
#define ITERATE_RDEV_PENDING(rdev,tmp) \
|
| 301 |
|
|
ITERATE_RDEV_GENERIC(pending_raid_disks,pending,rdev,tmp)
|
| 302 |
|
|
|
| 303 |
|
|
/*
|
| 304 |
|
|
* iterates through all used mddevs in the system.
|
| 305 |
|
|
*/
|
| 306 |
|
|
#define ITERATE_MDDEV(mddev,tmp) \
|
| 307 |
|
|
\
|
| 308 |
|
|
for (tmp = all_mddevs.next; \
|
| 309 |
|
|
mddev = md_list_entry(tmp, mddev_t, all_mddevs), \
|
| 310 |
|
|
tmp = tmp->next, tmp->prev != &all_mddevs \
|
| 311 |
|
|
; )
|
| 312 |
|
|
|
| 313 |
|
|
static inline int lock_mddev (mddev_t * mddev)
|
| 314 |
|
|
{
|
| 315 |
|
|
return down_interruptible(&mddev->reconfig_sem);
|
| 316 |
|
|
}
|
| 317 |
|
|
|
| 318 |
|
|
static inline void unlock_mddev (mddev_t * mddev)
|
| 319 |
|
|
{
|
| 320 |
|
|
up(&mddev->reconfig_sem);
|
| 321 |
|
|
}
|
| 322 |
|
|
|
| 323 |
|
|
#define xchg_values(x,y) do { __typeof__(x) __tmp = x; \
|
| 324 |
|
|
x = y; y = __tmp; } while (0)
|
| 325 |
|
|
|
| 326 |
|
|
typedef struct mdk_thread_s {
|
| 327 |
|
|
void (*run) (void *data);
|
| 328 |
|
|
void *data;
|
| 329 |
|
|
md_wait_queue_head_t wqueue;
|
| 330 |
|
|
unsigned long flags;
|
| 331 |
|
|
struct completion *event;
|
| 332 |
|
|
struct task_struct *tsk;
|
| 333 |
|
|
const char *name;
|
| 334 |
|
|
} mdk_thread_t;
|
| 335 |
|
|
|
| 336 |
|
|
#define THREAD_WAKEUP 0
|
| 337 |
|
|
|
| 338 |
|
|
#define MAX_DISKNAME_LEN 64
|
| 339 |
|
|
|
| 340 |
|
|
typedef struct dev_name_s {
|
| 341 |
|
|
struct md_list_head list;
|
| 342 |
|
|
kdev_t dev;
|
| 343 |
|
|
char namebuf [MAX_DISKNAME_LEN];
|
| 344 |
|
|
char *name;
|
| 345 |
|
|
} dev_name_t;
|
| 346 |
|
|
|
| 347 |
|
|
|
| 348 |
|
|
#define __wait_event_lock_irq(wq, condition, lock) \
|
| 349 |
|
|
do { \
|
| 350 |
|
|
wait_queue_t __wait; \
|
| 351 |
|
|
init_waitqueue_entry(&__wait, current); \
|
| 352 |
|
|
\
|
| 353 |
|
|
add_wait_queue(&wq, &__wait); \
|
| 354 |
|
|
for (;;) { \
|
| 355 |
|
|
set_current_state(TASK_UNINTERRUPTIBLE); \
|
| 356 |
|
|
if (condition) \
|
| 357 |
|
|
break; \
|
| 358 |
|
|
spin_unlock_irq(&lock); \
|
| 359 |
|
|
run_task_queue(&tq_disk); \
|
| 360 |
|
|
schedule(); \
|
| 361 |
|
|
spin_lock_irq(&lock); \
|
| 362 |
|
|
} \
|
| 363 |
|
|
current->state = TASK_RUNNING; \
|
| 364 |
|
|
remove_wait_queue(&wq, &__wait); \
|
| 365 |
|
|
} while (0)
|
| 366 |
|
|
|
| 367 |
|
|
#define wait_event_lock_irq(wq, condition, lock) \
|
| 368 |
|
|
do { \
|
| 369 |
|
|
if (condition) \
|
| 370 |
|
|
break; \
|
| 371 |
|
|
__wait_event_lock_irq(wq, condition, lock); \
|
| 372 |
|
|
} while (0)
|
| 373 |
|
|
|
| 374 |
|
|
|
| 375 |
|
|
#define __wait_disk_event(wq, condition) \
|
| 376 |
|
|
do { \
|
| 377 |
|
|
wait_queue_t __wait; \
|
| 378 |
|
|
init_waitqueue_entry(&__wait, current); \
|
| 379 |
|
|
\
|
| 380 |
|
|
add_wait_queue(&wq, &__wait); \
|
| 381 |
|
|
for (;;) { \
|
| 382 |
|
|
set_current_state(TASK_UNINTERRUPTIBLE); \
|
| 383 |
|
|
if (condition) \
|
| 384 |
|
|
break; \
|
| 385 |
|
|
run_task_queue(&tq_disk); \
|
| 386 |
|
|
schedule(); \
|
| 387 |
|
|
} \
|
| 388 |
|
|
current->state = TASK_RUNNING; \
|
| 389 |
|
|
remove_wait_queue(&wq, &__wait); \
|
| 390 |
|
|
} while (0)
|
| 391 |
|
|
|
| 392 |
|
|
#define wait_disk_event(wq, condition) \
|
| 393 |
|
|
do { \
|
| 394 |
|
|
if (condition) \
|
| 395 |
|
|
break; \
|
| 396 |
|
|
__wait_disk_event(wq, condition); \
|
| 397 |
|
|
} while (0)
|
| 398 |
|
|
|
| 399 |
|
|
#endif
|
| 400 |
|
|
|
