include/linux/raid/md_k.h

   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
  18 /* and dm-bio-list.h is not under include/linux because.... ??? */
  19 #include "../../../drivers/md/dm-bio-list.h"
  20
  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
  41 static 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
  59 static 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
  77 typedef struct mddev_s mddev_t;
  78 typedef 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
  92 static inline int disk_faulty(mdp_disk_t * d)
  93 {
  94         return d->state & (1 << MD_DISK_FAULTY);
  95 }
  96
  97 static inline int disk_active(mdp_disk_t * d)
  98 {
  99         return d->state & (1 << MD_DISK_ACTIVE);
 100 }
 101
 102 static inline int disk_sync(mdp_disk_t * d)
 103 {
 104         return d->state & (1 << MD_DISK_SYNC);
 105 }
 106
 107 static inline int disk_spare(mdp_disk_t * d)
 108 {
 109         return !disk_sync(d) && !disk_active(d) && !disk_faulty(d);
 110 }
 111
 112 static inline int disk_removed(mdp_disk_t * d)
 113 {
 114         return d->state & (1 << MD_DISK_REMOVED);
 115 }
 116
 117 static inline void mark_disk_faulty(mdp_disk_t * d)
 118 {
 119         d->state |= (1 << MD_DISK_FAULTY);
 120 }
 121
 122 static inline void mark_disk_active(mdp_disk_t * d)
 123 {
 124         d->state |= (1 << MD_DISK_ACTIVE);
 125 }
 126
 127 static inline void mark_disk_sync(mdp_disk_t * d)
 128 {
 129         d->state |= (1 << MD_DISK_SYNC);
 130 }
 131
 132 static inline void mark_disk_spare(mdp_disk_t * d)
 133 {
 134         d->state = 0;
 135 }
 136
 137 static inline void mark_disk_removed(mdp_disk_t * d)
 138 {
 139         d->state = (1 << MD_DISK_FAULTY) | (1 << MD_DISK_REMOVED);
 140 }
 141
 142 static inline void mark_disk_inactive(mdp_disk_t * d)
 143 {
 144         d->state &= ~(1 << MD_DISK_ACTIVE);
 145 }
 146
 147 static 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  */
 155 struct 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         int saved_raid_disk;            /* role that device used to have in the
 187                                          * array and could again if we did a partial
 188                                          * resync from the bitmap
 189                                          */
 190
 191         atomic_t        nr_pending;     /* number of pending requests.
 192                                          * only maintained for arrays that
 193                                          * support hot removal
 194                                          */
 195 };
 196
 197 typedef struct mdk_personality_s mdk_personality_t;
 198
 199 struct mddev_s
 200 {
 201         void                            *private;
 202         mdk_personality_t               *pers;
 203         dev_t                           unit;
 204         int                             md_minor;
 205         struct list_head                disks;
 206         int                             sb_dirty;
 207         int                             ro;
 208
 209         struct gendisk                  *gendisk;
 210
 211         /* Superblock information */
 212         int                             major_version,
 213                                         minor_version,
 214                                         patch_version;
 215         int                             persistent;
 216         int                             chunk_size;
 217         time_t                          ctime, utime;
 218         int                             level, layout;
 219         int                             raid_disks;
 220         int                             max_disks;
 221         sector_t                        size; /* used size of component devices */
 222         sector_t                        array_size; /* exported array size */
 223         __u64                           events;
 224
 225         char                            uuid[16];
 226
 227         struct mdk_thread_s             *thread;        /* management thread */
 228         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
 229         sector_t                        curr_resync;    /* blocks scheduled */
 230         unsigned long                   resync_mark;    /* a recent timestamp */
 231         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
 232
 233         sector_t                        resync_max_sectors; /* may be set by personality */
 234         /* recovery/resync flags
 235          * NEEDED:   we might need to start a resync/recover
 236          * RUNNING:  a thread is running, or about to be started
 237          * SYNC:     actually doing a resync, not a recovery
 238          * ERR:      and IO error was detected - abort the resync/recovery
 239          * INTR:     someone requested a (clean) early abort.
 240          * DONE:     thread is done and is waiting to be reaped
 241          */
 242 #define MD_RECOVERY_RUNNING     0
 243 #define MD_RECOVERY_SYNC        1
 244 #define MD_RECOVERY_ERR         2
 245 #define MD_RECOVERY_INTR        3
 246 #define MD_RECOVERY_DONE        4
 247 #define MD_RECOVERY_NEEDED      5
 248         unsigned long                   recovery;
 249
 250         int                             in_sync;        /* know to not need resync */
 251         struct semaphore                reconfig_sem;
 252         atomic_t                        active;
 253
 254         int                             changed;        /* true if we might need to reread partition info */
 255         int                             degraded;       /* whether md should consider
 256                                                          * adding a spare
 257                                                          */
 258
 259         atomic_t                        recovery_active; /* blocks scheduled, but not written */
 260         wait_queue_head_t               recovery_wait;
 261         sector_t                        recovery_cp;
 262
 263         spinlock_t                      write_lock;
 264         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
 265         atomic_t                        pending_writes; /* number of active superblock writes */
 266
 267         unsigned int                    safemode;       /* if set, update "clean" superblock
 268                                                          * when no writes pending.
 269                                                          */
 270         unsigned int                    safemode_delay;
 271         struct timer_list               safemode_timer;
 272         atomic_t                        writes_pending;
 273         request_queue_t                 *queue; /* for plugging ... */
 274
 275         struct bitmap                   *bitmap; /* the bitmap for the device */
 276         struct file                     *bitmap_file; /* the bitmap file */
 277         long                            bitmap_offset; /* offset from superblock of
 278                                                         * start of bitmap. May be
 279                                                         * negative, but not '0'
 280                                                         */
 281         long                            default_bitmap_offset; /* this is the offset to use when
 282                                                                 * hot-adding a bitmap.  It should
 283                                                                 * eventually be settable by sysfs.
 284                                                                 */
 285
 286         struct list_head                all_mddevs;
 287 };
 288
 289
 290 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
 291 {
 292         int faulty = rdev->faulty;
 293         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
 294                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 295 }
 296
 297 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
 298 {
 299         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 300 }
 301
 302 struct mdk_personality_s
 303 {
 304         char *name;
 305         struct module *owner;
 306         int (*make_request)(request_queue_t *q, struct bio *bio);
 307         int (*run)(mddev_t *mddev);
 308         int (*stop)(mddev_t *mddev);
 309         void (*status)(struct seq_file *seq, mddev_t *mddev);
 310         /* error_handler must set ->faulty and clear ->in_sync
 311          * if appropriate, and should abort recovery if needed
 312          */
 313         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
 314         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
 315         int (*hot_remove_disk) (mddev_t *mddev, int number);
 316         int (*spare_active) (mddev_t *mddev);
 317         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
 318         int (*resize) (mddev_t *mddev, sector_t sectors);
 319         int (*reshape) (mddev_t *mddev, int raid_disks);
 320         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
 321         /* quiesce moves between quiescence states
 322          * 0 - fully active
 323          * 1 - no new requests allowed
 324          * others - reserved
 325          */
 326         void (*quiesce) (mddev_t *mddev, int state);
 327 };
 328
 329
 330 static inline char * mdname (mddev_t * mddev)
 331 {
 332         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
 333 }
 334
 335 extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr);
 336
 337 /*
 338  * iterates through some rdev ringlist. It's safe to remove the
 339  * current 'rdev'. Dont touch 'tmp' though.
 340  */
 341 #define ITERATE_RDEV_GENERIC(head,rdev,tmp)                             \
 342                                                                         \
 343         for ((tmp) = (head).next;                                       \
 344                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
 345                         (tmp) = (tmp)->next, (tmp)->prev != &(head)     \
 346                 ; )
 347 /*
 348  * iterates through the 'same array disks' ringlist
 349  */
 350 #define ITERATE_RDEV(mddev,rdev,tmp)                                    \
 351         ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
 352
 353 /*
 354  * Iterates through 'pending RAID disks'
 355  */
 356 #define ITERATE_RDEV_PENDING(rdev,tmp)                                  \
 357         ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
 358
 359 typedef struct mdk_thread_s {
 360         void                    (*run) (mddev_t *mddev);
 361         mddev_t                 *mddev;
 362         wait_queue_head_t       wqueue;
 363         unsigned long           flags;
 364         struct completion       *event;
 365         struct task_struct      *tsk;
 366         unsigned long           timeout;
 367         const char              *name;
 368 } mdk_thread_t;
 369
 370 #define THREAD_WAKEUP  0
 371
 372 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
 373 do {                                                                    \
 374         wait_queue_t __wait;                                            \
 375         init_waitqueue_entry(&__wait, current);                         \
 376                                                                         \
 377         add_wait_queue(&wq, &__wait);                                   \
 378         for (;;) {                                                      \
 379                 set_current_state(TASK_UNINTERRUPTIBLE);                \
 380                 if (condition)                                          \
 381                         break;                                          \
 382                 spin_unlock_irq(&lock);                                 \
 383                 cmd;                                                    \
 384                 schedule();                                             \
 385                 spin_lock_irq(&lock);                                   \
 386         }                                                               \
 387         current->state = TASK_RUNNING;                                  \
 388         remove_wait_queue(&wq, &__wait);                                \
 389 } while (0)
 390
 391 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
 392 do {                                                                    \
 393         if (condition)                                                  \
 394                 break;                                                  \
 395         __wait_event_lock_irq(wq, condition, lock, cmd);                \
 396 } while (0)
 397
 398 #endif
 399