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  * MD's 'extended' device
  90  */
  91 struct mdk_rdev_s
  92 {
  93         struct list_head same_set;      /* RAID devices within the same set */
  94
  95         sector_t size;                  /* Device size (in blocks) */
  96         mddev_t *mddev;                 /* RAID array if running */
  97         unsigned long last_events;      /* IO event timestamp */
  98
  99         struct block_device *bdev;      /* block device handle */
 100
 101         struct page     *sb_page;
 102         int             sb_loaded;
 103         sector_t        data_offset;    /* start of data in array */
 104         sector_t        sb_offset;
 105         int             preferred_minor;        /* autorun support */
 106
 107         /* A device can be in one of three states based on two flags:
 108          * Not working:   faulty==1 in_sync==0
 109          * Fully working: faulty==0 in_sync==1
 110          * Working, but not
 111          * in sync with array
 112          *                faulty==0 in_sync==0
 113          *
 114          * It can never have faulty==1, in_sync==1
 115          * This reduces the burden of testing multiple flags in many cases
 116          */
 117         int faulty;                     /* if faulty do not issue IO requests */
 118         int in_sync;                    /* device is a full member of the array */
 119
 120         unsigned long   flags;          /* Should include faulty and in_sync here. */
 121 #define WriteMostly     4               /* Avoid reading if at all possible */
 122
 123         int desc_nr;                    /* descriptor index in the superblock */
 124         int raid_disk;                  /* role of device in array */
 125         int saved_raid_disk;            /* role that device used to have in the
 126                                          * array and could again if we did a partial
 127                                          * resync from the bitmap
 128                                          */
 129
 130         atomic_t        nr_pending;     /* number of pending requests.
 131                                          * only maintained for arrays that
 132                                          * support hot removal
 133                                          */
 134 };
 135
 136 typedef struct mdk_personality_s mdk_personality_t;
 137
 138 struct mddev_s
 139 {
 140         void                            *private;
 141         mdk_personality_t               *pers;
 142         dev_t                           unit;
 143         int                             md_minor;
 144         struct list_head                disks;
 145         int                             sb_dirty;
 146         int                             ro;
 147
 148         struct gendisk                  *gendisk;
 149
 150         /* Superblock information */
 151         int                             major_version,
 152                                         minor_version,
 153                                         patch_version;
 154         int                             persistent;
 155         int                             chunk_size;
 156         time_t                          ctime, utime;
 157         int                             level, layout;
 158         int                             raid_disks;
 159         int                             max_disks;
 160         sector_t                        size; /* used size of component devices */
 161         sector_t                        array_size; /* exported array size */
 162         __u64                           events;
 163
 164         char                            uuid[16];
 165
 166         struct mdk_thread_s             *thread;        /* management thread */
 167         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
 168         sector_t                        curr_resync;    /* blocks scheduled */
 169         unsigned long                   resync_mark;    /* a recent timestamp */
 170         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
 171
 172         sector_t                        resync_max_sectors; /* may be set by personality */
 173         /* recovery/resync flags
 174          * NEEDED:   we might need to start a resync/recover
 175          * RUNNING:  a thread is running, or about to be started
 176          * SYNC:     actually doing a resync, not a recovery
 177          * ERR:      and IO error was detected - abort the resync/recovery
 178          * INTR:     someone requested a (clean) early abort.
 179          * DONE:     thread is done and is waiting to be reaped
 180          */
 181 #define MD_RECOVERY_RUNNING     0
 182 #define MD_RECOVERY_SYNC        1
 183 #define MD_RECOVERY_ERR         2
 184 #define MD_RECOVERY_INTR        3
 185 #define MD_RECOVERY_DONE        4
 186 #define MD_RECOVERY_NEEDED      5
 187         unsigned long                   recovery;
 188
 189         int                             in_sync;        /* know to not need resync */
 190         struct semaphore                reconfig_sem;
 191         atomic_t                        active;
 192
 193         int                             changed;        /* true if we might need to reread partition info */
 194         int                             degraded;       /* whether md should consider
 195                                                          * adding a spare
 196                                                          */
 197
 198         atomic_t                        recovery_active; /* blocks scheduled, but not written */
 199         wait_queue_head_t               recovery_wait;
 200         sector_t                        recovery_cp;
 201
 202         spinlock_t                      write_lock;
 203         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
 204         atomic_t                        pending_writes; /* number of active superblock writes */
 205
 206         unsigned int                    safemode;       /* if set, update "clean" superblock
 207                                                          * when no writes pending.
 208                                                          */
 209         unsigned int                    safemode_delay;
 210         struct timer_list               safemode_timer;
 211         atomic_t                        writes_pending;
 212         request_queue_t                 *queue; /* for plugging ... */
 213
 214         atomic_t                        write_behind; /* outstanding async IO */
 215         unsigned int                    max_write_behind; /* 0 = sync */
 216
 217         struct bitmap                   *bitmap; /* the bitmap for the device */
 218         struct file                     *bitmap_file; /* the bitmap file */
 219         long                            bitmap_offset; /* offset from superblock of
 220                                                         * start of bitmap. May be
 221                                                         * negative, but not '0'
 222                                                         */
 223         long                            default_bitmap_offset; /* this is the offset to use when
 224                                                                 * hot-adding a bitmap.  It should
 225                                                                 * eventually be settable by sysfs.
 226                                                                 */
 227
 228         struct list_head                all_mddevs;
 229 };
 230
 231
 232 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
 233 {
 234         int faulty = rdev->faulty;
 235         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
 236                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 237 }
 238
 239 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
 240 {
 241         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 242 }
 243
 244 struct mdk_personality_s
 245 {
 246         char *name;
 247         struct module *owner;
 248         int (*make_request)(request_queue_t *q, struct bio *bio);
 249         int (*run)(mddev_t *mddev);
 250         int (*stop)(mddev_t *mddev);
 251         void (*status)(struct seq_file *seq, mddev_t *mddev);
 252         /* error_handler must set ->faulty and clear ->in_sync
 253          * if appropriate, and should abort recovery if needed
 254          */
 255         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
 256         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
 257         int (*hot_remove_disk) (mddev_t *mddev, int number);
 258         int (*spare_active) (mddev_t *mddev);
 259         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
 260         int (*resize) (mddev_t *mddev, sector_t sectors);
 261         int (*reshape) (mddev_t *mddev, int raid_disks);
 262         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
 263         /* quiesce moves between quiescence states
 264          * 0 - fully active
 265          * 1 - no new requests allowed
 266          * others - reserved
 267          */
 268         void (*quiesce) (mddev_t *mddev, int state);
 269 };
 270
 271
 272 static inline char * mdname (mddev_t * mddev)
 273 {
 274         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
 275 }
 276
 277 extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr);
 278
 279 /*
 280  * iterates through some rdev ringlist. It's safe to remove the
 281  * current 'rdev'. Dont touch 'tmp' though.
 282  */
 283 #define ITERATE_RDEV_GENERIC(head,rdev,tmp)                             \
 284                                                                         \
 285         for ((tmp) = (head).next;                                       \
 286                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
 287                         (tmp) = (tmp)->next, (tmp)->prev != &(head)     \
 288                 ; )
 289 /*
 290  * iterates through the 'same array disks' ringlist
 291  */
 292 #define ITERATE_RDEV(mddev,rdev,tmp)                                    \
 293         ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
 294
 295 /*
 296  * Iterates through 'pending RAID disks'
 297  */
 298 #define ITERATE_RDEV_PENDING(rdev,tmp)                                  \
 299         ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
 300
 301 typedef struct mdk_thread_s {
 302         void                    (*run) (mddev_t *mddev);
 303         mddev_t                 *mddev;
 304         wait_queue_head_t       wqueue;
 305         unsigned long           flags;
 306         struct completion       *event;
 307         struct task_struct      *tsk;
 308         unsigned long           timeout;
 309         const char              *name;
 310 } mdk_thread_t;
 311
 312 #define THREAD_WAKEUP  0
 313
 314 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
 315 do {                                                                    \
 316         wait_queue_t __wait;                                            \
 317         init_waitqueue_entry(&__wait, current);                         \
 318                                                                         \
 319         add_wait_queue(&wq, &__wait);                                   \
 320         for (;;) {                                                      \
 321                 set_current_state(TASK_UNINTERRUPTIBLE);                \
 322                 if (condition)                                          \
 323                         break;                                          \
 324                 spin_unlock_irq(&lock);                                 \
 325                 cmd;                                                    \
 326                 schedule();                                             \
 327                 spin_lock_irq(&lock);                                   \
 328         }                                                               \
 329         current->state = TASK_RUNNING;                                  \
 330         remove_wait_queue(&wq, &__wait);                                \
 331 } while (0)
 332
 333 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
 334 do {                                                                    \
 335         if (condition)                                                  \
 336                 break;                                                  \
 337         __wait_event_lock_irq(wq, condition, lock, cmd);                \
 338 } while (0)
 339
 340 #endif
 341