git.oblomov.eu Git - linux-2.6/blob - 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 LEVEL_MULTIPATH         (-4)
  22 #define LEVEL_LINEAR            (-1)
  23 #define LEVEL_FAULTY            (-5)
  24
  25 /* we need a value for 'no level specified' and 0
  26  * means 'raid0', so we need something else.  This is
  27  * for internal use only
  28  */
  29 #define LEVEL_NONE              (-1000000)
  30
  31 #define MaxSector (~(sector_t)0)
  32 #define MD_THREAD_NAME_MAX 14
  33
  34 typedef struct mddev_s mddev_t;
  35 typedef struct mdk_rdev_s mdk_rdev_t;
  36
  37 #define MAX_MD_DEVS  256        /* Max number of md dev */
  38
  39 /*
  40  * options passed in raidrun:
  41  */
  42
  43 #define MAX_CHUNK_SIZE (4096*1024)
  44
  45 /*
  46  * MD's 'extended' device
  47  */
  48 struct mdk_rdev_s
  49 {
  50         struct list_head same_set;      /* RAID devices within the same set */
  51
  52         sector_t size;                  /* Device size (in blocks) */
  53         mddev_t *mddev;                 /* RAID array if running */
  54         unsigned long last_events;      /* IO event timestamp */
  55
  56         struct block_device *bdev;      /* block device handle */
  57
  58         struct page     *sb_page;
  59         int             sb_loaded;
  60         sector_t        data_offset;    /* start of data in array */
  61         sector_t        sb_offset;
  62         int             sb_size;        /* bytes in the superblock */
  63         int             preferred_minor;        /* autorun support */
  64
  65         struct kobject  kobj;
  66
  67         /* A device can be in one of three states based on two flags:
  68          * Not working:   faulty==1 in_sync==0
  69          * Fully working: faulty==0 in_sync==1
  70          * Working, but not
  71          * in sync with array
  72          *                faulty==0 in_sync==0
  73          *
  74          * It can never have faulty==1, in_sync==1
  75          * This reduces the burden of testing multiple flags in many cases
  76          */
  77
  78         unsigned long   flags;
  79 #define Faulty          1               /* device is known to have a fault */
  80 #define In_sync         2               /* device is in_sync with rest of array */
  81 #define WriteMostly     4               /* Avoid reading if at all possible */
  82 #define BarriersNotsupp 5               /* BIO_RW_BARRIER is not supported */
  83
  84         int desc_nr;                    /* descriptor index in the superblock */
  85         int raid_disk;                  /* role of device in array */
  86         int saved_raid_disk;            /* role that device used to have in the
  87                                          * array and could again if we did a partial
  88                                          * resync from the bitmap
  89                                          */
  90
  91         atomic_t        nr_pending;     /* number of pending requests.
  92                                          * only maintained for arrays that
  93                                          * support hot removal
  94                                          */
  95         atomic_t        read_errors;    /* number of consecutive read errors that
  96                                          * we have tried to ignore.
  97                                          */
  98         atomic_t        corrected_errors; /* number of corrected read errors,
  99                                            * for reporting to userspace and storing
 100                                            * in superblock.
 101                                            */
 102 };
 103
 104 struct mddev_s
 105 {
 106         void                            *private;
 107         struct mdk_personality          *pers;
 108         dev_t                           unit;
 109         int                             md_minor;
 110         struct list_head                disks;
 111         int                             sb_dirty;
 112         int                             ro;
 113
 114         struct gendisk                  *gendisk;
 115
 116         struct kobject                  kobj;
 117
 118         /* Superblock information */
 119         int                             major_version,
 120                                         minor_version,
 121                                         patch_version;
 122         int                             persistent;
 123         int                             chunk_size;
 124         time_t                          ctime, utime;
 125         int                             level, layout;
 126         char                            clevel[16];
 127         int                             raid_disks;
 128         int                             max_disks;
 129         sector_t                        size; /* used size of component devices */
 130         sector_t                        array_size; /* exported array size */
 131         __u64                           events;
 132
 133         char                            uuid[16];
 134
 135         /* If the array is being reshaped, we need to record the
 136          * new shape and an indication of where we are up to.
 137          * This is written to the superblock.
 138          * If reshape_position is MaxSector, then no reshape is happening (yet).
 139          */
 140         sector_t                        reshape_position;
 141         int                             delta_disks, new_level, new_layout, new_chunk;
 142
 143         struct mdk_thread_s             *thread;        /* management thread */
 144         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
 145         sector_t                        curr_resync;    /* blocks scheduled */
 146         unsigned long                   resync_mark;    /* a recent timestamp */
 147         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
 148
 149         sector_t                        resync_max_sectors; /* may be set by personality */
 150
 151         sector_t                        resync_mismatches; /* count of sectors where
 152                                                             * parity/replica mismatch found
 153                                                             */
 154
 155         /* allow user-space to request suspension of IO to regions of the array */
 156         sector_t                        suspend_lo;
 157         sector_t                        suspend_hi;
 158         /* if zero, use the system-wide default */
 159         int                             sync_speed_min;
 160         int                             sync_speed_max;
 161
 162         int                             ok_start_degraded;
 163         /* recovery/resync flags
 164          * NEEDED:   we might need to start a resync/recover
 165          * RUNNING:  a thread is running, or about to be started
 166          * SYNC:     actually doing a resync, not a recovery
 167          * ERR:      and IO error was detected - abort the resync/recovery
 168          * INTR:     someone requested a (clean) early abort.
 169          * DONE:     thread is done and is waiting to be reaped
 170          * REQUEST:  user-space has requested a sync (used with SYNC)
 171          * CHECK:    user-space request for for check-only, no repair
 172          * RESHAPE:  A reshape is happening
 173          *
 174          * If neither SYNC or RESHAPE are set, then it is a recovery.
 175          */
 176 #define MD_RECOVERY_RUNNING     0
 177 #define MD_RECOVERY_SYNC        1
 178 #define MD_RECOVERY_ERR         2
 179 #define MD_RECOVERY_INTR        3
 180 #define MD_RECOVERY_DONE        4
 181 #define MD_RECOVERY_NEEDED      5
 182 #define MD_RECOVERY_REQUESTED   6
 183 #define MD_RECOVERY_CHECK       7
 184 #define MD_RECOVERY_RESHAPE     8
 185         unsigned long                   recovery;
 186
 187         int                             in_sync;        /* know to not need resync */
 188         struct mutex                    reconfig_mutex;
 189         atomic_t                        active;
 190
 191         int                             changed;        /* true if we might need to reread partition info */
 192         int                             degraded;       /* whether md should consider
 193                                                          * adding a spare
 194                                                          */
 195         int                             barriers_work;  /* initialised to true, cleared as soon
 196                                                          * as a barrier request to slave
 197                                                          * fails.  Only supported
 198                                                          */
 199         struct bio                      *biolist;       /* bios that need to be retried
 200                                                          * because BIO_RW_BARRIER is not supported
 201                                                          */
 202
 203         atomic_t                        recovery_active; /* blocks scheduled, but not written */
 204         wait_queue_head_t               recovery_wait;
 205         sector_t                        recovery_cp;
 206
 207         spinlock_t                      write_lock;
 208         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
 209         atomic_t                        pending_writes; /* number of active superblock writes */
 210
 211         unsigned int                    safemode;       /* if set, update "clean" superblock
 212                                                          * when no writes pending.
 213                                                          */
 214         unsigned int                    safemode_delay;
 215         struct timer_list               safemode_timer;
 216         atomic_t                        writes_pending;
 217         request_queue_t                 *queue; /* for plugging ... */
 218
 219         atomic_t                        write_behind; /* outstanding async IO */
 220         unsigned int                    max_write_behind; /* 0 = sync */
 221
 222         struct bitmap                   *bitmap; /* the bitmap for the device */
 223         struct file                     *bitmap_file; /* the bitmap file */
 224         long                            bitmap_offset; /* offset from superblock of
 225                                                         * start of bitmap. May be
 226                                                         * negative, but not '0'
 227                                                         */
 228         long                            default_bitmap_offset; /* this is the offset to use when
 229                                                                 * hot-adding a bitmap.  It should
 230                                                                 * eventually be settable by sysfs.
 231                                                                 */
 232
 233         struct list_head                all_mddevs;
 234 };
 235
 236
 237 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
 238 {
 239         int faulty = test_bit(Faulty, &rdev->flags);
 240         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
 241                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 242 }
 243
 244 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
 245 {
 246         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 247 }
 248
 249 struct mdk_personality
 250 {
 251         char *name;
 252         int level;
 253         struct list_head list;
 254         struct module *owner;
 255         int (*make_request)(request_queue_t *q, struct bio *bio);
 256         int (*run)(mddev_t *mddev);
 257         int (*stop)(mddev_t *mddev);
 258         void (*status)(struct seq_file *seq, mddev_t *mddev);
 259         /* error_handler must set ->faulty and clear ->in_sync
 260          * if appropriate, and should abort recovery if needed
 261          */
 262         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
 263         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
 264         int (*hot_remove_disk) (mddev_t *mddev, int number);
 265         int (*spare_active) (mddev_t *mddev);
 266         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
 267         int (*resize) (mddev_t *mddev, sector_t sectors);
 268         int (*check_reshape) (mddev_t *mddev);
 269         int (*start_reshape) (mddev_t *mddev);
 270         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
 271         /* quiesce moves between quiescence states
 272          * 0 - fully active
 273          * 1 - no new requests allowed
 274          * others - reserved
 275          */
 276         void (*quiesce) (mddev_t *mddev, int state);
 277 };
 278
 279
 280 struct md_sysfs_entry {
 281         struct attribute attr;
 282         ssize_t (*show)(mddev_t *, char *);
 283         ssize_t (*store)(mddev_t *, const char *, size_t);
 284 };
 285
 286
 287 static inline char * mdname (mddev_t * mddev)
 288 {
 289         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
 290 }
 291
 292 /*
 293  * iterates through some rdev ringlist. It's safe to remove the
 294  * current 'rdev'. Dont touch 'tmp' though.
 295  */
 296 #define ITERATE_RDEV_GENERIC(head,rdev,tmp)                             \
 297                                                                         \
 298         for ((tmp) = (head).next;                                       \
 299                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
 300                         (tmp) = (tmp)->next, (tmp)->prev != &(head)     \
 301                 ; )
 302 /*
 303  * iterates through the 'same array disks' ringlist
 304  */
 305 #define ITERATE_RDEV(mddev,rdev,tmp)                                    \
 306         ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
 307
 308 /*
 309  * Iterates through 'pending RAID disks'
 310  */
 311 #define ITERATE_RDEV_PENDING(rdev,tmp)                                  \
 312         ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
 313
 314 typedef struct mdk_thread_s {
 315         void                    (*run) (mddev_t *mddev);
 316         mddev_t                 *mddev;
 317         wait_queue_head_t       wqueue;
 318         unsigned long           flags;
 319         struct task_struct      *tsk;
 320         unsigned long           timeout;
 321 } mdk_thread_t;
 322
 323 #define THREAD_WAKEUP  0
 324
 325 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
 326 do {                                                                    \
 327         wait_queue_t __wait;                                            \
 328         init_waitqueue_entry(&__wait, current);                         \
 329                                                                         \
 330         add_wait_queue(&wq, &__wait);                                   \
 331         for (;;) {                                                      \
 332                 set_current_state(TASK_UNINTERRUPTIBLE);                \
 333                 if (condition)                                          \
 334                         break;                                          \
 335                 spin_unlock_irq(&lock);                                 \
 336                 cmd;                                                    \
 337                 schedule();                                             \
 338                 spin_lock_irq(&lock);                                   \
 339         }                                                               \
 340         current->state = TASK_RUNNING;                                  \
 341         remove_wait_queue(&wq, &__wait);                                \
 342 } while (0)
 343
 344 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
 345 do {                                                                    \
 346         if (condition)                                                  \
 347                 break;                                                  \
 348         __wait_event_lock_irq(wq, condition, lock, cmd);                \
 349 } while (0)
 350
 351 static inline void safe_put_page(struct page *p)
 352 {
 353         if (p) put_page(p);
 354 }
 355
 356 #endif
 357