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 /* Currently this must fix in an 'int' */
  44 #define MAX_CHUNK_SIZE (1<<30)
  45
  46 /*
  47  * MD's 'extended' device
  48  */
  49 struct mdk_rdev_s
  50 {
  51         struct list_head same_set;      /* RAID devices within the same set */
  52
  53         sector_t size;                  /* Device size (in blocks) */
  54         mddev_t *mddev;                 /* RAID array if running */
  55         unsigned long last_events;      /* IO event timestamp */
  56
  57         struct block_device *bdev;      /* block device handle */
  58
  59         struct page     *sb_page;
  60         int             sb_loaded;
  61         sector_t        data_offset;    /* start of data in array */
  62         sector_t        sb_offset;
  63         int             sb_size;        /* bytes in the superblock */
  64         int             preferred_minor;        /* autorun support */
  65
  66         struct kobject  kobj;
  67
  68         /* A device can be in one of three states based on two flags:
  69          * Not working:   faulty==1 in_sync==0
  70          * Fully working: faulty==0 in_sync==1
  71          * Working, but not
  72          * in sync with array
  73          *                faulty==0 in_sync==0
  74          *
  75          * It can never have faulty==1, in_sync==1
  76          * This reduces the burden of testing multiple flags in many cases
  77          */
  78
  79         unsigned long   flags;
  80 #define Faulty          1               /* device is known to have a fault */
  81 #define In_sync         2               /* device is in_sync with rest of array */
  82 #define WriteMostly     4               /* Avoid reading if at all possible */
  83 #define BarriersNotsupp 5               /* BIO_RW_BARRIER is not supported */
  84
  85         int desc_nr;                    /* descriptor index in the superblock */
  86         int raid_disk;                  /* role of device in array */
  87         int saved_raid_disk;            /* role that device used to have in the
  88                                          * array and could again if we did a partial
  89                                          * resync from the bitmap
  90                                          */
  91         sector_t        recovery_offset;/* If this device has been partially
  92                                          * recovered, this is where we were
  93                                          * up to.
  94                                          */
  95
  96         atomic_t        nr_pending;     /* number of pending requests.
  97                                          * only maintained for arrays that
  98                                          * support hot removal
  99                                          */
 100         atomic_t        read_errors;    /* number of consecutive read errors that
 101                                          * we have tried to ignore.
 102                                          */
 103         atomic_t        corrected_errors; /* number of corrected read errors,
 104                                            * for reporting to userspace and storing
 105                                            * in superblock.
 106                                            */
 107 };
 108
 109 struct mddev_s
 110 {
 111         void                            *private;
 112         struct mdk_personality          *pers;
 113         dev_t                           unit;
 114         int                             md_minor;
 115         struct list_head                disks;
 116         int                             sb_dirty;
 117         int                             ro;
 118
 119         struct gendisk                  *gendisk;
 120
 121         struct kobject                  kobj;
 122
 123         /* Superblock information */
 124         int                             major_version,
 125                                         minor_version,
 126                                         patch_version;
 127         int                             persistent;
 128         int                             chunk_size;
 129         time_t                          ctime, utime;
 130         int                             level, layout;
 131         char                            clevel[16];
 132         int                             raid_disks;
 133         int                             max_disks;
 134         sector_t                        size; /* used size of component devices */
 135         sector_t                        array_size; /* exported array size */
 136         __u64                           events;
 137
 138         char                            uuid[16];
 139
 140         /* If the array is being reshaped, we need to record the
 141          * new shape and an indication of where we are up to.
 142          * This is written to the superblock.
 143          * If reshape_position is MaxSector, then no reshape is happening (yet).
 144          */
 145         sector_t                        reshape_position;
 146         int                             delta_disks, new_level, new_layout, new_chunk;
 147
 148         struct mdk_thread_s             *thread;        /* management thread */
 149         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
 150         sector_t                        curr_resync;    /* blocks scheduled */
 151         unsigned long                   resync_mark;    /* a recent timestamp */
 152         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
 153
 154         sector_t                        resync_max_sectors; /* may be set by personality */
 155
 156         sector_t                        resync_mismatches; /* count of sectors where
 157                                                             * parity/replica mismatch found
 158                                                             */
 159
 160         /* allow user-space to request suspension of IO to regions of the array */
 161         sector_t                        suspend_lo;
 162         sector_t                        suspend_hi;
 163         /* if zero, use the system-wide default */
 164         int                             sync_speed_min;
 165         int                             sync_speed_max;
 166
 167         int                             ok_start_degraded;
 168         /* recovery/resync flags
 169          * NEEDED:   we might need to start a resync/recover
 170          * RUNNING:  a thread is running, or about to be started
 171          * SYNC:     actually doing a resync, not a recovery
 172          * ERR:      and IO error was detected - abort the resync/recovery
 173          * INTR:     someone requested a (clean) early abort.
 174          * DONE:     thread is done and is waiting to be reaped
 175          * REQUEST:  user-space has requested a sync (used with SYNC)
 176          * CHECK:    user-space request for for check-only, no repair
 177          * RESHAPE:  A reshape is happening
 178          *
 179          * If neither SYNC or RESHAPE are set, then it is a recovery.
 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 #define MD_RECOVERY_REQUESTED   6
 188 #define MD_RECOVERY_CHECK       7
 189 #define MD_RECOVERY_RESHAPE     8
 190 #define MD_RECOVERY_FROZEN      9
 191
 192         unsigned long                   recovery;
 193
 194         int                             in_sync;        /* know to not need resync */
 195         struct mutex                    reconfig_mutex;
 196         atomic_t                        active;
 197
 198         int                             changed;        /* true if we might need to reread partition info */
 199         int                             degraded;       /* whether md should consider
 200                                                          * adding a spare
 201                                                          */
 202         int                             barriers_work;  /* initialised to true, cleared as soon
 203                                                          * as a barrier request to slave
 204                                                          * fails.  Only supported
 205                                                          */
 206         struct bio                      *biolist;       /* bios that need to be retried
 207                                                          * because BIO_RW_BARRIER is not supported
 208                                                          */
 209
 210         atomic_t                        recovery_active; /* blocks scheduled, but not written */
 211         wait_queue_head_t               recovery_wait;
 212         sector_t                        recovery_cp;
 213
 214         spinlock_t                      write_lock;
 215         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
 216         atomic_t                        pending_writes; /* number of active superblock writes */
 217
 218         unsigned int                    safemode;       /* if set, update "clean" superblock
 219                                                          * when no writes pending.
 220                                                          */
 221         unsigned int                    safemode_delay;
 222         struct timer_list               safemode_timer;
 223         atomic_t                        writes_pending;
 224         request_queue_t                 *queue; /* for plugging ... */
 225
 226         atomic_t                        write_behind; /* outstanding async IO */
 227         unsigned int                    max_write_behind; /* 0 = sync */
 228
 229         struct bitmap                   *bitmap; /* the bitmap for the device */
 230         struct file                     *bitmap_file; /* the bitmap file */
 231         long                            bitmap_offset; /* offset from superblock of
 232                                                         * start of bitmap. May be
 233                                                         * negative, but not '0'
 234                                                         */
 235         long                            default_bitmap_offset; /* this is the offset to use when
 236                                                                 * hot-adding a bitmap.  It should
 237                                                                 * eventually be settable by sysfs.
 238                                                                 */
 239
 240         struct list_head                all_mddevs;
 241 };
 242
 243
 244 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
 245 {
 246         int faulty = test_bit(Faulty, &rdev->flags);
 247         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
 248                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 249 }
 250
 251 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
 252 {
 253         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 254 }
 255
 256 struct mdk_personality
 257 {
 258         char *name;
 259         int level;
 260         struct list_head list;
 261         struct module *owner;
 262         int (*make_request)(request_queue_t *q, struct bio *bio);
 263         int (*run)(mddev_t *mddev);
 264         int (*stop)(mddev_t *mddev);
 265         void (*status)(struct seq_file *seq, mddev_t *mddev);
 266         /* error_handler must set ->faulty and clear ->in_sync
 267          * if appropriate, and should abort recovery if needed
 268          */
 269         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
 270         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
 271         int (*hot_remove_disk) (mddev_t *mddev, int number);
 272         int (*spare_active) (mddev_t *mddev);
 273         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
 274         int (*resize) (mddev_t *mddev, sector_t sectors);
 275         int (*check_reshape) (mddev_t *mddev);
 276         int (*start_reshape) (mddev_t *mddev);
 277         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
 278         /* quiesce moves between quiescence states
 279          * 0 - fully active
 280          * 1 - no new requests allowed
 281          * others - reserved
 282          */
 283         void (*quiesce) (mddev_t *mddev, int state);
 284 };
 285
 286
 287 struct md_sysfs_entry {
 288         struct attribute attr;
 289         ssize_t (*show)(mddev_t *, char *);
 290         ssize_t (*store)(mddev_t *, const char *, size_t);
 291 };
 292
 293
 294 static inline char * mdname (mddev_t * mddev)
 295 {
 296         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
 297 }
 298
 299 /*
 300  * iterates through some rdev ringlist. It's safe to remove the
 301  * current 'rdev'. Dont touch 'tmp' though.
 302  */
 303 #define ITERATE_RDEV_GENERIC(head,rdev,tmp)                             \
 304                                                                         \
 305         for ((tmp) = (head).next;                                       \
 306                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
 307                         (tmp) = (tmp)->next, (tmp)->prev != &(head)     \
 308                 ; )
 309 /*
 310  * iterates through the 'same array disks' ringlist
 311  */
 312 #define ITERATE_RDEV(mddev,rdev,tmp)                                    \
 313         ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
 314
 315 /*
 316  * Iterates through 'pending RAID disks'
 317  */
 318 #define ITERATE_RDEV_PENDING(rdev,tmp)                                  \
 319         ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
 320
 321 typedef struct mdk_thread_s {
 322         void                    (*run) (mddev_t *mddev);
 323         mddev_t                 *mddev;
 324         wait_queue_head_t       wqueue;
 325         unsigned long           flags;
 326         struct task_struct      *tsk;
 327         unsigned long           timeout;
 328 } mdk_thread_t;
 329
 330 #define THREAD_WAKEUP  0
 331
 332 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
 333 do {                                                                    \
 334         wait_queue_t __wait;                                            \
 335         init_waitqueue_entry(&__wait, current);                         \
 336                                                                         \
 337         add_wait_queue(&wq, &__wait);                                   \
 338         for (;;) {                                                      \
 339                 set_current_state(TASK_UNINTERRUPTIBLE);                \
 340                 if (condition)                                          \
 341                         break;                                          \
 342                 spin_unlock_irq(&lock);                                 \
 343                 cmd;                                                    \
 344                 schedule();                                             \
 345                 spin_lock_irq(&lock);                                   \
 346         }                                                               \
 347         current->state = TASK_RUNNING;                                  \
 348         remove_wait_queue(&wq, &__wait);                                \
 349 } while (0)
 350
 351 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
 352 do {                                                                    \
 353         if (condition)                                                  \
 354                 break;                                                  \
 355         __wait_event_lock_irq(wq, condition, lock, cmd);                \
 356 } while (0)
 357
 358 static inline void safe_put_page(struct page *p)
 359 {
 360         if (p) put_page(p);
 361 }
 362
 363 #endif
 364