drivers/md/md.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_MD_H
  16 #define _MD_MD_H
  17
  18 #include <linux/blkdev.h>
  19 #include <linux/kobject.h>
  20 #include <linux/list.h>
  21 #include <linux/mm.h>
  22 #include <linux/mutex.h>
  23 #include <linux/timer.h>
  24 #include <linux/wait.h>
  25 #include <linux/workqueue.h>
  26
  27 #define MaxSector (~(sector_t)0)
  28
  29 typedef struct mddev_s mddev_t;
  30 typedef struct mdk_rdev_s mdk_rdev_t;
  31
  32 /*
  33  * MD's 'extended' device
  34  */
  35 struct mdk_rdev_s
  36 {
  37         struct list_head same_set;      /* RAID devices within the same set */
  38
  39         sector_t sectors;               /* Device size (in 512bytes sectors) */
  40         mddev_t *mddev;                 /* RAID array if running */
  41         int last_events;                /* IO event timestamp */
  42
  43         struct block_device *bdev;      /* block device handle */
  44
  45         struct page     *sb_page;
  46         int             sb_loaded;
  47         __u64           sb_events;
  48         sector_t        data_offset;    /* start of data in array */
  49         sector_t        sb_start;       /* offset of the super block (in 512byte sectors) */
  50         int             sb_size;        /* bytes in the superblock */
  51         int             preferred_minor;        /* autorun support */
  52
  53         struct kobject  kobj;
  54
  55         /* A device can be in one of three states based on two flags:
  56          * Not working:   faulty==1 in_sync==0
  57          * Fully working: faulty==0 in_sync==1
  58          * Working, but not
  59          * in sync with array
  60          *                faulty==0 in_sync==0
  61          *
  62          * It can never have faulty==1, in_sync==1
  63          * This reduces the burden of testing multiple flags in many cases
  64          */
  65
  66         unsigned long   flags;
  67 #define Faulty          1               /* device is known to have a fault */
  68 #define In_sync         2               /* device is in_sync with rest of array */
  69 #define WriteMostly     4               /* Avoid reading if at all possible */
  70 #define BarriersNotsupp 5               /* BIO_RW_BARRIER is not supported */
  71 #define AllReserved     6               /* If whole device is reserved for
  72                                          * one array */
  73 #define AutoDetected    7               /* added by auto-detect */
  74 #define Blocked         8               /* An error occured on an externally
  75                                          * managed array, don't allow writes
  76                                          * until it is cleared */
  77 #define StateChanged    9               /* Faulty or Blocked has changed during
  78                                          * interrupt, so it needs to be
  79                                          * notified by the thread */
  80         wait_queue_head_t blocked_wait;
  81
  82         int desc_nr;                    /* descriptor index in the superblock */
  83         int raid_disk;                  /* role of device in array */
  84         int saved_raid_disk;            /* role that device used to have in the
  85                                          * array and could again if we did a partial
  86                                          * resync from the bitmap
  87                                          */
  88         sector_t        recovery_offset;/* If this device has been partially
  89                                          * recovered, this is where we were
  90                                          * up to.
  91                                          */
  92
  93         atomic_t        nr_pending;     /* number of pending requests.
  94                                          * only maintained for arrays that
  95                                          * support hot removal
  96                                          */
  97         atomic_t        read_errors;    /* number of consecutive read errors that
  98                                          * we have tried to ignore.
  99                                          */
 100         struct timespec last_read_error;        /* monotonic time since our
 101                                                  * last read error
 102                                                  */
 103         atomic_t        corrected_errors; /* number of corrected read errors,
 104                                            * for reporting to userspace and storing
 105                                            * in superblock.
 106                                            */
 107         struct work_struct del_work;    /* used for delayed sysfs removal */
 108
 109         struct sysfs_dirent *sysfs_state; /* handle for 'state'
 110                                            * sysfs entry */
 111 };
 112
 113 struct mddev_s
 114 {
 115         void                            *private;
 116         struct mdk_personality          *pers;
 117         dev_t                           unit;
 118         int                             md_minor;
 119         struct list_head                disks;
 120         unsigned long                   flags;
 121 #define MD_CHANGE_DEVS  0       /* Some device status has changed */
 122 #define MD_CHANGE_CLEAN 1       /* transition to or from 'clean' */
 123 #define MD_CHANGE_PENDING 2     /* superblock update in progress */
 124
 125         int                             suspended;
 126         atomic_t                        active_io;
 127         int                             ro;
 128
 129         struct gendisk                  *gendisk;
 130
 131         struct kobject                  kobj;
 132         int                             hold_active;
 133 #define UNTIL_IOCTL     1
 134 #define UNTIL_STOP      2
 135
 136         /* Superblock information */
 137         int                             major_version,
 138                                         minor_version,
 139                                         patch_version;
 140         int                             persistent;
 141         int                             external;       /* metadata is
 142                                                          * managed externally */
 143         char                            metadata_type[17]; /* externally set*/
 144         int                             chunk_sectors;
 145         time_t                          ctime, utime;
 146         int                             level, layout;
 147         char                            clevel[16];
 148         int                             raid_disks;
 149         int                             max_disks;
 150         sector_t                        dev_sectors;    /* used size of
 151                                                          * component devices */
 152         sector_t                        array_sectors; /* exported array size */
 153         int                             external_size; /* size managed
 154                                                         * externally */
 155         __u64                           events;
 156
 157         char                            uuid[16];
 158
 159         /* If the array is being reshaped, we need to record the
 160          * new shape and an indication of where we are up to.
 161          * This is written to the superblock.
 162          * If reshape_position is MaxSector, then no reshape is happening (yet).
 163          */
 164         sector_t                        reshape_position;
 165         int                             delta_disks, new_level, new_layout;
 166         int                             new_chunk_sectors;
 167
 168         struct mdk_thread_s             *thread;        /* management thread */
 169         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
 170         sector_t                        curr_resync;    /* last block scheduled */
 171         /* As resync requests can complete out of order, we cannot easily track
 172          * how much resync has been completed.  So we occasionally pause until
 173          * everything completes, then set curr_resync_completed to curr_resync.
 174          * As such it may be well behind the real resync mark, but it is a value
 175          * we are certain of.
 176          */
 177         sector_t                        curr_resync_completed;
 178         unsigned long                   resync_mark;    /* a recent timestamp */
 179         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
 180         sector_t                        curr_mark_cnt; /* blocks scheduled now */
 181
 182         sector_t                        resync_max_sectors; /* may be set by personality */
 183
 184         sector_t                        resync_mismatches; /* count of sectors where
 185                                                             * parity/replica mismatch found
 186                                                             */
 187
 188         /* allow user-space to request suspension of IO to regions of the array */
 189         sector_t                        suspend_lo;
 190         sector_t                        suspend_hi;
 191         /* if zero, use the system-wide default */
 192         int                             sync_speed_min;
 193         int                             sync_speed_max;
 194
 195         /* resync even though the same disks are shared among md-devices */
 196         int                             parallel_resync;
 197
 198         int                             ok_start_degraded;
 199         /* recovery/resync flags
 200          * NEEDED:   we might need to start a resync/recover
 201          * RUNNING:  a thread is running, or about to be started
 202          * SYNC:     actually doing a resync, not a recovery
 203          * RECOVER:  doing recovery, or need to try it.
 204          * INTR:     resync needs to be aborted for some reason
 205          * DONE:     thread is done and is waiting to be reaped
 206          * REQUEST:  user-space has requested a sync (used with SYNC)
 207          * CHECK:    user-space request for check-only, no repair
 208          * RESHAPE:  A reshape is happening
 209          *
 210          * If neither SYNC or RESHAPE are set, then it is a recovery.
 211          */
 212 #define MD_RECOVERY_RUNNING     0
 213 #define MD_RECOVERY_SYNC        1
 214 #define MD_RECOVERY_RECOVER     2
 215 #define MD_RECOVERY_INTR        3
 216 #define MD_RECOVERY_DONE        4
 217 #define MD_RECOVERY_NEEDED      5
 218 #define MD_RECOVERY_REQUESTED   6
 219 #define MD_RECOVERY_CHECK       7
 220 #define MD_RECOVERY_RESHAPE     8
 221 #define MD_RECOVERY_FROZEN      9
 222
 223         unsigned long                   recovery;
 224         int                             recovery_disabled; /* if we detect that recovery
 225                                                             * will always fail, set this
 226                                                             * so we don't loop trying */
 227
 228         int                             in_sync;        /* know to not need resync */
 229         /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
 230          * that we are never stopping an array while it is open.
 231          * 'reconfig_mutex' protects all other reconfiguration.
 232          * These locks are separate due to conflicting interactions
 233          * with bdev->bd_mutex.
 234          * Lock ordering is:
 235          *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
 236          *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
 237          */
 238         struct mutex                    open_mutex;
 239         struct mutex                    reconfig_mutex;
 240         atomic_t                        active;         /* general refcount */
 241         atomic_t                        openers;        /* number of active opens */
 242
 243         int                             changed;        /* true if we might need to reread partition info */
 244         int                             degraded;       /* whether md should consider
 245                                                          * adding a spare
 246                                                          */
 247         int                             barriers_work;  /* initialised to true, cleared as soon
 248                                                          * as a barrier request to slave
 249                                                          * fails.  Only supported
 250                                                          */
 251         struct bio                      *biolist;       /* bios that need to be retried
 252                                                          * because BIO_RW_BARRIER is not supported
 253                                                          */
 254
 255         atomic_t                        recovery_active; /* blocks scheduled, but not written */
 256         wait_queue_head_t               recovery_wait;
 257         sector_t                        recovery_cp;
 258         sector_t                        resync_min;     /* user requested sync
 259                                                          * starts here */
 260         sector_t                        resync_max;     /* resync should pause
 261                                                          * when it gets here */
 262
 263         struct sysfs_dirent             *sysfs_state;   /* handle for 'array_state'
 264                                                          * file in sysfs.
 265                                                          */
 266         struct sysfs_dirent             *sysfs_action;  /* handle for 'sync_action' */
 267
 268         struct work_struct del_work;    /* used for delayed sysfs removal */
 269
 270         spinlock_t                      write_lock;
 271         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
 272         atomic_t                        pending_writes; /* number of active superblock writes */
 273
 274         unsigned int                    safemode;       /* if set, update "clean" superblock
 275                                                          * when no writes pending.
 276                                                          */
 277         unsigned int                    safemode_delay;
 278         struct timer_list               safemode_timer;
 279         atomic_t                        writes_pending;
 280         struct request_queue            *queue; /* for plugging ... */
 281
 282         atomic_t                        write_behind; /* outstanding async IO */
 283         unsigned int                    max_write_behind; /* 0 = sync */
 284
 285         struct bitmap                   *bitmap; /* the bitmap for the device */
 286         struct {
 287                 struct file             *file; /* the bitmap file */
 288                 loff_t                  offset; /* offset from superblock of
 289                                                  * start of bitmap. May be
 290                                                  * negative, but not '0'
 291                                                  * For external metadata, offset
 292                                                  * from start of device.
 293                                                  */
 294                 loff_t                  default_offset; /* this is the offset to use when
 295                                                          * hot-adding a bitmap.  It should
 296                                                          * eventually be settable by sysfs.
 297                                                          */
 298                 struct mutex            mutex;
 299                 unsigned long           chunksize;
 300                 unsigned long           daemon_sleep; /* how many seconds between updates? */
 301                 unsigned long           max_write_behind; /* write-behind mode */
 302                 int                     external;
 303         } bitmap_info;
 304
 305         atomic_t                        max_corr_read_errors; /* max read retries */
 306         struct list_head                all_mddevs;
 307
 308         /* Generic barrier handling.
 309          * If there is a pending barrier request, all other
 310          * writes are blocked while the devices are flushed.
 311          * The last to finish a flush schedules a worker to
 312          * submit the barrier request (without the barrier flag),
 313          * then submit more flush requests.
 314          */
 315         struct bio *barrier;
 316         atomic_t flush_pending;
 317         struct work_struct barrier_work;
 318 };
 319
 320
 321 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
 322 {
 323         int faulty = test_bit(Faulty, &rdev->flags);
 324         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
 325                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 326 }
 327
 328 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
 329 {
 330         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 331 }
 332
 333 struct mdk_personality
 334 {
 335         char *name;
 336         int level;
 337         struct list_head list;
 338         struct module *owner;
 339         int (*make_request)(struct request_queue *q, struct bio *bio);
 340         int (*run)(mddev_t *mddev);
 341         int (*stop)(mddev_t *mddev);
 342         void (*status)(struct seq_file *seq, mddev_t *mddev);
 343         /* error_handler must set ->faulty and clear ->in_sync
 344          * if appropriate, and should abort recovery if needed
 345          */
 346         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
 347         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
 348         int (*hot_remove_disk) (mddev_t *mddev, int number);
 349         int (*spare_active) (mddev_t *mddev);
 350         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
 351         int (*resize) (mddev_t *mddev, sector_t sectors);
 352         sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
 353         int (*check_reshape) (mddev_t *mddev);
 354         int (*start_reshape) (mddev_t *mddev);
 355         void (*finish_reshape) (mddev_t *mddev);
 356         /* quiesce moves between quiescence states
 357          * 0 - fully active
 358          * 1 - no new requests allowed
 359          * others - reserved
 360          */
 361         void (*quiesce) (mddev_t *mddev, int state);
 362         /* takeover is used to transition an array from one
 363          * personality to another.  The new personality must be able
 364          * to handle the data in the current layout.
 365          * e.g. 2drive raid1 -> 2drive raid5
 366          *      ndrive raid5 -> degraded n+1drive raid6 with special layout
 367          * If the takeover succeeds, a new 'private' structure is returned.
 368          * This needs to be installed and then ->run used to activate the
 369          * array.
 370          */
 371         void *(*takeover) (mddev_t *mddev);
 372 };
 373
 374
 375 struct md_sysfs_entry {
 376         struct attribute attr;
 377         ssize_t (*show)(mddev_t *, char *);
 378         ssize_t (*store)(mddev_t *, const char *, size_t);
 379 };
 380 extern struct attribute_group md_bitmap_group;
 381
 382 static inline char * mdname (mddev_t * mddev)
 383 {
 384         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
 385 }
 386
 387 /*
 388  * iterates through some rdev ringlist. It's safe to remove the
 389  * current 'rdev'. Dont touch 'tmp' though.
 390  */
 391 #define rdev_for_each_list(rdev, tmp, head)                             \
 392         list_for_each_entry_safe(rdev, tmp, head, same_set)
 393
 394 /*
 395  * iterates through the 'same array disks' ringlist
 396  */
 397 #define rdev_for_each(rdev, tmp, mddev)                         \
 398         list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
 399
 400 #define rdev_for_each_rcu(rdev, mddev)                          \
 401         list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
 402
 403 typedef struct mdk_thread_s {
 404         void                    (*run) (mddev_t *mddev);
 405         mddev_t                 *mddev;
 406         wait_queue_head_t       wqueue;
 407         unsigned long           flags;
 408         struct task_struct      *tsk;
 409         unsigned long           timeout;
 410 } mdk_thread_t;
 411
 412 #define THREAD_WAKEUP  0
 413
 414 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
 415 do {                                                                    \
 416         wait_queue_t __wait;                                            \
 417         init_waitqueue_entry(&__wait, current);                         \
 418                                                                         \
 419         add_wait_queue(&wq, &__wait);                                   \
 420         for (;;) {                                                      \
 421                 set_current_state(TASK_UNINTERRUPTIBLE);                \
 422                 if (condition)                                          \
 423                         break;                                          \
 424                 spin_unlock_irq(&lock);                                 \
 425                 cmd;                                                    \
 426                 schedule();                                             \
 427                 spin_lock_irq(&lock);                                   \
 428         }                                                               \
 429         current->state = TASK_RUNNING;                                  \
 430         remove_wait_queue(&wq, &__wait);                                \
 431 } while (0)
 432
 433 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
 434 do {                                                                    \
 435         if (condition)                                                  \
 436                 break;                                                  \
 437         __wait_event_lock_irq(wq, condition, lock, cmd);                \
 438 } while (0)
 439
 440 static inline void safe_put_page(struct page *p)
 441 {
 442         if (p) put_page(p);
 443 }
 444
 445 extern int register_md_personality(struct mdk_personality *p);
 446 extern int unregister_md_personality(struct mdk_personality *p);
 447 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
 448                                 mddev_t *mddev, const char *name);
 449 extern void md_unregister_thread(mdk_thread_t *thread);
 450 extern void md_wakeup_thread(mdk_thread_t *thread);
 451 extern void md_check_recovery(mddev_t *mddev);
 452 extern void md_write_start(mddev_t *mddev, struct bio *bi);
 453 extern void md_write_end(mddev_t *mddev);
 454 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
 455 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
 456
 457 extern int mddev_congested(mddev_t *mddev, int bits);
 458 extern void md_barrier_request(mddev_t *mddev, struct bio *bio);
 459 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 460                            sector_t sector, int size, struct page *page);
 461 extern void md_super_wait(mddev_t *mddev);
 462 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 463                         struct page *page, int rw);
 464 extern void md_do_sync(mddev_t *mddev);
 465 extern void md_new_event(mddev_t *mddev);
 466 extern int md_allow_write(mddev_t *mddev);
 467 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
 468 extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
 469 extern int md_check_no_bitmap(mddev_t *mddev);
 470 extern int md_integrity_register(mddev_t *mddev);
 471 extern void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
 472 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
 473 extern void restore_bitmap_write_access(struct file *file);
 474
 475 #endif /* _MD_MD_H */