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 /* generic plugging support - like that provided with request_queue,
  33  * but does not require a request_queue
  34  */
  35 struct plug_handle {
  36         void                    (*unplug_fn)(struct plug_handle *);
  37         struct timer_list       unplug_timer;
  38         struct work_struct      unplug_work;
  39         unsigned long           unplug_flag;
  40 };
  41 #define PLUGGED_FLAG 1
  42 void plugger_init(struct plug_handle *plug,
  43                   void (*unplug_fn)(struct plug_handle *));
  44 void plugger_set_plug(struct plug_handle *plug);
  45 int plugger_remove_plug(struct plug_handle *plug);
  46 static inline void plugger_flush(struct plug_handle *plug)
  47 {
  48         del_timer_sync(&plug->unplug_timer);
  49         cancel_work_sync(&plug->unplug_work);
  50 }
  51
  52 /*
  53  * MD's 'extended' device
  54  */
  55 struct mdk_rdev_s
  56 {
  57         struct list_head same_set;      /* RAID devices within the same set */
  58
  59         sector_t sectors;               /* Device size (in 512bytes sectors) */
  60         mddev_t *mddev;                 /* RAID array if running */
  61         int last_events;                /* IO event timestamp */
  62
  63         struct block_device *bdev;      /* block device handle */
  64
  65         struct page     *sb_page;
  66         int             sb_loaded;
  67         __u64           sb_events;
  68         sector_t        data_offset;    /* start of data in array */
  69         sector_t        sb_start;       /* offset of the super block (in 512byte sectors) */
  70         int             sb_size;        /* bytes in the superblock */
  71         int             preferred_minor;        /* autorun support */
  72
  73         struct kobject  kobj;
  74
  75         /* A device can be in one of three states based on two flags:
  76          * Not working:   faulty==1 in_sync==0
  77          * Fully working: faulty==0 in_sync==1
  78          * Working, but not
  79          * in sync with array
  80          *                faulty==0 in_sync==0
  81          *
  82          * It can never have faulty==1, in_sync==1
  83          * This reduces the burden of testing multiple flags in many cases
  84          */
  85
  86         unsigned long   flags;
  87 #define Faulty          1               /* device is known to have a fault */
  88 #define In_sync         2               /* device is in_sync with rest of array */
  89 #define WriteMostly     4               /* Avoid reading if at all possible */
  90 #define BarriersNotsupp 5               /* REQ_HARDBARRIER is not supported */
  91 #define AllReserved     6               /* If whole device is reserved for
  92                                          * one array */
  93 #define AutoDetected    7               /* added by auto-detect */
  94 #define Blocked         8               /* An error occured on an externally
  95                                          * managed array, don't allow writes
  96                                          * until it is cleared */
  97         wait_queue_head_t blocked_wait;
  98
  99         int desc_nr;                    /* descriptor index in the superblock */
 100         int raid_disk;                  /* role of device in array */
 101         int new_raid_disk;              /* role that the device will have in
 102                                          * the array after a level-change completes.
 103                                          */
 104         int saved_raid_disk;            /* role that device used to have in the
 105                                          * array and could again if we did a partial
 106                                          * resync from the bitmap
 107                                          */
 108         sector_t        recovery_offset;/* If this device has been partially
 109                                          * recovered, this is where we were
 110                                          * up to.
 111                                          */
 112
 113         atomic_t        nr_pending;     /* number of pending requests.
 114                                          * only maintained for arrays that
 115                                          * support hot removal
 116                                          */
 117         atomic_t        read_errors;    /* number of consecutive read errors that
 118                                          * we have tried to ignore.
 119                                          */
 120         struct timespec last_read_error;        /* monotonic time since our
 121                                                  * last read error
 122                                                  */
 123         atomic_t        corrected_errors; /* number of corrected read errors,
 124                                            * for reporting to userspace and storing
 125                                            * in superblock.
 126                                            */
 127         struct work_struct del_work;    /* used for delayed sysfs removal */
 128
 129         struct sysfs_dirent *sysfs_state; /* handle for 'state'
 130                                            * sysfs entry */
 131 };
 132
 133 struct mddev_s
 134 {
 135         void                            *private;
 136         struct mdk_personality          *pers;
 137         dev_t                           unit;
 138         int                             md_minor;
 139         struct list_head                disks;
 140         unsigned long                   flags;
 141 #define MD_CHANGE_DEVS  0       /* Some device status has changed */
 142 #define MD_CHANGE_CLEAN 1       /* transition to or from 'clean' */
 143 #define MD_CHANGE_PENDING 2     /* switch from 'clean' to 'active' in progress */
 144
 145         int                             suspended;
 146         atomic_t                        active_io;
 147         int                             ro;
 148         int                             sysfs_active; /* set when sysfs deletes
 149                                                        * are happening, so run/
 150                                                        * takeover/stop are not safe
 151                                                        */
 152         int                             ready; /* See when safe to pass
 153                                                 * IO requests down */
 154         struct gendisk                  *gendisk;
 155
 156         struct kobject                  kobj;
 157         int                             hold_active;
 158 #define UNTIL_IOCTL     1
 159 #define UNTIL_STOP      2
 160
 161         /* Superblock information */
 162         int                             major_version,
 163                                         minor_version,
 164                                         patch_version;
 165         int                             persistent;
 166         int                             external;       /* metadata is
 167                                                          * managed externally */
 168         char                            metadata_type[17]; /* externally set*/
 169         int                             chunk_sectors;
 170         time_t                          ctime, utime;
 171         int                             level, layout;
 172         char                            clevel[16];
 173         int                             raid_disks;
 174         int                             max_disks;
 175         sector_t                        dev_sectors;    /* used size of
 176                                                          * component devices */
 177         sector_t                        array_sectors; /* exported array size */
 178         int                             external_size; /* size managed
 179                                                         * externally */
 180         __u64                           events;
 181         /* If the last 'event' was simply a clean->dirty transition, and
 182          * we didn't write it to the spares, then it is safe and simple
 183          * to just decrement the event count on a dirty->clean transition.
 184          * So we record that possibility here.
 185          */
 186         int                             can_decrease_events;
 187
 188         char                            uuid[16];
 189
 190         /* If the array is being reshaped, we need to record the
 191          * new shape and an indication of where we are up to.
 192          * This is written to the superblock.
 193          * If reshape_position is MaxSector, then no reshape is happening (yet).
 194          */
 195         sector_t                        reshape_position;
 196         int                             delta_disks, new_level, new_layout;
 197         int                             new_chunk_sectors;
 198
 199         struct mdk_thread_s             *thread;        /* management thread */
 200         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
 201         sector_t                        curr_resync;    /* last block scheduled */
 202         /* As resync requests can complete out of order, we cannot easily track
 203          * how much resync has been completed.  So we occasionally pause until
 204          * everything completes, then set curr_resync_completed to curr_resync.
 205          * As such it may be well behind the real resync mark, but it is a value
 206          * we are certain of.
 207          */
 208         sector_t                        curr_resync_completed;
 209         unsigned long                   resync_mark;    /* a recent timestamp */
 210         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
 211         sector_t                        curr_mark_cnt; /* blocks scheduled now */
 212
 213         sector_t                        resync_max_sectors; /* may be set by personality */
 214
 215         sector_t                        resync_mismatches; /* count of sectors where
 216                                                             * parity/replica mismatch found
 217                                                             */
 218
 219         /* allow user-space to request suspension of IO to regions of the array */
 220         sector_t                        suspend_lo;
 221         sector_t                        suspend_hi;
 222         /* if zero, use the system-wide default */
 223         int                             sync_speed_min;
 224         int                             sync_speed_max;
 225
 226         /* resync even though the same disks are shared among md-devices */
 227         int                             parallel_resync;
 228
 229         int                             ok_start_degraded;
 230         /* recovery/resync flags
 231          * NEEDED:   we might need to start a resync/recover
 232          * RUNNING:  a thread is running, or about to be started
 233          * SYNC:     actually doing a resync, not a recovery
 234          * RECOVER:  doing recovery, or need to try it.
 235          * INTR:     resync needs to be aborted for some reason
 236          * DONE:     thread is done and is waiting to be reaped
 237          * REQUEST:  user-space has requested a sync (used with SYNC)
 238          * CHECK:    user-space request for check-only, no repair
 239          * RESHAPE:  A reshape is happening
 240          *
 241          * If neither SYNC or RESHAPE are set, then it is a recovery.
 242          */
 243 #define MD_RECOVERY_RUNNING     0
 244 #define MD_RECOVERY_SYNC        1
 245 #define MD_RECOVERY_RECOVER     2
 246 #define MD_RECOVERY_INTR        3
 247 #define MD_RECOVERY_DONE        4
 248 #define MD_RECOVERY_NEEDED      5
 249 #define MD_RECOVERY_REQUESTED   6
 250 #define MD_RECOVERY_CHECK       7
 251 #define MD_RECOVERY_RESHAPE     8
 252 #define MD_RECOVERY_FROZEN      9
 253
 254         unsigned long                   recovery;
 255         int                             recovery_disabled; /* if we detect that recovery
 256                                                             * will always fail, set this
 257                                                             * so we don't loop trying */
 258
 259         int                             in_sync;        /* know to not need resync */
 260         /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
 261          * that we are never stopping an array while it is open.
 262          * 'reconfig_mutex' protects all other reconfiguration.
 263          * These locks are separate due to conflicting interactions
 264          * with bdev->bd_mutex.
 265          * Lock ordering is:
 266          *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
 267          *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
 268          */
 269         struct mutex                    open_mutex;
 270         struct mutex                    reconfig_mutex;
 271         atomic_t                        active;         /* general refcount */
 272         atomic_t                        openers;        /* number of active opens */
 273
 274         int                             degraded;       /* whether md should consider
 275                                                          * adding a spare
 276                                                          */
 277         int                             barriers_work;  /* initialised to true, cleared as soon
 278                                                          * as a barrier request to slave
 279                                                          * fails.  Only supported
 280                                                          */
 281         struct bio                      *biolist;       /* bios that need to be retried
 282                                                          * because REQ_HARDBARRIER is not supported
 283                                                          */
 284
 285         atomic_t                        recovery_active; /* blocks scheduled, but not written */
 286         wait_queue_head_t               recovery_wait;
 287         sector_t                        recovery_cp;
 288         sector_t                        resync_min;     /* user requested sync
 289                                                          * starts here */
 290         sector_t                        resync_max;     /* resync should pause
 291                                                          * when it gets here */
 292
 293         struct sysfs_dirent             *sysfs_state;   /* handle for 'array_state'
 294                                                          * file in sysfs.
 295                                                          */
 296         struct sysfs_dirent             *sysfs_action;  /* handle for 'sync_action' */
 297
 298         struct work_struct del_work;    /* used for delayed sysfs removal */
 299
 300         spinlock_t                      write_lock;
 301         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
 302         atomic_t                        pending_writes; /* number of active superblock writes */
 303
 304         unsigned int                    safemode;       /* if set, update "clean" superblock
 305                                                          * when no writes pending.
 306                                                          */
 307         unsigned int                    safemode_delay;
 308         struct timer_list               safemode_timer;
 309         atomic_t                        writes_pending;
 310         struct request_queue            *queue; /* for plugging ... */
 311
 312         struct bitmap                   *bitmap; /* the bitmap for the device */
 313         struct {
 314                 struct file             *file; /* the bitmap file */
 315                 loff_t                  offset; /* offset from superblock of
 316                                                  * start of bitmap. May be
 317                                                  * negative, but not '0'
 318                                                  * For external metadata, offset
 319                                                  * from start of device.
 320                                                  */
 321                 loff_t                  default_offset; /* this is the offset to use when
 322                                                          * hot-adding a bitmap.  It should
 323                                                          * eventually be settable by sysfs.
 324                                                          */
 325                 /* When md is serving under dm, it might use a
 326                  * dirty_log to store the bits.
 327                  */
 328                 struct dm_dirty_log *log;
 329
 330                 struct mutex            mutex;
 331                 unsigned long           chunksize;
 332                 unsigned long           daemon_sleep; /* how many jiffies between updates? */
 333                 unsigned long           max_write_behind; /* write-behind mode */
 334                 int                     external;
 335         } bitmap_info;
 336
 337         atomic_t                        max_corr_read_errors; /* max read retries */
 338         struct list_head                all_mddevs;
 339
 340         struct attribute_group          *to_remove;
 341         struct plug_handle              *plug; /* if used by personality */
 342
 343         /* Generic barrier handling.
 344          * If there is a pending barrier request, all other
 345          * writes are blocked while the devices are flushed.
 346          * The last to finish a flush schedules a worker to
 347          * submit the barrier request (without the barrier flag),
 348          * then submit more flush requests.
 349          */
 350         struct bio *barrier;
 351         atomic_t flush_pending;
 352         struct work_struct barrier_work;
 353         struct work_struct event_work;  /* used by dm to report failure event */
 354 };
 355
 356
 357 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
 358 {
 359         int faulty = test_bit(Faulty, &rdev->flags);
 360         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
 361                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 362 }
 363
 364 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
 365 {
 366         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 367 }
 368
 369 struct mdk_personality
 370 {
 371         char *name;
 372         int level;
 373         struct list_head list;
 374         struct module *owner;
 375         int (*make_request)(mddev_t *mddev, struct bio *bio);
 376         int (*run)(mddev_t *mddev);
 377         int (*stop)(mddev_t *mddev);
 378         void (*status)(struct seq_file *seq, mddev_t *mddev);
 379         /* error_handler must set ->faulty and clear ->in_sync
 380          * if appropriate, and should abort recovery if needed
 381          */
 382         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
 383         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
 384         int (*hot_remove_disk) (mddev_t *mddev, int number);
 385         int (*spare_active) (mddev_t *mddev);
 386         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
 387         int (*resize) (mddev_t *mddev, sector_t sectors);
 388         sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
 389         int (*check_reshape) (mddev_t *mddev);
 390         int (*start_reshape) (mddev_t *mddev);
 391         void (*finish_reshape) (mddev_t *mddev);
 392         /* quiesce moves between quiescence states
 393          * 0 - fully active
 394          * 1 - no new requests allowed
 395          * others - reserved
 396          */
 397         void (*quiesce) (mddev_t *mddev, int state);
 398         /* takeover is used to transition an array from one
 399          * personality to another.  The new personality must be able
 400          * to handle the data in the current layout.
 401          * e.g. 2drive raid1 -> 2drive raid5
 402          *      ndrive raid5 -> degraded n+1drive raid6 with special layout
 403          * If the takeover succeeds, a new 'private' structure is returned.
 404          * This needs to be installed and then ->run used to activate the
 405          * array.
 406          */
 407         void *(*takeover) (mddev_t *mddev);
 408 };
 409
 410
 411 struct md_sysfs_entry {
 412         struct attribute attr;
 413         ssize_t (*show)(mddev_t *, char *);
 414         ssize_t (*store)(mddev_t *, const char *, size_t);
 415 };
 416 extern struct attribute_group md_bitmap_group;
 417
 418 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
 419 {
 420         if (sd)
 421                 return sysfs_get_dirent(sd, NULL, name);
 422         return sd;
 423 }
 424 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
 425 {
 426         if (sd)
 427                 sysfs_notify_dirent(sd);
 428 }
 429
 430 static inline char * mdname (mddev_t * mddev)
 431 {
 432         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
 433 }
 434
 435 /*
 436  * iterates through some rdev ringlist. It's safe to remove the
 437  * current 'rdev'. Dont touch 'tmp' though.
 438  */
 439 #define rdev_for_each_list(rdev, tmp, head)                             \
 440         list_for_each_entry_safe(rdev, tmp, head, same_set)
 441
 442 /*
 443  * iterates through the 'same array disks' ringlist
 444  */
 445 #define rdev_for_each(rdev, tmp, mddev)                         \
 446         list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
 447
 448 #define rdev_for_each_rcu(rdev, mddev)                          \
 449         list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
 450
 451 typedef struct mdk_thread_s {
 452         void                    (*run) (mddev_t *mddev);
 453         mddev_t                 *mddev;
 454         wait_queue_head_t       wqueue;
 455         unsigned long           flags;
 456         struct task_struct      *tsk;
 457         unsigned long           timeout;
 458 } mdk_thread_t;
 459
 460 #define THREAD_WAKEUP  0
 461
 462 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
 463 do {                                                                    \
 464         wait_queue_t __wait;                                            \
 465         init_waitqueue_entry(&__wait, current);                         \
 466                                                                         \
 467         add_wait_queue(&wq, &__wait);                                   \
 468         for (;;) {                                                      \
 469                 set_current_state(TASK_UNINTERRUPTIBLE);                \
 470                 if (condition)                                          \
 471                         break;                                          \
 472                 spin_unlock_irq(&lock);                                 \
 473                 cmd;                                                    \
 474                 schedule();                                             \
 475                 spin_lock_irq(&lock);                                   \
 476         }                                                               \
 477         current->state = TASK_RUNNING;                                  \
 478         remove_wait_queue(&wq, &__wait);                                \
 479 } while (0)
 480
 481 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
 482 do {                                                                    \
 483         if (condition)                                                  \
 484                 break;                                                  \
 485         __wait_event_lock_irq(wq, condition, lock, cmd);                \
 486 } while (0)
 487
 488 static inline void safe_put_page(struct page *p)
 489 {
 490         if (p) put_page(p);
 491 }
 492
 493 extern int register_md_personality(struct mdk_personality *p);
 494 extern int unregister_md_personality(struct mdk_personality *p);
 495 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
 496                                 mddev_t *mddev, const char *name);
 497 extern void md_unregister_thread(mdk_thread_t *thread);
 498 extern void md_wakeup_thread(mdk_thread_t *thread);
 499 extern void md_check_recovery(mddev_t *mddev);
 500 extern void md_write_start(mddev_t *mddev, struct bio *bi);
 501 extern void md_write_end(mddev_t *mddev);
 502 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
 503 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
 504
 505 extern int mddev_congested(mddev_t *mddev, int bits);
 506 extern void md_barrier_request(mddev_t *mddev, struct bio *bio);
 507 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 508                            sector_t sector, int size, struct page *page);
 509 extern void md_super_wait(mddev_t *mddev);
 510 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 511                         struct page *page, int rw);
 512 extern void md_do_sync(mddev_t *mddev);
 513 extern void md_new_event(mddev_t *mddev);
 514 extern int md_allow_write(mddev_t *mddev);
 515 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
 516 extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
 517 extern int md_check_no_bitmap(mddev_t *mddev);
 518 extern int md_integrity_register(mddev_t *mddev);
 519 extern void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
 520 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
 521 extern void restore_bitmap_write_access(struct file *file);
 522 extern void md_unplug(mddev_t *mddev);
 523
 524 extern void mddev_init(mddev_t *mddev);
 525 extern int md_run(mddev_t *mddev);
 526 extern void md_stop(mddev_t *mddev);
 527 extern void md_stop_writes(mddev_t *mddev);
 528 extern void md_rdev_init(mdk_rdev_t *rdev);
 529
 530 extern void mddev_suspend(mddev_t *mddev);
 531 extern void mddev_resume(mddev_t *mddev);
 532 #endif /* _MD_MD_H */