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