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