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