1 /* 2 md.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 /* Bad block numbers are stored sorted in a single page. 30 * 64bits is used for each block or extent. 31 * 54 bits are sector number, 9 bits are extent size, 32 * 1 bit is an 'acknowledged' flag. 33 */ 34 #define MD_MAX_BADBLOCKS (PAGE_SIZE/8) 35 36 /* 37 * MD's 'extended' device 38 */ 39 struct md_rdev { 40 struct list_head same_set; /* RAID devices within the same set */ 41 42 sector_t sectors; /* Device size (in 512bytes sectors) */ 43 struct mddev *mddev; /* RAID array if running */ 44 int last_events; /* IO event timestamp */ 45 46 /* 47 * If meta_bdev is non-NULL, it means that a separate device is 48 * being used to store the metadata (superblock/bitmap) which 49 * would otherwise be contained on the same device as the data (bdev). 50 */ 51 struct block_device *meta_bdev; 52 struct block_device *bdev; /* block device handle */ 53 54 struct page *sb_page, *bb_page; 55 int sb_loaded; 56 __u64 sb_events; 57 sector_t data_offset; /* start of data in array */ 58 sector_t sb_start; /* offset of the super block (in 512byte sectors) */ 59 int sb_size; /* bytes in the superblock */ 60 int preferred_minor; /* autorun support */ 61 62 struct kobject kobj; 63 64 /* A device can be in one of three states based on two flags: 65 * Not working: faulty==1 in_sync==0 66 * Fully working: faulty==0 in_sync==1 67 * Working, but not 68 * in sync with array 69 * faulty==0 in_sync==0 70 * 71 * It can never have faulty==1, in_sync==1 72 * This reduces the burden of testing multiple flags in many cases 73 */ 74 75 unsigned long flags; 76 #define Faulty 1 /* device is known to have a fault */ 77 #define In_sync 2 /* device is in_sync with rest of array */ 78 #define WriteMostly 4 /* Avoid reading if at all possible */ 79 #define AutoDetected 7 /* added by auto-detect */ 80 #define Blocked 8 /* An error occurred but has not yet 81 * been acknowledged by the metadata 82 * handler, so don't allow writes 83 * until it is cleared */ 84 #define WriteErrorSeen 9 /* A write error has been seen on this 85 * device 86 */ 87 #define FaultRecorded 10 /* Intermediate state for clearing 88 * Blocked. The Fault is/will-be 89 * recorded in the metadata, but that 90 * metadata hasn't been stored safely 91 * on disk yet. 92 */ 93 #define BlockedBadBlocks 11 /* A writer is blocked because they 94 * found an unacknowledged bad-block. 95 * This can safely be cleared at any 96 * time, and the writer will re-check. 97 * It may be set at any time, and at 98 * worst the writer will timeout and 99 * re-check. So setting it as 100 * accurately as possible is good, but 101 * not absolutely critical. 102 */ 103 wait_queue_head_t blocked_wait; 104 105 int desc_nr; /* descriptor index in the superblock */ 106 int raid_disk; /* role of device in array */ 107 int new_raid_disk; /* role that the device will have in 108 * the array after a level-change completes. 109 */ 110 int saved_raid_disk; /* role that device used to have in the 111 * array and could again if we did a partial 112 * resync from the bitmap 113 */ 114 sector_t recovery_offset;/* If this device has been partially 115 * recovered, this is where we were 116 * up to. 117 */ 118 119 atomic_t nr_pending; /* number of pending requests. 120 * only maintained for arrays that 121 * support hot removal 122 */ 123 atomic_t read_errors; /* number of consecutive read errors that 124 * we have tried to ignore. 125 */ 126 struct timespec last_read_error; /* monotonic time since our 127 * last read error 128 */ 129 atomic_t corrected_errors; /* number of corrected read errors, 130 * for reporting to userspace and storing 131 * in superblock. 132 */ 133 struct work_struct del_work; /* used for delayed sysfs removal */ 134 135 struct sysfs_dirent *sysfs_state; /* handle for 'state' 136 * sysfs entry */ 137 138 struct badblocks { 139 int count; /* count of bad blocks */ 140 int unacked_exist; /* there probably are unacknowledged 141 * bad blocks. This is only cleared 142 * when a read discovers none 143 */ 144 int shift; /* shift from sectors to block size 145 * a -ve shift means badblocks are 146 * disabled.*/ 147 u64 *page; /* badblock list */ 148 int changed; 149 seqlock_t lock; 150 151 sector_t sector; 152 sector_t size; /* in sectors */ 153 } badblocks; 154 }; 155 156 #define BB_LEN_MASK (0x00000000000001FFULL) 157 #define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL) 158 #define BB_ACK_MASK (0x8000000000000000ULL) 159 #define BB_MAX_LEN 512 160 #define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9) 161 #define BB_LEN(x) (((x) & BB_LEN_MASK) + 1) 162 #define BB_ACK(x) (!!((x) & BB_ACK_MASK)) 163 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63)) 164 165 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, 166 sector_t *first_bad, int *bad_sectors); 167 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors, 168 sector_t *first_bad, int *bad_sectors) 169 { 170 if (unlikely(rdev->badblocks.count)) { 171 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s, 172 sectors, 173 first_bad, bad_sectors); 174 if (rv) 175 *first_bad -= rdev->data_offset; 176 return rv; 177 } 178 return 0; 179 } 180 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, 181 int acknowledged); 182 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors); 183 extern void md_ack_all_badblocks(struct badblocks *bb); 184 185 struct mddev { 186 void *private; 187 struct md_personality *pers; 188 dev_t unit; 189 int md_minor; 190 struct list_head disks; 191 unsigned long flags; 192 #define MD_CHANGE_DEVS 0 /* Some device status has changed */ 193 #define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */ 194 #define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */ 195 #define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */ 196 197 int suspended; 198 atomic_t active_io; 199 int ro; 200 int sysfs_active; /* set when sysfs deletes 201 * are happening, so run/ 202 * takeover/stop are not safe 203 */ 204 int ready; /* See when safe to pass 205 * IO requests down */ 206 struct gendisk *gendisk; 207 208 struct kobject kobj; 209 int hold_active; 210 #define UNTIL_IOCTL 1 211 #define UNTIL_STOP 2 212 213 /* Superblock information */ 214 int major_version, 215 minor_version, 216 patch_version; 217 int persistent; 218 int external; /* metadata is 219 * managed externally */ 220 char metadata_type[17]; /* externally set*/ 221 int chunk_sectors; 222 time_t ctime, utime; 223 int level, layout; 224 char clevel[16]; 225 int raid_disks; 226 int max_disks; 227 sector_t dev_sectors; /* used size of 228 * component devices */ 229 sector_t array_sectors; /* exported array size */ 230 int external_size; /* size managed 231 * externally */ 232 __u64 events; 233 /* If the last 'event' was simply a clean->dirty transition, and 234 * we didn't write it to the spares, then it is safe and simple 235 * to just decrement the event count on a dirty->clean transition. 236 * So we record that possibility here. 237 */ 238 int can_decrease_events; 239 240 char uuid[16]; 241 242 /* If the array is being reshaped, we need to record the 243 * new shape and an indication of where we are up to. 244 * This is written to the superblock. 245 * If reshape_position is MaxSector, then no reshape is happening (yet). 246 */ 247 sector_t reshape_position; 248 int delta_disks, new_level, new_layout; 249 int new_chunk_sectors; 250 251 atomic_t plug_cnt; /* If device is expecting 252 * more bios soon. 253 */ 254 struct md_thread *thread; /* management thread */ 255 struct md_thread *sync_thread; /* doing resync or reconstruct */ 256 sector_t curr_resync; /* last block scheduled */ 257 /* As resync requests can complete out of order, we cannot easily track 258 * how much resync has been completed. So we occasionally pause until 259 * everything completes, then set curr_resync_completed to curr_resync. 260 * As such it may be well behind the real resync mark, but it is a value 261 * we are certain of. 262 */ 263 sector_t curr_resync_completed; 264 unsigned long resync_mark; /* a recent timestamp */ 265 sector_t resync_mark_cnt;/* blocks written at resync_mark */ 266 sector_t curr_mark_cnt; /* blocks scheduled now */ 267 268 sector_t resync_max_sectors; /* may be set by personality */ 269 270 sector_t resync_mismatches; /* count of sectors where 271 * parity/replica mismatch found 272 */ 273 274 /* allow user-space to request suspension of IO to regions of the array */ 275 sector_t suspend_lo; 276 sector_t suspend_hi; 277 /* if zero, use the system-wide default */ 278 int sync_speed_min; 279 int sync_speed_max; 280 281 /* resync even though the same disks are shared among md-devices */ 282 int parallel_resync; 283 284 int ok_start_degraded; 285 /* recovery/resync flags 286 * NEEDED: we might need to start a resync/recover 287 * RUNNING: a thread is running, or about to be started 288 * SYNC: actually doing a resync, not a recovery 289 * RECOVER: doing recovery, or need to try it. 290 * INTR: resync needs to be aborted for some reason 291 * DONE: thread is done and is waiting to be reaped 292 * REQUEST: user-space has requested a sync (used with SYNC) 293 * CHECK: user-space request for check-only, no repair 294 * RESHAPE: A reshape is happening 295 * 296 * If neither SYNC or RESHAPE are set, then it is a recovery. 297 */ 298 #define MD_RECOVERY_RUNNING 0 299 #define MD_RECOVERY_SYNC 1 300 #define MD_RECOVERY_RECOVER 2 301 #define MD_RECOVERY_INTR 3 302 #define MD_RECOVERY_DONE 4 303 #define MD_RECOVERY_NEEDED 5 304 #define MD_RECOVERY_REQUESTED 6 305 #define MD_RECOVERY_CHECK 7 306 #define MD_RECOVERY_RESHAPE 8 307 #define MD_RECOVERY_FROZEN 9 308 309 unsigned long recovery; 310 /* If a RAID personality determines that recovery (of a particular 311 * device) will fail due to a read error on the source device, it 312 * takes a copy of this number and does not attempt recovery again 313 * until this number changes. 314 */ 315 int recovery_disabled; 316 317 int in_sync; /* know to not need resync */ 318 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so 319 * that we are never stopping an array while it is open. 320 * 'reconfig_mutex' protects all other reconfiguration. 321 * These locks are separate due to conflicting interactions 322 * with bdev->bd_mutex. 323 * Lock ordering is: 324 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk 325 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open 326 */ 327 struct mutex open_mutex; 328 struct mutex reconfig_mutex; 329 atomic_t active; /* general refcount */ 330 atomic_t openers; /* number of active opens */ 331 332 int changed; /* True if we might need to 333 * reread partition info */ 334 int degraded; /* whether md should consider 335 * adding a spare 336 */ 337 338 atomic_t recovery_active; /* blocks scheduled, but not written */ 339 wait_queue_head_t recovery_wait; 340 sector_t recovery_cp; 341 sector_t resync_min; /* user requested sync 342 * starts here */ 343 sector_t resync_max; /* resync should pause 344 * when it gets here */ 345 346 struct sysfs_dirent *sysfs_state; /* handle for 'array_state' 347 * file in sysfs. 348 */ 349 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */ 350 351 struct work_struct del_work; /* used for delayed sysfs removal */ 352 353 spinlock_t write_lock; 354 wait_queue_head_t sb_wait; /* for waiting on superblock updates */ 355 atomic_t pending_writes; /* number of active superblock writes */ 356 357 unsigned int safemode; /* if set, update "clean" superblock 358 * when no writes pending. 359 */ 360 unsigned int safemode_delay; 361 struct timer_list safemode_timer; 362 atomic_t writes_pending; 363 struct request_queue *queue; /* for plugging ... */ 364 365 struct bitmap *bitmap; /* the bitmap for the device */ 366 struct { 367 struct file *file; /* the bitmap file */ 368 loff_t offset; /* offset from superblock of 369 * start of bitmap. May be 370 * negative, but not '0' 371 * For external metadata, offset 372 * from start of device. 373 */ 374 loff_t default_offset; /* this is the offset to use when 375 * hot-adding a bitmap. It should 376 * eventually be settable by sysfs. 377 */ 378 struct mutex mutex; 379 unsigned long chunksize; 380 unsigned long daemon_sleep; /* how many jiffies between updates? */ 381 unsigned long max_write_behind; /* write-behind mode */ 382 int external; 383 } bitmap_info; 384 385 atomic_t max_corr_read_errors; /* max read retries */ 386 struct list_head all_mddevs; 387 388 struct attribute_group *to_remove; 389 390 struct bio_set *bio_set; 391 392 /* Generic flush handling. 393 * The last to finish preflush schedules a worker to submit 394 * the rest of the request (without the REQ_FLUSH flag). 395 */ 396 struct bio *flush_bio; 397 atomic_t flush_pending; 398 struct work_struct flush_work; 399 struct work_struct event_work; /* used by dm to report failure event */ 400 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev); 401 }; 402 403 404 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev) 405 { 406 int faulty = test_bit(Faulty, &rdev->flags); 407 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) 408 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 409 } 410 411 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors) 412 { 413 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io); 414 } 415 416 struct md_personality 417 { 418 char *name; 419 int level; 420 struct list_head list; 421 struct module *owner; 422 void (*make_request)(struct mddev *mddev, struct bio *bio); 423 int (*run)(struct mddev *mddev); 424 int (*stop)(struct mddev *mddev); 425 void (*status)(struct seq_file *seq, struct mddev *mddev); 426 /* error_handler must set ->faulty and clear ->in_sync 427 * if appropriate, and should abort recovery if needed 428 */ 429 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev); 430 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev); 431 int (*hot_remove_disk) (struct mddev *mddev, int number); 432 int (*spare_active) (struct mddev *mddev); 433 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster); 434 int (*resize) (struct mddev *mddev, sector_t sectors); 435 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks); 436 int (*check_reshape) (struct mddev *mddev); 437 int (*start_reshape) (struct mddev *mddev); 438 void (*finish_reshape) (struct mddev *mddev); 439 /* quiesce moves between quiescence states 440 * 0 - fully active 441 * 1 - no new requests allowed 442 * others - reserved 443 */ 444 void (*quiesce) (struct mddev *mddev, int state); 445 /* takeover is used to transition an array from one 446 * personality to another. The new personality must be able 447 * to handle the data in the current layout. 448 * e.g. 2drive raid1 -> 2drive raid5 449 * ndrive raid5 -> degraded n+1drive raid6 with special layout 450 * If the takeover succeeds, a new 'private' structure is returned. 451 * This needs to be installed and then ->run used to activate the 452 * array. 453 */ 454 void *(*takeover) (struct mddev *mddev); 455 }; 456 457 458 struct md_sysfs_entry { 459 struct attribute attr; 460 ssize_t (*show)(struct mddev *, char *); 461 ssize_t (*store)(struct mddev *, const char *, size_t); 462 }; 463 extern struct attribute_group md_bitmap_group; 464 465 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name) 466 { 467 if (sd) 468 return sysfs_get_dirent(sd, NULL, name); 469 return sd; 470 } 471 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd) 472 { 473 if (sd) 474 sysfs_notify_dirent(sd); 475 } 476 477 static inline char * mdname (struct mddev * mddev) 478 { 479 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX"; 480 } 481 482 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev) 483 { 484 char nm[20]; 485 sprintf(nm, "rd%d", rdev->raid_disk); 486 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); 487 } 488 489 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev) 490 { 491 char nm[20]; 492 sprintf(nm, "rd%d", rdev->raid_disk); 493 sysfs_remove_link(&mddev->kobj, nm); 494 } 495 496 /* 497 * iterates through some rdev ringlist. It's safe to remove the 498 * current 'rdev'. Dont touch 'tmp' though. 499 */ 500 #define rdev_for_each_list(rdev, tmp, head) \ 501 list_for_each_entry_safe(rdev, tmp, head, same_set) 502 503 /* 504 * iterates through the 'same array disks' ringlist 505 */ 506 #define rdev_for_each(rdev, tmp, mddev) \ 507 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set) 508 509 #define rdev_for_each_rcu(rdev, mddev) \ 510 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set) 511 512 struct md_thread { 513 void (*run) (struct mddev *mddev); 514 struct mddev *mddev; 515 wait_queue_head_t wqueue; 516 unsigned long flags; 517 struct task_struct *tsk; 518 unsigned long timeout; 519 }; 520 521 #define THREAD_WAKEUP 0 522 523 #define __wait_event_lock_irq(wq, condition, lock, cmd) \ 524 do { \ 525 wait_queue_t __wait; \ 526 init_waitqueue_entry(&__wait, current); \ 527 \ 528 add_wait_queue(&wq, &__wait); \ 529 for (;;) { \ 530 set_current_state(TASK_UNINTERRUPTIBLE); \ 531 if (condition) \ 532 break; \ 533 spin_unlock_irq(&lock); \ 534 cmd; \ 535 schedule(); \ 536 spin_lock_irq(&lock); \ 537 } \ 538 current->state = TASK_RUNNING; \ 539 remove_wait_queue(&wq, &__wait); \ 540 } while (0) 541 542 #define wait_event_lock_irq(wq, condition, lock, cmd) \ 543 do { \ 544 if (condition) \ 545 break; \ 546 __wait_event_lock_irq(wq, condition, lock, cmd); \ 547 } while (0) 548 549 static inline void safe_put_page(struct page *p) 550 { 551 if (p) put_page(p); 552 } 553 554 extern int register_md_personality(struct md_personality *p); 555 extern int unregister_md_personality(struct md_personality *p); 556 extern struct md_thread *md_register_thread( 557 void (*run)(struct mddev *mddev), 558 struct mddev *mddev, 559 const char *name); 560 extern void md_unregister_thread(struct md_thread **threadp); 561 extern void md_wakeup_thread(struct md_thread *thread); 562 extern void md_check_recovery(struct mddev *mddev); 563 extern void md_write_start(struct mddev *mddev, struct bio *bi); 564 extern void md_write_end(struct mddev *mddev); 565 extern void md_done_sync(struct mddev *mddev, int blocks, int ok); 566 extern void md_error(struct mddev *mddev, struct md_rdev *rdev); 567 568 extern int mddev_congested(struct mddev *mddev, int bits); 569 extern void md_flush_request(struct mddev *mddev, struct bio *bio); 570 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev, 571 sector_t sector, int size, struct page *page); 572 extern void md_super_wait(struct mddev *mddev); 573 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, 574 struct page *page, int rw, bool metadata_op); 575 extern void md_do_sync(struct mddev *mddev); 576 extern void md_new_event(struct mddev *mddev); 577 extern int md_allow_write(struct mddev *mddev); 578 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev); 579 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors); 580 extern int md_check_no_bitmap(struct mddev *mddev); 581 extern int md_integrity_register(struct mddev *mddev); 582 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev); 583 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale); 584 extern void restore_bitmap_write_access(struct file *file); 585 586 extern void mddev_init(struct mddev *mddev); 587 extern int md_run(struct mddev *mddev); 588 extern void md_stop(struct mddev *mddev); 589 extern void md_stop_writes(struct mddev *mddev); 590 extern int md_rdev_init(struct md_rdev *rdev); 591 592 extern void mddev_suspend(struct mddev *mddev); 593 extern void mddev_resume(struct mddev *mddev); 594 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, 595 struct mddev *mddev); 596 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, 597 struct mddev *mddev); 598 extern int mddev_check_plugged(struct mddev *mddev); 599 extern void md_trim_bio(struct bio *bio, int offset, int size); 600 #endif /* _MD_MD_H */ 601