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 new_data_offset;/* only relevant while reshaping */ 59 sector_t sb_start; /* offset of the super block (in 512byte sectors) */ 60 int sb_size; /* bytes in the superblock */ 61 int preferred_minor; /* autorun support */ 62 63 struct kobject kobj; 64 65 /* A device can be in one of three states based on two flags: 66 * Not working: faulty==1 in_sync==0 67 * Fully working: faulty==0 in_sync==1 68 * Working, but not 69 * in sync with array 70 * faulty==0 in_sync==0 71 * 72 * It can never have faulty==1, in_sync==1 73 * This reduces the burden of testing multiple flags in many cases 74 */ 75 76 unsigned long flags; /* bit set of 'enum flag_bits' bits. */ 77 wait_queue_head_t blocked_wait; 78 79 int desc_nr; /* descriptor index in the superblock */ 80 int raid_disk; /* role of device in array */ 81 int new_raid_disk; /* role that the device will have in 82 * the array after a level-change completes. 83 */ 84 int saved_raid_disk; /* role that device used to have in the 85 * array and could again if we did a partial 86 * resync from the bitmap 87 */ 88 sector_t recovery_offset;/* If this device has been partially 89 * recovered, this is where we were 90 * up to. 91 */ 92 93 atomic_t nr_pending; /* number of pending requests. 94 * only maintained for arrays that 95 * support hot removal 96 */ 97 atomic_t read_errors; /* number of consecutive read errors that 98 * we have tried to ignore. 99 */ 100 struct timespec last_read_error; /* monotonic time since our 101 * last read error 102 */ 103 atomic_t corrected_errors; /* number of corrected read errors, 104 * for reporting to userspace and storing 105 * in superblock. 106 */ 107 struct work_struct del_work; /* used for delayed sysfs removal */ 108 109 struct sysfs_dirent *sysfs_state; /* handle for 'state' 110 * sysfs entry */ 111 112 struct badblocks { 113 int count; /* count of bad blocks */ 114 int unacked_exist; /* there probably are unacknowledged 115 * bad blocks. This is only cleared 116 * when a read discovers none 117 */ 118 int shift; /* shift from sectors to block size 119 * a -ve shift means badblocks are 120 * disabled.*/ 121 u64 *page; /* badblock list */ 122 int changed; 123 seqlock_t lock; 124 125 sector_t sector; 126 sector_t size; /* in sectors */ 127 } badblocks; 128 }; 129 enum flag_bits { 130 Faulty, /* device is known to have a fault */ 131 In_sync, /* device is in_sync with rest of array */ 132 Unmerged, /* device is being added to array and should 133 * be considerred for bvec_merge_fn but not 134 * yet for actual IO 135 */ 136 WriteMostly, /* Avoid reading if at all possible */ 137 AutoDetected, /* added by auto-detect */ 138 Blocked, /* An error occurred but has not yet 139 * been acknowledged by the metadata 140 * handler, so don't allow writes 141 * until it is cleared */ 142 WriteErrorSeen, /* A write error has been seen on this 143 * device 144 */ 145 FaultRecorded, /* Intermediate state for clearing 146 * Blocked. The Fault is/will-be 147 * recorded in the metadata, but that 148 * metadata hasn't been stored safely 149 * on disk yet. 150 */ 151 BlockedBadBlocks, /* A writer is blocked because they 152 * found an unacknowledged bad-block. 153 * This can safely be cleared at any 154 * time, and the writer will re-check. 155 * It may be set at any time, and at 156 * worst the writer will timeout and 157 * re-check. So setting it as 158 * accurately as possible is good, but 159 * not absolutely critical. 160 */ 161 WantReplacement, /* This device is a candidate to be 162 * hot-replaced, either because it has 163 * reported some faults, or because 164 * of explicit request. 165 */ 166 Replacement, /* This device is a replacement for 167 * a want_replacement device with same 168 * raid_disk number. 169 */ 170 }; 171 172 #define BB_LEN_MASK (0x00000000000001FFULL) 173 #define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL) 174 #define BB_ACK_MASK (0x8000000000000000ULL) 175 #define BB_MAX_LEN 512 176 #define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9) 177 #define BB_LEN(x) (((x) & BB_LEN_MASK) + 1) 178 #define BB_ACK(x) (!!((x) & BB_ACK_MASK)) 179 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63)) 180 181 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, 182 sector_t *first_bad, int *bad_sectors); 183 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors, 184 sector_t *first_bad, int *bad_sectors) 185 { 186 if (unlikely(rdev->badblocks.count)) { 187 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s, 188 sectors, 189 first_bad, bad_sectors); 190 if (rv) 191 *first_bad -= rdev->data_offset; 192 return rv; 193 } 194 return 0; 195 } 196 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, 197 int is_new); 198 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, 199 int is_new); 200 extern void md_ack_all_badblocks(struct badblocks *bb); 201 202 struct mddev { 203 void *private; 204 struct md_personality *pers; 205 dev_t unit; 206 int md_minor; 207 struct list_head disks; 208 unsigned long flags; 209 #define MD_CHANGE_DEVS 0 /* Some device status has changed */ 210 #define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */ 211 #define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */ 212 #define MD_UPDATE_SB_FLAGS (1 | 2 | 4) /* If these are set, md_update_sb needed */ 213 #define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */ 214 #define MD_STILL_CLOSED 4 /* If set, then array has not been opened since 215 * md_ioctl checked on it. 216 */ 217 218 int suspended; 219 atomic_t active_io; 220 int ro; 221 int sysfs_active; /* set when sysfs deletes 222 * are happening, so run/ 223 * takeover/stop are not safe 224 */ 225 int ready; /* See when safe to pass 226 * IO requests down */ 227 struct gendisk *gendisk; 228 229 struct kobject kobj; 230 int hold_active; 231 #define UNTIL_IOCTL 1 232 #define UNTIL_STOP 2 233 234 /* Superblock information */ 235 int major_version, 236 minor_version, 237 patch_version; 238 int persistent; 239 int external; /* metadata is 240 * managed externally */ 241 char metadata_type[17]; /* externally set*/ 242 int chunk_sectors; 243 time_t ctime, utime; 244 int level, layout; 245 char clevel[16]; 246 int raid_disks; 247 int max_disks; 248 sector_t dev_sectors; /* used size of 249 * component devices */ 250 sector_t array_sectors; /* exported array size */ 251 int external_size; /* size managed 252 * externally */ 253 __u64 events; 254 /* If the last 'event' was simply a clean->dirty transition, and 255 * we didn't write it to the spares, then it is safe and simple 256 * to just decrement the event count on a dirty->clean transition. 257 * So we record that possibility here. 258 */ 259 int can_decrease_events; 260 261 char uuid[16]; 262 263 /* If the array is being reshaped, we need to record the 264 * new shape and an indication of where we are up to. 265 * This is written to the superblock. 266 * If reshape_position is MaxSector, then no reshape is happening (yet). 267 */ 268 sector_t reshape_position; 269 int delta_disks, new_level, new_layout; 270 int new_chunk_sectors; 271 int reshape_backwards; 272 273 struct md_thread *thread; /* management thread */ 274 struct md_thread *sync_thread; /* doing resync or reconstruct */ 275 276 /* 'last_sync_action' is initialized to "none". It is set when a 277 * sync operation (i.e "data-check", "requested-resync", "resync", 278 * "recovery", or "reshape") is started. It holds this value even 279 * when the sync thread is "frozen" (interrupted) or "idle" (stopped 280 * or finished). It is overwritten when a new sync operation is begun. 281 */ 282 char *last_sync_action; 283 sector_t curr_resync; /* last block scheduled */ 284 /* As resync requests can complete out of order, we cannot easily track 285 * how much resync has been completed. So we occasionally pause until 286 * everything completes, then set curr_resync_completed to curr_resync. 287 * As such it may be well behind the real resync mark, but it is a value 288 * we are certain of. 289 */ 290 sector_t curr_resync_completed; 291 unsigned long resync_mark; /* a recent timestamp */ 292 sector_t resync_mark_cnt;/* blocks written at resync_mark */ 293 sector_t curr_mark_cnt; /* blocks scheduled now */ 294 295 sector_t resync_max_sectors; /* may be set by personality */ 296 297 atomic64_t resync_mismatches; /* count of sectors where 298 * parity/replica mismatch found 299 */ 300 301 /* allow user-space to request suspension of IO to regions of the array */ 302 sector_t suspend_lo; 303 sector_t suspend_hi; 304 /* if zero, use the system-wide default */ 305 int sync_speed_min; 306 int sync_speed_max; 307 308 /* resync even though the same disks are shared among md-devices */ 309 int parallel_resync; 310 311 int ok_start_degraded; 312 /* recovery/resync flags 313 * NEEDED: we might need to start a resync/recover 314 * RUNNING: a thread is running, or about to be started 315 * SYNC: actually doing a resync, not a recovery 316 * RECOVER: doing recovery, or need to try it. 317 * INTR: resync needs to be aborted for some reason 318 * DONE: thread is done and is waiting to be reaped 319 * REQUEST: user-space has requested a sync (used with SYNC) 320 * CHECK: user-space request for check-only, no repair 321 * RESHAPE: A reshape is happening 322 * ERROR: sync-action interrupted because io-error 323 * 324 * If neither SYNC or RESHAPE are set, then it is a recovery. 325 */ 326 #define MD_RECOVERY_RUNNING 0 327 #define MD_RECOVERY_SYNC 1 328 #define MD_RECOVERY_RECOVER 2 329 #define MD_RECOVERY_INTR 3 330 #define MD_RECOVERY_DONE 4 331 #define MD_RECOVERY_NEEDED 5 332 #define MD_RECOVERY_REQUESTED 6 333 #define MD_RECOVERY_CHECK 7 334 #define MD_RECOVERY_RESHAPE 8 335 #define MD_RECOVERY_FROZEN 9 336 #define MD_RECOVERY_ERROR 10 337 338 unsigned long recovery; 339 /* If a RAID personality determines that recovery (of a particular 340 * device) will fail due to a read error on the source device, it 341 * takes a copy of this number and does not attempt recovery again 342 * until this number changes. 343 */ 344 int recovery_disabled; 345 346 int in_sync; /* know to not need resync */ 347 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so 348 * that we are never stopping an array while it is open. 349 * 'reconfig_mutex' protects all other reconfiguration. 350 * These locks are separate due to conflicting interactions 351 * with bdev->bd_mutex. 352 * Lock ordering is: 353 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk 354 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open 355 */ 356 struct mutex open_mutex; 357 struct mutex reconfig_mutex; 358 atomic_t active; /* general refcount */ 359 atomic_t openers; /* number of active opens */ 360 361 int changed; /* True if we might need to 362 * reread partition info */ 363 int degraded; /* whether md should consider 364 * adding a spare 365 */ 366 int merge_check_needed; /* at least one 367 * member device 368 * has a 369 * merge_bvec_fn */ 370 371 atomic_t recovery_active; /* blocks scheduled, but not written */ 372 wait_queue_head_t recovery_wait; 373 sector_t recovery_cp; 374 sector_t resync_min; /* user requested sync 375 * starts here */ 376 sector_t resync_max; /* resync should pause 377 * when it gets here */ 378 379 struct sysfs_dirent *sysfs_state; /* handle for 'array_state' 380 * file in sysfs. 381 */ 382 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */ 383 384 struct work_struct del_work; /* used for delayed sysfs removal */ 385 386 spinlock_t write_lock; 387 wait_queue_head_t sb_wait; /* for waiting on superblock updates */ 388 atomic_t pending_writes; /* number of active superblock writes */ 389 390 unsigned int safemode; /* if set, update "clean" superblock 391 * when no writes pending. 392 */ 393 unsigned int safemode_delay; 394 struct timer_list safemode_timer; 395 atomic_t writes_pending; 396 struct request_queue *queue; /* for plugging ... */ 397 398 struct bitmap *bitmap; /* the bitmap for the device */ 399 struct { 400 struct file *file; /* the bitmap file */ 401 loff_t offset; /* offset from superblock of 402 * start of bitmap. May be 403 * negative, but not '0' 404 * For external metadata, offset 405 * from start of device. 406 */ 407 unsigned long space; /* space available at this offset */ 408 loff_t default_offset; /* this is the offset to use when 409 * hot-adding a bitmap. It should 410 * eventually be settable by sysfs. 411 */ 412 unsigned long default_space; /* space available at 413 * default offset */ 414 struct mutex mutex; 415 unsigned long chunksize; 416 unsigned long daemon_sleep; /* how many jiffies between updates? */ 417 unsigned long max_write_behind; /* write-behind mode */ 418 int external; 419 } bitmap_info; 420 421 atomic_t max_corr_read_errors; /* max read retries */ 422 struct list_head all_mddevs; 423 424 struct attribute_group *to_remove; 425 426 struct bio_set *bio_set; 427 428 /* Generic flush handling. 429 * The last to finish preflush schedules a worker to submit 430 * the rest of the request (without the REQ_FLUSH flag). 431 */ 432 struct bio *flush_bio; 433 atomic_t flush_pending; 434 struct work_struct flush_work; 435 struct work_struct event_work; /* used by dm to report failure event */ 436 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev); 437 }; 438 439 440 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev) 441 { 442 int faulty = test_bit(Faulty, &rdev->flags); 443 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) 444 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 445 } 446 447 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors) 448 { 449 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io); 450 } 451 452 struct md_personality 453 { 454 char *name; 455 int level; 456 struct list_head list; 457 struct module *owner; 458 void (*make_request)(struct mddev *mddev, struct bio *bio); 459 int (*run)(struct mddev *mddev); 460 int (*stop)(struct mddev *mddev); 461 void (*status)(struct seq_file *seq, struct mddev *mddev); 462 /* error_handler must set ->faulty and clear ->in_sync 463 * if appropriate, and should abort recovery if needed 464 */ 465 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev); 466 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev); 467 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev); 468 int (*spare_active) (struct mddev *mddev); 469 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster); 470 int (*resize) (struct mddev *mddev, sector_t sectors); 471 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks); 472 int (*check_reshape) (struct mddev *mddev); 473 int (*start_reshape) (struct mddev *mddev); 474 void (*finish_reshape) (struct mddev *mddev); 475 /* quiesce moves between quiescence states 476 * 0 - fully active 477 * 1 - no new requests allowed 478 * others - reserved 479 */ 480 void (*quiesce) (struct mddev *mddev, int state); 481 /* takeover is used to transition an array from one 482 * personality to another. The new personality must be able 483 * to handle the data in the current layout. 484 * e.g. 2drive raid1 -> 2drive raid5 485 * ndrive raid5 -> degraded n+1drive raid6 with special layout 486 * If the takeover succeeds, a new 'private' structure is returned. 487 * This needs to be installed and then ->run used to activate the 488 * array. 489 */ 490 void *(*takeover) (struct mddev *mddev); 491 }; 492 493 494 struct md_sysfs_entry { 495 struct attribute attr; 496 ssize_t (*show)(struct mddev *, char *); 497 ssize_t (*store)(struct mddev *, const char *, size_t); 498 }; 499 extern struct attribute_group md_bitmap_group; 500 501 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name) 502 { 503 if (sd) 504 return sysfs_get_dirent(sd, name); 505 return sd; 506 } 507 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd) 508 { 509 if (sd) 510 sysfs_notify_dirent(sd); 511 } 512 513 static inline char * mdname (struct mddev * mddev) 514 { 515 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX"; 516 } 517 518 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev) 519 { 520 char nm[20]; 521 if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) { 522 sprintf(nm, "rd%d", rdev->raid_disk); 523 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); 524 } else 525 return 0; 526 } 527 528 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev) 529 { 530 char nm[20]; 531 if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) { 532 sprintf(nm, "rd%d", rdev->raid_disk); 533 sysfs_remove_link(&mddev->kobj, nm); 534 } 535 } 536 537 /* 538 * iterates through some rdev ringlist. It's safe to remove the 539 * current 'rdev'. Dont touch 'tmp' though. 540 */ 541 #define rdev_for_each_list(rdev, tmp, head) \ 542 list_for_each_entry_safe(rdev, tmp, head, same_set) 543 544 /* 545 * iterates through the 'same array disks' ringlist 546 */ 547 #define rdev_for_each(rdev, mddev) \ 548 list_for_each_entry(rdev, &((mddev)->disks), same_set) 549 550 #define rdev_for_each_safe(rdev, tmp, mddev) \ 551 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set) 552 553 #define rdev_for_each_rcu(rdev, mddev) \ 554 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set) 555 556 struct md_thread { 557 void (*run) (struct md_thread *thread); 558 struct mddev *mddev; 559 wait_queue_head_t wqueue; 560 unsigned long flags; 561 struct task_struct *tsk; 562 unsigned long timeout; 563 void *private; 564 }; 565 566 #define THREAD_WAKEUP 0 567 568 static inline void safe_put_page(struct page *p) 569 { 570 if (p) put_page(p); 571 } 572 573 extern int register_md_personality(struct md_personality *p); 574 extern int unregister_md_personality(struct md_personality *p); 575 extern struct md_thread *md_register_thread( 576 void (*run)(struct md_thread *thread), 577 struct mddev *mddev, 578 const char *name); 579 extern void md_unregister_thread(struct md_thread **threadp); 580 extern void md_wakeup_thread(struct md_thread *thread); 581 extern void md_check_recovery(struct mddev *mddev); 582 extern void md_reap_sync_thread(struct mddev *mddev); 583 extern void md_write_start(struct mddev *mddev, struct bio *bi); 584 extern void md_write_end(struct mddev *mddev); 585 extern void md_done_sync(struct mddev *mddev, int blocks, int ok); 586 extern void md_error(struct mddev *mddev, struct md_rdev *rdev); 587 extern void md_finish_reshape(struct mddev *mddev); 588 589 extern int mddev_congested(struct mddev *mddev, int bits); 590 extern void md_flush_request(struct mddev *mddev, struct bio *bio); 591 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev, 592 sector_t sector, int size, struct page *page); 593 extern void md_super_wait(struct mddev *mddev); 594 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, 595 struct page *page, int rw, bool metadata_op); 596 extern void md_do_sync(struct md_thread *thread); 597 extern void md_new_event(struct mddev *mddev); 598 extern int md_allow_write(struct mddev *mddev); 599 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev); 600 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors); 601 extern int md_check_no_bitmap(struct mddev *mddev); 602 extern int md_integrity_register(struct mddev *mddev); 603 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev); 604 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale); 605 extern void restore_bitmap_write_access(struct file *file); 606 607 extern void mddev_init(struct mddev *mddev); 608 extern int md_run(struct mddev *mddev); 609 extern void md_stop(struct mddev *mddev); 610 extern void md_stop_writes(struct mddev *mddev); 611 extern int md_rdev_init(struct md_rdev *rdev); 612 extern void md_rdev_clear(struct md_rdev *rdev); 613 614 extern void mddev_suspend(struct mddev *mddev); 615 extern void mddev_resume(struct mddev *mddev); 616 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, 617 struct mddev *mddev); 618 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, 619 struct mddev *mddev); 620 621 extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule); 622 static inline int mddev_check_plugged(struct mddev *mddev) 623 { 624 return !!blk_check_plugged(md_unplug, mddev, 625 sizeof(struct blk_plug_cb)); 626 } 627 #endif /* _MD_MD_H */ 628