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