1 /* 2 md.c : Multiple Devices driver for Linux 3 Copyright (C) 1998, 1999, 2000 Ingo Molnar 4 5 completely rewritten, based on the MD driver code from Marc Zyngier 6 7 Changes: 8 9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar 10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> 11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> 12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su> 13 - kmod support by: Cyrus Durgin 14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> 15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> 16 17 - lots of fixes and improvements to the RAID1/RAID5 and generic 18 RAID code (such as request based resynchronization): 19 20 Neil Brown <neilb@cse.unsw.edu.au>. 21 22 - persistent bitmap code 23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. 24 25 This program is free software; you can redistribute it and/or modify 26 it under the terms of the GNU General Public License as published by 27 the Free Software Foundation; either version 2, or (at your option) 28 any later version. 29 30 You should have received a copy of the GNU General Public License 31 (for example /usr/src/linux/COPYING); if not, write to the Free 32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 33 */ 34 35 #include <linux/module.h> 36 #include <linux/config.h> 37 #include <linux/kthread.h> 38 #include <linux/linkage.h> 39 #include <linux/raid/md.h> 40 #include <linux/raid/bitmap.h> 41 #include <linux/sysctl.h> 42 #include <linux/devfs_fs_kernel.h> 43 #include <linux/buffer_head.h> /* for invalidate_bdev */ 44 #include <linux/suspend.h> 45 46 #include <linux/init.h> 47 48 #include <linux/file.h> 49 50 #ifdef CONFIG_KMOD 51 #include <linux/kmod.h> 52 #endif 53 54 #include <asm/unaligned.h> 55 56 #define MAJOR_NR MD_MAJOR 57 #define MD_DRIVER 58 59 /* 63 partitions with the alternate major number (mdp) */ 60 #define MdpMinorShift 6 61 62 #define DEBUG 0 63 #define dprintk(x...) ((void)(DEBUG && printk(x))) 64 65 66 #ifndef MODULE 67 static void autostart_arrays (int part); 68 #endif 69 70 static mdk_personality_t *pers[MAX_PERSONALITY]; 71 static DEFINE_SPINLOCK(pers_lock); 72 73 /* 74 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' 75 * is 1000 KB/sec, so the extra system load does not show up that much. 76 * Increase it if you want to have more _guaranteed_ speed. Note that 77 * the RAID driver will use the maximum available bandwidth if the IO 78 * subsystem is idle. There is also an 'absolute maximum' reconstruction 79 * speed limit - in case reconstruction slows down your system despite 80 * idle IO detection. 81 * 82 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. 83 */ 84 85 static int sysctl_speed_limit_min = 1000; 86 static int sysctl_speed_limit_max = 200000; 87 88 static struct ctl_table_header *raid_table_header; 89 90 static ctl_table raid_table[] = { 91 { 92 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, 93 .procname = "speed_limit_min", 94 .data = &sysctl_speed_limit_min, 95 .maxlen = sizeof(int), 96 .mode = 0644, 97 .proc_handler = &proc_dointvec, 98 }, 99 { 100 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, 101 .procname = "speed_limit_max", 102 .data = &sysctl_speed_limit_max, 103 .maxlen = sizeof(int), 104 .mode = 0644, 105 .proc_handler = &proc_dointvec, 106 }, 107 { .ctl_name = 0 } 108 }; 109 110 static ctl_table raid_dir_table[] = { 111 { 112 .ctl_name = DEV_RAID, 113 .procname = "raid", 114 .maxlen = 0, 115 .mode = 0555, 116 .child = raid_table, 117 }, 118 { .ctl_name = 0 } 119 }; 120 121 static ctl_table raid_root_table[] = { 122 { 123 .ctl_name = CTL_DEV, 124 .procname = "dev", 125 .maxlen = 0, 126 .mode = 0555, 127 .child = raid_dir_table, 128 }, 129 { .ctl_name = 0 } 130 }; 131 132 static struct block_device_operations md_fops; 133 134 static int start_readonly; 135 136 /* 137 * Enables to iterate over all existing md arrays 138 * all_mddevs_lock protects this list. 139 */ 140 static LIST_HEAD(all_mddevs); 141 static DEFINE_SPINLOCK(all_mddevs_lock); 142 143 144 /* 145 * iterates through all used mddevs in the system. 146 * We take care to grab the all_mddevs_lock whenever navigating 147 * the list, and to always hold a refcount when unlocked. 148 * Any code which breaks out of this loop while own 149 * a reference to the current mddev and must mddev_put it. 150 */ 151 #define ITERATE_MDDEV(mddev,tmp) \ 152 \ 153 for (({ spin_lock(&all_mddevs_lock); \ 154 tmp = all_mddevs.next; \ 155 mddev = NULL;}); \ 156 ({ if (tmp != &all_mddevs) \ 157 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ 158 spin_unlock(&all_mddevs_lock); \ 159 if (mddev) mddev_put(mddev); \ 160 mddev = list_entry(tmp, mddev_t, all_mddevs); \ 161 tmp != &all_mddevs;}); \ 162 ({ spin_lock(&all_mddevs_lock); \ 163 tmp = tmp->next;}) \ 164 ) 165 166 167 static int md_fail_request (request_queue_t *q, struct bio *bio) 168 { 169 bio_io_error(bio, bio->bi_size); 170 return 0; 171 } 172 173 static inline mddev_t *mddev_get(mddev_t *mddev) 174 { 175 atomic_inc(&mddev->active); 176 return mddev; 177 } 178 179 static void mddev_put(mddev_t *mddev) 180 { 181 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 182 return; 183 if (!mddev->raid_disks && list_empty(&mddev->disks)) { 184 list_del(&mddev->all_mddevs); 185 blk_put_queue(mddev->queue); 186 kobject_unregister(&mddev->kobj); 187 } 188 spin_unlock(&all_mddevs_lock); 189 } 190 191 static mddev_t * mddev_find(dev_t unit) 192 { 193 mddev_t *mddev, *new = NULL; 194 195 retry: 196 spin_lock(&all_mddevs_lock); 197 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 198 if (mddev->unit == unit) { 199 mddev_get(mddev); 200 spin_unlock(&all_mddevs_lock); 201 kfree(new); 202 return mddev; 203 } 204 205 if (new) { 206 list_add(&new->all_mddevs, &all_mddevs); 207 spin_unlock(&all_mddevs_lock); 208 return new; 209 } 210 spin_unlock(&all_mddevs_lock); 211 212 new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL); 213 if (!new) 214 return NULL; 215 216 memset(new, 0, sizeof(*new)); 217 218 new->unit = unit; 219 if (MAJOR(unit) == MD_MAJOR) 220 new->md_minor = MINOR(unit); 221 else 222 new->md_minor = MINOR(unit) >> MdpMinorShift; 223 224 init_MUTEX(&new->reconfig_sem); 225 INIT_LIST_HEAD(&new->disks); 226 INIT_LIST_HEAD(&new->all_mddevs); 227 init_timer(&new->safemode_timer); 228 atomic_set(&new->active, 1); 229 spin_lock_init(&new->write_lock); 230 init_waitqueue_head(&new->sb_wait); 231 232 new->queue = blk_alloc_queue(GFP_KERNEL); 233 if (!new->queue) { 234 kfree(new); 235 return NULL; 236 } 237 238 blk_queue_make_request(new->queue, md_fail_request); 239 240 goto retry; 241 } 242 243 static inline int mddev_lock(mddev_t * mddev) 244 { 245 return down_interruptible(&mddev->reconfig_sem); 246 } 247 248 static inline void mddev_lock_uninterruptible(mddev_t * mddev) 249 { 250 down(&mddev->reconfig_sem); 251 } 252 253 static inline int mddev_trylock(mddev_t * mddev) 254 { 255 return down_trylock(&mddev->reconfig_sem); 256 } 257 258 static inline void mddev_unlock(mddev_t * mddev) 259 { 260 up(&mddev->reconfig_sem); 261 262 md_wakeup_thread(mddev->thread); 263 } 264 265 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) 266 { 267 mdk_rdev_t * rdev; 268 struct list_head *tmp; 269 270 ITERATE_RDEV(mddev,rdev,tmp) { 271 if (rdev->desc_nr == nr) 272 return rdev; 273 } 274 return NULL; 275 } 276 277 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) 278 { 279 struct list_head *tmp; 280 mdk_rdev_t *rdev; 281 282 ITERATE_RDEV(mddev,rdev,tmp) { 283 if (rdev->bdev->bd_dev == dev) 284 return rdev; 285 } 286 return NULL; 287 } 288 289 static inline sector_t calc_dev_sboffset(struct block_device *bdev) 290 { 291 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; 292 return MD_NEW_SIZE_BLOCKS(size); 293 } 294 295 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size) 296 { 297 sector_t size; 298 299 size = rdev->sb_offset; 300 301 if (chunk_size) 302 size &= ~((sector_t)chunk_size/1024 - 1); 303 return size; 304 } 305 306 static int alloc_disk_sb(mdk_rdev_t * rdev) 307 { 308 if (rdev->sb_page) 309 MD_BUG(); 310 311 rdev->sb_page = alloc_page(GFP_KERNEL); 312 if (!rdev->sb_page) { 313 printk(KERN_ALERT "md: out of memory.\n"); 314 return -EINVAL; 315 } 316 317 return 0; 318 } 319 320 static void free_disk_sb(mdk_rdev_t * rdev) 321 { 322 if (rdev->sb_page) { 323 page_cache_release(rdev->sb_page); 324 rdev->sb_loaded = 0; 325 rdev->sb_page = NULL; 326 rdev->sb_offset = 0; 327 rdev->size = 0; 328 } 329 } 330 331 332 static int super_written(struct bio *bio, unsigned int bytes_done, int error) 333 { 334 mdk_rdev_t *rdev = bio->bi_private; 335 mddev_t *mddev = rdev->mddev; 336 if (bio->bi_size) 337 return 1; 338 339 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) 340 md_error(mddev, rdev); 341 342 if (atomic_dec_and_test(&mddev->pending_writes)) 343 wake_up(&mddev->sb_wait); 344 bio_put(bio); 345 return 0; 346 } 347 348 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error) 349 { 350 struct bio *bio2 = bio->bi_private; 351 mdk_rdev_t *rdev = bio2->bi_private; 352 mddev_t *mddev = rdev->mddev; 353 if (bio->bi_size) 354 return 1; 355 356 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && 357 error == -EOPNOTSUPP) { 358 unsigned long flags; 359 /* barriers don't appear to be supported :-( */ 360 set_bit(BarriersNotsupp, &rdev->flags); 361 mddev->barriers_work = 0; 362 spin_lock_irqsave(&mddev->write_lock, flags); 363 bio2->bi_next = mddev->biolist; 364 mddev->biolist = bio2; 365 spin_unlock_irqrestore(&mddev->write_lock, flags); 366 wake_up(&mddev->sb_wait); 367 bio_put(bio); 368 return 0; 369 } 370 bio_put(bio2); 371 bio->bi_private = rdev; 372 return super_written(bio, bytes_done, error); 373 } 374 375 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, 376 sector_t sector, int size, struct page *page) 377 { 378 /* write first size bytes of page to sector of rdev 379 * Increment mddev->pending_writes before returning 380 * and decrement it on completion, waking up sb_wait 381 * if zero is reached. 382 * If an error occurred, call md_error 383 * 384 * As we might need to resubmit the request if BIO_RW_BARRIER 385 * causes ENOTSUPP, we allocate a spare bio... 386 */ 387 struct bio *bio = bio_alloc(GFP_NOIO, 1); 388 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC); 389 390 bio->bi_bdev = rdev->bdev; 391 bio->bi_sector = sector; 392 bio_add_page(bio, page, size, 0); 393 bio->bi_private = rdev; 394 bio->bi_end_io = super_written; 395 bio->bi_rw = rw; 396 397 atomic_inc(&mddev->pending_writes); 398 if (!test_bit(BarriersNotsupp, &rdev->flags)) { 399 struct bio *rbio; 400 rw |= (1<<BIO_RW_BARRIER); 401 rbio = bio_clone(bio, GFP_NOIO); 402 rbio->bi_private = bio; 403 rbio->bi_end_io = super_written_barrier; 404 submit_bio(rw, rbio); 405 } else 406 submit_bio(rw, bio); 407 } 408 409 void md_super_wait(mddev_t *mddev) 410 { 411 /* wait for all superblock writes that were scheduled to complete. 412 * if any had to be retried (due to BARRIER problems), retry them 413 */ 414 DEFINE_WAIT(wq); 415 for(;;) { 416 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 417 if (atomic_read(&mddev->pending_writes)==0) 418 break; 419 while (mddev->biolist) { 420 struct bio *bio; 421 spin_lock_irq(&mddev->write_lock); 422 bio = mddev->biolist; 423 mddev->biolist = bio->bi_next ; 424 bio->bi_next = NULL; 425 spin_unlock_irq(&mddev->write_lock); 426 submit_bio(bio->bi_rw, bio); 427 } 428 schedule(); 429 } 430 finish_wait(&mddev->sb_wait, &wq); 431 } 432 433 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error) 434 { 435 if (bio->bi_size) 436 return 1; 437 438 complete((struct completion*)bio->bi_private); 439 return 0; 440 } 441 442 int sync_page_io(struct block_device *bdev, sector_t sector, int size, 443 struct page *page, int rw) 444 { 445 struct bio *bio = bio_alloc(GFP_NOIO, 1); 446 struct completion event; 447 int ret; 448 449 rw |= (1 << BIO_RW_SYNC); 450 451 bio->bi_bdev = bdev; 452 bio->bi_sector = sector; 453 bio_add_page(bio, page, size, 0); 454 init_completion(&event); 455 bio->bi_private = &event; 456 bio->bi_end_io = bi_complete; 457 submit_bio(rw, bio); 458 wait_for_completion(&event); 459 460 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 461 bio_put(bio); 462 return ret; 463 } 464 465 static int read_disk_sb(mdk_rdev_t * rdev, int size) 466 { 467 char b[BDEVNAME_SIZE]; 468 if (!rdev->sb_page) { 469 MD_BUG(); 470 return -EINVAL; 471 } 472 if (rdev->sb_loaded) 473 return 0; 474 475 476 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ)) 477 goto fail; 478 rdev->sb_loaded = 1; 479 return 0; 480 481 fail: 482 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 483 bdevname(rdev->bdev,b)); 484 return -EINVAL; 485 } 486 487 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 488 { 489 if ( (sb1->set_uuid0 == sb2->set_uuid0) && 490 (sb1->set_uuid1 == sb2->set_uuid1) && 491 (sb1->set_uuid2 == sb2->set_uuid2) && 492 (sb1->set_uuid3 == sb2->set_uuid3)) 493 494 return 1; 495 496 return 0; 497 } 498 499 500 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 501 { 502 int ret; 503 mdp_super_t *tmp1, *tmp2; 504 505 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 506 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 507 508 if (!tmp1 || !tmp2) { 509 ret = 0; 510 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n"); 511 goto abort; 512 } 513 514 *tmp1 = *sb1; 515 *tmp2 = *sb2; 516 517 /* 518 * nr_disks is not constant 519 */ 520 tmp1->nr_disks = 0; 521 tmp2->nr_disks = 0; 522 523 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4)) 524 ret = 0; 525 else 526 ret = 1; 527 528 abort: 529 kfree(tmp1); 530 kfree(tmp2); 531 return ret; 532 } 533 534 static unsigned int calc_sb_csum(mdp_super_t * sb) 535 { 536 unsigned int disk_csum, csum; 537 538 disk_csum = sb->sb_csum; 539 sb->sb_csum = 0; 540 csum = csum_partial((void *)sb, MD_SB_BYTES, 0); 541 sb->sb_csum = disk_csum; 542 return csum; 543 } 544 545 546 /* 547 * Handle superblock details. 548 * We want to be able to handle multiple superblock formats 549 * so we have a common interface to them all, and an array of 550 * different handlers. 551 * We rely on user-space to write the initial superblock, and support 552 * reading and updating of superblocks. 553 * Interface methods are: 554 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) 555 * loads and validates a superblock on dev. 556 * if refdev != NULL, compare superblocks on both devices 557 * Return: 558 * 0 - dev has a superblock that is compatible with refdev 559 * 1 - dev has a superblock that is compatible and newer than refdev 560 * so dev should be used as the refdev in future 561 * -EINVAL superblock incompatible or invalid 562 * -othererror e.g. -EIO 563 * 564 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) 565 * Verify that dev is acceptable into mddev. 566 * The first time, mddev->raid_disks will be 0, and data from 567 * dev should be merged in. Subsequent calls check that dev 568 * is new enough. Return 0 or -EINVAL 569 * 570 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) 571 * Update the superblock for rdev with data in mddev 572 * This does not write to disc. 573 * 574 */ 575 576 struct super_type { 577 char *name; 578 struct module *owner; 579 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version); 580 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); 581 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); 582 }; 583 584 /* 585 * load_super for 0.90.0 586 */ 587 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 588 { 589 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 590 mdp_super_t *sb; 591 int ret; 592 sector_t sb_offset; 593 594 /* 595 * Calculate the position of the superblock, 596 * it's at the end of the disk. 597 * 598 * It also happens to be a multiple of 4Kb. 599 */ 600 sb_offset = calc_dev_sboffset(rdev->bdev); 601 rdev->sb_offset = sb_offset; 602 603 ret = read_disk_sb(rdev, MD_SB_BYTES); 604 if (ret) return ret; 605 606 ret = -EINVAL; 607 608 bdevname(rdev->bdev, b); 609 sb = (mdp_super_t*)page_address(rdev->sb_page); 610 611 if (sb->md_magic != MD_SB_MAGIC) { 612 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 613 b); 614 goto abort; 615 } 616 617 if (sb->major_version != 0 || 618 sb->minor_version != 90) { 619 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 620 sb->major_version, sb->minor_version, 621 b); 622 goto abort; 623 } 624 625 if (sb->raid_disks <= 0) 626 goto abort; 627 628 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) { 629 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 630 b); 631 goto abort; 632 } 633 634 rdev->preferred_minor = sb->md_minor; 635 rdev->data_offset = 0; 636 rdev->sb_size = MD_SB_BYTES; 637 638 if (sb->level == LEVEL_MULTIPATH) 639 rdev->desc_nr = -1; 640 else 641 rdev->desc_nr = sb->this_disk.number; 642 643 if (refdev == 0) 644 ret = 1; 645 else { 646 __u64 ev1, ev2; 647 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); 648 if (!uuid_equal(refsb, sb)) { 649 printk(KERN_WARNING "md: %s has different UUID to %s\n", 650 b, bdevname(refdev->bdev,b2)); 651 goto abort; 652 } 653 if (!sb_equal(refsb, sb)) { 654 printk(KERN_WARNING "md: %s has same UUID" 655 " but different superblock to %s\n", 656 b, bdevname(refdev->bdev, b2)); 657 goto abort; 658 } 659 ev1 = md_event(sb); 660 ev2 = md_event(refsb); 661 if (ev1 > ev2) 662 ret = 1; 663 else 664 ret = 0; 665 } 666 rdev->size = calc_dev_size(rdev, sb->chunk_size); 667 668 abort: 669 return ret; 670 } 671 672 /* 673 * validate_super for 0.90.0 674 */ 675 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) 676 { 677 mdp_disk_t *desc; 678 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); 679 680 rdev->raid_disk = -1; 681 rdev->flags = 0; 682 if (mddev->raid_disks == 0) { 683 mddev->major_version = 0; 684 mddev->minor_version = sb->minor_version; 685 mddev->patch_version = sb->patch_version; 686 mddev->persistent = ! sb->not_persistent; 687 mddev->chunk_size = sb->chunk_size; 688 mddev->ctime = sb->ctime; 689 mddev->utime = sb->utime; 690 mddev->level = sb->level; 691 mddev->layout = sb->layout; 692 mddev->raid_disks = sb->raid_disks; 693 mddev->size = sb->size; 694 mddev->events = md_event(sb); 695 mddev->bitmap_offset = 0; 696 mddev->default_bitmap_offset = MD_SB_BYTES >> 9; 697 698 if (sb->state & (1<<MD_SB_CLEAN)) 699 mddev->recovery_cp = MaxSector; 700 else { 701 if (sb->events_hi == sb->cp_events_hi && 702 sb->events_lo == sb->cp_events_lo) { 703 mddev->recovery_cp = sb->recovery_cp; 704 } else 705 mddev->recovery_cp = 0; 706 } 707 708 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 709 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 710 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 711 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 712 713 mddev->max_disks = MD_SB_DISKS; 714 715 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 716 mddev->bitmap_file == NULL) { 717 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) { 718 /* FIXME use a better test */ 719 printk(KERN_WARNING "md: bitmaps only support for raid1\n"); 720 return -EINVAL; 721 } 722 mddev->bitmap_offset = mddev->default_bitmap_offset; 723 } 724 725 } else if (mddev->pers == NULL) { 726 /* Insist on good event counter while assembling */ 727 __u64 ev1 = md_event(sb); 728 ++ev1; 729 if (ev1 < mddev->events) 730 return -EINVAL; 731 } else if (mddev->bitmap) { 732 /* if adding to array with a bitmap, then we can accept an 733 * older device ... but not too old. 734 */ 735 __u64 ev1 = md_event(sb); 736 if (ev1 < mddev->bitmap->events_cleared) 737 return 0; 738 } else /* just a hot-add of a new device, leave raid_disk at -1 */ 739 return 0; 740 741 if (mddev->level != LEVEL_MULTIPATH) { 742 desc = sb->disks + rdev->desc_nr; 743 744 if (desc->state & (1<<MD_DISK_FAULTY)) 745 set_bit(Faulty, &rdev->flags); 746 else if (desc->state & (1<<MD_DISK_SYNC) && 747 desc->raid_disk < mddev->raid_disks) { 748 set_bit(In_sync, &rdev->flags); 749 rdev->raid_disk = desc->raid_disk; 750 } 751 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 752 set_bit(WriteMostly, &rdev->flags); 753 } else /* MULTIPATH are always insync */ 754 set_bit(In_sync, &rdev->flags); 755 return 0; 756 } 757 758 /* 759 * sync_super for 0.90.0 760 */ 761 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) 762 { 763 mdp_super_t *sb; 764 struct list_head *tmp; 765 mdk_rdev_t *rdev2; 766 int next_spare = mddev->raid_disks; 767 768 769 /* make rdev->sb match mddev data.. 770 * 771 * 1/ zero out disks 772 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 773 * 3/ any empty disks < next_spare become removed 774 * 775 * disks[0] gets initialised to REMOVED because 776 * we cannot be sure from other fields if it has 777 * been initialised or not. 778 */ 779 int i; 780 int active=0, working=0,failed=0,spare=0,nr_disks=0; 781 782 rdev->sb_size = MD_SB_BYTES; 783 784 sb = (mdp_super_t*)page_address(rdev->sb_page); 785 786 memset(sb, 0, sizeof(*sb)); 787 788 sb->md_magic = MD_SB_MAGIC; 789 sb->major_version = mddev->major_version; 790 sb->minor_version = mddev->minor_version; 791 sb->patch_version = mddev->patch_version; 792 sb->gvalid_words = 0; /* ignored */ 793 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 794 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 795 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 796 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 797 798 sb->ctime = mddev->ctime; 799 sb->level = mddev->level; 800 sb->size = mddev->size; 801 sb->raid_disks = mddev->raid_disks; 802 sb->md_minor = mddev->md_minor; 803 sb->not_persistent = !mddev->persistent; 804 sb->utime = mddev->utime; 805 sb->state = 0; 806 sb->events_hi = (mddev->events>>32); 807 sb->events_lo = (u32)mddev->events; 808 809 if (mddev->in_sync) 810 { 811 sb->recovery_cp = mddev->recovery_cp; 812 sb->cp_events_hi = (mddev->events>>32); 813 sb->cp_events_lo = (u32)mddev->events; 814 if (mddev->recovery_cp == MaxSector) 815 sb->state = (1<< MD_SB_CLEAN); 816 } else 817 sb->recovery_cp = 0; 818 819 sb->layout = mddev->layout; 820 sb->chunk_size = mddev->chunk_size; 821 822 if (mddev->bitmap && mddev->bitmap_file == NULL) 823 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 824 825 sb->disks[0].state = (1<<MD_DISK_REMOVED); 826 ITERATE_RDEV(mddev,rdev2,tmp) { 827 mdp_disk_t *d; 828 int desc_nr; 829 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) 830 && !test_bit(Faulty, &rdev2->flags)) 831 desc_nr = rdev2->raid_disk; 832 else 833 desc_nr = next_spare++; 834 rdev2->desc_nr = desc_nr; 835 d = &sb->disks[rdev2->desc_nr]; 836 nr_disks++; 837 d->number = rdev2->desc_nr; 838 d->major = MAJOR(rdev2->bdev->bd_dev); 839 d->minor = MINOR(rdev2->bdev->bd_dev); 840 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) 841 && !test_bit(Faulty, &rdev2->flags)) 842 d->raid_disk = rdev2->raid_disk; 843 else 844 d->raid_disk = rdev2->desc_nr; /* compatibility */ 845 if (test_bit(Faulty, &rdev2->flags)) { 846 d->state = (1<<MD_DISK_FAULTY); 847 failed++; 848 } else if (test_bit(In_sync, &rdev2->flags)) { 849 d->state = (1<<MD_DISK_ACTIVE); 850 d->state |= (1<<MD_DISK_SYNC); 851 active++; 852 working++; 853 } else { 854 d->state = 0; 855 spare++; 856 working++; 857 } 858 if (test_bit(WriteMostly, &rdev2->flags)) 859 d->state |= (1<<MD_DISK_WRITEMOSTLY); 860 } 861 /* now set the "removed" and "faulty" bits on any missing devices */ 862 for (i=0 ; i < mddev->raid_disks ; i++) { 863 mdp_disk_t *d = &sb->disks[i]; 864 if (d->state == 0 && d->number == 0) { 865 d->number = i; 866 d->raid_disk = i; 867 d->state = (1<<MD_DISK_REMOVED); 868 d->state |= (1<<MD_DISK_FAULTY); 869 failed++; 870 } 871 } 872 sb->nr_disks = nr_disks; 873 sb->active_disks = active; 874 sb->working_disks = working; 875 sb->failed_disks = failed; 876 sb->spare_disks = spare; 877 878 sb->this_disk = sb->disks[rdev->desc_nr]; 879 sb->sb_csum = calc_sb_csum(sb); 880 } 881 882 /* 883 * version 1 superblock 884 */ 885 886 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb) 887 { 888 unsigned int disk_csum, csum; 889 unsigned long long newcsum; 890 int size = 256 + le32_to_cpu(sb->max_dev)*2; 891 unsigned int *isuper = (unsigned int*)sb; 892 int i; 893 894 disk_csum = sb->sb_csum; 895 sb->sb_csum = 0; 896 newcsum = 0; 897 for (i=0; size>=4; size -= 4 ) 898 newcsum += le32_to_cpu(*isuper++); 899 900 if (size == 2) 901 newcsum += le16_to_cpu(*(unsigned short*) isuper); 902 903 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 904 sb->sb_csum = disk_csum; 905 return cpu_to_le32(csum); 906 } 907 908 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 909 { 910 struct mdp_superblock_1 *sb; 911 int ret; 912 sector_t sb_offset; 913 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 914 int bmask; 915 916 /* 917 * Calculate the position of the superblock. 918 * It is always aligned to a 4K boundary and 919 * depeding on minor_version, it can be: 920 * 0: At least 8K, but less than 12K, from end of device 921 * 1: At start of device 922 * 2: 4K from start of device. 923 */ 924 switch(minor_version) { 925 case 0: 926 sb_offset = rdev->bdev->bd_inode->i_size >> 9; 927 sb_offset -= 8*2; 928 sb_offset &= ~(sector_t)(4*2-1); 929 /* convert from sectors to K */ 930 sb_offset /= 2; 931 break; 932 case 1: 933 sb_offset = 0; 934 break; 935 case 2: 936 sb_offset = 4; 937 break; 938 default: 939 return -EINVAL; 940 } 941 rdev->sb_offset = sb_offset; 942 943 /* superblock is rarely larger than 1K, but it can be larger, 944 * and it is safe to read 4k, so we do that 945 */ 946 ret = read_disk_sb(rdev, 4096); 947 if (ret) return ret; 948 949 950 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 951 952 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 953 sb->major_version != cpu_to_le32(1) || 954 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 955 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) || 956 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 957 return -EINVAL; 958 959 if (calc_sb_1_csum(sb) != sb->sb_csum) { 960 printk("md: invalid superblock checksum on %s\n", 961 bdevname(rdev->bdev,b)); 962 return -EINVAL; 963 } 964 if (le64_to_cpu(sb->data_size) < 10) { 965 printk("md: data_size too small on %s\n", 966 bdevname(rdev->bdev,b)); 967 return -EINVAL; 968 } 969 rdev->preferred_minor = 0xffff; 970 rdev->data_offset = le64_to_cpu(sb->data_offset); 971 972 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 973 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1; 974 if (rdev->sb_size & bmask) 975 rdev-> sb_size = (rdev->sb_size | bmask)+1; 976 977 if (refdev == 0) 978 return 1; 979 else { 980 __u64 ev1, ev2; 981 struct mdp_superblock_1 *refsb = 982 (struct mdp_superblock_1*)page_address(refdev->sb_page); 983 984 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 985 sb->level != refsb->level || 986 sb->layout != refsb->layout || 987 sb->chunksize != refsb->chunksize) { 988 printk(KERN_WARNING "md: %s has strangely different" 989 " superblock to %s\n", 990 bdevname(rdev->bdev,b), 991 bdevname(refdev->bdev,b2)); 992 return -EINVAL; 993 } 994 ev1 = le64_to_cpu(sb->events); 995 ev2 = le64_to_cpu(refsb->events); 996 997 if (ev1 > ev2) 998 return 1; 999 } 1000 if (minor_version) 1001 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2; 1002 else 1003 rdev->size = rdev->sb_offset; 1004 if (rdev->size < le64_to_cpu(sb->data_size)/2) 1005 return -EINVAL; 1006 rdev->size = le64_to_cpu(sb->data_size)/2; 1007 if (le32_to_cpu(sb->chunksize)) 1008 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1); 1009 return 0; 1010 } 1011 1012 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1013 { 1014 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1015 1016 rdev->raid_disk = -1; 1017 rdev->flags = 0; 1018 if (mddev->raid_disks == 0) { 1019 mddev->major_version = 1; 1020 mddev->patch_version = 0; 1021 mddev->persistent = 1; 1022 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9; 1023 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1024 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1025 mddev->level = le32_to_cpu(sb->level); 1026 mddev->layout = le32_to_cpu(sb->layout); 1027 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1028 mddev->size = le64_to_cpu(sb->size)/2; 1029 mddev->events = le64_to_cpu(sb->events); 1030 mddev->bitmap_offset = 0; 1031 mddev->default_bitmap_offset = 0; 1032 mddev->default_bitmap_offset = 1024; 1033 1034 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1035 memcpy(mddev->uuid, sb->set_uuid, 16); 1036 1037 mddev->max_disks = (4096-256)/2; 1038 1039 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1040 mddev->bitmap_file == NULL ) { 1041 if (mddev->level != 1) { 1042 printk(KERN_WARNING "md: bitmaps only supported for raid1\n"); 1043 return -EINVAL; 1044 } 1045 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); 1046 } 1047 } else if (mddev->pers == NULL) { 1048 /* Insist of good event counter while assembling */ 1049 __u64 ev1 = le64_to_cpu(sb->events); 1050 ++ev1; 1051 if (ev1 < mddev->events) 1052 return -EINVAL; 1053 } else if (mddev->bitmap) { 1054 /* If adding to array with a bitmap, then we can accept an 1055 * older device, but not too old. 1056 */ 1057 __u64 ev1 = le64_to_cpu(sb->events); 1058 if (ev1 < mddev->bitmap->events_cleared) 1059 return 0; 1060 } else /* just a hot-add of a new device, leave raid_disk at -1 */ 1061 return 0; 1062 1063 if (mddev->level != LEVEL_MULTIPATH) { 1064 int role; 1065 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1066 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1067 switch(role) { 1068 case 0xffff: /* spare */ 1069 break; 1070 case 0xfffe: /* faulty */ 1071 set_bit(Faulty, &rdev->flags); 1072 break; 1073 default: 1074 set_bit(In_sync, &rdev->flags); 1075 rdev->raid_disk = role; 1076 break; 1077 } 1078 if (sb->devflags & WriteMostly1) 1079 set_bit(WriteMostly, &rdev->flags); 1080 } else /* MULTIPATH are always insync */ 1081 set_bit(In_sync, &rdev->flags); 1082 1083 return 0; 1084 } 1085 1086 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1087 { 1088 struct mdp_superblock_1 *sb; 1089 struct list_head *tmp; 1090 mdk_rdev_t *rdev2; 1091 int max_dev, i; 1092 /* make rdev->sb match mddev and rdev data. */ 1093 1094 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1095 1096 sb->feature_map = 0; 1097 sb->pad0 = 0; 1098 memset(sb->pad1, 0, sizeof(sb->pad1)); 1099 memset(sb->pad2, 0, sizeof(sb->pad2)); 1100 memset(sb->pad3, 0, sizeof(sb->pad3)); 1101 1102 sb->utime = cpu_to_le64((__u64)mddev->utime); 1103 sb->events = cpu_to_le64(mddev->events); 1104 if (mddev->in_sync) 1105 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1106 else 1107 sb->resync_offset = cpu_to_le64(0); 1108 1109 if (mddev->bitmap && mddev->bitmap_file == NULL) { 1110 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); 1111 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); 1112 } 1113 1114 max_dev = 0; 1115 ITERATE_RDEV(mddev,rdev2,tmp) 1116 if (rdev2->desc_nr+1 > max_dev) 1117 max_dev = rdev2->desc_nr+1; 1118 1119 sb->max_dev = cpu_to_le32(max_dev); 1120 for (i=0; i<max_dev;i++) 1121 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1122 1123 ITERATE_RDEV(mddev,rdev2,tmp) { 1124 i = rdev2->desc_nr; 1125 if (test_bit(Faulty, &rdev2->flags)) 1126 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1127 else if (test_bit(In_sync, &rdev2->flags)) 1128 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1129 else 1130 sb->dev_roles[i] = cpu_to_le16(0xffff); 1131 } 1132 1133 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */ 1134 sb->sb_csum = calc_sb_1_csum(sb); 1135 } 1136 1137 1138 static struct super_type super_types[] = { 1139 [0] = { 1140 .name = "0.90.0", 1141 .owner = THIS_MODULE, 1142 .load_super = super_90_load, 1143 .validate_super = super_90_validate, 1144 .sync_super = super_90_sync, 1145 }, 1146 [1] = { 1147 .name = "md-1", 1148 .owner = THIS_MODULE, 1149 .load_super = super_1_load, 1150 .validate_super = super_1_validate, 1151 .sync_super = super_1_sync, 1152 }, 1153 }; 1154 1155 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev) 1156 { 1157 struct list_head *tmp; 1158 mdk_rdev_t *rdev; 1159 1160 ITERATE_RDEV(mddev,rdev,tmp) 1161 if (rdev->bdev->bd_contains == dev->bdev->bd_contains) 1162 return rdev; 1163 1164 return NULL; 1165 } 1166 1167 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) 1168 { 1169 struct list_head *tmp; 1170 mdk_rdev_t *rdev; 1171 1172 ITERATE_RDEV(mddev1,rdev,tmp) 1173 if (match_dev_unit(mddev2, rdev)) 1174 return 1; 1175 1176 return 0; 1177 } 1178 1179 static LIST_HEAD(pending_raid_disks); 1180 1181 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) 1182 { 1183 mdk_rdev_t *same_pdev; 1184 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1185 struct kobject *ko; 1186 1187 if (rdev->mddev) { 1188 MD_BUG(); 1189 return -EINVAL; 1190 } 1191 same_pdev = match_dev_unit(mddev, rdev); 1192 if (same_pdev) 1193 printk(KERN_WARNING 1194 "%s: WARNING: %s appears to be on the same physical" 1195 " disk as %s. True\n protection against single-disk" 1196 " failure might be compromised.\n", 1197 mdname(mddev), bdevname(rdev->bdev,b), 1198 bdevname(same_pdev->bdev,b2)); 1199 1200 /* Verify rdev->desc_nr is unique. 1201 * If it is -1, assign a free number, else 1202 * check number is not in use 1203 */ 1204 if (rdev->desc_nr < 0) { 1205 int choice = 0; 1206 if (mddev->pers) choice = mddev->raid_disks; 1207 while (find_rdev_nr(mddev, choice)) 1208 choice++; 1209 rdev->desc_nr = choice; 1210 } else { 1211 if (find_rdev_nr(mddev, rdev->desc_nr)) 1212 return -EBUSY; 1213 } 1214 bdevname(rdev->bdev,b); 1215 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0) 1216 return -ENOMEM; 1217 1218 list_add(&rdev->same_set, &mddev->disks); 1219 rdev->mddev = mddev; 1220 printk(KERN_INFO "md: bind<%s>\n", b); 1221 1222 rdev->kobj.parent = &mddev->kobj; 1223 kobject_add(&rdev->kobj); 1224 1225 if (rdev->bdev->bd_part) 1226 ko = &rdev->bdev->bd_part->kobj; 1227 else 1228 ko = &rdev->bdev->bd_disk->kobj; 1229 sysfs_create_link(&rdev->kobj, ko, "block"); 1230 return 0; 1231 } 1232 1233 static void unbind_rdev_from_array(mdk_rdev_t * rdev) 1234 { 1235 char b[BDEVNAME_SIZE]; 1236 if (!rdev->mddev) { 1237 MD_BUG(); 1238 return; 1239 } 1240 list_del_init(&rdev->same_set); 1241 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 1242 rdev->mddev = NULL; 1243 sysfs_remove_link(&rdev->kobj, "block"); 1244 kobject_del(&rdev->kobj); 1245 } 1246 1247 /* 1248 * prevent the device from being mounted, repartitioned or 1249 * otherwise reused by a RAID array (or any other kernel 1250 * subsystem), by bd_claiming the device. 1251 */ 1252 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev) 1253 { 1254 int err = 0; 1255 struct block_device *bdev; 1256 char b[BDEVNAME_SIZE]; 1257 1258 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); 1259 if (IS_ERR(bdev)) { 1260 printk(KERN_ERR "md: could not open %s.\n", 1261 __bdevname(dev, b)); 1262 return PTR_ERR(bdev); 1263 } 1264 err = bd_claim(bdev, rdev); 1265 if (err) { 1266 printk(KERN_ERR "md: could not bd_claim %s.\n", 1267 bdevname(bdev, b)); 1268 blkdev_put(bdev); 1269 return err; 1270 } 1271 rdev->bdev = bdev; 1272 return err; 1273 } 1274 1275 static void unlock_rdev(mdk_rdev_t *rdev) 1276 { 1277 struct block_device *bdev = rdev->bdev; 1278 rdev->bdev = NULL; 1279 if (!bdev) 1280 MD_BUG(); 1281 bd_release(bdev); 1282 blkdev_put(bdev); 1283 } 1284 1285 void md_autodetect_dev(dev_t dev); 1286 1287 static void export_rdev(mdk_rdev_t * rdev) 1288 { 1289 char b[BDEVNAME_SIZE]; 1290 printk(KERN_INFO "md: export_rdev(%s)\n", 1291 bdevname(rdev->bdev,b)); 1292 if (rdev->mddev) 1293 MD_BUG(); 1294 free_disk_sb(rdev); 1295 list_del_init(&rdev->same_set); 1296 #ifndef MODULE 1297 md_autodetect_dev(rdev->bdev->bd_dev); 1298 #endif 1299 unlock_rdev(rdev); 1300 kobject_put(&rdev->kobj); 1301 } 1302 1303 static void kick_rdev_from_array(mdk_rdev_t * rdev) 1304 { 1305 unbind_rdev_from_array(rdev); 1306 export_rdev(rdev); 1307 } 1308 1309 static void export_array(mddev_t *mddev) 1310 { 1311 struct list_head *tmp; 1312 mdk_rdev_t *rdev; 1313 1314 ITERATE_RDEV(mddev,rdev,tmp) { 1315 if (!rdev->mddev) { 1316 MD_BUG(); 1317 continue; 1318 } 1319 kick_rdev_from_array(rdev); 1320 } 1321 if (!list_empty(&mddev->disks)) 1322 MD_BUG(); 1323 mddev->raid_disks = 0; 1324 mddev->major_version = 0; 1325 } 1326 1327 static void print_desc(mdp_disk_t *desc) 1328 { 1329 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 1330 desc->major,desc->minor,desc->raid_disk,desc->state); 1331 } 1332 1333 static void print_sb(mdp_super_t *sb) 1334 { 1335 int i; 1336 1337 printk(KERN_INFO 1338 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 1339 sb->major_version, sb->minor_version, sb->patch_version, 1340 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 1341 sb->ctime); 1342 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 1343 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 1344 sb->md_minor, sb->layout, sb->chunk_size); 1345 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 1346 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 1347 sb->utime, sb->state, sb->active_disks, sb->working_disks, 1348 sb->failed_disks, sb->spare_disks, 1349 sb->sb_csum, (unsigned long)sb->events_lo); 1350 1351 printk(KERN_INFO); 1352 for (i = 0; i < MD_SB_DISKS; i++) { 1353 mdp_disk_t *desc; 1354 1355 desc = sb->disks + i; 1356 if (desc->number || desc->major || desc->minor || 1357 desc->raid_disk || (desc->state && (desc->state != 4))) { 1358 printk(" D %2d: ", i); 1359 print_desc(desc); 1360 } 1361 } 1362 printk(KERN_INFO "md: THIS: "); 1363 print_desc(&sb->this_disk); 1364 1365 } 1366 1367 static void print_rdev(mdk_rdev_t *rdev) 1368 { 1369 char b[BDEVNAME_SIZE]; 1370 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n", 1371 bdevname(rdev->bdev,b), (unsigned long long)rdev->size, 1372 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 1373 rdev->desc_nr); 1374 if (rdev->sb_loaded) { 1375 printk(KERN_INFO "md: rdev superblock:\n"); 1376 print_sb((mdp_super_t*)page_address(rdev->sb_page)); 1377 } else 1378 printk(KERN_INFO "md: no rdev superblock!\n"); 1379 } 1380 1381 void md_print_devices(void) 1382 { 1383 struct list_head *tmp, *tmp2; 1384 mdk_rdev_t *rdev; 1385 mddev_t *mddev; 1386 char b[BDEVNAME_SIZE]; 1387 1388 printk("\n"); 1389 printk("md: **********************************\n"); 1390 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 1391 printk("md: **********************************\n"); 1392 ITERATE_MDDEV(mddev,tmp) { 1393 1394 if (mddev->bitmap) 1395 bitmap_print_sb(mddev->bitmap); 1396 else 1397 printk("%s: ", mdname(mddev)); 1398 ITERATE_RDEV(mddev,rdev,tmp2) 1399 printk("<%s>", bdevname(rdev->bdev,b)); 1400 printk("\n"); 1401 1402 ITERATE_RDEV(mddev,rdev,tmp2) 1403 print_rdev(rdev); 1404 } 1405 printk("md: **********************************\n"); 1406 printk("\n"); 1407 } 1408 1409 1410 static void sync_sbs(mddev_t * mddev) 1411 { 1412 mdk_rdev_t *rdev; 1413 struct list_head *tmp; 1414 1415 ITERATE_RDEV(mddev,rdev,tmp) { 1416 super_types[mddev->major_version]. 1417 sync_super(mddev, rdev); 1418 rdev->sb_loaded = 1; 1419 } 1420 } 1421 1422 static void md_update_sb(mddev_t * mddev) 1423 { 1424 int err; 1425 struct list_head *tmp; 1426 mdk_rdev_t *rdev; 1427 int sync_req; 1428 1429 repeat: 1430 spin_lock_irq(&mddev->write_lock); 1431 sync_req = mddev->in_sync; 1432 mddev->utime = get_seconds(); 1433 mddev->events ++; 1434 1435 if (!mddev->events) { 1436 /* 1437 * oops, this 64-bit counter should never wrap. 1438 * Either we are in around ~1 trillion A.C., assuming 1439 * 1 reboot per second, or we have a bug: 1440 */ 1441 MD_BUG(); 1442 mddev->events --; 1443 } 1444 mddev->sb_dirty = 2; 1445 sync_sbs(mddev); 1446 1447 /* 1448 * do not write anything to disk if using 1449 * nonpersistent superblocks 1450 */ 1451 if (!mddev->persistent) { 1452 mddev->sb_dirty = 0; 1453 spin_unlock_irq(&mddev->write_lock); 1454 wake_up(&mddev->sb_wait); 1455 return; 1456 } 1457 spin_unlock_irq(&mddev->write_lock); 1458 1459 dprintk(KERN_INFO 1460 "md: updating %s RAID superblock on device (in sync %d)\n", 1461 mdname(mddev),mddev->in_sync); 1462 1463 err = bitmap_update_sb(mddev->bitmap); 1464 ITERATE_RDEV(mddev,rdev,tmp) { 1465 char b[BDEVNAME_SIZE]; 1466 dprintk(KERN_INFO "md: "); 1467 if (test_bit(Faulty, &rdev->flags)) 1468 dprintk("(skipping faulty "); 1469 1470 dprintk("%s ", bdevname(rdev->bdev,b)); 1471 if (!test_bit(Faulty, &rdev->flags)) { 1472 md_super_write(mddev,rdev, 1473 rdev->sb_offset<<1, rdev->sb_size, 1474 rdev->sb_page); 1475 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", 1476 bdevname(rdev->bdev,b), 1477 (unsigned long long)rdev->sb_offset); 1478 1479 } else 1480 dprintk(")\n"); 1481 if (mddev->level == LEVEL_MULTIPATH) 1482 /* only need to write one superblock... */ 1483 break; 1484 } 1485 md_super_wait(mddev); 1486 /* if there was a failure, sb_dirty was set to 1, and we re-write super */ 1487 1488 spin_lock_irq(&mddev->write_lock); 1489 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) { 1490 /* have to write it out again */ 1491 spin_unlock_irq(&mddev->write_lock); 1492 goto repeat; 1493 } 1494 mddev->sb_dirty = 0; 1495 spin_unlock_irq(&mddev->write_lock); 1496 wake_up(&mddev->sb_wait); 1497 1498 } 1499 1500 struct rdev_sysfs_entry { 1501 struct attribute attr; 1502 ssize_t (*show)(mdk_rdev_t *, char *); 1503 ssize_t (*store)(mdk_rdev_t *, const char *, size_t); 1504 }; 1505 1506 static ssize_t 1507 rdev_show_state(mdk_rdev_t *rdev, char *page) 1508 { 1509 char *sep = ""; 1510 int len=0; 1511 1512 if (test_bit(Faulty, &rdev->flags)) { 1513 len+= sprintf(page+len, "%sfaulty",sep); 1514 sep = ","; 1515 } 1516 if (test_bit(In_sync, &rdev->flags)) { 1517 len += sprintf(page+len, "%sin_sync",sep); 1518 sep = ","; 1519 } 1520 if (!test_bit(Faulty, &rdev->flags) && 1521 !test_bit(In_sync, &rdev->flags)) { 1522 len += sprintf(page+len, "%sspare", sep); 1523 sep = ","; 1524 } 1525 return len+sprintf(page+len, "\n"); 1526 } 1527 1528 static struct rdev_sysfs_entry rdev_state = { 1529 .attr = {.name = "state", .mode = S_IRUGO }, 1530 .show = rdev_show_state, 1531 }; 1532 1533 static ssize_t 1534 rdev_show_super(mdk_rdev_t *rdev, char *page) 1535 { 1536 if (rdev->sb_loaded && rdev->sb_size) { 1537 memcpy(page, page_address(rdev->sb_page), rdev->sb_size); 1538 return rdev->sb_size; 1539 } else 1540 return 0; 1541 } 1542 static struct rdev_sysfs_entry rdev_super = { 1543 .attr = {.name = "super", .mode = S_IRUGO }, 1544 .show = rdev_show_super, 1545 }; 1546 static struct attribute *rdev_default_attrs[] = { 1547 &rdev_state.attr, 1548 &rdev_super.attr, 1549 NULL, 1550 }; 1551 static ssize_t 1552 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 1553 { 1554 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 1555 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 1556 1557 if (!entry->show) 1558 return -EIO; 1559 return entry->show(rdev, page); 1560 } 1561 1562 static ssize_t 1563 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 1564 const char *page, size_t length) 1565 { 1566 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 1567 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 1568 1569 if (!entry->store) 1570 return -EIO; 1571 return entry->store(rdev, page, length); 1572 } 1573 1574 static void rdev_free(struct kobject *ko) 1575 { 1576 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); 1577 kfree(rdev); 1578 } 1579 static struct sysfs_ops rdev_sysfs_ops = { 1580 .show = rdev_attr_show, 1581 .store = rdev_attr_store, 1582 }; 1583 static struct kobj_type rdev_ktype = { 1584 .release = rdev_free, 1585 .sysfs_ops = &rdev_sysfs_ops, 1586 .default_attrs = rdev_default_attrs, 1587 }; 1588 1589 /* 1590 * Import a device. If 'super_format' >= 0, then sanity check the superblock 1591 * 1592 * mark the device faulty if: 1593 * 1594 * - the device is nonexistent (zero size) 1595 * - the device has no valid superblock 1596 * 1597 * a faulty rdev _never_ has rdev->sb set. 1598 */ 1599 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) 1600 { 1601 char b[BDEVNAME_SIZE]; 1602 int err; 1603 mdk_rdev_t *rdev; 1604 sector_t size; 1605 1606 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL); 1607 if (!rdev) { 1608 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 1609 return ERR_PTR(-ENOMEM); 1610 } 1611 memset(rdev, 0, sizeof(*rdev)); 1612 1613 if ((err = alloc_disk_sb(rdev))) 1614 goto abort_free; 1615 1616 err = lock_rdev(rdev, newdev); 1617 if (err) 1618 goto abort_free; 1619 1620 rdev->kobj.parent = NULL; 1621 rdev->kobj.ktype = &rdev_ktype; 1622 kobject_init(&rdev->kobj); 1623 1624 rdev->desc_nr = -1; 1625 rdev->flags = 0; 1626 rdev->data_offset = 0; 1627 atomic_set(&rdev->nr_pending, 0); 1628 atomic_set(&rdev->read_errors, 0); 1629 1630 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; 1631 if (!size) { 1632 printk(KERN_WARNING 1633 "md: %s has zero or unknown size, marking faulty!\n", 1634 bdevname(rdev->bdev,b)); 1635 err = -EINVAL; 1636 goto abort_free; 1637 } 1638 1639 if (super_format >= 0) { 1640 err = super_types[super_format]. 1641 load_super(rdev, NULL, super_minor); 1642 if (err == -EINVAL) { 1643 printk(KERN_WARNING 1644 "md: %s has invalid sb, not importing!\n", 1645 bdevname(rdev->bdev,b)); 1646 goto abort_free; 1647 } 1648 if (err < 0) { 1649 printk(KERN_WARNING 1650 "md: could not read %s's sb, not importing!\n", 1651 bdevname(rdev->bdev,b)); 1652 goto abort_free; 1653 } 1654 } 1655 INIT_LIST_HEAD(&rdev->same_set); 1656 1657 return rdev; 1658 1659 abort_free: 1660 if (rdev->sb_page) { 1661 if (rdev->bdev) 1662 unlock_rdev(rdev); 1663 free_disk_sb(rdev); 1664 } 1665 kfree(rdev); 1666 return ERR_PTR(err); 1667 } 1668 1669 /* 1670 * Check a full RAID array for plausibility 1671 */ 1672 1673 1674 static void analyze_sbs(mddev_t * mddev) 1675 { 1676 int i; 1677 struct list_head *tmp; 1678 mdk_rdev_t *rdev, *freshest; 1679 char b[BDEVNAME_SIZE]; 1680 1681 freshest = NULL; 1682 ITERATE_RDEV(mddev,rdev,tmp) 1683 switch (super_types[mddev->major_version]. 1684 load_super(rdev, freshest, mddev->minor_version)) { 1685 case 1: 1686 freshest = rdev; 1687 break; 1688 case 0: 1689 break; 1690 default: 1691 printk( KERN_ERR \ 1692 "md: fatal superblock inconsistency in %s" 1693 " -- removing from array\n", 1694 bdevname(rdev->bdev,b)); 1695 kick_rdev_from_array(rdev); 1696 } 1697 1698 1699 super_types[mddev->major_version]. 1700 validate_super(mddev, freshest); 1701 1702 i = 0; 1703 ITERATE_RDEV(mddev,rdev,tmp) { 1704 if (rdev != freshest) 1705 if (super_types[mddev->major_version]. 1706 validate_super(mddev, rdev)) { 1707 printk(KERN_WARNING "md: kicking non-fresh %s" 1708 " from array!\n", 1709 bdevname(rdev->bdev,b)); 1710 kick_rdev_from_array(rdev); 1711 continue; 1712 } 1713 if (mddev->level == LEVEL_MULTIPATH) { 1714 rdev->desc_nr = i++; 1715 rdev->raid_disk = rdev->desc_nr; 1716 set_bit(In_sync, &rdev->flags); 1717 } 1718 } 1719 1720 1721 1722 if (mddev->recovery_cp != MaxSector && 1723 mddev->level >= 1) 1724 printk(KERN_ERR "md: %s: raid array is not clean" 1725 " -- starting background reconstruction\n", 1726 mdname(mddev)); 1727 1728 } 1729 1730 static ssize_t 1731 md_show_level(mddev_t *mddev, char *page) 1732 { 1733 mdk_personality_t *p = mddev->pers; 1734 if (p == NULL) 1735 return 0; 1736 if (mddev->level >= 0) 1737 return sprintf(page, "RAID-%d\n", mddev->level); 1738 else 1739 return sprintf(page, "%s\n", p->name); 1740 } 1741 1742 static struct md_sysfs_entry md_level = { 1743 .attr = {.name = "level", .mode = S_IRUGO }, 1744 .show = md_show_level, 1745 }; 1746 1747 static ssize_t 1748 md_show_rdisks(mddev_t *mddev, char *page) 1749 { 1750 return sprintf(page, "%d\n", mddev->raid_disks); 1751 } 1752 1753 static struct md_sysfs_entry md_raid_disks = { 1754 .attr = {.name = "raid_disks", .mode = S_IRUGO }, 1755 .show = md_show_rdisks, 1756 }; 1757 1758 static ssize_t 1759 md_show_scan(mddev_t *mddev, char *page) 1760 { 1761 char *type = "none"; 1762 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 1763 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) { 1764 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 1765 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 1766 type = "resync"; 1767 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 1768 type = "check"; 1769 else 1770 type = "repair"; 1771 } else 1772 type = "recover"; 1773 } 1774 return sprintf(page, "%s\n", type); 1775 } 1776 1777 static ssize_t 1778 md_store_scan(mddev_t *mddev, const char *page, size_t len) 1779 { 1780 int canscan=0; 1781 1782 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 1783 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 1784 return -EBUSY; 1785 down(&mddev->reconfig_sem); 1786 if (mddev->pers && mddev->pers->sync_request) 1787 canscan=1; 1788 up(&mddev->reconfig_sem); 1789 if (!canscan) 1790 return -EINVAL; 1791 1792 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0) 1793 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 1794 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0) 1795 return -EINVAL; 1796 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 1797 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 1798 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 1799 md_wakeup_thread(mddev->thread); 1800 return len; 1801 } 1802 1803 static ssize_t 1804 md_show_mismatch(mddev_t *mddev, char *page) 1805 { 1806 return sprintf(page, "%llu\n", 1807 (unsigned long long) mddev->resync_mismatches); 1808 } 1809 1810 static struct md_sysfs_entry md_scan_mode = { 1811 .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR }, 1812 .show = md_show_scan, 1813 .store = md_store_scan, 1814 }; 1815 1816 static struct md_sysfs_entry md_mismatches = { 1817 .attr = {.name = "mismatch_cnt", .mode = S_IRUGO }, 1818 .show = md_show_mismatch, 1819 }; 1820 1821 static struct attribute *md_default_attrs[] = { 1822 &md_level.attr, 1823 &md_raid_disks.attr, 1824 &md_scan_mode.attr, 1825 &md_mismatches.attr, 1826 NULL, 1827 }; 1828 1829 static ssize_t 1830 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 1831 { 1832 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 1833 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 1834 1835 if (!entry->show) 1836 return -EIO; 1837 return entry->show(mddev, page); 1838 } 1839 1840 static ssize_t 1841 md_attr_store(struct kobject *kobj, struct attribute *attr, 1842 const char *page, size_t length) 1843 { 1844 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 1845 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 1846 1847 if (!entry->store) 1848 return -EIO; 1849 return entry->store(mddev, page, length); 1850 } 1851 1852 static void md_free(struct kobject *ko) 1853 { 1854 mddev_t *mddev = container_of(ko, mddev_t, kobj); 1855 kfree(mddev); 1856 } 1857 1858 static struct sysfs_ops md_sysfs_ops = { 1859 .show = md_attr_show, 1860 .store = md_attr_store, 1861 }; 1862 static struct kobj_type md_ktype = { 1863 .release = md_free, 1864 .sysfs_ops = &md_sysfs_ops, 1865 .default_attrs = md_default_attrs, 1866 }; 1867 1868 int mdp_major = 0; 1869 1870 static struct kobject *md_probe(dev_t dev, int *part, void *data) 1871 { 1872 static DECLARE_MUTEX(disks_sem); 1873 mddev_t *mddev = mddev_find(dev); 1874 struct gendisk *disk; 1875 int partitioned = (MAJOR(dev) != MD_MAJOR); 1876 int shift = partitioned ? MdpMinorShift : 0; 1877 int unit = MINOR(dev) >> shift; 1878 1879 if (!mddev) 1880 return NULL; 1881 1882 down(&disks_sem); 1883 if (mddev->gendisk) { 1884 up(&disks_sem); 1885 mddev_put(mddev); 1886 return NULL; 1887 } 1888 disk = alloc_disk(1 << shift); 1889 if (!disk) { 1890 up(&disks_sem); 1891 mddev_put(mddev); 1892 return NULL; 1893 } 1894 disk->major = MAJOR(dev); 1895 disk->first_minor = unit << shift; 1896 if (partitioned) { 1897 sprintf(disk->disk_name, "md_d%d", unit); 1898 sprintf(disk->devfs_name, "md/d%d", unit); 1899 } else { 1900 sprintf(disk->disk_name, "md%d", unit); 1901 sprintf(disk->devfs_name, "md/%d", unit); 1902 } 1903 disk->fops = &md_fops; 1904 disk->private_data = mddev; 1905 disk->queue = mddev->queue; 1906 add_disk(disk); 1907 mddev->gendisk = disk; 1908 up(&disks_sem); 1909 mddev->kobj.parent = &disk->kobj; 1910 mddev->kobj.k_name = NULL; 1911 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md"); 1912 mddev->kobj.ktype = &md_ktype; 1913 kobject_register(&mddev->kobj); 1914 return NULL; 1915 } 1916 1917 void md_wakeup_thread(mdk_thread_t *thread); 1918 1919 static void md_safemode_timeout(unsigned long data) 1920 { 1921 mddev_t *mddev = (mddev_t *) data; 1922 1923 mddev->safemode = 1; 1924 md_wakeup_thread(mddev->thread); 1925 } 1926 1927 1928 static int do_md_run(mddev_t * mddev) 1929 { 1930 int pnum, err; 1931 int chunk_size; 1932 struct list_head *tmp; 1933 mdk_rdev_t *rdev; 1934 struct gendisk *disk; 1935 char b[BDEVNAME_SIZE]; 1936 1937 if (list_empty(&mddev->disks)) 1938 /* cannot run an array with no devices.. */ 1939 return -EINVAL; 1940 1941 if (mddev->pers) 1942 return -EBUSY; 1943 1944 /* 1945 * Analyze all RAID superblock(s) 1946 */ 1947 if (!mddev->raid_disks) 1948 analyze_sbs(mddev); 1949 1950 chunk_size = mddev->chunk_size; 1951 pnum = level_to_pers(mddev->level); 1952 1953 if ((pnum != MULTIPATH) && (pnum != RAID1)) { 1954 if (!chunk_size) { 1955 /* 1956 * 'default chunksize' in the old md code used to 1957 * be PAGE_SIZE, baaad. 1958 * we abort here to be on the safe side. We don't 1959 * want to continue the bad practice. 1960 */ 1961 printk(KERN_ERR 1962 "no chunksize specified, see 'man raidtab'\n"); 1963 return -EINVAL; 1964 } 1965 if (chunk_size > MAX_CHUNK_SIZE) { 1966 printk(KERN_ERR "too big chunk_size: %d > %d\n", 1967 chunk_size, MAX_CHUNK_SIZE); 1968 return -EINVAL; 1969 } 1970 /* 1971 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE 1972 */ 1973 if ( (1 << ffz(~chunk_size)) != chunk_size) { 1974 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size); 1975 return -EINVAL; 1976 } 1977 if (chunk_size < PAGE_SIZE) { 1978 printk(KERN_ERR "too small chunk_size: %d < %ld\n", 1979 chunk_size, PAGE_SIZE); 1980 return -EINVAL; 1981 } 1982 1983 /* devices must have minimum size of one chunk */ 1984 ITERATE_RDEV(mddev,rdev,tmp) { 1985 if (test_bit(Faulty, &rdev->flags)) 1986 continue; 1987 if (rdev->size < chunk_size / 1024) { 1988 printk(KERN_WARNING 1989 "md: Dev %s smaller than chunk_size:" 1990 " %lluk < %dk\n", 1991 bdevname(rdev->bdev,b), 1992 (unsigned long long)rdev->size, 1993 chunk_size / 1024); 1994 return -EINVAL; 1995 } 1996 } 1997 } 1998 1999 #ifdef CONFIG_KMOD 2000 if (!pers[pnum]) 2001 { 2002 request_module("md-personality-%d", pnum); 2003 } 2004 #endif 2005 2006 /* 2007 * Drop all container device buffers, from now on 2008 * the only valid external interface is through the md 2009 * device. 2010 * Also find largest hardsector size 2011 */ 2012 ITERATE_RDEV(mddev,rdev,tmp) { 2013 if (test_bit(Faulty, &rdev->flags)) 2014 continue; 2015 sync_blockdev(rdev->bdev); 2016 invalidate_bdev(rdev->bdev, 0); 2017 } 2018 2019 md_probe(mddev->unit, NULL, NULL); 2020 disk = mddev->gendisk; 2021 if (!disk) 2022 return -ENOMEM; 2023 2024 spin_lock(&pers_lock); 2025 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) { 2026 spin_unlock(&pers_lock); 2027 printk(KERN_WARNING "md: personality %d is not loaded!\n", 2028 pnum); 2029 return -EINVAL; 2030 } 2031 2032 mddev->pers = pers[pnum]; 2033 spin_unlock(&pers_lock); 2034 2035 mddev->recovery = 0; 2036 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ 2037 mddev->barriers_work = 1; 2038 2039 if (start_readonly) 2040 mddev->ro = 2; /* read-only, but switch on first write */ 2041 2042 /* before we start the array running, initialise the bitmap */ 2043 err = bitmap_create(mddev); 2044 if (err) 2045 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 2046 mdname(mddev), err); 2047 else 2048 err = mddev->pers->run(mddev); 2049 if (err) { 2050 printk(KERN_ERR "md: pers->run() failed ...\n"); 2051 module_put(mddev->pers->owner); 2052 mddev->pers = NULL; 2053 bitmap_destroy(mddev); 2054 return err; 2055 } 2056 atomic_set(&mddev->writes_pending,0); 2057 mddev->safemode = 0; 2058 mddev->safemode_timer.function = md_safemode_timeout; 2059 mddev->safemode_timer.data = (unsigned long) mddev; 2060 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */ 2061 mddev->in_sync = 1; 2062 2063 ITERATE_RDEV(mddev,rdev,tmp) 2064 if (rdev->raid_disk >= 0) { 2065 char nm[20]; 2066 sprintf(nm, "rd%d", rdev->raid_disk); 2067 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); 2068 } 2069 2070 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 2071 md_wakeup_thread(mddev->thread); 2072 2073 if (mddev->sb_dirty) 2074 md_update_sb(mddev); 2075 2076 set_capacity(disk, mddev->array_size<<1); 2077 2078 /* If we call blk_queue_make_request here, it will 2079 * re-initialise max_sectors etc which may have been 2080 * refined inside -> run. So just set the bits we need to set. 2081 * Most initialisation happended when we called 2082 * blk_queue_make_request(..., md_fail_request) 2083 * earlier. 2084 */ 2085 mddev->queue->queuedata = mddev; 2086 mddev->queue->make_request_fn = mddev->pers->make_request; 2087 2088 mddev->changed = 1; 2089 return 0; 2090 } 2091 2092 static int restart_array(mddev_t *mddev) 2093 { 2094 struct gendisk *disk = mddev->gendisk; 2095 int err; 2096 2097 /* 2098 * Complain if it has no devices 2099 */ 2100 err = -ENXIO; 2101 if (list_empty(&mddev->disks)) 2102 goto out; 2103 2104 if (mddev->pers) { 2105 err = -EBUSY; 2106 if (!mddev->ro) 2107 goto out; 2108 2109 mddev->safemode = 0; 2110 mddev->ro = 0; 2111 set_disk_ro(disk, 0); 2112 2113 printk(KERN_INFO "md: %s switched to read-write mode.\n", 2114 mdname(mddev)); 2115 /* 2116 * Kick recovery or resync if necessary 2117 */ 2118 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 2119 md_wakeup_thread(mddev->thread); 2120 err = 0; 2121 } else { 2122 printk(KERN_ERR "md: %s has no personality assigned.\n", 2123 mdname(mddev)); 2124 err = -EINVAL; 2125 } 2126 2127 out: 2128 return err; 2129 } 2130 2131 static int do_md_stop(mddev_t * mddev, int ro) 2132 { 2133 int err = 0; 2134 struct gendisk *disk = mddev->gendisk; 2135 2136 if (mddev->pers) { 2137 if (atomic_read(&mddev->active)>2) { 2138 printk("md: %s still in use.\n",mdname(mddev)); 2139 return -EBUSY; 2140 } 2141 2142 if (mddev->sync_thread) { 2143 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 2144 md_unregister_thread(mddev->sync_thread); 2145 mddev->sync_thread = NULL; 2146 } 2147 2148 del_timer_sync(&mddev->safemode_timer); 2149 2150 invalidate_partition(disk, 0); 2151 2152 if (ro) { 2153 err = -ENXIO; 2154 if (mddev->ro==1) 2155 goto out; 2156 mddev->ro = 1; 2157 } else { 2158 bitmap_flush(mddev); 2159 md_super_wait(mddev); 2160 if (mddev->ro) 2161 set_disk_ro(disk, 0); 2162 blk_queue_make_request(mddev->queue, md_fail_request); 2163 mddev->pers->stop(mddev); 2164 module_put(mddev->pers->owner); 2165 mddev->pers = NULL; 2166 if (mddev->ro) 2167 mddev->ro = 0; 2168 } 2169 if (!mddev->in_sync) { 2170 /* mark array as shutdown cleanly */ 2171 mddev->in_sync = 1; 2172 md_update_sb(mddev); 2173 } 2174 if (ro) 2175 set_disk_ro(disk, 1); 2176 } 2177 2178 bitmap_destroy(mddev); 2179 if (mddev->bitmap_file) { 2180 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1); 2181 fput(mddev->bitmap_file); 2182 mddev->bitmap_file = NULL; 2183 } 2184 mddev->bitmap_offset = 0; 2185 2186 /* 2187 * Free resources if final stop 2188 */ 2189 if (!ro) { 2190 mdk_rdev_t *rdev; 2191 struct list_head *tmp; 2192 struct gendisk *disk; 2193 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 2194 2195 ITERATE_RDEV(mddev,rdev,tmp) 2196 if (rdev->raid_disk >= 0) { 2197 char nm[20]; 2198 sprintf(nm, "rd%d", rdev->raid_disk); 2199 sysfs_remove_link(&mddev->kobj, nm); 2200 } 2201 2202 export_array(mddev); 2203 2204 mddev->array_size = 0; 2205 disk = mddev->gendisk; 2206 if (disk) 2207 set_capacity(disk, 0); 2208 mddev->changed = 1; 2209 } else 2210 printk(KERN_INFO "md: %s switched to read-only mode.\n", 2211 mdname(mddev)); 2212 err = 0; 2213 out: 2214 return err; 2215 } 2216 2217 static void autorun_array(mddev_t *mddev) 2218 { 2219 mdk_rdev_t *rdev; 2220 struct list_head *tmp; 2221 int err; 2222 2223 if (list_empty(&mddev->disks)) 2224 return; 2225 2226 printk(KERN_INFO "md: running: "); 2227 2228 ITERATE_RDEV(mddev,rdev,tmp) { 2229 char b[BDEVNAME_SIZE]; 2230 printk("<%s>", bdevname(rdev->bdev,b)); 2231 } 2232 printk("\n"); 2233 2234 err = do_md_run (mddev); 2235 if (err) { 2236 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 2237 do_md_stop (mddev, 0); 2238 } 2239 } 2240 2241 /* 2242 * lets try to run arrays based on all disks that have arrived 2243 * until now. (those are in pending_raid_disks) 2244 * 2245 * the method: pick the first pending disk, collect all disks with 2246 * the same UUID, remove all from the pending list and put them into 2247 * the 'same_array' list. Then order this list based on superblock 2248 * update time (freshest comes first), kick out 'old' disks and 2249 * compare superblocks. If everything's fine then run it. 2250 * 2251 * If "unit" is allocated, then bump its reference count 2252 */ 2253 static void autorun_devices(int part) 2254 { 2255 struct list_head candidates; 2256 struct list_head *tmp; 2257 mdk_rdev_t *rdev0, *rdev; 2258 mddev_t *mddev; 2259 char b[BDEVNAME_SIZE]; 2260 2261 printk(KERN_INFO "md: autorun ...\n"); 2262 while (!list_empty(&pending_raid_disks)) { 2263 dev_t dev; 2264 rdev0 = list_entry(pending_raid_disks.next, 2265 mdk_rdev_t, same_set); 2266 2267 printk(KERN_INFO "md: considering %s ...\n", 2268 bdevname(rdev0->bdev,b)); 2269 INIT_LIST_HEAD(&candidates); 2270 ITERATE_RDEV_PENDING(rdev,tmp) 2271 if (super_90_load(rdev, rdev0, 0) >= 0) { 2272 printk(KERN_INFO "md: adding %s ...\n", 2273 bdevname(rdev->bdev,b)); 2274 list_move(&rdev->same_set, &candidates); 2275 } 2276 /* 2277 * now we have a set of devices, with all of them having 2278 * mostly sane superblocks. It's time to allocate the 2279 * mddev. 2280 */ 2281 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) { 2282 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 2283 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 2284 break; 2285 } 2286 if (part) 2287 dev = MKDEV(mdp_major, 2288 rdev0->preferred_minor << MdpMinorShift); 2289 else 2290 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 2291 2292 md_probe(dev, NULL, NULL); 2293 mddev = mddev_find(dev); 2294 if (!mddev) { 2295 printk(KERN_ERR 2296 "md: cannot allocate memory for md drive.\n"); 2297 break; 2298 } 2299 if (mddev_lock(mddev)) 2300 printk(KERN_WARNING "md: %s locked, cannot run\n", 2301 mdname(mddev)); 2302 else if (mddev->raid_disks || mddev->major_version 2303 || !list_empty(&mddev->disks)) { 2304 printk(KERN_WARNING 2305 "md: %s already running, cannot run %s\n", 2306 mdname(mddev), bdevname(rdev0->bdev,b)); 2307 mddev_unlock(mddev); 2308 } else { 2309 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 2310 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) { 2311 list_del_init(&rdev->same_set); 2312 if (bind_rdev_to_array(rdev, mddev)) 2313 export_rdev(rdev); 2314 } 2315 autorun_array(mddev); 2316 mddev_unlock(mddev); 2317 } 2318 /* on success, candidates will be empty, on error 2319 * it won't... 2320 */ 2321 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) 2322 export_rdev(rdev); 2323 mddev_put(mddev); 2324 } 2325 printk(KERN_INFO "md: ... autorun DONE.\n"); 2326 } 2327 2328 /* 2329 * import RAID devices based on one partition 2330 * if possible, the array gets run as well. 2331 */ 2332 2333 static int autostart_array(dev_t startdev) 2334 { 2335 char b[BDEVNAME_SIZE]; 2336 int err = -EINVAL, i; 2337 mdp_super_t *sb = NULL; 2338 mdk_rdev_t *start_rdev = NULL, *rdev; 2339 2340 start_rdev = md_import_device(startdev, 0, 0); 2341 if (IS_ERR(start_rdev)) 2342 return err; 2343 2344 2345 /* NOTE: this can only work for 0.90.0 superblocks */ 2346 sb = (mdp_super_t*)page_address(start_rdev->sb_page); 2347 if (sb->major_version != 0 || 2348 sb->minor_version != 90 ) { 2349 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n"); 2350 export_rdev(start_rdev); 2351 return err; 2352 } 2353 2354 if (test_bit(Faulty, &start_rdev->flags)) { 2355 printk(KERN_WARNING 2356 "md: can not autostart based on faulty %s!\n", 2357 bdevname(start_rdev->bdev,b)); 2358 export_rdev(start_rdev); 2359 return err; 2360 } 2361 list_add(&start_rdev->same_set, &pending_raid_disks); 2362 2363 for (i = 0; i < MD_SB_DISKS; i++) { 2364 mdp_disk_t *desc = sb->disks + i; 2365 dev_t dev = MKDEV(desc->major, desc->minor); 2366 2367 if (!dev) 2368 continue; 2369 if (dev == startdev) 2370 continue; 2371 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor) 2372 continue; 2373 rdev = md_import_device(dev, 0, 0); 2374 if (IS_ERR(rdev)) 2375 continue; 2376 2377 list_add(&rdev->same_set, &pending_raid_disks); 2378 } 2379 2380 /* 2381 * possibly return codes 2382 */ 2383 autorun_devices(0); 2384 return 0; 2385 2386 } 2387 2388 2389 static int get_version(void __user * arg) 2390 { 2391 mdu_version_t ver; 2392 2393 ver.major = MD_MAJOR_VERSION; 2394 ver.minor = MD_MINOR_VERSION; 2395 ver.patchlevel = MD_PATCHLEVEL_VERSION; 2396 2397 if (copy_to_user(arg, &ver, sizeof(ver))) 2398 return -EFAULT; 2399 2400 return 0; 2401 } 2402 2403 static int get_array_info(mddev_t * mddev, void __user * arg) 2404 { 2405 mdu_array_info_t info; 2406 int nr,working,active,failed,spare; 2407 mdk_rdev_t *rdev; 2408 struct list_head *tmp; 2409 2410 nr=working=active=failed=spare=0; 2411 ITERATE_RDEV(mddev,rdev,tmp) { 2412 nr++; 2413 if (test_bit(Faulty, &rdev->flags)) 2414 failed++; 2415 else { 2416 working++; 2417 if (test_bit(In_sync, &rdev->flags)) 2418 active++; 2419 else 2420 spare++; 2421 } 2422 } 2423 2424 info.major_version = mddev->major_version; 2425 info.minor_version = mddev->minor_version; 2426 info.patch_version = MD_PATCHLEVEL_VERSION; 2427 info.ctime = mddev->ctime; 2428 info.level = mddev->level; 2429 info.size = mddev->size; 2430 info.nr_disks = nr; 2431 info.raid_disks = mddev->raid_disks; 2432 info.md_minor = mddev->md_minor; 2433 info.not_persistent= !mddev->persistent; 2434 2435 info.utime = mddev->utime; 2436 info.state = 0; 2437 if (mddev->in_sync) 2438 info.state = (1<<MD_SB_CLEAN); 2439 if (mddev->bitmap && mddev->bitmap_offset) 2440 info.state = (1<<MD_SB_BITMAP_PRESENT); 2441 info.active_disks = active; 2442 info.working_disks = working; 2443 info.failed_disks = failed; 2444 info.spare_disks = spare; 2445 2446 info.layout = mddev->layout; 2447 info.chunk_size = mddev->chunk_size; 2448 2449 if (copy_to_user(arg, &info, sizeof(info))) 2450 return -EFAULT; 2451 2452 return 0; 2453 } 2454 2455 static int get_bitmap_file(mddev_t * mddev, void __user * arg) 2456 { 2457 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 2458 char *ptr, *buf = NULL; 2459 int err = -ENOMEM; 2460 2461 file = kmalloc(sizeof(*file), GFP_KERNEL); 2462 if (!file) 2463 goto out; 2464 2465 /* bitmap disabled, zero the first byte and copy out */ 2466 if (!mddev->bitmap || !mddev->bitmap->file) { 2467 file->pathname[0] = '\0'; 2468 goto copy_out; 2469 } 2470 2471 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 2472 if (!buf) 2473 goto out; 2474 2475 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname)); 2476 if (!ptr) 2477 goto out; 2478 2479 strcpy(file->pathname, ptr); 2480 2481 copy_out: 2482 err = 0; 2483 if (copy_to_user(arg, file, sizeof(*file))) 2484 err = -EFAULT; 2485 out: 2486 kfree(buf); 2487 kfree(file); 2488 return err; 2489 } 2490 2491 static int get_disk_info(mddev_t * mddev, void __user * arg) 2492 { 2493 mdu_disk_info_t info; 2494 unsigned int nr; 2495 mdk_rdev_t *rdev; 2496 2497 if (copy_from_user(&info, arg, sizeof(info))) 2498 return -EFAULT; 2499 2500 nr = info.number; 2501 2502 rdev = find_rdev_nr(mddev, nr); 2503 if (rdev) { 2504 info.major = MAJOR(rdev->bdev->bd_dev); 2505 info.minor = MINOR(rdev->bdev->bd_dev); 2506 info.raid_disk = rdev->raid_disk; 2507 info.state = 0; 2508 if (test_bit(Faulty, &rdev->flags)) 2509 info.state |= (1<<MD_DISK_FAULTY); 2510 else if (test_bit(In_sync, &rdev->flags)) { 2511 info.state |= (1<<MD_DISK_ACTIVE); 2512 info.state |= (1<<MD_DISK_SYNC); 2513 } 2514 if (test_bit(WriteMostly, &rdev->flags)) 2515 info.state |= (1<<MD_DISK_WRITEMOSTLY); 2516 } else { 2517 info.major = info.minor = 0; 2518 info.raid_disk = -1; 2519 info.state = (1<<MD_DISK_REMOVED); 2520 } 2521 2522 if (copy_to_user(arg, &info, sizeof(info))) 2523 return -EFAULT; 2524 2525 return 0; 2526 } 2527 2528 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) 2529 { 2530 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 2531 mdk_rdev_t *rdev; 2532 dev_t dev = MKDEV(info->major,info->minor); 2533 2534 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 2535 return -EOVERFLOW; 2536 2537 if (!mddev->raid_disks) { 2538 int err; 2539 /* expecting a device which has a superblock */ 2540 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 2541 if (IS_ERR(rdev)) { 2542 printk(KERN_WARNING 2543 "md: md_import_device returned %ld\n", 2544 PTR_ERR(rdev)); 2545 return PTR_ERR(rdev); 2546 } 2547 if (!list_empty(&mddev->disks)) { 2548 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 2549 mdk_rdev_t, same_set); 2550 int err = super_types[mddev->major_version] 2551 .load_super(rdev, rdev0, mddev->minor_version); 2552 if (err < 0) { 2553 printk(KERN_WARNING 2554 "md: %s has different UUID to %s\n", 2555 bdevname(rdev->bdev,b), 2556 bdevname(rdev0->bdev,b2)); 2557 export_rdev(rdev); 2558 return -EINVAL; 2559 } 2560 } 2561 err = bind_rdev_to_array(rdev, mddev); 2562 if (err) 2563 export_rdev(rdev); 2564 return err; 2565 } 2566 2567 /* 2568 * add_new_disk can be used once the array is assembled 2569 * to add "hot spares". They must already have a superblock 2570 * written 2571 */ 2572 if (mddev->pers) { 2573 int err; 2574 if (!mddev->pers->hot_add_disk) { 2575 printk(KERN_WARNING 2576 "%s: personality does not support diskops!\n", 2577 mdname(mddev)); 2578 return -EINVAL; 2579 } 2580 if (mddev->persistent) 2581 rdev = md_import_device(dev, mddev->major_version, 2582 mddev->minor_version); 2583 else 2584 rdev = md_import_device(dev, -1, -1); 2585 if (IS_ERR(rdev)) { 2586 printk(KERN_WARNING 2587 "md: md_import_device returned %ld\n", 2588 PTR_ERR(rdev)); 2589 return PTR_ERR(rdev); 2590 } 2591 /* set save_raid_disk if appropriate */ 2592 if (!mddev->persistent) { 2593 if (info->state & (1<<MD_DISK_SYNC) && 2594 info->raid_disk < mddev->raid_disks) 2595 rdev->raid_disk = info->raid_disk; 2596 else 2597 rdev->raid_disk = -1; 2598 } else 2599 super_types[mddev->major_version]. 2600 validate_super(mddev, rdev); 2601 rdev->saved_raid_disk = rdev->raid_disk; 2602 2603 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 2604 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 2605 set_bit(WriteMostly, &rdev->flags); 2606 2607 rdev->raid_disk = -1; 2608 err = bind_rdev_to_array(rdev, mddev); 2609 if (err) 2610 export_rdev(rdev); 2611 2612 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 2613 md_wakeup_thread(mddev->thread); 2614 return err; 2615 } 2616 2617 /* otherwise, add_new_disk is only allowed 2618 * for major_version==0 superblocks 2619 */ 2620 if (mddev->major_version != 0) { 2621 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 2622 mdname(mddev)); 2623 return -EINVAL; 2624 } 2625 2626 if (!(info->state & (1<<MD_DISK_FAULTY))) { 2627 int err; 2628 rdev = md_import_device (dev, -1, 0); 2629 if (IS_ERR(rdev)) { 2630 printk(KERN_WARNING 2631 "md: error, md_import_device() returned %ld\n", 2632 PTR_ERR(rdev)); 2633 return PTR_ERR(rdev); 2634 } 2635 rdev->desc_nr = info->number; 2636 if (info->raid_disk < mddev->raid_disks) 2637 rdev->raid_disk = info->raid_disk; 2638 else 2639 rdev->raid_disk = -1; 2640 2641 rdev->flags = 0; 2642 2643 if (rdev->raid_disk < mddev->raid_disks) 2644 if (info->state & (1<<MD_DISK_SYNC)) 2645 set_bit(In_sync, &rdev->flags); 2646 2647 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 2648 set_bit(WriteMostly, &rdev->flags); 2649 2650 err = bind_rdev_to_array(rdev, mddev); 2651 if (err) { 2652 export_rdev(rdev); 2653 return err; 2654 } 2655 2656 if (!mddev->persistent) { 2657 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 2658 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; 2659 } else 2660 rdev->sb_offset = calc_dev_sboffset(rdev->bdev); 2661 rdev->size = calc_dev_size(rdev, mddev->chunk_size); 2662 2663 if (!mddev->size || (mddev->size > rdev->size)) 2664 mddev->size = rdev->size; 2665 } 2666 2667 return 0; 2668 } 2669 2670 static int hot_remove_disk(mddev_t * mddev, dev_t dev) 2671 { 2672 char b[BDEVNAME_SIZE]; 2673 mdk_rdev_t *rdev; 2674 2675 if (!mddev->pers) 2676 return -ENODEV; 2677 2678 rdev = find_rdev(mddev, dev); 2679 if (!rdev) 2680 return -ENXIO; 2681 2682 if (rdev->raid_disk >= 0) 2683 goto busy; 2684 2685 kick_rdev_from_array(rdev); 2686 md_update_sb(mddev); 2687 2688 return 0; 2689 busy: 2690 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n", 2691 bdevname(rdev->bdev,b), mdname(mddev)); 2692 return -EBUSY; 2693 } 2694 2695 static int hot_add_disk(mddev_t * mddev, dev_t dev) 2696 { 2697 char b[BDEVNAME_SIZE]; 2698 int err; 2699 unsigned int size; 2700 mdk_rdev_t *rdev; 2701 2702 if (!mddev->pers) 2703 return -ENODEV; 2704 2705 if (mddev->major_version != 0) { 2706 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 2707 " version-0 superblocks.\n", 2708 mdname(mddev)); 2709 return -EINVAL; 2710 } 2711 if (!mddev->pers->hot_add_disk) { 2712 printk(KERN_WARNING 2713 "%s: personality does not support diskops!\n", 2714 mdname(mddev)); 2715 return -EINVAL; 2716 } 2717 2718 rdev = md_import_device (dev, -1, 0); 2719 if (IS_ERR(rdev)) { 2720 printk(KERN_WARNING 2721 "md: error, md_import_device() returned %ld\n", 2722 PTR_ERR(rdev)); 2723 return -EINVAL; 2724 } 2725 2726 if (mddev->persistent) 2727 rdev->sb_offset = calc_dev_sboffset(rdev->bdev); 2728 else 2729 rdev->sb_offset = 2730 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; 2731 2732 size = calc_dev_size(rdev, mddev->chunk_size); 2733 rdev->size = size; 2734 2735 if (size < mddev->size) { 2736 printk(KERN_WARNING 2737 "%s: disk size %llu blocks < array size %llu\n", 2738 mdname(mddev), (unsigned long long)size, 2739 (unsigned long long)mddev->size); 2740 err = -ENOSPC; 2741 goto abort_export; 2742 } 2743 2744 if (test_bit(Faulty, &rdev->flags)) { 2745 printk(KERN_WARNING 2746 "md: can not hot-add faulty %s disk to %s!\n", 2747 bdevname(rdev->bdev,b), mdname(mddev)); 2748 err = -EINVAL; 2749 goto abort_export; 2750 } 2751 clear_bit(In_sync, &rdev->flags); 2752 rdev->desc_nr = -1; 2753 bind_rdev_to_array(rdev, mddev); 2754 2755 /* 2756 * The rest should better be atomic, we can have disk failures 2757 * noticed in interrupt contexts ... 2758 */ 2759 2760 if (rdev->desc_nr == mddev->max_disks) { 2761 printk(KERN_WARNING "%s: can not hot-add to full array!\n", 2762 mdname(mddev)); 2763 err = -EBUSY; 2764 goto abort_unbind_export; 2765 } 2766 2767 rdev->raid_disk = -1; 2768 2769 md_update_sb(mddev); 2770 2771 /* 2772 * Kick recovery, maybe this spare has to be added to the 2773 * array immediately. 2774 */ 2775 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 2776 md_wakeup_thread(mddev->thread); 2777 2778 return 0; 2779 2780 abort_unbind_export: 2781 unbind_rdev_from_array(rdev); 2782 2783 abort_export: 2784 export_rdev(rdev); 2785 return err; 2786 } 2787 2788 /* similar to deny_write_access, but accounts for our holding a reference 2789 * to the file ourselves */ 2790 static int deny_bitmap_write_access(struct file * file) 2791 { 2792 struct inode *inode = file->f_mapping->host; 2793 2794 spin_lock(&inode->i_lock); 2795 if (atomic_read(&inode->i_writecount) > 1) { 2796 spin_unlock(&inode->i_lock); 2797 return -ETXTBSY; 2798 } 2799 atomic_set(&inode->i_writecount, -1); 2800 spin_unlock(&inode->i_lock); 2801 2802 return 0; 2803 } 2804 2805 static int set_bitmap_file(mddev_t *mddev, int fd) 2806 { 2807 int err; 2808 2809 if (mddev->pers) { 2810 if (!mddev->pers->quiesce) 2811 return -EBUSY; 2812 if (mddev->recovery || mddev->sync_thread) 2813 return -EBUSY; 2814 /* we should be able to change the bitmap.. */ 2815 } 2816 2817 2818 if (fd >= 0) { 2819 if (mddev->bitmap) 2820 return -EEXIST; /* cannot add when bitmap is present */ 2821 mddev->bitmap_file = fget(fd); 2822 2823 if (mddev->bitmap_file == NULL) { 2824 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 2825 mdname(mddev)); 2826 return -EBADF; 2827 } 2828 2829 err = deny_bitmap_write_access(mddev->bitmap_file); 2830 if (err) { 2831 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 2832 mdname(mddev)); 2833 fput(mddev->bitmap_file); 2834 mddev->bitmap_file = NULL; 2835 return err; 2836 } 2837 mddev->bitmap_offset = 0; /* file overrides offset */ 2838 } else if (mddev->bitmap == NULL) 2839 return -ENOENT; /* cannot remove what isn't there */ 2840 err = 0; 2841 if (mddev->pers) { 2842 mddev->pers->quiesce(mddev, 1); 2843 if (fd >= 0) 2844 err = bitmap_create(mddev); 2845 if (fd < 0 || err) 2846 bitmap_destroy(mddev); 2847 mddev->pers->quiesce(mddev, 0); 2848 } else if (fd < 0) { 2849 if (mddev->bitmap_file) 2850 fput(mddev->bitmap_file); 2851 mddev->bitmap_file = NULL; 2852 } 2853 2854 return err; 2855 } 2856 2857 /* 2858 * set_array_info is used two different ways 2859 * The original usage is when creating a new array. 2860 * In this usage, raid_disks is > 0 and it together with 2861 * level, size, not_persistent,layout,chunksize determine the 2862 * shape of the array. 2863 * This will always create an array with a type-0.90.0 superblock. 2864 * The newer usage is when assembling an array. 2865 * In this case raid_disks will be 0, and the major_version field is 2866 * use to determine which style super-blocks are to be found on the devices. 2867 * The minor and patch _version numbers are also kept incase the 2868 * super_block handler wishes to interpret them. 2869 */ 2870 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) 2871 { 2872 2873 if (info->raid_disks == 0) { 2874 /* just setting version number for superblock loading */ 2875 if (info->major_version < 0 || 2876 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) || 2877 super_types[info->major_version].name == NULL) { 2878 /* maybe try to auto-load a module? */ 2879 printk(KERN_INFO 2880 "md: superblock version %d not known\n", 2881 info->major_version); 2882 return -EINVAL; 2883 } 2884 mddev->major_version = info->major_version; 2885 mddev->minor_version = info->minor_version; 2886 mddev->patch_version = info->patch_version; 2887 return 0; 2888 } 2889 mddev->major_version = MD_MAJOR_VERSION; 2890 mddev->minor_version = MD_MINOR_VERSION; 2891 mddev->patch_version = MD_PATCHLEVEL_VERSION; 2892 mddev->ctime = get_seconds(); 2893 2894 mddev->level = info->level; 2895 mddev->size = info->size; 2896 mddev->raid_disks = info->raid_disks; 2897 /* don't set md_minor, it is determined by which /dev/md* was 2898 * openned 2899 */ 2900 if (info->state & (1<<MD_SB_CLEAN)) 2901 mddev->recovery_cp = MaxSector; 2902 else 2903 mddev->recovery_cp = 0; 2904 mddev->persistent = ! info->not_persistent; 2905 2906 mddev->layout = info->layout; 2907 mddev->chunk_size = info->chunk_size; 2908 2909 mddev->max_disks = MD_SB_DISKS; 2910 2911 mddev->sb_dirty = 1; 2912 2913 /* 2914 * Generate a 128 bit UUID 2915 */ 2916 get_random_bytes(mddev->uuid, 16); 2917 2918 return 0; 2919 } 2920 2921 /* 2922 * update_array_info is used to change the configuration of an 2923 * on-line array. 2924 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 2925 * fields in the info are checked against the array. 2926 * Any differences that cannot be handled will cause an error. 2927 * Normally, only one change can be managed at a time. 2928 */ 2929 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) 2930 { 2931 int rv = 0; 2932 int cnt = 0; 2933 int state = 0; 2934 2935 /* calculate expected state,ignoring low bits */ 2936 if (mddev->bitmap && mddev->bitmap_offset) 2937 state |= (1 << MD_SB_BITMAP_PRESENT); 2938 2939 if (mddev->major_version != info->major_version || 2940 mddev->minor_version != info->minor_version || 2941 /* mddev->patch_version != info->patch_version || */ 2942 mddev->ctime != info->ctime || 2943 mddev->level != info->level || 2944 /* mddev->layout != info->layout || */ 2945 !mddev->persistent != info->not_persistent|| 2946 mddev->chunk_size != info->chunk_size || 2947 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 2948 ((state^info->state) & 0xfffffe00) 2949 ) 2950 return -EINVAL; 2951 /* Check there is only one change */ 2952 if (mddev->size != info->size) cnt++; 2953 if (mddev->raid_disks != info->raid_disks) cnt++; 2954 if (mddev->layout != info->layout) cnt++; 2955 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++; 2956 if (cnt == 0) return 0; 2957 if (cnt > 1) return -EINVAL; 2958 2959 if (mddev->layout != info->layout) { 2960 /* Change layout 2961 * we don't need to do anything at the md level, the 2962 * personality will take care of it all. 2963 */ 2964 if (mddev->pers->reconfig == NULL) 2965 return -EINVAL; 2966 else 2967 return mddev->pers->reconfig(mddev, info->layout, -1); 2968 } 2969 if (mddev->size != info->size) { 2970 mdk_rdev_t * rdev; 2971 struct list_head *tmp; 2972 if (mddev->pers->resize == NULL) 2973 return -EINVAL; 2974 /* The "size" is the amount of each device that is used. 2975 * This can only make sense for arrays with redundancy. 2976 * linear and raid0 always use whatever space is available 2977 * We can only consider changing the size if no resync 2978 * or reconstruction is happening, and if the new size 2979 * is acceptable. It must fit before the sb_offset or, 2980 * if that is <data_offset, it must fit before the 2981 * size of each device. 2982 * If size is zero, we find the largest size that fits. 2983 */ 2984 if (mddev->sync_thread) 2985 return -EBUSY; 2986 ITERATE_RDEV(mddev,rdev,tmp) { 2987 sector_t avail; 2988 int fit = (info->size == 0); 2989 if (rdev->sb_offset > rdev->data_offset) 2990 avail = (rdev->sb_offset*2) - rdev->data_offset; 2991 else 2992 avail = get_capacity(rdev->bdev->bd_disk) 2993 - rdev->data_offset; 2994 if (fit && (info->size == 0 || info->size > avail/2)) 2995 info->size = avail/2; 2996 if (avail < ((sector_t)info->size << 1)) 2997 return -ENOSPC; 2998 } 2999 rv = mddev->pers->resize(mddev, (sector_t)info->size *2); 3000 if (!rv) { 3001 struct block_device *bdev; 3002 3003 bdev = bdget_disk(mddev->gendisk, 0); 3004 if (bdev) { 3005 down(&bdev->bd_inode->i_sem); 3006 i_size_write(bdev->bd_inode, mddev->array_size << 10); 3007 up(&bdev->bd_inode->i_sem); 3008 bdput(bdev); 3009 } 3010 } 3011 } 3012 if (mddev->raid_disks != info->raid_disks) { 3013 /* change the number of raid disks */ 3014 if (mddev->pers->reshape == NULL) 3015 return -EINVAL; 3016 if (info->raid_disks <= 0 || 3017 info->raid_disks >= mddev->max_disks) 3018 return -EINVAL; 3019 if (mddev->sync_thread) 3020 return -EBUSY; 3021 rv = mddev->pers->reshape(mddev, info->raid_disks); 3022 if (!rv) { 3023 struct block_device *bdev; 3024 3025 bdev = bdget_disk(mddev->gendisk, 0); 3026 if (bdev) { 3027 down(&bdev->bd_inode->i_sem); 3028 i_size_write(bdev->bd_inode, mddev->array_size << 10); 3029 up(&bdev->bd_inode->i_sem); 3030 bdput(bdev); 3031 } 3032 } 3033 } 3034 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 3035 if (mddev->pers->quiesce == NULL) 3036 return -EINVAL; 3037 if (mddev->recovery || mddev->sync_thread) 3038 return -EBUSY; 3039 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 3040 /* add the bitmap */ 3041 if (mddev->bitmap) 3042 return -EEXIST; 3043 if (mddev->default_bitmap_offset == 0) 3044 return -EINVAL; 3045 mddev->bitmap_offset = mddev->default_bitmap_offset; 3046 mddev->pers->quiesce(mddev, 1); 3047 rv = bitmap_create(mddev); 3048 if (rv) 3049 bitmap_destroy(mddev); 3050 mddev->pers->quiesce(mddev, 0); 3051 } else { 3052 /* remove the bitmap */ 3053 if (!mddev->bitmap) 3054 return -ENOENT; 3055 if (mddev->bitmap->file) 3056 return -EINVAL; 3057 mddev->pers->quiesce(mddev, 1); 3058 bitmap_destroy(mddev); 3059 mddev->pers->quiesce(mddev, 0); 3060 mddev->bitmap_offset = 0; 3061 } 3062 } 3063 md_update_sb(mddev); 3064 return rv; 3065 } 3066 3067 static int set_disk_faulty(mddev_t *mddev, dev_t dev) 3068 { 3069 mdk_rdev_t *rdev; 3070 3071 if (mddev->pers == NULL) 3072 return -ENODEV; 3073 3074 rdev = find_rdev(mddev, dev); 3075 if (!rdev) 3076 return -ENODEV; 3077 3078 md_error(mddev, rdev); 3079 return 0; 3080 } 3081 3082 static int md_ioctl(struct inode *inode, struct file *file, 3083 unsigned int cmd, unsigned long arg) 3084 { 3085 int err = 0; 3086 void __user *argp = (void __user *)arg; 3087 struct hd_geometry __user *loc = argp; 3088 mddev_t *mddev = NULL; 3089 3090 if (!capable(CAP_SYS_ADMIN)) 3091 return -EACCES; 3092 3093 /* 3094 * Commands dealing with the RAID driver but not any 3095 * particular array: 3096 */ 3097 switch (cmd) 3098 { 3099 case RAID_VERSION: 3100 err = get_version(argp); 3101 goto done; 3102 3103 case PRINT_RAID_DEBUG: 3104 err = 0; 3105 md_print_devices(); 3106 goto done; 3107 3108 #ifndef MODULE 3109 case RAID_AUTORUN: 3110 err = 0; 3111 autostart_arrays(arg); 3112 goto done; 3113 #endif 3114 default:; 3115 } 3116 3117 /* 3118 * Commands creating/starting a new array: 3119 */ 3120 3121 mddev = inode->i_bdev->bd_disk->private_data; 3122 3123 if (!mddev) { 3124 BUG(); 3125 goto abort; 3126 } 3127 3128 3129 if (cmd == START_ARRAY) { 3130 /* START_ARRAY doesn't need to lock the array as autostart_array 3131 * does the locking, and it could even be a different array 3132 */ 3133 static int cnt = 3; 3134 if (cnt > 0 ) { 3135 printk(KERN_WARNING 3136 "md: %s(pid %d) used deprecated START_ARRAY ioctl. " 3137 "This will not be supported beyond 2.6\n", 3138 current->comm, current->pid); 3139 cnt--; 3140 } 3141 err = autostart_array(new_decode_dev(arg)); 3142 if (err) { 3143 printk(KERN_WARNING "md: autostart failed!\n"); 3144 goto abort; 3145 } 3146 goto done; 3147 } 3148 3149 err = mddev_lock(mddev); 3150 if (err) { 3151 printk(KERN_INFO 3152 "md: ioctl lock interrupted, reason %d, cmd %d\n", 3153 err, cmd); 3154 goto abort; 3155 } 3156 3157 switch (cmd) 3158 { 3159 case SET_ARRAY_INFO: 3160 { 3161 mdu_array_info_t info; 3162 if (!arg) 3163 memset(&info, 0, sizeof(info)); 3164 else if (copy_from_user(&info, argp, sizeof(info))) { 3165 err = -EFAULT; 3166 goto abort_unlock; 3167 } 3168 if (mddev->pers) { 3169 err = update_array_info(mddev, &info); 3170 if (err) { 3171 printk(KERN_WARNING "md: couldn't update" 3172 " array info. %d\n", err); 3173 goto abort_unlock; 3174 } 3175 goto done_unlock; 3176 } 3177 if (!list_empty(&mddev->disks)) { 3178 printk(KERN_WARNING 3179 "md: array %s already has disks!\n", 3180 mdname(mddev)); 3181 err = -EBUSY; 3182 goto abort_unlock; 3183 } 3184 if (mddev->raid_disks) { 3185 printk(KERN_WARNING 3186 "md: array %s already initialised!\n", 3187 mdname(mddev)); 3188 err = -EBUSY; 3189 goto abort_unlock; 3190 } 3191 err = set_array_info(mddev, &info); 3192 if (err) { 3193 printk(KERN_WARNING "md: couldn't set" 3194 " array info. %d\n", err); 3195 goto abort_unlock; 3196 } 3197 } 3198 goto done_unlock; 3199 3200 default:; 3201 } 3202 3203 /* 3204 * Commands querying/configuring an existing array: 3205 */ 3206 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 3207 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */ 3208 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 3209 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) { 3210 err = -ENODEV; 3211 goto abort_unlock; 3212 } 3213 3214 /* 3215 * Commands even a read-only array can execute: 3216 */ 3217 switch (cmd) 3218 { 3219 case GET_ARRAY_INFO: 3220 err = get_array_info(mddev, argp); 3221 goto done_unlock; 3222 3223 case GET_BITMAP_FILE: 3224 err = get_bitmap_file(mddev, argp); 3225 goto done_unlock; 3226 3227 case GET_DISK_INFO: 3228 err = get_disk_info(mddev, argp); 3229 goto done_unlock; 3230 3231 case RESTART_ARRAY_RW: 3232 err = restart_array(mddev); 3233 goto done_unlock; 3234 3235 case STOP_ARRAY: 3236 err = do_md_stop (mddev, 0); 3237 goto done_unlock; 3238 3239 case STOP_ARRAY_RO: 3240 err = do_md_stop (mddev, 1); 3241 goto done_unlock; 3242 3243 /* 3244 * We have a problem here : there is no easy way to give a CHS 3245 * virtual geometry. We currently pretend that we have a 2 heads 3246 * 4 sectors (with a BIG number of cylinders...). This drives 3247 * dosfs just mad... ;-) 3248 */ 3249 case HDIO_GETGEO: 3250 if (!loc) { 3251 err = -EINVAL; 3252 goto abort_unlock; 3253 } 3254 err = put_user (2, (char __user *) &loc->heads); 3255 if (err) 3256 goto abort_unlock; 3257 err = put_user (4, (char __user *) &loc->sectors); 3258 if (err) 3259 goto abort_unlock; 3260 err = put_user(get_capacity(mddev->gendisk)/8, 3261 (short __user *) &loc->cylinders); 3262 if (err) 3263 goto abort_unlock; 3264 err = put_user (get_start_sect(inode->i_bdev), 3265 (long __user *) &loc->start); 3266 goto done_unlock; 3267 } 3268 3269 /* 3270 * The remaining ioctls are changing the state of the 3271 * superblock, so we do not allow them on read-only arrays. 3272 * However non-MD ioctls (e.g. get-size) will still come through 3273 * here and hit the 'default' below, so only disallow 3274 * 'md' ioctls, and switch to rw mode if started auto-readonly. 3275 */ 3276 if (_IOC_TYPE(cmd) == MD_MAJOR && 3277 mddev->ro && mddev->pers) { 3278 if (mddev->ro == 2) { 3279 mddev->ro = 0; 3280 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3281 md_wakeup_thread(mddev->thread); 3282 3283 } else { 3284 err = -EROFS; 3285 goto abort_unlock; 3286 } 3287 } 3288 3289 switch (cmd) 3290 { 3291 case ADD_NEW_DISK: 3292 { 3293 mdu_disk_info_t info; 3294 if (copy_from_user(&info, argp, sizeof(info))) 3295 err = -EFAULT; 3296 else 3297 err = add_new_disk(mddev, &info); 3298 goto done_unlock; 3299 } 3300 3301 case HOT_REMOVE_DISK: 3302 err = hot_remove_disk(mddev, new_decode_dev(arg)); 3303 goto done_unlock; 3304 3305 case HOT_ADD_DISK: 3306 err = hot_add_disk(mddev, new_decode_dev(arg)); 3307 goto done_unlock; 3308 3309 case SET_DISK_FAULTY: 3310 err = set_disk_faulty(mddev, new_decode_dev(arg)); 3311 goto done_unlock; 3312 3313 case RUN_ARRAY: 3314 err = do_md_run (mddev); 3315 goto done_unlock; 3316 3317 case SET_BITMAP_FILE: 3318 err = set_bitmap_file(mddev, (int)arg); 3319 goto done_unlock; 3320 3321 default: 3322 if (_IOC_TYPE(cmd) == MD_MAJOR) 3323 printk(KERN_WARNING "md: %s(pid %d) used" 3324 " obsolete MD ioctl, upgrade your" 3325 " software to use new ictls.\n", 3326 current->comm, current->pid); 3327 err = -EINVAL; 3328 goto abort_unlock; 3329 } 3330 3331 done_unlock: 3332 abort_unlock: 3333 mddev_unlock(mddev); 3334 3335 return err; 3336 done: 3337 if (err) 3338 MD_BUG(); 3339 abort: 3340 return err; 3341 } 3342 3343 static int md_open(struct inode *inode, struct file *file) 3344 { 3345 /* 3346 * Succeed if we can lock the mddev, which confirms that 3347 * it isn't being stopped right now. 3348 */ 3349 mddev_t *mddev = inode->i_bdev->bd_disk->private_data; 3350 int err; 3351 3352 if ((err = mddev_lock(mddev))) 3353 goto out; 3354 3355 err = 0; 3356 mddev_get(mddev); 3357 mddev_unlock(mddev); 3358 3359 check_disk_change(inode->i_bdev); 3360 out: 3361 return err; 3362 } 3363 3364 static int md_release(struct inode *inode, struct file * file) 3365 { 3366 mddev_t *mddev = inode->i_bdev->bd_disk->private_data; 3367 3368 if (!mddev) 3369 BUG(); 3370 mddev_put(mddev); 3371 3372 return 0; 3373 } 3374 3375 static int md_media_changed(struct gendisk *disk) 3376 { 3377 mddev_t *mddev = disk->private_data; 3378 3379 return mddev->changed; 3380 } 3381 3382 static int md_revalidate(struct gendisk *disk) 3383 { 3384 mddev_t *mddev = disk->private_data; 3385 3386 mddev->changed = 0; 3387 return 0; 3388 } 3389 static struct block_device_operations md_fops = 3390 { 3391 .owner = THIS_MODULE, 3392 .open = md_open, 3393 .release = md_release, 3394 .ioctl = md_ioctl, 3395 .media_changed = md_media_changed, 3396 .revalidate_disk= md_revalidate, 3397 }; 3398 3399 static int md_thread(void * arg) 3400 { 3401 mdk_thread_t *thread = arg; 3402 3403 /* 3404 * md_thread is a 'system-thread', it's priority should be very 3405 * high. We avoid resource deadlocks individually in each 3406 * raid personality. (RAID5 does preallocation) We also use RR and 3407 * the very same RT priority as kswapd, thus we will never get 3408 * into a priority inversion deadlock. 3409 * 3410 * we definitely have to have equal or higher priority than 3411 * bdflush, otherwise bdflush will deadlock if there are too 3412 * many dirty RAID5 blocks. 3413 */ 3414 3415 allow_signal(SIGKILL); 3416 complete(thread->event); 3417 while (!kthread_should_stop()) { 3418 void (*run)(mddev_t *); 3419 3420 wait_event_interruptible_timeout(thread->wqueue, 3421 test_bit(THREAD_WAKEUP, &thread->flags) 3422 || kthread_should_stop(), 3423 thread->timeout); 3424 try_to_freeze(); 3425 3426 clear_bit(THREAD_WAKEUP, &thread->flags); 3427 3428 run = thread->run; 3429 if (run) 3430 run(thread->mddev); 3431 } 3432 3433 return 0; 3434 } 3435 3436 void md_wakeup_thread(mdk_thread_t *thread) 3437 { 3438 if (thread) { 3439 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); 3440 set_bit(THREAD_WAKEUP, &thread->flags); 3441 wake_up(&thread->wqueue); 3442 } 3443 } 3444 3445 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, 3446 const char *name) 3447 { 3448 mdk_thread_t *thread; 3449 struct completion event; 3450 3451 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL); 3452 if (!thread) 3453 return NULL; 3454 3455 memset(thread, 0, sizeof(mdk_thread_t)); 3456 init_waitqueue_head(&thread->wqueue); 3457 3458 init_completion(&event); 3459 thread->event = &event; 3460 thread->run = run; 3461 thread->mddev = mddev; 3462 thread->name = name; 3463 thread->timeout = MAX_SCHEDULE_TIMEOUT; 3464 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); 3465 if (IS_ERR(thread->tsk)) { 3466 kfree(thread); 3467 return NULL; 3468 } 3469 wait_for_completion(&event); 3470 return thread; 3471 } 3472 3473 void md_unregister_thread(mdk_thread_t *thread) 3474 { 3475 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid); 3476 3477 kthread_stop(thread->tsk); 3478 kfree(thread); 3479 } 3480 3481 void md_error(mddev_t *mddev, mdk_rdev_t *rdev) 3482 { 3483 if (!mddev) { 3484 MD_BUG(); 3485 return; 3486 } 3487 3488 if (!rdev || test_bit(Faulty, &rdev->flags)) 3489 return; 3490 /* 3491 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", 3492 mdname(mddev), 3493 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), 3494 __builtin_return_address(0),__builtin_return_address(1), 3495 __builtin_return_address(2),__builtin_return_address(3)); 3496 */ 3497 if (!mddev->pers->error_handler) 3498 return; 3499 mddev->pers->error_handler(mddev,rdev); 3500 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 3501 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3502 md_wakeup_thread(mddev->thread); 3503 } 3504 3505 /* seq_file implementation /proc/mdstat */ 3506 3507 static void status_unused(struct seq_file *seq) 3508 { 3509 int i = 0; 3510 mdk_rdev_t *rdev; 3511 struct list_head *tmp; 3512 3513 seq_printf(seq, "unused devices: "); 3514 3515 ITERATE_RDEV_PENDING(rdev,tmp) { 3516 char b[BDEVNAME_SIZE]; 3517 i++; 3518 seq_printf(seq, "%s ", 3519 bdevname(rdev->bdev,b)); 3520 } 3521 if (!i) 3522 seq_printf(seq, "<none>"); 3523 3524 seq_printf(seq, "\n"); 3525 } 3526 3527 3528 static void status_resync(struct seq_file *seq, mddev_t * mddev) 3529 { 3530 unsigned long max_blocks, resync, res, dt, db, rt; 3531 3532 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2; 3533 3534 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 3535 max_blocks = mddev->resync_max_sectors >> 1; 3536 else 3537 max_blocks = mddev->size; 3538 3539 /* 3540 * Should not happen. 3541 */ 3542 if (!max_blocks) { 3543 MD_BUG(); 3544 return; 3545 } 3546 res = (resync/1024)*1000/(max_blocks/1024 + 1); 3547 { 3548 int i, x = res/50, y = 20-x; 3549 seq_printf(seq, "["); 3550 for (i = 0; i < x; i++) 3551 seq_printf(seq, "="); 3552 seq_printf(seq, ">"); 3553 for (i = 0; i < y; i++) 3554 seq_printf(seq, "."); 3555 seq_printf(seq, "] "); 3556 } 3557 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)", 3558 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 3559 "resync" : "recovery"), 3560 res/10, res % 10, resync, max_blocks); 3561 3562 /* 3563 * We do not want to overflow, so the order of operands and 3564 * the * 100 / 100 trick are important. We do a +1 to be 3565 * safe against division by zero. We only estimate anyway. 3566 * 3567 * dt: time from mark until now 3568 * db: blocks written from mark until now 3569 * rt: remaining time 3570 */ 3571 dt = ((jiffies - mddev->resync_mark) / HZ); 3572 if (!dt) dt++; 3573 db = resync - (mddev->resync_mark_cnt/2); 3574 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100; 3575 3576 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6); 3577 3578 seq_printf(seq, " speed=%ldK/sec", db/dt); 3579 } 3580 3581 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 3582 { 3583 struct list_head *tmp; 3584 loff_t l = *pos; 3585 mddev_t *mddev; 3586 3587 if (l >= 0x10000) 3588 return NULL; 3589 if (!l--) 3590 /* header */ 3591 return (void*)1; 3592 3593 spin_lock(&all_mddevs_lock); 3594 list_for_each(tmp,&all_mddevs) 3595 if (!l--) { 3596 mddev = list_entry(tmp, mddev_t, all_mddevs); 3597 mddev_get(mddev); 3598 spin_unlock(&all_mddevs_lock); 3599 return mddev; 3600 } 3601 spin_unlock(&all_mddevs_lock); 3602 if (!l--) 3603 return (void*)2;/* tail */ 3604 return NULL; 3605 } 3606 3607 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3608 { 3609 struct list_head *tmp; 3610 mddev_t *next_mddev, *mddev = v; 3611 3612 ++*pos; 3613 if (v == (void*)2) 3614 return NULL; 3615 3616 spin_lock(&all_mddevs_lock); 3617 if (v == (void*)1) 3618 tmp = all_mddevs.next; 3619 else 3620 tmp = mddev->all_mddevs.next; 3621 if (tmp != &all_mddevs) 3622 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); 3623 else { 3624 next_mddev = (void*)2; 3625 *pos = 0x10000; 3626 } 3627 spin_unlock(&all_mddevs_lock); 3628 3629 if (v != (void*)1) 3630 mddev_put(mddev); 3631 return next_mddev; 3632 3633 } 3634 3635 static void md_seq_stop(struct seq_file *seq, void *v) 3636 { 3637 mddev_t *mddev = v; 3638 3639 if (mddev && v != (void*)1 && v != (void*)2) 3640 mddev_put(mddev); 3641 } 3642 3643 static int md_seq_show(struct seq_file *seq, void *v) 3644 { 3645 mddev_t *mddev = v; 3646 sector_t size; 3647 struct list_head *tmp2; 3648 mdk_rdev_t *rdev; 3649 int i; 3650 struct bitmap *bitmap; 3651 3652 if (v == (void*)1) { 3653 seq_printf(seq, "Personalities : "); 3654 spin_lock(&pers_lock); 3655 for (i = 0; i < MAX_PERSONALITY; i++) 3656 if (pers[i]) 3657 seq_printf(seq, "[%s] ", pers[i]->name); 3658 3659 spin_unlock(&pers_lock); 3660 seq_printf(seq, "\n"); 3661 return 0; 3662 } 3663 if (v == (void*)2) { 3664 status_unused(seq); 3665 return 0; 3666 } 3667 3668 if (mddev_lock(mddev)!=0) 3669 return -EINTR; 3670 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 3671 seq_printf(seq, "%s : %sactive", mdname(mddev), 3672 mddev->pers ? "" : "in"); 3673 if (mddev->pers) { 3674 if (mddev->ro==1) 3675 seq_printf(seq, " (read-only)"); 3676 if (mddev->ro==2) 3677 seq_printf(seq, "(auto-read-only)"); 3678 seq_printf(seq, " %s", mddev->pers->name); 3679 } 3680 3681 size = 0; 3682 ITERATE_RDEV(mddev,rdev,tmp2) { 3683 char b[BDEVNAME_SIZE]; 3684 seq_printf(seq, " %s[%d]", 3685 bdevname(rdev->bdev,b), rdev->desc_nr); 3686 if (test_bit(WriteMostly, &rdev->flags)) 3687 seq_printf(seq, "(W)"); 3688 if (test_bit(Faulty, &rdev->flags)) { 3689 seq_printf(seq, "(F)"); 3690 continue; 3691 } else if (rdev->raid_disk < 0) 3692 seq_printf(seq, "(S)"); /* spare */ 3693 size += rdev->size; 3694 } 3695 3696 if (!list_empty(&mddev->disks)) { 3697 if (mddev->pers) 3698 seq_printf(seq, "\n %llu blocks", 3699 (unsigned long long)mddev->array_size); 3700 else 3701 seq_printf(seq, "\n %llu blocks", 3702 (unsigned long long)size); 3703 } 3704 if (mddev->persistent) { 3705 if (mddev->major_version != 0 || 3706 mddev->minor_version != 90) { 3707 seq_printf(seq," super %d.%d", 3708 mddev->major_version, 3709 mddev->minor_version); 3710 } 3711 } else 3712 seq_printf(seq, " super non-persistent"); 3713 3714 if (mddev->pers) { 3715 mddev->pers->status (seq, mddev); 3716 seq_printf(seq, "\n "); 3717 if (mddev->curr_resync > 2) { 3718 status_resync (seq, mddev); 3719 seq_printf(seq, "\n "); 3720 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 3721 seq_printf(seq, "\tresync=DELAYED\n "); 3722 else if (mddev->recovery_cp < MaxSector) 3723 seq_printf(seq, "\tresync=PENDING\n "); 3724 } else 3725 seq_printf(seq, "\n "); 3726 3727 if ((bitmap = mddev->bitmap)) { 3728 unsigned long chunk_kb; 3729 unsigned long flags; 3730 spin_lock_irqsave(&bitmap->lock, flags); 3731 chunk_kb = bitmap->chunksize >> 10; 3732 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 3733 "%lu%s chunk", 3734 bitmap->pages - bitmap->missing_pages, 3735 bitmap->pages, 3736 (bitmap->pages - bitmap->missing_pages) 3737 << (PAGE_SHIFT - 10), 3738 chunk_kb ? chunk_kb : bitmap->chunksize, 3739 chunk_kb ? "KB" : "B"); 3740 if (bitmap->file) { 3741 seq_printf(seq, ", file: "); 3742 seq_path(seq, bitmap->file->f_vfsmnt, 3743 bitmap->file->f_dentry," \t\n"); 3744 } 3745 3746 seq_printf(seq, "\n"); 3747 spin_unlock_irqrestore(&bitmap->lock, flags); 3748 } 3749 3750 seq_printf(seq, "\n"); 3751 } 3752 mddev_unlock(mddev); 3753 3754 return 0; 3755 } 3756 3757 static struct seq_operations md_seq_ops = { 3758 .start = md_seq_start, 3759 .next = md_seq_next, 3760 .stop = md_seq_stop, 3761 .show = md_seq_show, 3762 }; 3763 3764 static int md_seq_open(struct inode *inode, struct file *file) 3765 { 3766 int error; 3767 3768 error = seq_open(file, &md_seq_ops); 3769 return error; 3770 } 3771 3772 static struct file_operations md_seq_fops = { 3773 .open = md_seq_open, 3774 .read = seq_read, 3775 .llseek = seq_lseek, 3776 .release = seq_release, 3777 }; 3778 3779 int register_md_personality(int pnum, mdk_personality_t *p) 3780 { 3781 if (pnum >= MAX_PERSONALITY) { 3782 printk(KERN_ERR 3783 "md: tried to install personality %s as nr %d, but max is %lu\n", 3784 p->name, pnum, MAX_PERSONALITY-1); 3785 return -EINVAL; 3786 } 3787 3788 spin_lock(&pers_lock); 3789 if (pers[pnum]) { 3790 spin_unlock(&pers_lock); 3791 return -EBUSY; 3792 } 3793 3794 pers[pnum] = p; 3795 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum); 3796 spin_unlock(&pers_lock); 3797 return 0; 3798 } 3799 3800 int unregister_md_personality(int pnum) 3801 { 3802 if (pnum >= MAX_PERSONALITY) 3803 return -EINVAL; 3804 3805 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name); 3806 spin_lock(&pers_lock); 3807 pers[pnum] = NULL; 3808 spin_unlock(&pers_lock); 3809 return 0; 3810 } 3811 3812 static int is_mddev_idle(mddev_t *mddev) 3813 { 3814 mdk_rdev_t * rdev; 3815 struct list_head *tmp; 3816 int idle; 3817 unsigned long curr_events; 3818 3819 idle = 1; 3820 ITERATE_RDEV(mddev,rdev,tmp) { 3821 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 3822 curr_events = disk_stat_read(disk, sectors[0]) + 3823 disk_stat_read(disk, sectors[1]) - 3824 atomic_read(&disk->sync_io); 3825 /* Allow some slack between valud of curr_events and last_events, 3826 * as there are some uninteresting races. 3827 * Note: the following is an unsigned comparison. 3828 */ 3829 if ((curr_events - rdev->last_events + 32) > 64) { 3830 rdev->last_events = curr_events; 3831 idle = 0; 3832 } 3833 } 3834 return idle; 3835 } 3836 3837 void md_done_sync(mddev_t *mddev, int blocks, int ok) 3838 { 3839 /* another "blocks" (512byte) blocks have been synced */ 3840 atomic_sub(blocks, &mddev->recovery_active); 3841 wake_up(&mddev->recovery_wait); 3842 if (!ok) { 3843 set_bit(MD_RECOVERY_ERR, &mddev->recovery); 3844 md_wakeup_thread(mddev->thread); 3845 // stop recovery, signal do_sync .... 3846 } 3847 } 3848 3849 3850 /* md_write_start(mddev, bi) 3851 * If we need to update some array metadata (e.g. 'active' flag 3852 * in superblock) before writing, schedule a superblock update 3853 * and wait for it to complete. 3854 */ 3855 void md_write_start(mddev_t *mddev, struct bio *bi) 3856 { 3857 if (bio_data_dir(bi) != WRITE) 3858 return; 3859 3860 BUG_ON(mddev->ro == 1); 3861 if (mddev->ro == 2) { 3862 /* need to switch to read/write */ 3863 mddev->ro = 0; 3864 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3865 md_wakeup_thread(mddev->thread); 3866 } 3867 atomic_inc(&mddev->writes_pending); 3868 if (mddev->in_sync) { 3869 spin_lock_irq(&mddev->write_lock); 3870 if (mddev->in_sync) { 3871 mddev->in_sync = 0; 3872 mddev->sb_dirty = 1; 3873 md_wakeup_thread(mddev->thread); 3874 } 3875 spin_unlock_irq(&mddev->write_lock); 3876 } 3877 wait_event(mddev->sb_wait, mddev->sb_dirty==0); 3878 } 3879 3880 void md_write_end(mddev_t *mddev) 3881 { 3882 if (atomic_dec_and_test(&mddev->writes_pending)) { 3883 if (mddev->safemode == 2) 3884 md_wakeup_thread(mddev->thread); 3885 else 3886 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 3887 } 3888 } 3889 3890 static DECLARE_WAIT_QUEUE_HEAD(resync_wait); 3891 3892 #define SYNC_MARKS 10 3893 #define SYNC_MARK_STEP (3*HZ) 3894 static void md_do_sync(mddev_t *mddev) 3895 { 3896 mddev_t *mddev2; 3897 unsigned int currspeed = 0, 3898 window; 3899 sector_t max_sectors,j, io_sectors; 3900 unsigned long mark[SYNC_MARKS]; 3901 sector_t mark_cnt[SYNC_MARKS]; 3902 int last_mark,m; 3903 struct list_head *tmp; 3904 sector_t last_check; 3905 int skipped = 0; 3906 3907 /* just incase thread restarts... */ 3908 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 3909 return; 3910 3911 /* we overload curr_resync somewhat here. 3912 * 0 == not engaged in resync at all 3913 * 2 == checking that there is no conflict with another sync 3914 * 1 == like 2, but have yielded to allow conflicting resync to 3915 * commense 3916 * other == active in resync - this many blocks 3917 * 3918 * Before starting a resync we must have set curr_resync to 3919 * 2, and then checked that every "conflicting" array has curr_resync 3920 * less than ours. When we find one that is the same or higher 3921 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 3922 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 3923 * This will mean we have to start checking from the beginning again. 3924 * 3925 */ 3926 3927 do { 3928 mddev->curr_resync = 2; 3929 3930 try_again: 3931 if (signal_pending(current) || 3932 kthread_should_stop()) { 3933 flush_signals(current); 3934 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 3935 goto skip; 3936 } 3937 ITERATE_MDDEV(mddev2,tmp) { 3938 if (mddev2 == mddev) 3939 continue; 3940 if (mddev2->curr_resync && 3941 match_mddev_units(mddev,mddev2)) { 3942 DEFINE_WAIT(wq); 3943 if (mddev < mddev2 && mddev->curr_resync == 2) { 3944 /* arbitrarily yield */ 3945 mddev->curr_resync = 1; 3946 wake_up(&resync_wait); 3947 } 3948 if (mddev > mddev2 && mddev->curr_resync == 1) 3949 /* no need to wait here, we can wait the next 3950 * time 'round when curr_resync == 2 3951 */ 3952 continue; 3953 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 3954 if (!signal_pending(current) && 3955 !kthread_should_stop() && 3956 mddev2->curr_resync >= mddev->curr_resync) { 3957 printk(KERN_INFO "md: delaying resync of %s" 3958 " until %s has finished resync (they" 3959 " share one or more physical units)\n", 3960 mdname(mddev), mdname(mddev2)); 3961 mddev_put(mddev2); 3962 schedule(); 3963 finish_wait(&resync_wait, &wq); 3964 goto try_again; 3965 } 3966 finish_wait(&resync_wait, &wq); 3967 } 3968 } 3969 } while (mddev->curr_resync < 2); 3970 3971 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 3972 /* resync follows the size requested by the personality, 3973 * which defaults to physical size, but can be virtual size 3974 */ 3975 max_sectors = mddev->resync_max_sectors; 3976 mddev->resync_mismatches = 0; 3977 } else 3978 /* recovery follows the physical size of devices */ 3979 max_sectors = mddev->size << 1; 3980 3981 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev)); 3982 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:" 3983 " %d KB/sec/disc.\n", sysctl_speed_limit_min); 3984 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 3985 "(but not more than %d KB/sec) for reconstruction.\n", 3986 sysctl_speed_limit_max); 3987 3988 is_mddev_idle(mddev); /* this also initializes IO event counters */ 3989 /* we don't use the checkpoint if there's a bitmap */ 3990 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap 3991 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 3992 j = mddev->recovery_cp; 3993 else 3994 j = 0; 3995 io_sectors = 0; 3996 for (m = 0; m < SYNC_MARKS; m++) { 3997 mark[m] = jiffies; 3998 mark_cnt[m] = io_sectors; 3999 } 4000 last_mark = 0; 4001 mddev->resync_mark = mark[last_mark]; 4002 mddev->resync_mark_cnt = mark_cnt[last_mark]; 4003 4004 /* 4005 * Tune reconstruction: 4006 */ 4007 window = 32*(PAGE_SIZE/512); 4008 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", 4009 window/2,(unsigned long long) max_sectors/2); 4010 4011 atomic_set(&mddev->recovery_active, 0); 4012 init_waitqueue_head(&mddev->recovery_wait); 4013 last_check = 0; 4014 4015 if (j>2) { 4016 printk(KERN_INFO 4017 "md: resuming recovery of %s from checkpoint.\n", 4018 mdname(mddev)); 4019 mddev->curr_resync = j; 4020 } 4021 4022 while (j < max_sectors) { 4023 sector_t sectors; 4024 4025 skipped = 0; 4026 sectors = mddev->pers->sync_request(mddev, j, &skipped, 4027 currspeed < sysctl_speed_limit_min); 4028 if (sectors == 0) { 4029 set_bit(MD_RECOVERY_ERR, &mddev->recovery); 4030 goto out; 4031 } 4032 4033 if (!skipped) { /* actual IO requested */ 4034 io_sectors += sectors; 4035 atomic_add(sectors, &mddev->recovery_active); 4036 } 4037 4038 j += sectors; 4039 if (j>1) mddev->curr_resync = j; 4040 4041 4042 if (last_check + window > io_sectors || j == max_sectors) 4043 continue; 4044 4045 last_check = io_sectors; 4046 4047 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) || 4048 test_bit(MD_RECOVERY_ERR, &mddev->recovery)) 4049 break; 4050 4051 repeat: 4052 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 4053 /* step marks */ 4054 int next = (last_mark+1) % SYNC_MARKS; 4055 4056 mddev->resync_mark = mark[next]; 4057 mddev->resync_mark_cnt = mark_cnt[next]; 4058 mark[next] = jiffies; 4059 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 4060 last_mark = next; 4061 } 4062 4063 4064 if (signal_pending(current) || kthread_should_stop()) { 4065 /* 4066 * got a signal, exit. 4067 */ 4068 printk(KERN_INFO 4069 "md: md_do_sync() got signal ... exiting\n"); 4070 flush_signals(current); 4071 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4072 goto out; 4073 } 4074 4075 /* 4076 * this loop exits only if either when we are slower than 4077 * the 'hard' speed limit, or the system was IO-idle for 4078 * a jiffy. 4079 * the system might be non-idle CPU-wise, but we only care 4080 * about not overloading the IO subsystem. (things like an 4081 * e2fsck being done on the RAID array should execute fast) 4082 */ 4083 mddev->queue->unplug_fn(mddev->queue); 4084 cond_resched(); 4085 4086 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 4087 /((jiffies-mddev->resync_mark)/HZ +1) +1; 4088 4089 if (currspeed > sysctl_speed_limit_min) { 4090 if ((currspeed > sysctl_speed_limit_max) || 4091 !is_mddev_idle(mddev)) { 4092 msleep_interruptible(250); 4093 goto repeat; 4094 } 4095 } 4096 } 4097 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev)); 4098 /* 4099 * this also signals 'finished resyncing' to md_stop 4100 */ 4101 out: 4102 mddev->queue->unplug_fn(mddev->queue); 4103 4104 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 4105 4106 /* tell personality that we are finished */ 4107 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 4108 4109 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && 4110 mddev->curr_resync > 2 && 4111 mddev->curr_resync >= mddev->recovery_cp) { 4112 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 4113 printk(KERN_INFO 4114 "md: checkpointing recovery of %s.\n", 4115 mdname(mddev)); 4116 mddev->recovery_cp = mddev->curr_resync; 4117 } else 4118 mddev->recovery_cp = MaxSector; 4119 } 4120 4121 skip: 4122 mddev->curr_resync = 0; 4123 wake_up(&resync_wait); 4124 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 4125 md_wakeup_thread(mddev->thread); 4126 } 4127 4128 4129 /* 4130 * This routine is regularly called by all per-raid-array threads to 4131 * deal with generic issues like resync and super-block update. 4132 * Raid personalities that don't have a thread (linear/raid0) do not 4133 * need this as they never do any recovery or update the superblock. 4134 * 4135 * It does not do any resync itself, but rather "forks" off other threads 4136 * to do that as needed. 4137 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 4138 * "->recovery" and create a thread at ->sync_thread. 4139 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR) 4140 * and wakeups up this thread which will reap the thread and finish up. 4141 * This thread also removes any faulty devices (with nr_pending == 0). 4142 * 4143 * The overall approach is: 4144 * 1/ if the superblock needs updating, update it. 4145 * 2/ If a recovery thread is running, don't do anything else. 4146 * 3/ If recovery has finished, clean up, possibly marking spares active. 4147 * 4/ If there are any faulty devices, remove them. 4148 * 5/ If array is degraded, try to add spares devices 4149 * 6/ If array has spares or is not in-sync, start a resync thread. 4150 */ 4151 void md_check_recovery(mddev_t *mddev) 4152 { 4153 mdk_rdev_t *rdev; 4154 struct list_head *rtmp; 4155 4156 4157 if (mddev->bitmap) 4158 bitmap_daemon_work(mddev->bitmap); 4159 4160 if (mddev->ro) 4161 return; 4162 4163 if (signal_pending(current)) { 4164 if (mddev->pers->sync_request) { 4165 printk(KERN_INFO "md: %s in immediate safe mode\n", 4166 mdname(mddev)); 4167 mddev->safemode = 2; 4168 } 4169 flush_signals(current); 4170 } 4171 4172 if ( ! ( 4173 mddev->sb_dirty || 4174 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 4175 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 4176 (mddev->safemode == 1) || 4177 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 4178 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 4179 )) 4180 return; 4181 4182 if (mddev_trylock(mddev)==0) { 4183 int spares =0; 4184 4185 spin_lock_irq(&mddev->write_lock); 4186 if (mddev->safemode && !atomic_read(&mddev->writes_pending) && 4187 !mddev->in_sync && mddev->recovery_cp == MaxSector) { 4188 mddev->in_sync = 1; 4189 mddev->sb_dirty = 1; 4190 } 4191 if (mddev->safemode == 1) 4192 mddev->safemode = 0; 4193 spin_unlock_irq(&mddev->write_lock); 4194 4195 if (mddev->sb_dirty) 4196 md_update_sb(mddev); 4197 4198 4199 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 4200 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 4201 /* resync/recovery still happening */ 4202 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4203 goto unlock; 4204 } 4205 if (mddev->sync_thread) { 4206 /* resync has finished, collect result */ 4207 md_unregister_thread(mddev->sync_thread); 4208 mddev->sync_thread = NULL; 4209 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && 4210 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 4211 /* success...*/ 4212 /* activate any spares */ 4213 mddev->pers->spare_active(mddev); 4214 } 4215 md_update_sb(mddev); 4216 4217 /* if array is no-longer degraded, then any saved_raid_disk 4218 * information must be scrapped 4219 */ 4220 if (!mddev->degraded) 4221 ITERATE_RDEV(mddev,rdev,rtmp) 4222 rdev->saved_raid_disk = -1; 4223 4224 mddev->recovery = 0; 4225 /* flag recovery needed just to double check */ 4226 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4227 goto unlock; 4228 } 4229 /* Clear some bits that don't mean anything, but 4230 * might be left set 4231 */ 4232 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4233 clear_bit(MD_RECOVERY_ERR, &mddev->recovery); 4234 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 4235 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 4236 4237 /* no recovery is running. 4238 * remove any failed drives, then 4239 * add spares if possible. 4240 * Spare are also removed and re-added, to allow 4241 * the personality to fail the re-add. 4242 */ 4243 ITERATE_RDEV(mddev,rdev,rtmp) 4244 if (rdev->raid_disk >= 0 && 4245 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) && 4246 atomic_read(&rdev->nr_pending)==0) { 4247 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) { 4248 char nm[20]; 4249 sprintf(nm,"rd%d", rdev->raid_disk); 4250 sysfs_remove_link(&mddev->kobj, nm); 4251 rdev->raid_disk = -1; 4252 } 4253 } 4254 4255 if (mddev->degraded) { 4256 ITERATE_RDEV(mddev,rdev,rtmp) 4257 if (rdev->raid_disk < 0 4258 && !test_bit(Faulty, &rdev->flags)) { 4259 if (mddev->pers->hot_add_disk(mddev,rdev)) { 4260 char nm[20]; 4261 sprintf(nm, "rd%d", rdev->raid_disk); 4262 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); 4263 spares++; 4264 } else 4265 break; 4266 } 4267 } 4268 4269 if (spares) { 4270 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 4271 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 4272 } else if (mddev->recovery_cp < MaxSector) { 4273 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 4274 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 4275 /* nothing to be done ... */ 4276 goto unlock; 4277 4278 if (mddev->pers->sync_request) { 4279 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 4280 if (spares && mddev->bitmap && ! mddev->bitmap->file) { 4281 /* We are adding a device or devices to an array 4282 * which has the bitmap stored on all devices. 4283 * So make sure all bitmap pages get written 4284 */ 4285 bitmap_write_all(mddev->bitmap); 4286 } 4287 mddev->sync_thread = md_register_thread(md_do_sync, 4288 mddev, 4289 "%s_resync"); 4290 if (!mddev->sync_thread) { 4291 printk(KERN_ERR "%s: could not start resync" 4292 " thread...\n", 4293 mdname(mddev)); 4294 /* leave the spares where they are, it shouldn't hurt */ 4295 mddev->recovery = 0; 4296 } else { 4297 md_wakeup_thread(mddev->sync_thread); 4298 } 4299 } 4300 unlock: 4301 mddev_unlock(mddev); 4302 } 4303 } 4304 4305 static int md_notify_reboot(struct notifier_block *this, 4306 unsigned long code, void *x) 4307 { 4308 struct list_head *tmp; 4309 mddev_t *mddev; 4310 4311 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { 4312 4313 printk(KERN_INFO "md: stopping all md devices.\n"); 4314 4315 ITERATE_MDDEV(mddev,tmp) 4316 if (mddev_trylock(mddev)==0) 4317 do_md_stop (mddev, 1); 4318 /* 4319 * certain more exotic SCSI devices are known to be 4320 * volatile wrt too early system reboots. While the 4321 * right place to handle this issue is the given 4322 * driver, we do want to have a safe RAID driver ... 4323 */ 4324 mdelay(1000*1); 4325 } 4326 return NOTIFY_DONE; 4327 } 4328 4329 static struct notifier_block md_notifier = { 4330 .notifier_call = md_notify_reboot, 4331 .next = NULL, 4332 .priority = INT_MAX, /* before any real devices */ 4333 }; 4334 4335 static void md_geninit(void) 4336 { 4337 struct proc_dir_entry *p; 4338 4339 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 4340 4341 p = create_proc_entry("mdstat", S_IRUGO, NULL); 4342 if (p) 4343 p->proc_fops = &md_seq_fops; 4344 } 4345 4346 static int __init md_init(void) 4347 { 4348 int minor; 4349 4350 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d," 4351 " MD_SB_DISKS=%d\n", 4352 MD_MAJOR_VERSION, MD_MINOR_VERSION, 4353 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS); 4354 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI, 4355 BITMAP_MINOR); 4356 4357 if (register_blkdev(MAJOR_NR, "md")) 4358 return -1; 4359 if ((mdp_major=register_blkdev(0, "mdp"))<=0) { 4360 unregister_blkdev(MAJOR_NR, "md"); 4361 return -1; 4362 } 4363 devfs_mk_dir("md"); 4364 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE, 4365 md_probe, NULL, NULL); 4366 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE, 4367 md_probe, NULL, NULL); 4368 4369 for (minor=0; minor < MAX_MD_DEVS; ++minor) 4370 devfs_mk_bdev(MKDEV(MAJOR_NR, minor), 4371 S_IFBLK|S_IRUSR|S_IWUSR, 4372 "md/%d", minor); 4373 4374 for (minor=0; minor < MAX_MD_DEVS; ++minor) 4375 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift), 4376 S_IFBLK|S_IRUSR|S_IWUSR, 4377 "md/mdp%d", minor); 4378 4379 4380 register_reboot_notifier(&md_notifier); 4381 raid_table_header = register_sysctl_table(raid_root_table, 1); 4382 4383 md_geninit(); 4384 return (0); 4385 } 4386 4387 4388 #ifndef MODULE 4389 4390 /* 4391 * Searches all registered partitions for autorun RAID arrays 4392 * at boot time. 4393 */ 4394 static dev_t detected_devices[128]; 4395 static int dev_cnt; 4396 4397 void md_autodetect_dev(dev_t dev) 4398 { 4399 if (dev_cnt >= 0 && dev_cnt < 127) 4400 detected_devices[dev_cnt++] = dev; 4401 } 4402 4403 4404 static void autostart_arrays(int part) 4405 { 4406 mdk_rdev_t *rdev; 4407 int i; 4408 4409 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 4410 4411 for (i = 0; i < dev_cnt; i++) { 4412 dev_t dev = detected_devices[i]; 4413 4414 rdev = md_import_device(dev,0, 0); 4415 if (IS_ERR(rdev)) 4416 continue; 4417 4418 if (test_bit(Faulty, &rdev->flags)) { 4419 MD_BUG(); 4420 continue; 4421 } 4422 list_add(&rdev->same_set, &pending_raid_disks); 4423 } 4424 dev_cnt = 0; 4425 4426 autorun_devices(part); 4427 } 4428 4429 #endif 4430 4431 static __exit void md_exit(void) 4432 { 4433 mddev_t *mddev; 4434 struct list_head *tmp; 4435 int i; 4436 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS); 4437 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift); 4438 for (i=0; i < MAX_MD_DEVS; i++) 4439 devfs_remove("md/%d", i); 4440 for (i=0; i < MAX_MD_DEVS; i++) 4441 devfs_remove("md/d%d", i); 4442 4443 devfs_remove("md"); 4444 4445 unregister_blkdev(MAJOR_NR,"md"); 4446 unregister_blkdev(mdp_major, "mdp"); 4447 unregister_reboot_notifier(&md_notifier); 4448 unregister_sysctl_table(raid_table_header); 4449 remove_proc_entry("mdstat", NULL); 4450 ITERATE_MDDEV(mddev,tmp) { 4451 struct gendisk *disk = mddev->gendisk; 4452 if (!disk) 4453 continue; 4454 export_array(mddev); 4455 del_gendisk(disk); 4456 put_disk(disk); 4457 mddev->gendisk = NULL; 4458 mddev_put(mddev); 4459 } 4460 } 4461 4462 module_init(md_init) 4463 module_exit(md_exit) 4464 4465 static int get_ro(char *buffer, struct kernel_param *kp) 4466 { 4467 return sprintf(buffer, "%d", start_readonly); 4468 } 4469 static int set_ro(const char *val, struct kernel_param *kp) 4470 { 4471 char *e; 4472 int num = simple_strtoul(val, &e, 10); 4473 if (*val && (*e == '\0' || *e == '\n')) { 4474 start_readonly = num; 4475 return 0;; 4476 } 4477 return -EINVAL; 4478 } 4479 4480 module_param_call(start_ro, set_ro, get_ro, NULL, 0600); 4481 4482 EXPORT_SYMBOL(register_md_personality); 4483 EXPORT_SYMBOL(unregister_md_personality); 4484 EXPORT_SYMBOL(md_error); 4485 EXPORT_SYMBOL(md_done_sync); 4486 EXPORT_SYMBOL(md_write_start); 4487 EXPORT_SYMBOL(md_write_end); 4488 EXPORT_SYMBOL(md_register_thread); 4489 EXPORT_SYMBOL(md_unregister_thread); 4490 EXPORT_SYMBOL(md_wakeup_thread); 4491 EXPORT_SYMBOL(md_print_devices); 4492 EXPORT_SYMBOL(md_check_recovery); 4493 MODULE_LICENSE("GPL"); 4494 MODULE_ALIAS("md"); 4495 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 4496