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/kthread.h> 36 #include <linux/blkdev.h> 37 #include <linux/sysctl.h> 38 #include <linux/seq_file.h> 39 #include <linux/buffer_head.h> /* for invalidate_bdev */ 40 #include <linux/poll.h> 41 #include <linux/ctype.h> 42 #include <linux/hdreg.h> 43 #include <linux/proc_fs.h> 44 #include <linux/random.h> 45 #include <linux/reboot.h> 46 #include <linux/file.h> 47 #include <linux/delay.h> 48 #include <linux/raid/md_p.h> 49 #include <linux/raid/md_u.h> 50 #include "md.h" 51 #include "bitmap.h" 52 53 #define DEBUG 0 54 #define dprintk(x...) ((void)(DEBUG && printk(x))) 55 56 57 #ifndef MODULE 58 static void autostart_arrays(int part); 59 #endif 60 61 static LIST_HEAD(pers_list); 62 static DEFINE_SPINLOCK(pers_lock); 63 64 static void md_print_devices(void); 65 66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait); 67 68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } 69 70 /* 71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' 72 * is 1000 KB/sec, so the extra system load does not show up that much. 73 * Increase it if you want to have more _guaranteed_ speed. Note that 74 * the RAID driver will use the maximum available bandwidth if the IO 75 * subsystem is idle. There is also an 'absolute maximum' reconstruction 76 * speed limit - in case reconstruction slows down your system despite 77 * idle IO detection. 78 * 79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. 80 * or /sys/block/mdX/md/sync_speed_{min,max} 81 */ 82 83 static int sysctl_speed_limit_min = 1000; 84 static int sysctl_speed_limit_max = 200000; 85 static inline int speed_min(mddev_t *mddev) 86 { 87 return mddev->sync_speed_min ? 88 mddev->sync_speed_min : sysctl_speed_limit_min; 89 } 90 91 static inline int speed_max(mddev_t *mddev) 92 { 93 return mddev->sync_speed_max ? 94 mddev->sync_speed_max : sysctl_speed_limit_max; 95 } 96 97 static struct ctl_table_header *raid_table_header; 98 99 static ctl_table raid_table[] = { 100 { 101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, 102 .procname = "speed_limit_min", 103 .data = &sysctl_speed_limit_min, 104 .maxlen = sizeof(int), 105 .mode = S_IRUGO|S_IWUSR, 106 .proc_handler = &proc_dointvec, 107 }, 108 { 109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, 110 .procname = "speed_limit_max", 111 .data = &sysctl_speed_limit_max, 112 .maxlen = sizeof(int), 113 .mode = S_IRUGO|S_IWUSR, 114 .proc_handler = &proc_dointvec, 115 }, 116 { .ctl_name = 0 } 117 }; 118 119 static ctl_table raid_dir_table[] = { 120 { 121 .ctl_name = DEV_RAID, 122 .procname = "raid", 123 .maxlen = 0, 124 .mode = S_IRUGO|S_IXUGO, 125 .child = raid_table, 126 }, 127 { .ctl_name = 0 } 128 }; 129 130 static ctl_table raid_root_table[] = { 131 { 132 .ctl_name = CTL_DEV, 133 .procname = "dev", 134 .maxlen = 0, 135 .mode = 0555, 136 .child = raid_dir_table, 137 }, 138 { .ctl_name = 0 } 139 }; 140 141 static struct block_device_operations md_fops; 142 143 static int start_readonly; 144 145 /* 146 * We have a system wide 'event count' that is incremented 147 * on any 'interesting' event, and readers of /proc/mdstat 148 * can use 'poll' or 'select' to find out when the event 149 * count increases. 150 * 151 * Events are: 152 * start array, stop array, error, add device, remove device, 153 * start build, activate spare 154 */ 155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); 156 static atomic_t md_event_count; 157 void md_new_event(mddev_t *mddev) 158 { 159 atomic_inc(&md_event_count); 160 wake_up(&md_event_waiters); 161 } 162 EXPORT_SYMBOL_GPL(md_new_event); 163 164 /* Alternate version that can be called from interrupts 165 * when calling sysfs_notify isn't needed. 166 */ 167 static void md_new_event_inintr(mddev_t *mddev) 168 { 169 atomic_inc(&md_event_count); 170 wake_up(&md_event_waiters); 171 } 172 173 /* 174 * Enables to iterate over all existing md arrays 175 * all_mddevs_lock protects this list. 176 */ 177 static LIST_HEAD(all_mddevs); 178 static DEFINE_SPINLOCK(all_mddevs_lock); 179 180 181 /* 182 * iterates through all used mddevs in the system. 183 * We take care to grab the all_mddevs_lock whenever navigating 184 * the list, and to always hold a refcount when unlocked. 185 * Any code which breaks out of this loop while own 186 * a reference to the current mddev and must mddev_put it. 187 */ 188 #define for_each_mddev(mddev,tmp) \ 189 \ 190 for (({ spin_lock(&all_mddevs_lock); \ 191 tmp = all_mddevs.next; \ 192 mddev = NULL;}); \ 193 ({ if (tmp != &all_mddevs) \ 194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ 195 spin_unlock(&all_mddevs_lock); \ 196 if (mddev) mddev_put(mddev); \ 197 mddev = list_entry(tmp, mddev_t, all_mddevs); \ 198 tmp != &all_mddevs;}); \ 199 ({ spin_lock(&all_mddevs_lock); \ 200 tmp = tmp->next;}) \ 201 ) 202 203 204 /* Rather than calling directly into the personality make_request function, 205 * IO requests come here first so that we can check if the device is 206 * being suspended pending a reconfiguration. 207 * We hold a refcount over the call to ->make_request. By the time that 208 * call has finished, the bio has been linked into some internal structure 209 * and so is visible to ->quiesce(), so we don't need the refcount any more. 210 */ 211 static int md_make_request(struct request_queue *q, struct bio *bio) 212 { 213 mddev_t *mddev = q->queuedata; 214 int rv; 215 if (mddev == NULL || mddev->pers == NULL) { 216 bio_io_error(bio); 217 return 0; 218 } 219 rcu_read_lock(); 220 if (mddev->suspended) { 221 DEFINE_WAIT(__wait); 222 for (;;) { 223 prepare_to_wait(&mddev->sb_wait, &__wait, 224 TASK_UNINTERRUPTIBLE); 225 if (!mddev->suspended) 226 break; 227 rcu_read_unlock(); 228 schedule(); 229 rcu_read_lock(); 230 } 231 finish_wait(&mddev->sb_wait, &__wait); 232 } 233 atomic_inc(&mddev->active_io); 234 rcu_read_unlock(); 235 rv = mddev->pers->make_request(q, bio); 236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) 237 wake_up(&mddev->sb_wait); 238 239 return rv; 240 } 241 242 static void mddev_suspend(mddev_t *mddev) 243 { 244 BUG_ON(mddev->suspended); 245 mddev->suspended = 1; 246 synchronize_rcu(); 247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); 248 mddev->pers->quiesce(mddev, 1); 249 md_unregister_thread(mddev->thread); 250 mddev->thread = NULL; 251 /* we now know that no code is executing in the personality module, 252 * except possibly the tail end of a ->bi_end_io function, but that 253 * is certain to complete before the module has a chance to get 254 * unloaded 255 */ 256 } 257 258 static void mddev_resume(mddev_t *mddev) 259 { 260 mddev->suspended = 0; 261 wake_up(&mddev->sb_wait); 262 mddev->pers->quiesce(mddev, 0); 263 } 264 265 266 static inline mddev_t *mddev_get(mddev_t *mddev) 267 { 268 atomic_inc(&mddev->active); 269 return mddev; 270 } 271 272 static void mddev_delayed_delete(struct work_struct *ws); 273 274 static void mddev_put(mddev_t *mddev) 275 { 276 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 277 return; 278 if (!mddev->raid_disks && list_empty(&mddev->disks) && 279 !mddev->hold_active) { 280 list_del(&mddev->all_mddevs); 281 if (mddev->gendisk) { 282 /* we did a probe so need to clean up. 283 * Call schedule_work inside the spinlock 284 * so that flush_scheduled_work() after 285 * mddev_find will succeed in waiting for the 286 * work to be done. 287 */ 288 INIT_WORK(&mddev->del_work, mddev_delayed_delete); 289 schedule_work(&mddev->del_work); 290 } else 291 kfree(mddev); 292 } 293 spin_unlock(&all_mddevs_lock); 294 } 295 296 static mddev_t * mddev_find(dev_t unit) 297 { 298 mddev_t *mddev, *new = NULL; 299 300 retry: 301 spin_lock(&all_mddevs_lock); 302 303 if (unit) { 304 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 305 if (mddev->unit == unit) { 306 mddev_get(mddev); 307 spin_unlock(&all_mddevs_lock); 308 kfree(new); 309 return mddev; 310 } 311 312 if (new) { 313 list_add(&new->all_mddevs, &all_mddevs); 314 spin_unlock(&all_mddevs_lock); 315 new->hold_active = UNTIL_IOCTL; 316 return new; 317 } 318 } else if (new) { 319 /* find an unused unit number */ 320 static int next_minor = 512; 321 int start = next_minor; 322 int is_free = 0; 323 int dev = 0; 324 while (!is_free) { 325 dev = MKDEV(MD_MAJOR, next_minor); 326 next_minor++; 327 if (next_minor > MINORMASK) 328 next_minor = 0; 329 if (next_minor == start) { 330 /* Oh dear, all in use. */ 331 spin_unlock(&all_mddevs_lock); 332 kfree(new); 333 return NULL; 334 } 335 336 is_free = 1; 337 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 338 if (mddev->unit == dev) { 339 is_free = 0; 340 break; 341 } 342 } 343 new->unit = dev; 344 new->md_minor = MINOR(dev); 345 new->hold_active = UNTIL_STOP; 346 list_add(&new->all_mddevs, &all_mddevs); 347 spin_unlock(&all_mddevs_lock); 348 return new; 349 } 350 spin_unlock(&all_mddevs_lock); 351 352 new = kzalloc(sizeof(*new), GFP_KERNEL); 353 if (!new) 354 return NULL; 355 356 new->unit = unit; 357 if (MAJOR(unit) == MD_MAJOR) 358 new->md_minor = MINOR(unit); 359 else 360 new->md_minor = MINOR(unit) >> MdpMinorShift; 361 362 mutex_init(&new->reconfig_mutex); 363 INIT_LIST_HEAD(&new->disks); 364 INIT_LIST_HEAD(&new->all_mddevs); 365 init_timer(&new->safemode_timer); 366 atomic_set(&new->active, 1); 367 atomic_set(&new->openers, 0); 368 atomic_set(&new->active_io, 0); 369 spin_lock_init(&new->write_lock); 370 init_waitqueue_head(&new->sb_wait); 371 init_waitqueue_head(&new->recovery_wait); 372 new->reshape_position = MaxSector; 373 new->resync_min = 0; 374 new->resync_max = MaxSector; 375 new->level = LEVEL_NONE; 376 377 goto retry; 378 } 379 380 static inline int mddev_lock(mddev_t * mddev) 381 { 382 return mutex_lock_interruptible(&mddev->reconfig_mutex); 383 } 384 385 static inline int mddev_is_locked(mddev_t *mddev) 386 { 387 return mutex_is_locked(&mddev->reconfig_mutex); 388 } 389 390 static inline int mddev_trylock(mddev_t * mddev) 391 { 392 return mutex_trylock(&mddev->reconfig_mutex); 393 } 394 395 static inline void mddev_unlock(mddev_t * mddev) 396 { 397 mutex_unlock(&mddev->reconfig_mutex); 398 399 md_wakeup_thread(mddev->thread); 400 } 401 402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) 403 { 404 mdk_rdev_t *rdev; 405 406 list_for_each_entry(rdev, &mddev->disks, same_set) 407 if (rdev->desc_nr == nr) 408 return rdev; 409 410 return NULL; 411 } 412 413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) 414 { 415 mdk_rdev_t *rdev; 416 417 list_for_each_entry(rdev, &mddev->disks, same_set) 418 if (rdev->bdev->bd_dev == dev) 419 return rdev; 420 421 return NULL; 422 } 423 424 static struct mdk_personality *find_pers(int level, char *clevel) 425 { 426 struct mdk_personality *pers; 427 list_for_each_entry(pers, &pers_list, list) { 428 if (level != LEVEL_NONE && pers->level == level) 429 return pers; 430 if (strcmp(pers->name, clevel)==0) 431 return pers; 432 } 433 return NULL; 434 } 435 436 /* return the offset of the super block in 512byte sectors */ 437 static inline sector_t calc_dev_sboffset(struct block_device *bdev) 438 { 439 sector_t num_sectors = bdev->bd_inode->i_size / 512; 440 return MD_NEW_SIZE_SECTORS(num_sectors); 441 } 442 443 static int alloc_disk_sb(mdk_rdev_t * rdev) 444 { 445 if (rdev->sb_page) 446 MD_BUG(); 447 448 rdev->sb_page = alloc_page(GFP_KERNEL); 449 if (!rdev->sb_page) { 450 printk(KERN_ALERT "md: out of memory.\n"); 451 return -ENOMEM; 452 } 453 454 return 0; 455 } 456 457 static void free_disk_sb(mdk_rdev_t * rdev) 458 { 459 if (rdev->sb_page) { 460 put_page(rdev->sb_page); 461 rdev->sb_loaded = 0; 462 rdev->sb_page = NULL; 463 rdev->sb_start = 0; 464 rdev->sectors = 0; 465 } 466 } 467 468 469 static void super_written(struct bio *bio, int error) 470 { 471 mdk_rdev_t *rdev = bio->bi_private; 472 mddev_t *mddev = rdev->mddev; 473 474 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { 475 printk("md: super_written gets error=%d, uptodate=%d\n", 476 error, test_bit(BIO_UPTODATE, &bio->bi_flags)); 477 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); 478 md_error(mddev, rdev); 479 } 480 481 if (atomic_dec_and_test(&mddev->pending_writes)) 482 wake_up(&mddev->sb_wait); 483 bio_put(bio); 484 } 485 486 static void super_written_barrier(struct bio *bio, int error) 487 { 488 struct bio *bio2 = bio->bi_private; 489 mdk_rdev_t *rdev = bio2->bi_private; 490 mddev_t *mddev = rdev->mddev; 491 492 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && 493 error == -EOPNOTSUPP) { 494 unsigned long flags; 495 /* barriers don't appear to be supported :-( */ 496 set_bit(BarriersNotsupp, &rdev->flags); 497 mddev->barriers_work = 0; 498 spin_lock_irqsave(&mddev->write_lock, flags); 499 bio2->bi_next = mddev->biolist; 500 mddev->biolist = bio2; 501 spin_unlock_irqrestore(&mddev->write_lock, flags); 502 wake_up(&mddev->sb_wait); 503 bio_put(bio); 504 } else { 505 bio_put(bio2); 506 bio->bi_private = rdev; 507 super_written(bio, error); 508 } 509 } 510 511 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, 512 sector_t sector, int size, struct page *page) 513 { 514 /* write first size bytes of page to sector of rdev 515 * Increment mddev->pending_writes before returning 516 * and decrement it on completion, waking up sb_wait 517 * if zero is reached. 518 * If an error occurred, call md_error 519 * 520 * As we might need to resubmit the request if BIO_RW_BARRIER 521 * causes ENOTSUPP, we allocate a spare bio... 522 */ 523 struct bio *bio = bio_alloc(GFP_NOIO, 1); 524 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); 525 526 bio->bi_bdev = rdev->bdev; 527 bio->bi_sector = sector; 528 bio_add_page(bio, page, size, 0); 529 bio->bi_private = rdev; 530 bio->bi_end_io = super_written; 531 bio->bi_rw = rw; 532 533 atomic_inc(&mddev->pending_writes); 534 if (!test_bit(BarriersNotsupp, &rdev->flags)) { 535 struct bio *rbio; 536 rw |= (1<<BIO_RW_BARRIER); 537 rbio = bio_clone(bio, GFP_NOIO); 538 rbio->bi_private = bio; 539 rbio->bi_end_io = super_written_barrier; 540 submit_bio(rw, rbio); 541 } else 542 submit_bio(rw, bio); 543 } 544 545 void md_super_wait(mddev_t *mddev) 546 { 547 /* wait for all superblock writes that were scheduled to complete. 548 * if any had to be retried (due to BARRIER problems), retry them 549 */ 550 DEFINE_WAIT(wq); 551 for(;;) { 552 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 553 if (atomic_read(&mddev->pending_writes)==0) 554 break; 555 while (mddev->biolist) { 556 struct bio *bio; 557 spin_lock_irq(&mddev->write_lock); 558 bio = mddev->biolist; 559 mddev->biolist = bio->bi_next ; 560 bio->bi_next = NULL; 561 spin_unlock_irq(&mddev->write_lock); 562 submit_bio(bio->bi_rw, bio); 563 } 564 schedule(); 565 } 566 finish_wait(&mddev->sb_wait, &wq); 567 } 568 569 static void bi_complete(struct bio *bio, int error) 570 { 571 complete((struct completion*)bio->bi_private); 572 } 573 574 int sync_page_io(struct block_device *bdev, sector_t sector, int size, 575 struct page *page, int rw) 576 { 577 struct bio *bio = bio_alloc(GFP_NOIO, 1); 578 struct completion event; 579 int ret; 580 581 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); 582 583 bio->bi_bdev = bdev; 584 bio->bi_sector = sector; 585 bio_add_page(bio, page, size, 0); 586 init_completion(&event); 587 bio->bi_private = &event; 588 bio->bi_end_io = bi_complete; 589 submit_bio(rw, bio); 590 wait_for_completion(&event); 591 592 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 593 bio_put(bio); 594 return ret; 595 } 596 EXPORT_SYMBOL_GPL(sync_page_io); 597 598 static int read_disk_sb(mdk_rdev_t * rdev, int size) 599 { 600 char b[BDEVNAME_SIZE]; 601 if (!rdev->sb_page) { 602 MD_BUG(); 603 return -EINVAL; 604 } 605 if (rdev->sb_loaded) 606 return 0; 607 608 609 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) 610 goto fail; 611 rdev->sb_loaded = 1; 612 return 0; 613 614 fail: 615 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 616 bdevname(rdev->bdev,b)); 617 return -EINVAL; 618 } 619 620 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 621 { 622 return sb1->set_uuid0 == sb2->set_uuid0 && 623 sb1->set_uuid1 == sb2->set_uuid1 && 624 sb1->set_uuid2 == sb2->set_uuid2 && 625 sb1->set_uuid3 == sb2->set_uuid3; 626 } 627 628 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 629 { 630 int ret; 631 mdp_super_t *tmp1, *tmp2; 632 633 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 634 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 635 636 if (!tmp1 || !tmp2) { 637 ret = 0; 638 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); 639 goto abort; 640 } 641 642 *tmp1 = *sb1; 643 *tmp2 = *sb2; 644 645 /* 646 * nr_disks is not constant 647 */ 648 tmp1->nr_disks = 0; 649 tmp2->nr_disks = 0; 650 651 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); 652 abort: 653 kfree(tmp1); 654 kfree(tmp2); 655 return ret; 656 } 657 658 659 static u32 md_csum_fold(u32 csum) 660 { 661 csum = (csum & 0xffff) + (csum >> 16); 662 return (csum & 0xffff) + (csum >> 16); 663 } 664 665 static unsigned int calc_sb_csum(mdp_super_t * sb) 666 { 667 u64 newcsum = 0; 668 u32 *sb32 = (u32*)sb; 669 int i; 670 unsigned int disk_csum, csum; 671 672 disk_csum = sb->sb_csum; 673 sb->sb_csum = 0; 674 675 for (i = 0; i < MD_SB_BYTES/4 ; i++) 676 newcsum += sb32[i]; 677 csum = (newcsum & 0xffffffff) + (newcsum>>32); 678 679 680 #ifdef CONFIG_ALPHA 681 /* This used to use csum_partial, which was wrong for several 682 * reasons including that different results are returned on 683 * different architectures. It isn't critical that we get exactly 684 * the same return value as before (we always csum_fold before 685 * testing, and that removes any differences). However as we 686 * know that csum_partial always returned a 16bit value on 687 * alphas, do a fold to maximise conformity to previous behaviour. 688 */ 689 sb->sb_csum = md_csum_fold(disk_csum); 690 #else 691 sb->sb_csum = disk_csum; 692 #endif 693 return csum; 694 } 695 696 697 /* 698 * Handle superblock details. 699 * We want to be able to handle multiple superblock formats 700 * so we have a common interface to them all, and an array of 701 * different handlers. 702 * We rely on user-space to write the initial superblock, and support 703 * reading and updating of superblocks. 704 * Interface methods are: 705 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) 706 * loads and validates a superblock on dev. 707 * if refdev != NULL, compare superblocks on both devices 708 * Return: 709 * 0 - dev has a superblock that is compatible with refdev 710 * 1 - dev has a superblock that is compatible and newer than refdev 711 * so dev should be used as the refdev in future 712 * -EINVAL superblock incompatible or invalid 713 * -othererror e.g. -EIO 714 * 715 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) 716 * Verify that dev is acceptable into mddev. 717 * The first time, mddev->raid_disks will be 0, and data from 718 * dev should be merged in. Subsequent calls check that dev 719 * is new enough. Return 0 or -EINVAL 720 * 721 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) 722 * Update the superblock for rdev with data in mddev 723 * This does not write to disc. 724 * 725 */ 726 727 struct super_type { 728 char *name; 729 struct module *owner; 730 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, 731 int minor_version); 732 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); 733 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); 734 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, 735 sector_t num_sectors); 736 }; 737 738 /* 739 * Check that the given mddev has no bitmap. 740 * 741 * This function is called from the run method of all personalities that do not 742 * support bitmaps. It prints an error message and returns non-zero if mddev 743 * has a bitmap. Otherwise, it returns 0. 744 * 745 */ 746 int md_check_no_bitmap(mddev_t *mddev) 747 { 748 if (!mddev->bitmap_file && !mddev->bitmap_offset) 749 return 0; 750 printk(KERN_ERR "%s: bitmaps are not supported for %s\n", 751 mdname(mddev), mddev->pers->name); 752 return 1; 753 } 754 EXPORT_SYMBOL(md_check_no_bitmap); 755 756 /* 757 * load_super for 0.90.0 758 */ 759 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 760 { 761 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 762 mdp_super_t *sb; 763 int ret; 764 765 /* 766 * Calculate the position of the superblock (512byte sectors), 767 * it's at the end of the disk. 768 * 769 * It also happens to be a multiple of 4Kb. 770 */ 771 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 772 773 ret = read_disk_sb(rdev, MD_SB_BYTES); 774 if (ret) return ret; 775 776 ret = -EINVAL; 777 778 bdevname(rdev->bdev, b); 779 sb = (mdp_super_t*)page_address(rdev->sb_page); 780 781 if (sb->md_magic != MD_SB_MAGIC) { 782 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 783 b); 784 goto abort; 785 } 786 787 if (sb->major_version != 0 || 788 sb->minor_version < 90 || 789 sb->minor_version > 91) { 790 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 791 sb->major_version, sb->minor_version, 792 b); 793 goto abort; 794 } 795 796 if (sb->raid_disks <= 0) 797 goto abort; 798 799 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { 800 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 801 b); 802 goto abort; 803 } 804 805 rdev->preferred_minor = sb->md_minor; 806 rdev->data_offset = 0; 807 rdev->sb_size = MD_SB_BYTES; 808 809 if (sb->level == LEVEL_MULTIPATH) 810 rdev->desc_nr = -1; 811 else 812 rdev->desc_nr = sb->this_disk.number; 813 814 if (!refdev) { 815 ret = 1; 816 } else { 817 __u64 ev1, ev2; 818 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); 819 if (!uuid_equal(refsb, sb)) { 820 printk(KERN_WARNING "md: %s has different UUID to %s\n", 821 b, bdevname(refdev->bdev,b2)); 822 goto abort; 823 } 824 if (!sb_equal(refsb, sb)) { 825 printk(KERN_WARNING "md: %s has same UUID" 826 " but different superblock to %s\n", 827 b, bdevname(refdev->bdev, b2)); 828 goto abort; 829 } 830 ev1 = md_event(sb); 831 ev2 = md_event(refsb); 832 if (ev1 > ev2) 833 ret = 1; 834 else 835 ret = 0; 836 } 837 rdev->sectors = rdev->sb_start; 838 839 if (rdev->sectors < sb->size * 2 && sb->level > 1) 840 /* "this cannot possibly happen" ... */ 841 ret = -EINVAL; 842 843 abort: 844 return ret; 845 } 846 847 /* 848 * validate_super for 0.90.0 849 */ 850 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) 851 { 852 mdp_disk_t *desc; 853 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); 854 __u64 ev1 = md_event(sb); 855 856 rdev->raid_disk = -1; 857 clear_bit(Faulty, &rdev->flags); 858 clear_bit(In_sync, &rdev->flags); 859 clear_bit(WriteMostly, &rdev->flags); 860 clear_bit(BarriersNotsupp, &rdev->flags); 861 862 if (mddev->raid_disks == 0) { 863 mddev->major_version = 0; 864 mddev->minor_version = sb->minor_version; 865 mddev->patch_version = sb->patch_version; 866 mddev->external = 0; 867 mddev->chunk_sectors = sb->chunk_size >> 9; 868 mddev->ctime = sb->ctime; 869 mddev->utime = sb->utime; 870 mddev->level = sb->level; 871 mddev->clevel[0] = 0; 872 mddev->layout = sb->layout; 873 mddev->raid_disks = sb->raid_disks; 874 mddev->dev_sectors = sb->size * 2; 875 mddev->events = ev1; 876 mddev->bitmap_offset = 0; 877 mddev->default_bitmap_offset = MD_SB_BYTES >> 9; 878 879 if (mddev->minor_version >= 91) { 880 mddev->reshape_position = sb->reshape_position; 881 mddev->delta_disks = sb->delta_disks; 882 mddev->new_level = sb->new_level; 883 mddev->new_layout = sb->new_layout; 884 mddev->new_chunk_sectors = sb->new_chunk >> 9; 885 } else { 886 mddev->reshape_position = MaxSector; 887 mddev->delta_disks = 0; 888 mddev->new_level = mddev->level; 889 mddev->new_layout = mddev->layout; 890 mddev->new_chunk_sectors = mddev->chunk_sectors; 891 } 892 893 if (sb->state & (1<<MD_SB_CLEAN)) 894 mddev->recovery_cp = MaxSector; 895 else { 896 if (sb->events_hi == sb->cp_events_hi && 897 sb->events_lo == sb->cp_events_lo) { 898 mddev->recovery_cp = sb->recovery_cp; 899 } else 900 mddev->recovery_cp = 0; 901 } 902 903 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 904 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 905 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 906 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 907 908 mddev->max_disks = MD_SB_DISKS; 909 910 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 911 mddev->bitmap_file == NULL) 912 mddev->bitmap_offset = mddev->default_bitmap_offset; 913 914 } else if (mddev->pers == NULL) { 915 /* Insist on good event counter while assembling */ 916 ++ev1; 917 if (ev1 < mddev->events) 918 return -EINVAL; 919 } else if (mddev->bitmap) { 920 /* if adding to array with a bitmap, then we can accept an 921 * older device ... but not too old. 922 */ 923 if (ev1 < mddev->bitmap->events_cleared) 924 return 0; 925 } else { 926 if (ev1 < mddev->events) 927 /* just a hot-add of a new device, leave raid_disk at -1 */ 928 return 0; 929 } 930 931 if (mddev->level != LEVEL_MULTIPATH) { 932 desc = sb->disks + rdev->desc_nr; 933 934 if (desc->state & (1<<MD_DISK_FAULTY)) 935 set_bit(Faulty, &rdev->flags); 936 else if (desc->state & (1<<MD_DISK_SYNC) /* && 937 desc->raid_disk < mddev->raid_disks */) { 938 set_bit(In_sync, &rdev->flags); 939 rdev->raid_disk = desc->raid_disk; 940 } 941 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 942 set_bit(WriteMostly, &rdev->flags); 943 } else /* MULTIPATH are always insync */ 944 set_bit(In_sync, &rdev->flags); 945 return 0; 946 } 947 948 /* 949 * sync_super for 0.90.0 950 */ 951 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) 952 { 953 mdp_super_t *sb; 954 mdk_rdev_t *rdev2; 955 int next_spare = mddev->raid_disks; 956 957 958 /* make rdev->sb match mddev data.. 959 * 960 * 1/ zero out disks 961 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 962 * 3/ any empty disks < next_spare become removed 963 * 964 * disks[0] gets initialised to REMOVED because 965 * we cannot be sure from other fields if it has 966 * been initialised or not. 967 */ 968 int i; 969 int active=0, working=0,failed=0,spare=0,nr_disks=0; 970 971 rdev->sb_size = MD_SB_BYTES; 972 973 sb = (mdp_super_t*)page_address(rdev->sb_page); 974 975 memset(sb, 0, sizeof(*sb)); 976 977 sb->md_magic = MD_SB_MAGIC; 978 sb->major_version = mddev->major_version; 979 sb->patch_version = mddev->patch_version; 980 sb->gvalid_words = 0; /* ignored */ 981 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 982 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 983 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 984 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 985 986 sb->ctime = mddev->ctime; 987 sb->level = mddev->level; 988 sb->size = mddev->dev_sectors / 2; 989 sb->raid_disks = mddev->raid_disks; 990 sb->md_minor = mddev->md_minor; 991 sb->not_persistent = 0; 992 sb->utime = mddev->utime; 993 sb->state = 0; 994 sb->events_hi = (mddev->events>>32); 995 sb->events_lo = (u32)mddev->events; 996 997 if (mddev->reshape_position == MaxSector) 998 sb->minor_version = 90; 999 else { 1000 sb->minor_version = 91; 1001 sb->reshape_position = mddev->reshape_position; 1002 sb->new_level = mddev->new_level; 1003 sb->delta_disks = mddev->delta_disks; 1004 sb->new_layout = mddev->new_layout; 1005 sb->new_chunk = mddev->new_chunk_sectors << 9; 1006 } 1007 mddev->minor_version = sb->minor_version; 1008 if (mddev->in_sync) 1009 { 1010 sb->recovery_cp = mddev->recovery_cp; 1011 sb->cp_events_hi = (mddev->events>>32); 1012 sb->cp_events_lo = (u32)mddev->events; 1013 if (mddev->recovery_cp == MaxSector) 1014 sb->state = (1<< MD_SB_CLEAN); 1015 } else 1016 sb->recovery_cp = 0; 1017 1018 sb->layout = mddev->layout; 1019 sb->chunk_size = mddev->chunk_sectors << 9; 1020 1021 if (mddev->bitmap && mddev->bitmap_file == NULL) 1022 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 1023 1024 sb->disks[0].state = (1<<MD_DISK_REMOVED); 1025 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1026 mdp_disk_t *d; 1027 int desc_nr; 1028 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) 1029 && !test_bit(Faulty, &rdev2->flags)) 1030 desc_nr = rdev2->raid_disk; 1031 else 1032 desc_nr = next_spare++; 1033 rdev2->desc_nr = desc_nr; 1034 d = &sb->disks[rdev2->desc_nr]; 1035 nr_disks++; 1036 d->number = rdev2->desc_nr; 1037 d->major = MAJOR(rdev2->bdev->bd_dev); 1038 d->minor = MINOR(rdev2->bdev->bd_dev); 1039 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) 1040 && !test_bit(Faulty, &rdev2->flags)) 1041 d->raid_disk = rdev2->raid_disk; 1042 else 1043 d->raid_disk = rdev2->desc_nr; /* compatibility */ 1044 if (test_bit(Faulty, &rdev2->flags)) 1045 d->state = (1<<MD_DISK_FAULTY); 1046 else if (test_bit(In_sync, &rdev2->flags)) { 1047 d->state = (1<<MD_DISK_ACTIVE); 1048 d->state |= (1<<MD_DISK_SYNC); 1049 active++; 1050 working++; 1051 } else { 1052 d->state = 0; 1053 spare++; 1054 working++; 1055 } 1056 if (test_bit(WriteMostly, &rdev2->flags)) 1057 d->state |= (1<<MD_DISK_WRITEMOSTLY); 1058 } 1059 /* now set the "removed" and "faulty" bits on any missing devices */ 1060 for (i=0 ; i < mddev->raid_disks ; i++) { 1061 mdp_disk_t *d = &sb->disks[i]; 1062 if (d->state == 0 && d->number == 0) { 1063 d->number = i; 1064 d->raid_disk = i; 1065 d->state = (1<<MD_DISK_REMOVED); 1066 d->state |= (1<<MD_DISK_FAULTY); 1067 failed++; 1068 } 1069 } 1070 sb->nr_disks = nr_disks; 1071 sb->active_disks = active; 1072 sb->working_disks = working; 1073 sb->failed_disks = failed; 1074 sb->spare_disks = spare; 1075 1076 sb->this_disk = sb->disks[rdev->desc_nr]; 1077 sb->sb_csum = calc_sb_csum(sb); 1078 } 1079 1080 /* 1081 * rdev_size_change for 0.90.0 1082 */ 1083 static unsigned long long 1084 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1085 { 1086 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1087 return 0; /* component must fit device */ 1088 if (rdev->mddev->bitmap_offset) 1089 return 0; /* can't move bitmap */ 1090 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 1091 if (!num_sectors || num_sectors > rdev->sb_start) 1092 num_sectors = rdev->sb_start; 1093 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1094 rdev->sb_page); 1095 md_super_wait(rdev->mddev); 1096 return num_sectors / 2; /* kB for sysfs */ 1097 } 1098 1099 1100 /* 1101 * version 1 superblock 1102 */ 1103 1104 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) 1105 { 1106 __le32 disk_csum; 1107 u32 csum; 1108 unsigned long long newcsum; 1109 int size = 256 + le32_to_cpu(sb->max_dev)*2; 1110 __le32 *isuper = (__le32*)sb; 1111 int i; 1112 1113 disk_csum = sb->sb_csum; 1114 sb->sb_csum = 0; 1115 newcsum = 0; 1116 for (i=0; size>=4; size -= 4 ) 1117 newcsum += le32_to_cpu(*isuper++); 1118 1119 if (size == 2) 1120 newcsum += le16_to_cpu(*(__le16*) isuper); 1121 1122 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 1123 sb->sb_csum = disk_csum; 1124 return cpu_to_le32(csum); 1125 } 1126 1127 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 1128 { 1129 struct mdp_superblock_1 *sb; 1130 int ret; 1131 sector_t sb_start; 1132 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1133 int bmask; 1134 1135 /* 1136 * Calculate the position of the superblock in 512byte sectors. 1137 * It is always aligned to a 4K boundary and 1138 * depeding on minor_version, it can be: 1139 * 0: At least 8K, but less than 12K, from end of device 1140 * 1: At start of device 1141 * 2: 4K from start of device. 1142 */ 1143 switch(minor_version) { 1144 case 0: 1145 sb_start = rdev->bdev->bd_inode->i_size >> 9; 1146 sb_start -= 8*2; 1147 sb_start &= ~(sector_t)(4*2-1); 1148 break; 1149 case 1: 1150 sb_start = 0; 1151 break; 1152 case 2: 1153 sb_start = 8; 1154 break; 1155 default: 1156 return -EINVAL; 1157 } 1158 rdev->sb_start = sb_start; 1159 1160 /* superblock is rarely larger than 1K, but it can be larger, 1161 * and it is safe to read 4k, so we do that 1162 */ 1163 ret = read_disk_sb(rdev, 4096); 1164 if (ret) return ret; 1165 1166 1167 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1168 1169 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 1170 sb->major_version != cpu_to_le32(1) || 1171 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 1172 le64_to_cpu(sb->super_offset) != rdev->sb_start || 1173 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 1174 return -EINVAL; 1175 1176 if (calc_sb_1_csum(sb) != sb->sb_csum) { 1177 printk("md: invalid superblock checksum on %s\n", 1178 bdevname(rdev->bdev,b)); 1179 return -EINVAL; 1180 } 1181 if (le64_to_cpu(sb->data_size) < 10) { 1182 printk("md: data_size too small on %s\n", 1183 bdevname(rdev->bdev,b)); 1184 return -EINVAL; 1185 } 1186 1187 rdev->preferred_minor = 0xffff; 1188 rdev->data_offset = le64_to_cpu(sb->data_offset); 1189 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); 1190 1191 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 1192 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1193 if (rdev->sb_size & bmask) 1194 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1195 1196 if (minor_version 1197 && rdev->data_offset < sb_start + (rdev->sb_size/512)) 1198 return -EINVAL; 1199 1200 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) 1201 rdev->desc_nr = -1; 1202 else 1203 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1204 1205 if (!refdev) { 1206 ret = 1; 1207 } else { 1208 __u64 ev1, ev2; 1209 struct mdp_superblock_1 *refsb = 1210 (struct mdp_superblock_1*)page_address(refdev->sb_page); 1211 1212 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 1213 sb->level != refsb->level || 1214 sb->layout != refsb->layout || 1215 sb->chunksize != refsb->chunksize) { 1216 printk(KERN_WARNING "md: %s has strangely different" 1217 " superblock to %s\n", 1218 bdevname(rdev->bdev,b), 1219 bdevname(refdev->bdev,b2)); 1220 return -EINVAL; 1221 } 1222 ev1 = le64_to_cpu(sb->events); 1223 ev2 = le64_to_cpu(refsb->events); 1224 1225 if (ev1 > ev2) 1226 ret = 1; 1227 else 1228 ret = 0; 1229 } 1230 if (minor_version) 1231 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - 1232 le64_to_cpu(sb->data_offset); 1233 else 1234 rdev->sectors = rdev->sb_start; 1235 if (rdev->sectors < le64_to_cpu(sb->data_size)) 1236 return -EINVAL; 1237 rdev->sectors = le64_to_cpu(sb->data_size); 1238 if (le64_to_cpu(sb->size) > rdev->sectors) 1239 return -EINVAL; 1240 return ret; 1241 } 1242 1243 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1244 { 1245 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1246 __u64 ev1 = le64_to_cpu(sb->events); 1247 1248 rdev->raid_disk = -1; 1249 clear_bit(Faulty, &rdev->flags); 1250 clear_bit(In_sync, &rdev->flags); 1251 clear_bit(WriteMostly, &rdev->flags); 1252 clear_bit(BarriersNotsupp, &rdev->flags); 1253 1254 if (mddev->raid_disks == 0) { 1255 mddev->major_version = 1; 1256 mddev->patch_version = 0; 1257 mddev->external = 0; 1258 mddev->chunk_sectors = le32_to_cpu(sb->chunksize); 1259 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1260 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1261 mddev->level = le32_to_cpu(sb->level); 1262 mddev->clevel[0] = 0; 1263 mddev->layout = le32_to_cpu(sb->layout); 1264 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1265 mddev->dev_sectors = le64_to_cpu(sb->size); 1266 mddev->events = ev1; 1267 mddev->bitmap_offset = 0; 1268 mddev->default_bitmap_offset = 1024 >> 9; 1269 1270 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1271 memcpy(mddev->uuid, sb->set_uuid, 16); 1272 1273 mddev->max_disks = (4096-256)/2; 1274 1275 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1276 mddev->bitmap_file == NULL ) 1277 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); 1278 1279 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { 1280 mddev->reshape_position = le64_to_cpu(sb->reshape_position); 1281 mddev->delta_disks = le32_to_cpu(sb->delta_disks); 1282 mddev->new_level = le32_to_cpu(sb->new_level); 1283 mddev->new_layout = le32_to_cpu(sb->new_layout); 1284 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); 1285 } else { 1286 mddev->reshape_position = MaxSector; 1287 mddev->delta_disks = 0; 1288 mddev->new_level = mddev->level; 1289 mddev->new_layout = mddev->layout; 1290 mddev->new_chunk_sectors = mddev->chunk_sectors; 1291 } 1292 1293 } else if (mddev->pers == NULL) { 1294 /* Insist of good event counter while assembling */ 1295 ++ev1; 1296 if (ev1 < mddev->events) 1297 return -EINVAL; 1298 } else if (mddev->bitmap) { 1299 /* If adding to array with a bitmap, then we can accept an 1300 * older device, but not too old. 1301 */ 1302 if (ev1 < mddev->bitmap->events_cleared) 1303 return 0; 1304 } else { 1305 if (ev1 < mddev->events) 1306 /* just a hot-add of a new device, leave raid_disk at -1 */ 1307 return 0; 1308 } 1309 if (mddev->level != LEVEL_MULTIPATH) { 1310 int role; 1311 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1312 switch(role) { 1313 case 0xffff: /* spare */ 1314 break; 1315 case 0xfffe: /* faulty */ 1316 set_bit(Faulty, &rdev->flags); 1317 break; 1318 default: 1319 if ((le32_to_cpu(sb->feature_map) & 1320 MD_FEATURE_RECOVERY_OFFSET)) 1321 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); 1322 else 1323 set_bit(In_sync, &rdev->flags); 1324 rdev->raid_disk = role; 1325 break; 1326 } 1327 if (sb->devflags & WriteMostly1) 1328 set_bit(WriteMostly, &rdev->flags); 1329 } else /* MULTIPATH are always insync */ 1330 set_bit(In_sync, &rdev->flags); 1331 1332 return 0; 1333 } 1334 1335 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1336 { 1337 struct mdp_superblock_1 *sb; 1338 mdk_rdev_t *rdev2; 1339 int max_dev, i; 1340 /* make rdev->sb match mddev and rdev data. */ 1341 1342 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1343 1344 sb->feature_map = 0; 1345 sb->pad0 = 0; 1346 sb->recovery_offset = cpu_to_le64(0); 1347 memset(sb->pad1, 0, sizeof(sb->pad1)); 1348 memset(sb->pad2, 0, sizeof(sb->pad2)); 1349 memset(sb->pad3, 0, sizeof(sb->pad3)); 1350 1351 sb->utime = cpu_to_le64((__u64)mddev->utime); 1352 sb->events = cpu_to_le64(mddev->events); 1353 if (mddev->in_sync) 1354 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1355 else 1356 sb->resync_offset = cpu_to_le64(0); 1357 1358 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); 1359 1360 sb->raid_disks = cpu_to_le32(mddev->raid_disks); 1361 sb->size = cpu_to_le64(mddev->dev_sectors); 1362 sb->chunksize = cpu_to_le32(mddev->chunk_sectors); 1363 sb->level = cpu_to_le32(mddev->level); 1364 sb->layout = cpu_to_le32(mddev->layout); 1365 1366 if (mddev->bitmap && mddev->bitmap_file == NULL) { 1367 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); 1368 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); 1369 } 1370 1371 if (rdev->raid_disk >= 0 && 1372 !test_bit(In_sync, &rdev->flags)) { 1373 if (mddev->curr_resync_completed > rdev->recovery_offset) 1374 rdev->recovery_offset = mddev->curr_resync_completed; 1375 if (rdev->recovery_offset > 0) { 1376 sb->feature_map |= 1377 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); 1378 sb->recovery_offset = 1379 cpu_to_le64(rdev->recovery_offset); 1380 } 1381 } 1382 1383 if (mddev->reshape_position != MaxSector) { 1384 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); 1385 sb->reshape_position = cpu_to_le64(mddev->reshape_position); 1386 sb->new_layout = cpu_to_le32(mddev->new_layout); 1387 sb->delta_disks = cpu_to_le32(mddev->delta_disks); 1388 sb->new_level = cpu_to_le32(mddev->new_level); 1389 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); 1390 } 1391 1392 max_dev = 0; 1393 list_for_each_entry(rdev2, &mddev->disks, same_set) 1394 if (rdev2->desc_nr+1 > max_dev) 1395 max_dev = rdev2->desc_nr+1; 1396 1397 if (max_dev > le32_to_cpu(sb->max_dev)) 1398 sb->max_dev = cpu_to_le32(max_dev); 1399 for (i=0; i<max_dev;i++) 1400 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1401 1402 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1403 i = rdev2->desc_nr; 1404 if (test_bit(Faulty, &rdev2->flags)) 1405 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1406 else if (test_bit(In_sync, &rdev2->flags)) 1407 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1408 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) 1409 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1410 else 1411 sb->dev_roles[i] = cpu_to_le16(0xffff); 1412 } 1413 1414 sb->sb_csum = calc_sb_1_csum(sb); 1415 } 1416 1417 static unsigned long long 1418 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1419 { 1420 struct mdp_superblock_1 *sb; 1421 sector_t max_sectors; 1422 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1423 return 0; /* component must fit device */ 1424 if (rdev->sb_start < rdev->data_offset) { 1425 /* minor versions 1 and 2; superblock before data */ 1426 max_sectors = rdev->bdev->bd_inode->i_size >> 9; 1427 max_sectors -= rdev->data_offset; 1428 if (!num_sectors || num_sectors > max_sectors) 1429 num_sectors = max_sectors; 1430 } else if (rdev->mddev->bitmap_offset) { 1431 /* minor version 0 with bitmap we can't move */ 1432 return 0; 1433 } else { 1434 /* minor version 0; superblock after data */ 1435 sector_t sb_start; 1436 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; 1437 sb_start &= ~(sector_t)(4*2 - 1); 1438 max_sectors = rdev->sectors + sb_start - rdev->sb_start; 1439 if (!num_sectors || num_sectors > max_sectors) 1440 num_sectors = max_sectors; 1441 rdev->sb_start = sb_start; 1442 } 1443 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); 1444 sb->data_size = cpu_to_le64(num_sectors); 1445 sb->super_offset = rdev->sb_start; 1446 sb->sb_csum = calc_sb_1_csum(sb); 1447 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1448 rdev->sb_page); 1449 md_super_wait(rdev->mddev); 1450 return num_sectors / 2; /* kB for sysfs */ 1451 } 1452 1453 static struct super_type super_types[] = { 1454 [0] = { 1455 .name = "0.90.0", 1456 .owner = THIS_MODULE, 1457 .load_super = super_90_load, 1458 .validate_super = super_90_validate, 1459 .sync_super = super_90_sync, 1460 .rdev_size_change = super_90_rdev_size_change, 1461 }, 1462 [1] = { 1463 .name = "md-1", 1464 .owner = THIS_MODULE, 1465 .load_super = super_1_load, 1466 .validate_super = super_1_validate, 1467 .sync_super = super_1_sync, 1468 .rdev_size_change = super_1_rdev_size_change, 1469 }, 1470 }; 1471 1472 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) 1473 { 1474 mdk_rdev_t *rdev, *rdev2; 1475 1476 rcu_read_lock(); 1477 rdev_for_each_rcu(rdev, mddev1) 1478 rdev_for_each_rcu(rdev2, mddev2) 1479 if (rdev->bdev->bd_contains == 1480 rdev2->bdev->bd_contains) { 1481 rcu_read_unlock(); 1482 return 1; 1483 } 1484 rcu_read_unlock(); 1485 return 0; 1486 } 1487 1488 static LIST_HEAD(pending_raid_disks); 1489 1490 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev) 1491 { 1492 struct mdk_personality *pers = mddev->pers; 1493 struct gendisk *disk = mddev->gendisk; 1494 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); 1495 struct blk_integrity *bi_mddev = blk_get_integrity(disk); 1496 1497 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */ 1498 if (pers && pers->level >= 4 && pers->level <= 6) 1499 return; 1500 1501 /* If rdev is integrity capable, register profile for mddev */ 1502 if (!bi_mddev && bi_rdev) { 1503 if (blk_integrity_register(disk, bi_rdev)) 1504 printk(KERN_ERR "%s: %s Could not register integrity!\n", 1505 __func__, disk->disk_name); 1506 else 1507 printk(KERN_NOTICE "Enabling data integrity on %s\n", 1508 disk->disk_name); 1509 return; 1510 } 1511 1512 /* Check that mddev and rdev have matching profiles */ 1513 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) { 1514 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__, 1515 disk->disk_name, rdev->bdev->bd_disk->disk_name); 1516 printk(KERN_NOTICE "Disabling data integrity on %s\n", 1517 disk->disk_name); 1518 blk_integrity_unregister(disk); 1519 } 1520 } 1521 1522 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) 1523 { 1524 char b[BDEVNAME_SIZE]; 1525 struct kobject *ko; 1526 char *s; 1527 int err; 1528 1529 if (rdev->mddev) { 1530 MD_BUG(); 1531 return -EINVAL; 1532 } 1533 1534 /* prevent duplicates */ 1535 if (find_rdev(mddev, rdev->bdev->bd_dev)) 1536 return -EEXIST; 1537 1538 /* make sure rdev->sectors exceeds mddev->dev_sectors */ 1539 if (rdev->sectors && (mddev->dev_sectors == 0 || 1540 rdev->sectors < mddev->dev_sectors)) { 1541 if (mddev->pers) { 1542 /* Cannot change size, so fail 1543 * If mddev->level <= 0, then we don't care 1544 * about aligning sizes (e.g. linear) 1545 */ 1546 if (mddev->level > 0) 1547 return -ENOSPC; 1548 } else 1549 mddev->dev_sectors = rdev->sectors; 1550 } 1551 1552 /* Verify rdev->desc_nr is unique. 1553 * If it is -1, assign a free number, else 1554 * check number is not in use 1555 */ 1556 if (rdev->desc_nr < 0) { 1557 int choice = 0; 1558 if (mddev->pers) choice = mddev->raid_disks; 1559 while (find_rdev_nr(mddev, choice)) 1560 choice++; 1561 rdev->desc_nr = choice; 1562 } else { 1563 if (find_rdev_nr(mddev, rdev->desc_nr)) 1564 return -EBUSY; 1565 } 1566 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { 1567 printk(KERN_WARNING "md: %s: array is limited to %d devices\n", 1568 mdname(mddev), mddev->max_disks); 1569 return -EBUSY; 1570 } 1571 bdevname(rdev->bdev,b); 1572 while ( (s=strchr(b, '/')) != NULL) 1573 *s = '!'; 1574 1575 rdev->mddev = mddev; 1576 printk(KERN_INFO "md: bind<%s>\n", b); 1577 1578 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) 1579 goto fail; 1580 1581 ko = &part_to_dev(rdev->bdev->bd_part)->kobj; 1582 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { 1583 kobject_del(&rdev->kobj); 1584 goto fail; 1585 } 1586 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state"); 1587 1588 list_add_rcu(&rdev->same_set, &mddev->disks); 1589 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); 1590 1591 /* May as well allow recovery to be retried once */ 1592 mddev->recovery_disabled = 0; 1593 1594 md_integrity_check(rdev, mddev); 1595 return 0; 1596 1597 fail: 1598 printk(KERN_WARNING "md: failed to register dev-%s for %s\n", 1599 b, mdname(mddev)); 1600 return err; 1601 } 1602 1603 static void md_delayed_delete(struct work_struct *ws) 1604 { 1605 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); 1606 kobject_del(&rdev->kobj); 1607 kobject_put(&rdev->kobj); 1608 } 1609 1610 static void unbind_rdev_from_array(mdk_rdev_t * rdev) 1611 { 1612 char b[BDEVNAME_SIZE]; 1613 if (!rdev->mddev) { 1614 MD_BUG(); 1615 return; 1616 } 1617 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); 1618 list_del_rcu(&rdev->same_set); 1619 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 1620 rdev->mddev = NULL; 1621 sysfs_remove_link(&rdev->kobj, "block"); 1622 sysfs_put(rdev->sysfs_state); 1623 rdev->sysfs_state = NULL; 1624 /* We need to delay this, otherwise we can deadlock when 1625 * writing to 'remove' to "dev/state". We also need 1626 * to delay it due to rcu usage. 1627 */ 1628 synchronize_rcu(); 1629 INIT_WORK(&rdev->del_work, md_delayed_delete); 1630 kobject_get(&rdev->kobj); 1631 schedule_work(&rdev->del_work); 1632 } 1633 1634 /* 1635 * prevent the device from being mounted, repartitioned or 1636 * otherwise reused by a RAID array (or any other kernel 1637 * subsystem), by bd_claiming the device. 1638 */ 1639 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) 1640 { 1641 int err = 0; 1642 struct block_device *bdev; 1643 char b[BDEVNAME_SIZE]; 1644 1645 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); 1646 if (IS_ERR(bdev)) { 1647 printk(KERN_ERR "md: could not open %s.\n", 1648 __bdevname(dev, b)); 1649 return PTR_ERR(bdev); 1650 } 1651 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); 1652 if (err) { 1653 printk(KERN_ERR "md: could not bd_claim %s.\n", 1654 bdevname(bdev, b)); 1655 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1656 return err; 1657 } 1658 if (!shared) 1659 set_bit(AllReserved, &rdev->flags); 1660 rdev->bdev = bdev; 1661 return err; 1662 } 1663 1664 static void unlock_rdev(mdk_rdev_t *rdev) 1665 { 1666 struct block_device *bdev = rdev->bdev; 1667 rdev->bdev = NULL; 1668 if (!bdev) 1669 MD_BUG(); 1670 bd_release(bdev); 1671 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1672 } 1673 1674 void md_autodetect_dev(dev_t dev); 1675 1676 static void export_rdev(mdk_rdev_t * rdev) 1677 { 1678 char b[BDEVNAME_SIZE]; 1679 printk(KERN_INFO "md: export_rdev(%s)\n", 1680 bdevname(rdev->bdev,b)); 1681 if (rdev->mddev) 1682 MD_BUG(); 1683 free_disk_sb(rdev); 1684 #ifndef MODULE 1685 if (test_bit(AutoDetected, &rdev->flags)) 1686 md_autodetect_dev(rdev->bdev->bd_dev); 1687 #endif 1688 unlock_rdev(rdev); 1689 kobject_put(&rdev->kobj); 1690 } 1691 1692 static void kick_rdev_from_array(mdk_rdev_t * rdev) 1693 { 1694 unbind_rdev_from_array(rdev); 1695 export_rdev(rdev); 1696 } 1697 1698 static void export_array(mddev_t *mddev) 1699 { 1700 mdk_rdev_t *rdev, *tmp; 1701 1702 rdev_for_each(rdev, tmp, mddev) { 1703 if (!rdev->mddev) { 1704 MD_BUG(); 1705 continue; 1706 } 1707 kick_rdev_from_array(rdev); 1708 } 1709 if (!list_empty(&mddev->disks)) 1710 MD_BUG(); 1711 mddev->raid_disks = 0; 1712 mddev->major_version = 0; 1713 } 1714 1715 static void print_desc(mdp_disk_t *desc) 1716 { 1717 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 1718 desc->major,desc->minor,desc->raid_disk,desc->state); 1719 } 1720 1721 static void print_sb_90(mdp_super_t *sb) 1722 { 1723 int i; 1724 1725 printk(KERN_INFO 1726 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 1727 sb->major_version, sb->minor_version, sb->patch_version, 1728 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 1729 sb->ctime); 1730 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 1731 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 1732 sb->md_minor, sb->layout, sb->chunk_size); 1733 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 1734 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 1735 sb->utime, sb->state, sb->active_disks, sb->working_disks, 1736 sb->failed_disks, sb->spare_disks, 1737 sb->sb_csum, (unsigned long)sb->events_lo); 1738 1739 printk(KERN_INFO); 1740 for (i = 0; i < MD_SB_DISKS; i++) { 1741 mdp_disk_t *desc; 1742 1743 desc = sb->disks + i; 1744 if (desc->number || desc->major || desc->minor || 1745 desc->raid_disk || (desc->state && (desc->state != 4))) { 1746 printk(" D %2d: ", i); 1747 print_desc(desc); 1748 } 1749 } 1750 printk(KERN_INFO "md: THIS: "); 1751 print_desc(&sb->this_disk); 1752 } 1753 1754 static void print_sb_1(struct mdp_superblock_1 *sb) 1755 { 1756 __u8 *uuid; 1757 1758 uuid = sb->set_uuid; 1759 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x" 1760 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n" 1761 KERN_INFO "md: Name: \"%s\" CT:%llu\n", 1762 le32_to_cpu(sb->major_version), 1763 le32_to_cpu(sb->feature_map), 1764 uuid[0], uuid[1], uuid[2], uuid[3], 1765 uuid[4], uuid[5], uuid[6], uuid[7], 1766 uuid[8], uuid[9], uuid[10], uuid[11], 1767 uuid[12], uuid[13], uuid[14], uuid[15], 1768 sb->set_name, 1769 (unsigned long long)le64_to_cpu(sb->ctime) 1770 & MD_SUPERBLOCK_1_TIME_SEC_MASK); 1771 1772 uuid = sb->device_uuid; 1773 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" 1774 " RO:%llu\n" 1775 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x" 1776 ":%02x%02x%02x%02x%02x%02x\n" 1777 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" 1778 KERN_INFO "md: (MaxDev:%u) \n", 1779 le32_to_cpu(sb->level), 1780 (unsigned long long)le64_to_cpu(sb->size), 1781 le32_to_cpu(sb->raid_disks), 1782 le32_to_cpu(sb->layout), 1783 le32_to_cpu(sb->chunksize), 1784 (unsigned long long)le64_to_cpu(sb->data_offset), 1785 (unsigned long long)le64_to_cpu(sb->data_size), 1786 (unsigned long long)le64_to_cpu(sb->super_offset), 1787 (unsigned long long)le64_to_cpu(sb->recovery_offset), 1788 le32_to_cpu(sb->dev_number), 1789 uuid[0], uuid[1], uuid[2], uuid[3], 1790 uuid[4], uuid[5], uuid[6], uuid[7], 1791 uuid[8], uuid[9], uuid[10], uuid[11], 1792 uuid[12], uuid[13], uuid[14], uuid[15], 1793 sb->devflags, 1794 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, 1795 (unsigned long long)le64_to_cpu(sb->events), 1796 (unsigned long long)le64_to_cpu(sb->resync_offset), 1797 le32_to_cpu(sb->sb_csum), 1798 le32_to_cpu(sb->max_dev) 1799 ); 1800 } 1801 1802 static void print_rdev(mdk_rdev_t *rdev, int major_version) 1803 { 1804 char b[BDEVNAME_SIZE]; 1805 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", 1806 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, 1807 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 1808 rdev->desc_nr); 1809 if (rdev->sb_loaded) { 1810 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); 1811 switch (major_version) { 1812 case 0: 1813 print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); 1814 break; 1815 case 1: 1816 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); 1817 break; 1818 } 1819 } else 1820 printk(KERN_INFO "md: no rdev superblock!\n"); 1821 } 1822 1823 static void md_print_devices(void) 1824 { 1825 struct list_head *tmp; 1826 mdk_rdev_t *rdev; 1827 mddev_t *mddev; 1828 char b[BDEVNAME_SIZE]; 1829 1830 printk("\n"); 1831 printk("md: **********************************\n"); 1832 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 1833 printk("md: **********************************\n"); 1834 for_each_mddev(mddev, tmp) { 1835 1836 if (mddev->bitmap) 1837 bitmap_print_sb(mddev->bitmap); 1838 else 1839 printk("%s: ", mdname(mddev)); 1840 list_for_each_entry(rdev, &mddev->disks, same_set) 1841 printk("<%s>", bdevname(rdev->bdev,b)); 1842 printk("\n"); 1843 1844 list_for_each_entry(rdev, &mddev->disks, same_set) 1845 print_rdev(rdev, mddev->major_version); 1846 } 1847 printk("md: **********************************\n"); 1848 printk("\n"); 1849 } 1850 1851 1852 static void sync_sbs(mddev_t * mddev, int nospares) 1853 { 1854 /* Update each superblock (in-memory image), but 1855 * if we are allowed to, skip spares which already 1856 * have the right event counter, or have one earlier 1857 * (which would mean they aren't being marked as dirty 1858 * with the rest of the array) 1859 */ 1860 mdk_rdev_t *rdev; 1861 1862 list_for_each_entry(rdev, &mddev->disks, same_set) { 1863 if (rdev->sb_events == mddev->events || 1864 (nospares && 1865 rdev->raid_disk < 0 && 1866 (rdev->sb_events&1)==0 && 1867 rdev->sb_events+1 == mddev->events)) { 1868 /* Don't update this superblock */ 1869 rdev->sb_loaded = 2; 1870 } else { 1871 super_types[mddev->major_version]. 1872 sync_super(mddev, rdev); 1873 rdev->sb_loaded = 1; 1874 } 1875 } 1876 } 1877 1878 static void md_update_sb(mddev_t * mddev, int force_change) 1879 { 1880 mdk_rdev_t *rdev; 1881 int sync_req; 1882 int nospares = 0; 1883 1884 mddev->utime = get_seconds(); 1885 if (mddev->external) 1886 return; 1887 repeat: 1888 spin_lock_irq(&mddev->write_lock); 1889 1890 set_bit(MD_CHANGE_PENDING, &mddev->flags); 1891 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) 1892 force_change = 1; 1893 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) 1894 /* just a clean<-> dirty transition, possibly leave spares alone, 1895 * though if events isn't the right even/odd, we will have to do 1896 * spares after all 1897 */ 1898 nospares = 1; 1899 if (force_change) 1900 nospares = 0; 1901 if (mddev->degraded) 1902 /* If the array is degraded, then skipping spares is both 1903 * dangerous and fairly pointless. 1904 * Dangerous because a device that was removed from the array 1905 * might have a event_count that still looks up-to-date, 1906 * so it can be re-added without a resync. 1907 * Pointless because if there are any spares to skip, 1908 * then a recovery will happen and soon that array won't 1909 * be degraded any more and the spare can go back to sleep then. 1910 */ 1911 nospares = 0; 1912 1913 sync_req = mddev->in_sync; 1914 1915 /* If this is just a dirty<->clean transition, and the array is clean 1916 * and 'events' is odd, we can roll back to the previous clean state */ 1917 if (nospares 1918 && (mddev->in_sync && mddev->recovery_cp == MaxSector) 1919 && (mddev->events & 1) 1920 && mddev->events != 1) 1921 mddev->events--; 1922 else { 1923 /* otherwise we have to go forward and ... */ 1924 mddev->events ++; 1925 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ 1926 /* .. if the array isn't clean, insist on an odd 'events' */ 1927 if ((mddev->events&1)==0) { 1928 mddev->events++; 1929 nospares = 0; 1930 } 1931 } else { 1932 /* otherwise insist on an even 'events' (for clean states) */ 1933 if ((mddev->events&1)) { 1934 mddev->events++; 1935 nospares = 0; 1936 } 1937 } 1938 } 1939 1940 if (!mddev->events) { 1941 /* 1942 * oops, this 64-bit counter should never wrap. 1943 * Either we are in around ~1 trillion A.C., assuming 1944 * 1 reboot per second, or we have a bug: 1945 */ 1946 MD_BUG(); 1947 mddev->events --; 1948 } 1949 1950 /* 1951 * do not write anything to disk if using 1952 * nonpersistent superblocks 1953 */ 1954 if (!mddev->persistent) { 1955 if (!mddev->external) 1956 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 1957 1958 spin_unlock_irq(&mddev->write_lock); 1959 wake_up(&mddev->sb_wait); 1960 return; 1961 } 1962 sync_sbs(mddev, nospares); 1963 spin_unlock_irq(&mddev->write_lock); 1964 1965 dprintk(KERN_INFO 1966 "md: updating %s RAID superblock on device (in sync %d)\n", 1967 mdname(mddev),mddev->in_sync); 1968 1969 bitmap_update_sb(mddev->bitmap); 1970 list_for_each_entry(rdev, &mddev->disks, same_set) { 1971 char b[BDEVNAME_SIZE]; 1972 dprintk(KERN_INFO "md: "); 1973 if (rdev->sb_loaded != 1) 1974 continue; /* no noise on spare devices */ 1975 if (test_bit(Faulty, &rdev->flags)) 1976 dprintk("(skipping faulty "); 1977 1978 dprintk("%s ", bdevname(rdev->bdev,b)); 1979 if (!test_bit(Faulty, &rdev->flags)) { 1980 md_super_write(mddev,rdev, 1981 rdev->sb_start, rdev->sb_size, 1982 rdev->sb_page); 1983 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", 1984 bdevname(rdev->bdev,b), 1985 (unsigned long long)rdev->sb_start); 1986 rdev->sb_events = mddev->events; 1987 1988 } else 1989 dprintk(")\n"); 1990 if (mddev->level == LEVEL_MULTIPATH) 1991 /* only need to write one superblock... */ 1992 break; 1993 } 1994 md_super_wait(mddev); 1995 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ 1996 1997 spin_lock_irq(&mddev->write_lock); 1998 if (mddev->in_sync != sync_req || 1999 test_bit(MD_CHANGE_DEVS, &mddev->flags)) { 2000 /* have to write it out again */ 2001 spin_unlock_irq(&mddev->write_lock); 2002 goto repeat; 2003 } 2004 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2005 spin_unlock_irq(&mddev->write_lock); 2006 wake_up(&mddev->sb_wait); 2007 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 2008 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 2009 2010 } 2011 2012 /* words written to sysfs files may, or may not, be \n terminated. 2013 * We want to accept with case. For this we use cmd_match. 2014 */ 2015 static int cmd_match(const char *cmd, const char *str) 2016 { 2017 /* See if cmd, written into a sysfs file, matches 2018 * str. They must either be the same, or cmd can 2019 * have a trailing newline 2020 */ 2021 while (*cmd && *str && *cmd == *str) { 2022 cmd++; 2023 str++; 2024 } 2025 if (*cmd == '\n') 2026 cmd++; 2027 if (*str || *cmd) 2028 return 0; 2029 return 1; 2030 } 2031 2032 struct rdev_sysfs_entry { 2033 struct attribute attr; 2034 ssize_t (*show)(mdk_rdev_t *, char *); 2035 ssize_t (*store)(mdk_rdev_t *, const char *, size_t); 2036 }; 2037 2038 static ssize_t 2039 state_show(mdk_rdev_t *rdev, char *page) 2040 { 2041 char *sep = ""; 2042 size_t len = 0; 2043 2044 if (test_bit(Faulty, &rdev->flags)) { 2045 len+= sprintf(page+len, "%sfaulty",sep); 2046 sep = ","; 2047 } 2048 if (test_bit(In_sync, &rdev->flags)) { 2049 len += sprintf(page+len, "%sin_sync",sep); 2050 sep = ","; 2051 } 2052 if (test_bit(WriteMostly, &rdev->flags)) { 2053 len += sprintf(page+len, "%swrite_mostly",sep); 2054 sep = ","; 2055 } 2056 if (test_bit(Blocked, &rdev->flags)) { 2057 len += sprintf(page+len, "%sblocked", sep); 2058 sep = ","; 2059 } 2060 if (!test_bit(Faulty, &rdev->flags) && 2061 !test_bit(In_sync, &rdev->flags)) { 2062 len += sprintf(page+len, "%sspare", sep); 2063 sep = ","; 2064 } 2065 return len+sprintf(page+len, "\n"); 2066 } 2067 2068 static ssize_t 2069 state_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2070 { 2071 /* can write 2072 * faulty - simulates and error 2073 * remove - disconnects the device 2074 * writemostly - sets write_mostly 2075 * -writemostly - clears write_mostly 2076 * blocked - sets the Blocked flag 2077 * -blocked - clears the Blocked flag 2078 * insync - sets Insync providing device isn't active 2079 */ 2080 int err = -EINVAL; 2081 if (cmd_match(buf, "faulty") && rdev->mddev->pers) { 2082 md_error(rdev->mddev, rdev); 2083 err = 0; 2084 } else if (cmd_match(buf, "remove")) { 2085 if (rdev->raid_disk >= 0) 2086 err = -EBUSY; 2087 else { 2088 mddev_t *mddev = rdev->mddev; 2089 kick_rdev_from_array(rdev); 2090 if (mddev->pers) 2091 md_update_sb(mddev, 1); 2092 md_new_event(mddev); 2093 err = 0; 2094 } 2095 } else if (cmd_match(buf, "writemostly")) { 2096 set_bit(WriteMostly, &rdev->flags); 2097 err = 0; 2098 } else if (cmd_match(buf, "-writemostly")) { 2099 clear_bit(WriteMostly, &rdev->flags); 2100 err = 0; 2101 } else if (cmd_match(buf, "blocked")) { 2102 set_bit(Blocked, &rdev->flags); 2103 err = 0; 2104 } else if (cmd_match(buf, "-blocked")) { 2105 clear_bit(Blocked, &rdev->flags); 2106 wake_up(&rdev->blocked_wait); 2107 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2108 md_wakeup_thread(rdev->mddev->thread); 2109 2110 err = 0; 2111 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { 2112 set_bit(In_sync, &rdev->flags); 2113 err = 0; 2114 } 2115 if (!err && rdev->sysfs_state) 2116 sysfs_notify_dirent(rdev->sysfs_state); 2117 return err ? err : len; 2118 } 2119 static struct rdev_sysfs_entry rdev_state = 2120 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); 2121 2122 static ssize_t 2123 errors_show(mdk_rdev_t *rdev, char *page) 2124 { 2125 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); 2126 } 2127 2128 static ssize_t 2129 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2130 { 2131 char *e; 2132 unsigned long n = simple_strtoul(buf, &e, 10); 2133 if (*buf && (*e == 0 || *e == '\n')) { 2134 atomic_set(&rdev->corrected_errors, n); 2135 return len; 2136 } 2137 return -EINVAL; 2138 } 2139 static struct rdev_sysfs_entry rdev_errors = 2140 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); 2141 2142 static ssize_t 2143 slot_show(mdk_rdev_t *rdev, char *page) 2144 { 2145 if (rdev->raid_disk < 0) 2146 return sprintf(page, "none\n"); 2147 else 2148 return sprintf(page, "%d\n", rdev->raid_disk); 2149 } 2150 2151 static ssize_t 2152 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2153 { 2154 char *e; 2155 int err; 2156 char nm[20]; 2157 int slot = simple_strtoul(buf, &e, 10); 2158 if (strncmp(buf, "none", 4)==0) 2159 slot = -1; 2160 else if (e==buf || (*e && *e!= '\n')) 2161 return -EINVAL; 2162 if (rdev->mddev->pers && slot == -1) { 2163 /* Setting 'slot' on an active array requires also 2164 * updating the 'rd%d' link, and communicating 2165 * with the personality with ->hot_*_disk. 2166 * For now we only support removing 2167 * failed/spare devices. This normally happens automatically, 2168 * but not when the metadata is externally managed. 2169 */ 2170 if (rdev->raid_disk == -1) 2171 return -EEXIST; 2172 /* personality does all needed checks */ 2173 if (rdev->mddev->pers->hot_add_disk == NULL) 2174 return -EINVAL; 2175 err = rdev->mddev->pers-> 2176 hot_remove_disk(rdev->mddev, rdev->raid_disk); 2177 if (err) 2178 return err; 2179 sprintf(nm, "rd%d", rdev->raid_disk); 2180 sysfs_remove_link(&rdev->mddev->kobj, nm); 2181 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2182 md_wakeup_thread(rdev->mddev->thread); 2183 } else if (rdev->mddev->pers) { 2184 mdk_rdev_t *rdev2; 2185 /* Activating a spare .. or possibly reactivating 2186 * if we ever get bitmaps working here. 2187 */ 2188 2189 if (rdev->raid_disk != -1) 2190 return -EBUSY; 2191 2192 if (rdev->mddev->pers->hot_add_disk == NULL) 2193 return -EINVAL; 2194 2195 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) 2196 if (rdev2->raid_disk == slot) 2197 return -EEXIST; 2198 2199 rdev->raid_disk = slot; 2200 if (test_bit(In_sync, &rdev->flags)) 2201 rdev->saved_raid_disk = slot; 2202 else 2203 rdev->saved_raid_disk = -1; 2204 err = rdev->mddev->pers-> 2205 hot_add_disk(rdev->mddev, rdev); 2206 if (err) { 2207 rdev->raid_disk = -1; 2208 return err; 2209 } else 2210 sysfs_notify_dirent(rdev->sysfs_state); 2211 sprintf(nm, "rd%d", rdev->raid_disk); 2212 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) 2213 printk(KERN_WARNING 2214 "md: cannot register " 2215 "%s for %s\n", 2216 nm, mdname(rdev->mddev)); 2217 2218 /* don't wakeup anyone, leave that to userspace. */ 2219 } else { 2220 if (slot >= rdev->mddev->raid_disks) 2221 return -ENOSPC; 2222 rdev->raid_disk = slot; 2223 /* assume it is working */ 2224 clear_bit(Faulty, &rdev->flags); 2225 clear_bit(WriteMostly, &rdev->flags); 2226 set_bit(In_sync, &rdev->flags); 2227 sysfs_notify_dirent(rdev->sysfs_state); 2228 } 2229 return len; 2230 } 2231 2232 2233 static struct rdev_sysfs_entry rdev_slot = 2234 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); 2235 2236 static ssize_t 2237 offset_show(mdk_rdev_t *rdev, char *page) 2238 { 2239 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); 2240 } 2241 2242 static ssize_t 2243 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2244 { 2245 char *e; 2246 unsigned long long offset = simple_strtoull(buf, &e, 10); 2247 if (e==buf || (*e && *e != '\n')) 2248 return -EINVAL; 2249 if (rdev->mddev->pers && rdev->raid_disk >= 0) 2250 return -EBUSY; 2251 if (rdev->sectors && rdev->mddev->external) 2252 /* Must set offset before size, so overlap checks 2253 * can be sane */ 2254 return -EBUSY; 2255 rdev->data_offset = offset; 2256 return len; 2257 } 2258 2259 static struct rdev_sysfs_entry rdev_offset = 2260 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); 2261 2262 static ssize_t 2263 rdev_size_show(mdk_rdev_t *rdev, char *page) 2264 { 2265 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); 2266 } 2267 2268 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) 2269 { 2270 /* check if two start/length pairs overlap */ 2271 if (s1+l1 <= s2) 2272 return 0; 2273 if (s2+l2 <= s1) 2274 return 0; 2275 return 1; 2276 } 2277 2278 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) 2279 { 2280 unsigned long long blocks; 2281 sector_t new; 2282 2283 if (strict_strtoull(buf, 10, &blocks) < 0) 2284 return -EINVAL; 2285 2286 if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) 2287 return -EINVAL; /* sector conversion overflow */ 2288 2289 new = blocks * 2; 2290 if (new != blocks * 2) 2291 return -EINVAL; /* unsigned long long to sector_t overflow */ 2292 2293 *sectors = new; 2294 return 0; 2295 } 2296 2297 static ssize_t 2298 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2299 { 2300 mddev_t *my_mddev = rdev->mddev; 2301 sector_t oldsectors = rdev->sectors; 2302 sector_t sectors; 2303 2304 if (strict_blocks_to_sectors(buf, §ors) < 0) 2305 return -EINVAL; 2306 if (my_mddev->pers && rdev->raid_disk >= 0) { 2307 if (my_mddev->persistent) { 2308 sectors = super_types[my_mddev->major_version]. 2309 rdev_size_change(rdev, sectors); 2310 if (!sectors) 2311 return -EBUSY; 2312 } else if (!sectors) 2313 sectors = (rdev->bdev->bd_inode->i_size >> 9) - 2314 rdev->data_offset; 2315 } 2316 if (sectors < my_mddev->dev_sectors) 2317 return -EINVAL; /* component must fit device */ 2318 2319 rdev->sectors = sectors; 2320 if (sectors > oldsectors && my_mddev->external) { 2321 /* need to check that all other rdevs with the same ->bdev 2322 * do not overlap. We need to unlock the mddev to avoid 2323 * a deadlock. We have already changed rdev->sectors, and if 2324 * we have to change it back, we will have the lock again. 2325 */ 2326 mddev_t *mddev; 2327 int overlap = 0; 2328 struct list_head *tmp; 2329 2330 mddev_unlock(my_mddev); 2331 for_each_mddev(mddev, tmp) { 2332 mdk_rdev_t *rdev2; 2333 2334 mddev_lock(mddev); 2335 list_for_each_entry(rdev2, &mddev->disks, same_set) 2336 if (test_bit(AllReserved, &rdev2->flags) || 2337 (rdev->bdev == rdev2->bdev && 2338 rdev != rdev2 && 2339 overlaps(rdev->data_offset, rdev->sectors, 2340 rdev2->data_offset, 2341 rdev2->sectors))) { 2342 overlap = 1; 2343 break; 2344 } 2345 mddev_unlock(mddev); 2346 if (overlap) { 2347 mddev_put(mddev); 2348 break; 2349 } 2350 } 2351 mddev_lock(my_mddev); 2352 if (overlap) { 2353 /* Someone else could have slipped in a size 2354 * change here, but doing so is just silly. 2355 * We put oldsectors back because we *know* it is 2356 * safe, and trust userspace not to race with 2357 * itself 2358 */ 2359 rdev->sectors = oldsectors; 2360 return -EBUSY; 2361 } 2362 } 2363 return len; 2364 } 2365 2366 static struct rdev_sysfs_entry rdev_size = 2367 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); 2368 2369 static struct attribute *rdev_default_attrs[] = { 2370 &rdev_state.attr, 2371 &rdev_errors.attr, 2372 &rdev_slot.attr, 2373 &rdev_offset.attr, 2374 &rdev_size.attr, 2375 NULL, 2376 }; 2377 static ssize_t 2378 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 2379 { 2380 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2381 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2382 mddev_t *mddev = rdev->mddev; 2383 ssize_t rv; 2384 2385 if (!entry->show) 2386 return -EIO; 2387 2388 rv = mddev ? mddev_lock(mddev) : -EBUSY; 2389 if (!rv) { 2390 if (rdev->mddev == NULL) 2391 rv = -EBUSY; 2392 else 2393 rv = entry->show(rdev, page); 2394 mddev_unlock(mddev); 2395 } 2396 return rv; 2397 } 2398 2399 static ssize_t 2400 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 2401 const char *page, size_t length) 2402 { 2403 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2404 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2405 ssize_t rv; 2406 mddev_t *mddev = rdev->mddev; 2407 2408 if (!entry->store) 2409 return -EIO; 2410 if (!capable(CAP_SYS_ADMIN)) 2411 return -EACCES; 2412 rv = mddev ? mddev_lock(mddev): -EBUSY; 2413 if (!rv) { 2414 if (rdev->mddev == NULL) 2415 rv = -EBUSY; 2416 else 2417 rv = entry->store(rdev, page, length); 2418 mddev_unlock(mddev); 2419 } 2420 return rv; 2421 } 2422 2423 static void rdev_free(struct kobject *ko) 2424 { 2425 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); 2426 kfree(rdev); 2427 } 2428 static struct sysfs_ops rdev_sysfs_ops = { 2429 .show = rdev_attr_show, 2430 .store = rdev_attr_store, 2431 }; 2432 static struct kobj_type rdev_ktype = { 2433 .release = rdev_free, 2434 .sysfs_ops = &rdev_sysfs_ops, 2435 .default_attrs = rdev_default_attrs, 2436 }; 2437 2438 /* 2439 * Import a device. If 'super_format' >= 0, then sanity check the superblock 2440 * 2441 * mark the device faulty if: 2442 * 2443 * - the device is nonexistent (zero size) 2444 * - the device has no valid superblock 2445 * 2446 * a faulty rdev _never_ has rdev->sb set. 2447 */ 2448 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) 2449 { 2450 char b[BDEVNAME_SIZE]; 2451 int err; 2452 mdk_rdev_t *rdev; 2453 sector_t size; 2454 2455 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); 2456 if (!rdev) { 2457 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 2458 return ERR_PTR(-ENOMEM); 2459 } 2460 2461 if ((err = alloc_disk_sb(rdev))) 2462 goto abort_free; 2463 2464 err = lock_rdev(rdev, newdev, super_format == -2); 2465 if (err) 2466 goto abort_free; 2467 2468 kobject_init(&rdev->kobj, &rdev_ktype); 2469 2470 rdev->desc_nr = -1; 2471 rdev->saved_raid_disk = -1; 2472 rdev->raid_disk = -1; 2473 rdev->flags = 0; 2474 rdev->data_offset = 0; 2475 rdev->sb_events = 0; 2476 atomic_set(&rdev->nr_pending, 0); 2477 atomic_set(&rdev->read_errors, 0); 2478 atomic_set(&rdev->corrected_errors, 0); 2479 2480 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; 2481 if (!size) { 2482 printk(KERN_WARNING 2483 "md: %s has zero or unknown size, marking faulty!\n", 2484 bdevname(rdev->bdev,b)); 2485 err = -EINVAL; 2486 goto abort_free; 2487 } 2488 2489 if (super_format >= 0) { 2490 err = super_types[super_format]. 2491 load_super(rdev, NULL, super_minor); 2492 if (err == -EINVAL) { 2493 printk(KERN_WARNING 2494 "md: %s does not have a valid v%d.%d " 2495 "superblock, not importing!\n", 2496 bdevname(rdev->bdev,b), 2497 super_format, super_minor); 2498 goto abort_free; 2499 } 2500 if (err < 0) { 2501 printk(KERN_WARNING 2502 "md: could not read %s's sb, not importing!\n", 2503 bdevname(rdev->bdev,b)); 2504 goto abort_free; 2505 } 2506 } 2507 2508 INIT_LIST_HEAD(&rdev->same_set); 2509 init_waitqueue_head(&rdev->blocked_wait); 2510 2511 return rdev; 2512 2513 abort_free: 2514 if (rdev->sb_page) { 2515 if (rdev->bdev) 2516 unlock_rdev(rdev); 2517 free_disk_sb(rdev); 2518 } 2519 kfree(rdev); 2520 return ERR_PTR(err); 2521 } 2522 2523 /* 2524 * Check a full RAID array for plausibility 2525 */ 2526 2527 2528 static void analyze_sbs(mddev_t * mddev) 2529 { 2530 int i; 2531 mdk_rdev_t *rdev, *freshest, *tmp; 2532 char b[BDEVNAME_SIZE]; 2533 2534 freshest = NULL; 2535 rdev_for_each(rdev, tmp, mddev) 2536 switch (super_types[mddev->major_version]. 2537 load_super(rdev, freshest, mddev->minor_version)) { 2538 case 1: 2539 freshest = rdev; 2540 break; 2541 case 0: 2542 break; 2543 default: 2544 printk( KERN_ERR \ 2545 "md: fatal superblock inconsistency in %s" 2546 " -- removing from array\n", 2547 bdevname(rdev->bdev,b)); 2548 kick_rdev_from_array(rdev); 2549 } 2550 2551 2552 super_types[mddev->major_version]. 2553 validate_super(mddev, freshest); 2554 2555 i = 0; 2556 rdev_for_each(rdev, tmp, mddev) { 2557 if (rdev->desc_nr >= mddev->max_disks || 2558 i > mddev->max_disks) { 2559 printk(KERN_WARNING 2560 "md: %s: %s: only %d devices permitted\n", 2561 mdname(mddev), bdevname(rdev->bdev, b), 2562 mddev->max_disks); 2563 kick_rdev_from_array(rdev); 2564 continue; 2565 } 2566 if (rdev != freshest) 2567 if (super_types[mddev->major_version]. 2568 validate_super(mddev, rdev)) { 2569 printk(KERN_WARNING "md: kicking non-fresh %s" 2570 " from array!\n", 2571 bdevname(rdev->bdev,b)); 2572 kick_rdev_from_array(rdev); 2573 continue; 2574 } 2575 if (mddev->level == LEVEL_MULTIPATH) { 2576 rdev->desc_nr = i++; 2577 rdev->raid_disk = rdev->desc_nr; 2578 set_bit(In_sync, &rdev->flags); 2579 } else if (rdev->raid_disk >= mddev->raid_disks) { 2580 rdev->raid_disk = -1; 2581 clear_bit(In_sync, &rdev->flags); 2582 } 2583 } 2584 } 2585 2586 static void md_safemode_timeout(unsigned long data); 2587 2588 static ssize_t 2589 safe_delay_show(mddev_t *mddev, char *page) 2590 { 2591 int msec = (mddev->safemode_delay*1000)/HZ; 2592 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); 2593 } 2594 static ssize_t 2595 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) 2596 { 2597 int scale=1; 2598 int dot=0; 2599 int i; 2600 unsigned long msec; 2601 char buf[30]; 2602 2603 /* remove a period, and count digits after it */ 2604 if (len >= sizeof(buf)) 2605 return -EINVAL; 2606 strlcpy(buf, cbuf, sizeof(buf)); 2607 for (i=0; i<len; i++) { 2608 if (dot) { 2609 if (isdigit(buf[i])) { 2610 buf[i-1] = buf[i]; 2611 scale *= 10; 2612 } 2613 buf[i] = 0; 2614 } else if (buf[i] == '.') { 2615 dot=1; 2616 buf[i] = 0; 2617 } 2618 } 2619 if (strict_strtoul(buf, 10, &msec) < 0) 2620 return -EINVAL; 2621 msec = (msec * 1000) / scale; 2622 if (msec == 0) 2623 mddev->safemode_delay = 0; 2624 else { 2625 unsigned long old_delay = mddev->safemode_delay; 2626 mddev->safemode_delay = (msec*HZ)/1000; 2627 if (mddev->safemode_delay == 0) 2628 mddev->safemode_delay = 1; 2629 if (mddev->safemode_delay < old_delay) 2630 md_safemode_timeout((unsigned long)mddev); 2631 } 2632 return len; 2633 } 2634 static struct md_sysfs_entry md_safe_delay = 2635 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); 2636 2637 static ssize_t 2638 level_show(mddev_t *mddev, char *page) 2639 { 2640 struct mdk_personality *p = mddev->pers; 2641 if (p) 2642 return sprintf(page, "%s\n", p->name); 2643 else if (mddev->clevel[0]) 2644 return sprintf(page, "%s\n", mddev->clevel); 2645 else if (mddev->level != LEVEL_NONE) 2646 return sprintf(page, "%d\n", mddev->level); 2647 else 2648 return 0; 2649 } 2650 2651 static ssize_t 2652 level_store(mddev_t *mddev, const char *buf, size_t len) 2653 { 2654 char level[16]; 2655 ssize_t rv = len; 2656 struct mdk_personality *pers; 2657 void *priv; 2658 2659 if (mddev->pers == NULL) { 2660 if (len == 0) 2661 return 0; 2662 if (len >= sizeof(mddev->clevel)) 2663 return -ENOSPC; 2664 strncpy(mddev->clevel, buf, len); 2665 if (mddev->clevel[len-1] == '\n') 2666 len--; 2667 mddev->clevel[len] = 0; 2668 mddev->level = LEVEL_NONE; 2669 return rv; 2670 } 2671 2672 /* request to change the personality. Need to ensure: 2673 * - array is not engaged in resync/recovery/reshape 2674 * - old personality can be suspended 2675 * - new personality will access other array. 2676 */ 2677 2678 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 2679 return -EBUSY; 2680 2681 if (!mddev->pers->quiesce) { 2682 printk(KERN_WARNING "md: %s: %s does not support online personality change\n", 2683 mdname(mddev), mddev->pers->name); 2684 return -EINVAL; 2685 } 2686 2687 /* Now find the new personality */ 2688 if (len == 0 || len >= sizeof(level)) 2689 return -EINVAL; 2690 strncpy(level, buf, len); 2691 if (level[len-1] == '\n') 2692 len--; 2693 level[len] = 0; 2694 2695 request_module("md-%s", level); 2696 spin_lock(&pers_lock); 2697 pers = find_pers(LEVEL_NONE, level); 2698 if (!pers || !try_module_get(pers->owner)) { 2699 spin_unlock(&pers_lock); 2700 printk(KERN_WARNING "md: personality %s not loaded\n", level); 2701 return -EINVAL; 2702 } 2703 spin_unlock(&pers_lock); 2704 2705 if (pers == mddev->pers) { 2706 /* Nothing to do! */ 2707 module_put(pers->owner); 2708 return rv; 2709 } 2710 if (!pers->takeover) { 2711 module_put(pers->owner); 2712 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", 2713 mdname(mddev), level); 2714 return -EINVAL; 2715 } 2716 2717 /* ->takeover must set new_* and/or delta_disks 2718 * if it succeeds, and may set them when it fails. 2719 */ 2720 priv = pers->takeover(mddev); 2721 if (IS_ERR(priv)) { 2722 mddev->new_level = mddev->level; 2723 mddev->new_layout = mddev->layout; 2724 mddev->new_chunk_sectors = mddev->chunk_sectors; 2725 mddev->raid_disks -= mddev->delta_disks; 2726 mddev->delta_disks = 0; 2727 module_put(pers->owner); 2728 printk(KERN_WARNING "md: %s: %s would not accept array\n", 2729 mdname(mddev), level); 2730 return PTR_ERR(priv); 2731 } 2732 2733 /* Looks like we have a winner */ 2734 mddev_suspend(mddev); 2735 mddev->pers->stop(mddev); 2736 module_put(mddev->pers->owner); 2737 mddev->pers = pers; 2738 mddev->private = priv; 2739 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 2740 mddev->level = mddev->new_level; 2741 mddev->layout = mddev->new_layout; 2742 mddev->chunk_sectors = mddev->new_chunk_sectors; 2743 mddev->delta_disks = 0; 2744 pers->run(mddev); 2745 mddev_resume(mddev); 2746 set_bit(MD_CHANGE_DEVS, &mddev->flags); 2747 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 2748 md_wakeup_thread(mddev->thread); 2749 return rv; 2750 } 2751 2752 static struct md_sysfs_entry md_level = 2753 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); 2754 2755 2756 static ssize_t 2757 layout_show(mddev_t *mddev, char *page) 2758 { 2759 /* just a number, not meaningful for all levels */ 2760 if (mddev->reshape_position != MaxSector && 2761 mddev->layout != mddev->new_layout) 2762 return sprintf(page, "%d (%d)\n", 2763 mddev->new_layout, mddev->layout); 2764 return sprintf(page, "%d\n", mddev->layout); 2765 } 2766 2767 static ssize_t 2768 layout_store(mddev_t *mddev, const char *buf, size_t len) 2769 { 2770 char *e; 2771 unsigned long n = simple_strtoul(buf, &e, 10); 2772 2773 if (!*buf || (*e && *e != '\n')) 2774 return -EINVAL; 2775 2776 if (mddev->pers) { 2777 int err; 2778 if (mddev->pers->check_reshape == NULL) 2779 return -EBUSY; 2780 mddev->new_layout = n; 2781 err = mddev->pers->check_reshape(mddev); 2782 if (err) { 2783 mddev->new_layout = mddev->layout; 2784 return err; 2785 } 2786 } else { 2787 mddev->new_layout = n; 2788 if (mddev->reshape_position == MaxSector) 2789 mddev->layout = n; 2790 } 2791 return len; 2792 } 2793 static struct md_sysfs_entry md_layout = 2794 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); 2795 2796 2797 static ssize_t 2798 raid_disks_show(mddev_t *mddev, char *page) 2799 { 2800 if (mddev->raid_disks == 0) 2801 return 0; 2802 if (mddev->reshape_position != MaxSector && 2803 mddev->delta_disks != 0) 2804 return sprintf(page, "%d (%d)\n", mddev->raid_disks, 2805 mddev->raid_disks - mddev->delta_disks); 2806 return sprintf(page, "%d\n", mddev->raid_disks); 2807 } 2808 2809 static int update_raid_disks(mddev_t *mddev, int raid_disks); 2810 2811 static ssize_t 2812 raid_disks_store(mddev_t *mddev, const char *buf, size_t len) 2813 { 2814 char *e; 2815 int rv = 0; 2816 unsigned long n = simple_strtoul(buf, &e, 10); 2817 2818 if (!*buf || (*e && *e != '\n')) 2819 return -EINVAL; 2820 2821 if (mddev->pers) 2822 rv = update_raid_disks(mddev, n); 2823 else if (mddev->reshape_position != MaxSector) { 2824 int olddisks = mddev->raid_disks - mddev->delta_disks; 2825 mddev->delta_disks = n - olddisks; 2826 mddev->raid_disks = n; 2827 } else 2828 mddev->raid_disks = n; 2829 return rv ? rv : len; 2830 } 2831 static struct md_sysfs_entry md_raid_disks = 2832 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); 2833 2834 static ssize_t 2835 chunk_size_show(mddev_t *mddev, char *page) 2836 { 2837 if (mddev->reshape_position != MaxSector && 2838 mddev->chunk_sectors != mddev->new_chunk_sectors) 2839 return sprintf(page, "%d (%d)\n", 2840 mddev->new_chunk_sectors << 9, 2841 mddev->chunk_sectors << 9); 2842 return sprintf(page, "%d\n", mddev->chunk_sectors << 9); 2843 } 2844 2845 static ssize_t 2846 chunk_size_store(mddev_t *mddev, const char *buf, size_t len) 2847 { 2848 char *e; 2849 unsigned long n = simple_strtoul(buf, &e, 10); 2850 2851 if (!*buf || (*e && *e != '\n')) 2852 return -EINVAL; 2853 2854 if (mddev->pers) { 2855 int err; 2856 if (mddev->pers->check_reshape == NULL) 2857 return -EBUSY; 2858 mddev->new_chunk_sectors = n >> 9; 2859 err = mddev->pers->check_reshape(mddev); 2860 if (err) { 2861 mddev->new_chunk_sectors = mddev->chunk_sectors; 2862 return err; 2863 } 2864 } else { 2865 mddev->new_chunk_sectors = n >> 9; 2866 if (mddev->reshape_position == MaxSector) 2867 mddev->chunk_sectors = n >> 9; 2868 } 2869 return len; 2870 } 2871 static struct md_sysfs_entry md_chunk_size = 2872 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); 2873 2874 static ssize_t 2875 resync_start_show(mddev_t *mddev, char *page) 2876 { 2877 if (mddev->recovery_cp == MaxSector) 2878 return sprintf(page, "none\n"); 2879 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); 2880 } 2881 2882 static ssize_t 2883 resync_start_store(mddev_t *mddev, const char *buf, size_t len) 2884 { 2885 char *e; 2886 unsigned long long n = simple_strtoull(buf, &e, 10); 2887 2888 if (mddev->pers) 2889 return -EBUSY; 2890 if (!*buf || (*e && *e != '\n')) 2891 return -EINVAL; 2892 2893 mddev->recovery_cp = n; 2894 return len; 2895 } 2896 static struct md_sysfs_entry md_resync_start = 2897 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); 2898 2899 /* 2900 * The array state can be: 2901 * 2902 * clear 2903 * No devices, no size, no level 2904 * Equivalent to STOP_ARRAY ioctl 2905 * inactive 2906 * May have some settings, but array is not active 2907 * all IO results in error 2908 * When written, doesn't tear down array, but just stops it 2909 * suspended (not supported yet) 2910 * All IO requests will block. The array can be reconfigured. 2911 * Writing this, if accepted, will block until array is quiescent 2912 * readonly 2913 * no resync can happen. no superblocks get written. 2914 * write requests fail 2915 * read-auto 2916 * like readonly, but behaves like 'clean' on a write request. 2917 * 2918 * clean - no pending writes, but otherwise active. 2919 * When written to inactive array, starts without resync 2920 * If a write request arrives then 2921 * if metadata is known, mark 'dirty' and switch to 'active'. 2922 * if not known, block and switch to write-pending 2923 * If written to an active array that has pending writes, then fails. 2924 * active 2925 * fully active: IO and resync can be happening. 2926 * When written to inactive array, starts with resync 2927 * 2928 * write-pending 2929 * clean, but writes are blocked waiting for 'active' to be written. 2930 * 2931 * active-idle 2932 * like active, but no writes have been seen for a while (100msec). 2933 * 2934 */ 2935 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, 2936 write_pending, active_idle, bad_word}; 2937 static char *array_states[] = { 2938 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", 2939 "write-pending", "active-idle", NULL }; 2940 2941 static int match_word(const char *word, char **list) 2942 { 2943 int n; 2944 for (n=0; list[n]; n++) 2945 if (cmd_match(word, list[n])) 2946 break; 2947 return n; 2948 } 2949 2950 static ssize_t 2951 array_state_show(mddev_t *mddev, char *page) 2952 { 2953 enum array_state st = inactive; 2954 2955 if (mddev->pers) 2956 switch(mddev->ro) { 2957 case 1: 2958 st = readonly; 2959 break; 2960 case 2: 2961 st = read_auto; 2962 break; 2963 case 0: 2964 if (mddev->in_sync) 2965 st = clean; 2966 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) 2967 st = write_pending; 2968 else if (mddev->safemode) 2969 st = active_idle; 2970 else 2971 st = active; 2972 } 2973 else { 2974 if (list_empty(&mddev->disks) && 2975 mddev->raid_disks == 0 && 2976 mddev->dev_sectors == 0) 2977 st = clear; 2978 else 2979 st = inactive; 2980 } 2981 return sprintf(page, "%s\n", array_states[st]); 2982 } 2983 2984 static int do_md_stop(mddev_t * mddev, int ro, int is_open); 2985 static int do_md_run(mddev_t * mddev); 2986 static int restart_array(mddev_t *mddev); 2987 2988 static ssize_t 2989 array_state_store(mddev_t *mddev, const char *buf, size_t len) 2990 { 2991 int err = -EINVAL; 2992 enum array_state st = match_word(buf, array_states); 2993 switch(st) { 2994 case bad_word: 2995 break; 2996 case clear: 2997 /* stopping an active array */ 2998 if (atomic_read(&mddev->openers) > 0) 2999 return -EBUSY; 3000 err = do_md_stop(mddev, 0, 0); 3001 break; 3002 case inactive: 3003 /* stopping an active array */ 3004 if (mddev->pers) { 3005 if (atomic_read(&mddev->openers) > 0) 3006 return -EBUSY; 3007 err = do_md_stop(mddev, 2, 0); 3008 } else 3009 err = 0; /* already inactive */ 3010 break; 3011 case suspended: 3012 break; /* not supported yet */ 3013 case readonly: 3014 if (mddev->pers) 3015 err = do_md_stop(mddev, 1, 0); 3016 else { 3017 mddev->ro = 1; 3018 set_disk_ro(mddev->gendisk, 1); 3019 err = do_md_run(mddev); 3020 } 3021 break; 3022 case read_auto: 3023 if (mddev->pers) { 3024 if (mddev->ro == 0) 3025 err = do_md_stop(mddev, 1, 0); 3026 else if (mddev->ro == 1) 3027 err = restart_array(mddev); 3028 if (err == 0) { 3029 mddev->ro = 2; 3030 set_disk_ro(mddev->gendisk, 0); 3031 } 3032 } else { 3033 mddev->ro = 2; 3034 err = do_md_run(mddev); 3035 } 3036 break; 3037 case clean: 3038 if (mddev->pers) { 3039 restart_array(mddev); 3040 spin_lock_irq(&mddev->write_lock); 3041 if (atomic_read(&mddev->writes_pending) == 0) { 3042 if (mddev->in_sync == 0) { 3043 mddev->in_sync = 1; 3044 if (mddev->safemode == 1) 3045 mddev->safemode = 0; 3046 if (mddev->persistent) 3047 set_bit(MD_CHANGE_CLEAN, 3048 &mddev->flags); 3049 } 3050 err = 0; 3051 } else 3052 err = -EBUSY; 3053 spin_unlock_irq(&mddev->write_lock); 3054 } else 3055 err = -EINVAL; 3056 break; 3057 case active: 3058 if (mddev->pers) { 3059 restart_array(mddev); 3060 if (mddev->external) 3061 clear_bit(MD_CHANGE_CLEAN, &mddev->flags); 3062 wake_up(&mddev->sb_wait); 3063 err = 0; 3064 } else { 3065 mddev->ro = 0; 3066 set_disk_ro(mddev->gendisk, 0); 3067 err = do_md_run(mddev); 3068 } 3069 break; 3070 case write_pending: 3071 case active_idle: 3072 /* these cannot be set */ 3073 break; 3074 } 3075 if (err) 3076 return err; 3077 else { 3078 sysfs_notify_dirent(mddev->sysfs_state); 3079 return len; 3080 } 3081 } 3082 static struct md_sysfs_entry md_array_state = 3083 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); 3084 3085 static ssize_t 3086 null_show(mddev_t *mddev, char *page) 3087 { 3088 return -EINVAL; 3089 } 3090 3091 static ssize_t 3092 new_dev_store(mddev_t *mddev, const char *buf, size_t len) 3093 { 3094 /* buf must be %d:%d\n? giving major and minor numbers */ 3095 /* The new device is added to the array. 3096 * If the array has a persistent superblock, we read the 3097 * superblock to initialise info and check validity. 3098 * Otherwise, only checking done is that in bind_rdev_to_array, 3099 * which mainly checks size. 3100 */ 3101 char *e; 3102 int major = simple_strtoul(buf, &e, 10); 3103 int minor; 3104 dev_t dev; 3105 mdk_rdev_t *rdev; 3106 int err; 3107 3108 if (!*buf || *e != ':' || !e[1] || e[1] == '\n') 3109 return -EINVAL; 3110 minor = simple_strtoul(e+1, &e, 10); 3111 if (*e && *e != '\n') 3112 return -EINVAL; 3113 dev = MKDEV(major, minor); 3114 if (major != MAJOR(dev) || 3115 minor != MINOR(dev)) 3116 return -EOVERFLOW; 3117 3118 3119 if (mddev->persistent) { 3120 rdev = md_import_device(dev, mddev->major_version, 3121 mddev->minor_version); 3122 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { 3123 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 3124 mdk_rdev_t, same_set); 3125 err = super_types[mddev->major_version] 3126 .load_super(rdev, rdev0, mddev->minor_version); 3127 if (err < 0) 3128 goto out; 3129 } 3130 } else if (mddev->external) 3131 rdev = md_import_device(dev, -2, -1); 3132 else 3133 rdev = md_import_device(dev, -1, -1); 3134 3135 if (IS_ERR(rdev)) 3136 return PTR_ERR(rdev); 3137 err = bind_rdev_to_array(rdev, mddev); 3138 out: 3139 if (err) 3140 export_rdev(rdev); 3141 return err ? err : len; 3142 } 3143 3144 static struct md_sysfs_entry md_new_device = 3145 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); 3146 3147 static ssize_t 3148 bitmap_store(mddev_t *mddev, const char *buf, size_t len) 3149 { 3150 char *end; 3151 unsigned long chunk, end_chunk; 3152 3153 if (!mddev->bitmap) 3154 goto out; 3155 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ 3156 while (*buf) { 3157 chunk = end_chunk = simple_strtoul(buf, &end, 0); 3158 if (buf == end) break; 3159 if (*end == '-') { /* range */ 3160 buf = end + 1; 3161 end_chunk = simple_strtoul(buf, &end, 0); 3162 if (buf == end) break; 3163 } 3164 if (*end && !isspace(*end)) break; 3165 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); 3166 buf = end; 3167 while (isspace(*buf)) buf++; 3168 } 3169 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ 3170 out: 3171 return len; 3172 } 3173 3174 static struct md_sysfs_entry md_bitmap = 3175 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); 3176 3177 static ssize_t 3178 size_show(mddev_t *mddev, char *page) 3179 { 3180 return sprintf(page, "%llu\n", 3181 (unsigned long long)mddev->dev_sectors / 2); 3182 } 3183 3184 static int update_size(mddev_t *mddev, sector_t num_sectors); 3185 3186 static ssize_t 3187 size_store(mddev_t *mddev, const char *buf, size_t len) 3188 { 3189 /* If array is inactive, we can reduce the component size, but 3190 * not increase it (except from 0). 3191 * If array is active, we can try an on-line resize 3192 */ 3193 sector_t sectors; 3194 int err = strict_blocks_to_sectors(buf, §ors); 3195 3196 if (err < 0) 3197 return err; 3198 if (mddev->pers) { 3199 err = update_size(mddev, sectors); 3200 md_update_sb(mddev, 1); 3201 } else { 3202 if (mddev->dev_sectors == 0 || 3203 mddev->dev_sectors > sectors) 3204 mddev->dev_sectors = sectors; 3205 else 3206 err = -ENOSPC; 3207 } 3208 return err ? err : len; 3209 } 3210 3211 static struct md_sysfs_entry md_size = 3212 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); 3213 3214 3215 /* Metdata version. 3216 * This is one of 3217 * 'none' for arrays with no metadata (good luck...) 3218 * 'external' for arrays with externally managed metadata, 3219 * or N.M for internally known formats 3220 */ 3221 static ssize_t 3222 metadata_show(mddev_t *mddev, char *page) 3223 { 3224 if (mddev->persistent) 3225 return sprintf(page, "%d.%d\n", 3226 mddev->major_version, mddev->minor_version); 3227 else if (mddev->external) 3228 return sprintf(page, "external:%s\n", mddev->metadata_type); 3229 else 3230 return sprintf(page, "none\n"); 3231 } 3232 3233 static ssize_t 3234 metadata_store(mddev_t *mddev, const char *buf, size_t len) 3235 { 3236 int major, minor; 3237 char *e; 3238 /* Changing the details of 'external' metadata is 3239 * always permitted. Otherwise there must be 3240 * no devices attached to the array. 3241 */ 3242 if (mddev->external && strncmp(buf, "external:", 9) == 0) 3243 ; 3244 else if (!list_empty(&mddev->disks)) 3245 return -EBUSY; 3246 3247 if (cmd_match(buf, "none")) { 3248 mddev->persistent = 0; 3249 mddev->external = 0; 3250 mddev->major_version = 0; 3251 mddev->minor_version = 90; 3252 return len; 3253 } 3254 if (strncmp(buf, "external:", 9) == 0) { 3255 size_t namelen = len-9; 3256 if (namelen >= sizeof(mddev->metadata_type)) 3257 namelen = sizeof(mddev->metadata_type)-1; 3258 strncpy(mddev->metadata_type, buf+9, namelen); 3259 mddev->metadata_type[namelen] = 0; 3260 if (namelen && mddev->metadata_type[namelen-1] == '\n') 3261 mddev->metadata_type[--namelen] = 0; 3262 mddev->persistent = 0; 3263 mddev->external = 1; 3264 mddev->major_version = 0; 3265 mddev->minor_version = 90; 3266 return len; 3267 } 3268 major = simple_strtoul(buf, &e, 10); 3269 if (e==buf || *e != '.') 3270 return -EINVAL; 3271 buf = e+1; 3272 minor = simple_strtoul(buf, &e, 10); 3273 if (e==buf || (*e && *e != '\n') ) 3274 return -EINVAL; 3275 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) 3276 return -ENOENT; 3277 mddev->major_version = major; 3278 mddev->minor_version = minor; 3279 mddev->persistent = 1; 3280 mddev->external = 0; 3281 return len; 3282 } 3283 3284 static struct md_sysfs_entry md_metadata = 3285 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 3286 3287 static ssize_t 3288 action_show(mddev_t *mddev, char *page) 3289 { 3290 char *type = "idle"; 3291 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 3292 type = "frozen"; 3293 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 3294 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { 3295 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 3296 type = "reshape"; 3297 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 3298 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 3299 type = "resync"; 3300 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 3301 type = "check"; 3302 else 3303 type = "repair"; 3304 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) 3305 type = "recover"; 3306 } 3307 return sprintf(page, "%s\n", type); 3308 } 3309 3310 static ssize_t 3311 action_store(mddev_t *mddev, const char *page, size_t len) 3312 { 3313 if (!mddev->pers || !mddev->pers->sync_request) 3314 return -EINVAL; 3315 3316 if (cmd_match(page, "frozen")) 3317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 3318 else 3319 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 3320 3321 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { 3322 if (mddev->sync_thread) { 3323 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 3324 md_unregister_thread(mddev->sync_thread); 3325 mddev->sync_thread = NULL; 3326 mddev->recovery = 0; 3327 } 3328 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 3329 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 3330 return -EBUSY; 3331 else if (cmd_match(page, "resync")) 3332 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3333 else if (cmd_match(page, "recover")) { 3334 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 3335 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3336 } else if (cmd_match(page, "reshape")) { 3337 int err; 3338 if (mddev->pers->start_reshape == NULL) 3339 return -EINVAL; 3340 err = mddev->pers->start_reshape(mddev); 3341 if (err) 3342 return err; 3343 sysfs_notify(&mddev->kobj, NULL, "degraded"); 3344 } else { 3345 if (cmd_match(page, "check")) 3346 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 3347 else if (!cmd_match(page, "repair")) 3348 return -EINVAL; 3349 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 3350 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 3351 } 3352 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3353 md_wakeup_thread(mddev->thread); 3354 sysfs_notify_dirent(mddev->sysfs_action); 3355 return len; 3356 } 3357 3358 static ssize_t 3359 mismatch_cnt_show(mddev_t *mddev, char *page) 3360 { 3361 return sprintf(page, "%llu\n", 3362 (unsigned long long) mddev->resync_mismatches); 3363 } 3364 3365 static struct md_sysfs_entry md_scan_mode = 3366 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); 3367 3368 3369 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); 3370 3371 static ssize_t 3372 sync_min_show(mddev_t *mddev, char *page) 3373 { 3374 return sprintf(page, "%d (%s)\n", speed_min(mddev), 3375 mddev->sync_speed_min ? "local": "system"); 3376 } 3377 3378 static ssize_t 3379 sync_min_store(mddev_t *mddev, const char *buf, size_t len) 3380 { 3381 int min; 3382 char *e; 3383 if (strncmp(buf, "system", 6)==0) { 3384 mddev->sync_speed_min = 0; 3385 return len; 3386 } 3387 min = simple_strtoul(buf, &e, 10); 3388 if (buf == e || (*e && *e != '\n') || min <= 0) 3389 return -EINVAL; 3390 mddev->sync_speed_min = min; 3391 return len; 3392 } 3393 3394 static struct md_sysfs_entry md_sync_min = 3395 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); 3396 3397 static ssize_t 3398 sync_max_show(mddev_t *mddev, char *page) 3399 { 3400 return sprintf(page, "%d (%s)\n", speed_max(mddev), 3401 mddev->sync_speed_max ? "local": "system"); 3402 } 3403 3404 static ssize_t 3405 sync_max_store(mddev_t *mddev, const char *buf, size_t len) 3406 { 3407 int max; 3408 char *e; 3409 if (strncmp(buf, "system", 6)==0) { 3410 mddev->sync_speed_max = 0; 3411 return len; 3412 } 3413 max = simple_strtoul(buf, &e, 10); 3414 if (buf == e || (*e && *e != '\n') || max <= 0) 3415 return -EINVAL; 3416 mddev->sync_speed_max = max; 3417 return len; 3418 } 3419 3420 static struct md_sysfs_entry md_sync_max = 3421 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); 3422 3423 static ssize_t 3424 degraded_show(mddev_t *mddev, char *page) 3425 { 3426 return sprintf(page, "%d\n", mddev->degraded); 3427 } 3428 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); 3429 3430 static ssize_t 3431 sync_force_parallel_show(mddev_t *mddev, char *page) 3432 { 3433 return sprintf(page, "%d\n", mddev->parallel_resync); 3434 } 3435 3436 static ssize_t 3437 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) 3438 { 3439 long n; 3440 3441 if (strict_strtol(buf, 10, &n)) 3442 return -EINVAL; 3443 3444 if (n != 0 && n != 1) 3445 return -EINVAL; 3446 3447 mddev->parallel_resync = n; 3448 3449 if (mddev->sync_thread) 3450 wake_up(&resync_wait); 3451 3452 return len; 3453 } 3454 3455 /* force parallel resync, even with shared block devices */ 3456 static struct md_sysfs_entry md_sync_force_parallel = 3457 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, 3458 sync_force_parallel_show, sync_force_parallel_store); 3459 3460 static ssize_t 3461 sync_speed_show(mddev_t *mddev, char *page) 3462 { 3463 unsigned long resync, dt, db; 3464 if (mddev->curr_resync == 0) 3465 return sprintf(page, "none\n"); 3466 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); 3467 dt = (jiffies - mddev->resync_mark) / HZ; 3468 if (!dt) dt++; 3469 db = resync - mddev->resync_mark_cnt; 3470 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ 3471 } 3472 3473 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); 3474 3475 static ssize_t 3476 sync_completed_show(mddev_t *mddev, char *page) 3477 { 3478 unsigned long max_sectors, resync; 3479 3480 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3481 return sprintf(page, "none\n"); 3482 3483 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 3484 max_sectors = mddev->resync_max_sectors; 3485 else 3486 max_sectors = mddev->dev_sectors; 3487 3488 resync = mddev->curr_resync_completed; 3489 return sprintf(page, "%lu / %lu\n", resync, max_sectors); 3490 } 3491 3492 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); 3493 3494 static ssize_t 3495 min_sync_show(mddev_t *mddev, char *page) 3496 { 3497 return sprintf(page, "%llu\n", 3498 (unsigned long long)mddev->resync_min); 3499 } 3500 static ssize_t 3501 min_sync_store(mddev_t *mddev, const char *buf, size_t len) 3502 { 3503 unsigned long long min; 3504 if (strict_strtoull(buf, 10, &min)) 3505 return -EINVAL; 3506 if (min > mddev->resync_max) 3507 return -EINVAL; 3508 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3509 return -EBUSY; 3510 3511 /* Must be a multiple of chunk_size */ 3512 if (mddev->chunk_sectors) { 3513 sector_t temp = min; 3514 if (sector_div(temp, mddev->chunk_sectors)) 3515 return -EINVAL; 3516 } 3517 mddev->resync_min = min; 3518 3519 return len; 3520 } 3521 3522 static struct md_sysfs_entry md_min_sync = 3523 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); 3524 3525 static ssize_t 3526 max_sync_show(mddev_t *mddev, char *page) 3527 { 3528 if (mddev->resync_max == MaxSector) 3529 return sprintf(page, "max\n"); 3530 else 3531 return sprintf(page, "%llu\n", 3532 (unsigned long long)mddev->resync_max); 3533 } 3534 static ssize_t 3535 max_sync_store(mddev_t *mddev, const char *buf, size_t len) 3536 { 3537 if (strncmp(buf, "max", 3) == 0) 3538 mddev->resync_max = MaxSector; 3539 else { 3540 unsigned long long max; 3541 if (strict_strtoull(buf, 10, &max)) 3542 return -EINVAL; 3543 if (max < mddev->resync_min) 3544 return -EINVAL; 3545 if (max < mddev->resync_max && 3546 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3547 return -EBUSY; 3548 3549 /* Must be a multiple of chunk_size */ 3550 if (mddev->chunk_sectors) { 3551 sector_t temp = max; 3552 if (sector_div(temp, mddev->chunk_sectors)) 3553 return -EINVAL; 3554 } 3555 mddev->resync_max = max; 3556 } 3557 wake_up(&mddev->recovery_wait); 3558 return len; 3559 } 3560 3561 static struct md_sysfs_entry md_max_sync = 3562 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); 3563 3564 static ssize_t 3565 suspend_lo_show(mddev_t *mddev, char *page) 3566 { 3567 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); 3568 } 3569 3570 static ssize_t 3571 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) 3572 { 3573 char *e; 3574 unsigned long long new = simple_strtoull(buf, &e, 10); 3575 3576 if (mddev->pers->quiesce == NULL) 3577 return -EINVAL; 3578 if (buf == e || (*e && *e != '\n')) 3579 return -EINVAL; 3580 if (new >= mddev->suspend_hi || 3581 (new > mddev->suspend_lo && new < mddev->suspend_hi)) { 3582 mddev->suspend_lo = new; 3583 mddev->pers->quiesce(mddev, 2); 3584 return len; 3585 } else 3586 return -EINVAL; 3587 } 3588 static struct md_sysfs_entry md_suspend_lo = 3589 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); 3590 3591 3592 static ssize_t 3593 suspend_hi_show(mddev_t *mddev, char *page) 3594 { 3595 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); 3596 } 3597 3598 static ssize_t 3599 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) 3600 { 3601 char *e; 3602 unsigned long long new = simple_strtoull(buf, &e, 10); 3603 3604 if (mddev->pers->quiesce == NULL) 3605 return -EINVAL; 3606 if (buf == e || (*e && *e != '\n')) 3607 return -EINVAL; 3608 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || 3609 (new > mddev->suspend_lo && new > mddev->suspend_hi)) { 3610 mddev->suspend_hi = new; 3611 mddev->pers->quiesce(mddev, 1); 3612 mddev->pers->quiesce(mddev, 0); 3613 return len; 3614 } else 3615 return -EINVAL; 3616 } 3617 static struct md_sysfs_entry md_suspend_hi = 3618 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); 3619 3620 static ssize_t 3621 reshape_position_show(mddev_t *mddev, char *page) 3622 { 3623 if (mddev->reshape_position != MaxSector) 3624 return sprintf(page, "%llu\n", 3625 (unsigned long long)mddev->reshape_position); 3626 strcpy(page, "none\n"); 3627 return 5; 3628 } 3629 3630 static ssize_t 3631 reshape_position_store(mddev_t *mddev, const char *buf, size_t len) 3632 { 3633 char *e; 3634 unsigned long long new = simple_strtoull(buf, &e, 10); 3635 if (mddev->pers) 3636 return -EBUSY; 3637 if (buf == e || (*e && *e != '\n')) 3638 return -EINVAL; 3639 mddev->reshape_position = new; 3640 mddev->delta_disks = 0; 3641 mddev->new_level = mddev->level; 3642 mddev->new_layout = mddev->layout; 3643 mddev->new_chunk_sectors = mddev->chunk_sectors; 3644 return len; 3645 } 3646 3647 static struct md_sysfs_entry md_reshape_position = 3648 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, 3649 reshape_position_store); 3650 3651 static ssize_t 3652 array_size_show(mddev_t *mddev, char *page) 3653 { 3654 if (mddev->external_size) 3655 return sprintf(page, "%llu\n", 3656 (unsigned long long)mddev->array_sectors/2); 3657 else 3658 return sprintf(page, "default\n"); 3659 } 3660 3661 static ssize_t 3662 array_size_store(mddev_t *mddev, const char *buf, size_t len) 3663 { 3664 sector_t sectors; 3665 3666 if (strncmp(buf, "default", 7) == 0) { 3667 if (mddev->pers) 3668 sectors = mddev->pers->size(mddev, 0, 0); 3669 else 3670 sectors = mddev->array_sectors; 3671 3672 mddev->external_size = 0; 3673 } else { 3674 if (strict_blocks_to_sectors(buf, §ors) < 0) 3675 return -EINVAL; 3676 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) 3677 return -E2BIG; 3678 3679 mddev->external_size = 1; 3680 } 3681 3682 mddev->array_sectors = sectors; 3683 set_capacity(mddev->gendisk, mddev->array_sectors); 3684 if (mddev->pers) { 3685 struct block_device *bdev = bdget_disk(mddev->gendisk, 0); 3686 3687 if (bdev) { 3688 mutex_lock(&bdev->bd_inode->i_mutex); 3689 i_size_write(bdev->bd_inode, 3690 (loff_t)mddev->array_sectors << 9); 3691 mutex_unlock(&bdev->bd_inode->i_mutex); 3692 bdput(bdev); 3693 } 3694 } 3695 3696 return len; 3697 } 3698 3699 static struct md_sysfs_entry md_array_size = 3700 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, 3701 array_size_store); 3702 3703 static struct attribute *md_default_attrs[] = { 3704 &md_level.attr, 3705 &md_layout.attr, 3706 &md_raid_disks.attr, 3707 &md_chunk_size.attr, 3708 &md_size.attr, 3709 &md_resync_start.attr, 3710 &md_metadata.attr, 3711 &md_new_device.attr, 3712 &md_safe_delay.attr, 3713 &md_array_state.attr, 3714 &md_reshape_position.attr, 3715 &md_array_size.attr, 3716 NULL, 3717 }; 3718 3719 static struct attribute *md_redundancy_attrs[] = { 3720 &md_scan_mode.attr, 3721 &md_mismatches.attr, 3722 &md_sync_min.attr, 3723 &md_sync_max.attr, 3724 &md_sync_speed.attr, 3725 &md_sync_force_parallel.attr, 3726 &md_sync_completed.attr, 3727 &md_min_sync.attr, 3728 &md_max_sync.attr, 3729 &md_suspend_lo.attr, 3730 &md_suspend_hi.attr, 3731 &md_bitmap.attr, 3732 &md_degraded.attr, 3733 NULL, 3734 }; 3735 static struct attribute_group md_redundancy_group = { 3736 .name = NULL, 3737 .attrs = md_redundancy_attrs, 3738 }; 3739 3740 3741 static ssize_t 3742 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 3743 { 3744 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 3745 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 3746 ssize_t rv; 3747 3748 if (!entry->show) 3749 return -EIO; 3750 rv = mddev_lock(mddev); 3751 if (!rv) { 3752 rv = entry->show(mddev, page); 3753 mddev_unlock(mddev); 3754 } 3755 return rv; 3756 } 3757 3758 static ssize_t 3759 md_attr_store(struct kobject *kobj, struct attribute *attr, 3760 const char *page, size_t length) 3761 { 3762 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 3763 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 3764 ssize_t rv; 3765 3766 if (!entry->store) 3767 return -EIO; 3768 if (!capable(CAP_SYS_ADMIN)) 3769 return -EACCES; 3770 rv = mddev_lock(mddev); 3771 if (mddev->hold_active == UNTIL_IOCTL) 3772 mddev->hold_active = 0; 3773 if (!rv) { 3774 rv = entry->store(mddev, page, length); 3775 mddev_unlock(mddev); 3776 } 3777 return rv; 3778 } 3779 3780 static void md_free(struct kobject *ko) 3781 { 3782 mddev_t *mddev = container_of(ko, mddev_t, kobj); 3783 3784 if (mddev->sysfs_state) 3785 sysfs_put(mddev->sysfs_state); 3786 3787 if (mddev->gendisk) { 3788 del_gendisk(mddev->gendisk); 3789 put_disk(mddev->gendisk); 3790 } 3791 if (mddev->queue) 3792 blk_cleanup_queue(mddev->queue); 3793 3794 kfree(mddev); 3795 } 3796 3797 static struct sysfs_ops md_sysfs_ops = { 3798 .show = md_attr_show, 3799 .store = md_attr_store, 3800 }; 3801 static struct kobj_type md_ktype = { 3802 .release = md_free, 3803 .sysfs_ops = &md_sysfs_ops, 3804 .default_attrs = md_default_attrs, 3805 }; 3806 3807 int mdp_major = 0; 3808 3809 static void mddev_delayed_delete(struct work_struct *ws) 3810 { 3811 mddev_t *mddev = container_of(ws, mddev_t, del_work); 3812 3813 if (mddev->private == &md_redundancy_group) { 3814 sysfs_remove_group(&mddev->kobj, &md_redundancy_group); 3815 if (mddev->sysfs_action) 3816 sysfs_put(mddev->sysfs_action); 3817 mddev->sysfs_action = NULL; 3818 mddev->private = NULL; 3819 } 3820 kobject_del(&mddev->kobj); 3821 kobject_put(&mddev->kobj); 3822 } 3823 3824 static int md_alloc(dev_t dev, char *name) 3825 { 3826 static DEFINE_MUTEX(disks_mutex); 3827 mddev_t *mddev = mddev_find(dev); 3828 struct gendisk *disk; 3829 int partitioned; 3830 int shift; 3831 int unit; 3832 int error; 3833 3834 if (!mddev) 3835 return -ENODEV; 3836 3837 partitioned = (MAJOR(mddev->unit) != MD_MAJOR); 3838 shift = partitioned ? MdpMinorShift : 0; 3839 unit = MINOR(mddev->unit) >> shift; 3840 3841 /* wait for any previous instance if this device 3842 * to be completed removed (mddev_delayed_delete). 3843 */ 3844 flush_scheduled_work(); 3845 3846 mutex_lock(&disks_mutex); 3847 if (mddev->gendisk) { 3848 mutex_unlock(&disks_mutex); 3849 mddev_put(mddev); 3850 return -EEXIST; 3851 } 3852 3853 if (name) { 3854 /* Need to ensure that 'name' is not a duplicate. 3855 */ 3856 mddev_t *mddev2; 3857 spin_lock(&all_mddevs_lock); 3858 3859 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) 3860 if (mddev2->gendisk && 3861 strcmp(mddev2->gendisk->disk_name, name) == 0) { 3862 spin_unlock(&all_mddevs_lock); 3863 return -EEXIST; 3864 } 3865 spin_unlock(&all_mddevs_lock); 3866 } 3867 3868 mddev->queue = blk_alloc_queue(GFP_KERNEL); 3869 if (!mddev->queue) { 3870 mutex_unlock(&disks_mutex); 3871 mddev_put(mddev); 3872 return -ENOMEM; 3873 } 3874 mddev->queue->queuedata = mddev; 3875 3876 /* Can be unlocked because the queue is new: no concurrency */ 3877 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); 3878 3879 blk_queue_make_request(mddev->queue, md_make_request); 3880 3881 disk = alloc_disk(1 << shift); 3882 if (!disk) { 3883 mutex_unlock(&disks_mutex); 3884 blk_cleanup_queue(mddev->queue); 3885 mddev->queue = NULL; 3886 mddev_put(mddev); 3887 return -ENOMEM; 3888 } 3889 disk->major = MAJOR(mddev->unit); 3890 disk->first_minor = unit << shift; 3891 if (name) 3892 strcpy(disk->disk_name, name); 3893 else if (partitioned) 3894 sprintf(disk->disk_name, "md_d%d", unit); 3895 else 3896 sprintf(disk->disk_name, "md%d", unit); 3897 disk->fops = &md_fops; 3898 disk->private_data = mddev; 3899 disk->queue = mddev->queue; 3900 /* Allow extended partitions. This makes the 3901 * 'mdp' device redundant, but we can't really 3902 * remove it now. 3903 */ 3904 disk->flags |= GENHD_FL_EXT_DEVT; 3905 add_disk(disk); 3906 mddev->gendisk = disk; 3907 error = kobject_init_and_add(&mddev->kobj, &md_ktype, 3908 &disk_to_dev(disk)->kobj, "%s", "md"); 3909 mutex_unlock(&disks_mutex); 3910 if (error) 3911 printk(KERN_WARNING "md: cannot register %s/md - name in use\n", 3912 disk->disk_name); 3913 else { 3914 kobject_uevent(&mddev->kobj, KOBJ_ADD); 3915 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state"); 3916 } 3917 mddev_put(mddev); 3918 return 0; 3919 } 3920 3921 static struct kobject *md_probe(dev_t dev, int *part, void *data) 3922 { 3923 md_alloc(dev, NULL); 3924 return NULL; 3925 } 3926 3927 static int add_named_array(const char *val, struct kernel_param *kp) 3928 { 3929 /* val must be "md_*" where * is not all digits. 3930 * We allocate an array with a large free minor number, and 3931 * set the name to val. val must not already be an active name. 3932 */ 3933 int len = strlen(val); 3934 char buf[DISK_NAME_LEN]; 3935 3936 while (len && val[len-1] == '\n') 3937 len--; 3938 if (len >= DISK_NAME_LEN) 3939 return -E2BIG; 3940 strlcpy(buf, val, len+1); 3941 if (strncmp(buf, "md_", 3) != 0) 3942 return -EINVAL; 3943 return md_alloc(0, buf); 3944 } 3945 3946 static void md_safemode_timeout(unsigned long data) 3947 { 3948 mddev_t *mddev = (mddev_t *) data; 3949 3950 if (!atomic_read(&mddev->writes_pending)) { 3951 mddev->safemode = 1; 3952 if (mddev->external) 3953 sysfs_notify_dirent(mddev->sysfs_state); 3954 } 3955 md_wakeup_thread(mddev->thread); 3956 } 3957 3958 static int start_dirty_degraded; 3959 3960 static int do_md_run(mddev_t * mddev) 3961 { 3962 int err; 3963 mdk_rdev_t *rdev; 3964 struct gendisk *disk; 3965 struct mdk_personality *pers; 3966 3967 if (list_empty(&mddev->disks)) 3968 /* cannot run an array with no devices.. */ 3969 return -EINVAL; 3970 3971 if (mddev->pers) 3972 return -EBUSY; 3973 3974 /* 3975 * Analyze all RAID superblock(s) 3976 */ 3977 if (!mddev->raid_disks) { 3978 if (!mddev->persistent) 3979 return -EINVAL; 3980 analyze_sbs(mddev); 3981 } 3982 3983 if (mddev->level != LEVEL_NONE) 3984 request_module("md-level-%d", mddev->level); 3985 else if (mddev->clevel[0]) 3986 request_module("md-%s", mddev->clevel); 3987 3988 /* 3989 * Drop all container device buffers, from now on 3990 * the only valid external interface is through the md 3991 * device. 3992 */ 3993 list_for_each_entry(rdev, &mddev->disks, same_set) { 3994 if (test_bit(Faulty, &rdev->flags)) 3995 continue; 3996 sync_blockdev(rdev->bdev); 3997 invalidate_bdev(rdev->bdev); 3998 3999 /* perform some consistency tests on the device. 4000 * We don't want the data to overlap the metadata, 4001 * Internal Bitmap issues have been handled elsewhere. 4002 */ 4003 if (rdev->data_offset < rdev->sb_start) { 4004 if (mddev->dev_sectors && 4005 rdev->data_offset + mddev->dev_sectors 4006 > rdev->sb_start) { 4007 printk("md: %s: data overlaps metadata\n", 4008 mdname(mddev)); 4009 return -EINVAL; 4010 } 4011 } else { 4012 if (rdev->sb_start + rdev->sb_size/512 4013 > rdev->data_offset) { 4014 printk("md: %s: metadata overlaps data\n", 4015 mdname(mddev)); 4016 return -EINVAL; 4017 } 4018 } 4019 sysfs_notify_dirent(rdev->sysfs_state); 4020 } 4021 4022 md_probe(mddev->unit, NULL, NULL); 4023 disk = mddev->gendisk; 4024 if (!disk) 4025 return -ENOMEM; 4026 4027 spin_lock(&pers_lock); 4028 pers = find_pers(mddev->level, mddev->clevel); 4029 if (!pers || !try_module_get(pers->owner)) { 4030 spin_unlock(&pers_lock); 4031 if (mddev->level != LEVEL_NONE) 4032 printk(KERN_WARNING "md: personality for level %d is not loaded!\n", 4033 mddev->level); 4034 else 4035 printk(KERN_WARNING "md: personality for level %s is not loaded!\n", 4036 mddev->clevel); 4037 return -EINVAL; 4038 } 4039 mddev->pers = pers; 4040 spin_unlock(&pers_lock); 4041 if (mddev->level != pers->level) { 4042 mddev->level = pers->level; 4043 mddev->new_level = pers->level; 4044 } 4045 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 4046 4047 if (pers->level >= 4 && pers->level <= 6) 4048 /* Cannot support integrity (yet) */ 4049 blk_integrity_unregister(mddev->gendisk); 4050 4051 if (mddev->reshape_position != MaxSector && 4052 pers->start_reshape == NULL) { 4053 /* This personality cannot handle reshaping... */ 4054 mddev->pers = NULL; 4055 module_put(pers->owner); 4056 return -EINVAL; 4057 } 4058 4059 if (pers->sync_request) { 4060 /* Warn if this is a potentially silly 4061 * configuration. 4062 */ 4063 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 4064 mdk_rdev_t *rdev2; 4065 int warned = 0; 4066 4067 list_for_each_entry(rdev, &mddev->disks, same_set) 4068 list_for_each_entry(rdev2, &mddev->disks, same_set) { 4069 if (rdev < rdev2 && 4070 rdev->bdev->bd_contains == 4071 rdev2->bdev->bd_contains) { 4072 printk(KERN_WARNING 4073 "%s: WARNING: %s appears to be" 4074 " on the same physical disk as" 4075 " %s.\n", 4076 mdname(mddev), 4077 bdevname(rdev->bdev,b), 4078 bdevname(rdev2->bdev,b2)); 4079 warned = 1; 4080 } 4081 } 4082 4083 if (warned) 4084 printk(KERN_WARNING 4085 "True protection against single-disk" 4086 " failure might be compromised.\n"); 4087 } 4088 4089 mddev->recovery = 0; 4090 /* may be over-ridden by personality */ 4091 mddev->resync_max_sectors = mddev->dev_sectors; 4092 4093 mddev->barriers_work = 1; 4094 mddev->ok_start_degraded = start_dirty_degraded; 4095 4096 if (start_readonly) 4097 mddev->ro = 2; /* read-only, but switch on first write */ 4098 4099 err = mddev->pers->run(mddev); 4100 if (err) 4101 printk(KERN_ERR "md: pers->run() failed ...\n"); 4102 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { 4103 WARN_ONCE(!mddev->external_size, "%s: default size too small," 4104 " but 'external_size' not in effect?\n", __func__); 4105 printk(KERN_ERR 4106 "md: invalid array_size %llu > default size %llu\n", 4107 (unsigned long long)mddev->array_sectors / 2, 4108 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); 4109 err = -EINVAL; 4110 mddev->pers->stop(mddev); 4111 } 4112 if (err == 0 && mddev->pers->sync_request) { 4113 err = bitmap_create(mddev); 4114 if (err) { 4115 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 4116 mdname(mddev), err); 4117 mddev->pers->stop(mddev); 4118 } 4119 } 4120 if (err) { 4121 module_put(mddev->pers->owner); 4122 mddev->pers = NULL; 4123 bitmap_destroy(mddev); 4124 return err; 4125 } 4126 if (mddev->pers->sync_request) { 4127 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 4128 printk(KERN_WARNING 4129 "md: cannot register extra attributes for %s\n", 4130 mdname(mddev)); 4131 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); 4132 } else if (mddev->ro == 2) /* auto-readonly not meaningful */ 4133 mddev->ro = 0; 4134 4135 atomic_set(&mddev->writes_pending,0); 4136 mddev->safemode = 0; 4137 mddev->safemode_timer.function = md_safemode_timeout; 4138 mddev->safemode_timer.data = (unsigned long) mddev; 4139 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ 4140 mddev->in_sync = 1; 4141 4142 list_for_each_entry(rdev, &mddev->disks, same_set) 4143 if (rdev->raid_disk >= 0) { 4144 char nm[20]; 4145 sprintf(nm, "rd%d", rdev->raid_disk); 4146 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) 4147 printk("md: cannot register %s for %s\n", 4148 nm, mdname(mddev)); 4149 } 4150 4151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4152 4153 if (mddev->flags) 4154 md_update_sb(mddev, 0); 4155 4156 set_capacity(disk, mddev->array_sectors); 4157 4158 /* If there is a partially-recovered drive we need to 4159 * start recovery here. If we leave it to md_check_recovery, 4160 * it will remove the drives and not do the right thing 4161 */ 4162 if (mddev->degraded && !mddev->sync_thread) { 4163 int spares = 0; 4164 list_for_each_entry(rdev, &mddev->disks, same_set) 4165 if (rdev->raid_disk >= 0 && 4166 !test_bit(In_sync, &rdev->flags) && 4167 !test_bit(Faulty, &rdev->flags)) 4168 /* complete an interrupted recovery */ 4169 spares++; 4170 if (spares && mddev->pers->sync_request) { 4171 mddev->recovery = 0; 4172 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 4173 mddev->sync_thread = md_register_thread(md_do_sync, 4174 mddev, 4175 "%s_resync"); 4176 if (!mddev->sync_thread) { 4177 printk(KERN_ERR "%s: could not start resync" 4178 " thread...\n", 4179 mdname(mddev)); 4180 /* leave the spares where they are, it shouldn't hurt */ 4181 mddev->recovery = 0; 4182 } 4183 } 4184 } 4185 md_wakeup_thread(mddev->thread); 4186 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 4187 4188 mddev->changed = 1; 4189 md_new_event(mddev); 4190 sysfs_notify_dirent(mddev->sysfs_state); 4191 if (mddev->sysfs_action) 4192 sysfs_notify_dirent(mddev->sysfs_action); 4193 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4194 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4195 return 0; 4196 } 4197 4198 static int restart_array(mddev_t *mddev) 4199 { 4200 struct gendisk *disk = mddev->gendisk; 4201 4202 /* Complain if it has no devices */ 4203 if (list_empty(&mddev->disks)) 4204 return -ENXIO; 4205 if (!mddev->pers) 4206 return -EINVAL; 4207 if (!mddev->ro) 4208 return -EBUSY; 4209 mddev->safemode = 0; 4210 mddev->ro = 0; 4211 set_disk_ro(disk, 0); 4212 printk(KERN_INFO "md: %s switched to read-write mode.\n", 4213 mdname(mddev)); 4214 /* Kick recovery or resync if necessary */ 4215 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4216 md_wakeup_thread(mddev->thread); 4217 md_wakeup_thread(mddev->sync_thread); 4218 sysfs_notify_dirent(mddev->sysfs_state); 4219 return 0; 4220 } 4221 4222 /* similar to deny_write_access, but accounts for our holding a reference 4223 * to the file ourselves */ 4224 static int deny_bitmap_write_access(struct file * file) 4225 { 4226 struct inode *inode = file->f_mapping->host; 4227 4228 spin_lock(&inode->i_lock); 4229 if (atomic_read(&inode->i_writecount) > 1) { 4230 spin_unlock(&inode->i_lock); 4231 return -ETXTBSY; 4232 } 4233 atomic_set(&inode->i_writecount, -1); 4234 spin_unlock(&inode->i_lock); 4235 4236 return 0; 4237 } 4238 4239 static void restore_bitmap_write_access(struct file *file) 4240 { 4241 struct inode *inode = file->f_mapping->host; 4242 4243 spin_lock(&inode->i_lock); 4244 atomic_set(&inode->i_writecount, 1); 4245 spin_unlock(&inode->i_lock); 4246 } 4247 4248 /* mode: 4249 * 0 - completely stop and dis-assemble array 4250 * 1 - switch to readonly 4251 * 2 - stop but do not disassemble array 4252 */ 4253 static int do_md_stop(mddev_t * mddev, int mode, int is_open) 4254 { 4255 int err = 0; 4256 struct gendisk *disk = mddev->gendisk; 4257 mdk_rdev_t *rdev; 4258 4259 if (atomic_read(&mddev->openers) > is_open) { 4260 printk("md: %s still in use.\n",mdname(mddev)); 4261 return -EBUSY; 4262 } 4263 4264 if (mddev->pers) { 4265 4266 if (mddev->sync_thread) { 4267 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4268 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4269 md_unregister_thread(mddev->sync_thread); 4270 mddev->sync_thread = NULL; 4271 } 4272 4273 del_timer_sync(&mddev->safemode_timer); 4274 4275 switch(mode) { 4276 case 1: /* readonly */ 4277 err = -ENXIO; 4278 if (mddev->ro==1) 4279 goto out; 4280 mddev->ro = 1; 4281 break; 4282 case 0: /* disassemble */ 4283 case 2: /* stop */ 4284 bitmap_flush(mddev); 4285 md_super_wait(mddev); 4286 if (mddev->ro) 4287 set_disk_ro(disk, 0); 4288 4289 mddev->pers->stop(mddev); 4290 mddev->queue->merge_bvec_fn = NULL; 4291 mddev->queue->unplug_fn = NULL; 4292 mddev->queue->backing_dev_info.congested_fn = NULL; 4293 module_put(mddev->pers->owner); 4294 if (mddev->pers->sync_request) 4295 mddev->private = &md_redundancy_group; 4296 mddev->pers = NULL; 4297 /* tell userspace to handle 'inactive' */ 4298 sysfs_notify_dirent(mddev->sysfs_state); 4299 4300 list_for_each_entry(rdev, &mddev->disks, same_set) 4301 if (rdev->raid_disk >= 0) { 4302 char nm[20]; 4303 sprintf(nm, "rd%d", rdev->raid_disk); 4304 sysfs_remove_link(&mddev->kobj, nm); 4305 } 4306 4307 set_capacity(disk, 0); 4308 mddev->changed = 1; 4309 4310 if (mddev->ro) 4311 mddev->ro = 0; 4312 } 4313 if (!mddev->in_sync || mddev->flags) { 4314 /* mark array as shutdown cleanly */ 4315 mddev->in_sync = 1; 4316 md_update_sb(mddev, 1); 4317 } 4318 if (mode == 1) 4319 set_disk_ro(disk, 1); 4320 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4321 } 4322 4323 /* 4324 * Free resources if final stop 4325 */ 4326 if (mode == 0) { 4327 4328 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 4329 4330 bitmap_destroy(mddev); 4331 if (mddev->bitmap_file) { 4332 restore_bitmap_write_access(mddev->bitmap_file); 4333 fput(mddev->bitmap_file); 4334 mddev->bitmap_file = NULL; 4335 } 4336 mddev->bitmap_offset = 0; 4337 4338 /* make sure all md_delayed_delete calls have finished */ 4339 flush_scheduled_work(); 4340 4341 export_array(mddev); 4342 4343 mddev->array_sectors = 0; 4344 mddev->external_size = 0; 4345 mddev->dev_sectors = 0; 4346 mddev->raid_disks = 0; 4347 mddev->recovery_cp = 0; 4348 mddev->resync_min = 0; 4349 mddev->resync_max = MaxSector; 4350 mddev->reshape_position = MaxSector; 4351 mddev->external = 0; 4352 mddev->persistent = 0; 4353 mddev->level = LEVEL_NONE; 4354 mddev->clevel[0] = 0; 4355 mddev->flags = 0; 4356 mddev->ro = 0; 4357 mddev->metadata_type[0] = 0; 4358 mddev->chunk_sectors = 0; 4359 mddev->ctime = mddev->utime = 0; 4360 mddev->layout = 0; 4361 mddev->max_disks = 0; 4362 mddev->events = 0; 4363 mddev->delta_disks = 0; 4364 mddev->new_level = LEVEL_NONE; 4365 mddev->new_layout = 0; 4366 mddev->new_chunk_sectors = 0; 4367 mddev->curr_resync = 0; 4368 mddev->resync_mismatches = 0; 4369 mddev->suspend_lo = mddev->suspend_hi = 0; 4370 mddev->sync_speed_min = mddev->sync_speed_max = 0; 4371 mddev->recovery = 0; 4372 mddev->in_sync = 0; 4373 mddev->changed = 0; 4374 mddev->degraded = 0; 4375 mddev->barriers_work = 0; 4376 mddev->safemode = 0; 4377 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4378 if (mddev->hold_active == UNTIL_STOP) 4379 mddev->hold_active = 0; 4380 4381 } else if (mddev->pers) 4382 printk(KERN_INFO "md: %s switched to read-only mode.\n", 4383 mdname(mddev)); 4384 err = 0; 4385 blk_integrity_unregister(disk); 4386 md_new_event(mddev); 4387 sysfs_notify_dirent(mddev->sysfs_state); 4388 out: 4389 return err; 4390 } 4391 4392 #ifndef MODULE 4393 static void autorun_array(mddev_t *mddev) 4394 { 4395 mdk_rdev_t *rdev; 4396 int err; 4397 4398 if (list_empty(&mddev->disks)) 4399 return; 4400 4401 printk(KERN_INFO "md: running: "); 4402 4403 list_for_each_entry(rdev, &mddev->disks, same_set) { 4404 char b[BDEVNAME_SIZE]; 4405 printk("<%s>", bdevname(rdev->bdev,b)); 4406 } 4407 printk("\n"); 4408 4409 err = do_md_run(mddev); 4410 if (err) { 4411 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 4412 do_md_stop(mddev, 0, 0); 4413 } 4414 } 4415 4416 /* 4417 * lets try to run arrays based on all disks that have arrived 4418 * until now. (those are in pending_raid_disks) 4419 * 4420 * the method: pick the first pending disk, collect all disks with 4421 * the same UUID, remove all from the pending list and put them into 4422 * the 'same_array' list. Then order this list based on superblock 4423 * update time (freshest comes first), kick out 'old' disks and 4424 * compare superblocks. If everything's fine then run it. 4425 * 4426 * If "unit" is allocated, then bump its reference count 4427 */ 4428 static void autorun_devices(int part) 4429 { 4430 mdk_rdev_t *rdev0, *rdev, *tmp; 4431 mddev_t *mddev; 4432 char b[BDEVNAME_SIZE]; 4433 4434 printk(KERN_INFO "md: autorun ...\n"); 4435 while (!list_empty(&pending_raid_disks)) { 4436 int unit; 4437 dev_t dev; 4438 LIST_HEAD(candidates); 4439 rdev0 = list_entry(pending_raid_disks.next, 4440 mdk_rdev_t, same_set); 4441 4442 printk(KERN_INFO "md: considering %s ...\n", 4443 bdevname(rdev0->bdev,b)); 4444 INIT_LIST_HEAD(&candidates); 4445 rdev_for_each_list(rdev, tmp, &pending_raid_disks) 4446 if (super_90_load(rdev, rdev0, 0) >= 0) { 4447 printk(KERN_INFO "md: adding %s ...\n", 4448 bdevname(rdev->bdev,b)); 4449 list_move(&rdev->same_set, &candidates); 4450 } 4451 /* 4452 * now we have a set of devices, with all of them having 4453 * mostly sane superblocks. It's time to allocate the 4454 * mddev. 4455 */ 4456 if (part) { 4457 dev = MKDEV(mdp_major, 4458 rdev0->preferred_minor << MdpMinorShift); 4459 unit = MINOR(dev) >> MdpMinorShift; 4460 } else { 4461 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 4462 unit = MINOR(dev); 4463 } 4464 if (rdev0->preferred_minor != unit) { 4465 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 4466 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 4467 break; 4468 } 4469 4470 md_probe(dev, NULL, NULL); 4471 mddev = mddev_find(dev); 4472 if (!mddev || !mddev->gendisk) { 4473 if (mddev) 4474 mddev_put(mddev); 4475 printk(KERN_ERR 4476 "md: cannot allocate memory for md drive.\n"); 4477 break; 4478 } 4479 if (mddev_lock(mddev)) 4480 printk(KERN_WARNING "md: %s locked, cannot run\n", 4481 mdname(mddev)); 4482 else if (mddev->raid_disks || mddev->major_version 4483 || !list_empty(&mddev->disks)) { 4484 printk(KERN_WARNING 4485 "md: %s already running, cannot run %s\n", 4486 mdname(mddev), bdevname(rdev0->bdev,b)); 4487 mddev_unlock(mddev); 4488 } else { 4489 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 4490 mddev->persistent = 1; 4491 rdev_for_each_list(rdev, tmp, &candidates) { 4492 list_del_init(&rdev->same_set); 4493 if (bind_rdev_to_array(rdev, mddev)) 4494 export_rdev(rdev); 4495 } 4496 autorun_array(mddev); 4497 mddev_unlock(mddev); 4498 } 4499 /* on success, candidates will be empty, on error 4500 * it won't... 4501 */ 4502 rdev_for_each_list(rdev, tmp, &candidates) { 4503 list_del_init(&rdev->same_set); 4504 export_rdev(rdev); 4505 } 4506 mddev_put(mddev); 4507 } 4508 printk(KERN_INFO "md: ... autorun DONE.\n"); 4509 } 4510 #endif /* !MODULE */ 4511 4512 static int get_version(void __user * arg) 4513 { 4514 mdu_version_t ver; 4515 4516 ver.major = MD_MAJOR_VERSION; 4517 ver.minor = MD_MINOR_VERSION; 4518 ver.patchlevel = MD_PATCHLEVEL_VERSION; 4519 4520 if (copy_to_user(arg, &ver, sizeof(ver))) 4521 return -EFAULT; 4522 4523 return 0; 4524 } 4525 4526 static int get_array_info(mddev_t * mddev, void __user * arg) 4527 { 4528 mdu_array_info_t info; 4529 int nr,working,active,failed,spare; 4530 mdk_rdev_t *rdev; 4531 4532 nr=working=active=failed=spare=0; 4533 list_for_each_entry(rdev, &mddev->disks, same_set) { 4534 nr++; 4535 if (test_bit(Faulty, &rdev->flags)) 4536 failed++; 4537 else { 4538 working++; 4539 if (test_bit(In_sync, &rdev->flags)) 4540 active++; 4541 else 4542 spare++; 4543 } 4544 } 4545 4546 info.major_version = mddev->major_version; 4547 info.minor_version = mddev->minor_version; 4548 info.patch_version = MD_PATCHLEVEL_VERSION; 4549 info.ctime = mddev->ctime; 4550 info.level = mddev->level; 4551 info.size = mddev->dev_sectors / 2; 4552 if (info.size != mddev->dev_sectors / 2) /* overflow */ 4553 info.size = -1; 4554 info.nr_disks = nr; 4555 info.raid_disks = mddev->raid_disks; 4556 info.md_minor = mddev->md_minor; 4557 info.not_persistent= !mddev->persistent; 4558 4559 info.utime = mddev->utime; 4560 info.state = 0; 4561 if (mddev->in_sync) 4562 info.state = (1<<MD_SB_CLEAN); 4563 if (mddev->bitmap && mddev->bitmap_offset) 4564 info.state = (1<<MD_SB_BITMAP_PRESENT); 4565 info.active_disks = active; 4566 info.working_disks = working; 4567 info.failed_disks = failed; 4568 info.spare_disks = spare; 4569 4570 info.layout = mddev->layout; 4571 info.chunk_size = mddev->chunk_sectors << 9; 4572 4573 if (copy_to_user(arg, &info, sizeof(info))) 4574 return -EFAULT; 4575 4576 return 0; 4577 } 4578 4579 static int get_bitmap_file(mddev_t * mddev, void __user * arg) 4580 { 4581 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 4582 char *ptr, *buf = NULL; 4583 int err = -ENOMEM; 4584 4585 if (md_allow_write(mddev)) 4586 file = kmalloc(sizeof(*file), GFP_NOIO); 4587 else 4588 file = kmalloc(sizeof(*file), GFP_KERNEL); 4589 4590 if (!file) 4591 goto out; 4592 4593 /* bitmap disabled, zero the first byte and copy out */ 4594 if (!mddev->bitmap || !mddev->bitmap->file) { 4595 file->pathname[0] = '\0'; 4596 goto copy_out; 4597 } 4598 4599 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 4600 if (!buf) 4601 goto out; 4602 4603 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); 4604 if (IS_ERR(ptr)) 4605 goto out; 4606 4607 strcpy(file->pathname, ptr); 4608 4609 copy_out: 4610 err = 0; 4611 if (copy_to_user(arg, file, sizeof(*file))) 4612 err = -EFAULT; 4613 out: 4614 kfree(buf); 4615 kfree(file); 4616 return err; 4617 } 4618 4619 static int get_disk_info(mddev_t * mddev, void __user * arg) 4620 { 4621 mdu_disk_info_t info; 4622 mdk_rdev_t *rdev; 4623 4624 if (copy_from_user(&info, arg, sizeof(info))) 4625 return -EFAULT; 4626 4627 rdev = find_rdev_nr(mddev, info.number); 4628 if (rdev) { 4629 info.major = MAJOR(rdev->bdev->bd_dev); 4630 info.minor = MINOR(rdev->bdev->bd_dev); 4631 info.raid_disk = rdev->raid_disk; 4632 info.state = 0; 4633 if (test_bit(Faulty, &rdev->flags)) 4634 info.state |= (1<<MD_DISK_FAULTY); 4635 else if (test_bit(In_sync, &rdev->flags)) { 4636 info.state |= (1<<MD_DISK_ACTIVE); 4637 info.state |= (1<<MD_DISK_SYNC); 4638 } 4639 if (test_bit(WriteMostly, &rdev->flags)) 4640 info.state |= (1<<MD_DISK_WRITEMOSTLY); 4641 } else { 4642 info.major = info.minor = 0; 4643 info.raid_disk = -1; 4644 info.state = (1<<MD_DISK_REMOVED); 4645 } 4646 4647 if (copy_to_user(arg, &info, sizeof(info))) 4648 return -EFAULT; 4649 4650 return 0; 4651 } 4652 4653 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) 4654 { 4655 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 4656 mdk_rdev_t *rdev; 4657 dev_t dev = MKDEV(info->major,info->minor); 4658 4659 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 4660 return -EOVERFLOW; 4661 4662 if (!mddev->raid_disks) { 4663 int err; 4664 /* expecting a device which has a superblock */ 4665 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 4666 if (IS_ERR(rdev)) { 4667 printk(KERN_WARNING 4668 "md: md_import_device returned %ld\n", 4669 PTR_ERR(rdev)); 4670 return PTR_ERR(rdev); 4671 } 4672 if (!list_empty(&mddev->disks)) { 4673 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 4674 mdk_rdev_t, same_set); 4675 int err = super_types[mddev->major_version] 4676 .load_super(rdev, rdev0, mddev->minor_version); 4677 if (err < 0) { 4678 printk(KERN_WARNING 4679 "md: %s has different UUID to %s\n", 4680 bdevname(rdev->bdev,b), 4681 bdevname(rdev0->bdev,b2)); 4682 export_rdev(rdev); 4683 return -EINVAL; 4684 } 4685 } 4686 err = bind_rdev_to_array(rdev, mddev); 4687 if (err) 4688 export_rdev(rdev); 4689 return err; 4690 } 4691 4692 /* 4693 * add_new_disk can be used once the array is assembled 4694 * to add "hot spares". They must already have a superblock 4695 * written 4696 */ 4697 if (mddev->pers) { 4698 int err; 4699 if (!mddev->pers->hot_add_disk) { 4700 printk(KERN_WARNING 4701 "%s: personality does not support diskops!\n", 4702 mdname(mddev)); 4703 return -EINVAL; 4704 } 4705 if (mddev->persistent) 4706 rdev = md_import_device(dev, mddev->major_version, 4707 mddev->minor_version); 4708 else 4709 rdev = md_import_device(dev, -1, -1); 4710 if (IS_ERR(rdev)) { 4711 printk(KERN_WARNING 4712 "md: md_import_device returned %ld\n", 4713 PTR_ERR(rdev)); 4714 return PTR_ERR(rdev); 4715 } 4716 /* set save_raid_disk if appropriate */ 4717 if (!mddev->persistent) { 4718 if (info->state & (1<<MD_DISK_SYNC) && 4719 info->raid_disk < mddev->raid_disks) 4720 rdev->raid_disk = info->raid_disk; 4721 else 4722 rdev->raid_disk = -1; 4723 } else 4724 super_types[mddev->major_version]. 4725 validate_super(mddev, rdev); 4726 rdev->saved_raid_disk = rdev->raid_disk; 4727 4728 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 4729 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 4730 set_bit(WriteMostly, &rdev->flags); 4731 else 4732 clear_bit(WriteMostly, &rdev->flags); 4733 4734 rdev->raid_disk = -1; 4735 err = bind_rdev_to_array(rdev, mddev); 4736 if (!err && !mddev->pers->hot_remove_disk) { 4737 /* If there is hot_add_disk but no hot_remove_disk 4738 * then added disks for geometry changes, 4739 * and should be added immediately. 4740 */ 4741 super_types[mddev->major_version]. 4742 validate_super(mddev, rdev); 4743 err = mddev->pers->hot_add_disk(mddev, rdev); 4744 if (err) 4745 unbind_rdev_from_array(rdev); 4746 } 4747 if (err) 4748 export_rdev(rdev); 4749 else 4750 sysfs_notify_dirent(rdev->sysfs_state); 4751 4752 md_update_sb(mddev, 1); 4753 if (mddev->degraded) 4754 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 4755 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4756 md_wakeup_thread(mddev->thread); 4757 return err; 4758 } 4759 4760 /* otherwise, add_new_disk is only allowed 4761 * for major_version==0 superblocks 4762 */ 4763 if (mddev->major_version != 0) { 4764 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 4765 mdname(mddev)); 4766 return -EINVAL; 4767 } 4768 4769 if (!(info->state & (1<<MD_DISK_FAULTY))) { 4770 int err; 4771 rdev = md_import_device(dev, -1, 0); 4772 if (IS_ERR(rdev)) { 4773 printk(KERN_WARNING 4774 "md: error, md_import_device() returned %ld\n", 4775 PTR_ERR(rdev)); 4776 return PTR_ERR(rdev); 4777 } 4778 rdev->desc_nr = info->number; 4779 if (info->raid_disk < mddev->raid_disks) 4780 rdev->raid_disk = info->raid_disk; 4781 else 4782 rdev->raid_disk = -1; 4783 4784 if (rdev->raid_disk < mddev->raid_disks) 4785 if (info->state & (1<<MD_DISK_SYNC)) 4786 set_bit(In_sync, &rdev->flags); 4787 4788 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 4789 set_bit(WriteMostly, &rdev->flags); 4790 4791 if (!mddev->persistent) { 4792 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 4793 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; 4794 } else 4795 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 4796 rdev->sectors = rdev->sb_start; 4797 4798 err = bind_rdev_to_array(rdev, mddev); 4799 if (err) { 4800 export_rdev(rdev); 4801 return err; 4802 } 4803 } 4804 4805 return 0; 4806 } 4807 4808 static int hot_remove_disk(mddev_t * mddev, dev_t dev) 4809 { 4810 char b[BDEVNAME_SIZE]; 4811 mdk_rdev_t *rdev; 4812 4813 rdev = find_rdev(mddev, dev); 4814 if (!rdev) 4815 return -ENXIO; 4816 4817 if (rdev->raid_disk >= 0) 4818 goto busy; 4819 4820 kick_rdev_from_array(rdev); 4821 md_update_sb(mddev, 1); 4822 md_new_event(mddev); 4823 4824 return 0; 4825 busy: 4826 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", 4827 bdevname(rdev->bdev,b), mdname(mddev)); 4828 return -EBUSY; 4829 } 4830 4831 static int hot_add_disk(mddev_t * mddev, dev_t dev) 4832 { 4833 char b[BDEVNAME_SIZE]; 4834 int err; 4835 mdk_rdev_t *rdev; 4836 4837 if (!mddev->pers) 4838 return -ENODEV; 4839 4840 if (mddev->major_version != 0) { 4841 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 4842 " version-0 superblocks.\n", 4843 mdname(mddev)); 4844 return -EINVAL; 4845 } 4846 if (!mddev->pers->hot_add_disk) { 4847 printk(KERN_WARNING 4848 "%s: personality does not support diskops!\n", 4849 mdname(mddev)); 4850 return -EINVAL; 4851 } 4852 4853 rdev = md_import_device(dev, -1, 0); 4854 if (IS_ERR(rdev)) { 4855 printk(KERN_WARNING 4856 "md: error, md_import_device() returned %ld\n", 4857 PTR_ERR(rdev)); 4858 return -EINVAL; 4859 } 4860 4861 if (mddev->persistent) 4862 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 4863 else 4864 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; 4865 4866 rdev->sectors = rdev->sb_start; 4867 4868 if (test_bit(Faulty, &rdev->flags)) { 4869 printk(KERN_WARNING 4870 "md: can not hot-add faulty %s disk to %s!\n", 4871 bdevname(rdev->bdev,b), mdname(mddev)); 4872 err = -EINVAL; 4873 goto abort_export; 4874 } 4875 clear_bit(In_sync, &rdev->flags); 4876 rdev->desc_nr = -1; 4877 rdev->saved_raid_disk = -1; 4878 err = bind_rdev_to_array(rdev, mddev); 4879 if (err) 4880 goto abort_export; 4881 4882 /* 4883 * The rest should better be atomic, we can have disk failures 4884 * noticed in interrupt contexts ... 4885 */ 4886 4887 rdev->raid_disk = -1; 4888 4889 md_update_sb(mddev, 1); 4890 4891 /* 4892 * Kick recovery, maybe this spare has to be added to the 4893 * array immediately. 4894 */ 4895 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4896 md_wakeup_thread(mddev->thread); 4897 md_new_event(mddev); 4898 return 0; 4899 4900 abort_export: 4901 export_rdev(rdev); 4902 return err; 4903 } 4904 4905 static int set_bitmap_file(mddev_t *mddev, int fd) 4906 { 4907 int err; 4908 4909 if (mddev->pers) { 4910 if (!mddev->pers->quiesce) 4911 return -EBUSY; 4912 if (mddev->recovery || mddev->sync_thread) 4913 return -EBUSY; 4914 /* we should be able to change the bitmap.. */ 4915 } 4916 4917 4918 if (fd >= 0) { 4919 if (mddev->bitmap) 4920 return -EEXIST; /* cannot add when bitmap is present */ 4921 mddev->bitmap_file = fget(fd); 4922 4923 if (mddev->bitmap_file == NULL) { 4924 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 4925 mdname(mddev)); 4926 return -EBADF; 4927 } 4928 4929 err = deny_bitmap_write_access(mddev->bitmap_file); 4930 if (err) { 4931 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 4932 mdname(mddev)); 4933 fput(mddev->bitmap_file); 4934 mddev->bitmap_file = NULL; 4935 return err; 4936 } 4937 mddev->bitmap_offset = 0; /* file overrides offset */ 4938 } else if (mddev->bitmap == NULL) 4939 return -ENOENT; /* cannot remove what isn't there */ 4940 err = 0; 4941 if (mddev->pers) { 4942 mddev->pers->quiesce(mddev, 1); 4943 if (fd >= 0) 4944 err = bitmap_create(mddev); 4945 if (fd < 0 || err) { 4946 bitmap_destroy(mddev); 4947 fd = -1; /* make sure to put the file */ 4948 } 4949 mddev->pers->quiesce(mddev, 0); 4950 } 4951 if (fd < 0) { 4952 if (mddev->bitmap_file) { 4953 restore_bitmap_write_access(mddev->bitmap_file); 4954 fput(mddev->bitmap_file); 4955 } 4956 mddev->bitmap_file = NULL; 4957 } 4958 4959 return err; 4960 } 4961 4962 /* 4963 * set_array_info is used two different ways 4964 * The original usage is when creating a new array. 4965 * In this usage, raid_disks is > 0 and it together with 4966 * level, size, not_persistent,layout,chunksize determine the 4967 * shape of the array. 4968 * This will always create an array with a type-0.90.0 superblock. 4969 * The newer usage is when assembling an array. 4970 * In this case raid_disks will be 0, and the major_version field is 4971 * use to determine which style super-blocks are to be found on the devices. 4972 * The minor and patch _version numbers are also kept incase the 4973 * super_block handler wishes to interpret them. 4974 */ 4975 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) 4976 { 4977 4978 if (info->raid_disks == 0) { 4979 /* just setting version number for superblock loading */ 4980 if (info->major_version < 0 || 4981 info->major_version >= ARRAY_SIZE(super_types) || 4982 super_types[info->major_version].name == NULL) { 4983 /* maybe try to auto-load a module? */ 4984 printk(KERN_INFO 4985 "md: superblock version %d not known\n", 4986 info->major_version); 4987 return -EINVAL; 4988 } 4989 mddev->major_version = info->major_version; 4990 mddev->minor_version = info->minor_version; 4991 mddev->patch_version = info->patch_version; 4992 mddev->persistent = !info->not_persistent; 4993 return 0; 4994 } 4995 mddev->major_version = MD_MAJOR_VERSION; 4996 mddev->minor_version = MD_MINOR_VERSION; 4997 mddev->patch_version = MD_PATCHLEVEL_VERSION; 4998 mddev->ctime = get_seconds(); 4999 5000 mddev->level = info->level; 5001 mddev->clevel[0] = 0; 5002 mddev->dev_sectors = 2 * (sector_t)info->size; 5003 mddev->raid_disks = info->raid_disks; 5004 /* don't set md_minor, it is determined by which /dev/md* was 5005 * openned 5006 */ 5007 if (info->state & (1<<MD_SB_CLEAN)) 5008 mddev->recovery_cp = MaxSector; 5009 else 5010 mddev->recovery_cp = 0; 5011 mddev->persistent = ! info->not_persistent; 5012 mddev->external = 0; 5013 5014 mddev->layout = info->layout; 5015 mddev->chunk_sectors = info->chunk_size >> 9; 5016 5017 mddev->max_disks = MD_SB_DISKS; 5018 5019 if (mddev->persistent) 5020 mddev->flags = 0; 5021 set_bit(MD_CHANGE_DEVS, &mddev->flags); 5022 5023 mddev->default_bitmap_offset = MD_SB_BYTES >> 9; 5024 mddev->bitmap_offset = 0; 5025 5026 mddev->reshape_position = MaxSector; 5027 5028 /* 5029 * Generate a 128 bit UUID 5030 */ 5031 get_random_bytes(mddev->uuid, 16); 5032 5033 mddev->new_level = mddev->level; 5034 mddev->new_chunk_sectors = mddev->chunk_sectors; 5035 mddev->new_layout = mddev->layout; 5036 mddev->delta_disks = 0; 5037 5038 return 0; 5039 } 5040 5041 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) 5042 { 5043 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); 5044 5045 if (mddev->external_size) 5046 return; 5047 5048 mddev->array_sectors = array_sectors; 5049 } 5050 EXPORT_SYMBOL(md_set_array_sectors); 5051 5052 static int update_size(mddev_t *mddev, sector_t num_sectors) 5053 { 5054 mdk_rdev_t *rdev; 5055 int rv; 5056 int fit = (num_sectors == 0); 5057 5058 if (mddev->pers->resize == NULL) 5059 return -EINVAL; 5060 /* The "num_sectors" is the number of sectors of each device that 5061 * is used. This can only make sense for arrays with redundancy. 5062 * linear and raid0 always use whatever space is available. We can only 5063 * consider changing this number if no resync or reconstruction is 5064 * happening, and if the new size is acceptable. It must fit before the 5065 * sb_start or, if that is <data_offset, it must fit before the size 5066 * of each device. If num_sectors is zero, we find the largest size 5067 * that fits. 5068 5069 */ 5070 if (mddev->sync_thread) 5071 return -EBUSY; 5072 if (mddev->bitmap) 5073 /* Sorry, cannot grow a bitmap yet, just remove it, 5074 * grow, and re-add. 5075 */ 5076 return -EBUSY; 5077 list_for_each_entry(rdev, &mddev->disks, same_set) { 5078 sector_t avail = rdev->sectors; 5079 5080 if (fit && (num_sectors == 0 || num_sectors > avail)) 5081 num_sectors = avail; 5082 if (avail < num_sectors) 5083 return -ENOSPC; 5084 } 5085 rv = mddev->pers->resize(mddev, num_sectors); 5086 if (!rv) { 5087 struct block_device *bdev; 5088 5089 bdev = bdget_disk(mddev->gendisk, 0); 5090 if (bdev) { 5091 mutex_lock(&bdev->bd_inode->i_mutex); 5092 i_size_write(bdev->bd_inode, 5093 (loff_t)mddev->array_sectors << 9); 5094 mutex_unlock(&bdev->bd_inode->i_mutex); 5095 bdput(bdev); 5096 } 5097 } 5098 return rv; 5099 } 5100 5101 static int update_raid_disks(mddev_t *mddev, int raid_disks) 5102 { 5103 int rv; 5104 /* change the number of raid disks */ 5105 if (mddev->pers->check_reshape == NULL) 5106 return -EINVAL; 5107 if (raid_disks <= 0 || 5108 raid_disks >= mddev->max_disks) 5109 return -EINVAL; 5110 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 5111 return -EBUSY; 5112 mddev->delta_disks = raid_disks - mddev->raid_disks; 5113 5114 rv = mddev->pers->check_reshape(mddev); 5115 return rv; 5116 } 5117 5118 5119 /* 5120 * update_array_info is used to change the configuration of an 5121 * on-line array. 5122 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 5123 * fields in the info are checked against the array. 5124 * Any differences that cannot be handled will cause an error. 5125 * Normally, only one change can be managed at a time. 5126 */ 5127 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) 5128 { 5129 int rv = 0; 5130 int cnt = 0; 5131 int state = 0; 5132 5133 /* calculate expected state,ignoring low bits */ 5134 if (mddev->bitmap && mddev->bitmap_offset) 5135 state |= (1 << MD_SB_BITMAP_PRESENT); 5136 5137 if (mddev->major_version != info->major_version || 5138 mddev->minor_version != info->minor_version || 5139 /* mddev->patch_version != info->patch_version || */ 5140 mddev->ctime != info->ctime || 5141 mddev->level != info->level || 5142 /* mddev->layout != info->layout || */ 5143 !mddev->persistent != info->not_persistent|| 5144 mddev->chunk_sectors != info->chunk_size >> 9 || 5145 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 5146 ((state^info->state) & 0xfffffe00) 5147 ) 5148 return -EINVAL; 5149 /* Check there is only one change */ 5150 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5151 cnt++; 5152 if (mddev->raid_disks != info->raid_disks) 5153 cnt++; 5154 if (mddev->layout != info->layout) 5155 cnt++; 5156 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) 5157 cnt++; 5158 if (cnt == 0) 5159 return 0; 5160 if (cnt > 1) 5161 return -EINVAL; 5162 5163 if (mddev->layout != info->layout) { 5164 /* Change layout 5165 * we don't need to do anything at the md level, the 5166 * personality will take care of it all. 5167 */ 5168 if (mddev->pers->check_reshape == NULL) 5169 return -EINVAL; 5170 else { 5171 mddev->new_layout = info->layout; 5172 rv = mddev->pers->check_reshape(mddev); 5173 if (rv) 5174 mddev->new_layout = mddev->layout; 5175 return rv; 5176 } 5177 } 5178 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5179 rv = update_size(mddev, (sector_t)info->size * 2); 5180 5181 if (mddev->raid_disks != info->raid_disks) 5182 rv = update_raid_disks(mddev, info->raid_disks); 5183 5184 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 5185 if (mddev->pers->quiesce == NULL) 5186 return -EINVAL; 5187 if (mddev->recovery || mddev->sync_thread) 5188 return -EBUSY; 5189 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 5190 /* add the bitmap */ 5191 if (mddev->bitmap) 5192 return -EEXIST; 5193 if (mddev->default_bitmap_offset == 0) 5194 return -EINVAL; 5195 mddev->bitmap_offset = mddev->default_bitmap_offset; 5196 mddev->pers->quiesce(mddev, 1); 5197 rv = bitmap_create(mddev); 5198 if (rv) 5199 bitmap_destroy(mddev); 5200 mddev->pers->quiesce(mddev, 0); 5201 } else { 5202 /* remove the bitmap */ 5203 if (!mddev->bitmap) 5204 return -ENOENT; 5205 if (mddev->bitmap->file) 5206 return -EINVAL; 5207 mddev->pers->quiesce(mddev, 1); 5208 bitmap_destroy(mddev); 5209 mddev->pers->quiesce(mddev, 0); 5210 mddev->bitmap_offset = 0; 5211 } 5212 } 5213 md_update_sb(mddev, 1); 5214 return rv; 5215 } 5216 5217 static int set_disk_faulty(mddev_t *mddev, dev_t dev) 5218 { 5219 mdk_rdev_t *rdev; 5220 5221 if (mddev->pers == NULL) 5222 return -ENODEV; 5223 5224 rdev = find_rdev(mddev, dev); 5225 if (!rdev) 5226 return -ENODEV; 5227 5228 md_error(mddev, rdev); 5229 return 0; 5230 } 5231 5232 /* 5233 * We have a problem here : there is no easy way to give a CHS 5234 * virtual geometry. We currently pretend that we have a 2 heads 5235 * 4 sectors (with a BIG number of cylinders...). This drives 5236 * dosfs just mad... ;-) 5237 */ 5238 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) 5239 { 5240 mddev_t *mddev = bdev->bd_disk->private_data; 5241 5242 geo->heads = 2; 5243 geo->sectors = 4; 5244 geo->cylinders = get_capacity(mddev->gendisk) / 8; 5245 return 0; 5246 } 5247 5248 static int md_ioctl(struct block_device *bdev, fmode_t mode, 5249 unsigned int cmd, unsigned long arg) 5250 { 5251 int err = 0; 5252 void __user *argp = (void __user *)arg; 5253 mddev_t *mddev = NULL; 5254 5255 if (!capable(CAP_SYS_ADMIN)) 5256 return -EACCES; 5257 5258 /* 5259 * Commands dealing with the RAID driver but not any 5260 * particular array: 5261 */ 5262 switch (cmd) 5263 { 5264 case RAID_VERSION: 5265 err = get_version(argp); 5266 goto done; 5267 5268 case PRINT_RAID_DEBUG: 5269 err = 0; 5270 md_print_devices(); 5271 goto done; 5272 5273 #ifndef MODULE 5274 case RAID_AUTORUN: 5275 err = 0; 5276 autostart_arrays(arg); 5277 goto done; 5278 #endif 5279 default:; 5280 } 5281 5282 /* 5283 * Commands creating/starting a new array: 5284 */ 5285 5286 mddev = bdev->bd_disk->private_data; 5287 5288 if (!mddev) { 5289 BUG(); 5290 goto abort; 5291 } 5292 5293 err = mddev_lock(mddev); 5294 if (err) { 5295 printk(KERN_INFO 5296 "md: ioctl lock interrupted, reason %d, cmd %d\n", 5297 err, cmd); 5298 goto abort; 5299 } 5300 5301 switch (cmd) 5302 { 5303 case SET_ARRAY_INFO: 5304 { 5305 mdu_array_info_t info; 5306 if (!arg) 5307 memset(&info, 0, sizeof(info)); 5308 else if (copy_from_user(&info, argp, sizeof(info))) { 5309 err = -EFAULT; 5310 goto abort_unlock; 5311 } 5312 if (mddev->pers) { 5313 err = update_array_info(mddev, &info); 5314 if (err) { 5315 printk(KERN_WARNING "md: couldn't update" 5316 " array info. %d\n", err); 5317 goto abort_unlock; 5318 } 5319 goto done_unlock; 5320 } 5321 if (!list_empty(&mddev->disks)) { 5322 printk(KERN_WARNING 5323 "md: array %s already has disks!\n", 5324 mdname(mddev)); 5325 err = -EBUSY; 5326 goto abort_unlock; 5327 } 5328 if (mddev->raid_disks) { 5329 printk(KERN_WARNING 5330 "md: array %s already initialised!\n", 5331 mdname(mddev)); 5332 err = -EBUSY; 5333 goto abort_unlock; 5334 } 5335 err = set_array_info(mddev, &info); 5336 if (err) { 5337 printk(KERN_WARNING "md: couldn't set" 5338 " array info. %d\n", err); 5339 goto abort_unlock; 5340 } 5341 } 5342 goto done_unlock; 5343 5344 default:; 5345 } 5346 5347 /* 5348 * Commands querying/configuring an existing array: 5349 */ 5350 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 5351 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ 5352 if ((!mddev->raid_disks && !mddev->external) 5353 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 5354 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE 5355 && cmd != GET_BITMAP_FILE) { 5356 err = -ENODEV; 5357 goto abort_unlock; 5358 } 5359 5360 /* 5361 * Commands even a read-only array can execute: 5362 */ 5363 switch (cmd) 5364 { 5365 case GET_ARRAY_INFO: 5366 err = get_array_info(mddev, argp); 5367 goto done_unlock; 5368 5369 case GET_BITMAP_FILE: 5370 err = get_bitmap_file(mddev, argp); 5371 goto done_unlock; 5372 5373 case GET_DISK_INFO: 5374 err = get_disk_info(mddev, argp); 5375 goto done_unlock; 5376 5377 case RESTART_ARRAY_RW: 5378 err = restart_array(mddev); 5379 goto done_unlock; 5380 5381 case STOP_ARRAY: 5382 err = do_md_stop(mddev, 0, 1); 5383 goto done_unlock; 5384 5385 case STOP_ARRAY_RO: 5386 err = do_md_stop(mddev, 1, 1); 5387 goto done_unlock; 5388 5389 } 5390 5391 /* 5392 * The remaining ioctls are changing the state of the 5393 * superblock, so we do not allow them on read-only arrays. 5394 * However non-MD ioctls (e.g. get-size) will still come through 5395 * here and hit the 'default' below, so only disallow 5396 * 'md' ioctls, and switch to rw mode if started auto-readonly. 5397 */ 5398 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { 5399 if (mddev->ro == 2) { 5400 mddev->ro = 0; 5401 sysfs_notify_dirent(mddev->sysfs_state); 5402 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5403 md_wakeup_thread(mddev->thread); 5404 } else { 5405 err = -EROFS; 5406 goto abort_unlock; 5407 } 5408 } 5409 5410 switch (cmd) 5411 { 5412 case ADD_NEW_DISK: 5413 { 5414 mdu_disk_info_t info; 5415 if (copy_from_user(&info, argp, sizeof(info))) 5416 err = -EFAULT; 5417 else 5418 err = add_new_disk(mddev, &info); 5419 goto done_unlock; 5420 } 5421 5422 case HOT_REMOVE_DISK: 5423 err = hot_remove_disk(mddev, new_decode_dev(arg)); 5424 goto done_unlock; 5425 5426 case HOT_ADD_DISK: 5427 err = hot_add_disk(mddev, new_decode_dev(arg)); 5428 goto done_unlock; 5429 5430 case SET_DISK_FAULTY: 5431 err = set_disk_faulty(mddev, new_decode_dev(arg)); 5432 goto done_unlock; 5433 5434 case RUN_ARRAY: 5435 err = do_md_run(mddev); 5436 goto done_unlock; 5437 5438 case SET_BITMAP_FILE: 5439 err = set_bitmap_file(mddev, (int)arg); 5440 goto done_unlock; 5441 5442 default: 5443 err = -EINVAL; 5444 goto abort_unlock; 5445 } 5446 5447 done_unlock: 5448 abort_unlock: 5449 if (mddev->hold_active == UNTIL_IOCTL && 5450 err != -EINVAL) 5451 mddev->hold_active = 0; 5452 mddev_unlock(mddev); 5453 5454 return err; 5455 done: 5456 if (err) 5457 MD_BUG(); 5458 abort: 5459 return err; 5460 } 5461 5462 static int md_open(struct block_device *bdev, fmode_t mode) 5463 { 5464 /* 5465 * Succeed if we can lock the mddev, which confirms that 5466 * it isn't being stopped right now. 5467 */ 5468 mddev_t *mddev = mddev_find(bdev->bd_dev); 5469 int err; 5470 5471 if (mddev->gendisk != bdev->bd_disk) { 5472 /* we are racing with mddev_put which is discarding this 5473 * bd_disk. 5474 */ 5475 mddev_put(mddev); 5476 /* Wait until bdev->bd_disk is definitely gone */ 5477 flush_scheduled_work(); 5478 /* Then retry the open from the top */ 5479 return -ERESTARTSYS; 5480 } 5481 BUG_ON(mddev != bdev->bd_disk->private_data); 5482 5483 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1))) 5484 goto out; 5485 5486 err = 0; 5487 atomic_inc(&mddev->openers); 5488 mddev_unlock(mddev); 5489 5490 check_disk_change(bdev); 5491 out: 5492 return err; 5493 } 5494 5495 static int md_release(struct gendisk *disk, fmode_t mode) 5496 { 5497 mddev_t *mddev = disk->private_data; 5498 5499 BUG_ON(!mddev); 5500 atomic_dec(&mddev->openers); 5501 mddev_put(mddev); 5502 5503 return 0; 5504 } 5505 5506 static int md_media_changed(struct gendisk *disk) 5507 { 5508 mddev_t *mddev = disk->private_data; 5509 5510 return mddev->changed; 5511 } 5512 5513 static int md_revalidate(struct gendisk *disk) 5514 { 5515 mddev_t *mddev = disk->private_data; 5516 5517 mddev->changed = 0; 5518 return 0; 5519 } 5520 static struct block_device_operations md_fops = 5521 { 5522 .owner = THIS_MODULE, 5523 .open = md_open, 5524 .release = md_release, 5525 .ioctl = md_ioctl, 5526 .getgeo = md_getgeo, 5527 .media_changed = md_media_changed, 5528 .revalidate_disk= md_revalidate, 5529 }; 5530 5531 static int md_thread(void * arg) 5532 { 5533 mdk_thread_t *thread = arg; 5534 5535 /* 5536 * md_thread is a 'system-thread', it's priority should be very 5537 * high. We avoid resource deadlocks individually in each 5538 * raid personality. (RAID5 does preallocation) We also use RR and 5539 * the very same RT priority as kswapd, thus we will never get 5540 * into a priority inversion deadlock. 5541 * 5542 * we definitely have to have equal or higher priority than 5543 * bdflush, otherwise bdflush will deadlock if there are too 5544 * many dirty RAID5 blocks. 5545 */ 5546 5547 allow_signal(SIGKILL); 5548 while (!kthread_should_stop()) { 5549 5550 /* We need to wait INTERRUPTIBLE so that 5551 * we don't add to the load-average. 5552 * That means we need to be sure no signals are 5553 * pending 5554 */ 5555 if (signal_pending(current)) 5556 flush_signals(current); 5557 5558 wait_event_interruptible_timeout 5559 (thread->wqueue, 5560 test_bit(THREAD_WAKEUP, &thread->flags) 5561 || kthread_should_stop(), 5562 thread->timeout); 5563 5564 clear_bit(THREAD_WAKEUP, &thread->flags); 5565 5566 thread->run(thread->mddev); 5567 } 5568 5569 return 0; 5570 } 5571 5572 void md_wakeup_thread(mdk_thread_t *thread) 5573 { 5574 if (thread) { 5575 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); 5576 set_bit(THREAD_WAKEUP, &thread->flags); 5577 wake_up(&thread->wqueue); 5578 } 5579 } 5580 5581 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, 5582 const char *name) 5583 { 5584 mdk_thread_t *thread; 5585 5586 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); 5587 if (!thread) 5588 return NULL; 5589 5590 init_waitqueue_head(&thread->wqueue); 5591 5592 thread->run = run; 5593 thread->mddev = mddev; 5594 thread->timeout = MAX_SCHEDULE_TIMEOUT; 5595 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); 5596 if (IS_ERR(thread->tsk)) { 5597 kfree(thread); 5598 return NULL; 5599 } 5600 return thread; 5601 } 5602 5603 void md_unregister_thread(mdk_thread_t *thread) 5604 { 5605 if (!thread) 5606 return; 5607 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); 5608 5609 kthread_stop(thread->tsk); 5610 kfree(thread); 5611 } 5612 5613 void md_error(mddev_t *mddev, mdk_rdev_t *rdev) 5614 { 5615 if (!mddev) { 5616 MD_BUG(); 5617 return; 5618 } 5619 5620 if (!rdev || test_bit(Faulty, &rdev->flags)) 5621 return; 5622 5623 if (mddev->external) 5624 set_bit(Blocked, &rdev->flags); 5625 /* 5626 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", 5627 mdname(mddev), 5628 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), 5629 __builtin_return_address(0),__builtin_return_address(1), 5630 __builtin_return_address(2),__builtin_return_address(3)); 5631 */ 5632 if (!mddev->pers) 5633 return; 5634 if (!mddev->pers->error_handler) 5635 return; 5636 mddev->pers->error_handler(mddev,rdev); 5637 if (mddev->degraded) 5638 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 5639 set_bit(StateChanged, &rdev->flags); 5640 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 5641 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5642 md_wakeup_thread(mddev->thread); 5643 md_new_event_inintr(mddev); 5644 } 5645 5646 /* seq_file implementation /proc/mdstat */ 5647 5648 static void status_unused(struct seq_file *seq) 5649 { 5650 int i = 0; 5651 mdk_rdev_t *rdev; 5652 5653 seq_printf(seq, "unused devices: "); 5654 5655 list_for_each_entry(rdev, &pending_raid_disks, same_set) { 5656 char b[BDEVNAME_SIZE]; 5657 i++; 5658 seq_printf(seq, "%s ", 5659 bdevname(rdev->bdev,b)); 5660 } 5661 if (!i) 5662 seq_printf(seq, "<none>"); 5663 5664 seq_printf(seq, "\n"); 5665 } 5666 5667 5668 static void status_resync(struct seq_file *seq, mddev_t * mddev) 5669 { 5670 sector_t max_sectors, resync, res; 5671 unsigned long dt, db; 5672 sector_t rt; 5673 int scale; 5674 unsigned int per_milli; 5675 5676 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); 5677 5678 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 5679 max_sectors = mddev->resync_max_sectors; 5680 else 5681 max_sectors = mddev->dev_sectors; 5682 5683 /* 5684 * Should not happen. 5685 */ 5686 if (!max_sectors) { 5687 MD_BUG(); 5688 return; 5689 } 5690 /* Pick 'scale' such that (resync>>scale)*1000 will fit 5691 * in a sector_t, and (max_sectors>>scale) will fit in a 5692 * u32, as those are the requirements for sector_div. 5693 * Thus 'scale' must be at least 10 5694 */ 5695 scale = 10; 5696 if (sizeof(sector_t) > sizeof(unsigned long)) { 5697 while ( max_sectors/2 > (1ULL<<(scale+32))) 5698 scale++; 5699 } 5700 res = (resync>>scale)*1000; 5701 sector_div(res, (u32)((max_sectors>>scale)+1)); 5702 5703 per_milli = res; 5704 { 5705 int i, x = per_milli/50, y = 20-x; 5706 seq_printf(seq, "["); 5707 for (i = 0; i < x; i++) 5708 seq_printf(seq, "="); 5709 seq_printf(seq, ">"); 5710 for (i = 0; i < y; i++) 5711 seq_printf(seq, "."); 5712 seq_printf(seq, "] "); 5713 } 5714 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", 5715 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? 5716 "reshape" : 5717 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? 5718 "check" : 5719 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 5720 "resync" : "recovery"))), 5721 per_milli/10, per_milli % 10, 5722 (unsigned long long) resync/2, 5723 (unsigned long long) max_sectors/2); 5724 5725 /* 5726 * dt: time from mark until now 5727 * db: blocks written from mark until now 5728 * rt: remaining time 5729 * 5730 * rt is a sector_t, so could be 32bit or 64bit. 5731 * So we divide before multiply in case it is 32bit and close 5732 * to the limit. 5733 * We scale the divisor (db) by 32 to avoid loosing precision 5734 * near the end of resync when the number of remaining sectors 5735 * is close to 'db'. 5736 * We then divide rt by 32 after multiplying by db to compensate. 5737 * The '+1' avoids division by zero if db is very small. 5738 */ 5739 dt = ((jiffies - mddev->resync_mark) / HZ); 5740 if (!dt) dt++; 5741 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) 5742 - mddev->resync_mark_cnt; 5743 5744 rt = max_sectors - resync; /* number of remaining sectors */ 5745 sector_div(rt, db/32+1); 5746 rt *= dt; 5747 rt >>= 5; 5748 5749 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, 5750 ((unsigned long)rt % 60)/6); 5751 5752 seq_printf(seq, " speed=%ldK/sec", db/2/dt); 5753 } 5754 5755 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 5756 { 5757 struct list_head *tmp; 5758 loff_t l = *pos; 5759 mddev_t *mddev; 5760 5761 if (l >= 0x10000) 5762 return NULL; 5763 if (!l--) 5764 /* header */ 5765 return (void*)1; 5766 5767 spin_lock(&all_mddevs_lock); 5768 list_for_each(tmp,&all_mddevs) 5769 if (!l--) { 5770 mddev = list_entry(tmp, mddev_t, all_mddevs); 5771 mddev_get(mddev); 5772 spin_unlock(&all_mddevs_lock); 5773 return mddev; 5774 } 5775 spin_unlock(&all_mddevs_lock); 5776 if (!l--) 5777 return (void*)2;/* tail */ 5778 return NULL; 5779 } 5780 5781 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 5782 { 5783 struct list_head *tmp; 5784 mddev_t *next_mddev, *mddev = v; 5785 5786 ++*pos; 5787 if (v == (void*)2) 5788 return NULL; 5789 5790 spin_lock(&all_mddevs_lock); 5791 if (v == (void*)1) 5792 tmp = all_mddevs.next; 5793 else 5794 tmp = mddev->all_mddevs.next; 5795 if (tmp != &all_mddevs) 5796 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); 5797 else { 5798 next_mddev = (void*)2; 5799 *pos = 0x10000; 5800 } 5801 spin_unlock(&all_mddevs_lock); 5802 5803 if (v != (void*)1) 5804 mddev_put(mddev); 5805 return next_mddev; 5806 5807 } 5808 5809 static void md_seq_stop(struct seq_file *seq, void *v) 5810 { 5811 mddev_t *mddev = v; 5812 5813 if (mddev && v != (void*)1 && v != (void*)2) 5814 mddev_put(mddev); 5815 } 5816 5817 struct mdstat_info { 5818 int event; 5819 }; 5820 5821 static int md_seq_show(struct seq_file *seq, void *v) 5822 { 5823 mddev_t *mddev = v; 5824 sector_t sectors; 5825 mdk_rdev_t *rdev; 5826 struct mdstat_info *mi = seq->private; 5827 struct bitmap *bitmap; 5828 5829 if (v == (void*)1) { 5830 struct mdk_personality *pers; 5831 seq_printf(seq, "Personalities : "); 5832 spin_lock(&pers_lock); 5833 list_for_each_entry(pers, &pers_list, list) 5834 seq_printf(seq, "[%s] ", pers->name); 5835 5836 spin_unlock(&pers_lock); 5837 seq_printf(seq, "\n"); 5838 mi->event = atomic_read(&md_event_count); 5839 return 0; 5840 } 5841 if (v == (void*)2) { 5842 status_unused(seq); 5843 return 0; 5844 } 5845 5846 if (mddev_lock(mddev) < 0) 5847 return -EINTR; 5848 5849 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 5850 seq_printf(seq, "%s : %sactive", mdname(mddev), 5851 mddev->pers ? "" : "in"); 5852 if (mddev->pers) { 5853 if (mddev->ro==1) 5854 seq_printf(seq, " (read-only)"); 5855 if (mddev->ro==2) 5856 seq_printf(seq, " (auto-read-only)"); 5857 seq_printf(seq, " %s", mddev->pers->name); 5858 } 5859 5860 sectors = 0; 5861 list_for_each_entry(rdev, &mddev->disks, same_set) { 5862 char b[BDEVNAME_SIZE]; 5863 seq_printf(seq, " %s[%d]", 5864 bdevname(rdev->bdev,b), rdev->desc_nr); 5865 if (test_bit(WriteMostly, &rdev->flags)) 5866 seq_printf(seq, "(W)"); 5867 if (test_bit(Faulty, &rdev->flags)) { 5868 seq_printf(seq, "(F)"); 5869 continue; 5870 } else if (rdev->raid_disk < 0) 5871 seq_printf(seq, "(S)"); /* spare */ 5872 sectors += rdev->sectors; 5873 } 5874 5875 if (!list_empty(&mddev->disks)) { 5876 if (mddev->pers) 5877 seq_printf(seq, "\n %llu blocks", 5878 (unsigned long long) 5879 mddev->array_sectors / 2); 5880 else 5881 seq_printf(seq, "\n %llu blocks", 5882 (unsigned long long)sectors / 2); 5883 } 5884 if (mddev->persistent) { 5885 if (mddev->major_version != 0 || 5886 mddev->minor_version != 90) { 5887 seq_printf(seq," super %d.%d", 5888 mddev->major_version, 5889 mddev->minor_version); 5890 } 5891 } else if (mddev->external) 5892 seq_printf(seq, " super external:%s", 5893 mddev->metadata_type); 5894 else 5895 seq_printf(seq, " super non-persistent"); 5896 5897 if (mddev->pers) { 5898 mddev->pers->status(seq, mddev); 5899 seq_printf(seq, "\n "); 5900 if (mddev->pers->sync_request) { 5901 if (mddev->curr_resync > 2) { 5902 status_resync(seq, mddev); 5903 seq_printf(seq, "\n "); 5904 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 5905 seq_printf(seq, "\tresync=DELAYED\n "); 5906 else if (mddev->recovery_cp < MaxSector) 5907 seq_printf(seq, "\tresync=PENDING\n "); 5908 } 5909 } else 5910 seq_printf(seq, "\n "); 5911 5912 if ((bitmap = mddev->bitmap)) { 5913 unsigned long chunk_kb; 5914 unsigned long flags; 5915 spin_lock_irqsave(&bitmap->lock, flags); 5916 chunk_kb = bitmap->chunksize >> 10; 5917 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 5918 "%lu%s chunk", 5919 bitmap->pages - bitmap->missing_pages, 5920 bitmap->pages, 5921 (bitmap->pages - bitmap->missing_pages) 5922 << (PAGE_SHIFT - 10), 5923 chunk_kb ? chunk_kb : bitmap->chunksize, 5924 chunk_kb ? "KB" : "B"); 5925 if (bitmap->file) { 5926 seq_printf(seq, ", file: "); 5927 seq_path(seq, &bitmap->file->f_path, " \t\n"); 5928 } 5929 5930 seq_printf(seq, "\n"); 5931 spin_unlock_irqrestore(&bitmap->lock, flags); 5932 } 5933 5934 seq_printf(seq, "\n"); 5935 } 5936 mddev_unlock(mddev); 5937 5938 return 0; 5939 } 5940 5941 static const struct seq_operations md_seq_ops = { 5942 .start = md_seq_start, 5943 .next = md_seq_next, 5944 .stop = md_seq_stop, 5945 .show = md_seq_show, 5946 }; 5947 5948 static int md_seq_open(struct inode *inode, struct file *file) 5949 { 5950 int error; 5951 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); 5952 if (mi == NULL) 5953 return -ENOMEM; 5954 5955 error = seq_open(file, &md_seq_ops); 5956 if (error) 5957 kfree(mi); 5958 else { 5959 struct seq_file *p = file->private_data; 5960 p->private = mi; 5961 mi->event = atomic_read(&md_event_count); 5962 } 5963 return error; 5964 } 5965 5966 static unsigned int mdstat_poll(struct file *filp, poll_table *wait) 5967 { 5968 struct seq_file *m = filp->private_data; 5969 struct mdstat_info *mi = m->private; 5970 int mask; 5971 5972 poll_wait(filp, &md_event_waiters, wait); 5973 5974 /* always allow read */ 5975 mask = POLLIN | POLLRDNORM; 5976 5977 if (mi->event != atomic_read(&md_event_count)) 5978 mask |= POLLERR | POLLPRI; 5979 return mask; 5980 } 5981 5982 static const struct file_operations md_seq_fops = { 5983 .owner = THIS_MODULE, 5984 .open = md_seq_open, 5985 .read = seq_read, 5986 .llseek = seq_lseek, 5987 .release = seq_release_private, 5988 .poll = mdstat_poll, 5989 }; 5990 5991 int register_md_personality(struct mdk_personality *p) 5992 { 5993 spin_lock(&pers_lock); 5994 list_add_tail(&p->list, &pers_list); 5995 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); 5996 spin_unlock(&pers_lock); 5997 return 0; 5998 } 5999 6000 int unregister_md_personality(struct mdk_personality *p) 6001 { 6002 printk(KERN_INFO "md: %s personality unregistered\n", p->name); 6003 spin_lock(&pers_lock); 6004 list_del_init(&p->list); 6005 spin_unlock(&pers_lock); 6006 return 0; 6007 } 6008 6009 static int is_mddev_idle(mddev_t *mddev, int init) 6010 { 6011 mdk_rdev_t * rdev; 6012 int idle; 6013 int curr_events; 6014 6015 idle = 1; 6016 rcu_read_lock(); 6017 rdev_for_each_rcu(rdev, mddev) { 6018 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 6019 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 6020 (int)part_stat_read(&disk->part0, sectors[1]) - 6021 atomic_read(&disk->sync_io); 6022 /* sync IO will cause sync_io to increase before the disk_stats 6023 * as sync_io is counted when a request starts, and 6024 * disk_stats is counted when it completes. 6025 * So resync activity will cause curr_events to be smaller than 6026 * when there was no such activity. 6027 * non-sync IO will cause disk_stat to increase without 6028 * increasing sync_io so curr_events will (eventually) 6029 * be larger than it was before. Once it becomes 6030 * substantially larger, the test below will cause 6031 * the array to appear non-idle, and resync will slow 6032 * down. 6033 * If there is a lot of outstanding resync activity when 6034 * we set last_event to curr_events, then all that activity 6035 * completing might cause the array to appear non-idle 6036 * and resync will be slowed down even though there might 6037 * not have been non-resync activity. This will only 6038 * happen once though. 'last_events' will soon reflect 6039 * the state where there is little or no outstanding 6040 * resync requests, and further resync activity will 6041 * always make curr_events less than last_events. 6042 * 6043 */ 6044 if (init || curr_events - rdev->last_events > 64) { 6045 rdev->last_events = curr_events; 6046 idle = 0; 6047 } 6048 } 6049 rcu_read_unlock(); 6050 return idle; 6051 } 6052 6053 void md_done_sync(mddev_t *mddev, int blocks, int ok) 6054 { 6055 /* another "blocks" (512byte) blocks have been synced */ 6056 atomic_sub(blocks, &mddev->recovery_active); 6057 wake_up(&mddev->recovery_wait); 6058 if (!ok) { 6059 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6060 md_wakeup_thread(mddev->thread); 6061 // stop recovery, signal do_sync .... 6062 } 6063 } 6064 6065 6066 /* md_write_start(mddev, bi) 6067 * If we need to update some array metadata (e.g. 'active' flag 6068 * in superblock) before writing, schedule a superblock update 6069 * and wait for it to complete. 6070 */ 6071 void md_write_start(mddev_t *mddev, struct bio *bi) 6072 { 6073 int did_change = 0; 6074 if (bio_data_dir(bi) != WRITE) 6075 return; 6076 6077 BUG_ON(mddev->ro == 1); 6078 if (mddev->ro == 2) { 6079 /* need to switch to read/write */ 6080 mddev->ro = 0; 6081 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6082 md_wakeup_thread(mddev->thread); 6083 md_wakeup_thread(mddev->sync_thread); 6084 did_change = 1; 6085 } 6086 atomic_inc(&mddev->writes_pending); 6087 if (mddev->safemode == 1) 6088 mddev->safemode = 0; 6089 if (mddev->in_sync) { 6090 spin_lock_irq(&mddev->write_lock); 6091 if (mddev->in_sync) { 6092 mddev->in_sync = 0; 6093 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6094 md_wakeup_thread(mddev->thread); 6095 did_change = 1; 6096 } 6097 spin_unlock_irq(&mddev->write_lock); 6098 } 6099 if (did_change) 6100 sysfs_notify_dirent(mddev->sysfs_state); 6101 wait_event(mddev->sb_wait, 6102 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && 6103 !test_bit(MD_CHANGE_PENDING, &mddev->flags)); 6104 } 6105 6106 void md_write_end(mddev_t *mddev) 6107 { 6108 if (atomic_dec_and_test(&mddev->writes_pending)) { 6109 if (mddev->safemode == 2) 6110 md_wakeup_thread(mddev->thread); 6111 else if (mddev->safemode_delay) 6112 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 6113 } 6114 } 6115 6116 /* md_allow_write(mddev) 6117 * Calling this ensures that the array is marked 'active' so that writes 6118 * may proceed without blocking. It is important to call this before 6119 * attempting a GFP_KERNEL allocation while holding the mddev lock. 6120 * Must be called with mddev_lock held. 6121 * 6122 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock 6123 * is dropped, so return -EAGAIN after notifying userspace. 6124 */ 6125 int md_allow_write(mddev_t *mddev) 6126 { 6127 if (!mddev->pers) 6128 return 0; 6129 if (mddev->ro) 6130 return 0; 6131 if (!mddev->pers->sync_request) 6132 return 0; 6133 6134 spin_lock_irq(&mddev->write_lock); 6135 if (mddev->in_sync) { 6136 mddev->in_sync = 0; 6137 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6138 if (mddev->safemode_delay && 6139 mddev->safemode == 0) 6140 mddev->safemode = 1; 6141 spin_unlock_irq(&mddev->write_lock); 6142 md_update_sb(mddev, 0); 6143 sysfs_notify_dirent(mddev->sysfs_state); 6144 } else 6145 spin_unlock_irq(&mddev->write_lock); 6146 6147 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) 6148 return -EAGAIN; 6149 else 6150 return 0; 6151 } 6152 EXPORT_SYMBOL_GPL(md_allow_write); 6153 6154 #define SYNC_MARKS 10 6155 #define SYNC_MARK_STEP (3*HZ) 6156 void md_do_sync(mddev_t *mddev) 6157 { 6158 mddev_t *mddev2; 6159 unsigned int currspeed = 0, 6160 window; 6161 sector_t max_sectors,j, io_sectors; 6162 unsigned long mark[SYNC_MARKS]; 6163 sector_t mark_cnt[SYNC_MARKS]; 6164 int last_mark,m; 6165 struct list_head *tmp; 6166 sector_t last_check; 6167 int skipped = 0; 6168 mdk_rdev_t *rdev; 6169 char *desc; 6170 6171 /* just incase thread restarts... */ 6172 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 6173 return; 6174 if (mddev->ro) /* never try to sync a read-only array */ 6175 return; 6176 6177 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6178 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 6179 desc = "data-check"; 6180 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6181 desc = "requested-resync"; 6182 else 6183 desc = "resync"; 6184 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6185 desc = "reshape"; 6186 else 6187 desc = "recovery"; 6188 6189 /* we overload curr_resync somewhat here. 6190 * 0 == not engaged in resync at all 6191 * 2 == checking that there is no conflict with another sync 6192 * 1 == like 2, but have yielded to allow conflicting resync to 6193 * commense 6194 * other == active in resync - this many blocks 6195 * 6196 * Before starting a resync we must have set curr_resync to 6197 * 2, and then checked that every "conflicting" array has curr_resync 6198 * less than ours. When we find one that is the same or higher 6199 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 6200 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 6201 * This will mean we have to start checking from the beginning again. 6202 * 6203 */ 6204 6205 do { 6206 mddev->curr_resync = 2; 6207 6208 try_again: 6209 if (kthread_should_stop()) { 6210 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6211 goto skip; 6212 } 6213 for_each_mddev(mddev2, tmp) { 6214 if (mddev2 == mddev) 6215 continue; 6216 if (!mddev->parallel_resync 6217 && mddev2->curr_resync 6218 && match_mddev_units(mddev, mddev2)) { 6219 DEFINE_WAIT(wq); 6220 if (mddev < mddev2 && mddev->curr_resync == 2) { 6221 /* arbitrarily yield */ 6222 mddev->curr_resync = 1; 6223 wake_up(&resync_wait); 6224 } 6225 if (mddev > mddev2 && mddev->curr_resync == 1) 6226 /* no need to wait here, we can wait the next 6227 * time 'round when curr_resync == 2 6228 */ 6229 continue; 6230 /* We need to wait 'interruptible' so as not to 6231 * contribute to the load average, and not to 6232 * be caught by 'softlockup' 6233 */ 6234 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 6235 if (!kthread_should_stop() && 6236 mddev2->curr_resync >= mddev->curr_resync) { 6237 printk(KERN_INFO "md: delaying %s of %s" 6238 " until %s has finished (they" 6239 " share one or more physical units)\n", 6240 desc, mdname(mddev), mdname(mddev2)); 6241 mddev_put(mddev2); 6242 if (signal_pending(current)) 6243 flush_signals(current); 6244 schedule(); 6245 finish_wait(&resync_wait, &wq); 6246 goto try_again; 6247 } 6248 finish_wait(&resync_wait, &wq); 6249 } 6250 } 6251 } while (mddev->curr_resync < 2); 6252 6253 j = 0; 6254 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6255 /* resync follows the size requested by the personality, 6256 * which defaults to physical size, but can be virtual size 6257 */ 6258 max_sectors = mddev->resync_max_sectors; 6259 mddev->resync_mismatches = 0; 6260 /* we don't use the checkpoint if there's a bitmap */ 6261 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6262 j = mddev->resync_min; 6263 else if (!mddev->bitmap) 6264 j = mddev->recovery_cp; 6265 6266 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6267 max_sectors = mddev->dev_sectors; 6268 else { 6269 /* recovery follows the physical size of devices */ 6270 max_sectors = mddev->dev_sectors; 6271 j = MaxSector; 6272 list_for_each_entry(rdev, &mddev->disks, same_set) 6273 if (rdev->raid_disk >= 0 && 6274 !test_bit(Faulty, &rdev->flags) && 6275 !test_bit(In_sync, &rdev->flags) && 6276 rdev->recovery_offset < j) 6277 j = rdev->recovery_offset; 6278 } 6279 6280 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); 6281 printk(KERN_INFO "md: minimum _guaranteed_ speed:" 6282 " %d KB/sec/disk.\n", speed_min(mddev)); 6283 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 6284 "(but not more than %d KB/sec) for %s.\n", 6285 speed_max(mddev), desc); 6286 6287 is_mddev_idle(mddev, 1); /* this initializes IO event counters */ 6288 6289 io_sectors = 0; 6290 for (m = 0; m < SYNC_MARKS; m++) { 6291 mark[m] = jiffies; 6292 mark_cnt[m] = io_sectors; 6293 } 6294 last_mark = 0; 6295 mddev->resync_mark = mark[last_mark]; 6296 mddev->resync_mark_cnt = mark_cnt[last_mark]; 6297 6298 /* 6299 * Tune reconstruction: 6300 */ 6301 window = 32*(PAGE_SIZE/512); 6302 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", 6303 window/2,(unsigned long long) max_sectors/2); 6304 6305 atomic_set(&mddev->recovery_active, 0); 6306 last_check = 0; 6307 6308 if (j>2) { 6309 printk(KERN_INFO 6310 "md: resuming %s of %s from checkpoint.\n", 6311 desc, mdname(mddev)); 6312 mddev->curr_resync = j; 6313 } 6314 6315 while (j < max_sectors) { 6316 sector_t sectors; 6317 6318 skipped = 0; 6319 6320 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 6321 ((mddev->curr_resync > mddev->curr_resync_completed && 6322 (mddev->curr_resync - mddev->curr_resync_completed) 6323 > (max_sectors >> 4)) || 6324 (j - mddev->curr_resync_completed)*2 6325 >= mddev->resync_max - mddev->curr_resync_completed 6326 )) { 6327 /* time to update curr_resync_completed */ 6328 blk_unplug(mddev->queue); 6329 wait_event(mddev->recovery_wait, 6330 atomic_read(&mddev->recovery_active) == 0); 6331 mddev->curr_resync_completed = 6332 mddev->curr_resync; 6333 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6334 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 6335 } 6336 6337 if (j >= mddev->resync_max) 6338 wait_event(mddev->recovery_wait, 6339 mddev->resync_max > j 6340 || kthread_should_stop()); 6341 6342 if (kthread_should_stop()) 6343 goto interrupted; 6344 6345 sectors = mddev->pers->sync_request(mddev, j, &skipped, 6346 currspeed < speed_min(mddev)); 6347 if (sectors == 0) { 6348 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6349 goto out; 6350 } 6351 6352 if (!skipped) { /* actual IO requested */ 6353 io_sectors += sectors; 6354 atomic_add(sectors, &mddev->recovery_active); 6355 } 6356 6357 j += sectors; 6358 if (j>1) mddev->curr_resync = j; 6359 mddev->curr_mark_cnt = io_sectors; 6360 if (last_check == 0) 6361 /* this is the earliers that rebuilt will be 6362 * visible in /proc/mdstat 6363 */ 6364 md_new_event(mddev); 6365 6366 if (last_check + window > io_sectors || j == max_sectors) 6367 continue; 6368 6369 last_check = io_sectors; 6370 6371 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6372 break; 6373 6374 repeat: 6375 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 6376 /* step marks */ 6377 int next = (last_mark+1) % SYNC_MARKS; 6378 6379 mddev->resync_mark = mark[next]; 6380 mddev->resync_mark_cnt = mark_cnt[next]; 6381 mark[next] = jiffies; 6382 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 6383 last_mark = next; 6384 } 6385 6386 6387 if (kthread_should_stop()) 6388 goto interrupted; 6389 6390 6391 /* 6392 * this loop exits only if either when we are slower than 6393 * the 'hard' speed limit, or the system was IO-idle for 6394 * a jiffy. 6395 * the system might be non-idle CPU-wise, but we only care 6396 * about not overloading the IO subsystem. (things like an 6397 * e2fsck being done on the RAID array should execute fast) 6398 */ 6399 blk_unplug(mddev->queue); 6400 cond_resched(); 6401 6402 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 6403 /((jiffies-mddev->resync_mark)/HZ +1) +1; 6404 6405 if (currspeed > speed_min(mddev)) { 6406 if ((currspeed > speed_max(mddev)) || 6407 !is_mddev_idle(mddev, 0)) { 6408 msleep(500); 6409 goto repeat; 6410 } 6411 } 6412 } 6413 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); 6414 /* 6415 * this also signals 'finished resyncing' to md_stop 6416 */ 6417 out: 6418 blk_unplug(mddev->queue); 6419 6420 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 6421 6422 /* tell personality that we are finished */ 6423 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 6424 6425 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && 6426 mddev->curr_resync > 2) { 6427 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6428 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 6429 if (mddev->curr_resync >= mddev->recovery_cp) { 6430 printk(KERN_INFO 6431 "md: checkpointing %s of %s.\n", 6432 desc, mdname(mddev)); 6433 mddev->recovery_cp = mddev->curr_resync; 6434 } 6435 } else 6436 mddev->recovery_cp = MaxSector; 6437 } else { 6438 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6439 mddev->curr_resync = MaxSector; 6440 list_for_each_entry(rdev, &mddev->disks, same_set) 6441 if (rdev->raid_disk >= 0 && 6442 !test_bit(Faulty, &rdev->flags) && 6443 !test_bit(In_sync, &rdev->flags) && 6444 rdev->recovery_offset < mddev->curr_resync) 6445 rdev->recovery_offset = mddev->curr_resync; 6446 } 6447 } 6448 set_bit(MD_CHANGE_DEVS, &mddev->flags); 6449 6450 skip: 6451 mddev->curr_resync = 0; 6452 mddev->curr_resync_completed = 0; 6453 mddev->resync_min = 0; 6454 mddev->resync_max = MaxSector; 6455 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 6456 wake_up(&resync_wait); 6457 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 6458 md_wakeup_thread(mddev->thread); 6459 return; 6460 6461 interrupted: 6462 /* 6463 * got a signal, exit. 6464 */ 6465 printk(KERN_INFO 6466 "md: md_do_sync() got signal ... exiting\n"); 6467 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6468 goto out; 6469 6470 } 6471 EXPORT_SYMBOL_GPL(md_do_sync); 6472 6473 6474 static int remove_and_add_spares(mddev_t *mddev) 6475 { 6476 mdk_rdev_t *rdev; 6477 int spares = 0; 6478 6479 mddev->curr_resync_completed = 0; 6480 6481 list_for_each_entry(rdev, &mddev->disks, same_set) 6482 if (rdev->raid_disk >= 0 && 6483 !test_bit(Blocked, &rdev->flags) && 6484 (test_bit(Faulty, &rdev->flags) || 6485 ! test_bit(In_sync, &rdev->flags)) && 6486 atomic_read(&rdev->nr_pending)==0) { 6487 if (mddev->pers->hot_remove_disk( 6488 mddev, rdev->raid_disk)==0) { 6489 char nm[20]; 6490 sprintf(nm,"rd%d", rdev->raid_disk); 6491 sysfs_remove_link(&mddev->kobj, nm); 6492 rdev->raid_disk = -1; 6493 } 6494 } 6495 6496 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { 6497 list_for_each_entry(rdev, &mddev->disks, same_set) { 6498 if (rdev->raid_disk >= 0 && 6499 !test_bit(In_sync, &rdev->flags) && 6500 !test_bit(Blocked, &rdev->flags)) 6501 spares++; 6502 if (rdev->raid_disk < 0 6503 && !test_bit(Faulty, &rdev->flags)) { 6504 rdev->recovery_offset = 0; 6505 if (mddev->pers-> 6506 hot_add_disk(mddev, rdev) == 0) { 6507 char nm[20]; 6508 sprintf(nm, "rd%d", rdev->raid_disk); 6509 if (sysfs_create_link(&mddev->kobj, 6510 &rdev->kobj, nm)) 6511 printk(KERN_WARNING 6512 "md: cannot register " 6513 "%s for %s\n", 6514 nm, mdname(mddev)); 6515 spares++; 6516 md_new_event(mddev); 6517 } else 6518 break; 6519 } 6520 } 6521 } 6522 return spares; 6523 } 6524 /* 6525 * This routine is regularly called by all per-raid-array threads to 6526 * deal with generic issues like resync and super-block update. 6527 * Raid personalities that don't have a thread (linear/raid0) do not 6528 * need this as they never do any recovery or update the superblock. 6529 * 6530 * It does not do any resync itself, but rather "forks" off other threads 6531 * to do that as needed. 6532 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 6533 * "->recovery" and create a thread at ->sync_thread. 6534 * When the thread finishes it sets MD_RECOVERY_DONE 6535 * and wakeups up this thread which will reap the thread and finish up. 6536 * This thread also removes any faulty devices (with nr_pending == 0). 6537 * 6538 * The overall approach is: 6539 * 1/ if the superblock needs updating, update it. 6540 * 2/ If a recovery thread is running, don't do anything else. 6541 * 3/ If recovery has finished, clean up, possibly marking spares active. 6542 * 4/ If there are any faulty devices, remove them. 6543 * 5/ If array is degraded, try to add spares devices 6544 * 6/ If array has spares or is not in-sync, start a resync thread. 6545 */ 6546 void md_check_recovery(mddev_t *mddev) 6547 { 6548 mdk_rdev_t *rdev; 6549 6550 6551 if (mddev->bitmap) 6552 bitmap_daemon_work(mddev->bitmap); 6553 6554 if (mddev->ro) 6555 return; 6556 6557 if (signal_pending(current)) { 6558 if (mddev->pers->sync_request && !mddev->external) { 6559 printk(KERN_INFO "md: %s in immediate safe mode\n", 6560 mdname(mddev)); 6561 mddev->safemode = 2; 6562 } 6563 flush_signals(current); 6564 } 6565 6566 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 6567 return; 6568 if ( ! ( 6569 (mddev->flags && !mddev->external) || 6570 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 6571 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 6572 (mddev->external == 0 && mddev->safemode == 1) || 6573 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 6574 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 6575 )) 6576 return; 6577 6578 if (mddev_trylock(mddev)) { 6579 int spares = 0; 6580 6581 if (mddev->ro) { 6582 /* Only thing we do on a ro array is remove 6583 * failed devices. 6584 */ 6585 remove_and_add_spares(mddev); 6586 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6587 goto unlock; 6588 } 6589 6590 if (!mddev->external) { 6591 int did_change = 0; 6592 spin_lock_irq(&mddev->write_lock); 6593 if (mddev->safemode && 6594 !atomic_read(&mddev->writes_pending) && 6595 !mddev->in_sync && 6596 mddev->recovery_cp == MaxSector) { 6597 mddev->in_sync = 1; 6598 did_change = 1; 6599 if (mddev->persistent) 6600 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6601 } 6602 if (mddev->safemode == 1) 6603 mddev->safemode = 0; 6604 spin_unlock_irq(&mddev->write_lock); 6605 if (did_change) 6606 sysfs_notify_dirent(mddev->sysfs_state); 6607 } 6608 6609 if (mddev->flags) 6610 md_update_sb(mddev, 0); 6611 6612 list_for_each_entry(rdev, &mddev->disks, same_set) 6613 if (test_and_clear_bit(StateChanged, &rdev->flags)) 6614 sysfs_notify_dirent(rdev->sysfs_state); 6615 6616 6617 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 6618 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 6619 /* resync/recovery still happening */ 6620 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6621 goto unlock; 6622 } 6623 if (mddev->sync_thread) { 6624 /* resync has finished, collect result */ 6625 md_unregister_thread(mddev->sync_thread); 6626 mddev->sync_thread = NULL; 6627 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && 6628 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 6629 /* success...*/ 6630 /* activate any spares */ 6631 if (mddev->pers->spare_active(mddev)) 6632 sysfs_notify(&mddev->kobj, NULL, 6633 "degraded"); 6634 } 6635 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 6636 mddev->pers->finish_reshape) 6637 mddev->pers->finish_reshape(mddev); 6638 md_update_sb(mddev, 1); 6639 6640 /* if array is no-longer degraded, then any saved_raid_disk 6641 * information must be scrapped 6642 */ 6643 if (!mddev->degraded) 6644 list_for_each_entry(rdev, &mddev->disks, same_set) 6645 rdev->saved_raid_disk = -1; 6646 6647 mddev->recovery = 0; 6648 /* flag recovery needed just to double check */ 6649 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6650 sysfs_notify_dirent(mddev->sysfs_action); 6651 md_new_event(mddev); 6652 goto unlock; 6653 } 6654 /* Set RUNNING before clearing NEEDED to avoid 6655 * any transients in the value of "sync_action". 6656 */ 6657 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 6658 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6659 /* Clear some bits that don't mean anything, but 6660 * might be left set 6661 */ 6662 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 6663 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 6664 6665 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 6666 goto unlock; 6667 /* no recovery is running. 6668 * remove any failed drives, then 6669 * add spares if possible. 6670 * Spare are also removed and re-added, to allow 6671 * the personality to fail the re-add. 6672 */ 6673 6674 if (mddev->reshape_position != MaxSector) { 6675 if (mddev->pers->check_reshape == NULL || 6676 mddev->pers->check_reshape(mddev) != 0) 6677 /* Cannot proceed */ 6678 goto unlock; 6679 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 6680 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6681 } else if ((spares = remove_and_add_spares(mddev))) { 6682 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 6683 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 6684 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 6685 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6686 } else if (mddev->recovery_cp < MaxSector) { 6687 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 6688 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6689 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 6690 /* nothing to be done ... */ 6691 goto unlock; 6692 6693 if (mddev->pers->sync_request) { 6694 if (spares && mddev->bitmap && ! mddev->bitmap->file) { 6695 /* We are adding a device or devices to an array 6696 * which has the bitmap stored on all devices. 6697 * So make sure all bitmap pages get written 6698 */ 6699 bitmap_write_all(mddev->bitmap); 6700 } 6701 mddev->sync_thread = md_register_thread(md_do_sync, 6702 mddev, 6703 "%s_resync"); 6704 if (!mddev->sync_thread) { 6705 printk(KERN_ERR "%s: could not start resync" 6706 " thread...\n", 6707 mdname(mddev)); 6708 /* leave the spares where they are, it shouldn't hurt */ 6709 mddev->recovery = 0; 6710 } else 6711 md_wakeup_thread(mddev->sync_thread); 6712 sysfs_notify_dirent(mddev->sysfs_action); 6713 md_new_event(mddev); 6714 } 6715 unlock: 6716 if (!mddev->sync_thread) { 6717 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 6718 if (test_and_clear_bit(MD_RECOVERY_RECOVER, 6719 &mddev->recovery)) 6720 if (mddev->sysfs_action) 6721 sysfs_notify_dirent(mddev->sysfs_action); 6722 } 6723 mddev_unlock(mddev); 6724 } 6725 } 6726 6727 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 6728 { 6729 sysfs_notify_dirent(rdev->sysfs_state); 6730 wait_event_timeout(rdev->blocked_wait, 6731 !test_bit(Blocked, &rdev->flags), 6732 msecs_to_jiffies(5000)); 6733 rdev_dec_pending(rdev, mddev); 6734 } 6735 EXPORT_SYMBOL(md_wait_for_blocked_rdev); 6736 6737 static int md_notify_reboot(struct notifier_block *this, 6738 unsigned long code, void *x) 6739 { 6740 struct list_head *tmp; 6741 mddev_t *mddev; 6742 6743 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { 6744 6745 printk(KERN_INFO "md: stopping all md devices.\n"); 6746 6747 for_each_mddev(mddev, tmp) 6748 if (mddev_trylock(mddev)) { 6749 /* Force a switch to readonly even array 6750 * appears to still be in use. Hence 6751 * the '100'. 6752 */ 6753 do_md_stop(mddev, 1, 100); 6754 mddev_unlock(mddev); 6755 } 6756 /* 6757 * certain more exotic SCSI devices are known to be 6758 * volatile wrt too early system reboots. While the 6759 * right place to handle this issue is the given 6760 * driver, we do want to have a safe RAID driver ... 6761 */ 6762 mdelay(1000*1); 6763 } 6764 return NOTIFY_DONE; 6765 } 6766 6767 static struct notifier_block md_notifier = { 6768 .notifier_call = md_notify_reboot, 6769 .next = NULL, 6770 .priority = INT_MAX, /* before any real devices */ 6771 }; 6772 6773 static void md_geninit(void) 6774 { 6775 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 6776 6777 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); 6778 } 6779 6780 static int __init md_init(void) 6781 { 6782 if (register_blkdev(MD_MAJOR, "md")) 6783 return -1; 6784 if ((mdp_major=register_blkdev(0, "mdp"))<=0) { 6785 unregister_blkdev(MD_MAJOR, "md"); 6786 return -1; 6787 } 6788 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, 6789 md_probe, NULL, NULL); 6790 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, 6791 md_probe, NULL, NULL); 6792 6793 register_reboot_notifier(&md_notifier); 6794 raid_table_header = register_sysctl_table(raid_root_table); 6795 6796 md_geninit(); 6797 return 0; 6798 } 6799 6800 6801 #ifndef MODULE 6802 6803 /* 6804 * Searches all registered partitions for autorun RAID arrays 6805 * at boot time. 6806 */ 6807 6808 static LIST_HEAD(all_detected_devices); 6809 struct detected_devices_node { 6810 struct list_head list; 6811 dev_t dev; 6812 }; 6813 6814 void md_autodetect_dev(dev_t dev) 6815 { 6816 struct detected_devices_node *node_detected_dev; 6817 6818 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); 6819 if (node_detected_dev) { 6820 node_detected_dev->dev = dev; 6821 list_add_tail(&node_detected_dev->list, &all_detected_devices); 6822 } else { 6823 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" 6824 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); 6825 } 6826 } 6827 6828 6829 static void autostart_arrays(int part) 6830 { 6831 mdk_rdev_t *rdev; 6832 struct detected_devices_node *node_detected_dev; 6833 dev_t dev; 6834 int i_scanned, i_passed; 6835 6836 i_scanned = 0; 6837 i_passed = 0; 6838 6839 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 6840 6841 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { 6842 i_scanned++; 6843 node_detected_dev = list_entry(all_detected_devices.next, 6844 struct detected_devices_node, list); 6845 list_del(&node_detected_dev->list); 6846 dev = node_detected_dev->dev; 6847 kfree(node_detected_dev); 6848 rdev = md_import_device(dev,0, 90); 6849 if (IS_ERR(rdev)) 6850 continue; 6851 6852 if (test_bit(Faulty, &rdev->flags)) { 6853 MD_BUG(); 6854 continue; 6855 } 6856 set_bit(AutoDetected, &rdev->flags); 6857 list_add(&rdev->same_set, &pending_raid_disks); 6858 i_passed++; 6859 } 6860 6861 printk(KERN_INFO "md: Scanned %d and added %d devices.\n", 6862 i_scanned, i_passed); 6863 6864 autorun_devices(part); 6865 } 6866 6867 #endif /* !MODULE */ 6868 6869 static __exit void md_exit(void) 6870 { 6871 mddev_t *mddev; 6872 struct list_head *tmp; 6873 6874 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); 6875 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); 6876 6877 unregister_blkdev(MD_MAJOR,"md"); 6878 unregister_blkdev(mdp_major, "mdp"); 6879 unregister_reboot_notifier(&md_notifier); 6880 unregister_sysctl_table(raid_table_header); 6881 remove_proc_entry("mdstat", NULL); 6882 for_each_mddev(mddev, tmp) { 6883 export_array(mddev); 6884 mddev->hold_active = 0; 6885 } 6886 } 6887 6888 subsys_initcall(md_init); 6889 module_exit(md_exit) 6890 6891 static int get_ro(char *buffer, struct kernel_param *kp) 6892 { 6893 return sprintf(buffer, "%d", start_readonly); 6894 } 6895 static int set_ro(const char *val, struct kernel_param *kp) 6896 { 6897 char *e; 6898 int num = simple_strtoul(val, &e, 10); 6899 if (*val && (*e == '\0' || *e == '\n')) { 6900 start_readonly = num; 6901 return 0; 6902 } 6903 return -EINVAL; 6904 } 6905 6906 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); 6907 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); 6908 6909 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); 6910 6911 EXPORT_SYMBOL(register_md_personality); 6912 EXPORT_SYMBOL(unregister_md_personality); 6913 EXPORT_SYMBOL(md_error); 6914 EXPORT_SYMBOL(md_done_sync); 6915 EXPORT_SYMBOL(md_write_start); 6916 EXPORT_SYMBOL(md_write_end); 6917 EXPORT_SYMBOL(md_register_thread); 6918 EXPORT_SYMBOL(md_unregister_thread); 6919 EXPORT_SYMBOL(md_wakeup_thread); 6920 EXPORT_SYMBOL(md_check_recovery); 6921 MODULE_LICENSE("GPL"); 6922 MODULE_ALIAS("md"); 6923 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 6924