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/fs.h> 40 #include <linux/poll.h> 41 #include <linux/ctype.h> 42 #include <linux/string.h> 43 #include <linux/hdreg.h> 44 #include <linux/proc_fs.h> 45 #include <linux/random.h> 46 #include <linux/module.h> 47 #include <linux/reboot.h> 48 #include <linux/file.h> 49 #include <linux/compat.h> 50 #include <linux/delay.h> 51 #include <linux/raid/md_p.h> 52 #include <linux/raid/md_u.h> 53 #include <linux/slab.h> 54 #include "md.h" 55 #include "bitmap.h" 56 57 #ifndef MODULE 58 static void autostart_arrays(int part); 59 #endif 60 61 /* pers_list is a list of registered personalities protected 62 * by pers_lock. 63 * pers_lock does extra service to protect accesses to 64 * mddev->thread when the mutex cannot be held. 65 */ 66 static LIST_HEAD(pers_list); 67 static DEFINE_SPINLOCK(pers_lock); 68 69 static void md_print_devices(void); 70 71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait); 72 static struct workqueue_struct *md_wq; 73 static struct workqueue_struct *md_misc_wq; 74 75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } 76 77 /* 78 * Default number of read corrections we'll attempt on an rdev 79 * before ejecting it from the array. We divide the read error 80 * count by 2 for every hour elapsed between read errors. 81 */ 82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 83 /* 84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' 85 * is 1000 KB/sec, so the extra system load does not show up that much. 86 * Increase it if you want to have more _guaranteed_ speed. Note that 87 * the RAID driver will use the maximum available bandwidth if the IO 88 * subsystem is idle. There is also an 'absolute maximum' reconstruction 89 * speed limit - in case reconstruction slows down your system despite 90 * idle IO detection. 91 * 92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. 93 * or /sys/block/mdX/md/sync_speed_{min,max} 94 */ 95 96 static int sysctl_speed_limit_min = 1000; 97 static int sysctl_speed_limit_max = 200000; 98 static inline int speed_min(struct mddev *mddev) 99 { 100 return mddev->sync_speed_min ? 101 mddev->sync_speed_min : sysctl_speed_limit_min; 102 } 103 104 static inline int speed_max(struct mddev *mddev) 105 { 106 return mddev->sync_speed_max ? 107 mddev->sync_speed_max : sysctl_speed_limit_max; 108 } 109 110 static struct ctl_table_header *raid_table_header; 111 112 static ctl_table raid_table[] = { 113 { 114 .procname = "speed_limit_min", 115 .data = &sysctl_speed_limit_min, 116 .maxlen = sizeof(int), 117 .mode = S_IRUGO|S_IWUSR, 118 .proc_handler = proc_dointvec, 119 }, 120 { 121 .procname = "speed_limit_max", 122 .data = &sysctl_speed_limit_max, 123 .maxlen = sizeof(int), 124 .mode = S_IRUGO|S_IWUSR, 125 .proc_handler = proc_dointvec, 126 }, 127 { } 128 }; 129 130 static ctl_table raid_dir_table[] = { 131 { 132 .procname = "raid", 133 .maxlen = 0, 134 .mode = S_IRUGO|S_IXUGO, 135 .child = raid_table, 136 }, 137 { } 138 }; 139 140 static ctl_table raid_root_table[] = { 141 { 142 .procname = "dev", 143 .maxlen = 0, 144 .mode = 0555, 145 .child = raid_dir_table, 146 }, 147 { } 148 }; 149 150 static const struct block_device_operations md_fops; 151 152 static int start_readonly; 153 154 /* bio_clone_mddev 155 * like bio_clone, but with a local bio set 156 */ 157 158 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, 159 struct mddev *mddev) 160 { 161 struct bio *b; 162 163 if (!mddev || !mddev->bio_set) 164 return bio_alloc(gfp_mask, nr_iovecs); 165 166 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); 167 if (!b) 168 return NULL; 169 return b; 170 } 171 EXPORT_SYMBOL_GPL(bio_alloc_mddev); 172 173 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, 174 struct mddev *mddev) 175 { 176 struct bio *b; 177 178 if (!mddev || !mddev->bio_set) 179 return bio_clone(bio, gfp_mask); 180 181 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, mddev->bio_set); 182 if (!b) 183 return NULL; 184 185 __bio_clone(b, bio); 186 if (bio_integrity(bio)) { 187 int ret; 188 189 ret = bio_integrity_clone(b, bio, gfp_mask); 190 191 if (ret < 0) { 192 bio_put(b); 193 return NULL; 194 } 195 } 196 197 return b; 198 } 199 EXPORT_SYMBOL_GPL(bio_clone_mddev); 200 201 void md_trim_bio(struct bio *bio, int offset, int size) 202 { 203 /* 'bio' is a cloned bio which we need to trim to match 204 * the given offset and size. 205 * This requires adjusting bi_sector, bi_size, and bi_io_vec 206 */ 207 int i; 208 struct bio_vec *bvec; 209 int sofar = 0; 210 211 size <<= 9; 212 if (offset == 0 && size == bio->bi_size) 213 return; 214 215 bio->bi_sector += offset; 216 bio->bi_size = size; 217 offset <<= 9; 218 clear_bit(BIO_SEG_VALID, &bio->bi_flags); 219 220 while (bio->bi_idx < bio->bi_vcnt && 221 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) { 222 /* remove this whole bio_vec */ 223 offset -= bio->bi_io_vec[bio->bi_idx].bv_len; 224 bio->bi_idx++; 225 } 226 if (bio->bi_idx < bio->bi_vcnt) { 227 bio->bi_io_vec[bio->bi_idx].bv_offset += offset; 228 bio->bi_io_vec[bio->bi_idx].bv_len -= offset; 229 } 230 /* avoid any complications with bi_idx being non-zero*/ 231 if (bio->bi_idx) { 232 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx, 233 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec)); 234 bio->bi_vcnt -= bio->bi_idx; 235 bio->bi_idx = 0; 236 } 237 /* Make sure vcnt and last bv are not too big */ 238 bio_for_each_segment(bvec, bio, i) { 239 if (sofar + bvec->bv_len > size) 240 bvec->bv_len = size - sofar; 241 if (bvec->bv_len == 0) { 242 bio->bi_vcnt = i; 243 break; 244 } 245 sofar += bvec->bv_len; 246 } 247 } 248 EXPORT_SYMBOL_GPL(md_trim_bio); 249 250 /* 251 * We have a system wide 'event count' that is incremented 252 * on any 'interesting' event, and readers of /proc/mdstat 253 * can use 'poll' or 'select' to find out when the event 254 * count increases. 255 * 256 * Events are: 257 * start array, stop array, error, add device, remove device, 258 * start build, activate spare 259 */ 260 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); 261 static atomic_t md_event_count; 262 void md_new_event(struct mddev *mddev) 263 { 264 atomic_inc(&md_event_count); 265 wake_up(&md_event_waiters); 266 } 267 EXPORT_SYMBOL_GPL(md_new_event); 268 269 /* Alternate version that can be called from interrupts 270 * when calling sysfs_notify isn't needed. 271 */ 272 static void md_new_event_inintr(struct mddev *mddev) 273 { 274 atomic_inc(&md_event_count); 275 wake_up(&md_event_waiters); 276 } 277 278 /* 279 * Enables to iterate over all existing md arrays 280 * all_mddevs_lock protects this list. 281 */ 282 static LIST_HEAD(all_mddevs); 283 static DEFINE_SPINLOCK(all_mddevs_lock); 284 285 286 /* 287 * iterates through all used mddevs in the system. 288 * We take care to grab the all_mddevs_lock whenever navigating 289 * the list, and to always hold a refcount when unlocked. 290 * Any code which breaks out of this loop while own 291 * a reference to the current mddev and must mddev_put it. 292 */ 293 #define for_each_mddev(_mddev,_tmp) \ 294 \ 295 for (({ spin_lock(&all_mddevs_lock); \ 296 _tmp = all_mddevs.next; \ 297 _mddev = NULL;}); \ 298 ({ if (_tmp != &all_mddevs) \ 299 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ 300 spin_unlock(&all_mddevs_lock); \ 301 if (_mddev) mddev_put(_mddev); \ 302 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ 303 _tmp != &all_mddevs;}); \ 304 ({ spin_lock(&all_mddevs_lock); \ 305 _tmp = _tmp->next;}) \ 306 ) 307 308 309 /* Rather than calling directly into the personality make_request function, 310 * IO requests come here first so that we can check if the device is 311 * being suspended pending a reconfiguration. 312 * We hold a refcount over the call to ->make_request. By the time that 313 * call has finished, the bio has been linked into some internal structure 314 * and so is visible to ->quiesce(), so we don't need the refcount any more. 315 */ 316 static void md_make_request(struct request_queue *q, struct bio *bio) 317 { 318 const int rw = bio_data_dir(bio); 319 struct mddev *mddev = q->queuedata; 320 int cpu; 321 unsigned int sectors; 322 323 if (mddev == NULL || mddev->pers == NULL 324 || !mddev->ready) { 325 bio_io_error(bio); 326 return; 327 } 328 smp_rmb(); /* Ensure implications of 'active' are visible */ 329 rcu_read_lock(); 330 if (mddev->suspended) { 331 DEFINE_WAIT(__wait); 332 for (;;) { 333 prepare_to_wait(&mddev->sb_wait, &__wait, 334 TASK_UNINTERRUPTIBLE); 335 if (!mddev->suspended) 336 break; 337 rcu_read_unlock(); 338 schedule(); 339 rcu_read_lock(); 340 } 341 finish_wait(&mddev->sb_wait, &__wait); 342 } 343 atomic_inc(&mddev->active_io); 344 rcu_read_unlock(); 345 346 /* 347 * save the sectors now since our bio can 348 * go away inside make_request 349 */ 350 sectors = bio_sectors(bio); 351 mddev->pers->make_request(mddev, bio); 352 353 cpu = part_stat_lock(); 354 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); 355 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); 356 part_stat_unlock(); 357 358 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) 359 wake_up(&mddev->sb_wait); 360 } 361 362 /* mddev_suspend makes sure no new requests are submitted 363 * to the device, and that any requests that have been submitted 364 * are completely handled. 365 * Once ->stop is called and completes, the module will be completely 366 * unused. 367 */ 368 void mddev_suspend(struct mddev *mddev) 369 { 370 BUG_ON(mddev->suspended); 371 mddev->suspended = 1; 372 synchronize_rcu(); 373 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); 374 mddev->pers->quiesce(mddev, 1); 375 376 del_timer_sync(&mddev->safemode_timer); 377 } 378 EXPORT_SYMBOL_GPL(mddev_suspend); 379 380 void mddev_resume(struct mddev *mddev) 381 { 382 mddev->suspended = 0; 383 wake_up(&mddev->sb_wait); 384 mddev->pers->quiesce(mddev, 0); 385 386 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 387 md_wakeup_thread(mddev->thread); 388 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 389 } 390 EXPORT_SYMBOL_GPL(mddev_resume); 391 392 int mddev_congested(struct mddev *mddev, int bits) 393 { 394 return mddev->suspended; 395 } 396 EXPORT_SYMBOL(mddev_congested); 397 398 /* 399 * Generic flush handling for md 400 */ 401 402 static void md_end_flush(struct bio *bio, int err) 403 { 404 struct md_rdev *rdev = bio->bi_private; 405 struct mddev *mddev = rdev->mddev; 406 407 rdev_dec_pending(rdev, mddev); 408 409 if (atomic_dec_and_test(&mddev->flush_pending)) { 410 /* The pre-request flush has finished */ 411 queue_work(md_wq, &mddev->flush_work); 412 } 413 bio_put(bio); 414 } 415 416 static void md_submit_flush_data(struct work_struct *ws); 417 418 static void submit_flushes(struct work_struct *ws) 419 { 420 struct mddev *mddev = container_of(ws, struct mddev, flush_work); 421 struct md_rdev *rdev; 422 423 INIT_WORK(&mddev->flush_work, md_submit_flush_data); 424 atomic_set(&mddev->flush_pending, 1); 425 rcu_read_lock(); 426 rdev_for_each_rcu(rdev, mddev) 427 if (rdev->raid_disk >= 0 && 428 !test_bit(Faulty, &rdev->flags)) { 429 /* Take two references, one is dropped 430 * when request finishes, one after 431 * we reclaim rcu_read_lock 432 */ 433 struct bio *bi; 434 atomic_inc(&rdev->nr_pending); 435 atomic_inc(&rdev->nr_pending); 436 rcu_read_unlock(); 437 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); 438 bi->bi_end_io = md_end_flush; 439 bi->bi_private = rdev; 440 bi->bi_bdev = rdev->bdev; 441 atomic_inc(&mddev->flush_pending); 442 submit_bio(WRITE_FLUSH, bi); 443 rcu_read_lock(); 444 rdev_dec_pending(rdev, mddev); 445 } 446 rcu_read_unlock(); 447 if (atomic_dec_and_test(&mddev->flush_pending)) 448 queue_work(md_wq, &mddev->flush_work); 449 } 450 451 static void md_submit_flush_data(struct work_struct *ws) 452 { 453 struct mddev *mddev = container_of(ws, struct mddev, flush_work); 454 struct bio *bio = mddev->flush_bio; 455 456 if (bio->bi_size == 0) 457 /* an empty barrier - all done */ 458 bio_endio(bio, 0); 459 else { 460 bio->bi_rw &= ~REQ_FLUSH; 461 mddev->pers->make_request(mddev, bio); 462 } 463 464 mddev->flush_bio = NULL; 465 wake_up(&mddev->sb_wait); 466 } 467 468 void md_flush_request(struct mddev *mddev, struct bio *bio) 469 { 470 spin_lock_irq(&mddev->write_lock); 471 wait_event_lock_irq(mddev->sb_wait, 472 !mddev->flush_bio, 473 mddev->write_lock, /*nothing*/); 474 mddev->flush_bio = bio; 475 spin_unlock_irq(&mddev->write_lock); 476 477 INIT_WORK(&mddev->flush_work, submit_flushes); 478 queue_work(md_wq, &mddev->flush_work); 479 } 480 EXPORT_SYMBOL(md_flush_request); 481 482 void md_unplug(struct blk_plug_cb *cb, bool from_schedule) 483 { 484 struct mddev *mddev = cb->data; 485 md_wakeup_thread(mddev->thread); 486 kfree(cb); 487 } 488 EXPORT_SYMBOL(md_unplug); 489 490 static inline struct mddev *mddev_get(struct mddev *mddev) 491 { 492 atomic_inc(&mddev->active); 493 return mddev; 494 } 495 496 static void mddev_delayed_delete(struct work_struct *ws); 497 498 static void mddev_put(struct mddev *mddev) 499 { 500 struct bio_set *bs = NULL; 501 502 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 503 return; 504 if (!mddev->raid_disks && list_empty(&mddev->disks) && 505 mddev->ctime == 0 && !mddev->hold_active) { 506 /* Array is not configured at all, and not held active, 507 * so destroy it */ 508 list_del_init(&mddev->all_mddevs); 509 bs = mddev->bio_set; 510 mddev->bio_set = NULL; 511 if (mddev->gendisk) { 512 /* We did a probe so need to clean up. Call 513 * queue_work inside the spinlock so that 514 * flush_workqueue() after mddev_find will 515 * succeed in waiting for the work to be done. 516 */ 517 INIT_WORK(&mddev->del_work, mddev_delayed_delete); 518 queue_work(md_misc_wq, &mddev->del_work); 519 } else 520 kfree(mddev); 521 } 522 spin_unlock(&all_mddevs_lock); 523 if (bs) 524 bioset_free(bs); 525 } 526 527 void mddev_init(struct mddev *mddev) 528 { 529 mutex_init(&mddev->open_mutex); 530 mutex_init(&mddev->reconfig_mutex); 531 mutex_init(&mddev->bitmap_info.mutex); 532 INIT_LIST_HEAD(&mddev->disks); 533 INIT_LIST_HEAD(&mddev->all_mddevs); 534 init_timer(&mddev->safemode_timer); 535 atomic_set(&mddev->active, 1); 536 atomic_set(&mddev->openers, 0); 537 atomic_set(&mddev->active_io, 0); 538 spin_lock_init(&mddev->write_lock); 539 atomic_set(&mddev->flush_pending, 0); 540 init_waitqueue_head(&mddev->sb_wait); 541 init_waitqueue_head(&mddev->recovery_wait); 542 mddev->reshape_position = MaxSector; 543 mddev->reshape_backwards = 0; 544 mddev->resync_min = 0; 545 mddev->resync_max = MaxSector; 546 mddev->level = LEVEL_NONE; 547 } 548 EXPORT_SYMBOL_GPL(mddev_init); 549 550 static struct mddev * mddev_find(dev_t unit) 551 { 552 struct mddev *mddev, *new = NULL; 553 554 if (unit && MAJOR(unit) != MD_MAJOR) 555 unit &= ~((1<<MdpMinorShift)-1); 556 557 retry: 558 spin_lock(&all_mddevs_lock); 559 560 if (unit) { 561 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 562 if (mddev->unit == unit) { 563 mddev_get(mddev); 564 spin_unlock(&all_mddevs_lock); 565 kfree(new); 566 return mddev; 567 } 568 569 if (new) { 570 list_add(&new->all_mddevs, &all_mddevs); 571 spin_unlock(&all_mddevs_lock); 572 new->hold_active = UNTIL_IOCTL; 573 return new; 574 } 575 } else if (new) { 576 /* find an unused unit number */ 577 static int next_minor = 512; 578 int start = next_minor; 579 int is_free = 0; 580 int dev = 0; 581 while (!is_free) { 582 dev = MKDEV(MD_MAJOR, next_minor); 583 next_minor++; 584 if (next_minor > MINORMASK) 585 next_minor = 0; 586 if (next_minor == start) { 587 /* Oh dear, all in use. */ 588 spin_unlock(&all_mddevs_lock); 589 kfree(new); 590 return NULL; 591 } 592 593 is_free = 1; 594 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 595 if (mddev->unit == dev) { 596 is_free = 0; 597 break; 598 } 599 } 600 new->unit = dev; 601 new->md_minor = MINOR(dev); 602 new->hold_active = UNTIL_STOP; 603 list_add(&new->all_mddevs, &all_mddevs); 604 spin_unlock(&all_mddevs_lock); 605 return new; 606 } 607 spin_unlock(&all_mddevs_lock); 608 609 new = kzalloc(sizeof(*new), GFP_KERNEL); 610 if (!new) 611 return NULL; 612 613 new->unit = unit; 614 if (MAJOR(unit) == MD_MAJOR) 615 new->md_minor = MINOR(unit); 616 else 617 new->md_minor = MINOR(unit) >> MdpMinorShift; 618 619 mddev_init(new); 620 621 goto retry; 622 } 623 624 static inline int mddev_lock(struct mddev * mddev) 625 { 626 return mutex_lock_interruptible(&mddev->reconfig_mutex); 627 } 628 629 static inline int mddev_is_locked(struct mddev *mddev) 630 { 631 return mutex_is_locked(&mddev->reconfig_mutex); 632 } 633 634 static inline int mddev_trylock(struct mddev * mddev) 635 { 636 return mutex_trylock(&mddev->reconfig_mutex); 637 } 638 639 static struct attribute_group md_redundancy_group; 640 641 static void mddev_unlock(struct mddev * mddev) 642 { 643 if (mddev->to_remove) { 644 /* These cannot be removed under reconfig_mutex as 645 * an access to the files will try to take reconfig_mutex 646 * while holding the file unremovable, which leads to 647 * a deadlock. 648 * So hold set sysfs_active while the remove in happeing, 649 * and anything else which might set ->to_remove or my 650 * otherwise change the sysfs namespace will fail with 651 * -EBUSY if sysfs_active is still set. 652 * We set sysfs_active under reconfig_mutex and elsewhere 653 * test it under the same mutex to ensure its correct value 654 * is seen. 655 */ 656 struct attribute_group *to_remove = mddev->to_remove; 657 mddev->to_remove = NULL; 658 mddev->sysfs_active = 1; 659 mutex_unlock(&mddev->reconfig_mutex); 660 661 if (mddev->kobj.sd) { 662 if (to_remove != &md_redundancy_group) 663 sysfs_remove_group(&mddev->kobj, to_remove); 664 if (mddev->pers == NULL || 665 mddev->pers->sync_request == NULL) { 666 sysfs_remove_group(&mddev->kobj, &md_redundancy_group); 667 if (mddev->sysfs_action) 668 sysfs_put(mddev->sysfs_action); 669 mddev->sysfs_action = NULL; 670 } 671 } 672 mddev->sysfs_active = 0; 673 } else 674 mutex_unlock(&mddev->reconfig_mutex); 675 676 /* As we've dropped the mutex we need a spinlock to 677 * make sure the thread doesn't disappear 678 */ 679 spin_lock(&pers_lock); 680 md_wakeup_thread(mddev->thread); 681 spin_unlock(&pers_lock); 682 } 683 684 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr) 685 { 686 struct md_rdev *rdev; 687 688 rdev_for_each(rdev, mddev) 689 if (rdev->desc_nr == nr) 690 return rdev; 691 692 return NULL; 693 } 694 695 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev) 696 { 697 struct md_rdev *rdev; 698 699 rdev_for_each(rdev, mddev) 700 if (rdev->bdev->bd_dev == dev) 701 return rdev; 702 703 return NULL; 704 } 705 706 static struct md_personality *find_pers(int level, char *clevel) 707 { 708 struct md_personality *pers; 709 list_for_each_entry(pers, &pers_list, list) { 710 if (level != LEVEL_NONE && pers->level == level) 711 return pers; 712 if (strcmp(pers->name, clevel)==0) 713 return pers; 714 } 715 return NULL; 716 } 717 718 /* return the offset of the super block in 512byte sectors */ 719 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) 720 { 721 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; 722 return MD_NEW_SIZE_SECTORS(num_sectors); 723 } 724 725 static int alloc_disk_sb(struct md_rdev * rdev) 726 { 727 if (rdev->sb_page) 728 MD_BUG(); 729 730 rdev->sb_page = alloc_page(GFP_KERNEL); 731 if (!rdev->sb_page) { 732 printk(KERN_ALERT "md: out of memory.\n"); 733 return -ENOMEM; 734 } 735 736 return 0; 737 } 738 739 void md_rdev_clear(struct md_rdev *rdev) 740 { 741 if (rdev->sb_page) { 742 put_page(rdev->sb_page); 743 rdev->sb_loaded = 0; 744 rdev->sb_page = NULL; 745 rdev->sb_start = 0; 746 rdev->sectors = 0; 747 } 748 if (rdev->bb_page) { 749 put_page(rdev->bb_page); 750 rdev->bb_page = NULL; 751 } 752 kfree(rdev->badblocks.page); 753 rdev->badblocks.page = NULL; 754 } 755 EXPORT_SYMBOL_GPL(md_rdev_clear); 756 757 static void super_written(struct bio *bio, int error) 758 { 759 struct md_rdev *rdev = bio->bi_private; 760 struct mddev *mddev = rdev->mddev; 761 762 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { 763 printk("md: super_written gets error=%d, uptodate=%d\n", 764 error, test_bit(BIO_UPTODATE, &bio->bi_flags)); 765 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); 766 md_error(mddev, rdev); 767 } 768 769 if (atomic_dec_and_test(&mddev->pending_writes)) 770 wake_up(&mddev->sb_wait); 771 bio_put(bio); 772 } 773 774 void md_super_write(struct mddev *mddev, struct md_rdev *rdev, 775 sector_t sector, int size, struct page *page) 776 { 777 /* write first size bytes of page to sector of rdev 778 * Increment mddev->pending_writes before returning 779 * and decrement it on completion, waking up sb_wait 780 * if zero is reached. 781 * If an error occurred, call md_error 782 */ 783 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); 784 785 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; 786 bio->bi_sector = sector; 787 bio_add_page(bio, page, size, 0); 788 bio->bi_private = rdev; 789 bio->bi_end_io = super_written; 790 791 atomic_inc(&mddev->pending_writes); 792 submit_bio(WRITE_FLUSH_FUA, bio); 793 } 794 795 void md_super_wait(struct mddev *mddev) 796 { 797 /* wait for all superblock writes that were scheduled to complete */ 798 DEFINE_WAIT(wq); 799 for(;;) { 800 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 801 if (atomic_read(&mddev->pending_writes)==0) 802 break; 803 schedule(); 804 } 805 finish_wait(&mddev->sb_wait, &wq); 806 } 807 808 static void bi_complete(struct bio *bio, int error) 809 { 810 complete((struct completion*)bio->bi_private); 811 } 812 813 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, 814 struct page *page, int rw, bool metadata_op) 815 { 816 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); 817 struct completion event; 818 int ret; 819 820 rw |= REQ_SYNC; 821 822 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? 823 rdev->meta_bdev : rdev->bdev; 824 if (metadata_op) 825 bio->bi_sector = sector + rdev->sb_start; 826 else if (rdev->mddev->reshape_position != MaxSector && 827 (rdev->mddev->reshape_backwards == 828 (sector >= rdev->mddev->reshape_position))) 829 bio->bi_sector = sector + rdev->new_data_offset; 830 else 831 bio->bi_sector = sector + rdev->data_offset; 832 bio_add_page(bio, page, size, 0); 833 init_completion(&event); 834 bio->bi_private = &event; 835 bio->bi_end_io = bi_complete; 836 submit_bio(rw, bio); 837 wait_for_completion(&event); 838 839 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 840 bio_put(bio); 841 return ret; 842 } 843 EXPORT_SYMBOL_GPL(sync_page_io); 844 845 static int read_disk_sb(struct md_rdev * rdev, int size) 846 { 847 char b[BDEVNAME_SIZE]; 848 if (!rdev->sb_page) { 849 MD_BUG(); 850 return -EINVAL; 851 } 852 if (rdev->sb_loaded) 853 return 0; 854 855 856 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) 857 goto fail; 858 rdev->sb_loaded = 1; 859 return 0; 860 861 fail: 862 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 863 bdevname(rdev->bdev,b)); 864 return -EINVAL; 865 } 866 867 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 868 { 869 return sb1->set_uuid0 == sb2->set_uuid0 && 870 sb1->set_uuid1 == sb2->set_uuid1 && 871 sb1->set_uuid2 == sb2->set_uuid2 && 872 sb1->set_uuid3 == sb2->set_uuid3; 873 } 874 875 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 876 { 877 int ret; 878 mdp_super_t *tmp1, *tmp2; 879 880 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 881 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 882 883 if (!tmp1 || !tmp2) { 884 ret = 0; 885 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); 886 goto abort; 887 } 888 889 *tmp1 = *sb1; 890 *tmp2 = *sb2; 891 892 /* 893 * nr_disks is not constant 894 */ 895 tmp1->nr_disks = 0; 896 tmp2->nr_disks = 0; 897 898 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); 899 abort: 900 kfree(tmp1); 901 kfree(tmp2); 902 return ret; 903 } 904 905 906 static u32 md_csum_fold(u32 csum) 907 { 908 csum = (csum & 0xffff) + (csum >> 16); 909 return (csum & 0xffff) + (csum >> 16); 910 } 911 912 static unsigned int calc_sb_csum(mdp_super_t * sb) 913 { 914 u64 newcsum = 0; 915 u32 *sb32 = (u32*)sb; 916 int i; 917 unsigned int disk_csum, csum; 918 919 disk_csum = sb->sb_csum; 920 sb->sb_csum = 0; 921 922 for (i = 0; i < MD_SB_BYTES/4 ; i++) 923 newcsum += sb32[i]; 924 csum = (newcsum & 0xffffffff) + (newcsum>>32); 925 926 927 #ifdef CONFIG_ALPHA 928 /* This used to use csum_partial, which was wrong for several 929 * reasons including that different results are returned on 930 * different architectures. It isn't critical that we get exactly 931 * the same return value as before (we always csum_fold before 932 * testing, and that removes any differences). However as we 933 * know that csum_partial always returned a 16bit value on 934 * alphas, do a fold to maximise conformity to previous behaviour. 935 */ 936 sb->sb_csum = md_csum_fold(disk_csum); 937 #else 938 sb->sb_csum = disk_csum; 939 #endif 940 return csum; 941 } 942 943 944 /* 945 * Handle superblock details. 946 * We want to be able to handle multiple superblock formats 947 * so we have a common interface to them all, and an array of 948 * different handlers. 949 * We rely on user-space to write the initial superblock, and support 950 * reading and updating of superblocks. 951 * Interface methods are: 952 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) 953 * loads and validates a superblock on dev. 954 * if refdev != NULL, compare superblocks on both devices 955 * Return: 956 * 0 - dev has a superblock that is compatible with refdev 957 * 1 - dev has a superblock that is compatible and newer than refdev 958 * so dev should be used as the refdev in future 959 * -EINVAL superblock incompatible or invalid 960 * -othererror e.g. -EIO 961 * 962 * int validate_super(struct mddev *mddev, struct md_rdev *dev) 963 * Verify that dev is acceptable into mddev. 964 * The first time, mddev->raid_disks will be 0, and data from 965 * dev should be merged in. Subsequent calls check that dev 966 * is new enough. Return 0 or -EINVAL 967 * 968 * void sync_super(struct mddev *mddev, struct md_rdev *dev) 969 * Update the superblock for rdev with data in mddev 970 * This does not write to disc. 971 * 972 */ 973 974 struct super_type { 975 char *name; 976 struct module *owner; 977 int (*load_super)(struct md_rdev *rdev, 978 struct md_rdev *refdev, 979 int minor_version); 980 int (*validate_super)(struct mddev *mddev, 981 struct md_rdev *rdev); 982 void (*sync_super)(struct mddev *mddev, 983 struct md_rdev *rdev); 984 unsigned long long (*rdev_size_change)(struct md_rdev *rdev, 985 sector_t num_sectors); 986 int (*allow_new_offset)(struct md_rdev *rdev, 987 unsigned long long new_offset); 988 }; 989 990 /* 991 * Check that the given mddev has no bitmap. 992 * 993 * This function is called from the run method of all personalities that do not 994 * support bitmaps. It prints an error message and returns non-zero if mddev 995 * has a bitmap. Otherwise, it returns 0. 996 * 997 */ 998 int md_check_no_bitmap(struct mddev *mddev) 999 { 1000 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) 1001 return 0; 1002 printk(KERN_ERR "%s: bitmaps are not supported for %s\n", 1003 mdname(mddev), mddev->pers->name); 1004 return 1; 1005 } 1006 EXPORT_SYMBOL(md_check_no_bitmap); 1007 1008 /* 1009 * load_super for 0.90.0 1010 */ 1011 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) 1012 { 1013 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1014 mdp_super_t *sb; 1015 int ret; 1016 1017 /* 1018 * Calculate the position of the superblock (512byte sectors), 1019 * it's at the end of the disk. 1020 * 1021 * It also happens to be a multiple of 4Kb. 1022 */ 1023 rdev->sb_start = calc_dev_sboffset(rdev); 1024 1025 ret = read_disk_sb(rdev, MD_SB_BYTES); 1026 if (ret) return ret; 1027 1028 ret = -EINVAL; 1029 1030 bdevname(rdev->bdev, b); 1031 sb = page_address(rdev->sb_page); 1032 1033 if (sb->md_magic != MD_SB_MAGIC) { 1034 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 1035 b); 1036 goto abort; 1037 } 1038 1039 if (sb->major_version != 0 || 1040 sb->minor_version < 90 || 1041 sb->minor_version > 91) { 1042 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 1043 sb->major_version, sb->minor_version, 1044 b); 1045 goto abort; 1046 } 1047 1048 if (sb->raid_disks <= 0) 1049 goto abort; 1050 1051 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { 1052 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 1053 b); 1054 goto abort; 1055 } 1056 1057 rdev->preferred_minor = sb->md_minor; 1058 rdev->data_offset = 0; 1059 rdev->new_data_offset = 0; 1060 rdev->sb_size = MD_SB_BYTES; 1061 rdev->badblocks.shift = -1; 1062 1063 if (sb->level == LEVEL_MULTIPATH) 1064 rdev->desc_nr = -1; 1065 else 1066 rdev->desc_nr = sb->this_disk.number; 1067 1068 if (!refdev) { 1069 ret = 1; 1070 } else { 1071 __u64 ev1, ev2; 1072 mdp_super_t *refsb = page_address(refdev->sb_page); 1073 if (!uuid_equal(refsb, sb)) { 1074 printk(KERN_WARNING "md: %s has different UUID to %s\n", 1075 b, bdevname(refdev->bdev,b2)); 1076 goto abort; 1077 } 1078 if (!sb_equal(refsb, sb)) { 1079 printk(KERN_WARNING "md: %s has same UUID" 1080 " but different superblock to %s\n", 1081 b, bdevname(refdev->bdev, b2)); 1082 goto abort; 1083 } 1084 ev1 = md_event(sb); 1085 ev2 = md_event(refsb); 1086 if (ev1 > ev2) 1087 ret = 1; 1088 else 1089 ret = 0; 1090 } 1091 rdev->sectors = rdev->sb_start; 1092 /* Limit to 4TB as metadata cannot record more than that. 1093 * (not needed for Linear and RAID0 as metadata doesn't 1094 * record this size) 1095 */ 1096 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) 1097 rdev->sectors = (2ULL << 32) - 2; 1098 1099 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) 1100 /* "this cannot possibly happen" ... */ 1101 ret = -EINVAL; 1102 1103 abort: 1104 return ret; 1105 } 1106 1107 /* 1108 * validate_super for 0.90.0 1109 */ 1110 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) 1111 { 1112 mdp_disk_t *desc; 1113 mdp_super_t *sb = page_address(rdev->sb_page); 1114 __u64 ev1 = md_event(sb); 1115 1116 rdev->raid_disk = -1; 1117 clear_bit(Faulty, &rdev->flags); 1118 clear_bit(In_sync, &rdev->flags); 1119 clear_bit(WriteMostly, &rdev->flags); 1120 1121 if (mddev->raid_disks == 0) { 1122 mddev->major_version = 0; 1123 mddev->minor_version = sb->minor_version; 1124 mddev->patch_version = sb->patch_version; 1125 mddev->external = 0; 1126 mddev->chunk_sectors = sb->chunk_size >> 9; 1127 mddev->ctime = sb->ctime; 1128 mddev->utime = sb->utime; 1129 mddev->level = sb->level; 1130 mddev->clevel[0] = 0; 1131 mddev->layout = sb->layout; 1132 mddev->raid_disks = sb->raid_disks; 1133 mddev->dev_sectors = ((sector_t)sb->size) * 2; 1134 mddev->events = ev1; 1135 mddev->bitmap_info.offset = 0; 1136 mddev->bitmap_info.space = 0; 1137 /* bitmap can use 60 K after the 4K superblocks */ 1138 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 1139 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); 1140 mddev->reshape_backwards = 0; 1141 1142 if (mddev->minor_version >= 91) { 1143 mddev->reshape_position = sb->reshape_position; 1144 mddev->delta_disks = sb->delta_disks; 1145 mddev->new_level = sb->new_level; 1146 mddev->new_layout = sb->new_layout; 1147 mddev->new_chunk_sectors = sb->new_chunk >> 9; 1148 if (mddev->delta_disks < 0) 1149 mddev->reshape_backwards = 1; 1150 } else { 1151 mddev->reshape_position = MaxSector; 1152 mddev->delta_disks = 0; 1153 mddev->new_level = mddev->level; 1154 mddev->new_layout = mddev->layout; 1155 mddev->new_chunk_sectors = mddev->chunk_sectors; 1156 } 1157 1158 if (sb->state & (1<<MD_SB_CLEAN)) 1159 mddev->recovery_cp = MaxSector; 1160 else { 1161 if (sb->events_hi == sb->cp_events_hi && 1162 sb->events_lo == sb->cp_events_lo) { 1163 mddev->recovery_cp = sb->recovery_cp; 1164 } else 1165 mddev->recovery_cp = 0; 1166 } 1167 1168 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 1169 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 1170 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 1171 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 1172 1173 mddev->max_disks = MD_SB_DISKS; 1174 1175 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 1176 mddev->bitmap_info.file == NULL) { 1177 mddev->bitmap_info.offset = 1178 mddev->bitmap_info.default_offset; 1179 mddev->bitmap_info.space = 1180 mddev->bitmap_info.space; 1181 } 1182 1183 } else if (mddev->pers == NULL) { 1184 /* Insist on good event counter while assembling, except 1185 * for spares (which don't need an event count) */ 1186 ++ev1; 1187 if (sb->disks[rdev->desc_nr].state & ( 1188 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) 1189 if (ev1 < mddev->events) 1190 return -EINVAL; 1191 } else if (mddev->bitmap) { 1192 /* if adding to array with a bitmap, then we can accept an 1193 * older device ... but not too old. 1194 */ 1195 if (ev1 < mddev->bitmap->events_cleared) 1196 return 0; 1197 } else { 1198 if (ev1 < mddev->events) 1199 /* just a hot-add of a new device, leave raid_disk at -1 */ 1200 return 0; 1201 } 1202 1203 if (mddev->level != LEVEL_MULTIPATH) { 1204 desc = sb->disks + rdev->desc_nr; 1205 1206 if (desc->state & (1<<MD_DISK_FAULTY)) 1207 set_bit(Faulty, &rdev->flags); 1208 else if (desc->state & (1<<MD_DISK_SYNC) /* && 1209 desc->raid_disk < mddev->raid_disks */) { 1210 set_bit(In_sync, &rdev->flags); 1211 rdev->raid_disk = desc->raid_disk; 1212 } else if (desc->state & (1<<MD_DISK_ACTIVE)) { 1213 /* active but not in sync implies recovery up to 1214 * reshape position. We don't know exactly where 1215 * that is, so set to zero for now */ 1216 if (mddev->minor_version >= 91) { 1217 rdev->recovery_offset = 0; 1218 rdev->raid_disk = desc->raid_disk; 1219 } 1220 } 1221 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 1222 set_bit(WriteMostly, &rdev->flags); 1223 } else /* MULTIPATH are always insync */ 1224 set_bit(In_sync, &rdev->flags); 1225 return 0; 1226 } 1227 1228 /* 1229 * sync_super for 0.90.0 1230 */ 1231 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) 1232 { 1233 mdp_super_t *sb; 1234 struct md_rdev *rdev2; 1235 int next_spare = mddev->raid_disks; 1236 1237 1238 /* make rdev->sb match mddev data.. 1239 * 1240 * 1/ zero out disks 1241 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 1242 * 3/ any empty disks < next_spare become removed 1243 * 1244 * disks[0] gets initialised to REMOVED because 1245 * we cannot be sure from other fields if it has 1246 * been initialised or not. 1247 */ 1248 int i; 1249 int active=0, working=0,failed=0,spare=0,nr_disks=0; 1250 1251 rdev->sb_size = MD_SB_BYTES; 1252 1253 sb = page_address(rdev->sb_page); 1254 1255 memset(sb, 0, sizeof(*sb)); 1256 1257 sb->md_magic = MD_SB_MAGIC; 1258 sb->major_version = mddev->major_version; 1259 sb->patch_version = mddev->patch_version; 1260 sb->gvalid_words = 0; /* ignored */ 1261 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 1262 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 1263 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 1264 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 1265 1266 sb->ctime = mddev->ctime; 1267 sb->level = mddev->level; 1268 sb->size = mddev->dev_sectors / 2; 1269 sb->raid_disks = mddev->raid_disks; 1270 sb->md_minor = mddev->md_minor; 1271 sb->not_persistent = 0; 1272 sb->utime = mddev->utime; 1273 sb->state = 0; 1274 sb->events_hi = (mddev->events>>32); 1275 sb->events_lo = (u32)mddev->events; 1276 1277 if (mddev->reshape_position == MaxSector) 1278 sb->minor_version = 90; 1279 else { 1280 sb->minor_version = 91; 1281 sb->reshape_position = mddev->reshape_position; 1282 sb->new_level = mddev->new_level; 1283 sb->delta_disks = mddev->delta_disks; 1284 sb->new_layout = mddev->new_layout; 1285 sb->new_chunk = mddev->new_chunk_sectors << 9; 1286 } 1287 mddev->minor_version = sb->minor_version; 1288 if (mddev->in_sync) 1289 { 1290 sb->recovery_cp = mddev->recovery_cp; 1291 sb->cp_events_hi = (mddev->events>>32); 1292 sb->cp_events_lo = (u32)mddev->events; 1293 if (mddev->recovery_cp == MaxSector) 1294 sb->state = (1<< MD_SB_CLEAN); 1295 } else 1296 sb->recovery_cp = 0; 1297 1298 sb->layout = mddev->layout; 1299 sb->chunk_size = mddev->chunk_sectors << 9; 1300 1301 if (mddev->bitmap && mddev->bitmap_info.file == NULL) 1302 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 1303 1304 sb->disks[0].state = (1<<MD_DISK_REMOVED); 1305 rdev_for_each(rdev2, mddev) { 1306 mdp_disk_t *d; 1307 int desc_nr; 1308 int is_active = test_bit(In_sync, &rdev2->flags); 1309 1310 if (rdev2->raid_disk >= 0 && 1311 sb->minor_version >= 91) 1312 /* we have nowhere to store the recovery_offset, 1313 * but if it is not below the reshape_position, 1314 * we can piggy-back on that. 1315 */ 1316 is_active = 1; 1317 if (rdev2->raid_disk < 0 || 1318 test_bit(Faulty, &rdev2->flags)) 1319 is_active = 0; 1320 if (is_active) 1321 desc_nr = rdev2->raid_disk; 1322 else 1323 desc_nr = next_spare++; 1324 rdev2->desc_nr = desc_nr; 1325 d = &sb->disks[rdev2->desc_nr]; 1326 nr_disks++; 1327 d->number = rdev2->desc_nr; 1328 d->major = MAJOR(rdev2->bdev->bd_dev); 1329 d->minor = MINOR(rdev2->bdev->bd_dev); 1330 if (is_active) 1331 d->raid_disk = rdev2->raid_disk; 1332 else 1333 d->raid_disk = rdev2->desc_nr; /* compatibility */ 1334 if (test_bit(Faulty, &rdev2->flags)) 1335 d->state = (1<<MD_DISK_FAULTY); 1336 else if (is_active) { 1337 d->state = (1<<MD_DISK_ACTIVE); 1338 if (test_bit(In_sync, &rdev2->flags)) 1339 d->state |= (1<<MD_DISK_SYNC); 1340 active++; 1341 working++; 1342 } else { 1343 d->state = 0; 1344 spare++; 1345 working++; 1346 } 1347 if (test_bit(WriteMostly, &rdev2->flags)) 1348 d->state |= (1<<MD_DISK_WRITEMOSTLY); 1349 } 1350 /* now set the "removed" and "faulty" bits on any missing devices */ 1351 for (i=0 ; i < mddev->raid_disks ; i++) { 1352 mdp_disk_t *d = &sb->disks[i]; 1353 if (d->state == 0 && d->number == 0) { 1354 d->number = i; 1355 d->raid_disk = i; 1356 d->state = (1<<MD_DISK_REMOVED); 1357 d->state |= (1<<MD_DISK_FAULTY); 1358 failed++; 1359 } 1360 } 1361 sb->nr_disks = nr_disks; 1362 sb->active_disks = active; 1363 sb->working_disks = working; 1364 sb->failed_disks = failed; 1365 sb->spare_disks = spare; 1366 1367 sb->this_disk = sb->disks[rdev->desc_nr]; 1368 sb->sb_csum = calc_sb_csum(sb); 1369 } 1370 1371 /* 1372 * rdev_size_change for 0.90.0 1373 */ 1374 static unsigned long long 1375 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) 1376 { 1377 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1378 return 0; /* component must fit device */ 1379 if (rdev->mddev->bitmap_info.offset) 1380 return 0; /* can't move bitmap */ 1381 rdev->sb_start = calc_dev_sboffset(rdev); 1382 if (!num_sectors || num_sectors > rdev->sb_start) 1383 num_sectors = rdev->sb_start; 1384 /* Limit to 4TB as metadata cannot record more than that. 1385 * 4TB == 2^32 KB, or 2*2^32 sectors. 1386 */ 1387 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) 1388 num_sectors = (2ULL << 32) - 2; 1389 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1390 rdev->sb_page); 1391 md_super_wait(rdev->mddev); 1392 return num_sectors; 1393 } 1394 1395 static int 1396 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) 1397 { 1398 /* non-zero offset changes not possible with v0.90 */ 1399 return new_offset == 0; 1400 } 1401 1402 /* 1403 * version 1 superblock 1404 */ 1405 1406 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) 1407 { 1408 __le32 disk_csum; 1409 u32 csum; 1410 unsigned long long newcsum; 1411 int size = 256 + le32_to_cpu(sb->max_dev)*2; 1412 __le32 *isuper = (__le32*)sb; 1413 int i; 1414 1415 disk_csum = sb->sb_csum; 1416 sb->sb_csum = 0; 1417 newcsum = 0; 1418 for (i=0; size>=4; size -= 4 ) 1419 newcsum += le32_to_cpu(*isuper++); 1420 1421 if (size == 2) 1422 newcsum += le16_to_cpu(*(__le16*) isuper); 1423 1424 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 1425 sb->sb_csum = disk_csum; 1426 return cpu_to_le32(csum); 1427 } 1428 1429 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, 1430 int acknowledged); 1431 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) 1432 { 1433 struct mdp_superblock_1 *sb; 1434 int ret; 1435 sector_t sb_start; 1436 sector_t sectors; 1437 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1438 int bmask; 1439 1440 /* 1441 * Calculate the position of the superblock in 512byte sectors. 1442 * It is always aligned to a 4K boundary and 1443 * depeding on minor_version, it can be: 1444 * 0: At least 8K, but less than 12K, from end of device 1445 * 1: At start of device 1446 * 2: 4K from start of device. 1447 */ 1448 switch(minor_version) { 1449 case 0: 1450 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; 1451 sb_start -= 8*2; 1452 sb_start &= ~(sector_t)(4*2-1); 1453 break; 1454 case 1: 1455 sb_start = 0; 1456 break; 1457 case 2: 1458 sb_start = 8; 1459 break; 1460 default: 1461 return -EINVAL; 1462 } 1463 rdev->sb_start = sb_start; 1464 1465 /* superblock is rarely larger than 1K, but it can be larger, 1466 * and it is safe to read 4k, so we do that 1467 */ 1468 ret = read_disk_sb(rdev, 4096); 1469 if (ret) return ret; 1470 1471 1472 sb = page_address(rdev->sb_page); 1473 1474 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 1475 sb->major_version != cpu_to_le32(1) || 1476 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 1477 le64_to_cpu(sb->super_offset) != rdev->sb_start || 1478 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 1479 return -EINVAL; 1480 1481 if (calc_sb_1_csum(sb) != sb->sb_csum) { 1482 printk("md: invalid superblock checksum on %s\n", 1483 bdevname(rdev->bdev,b)); 1484 return -EINVAL; 1485 } 1486 if (le64_to_cpu(sb->data_size) < 10) { 1487 printk("md: data_size too small on %s\n", 1488 bdevname(rdev->bdev,b)); 1489 return -EINVAL; 1490 } 1491 if (sb->pad0 || 1492 sb->pad3[0] || 1493 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) 1494 /* Some padding is non-zero, might be a new feature */ 1495 return -EINVAL; 1496 1497 rdev->preferred_minor = 0xffff; 1498 rdev->data_offset = le64_to_cpu(sb->data_offset); 1499 rdev->new_data_offset = rdev->data_offset; 1500 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && 1501 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) 1502 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); 1503 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); 1504 1505 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 1506 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1507 if (rdev->sb_size & bmask) 1508 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1509 1510 if (minor_version 1511 && rdev->data_offset < sb_start + (rdev->sb_size/512)) 1512 return -EINVAL; 1513 if (minor_version 1514 && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) 1515 return -EINVAL; 1516 1517 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) 1518 rdev->desc_nr = -1; 1519 else 1520 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1521 1522 if (!rdev->bb_page) { 1523 rdev->bb_page = alloc_page(GFP_KERNEL); 1524 if (!rdev->bb_page) 1525 return -ENOMEM; 1526 } 1527 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && 1528 rdev->badblocks.count == 0) { 1529 /* need to load the bad block list. 1530 * Currently we limit it to one page. 1531 */ 1532 s32 offset; 1533 sector_t bb_sector; 1534 u64 *bbp; 1535 int i; 1536 int sectors = le16_to_cpu(sb->bblog_size); 1537 if (sectors > (PAGE_SIZE / 512)) 1538 return -EINVAL; 1539 offset = le32_to_cpu(sb->bblog_offset); 1540 if (offset == 0) 1541 return -EINVAL; 1542 bb_sector = (long long)offset; 1543 if (!sync_page_io(rdev, bb_sector, sectors << 9, 1544 rdev->bb_page, READ, true)) 1545 return -EIO; 1546 bbp = (u64 *)page_address(rdev->bb_page); 1547 rdev->badblocks.shift = sb->bblog_shift; 1548 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { 1549 u64 bb = le64_to_cpu(*bbp); 1550 int count = bb & (0x3ff); 1551 u64 sector = bb >> 10; 1552 sector <<= sb->bblog_shift; 1553 count <<= sb->bblog_shift; 1554 if (bb + 1 == 0) 1555 break; 1556 if (md_set_badblocks(&rdev->badblocks, 1557 sector, count, 1) == 0) 1558 return -EINVAL; 1559 } 1560 } else if (sb->bblog_offset == 0) 1561 rdev->badblocks.shift = -1; 1562 1563 if (!refdev) { 1564 ret = 1; 1565 } else { 1566 __u64 ev1, ev2; 1567 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); 1568 1569 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 1570 sb->level != refsb->level || 1571 sb->layout != refsb->layout || 1572 sb->chunksize != refsb->chunksize) { 1573 printk(KERN_WARNING "md: %s has strangely different" 1574 " superblock to %s\n", 1575 bdevname(rdev->bdev,b), 1576 bdevname(refdev->bdev,b2)); 1577 return -EINVAL; 1578 } 1579 ev1 = le64_to_cpu(sb->events); 1580 ev2 = le64_to_cpu(refsb->events); 1581 1582 if (ev1 > ev2) 1583 ret = 1; 1584 else 1585 ret = 0; 1586 } 1587 if (minor_version) { 1588 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); 1589 sectors -= rdev->data_offset; 1590 } else 1591 sectors = rdev->sb_start; 1592 if (sectors < le64_to_cpu(sb->data_size)) 1593 return -EINVAL; 1594 rdev->sectors = le64_to_cpu(sb->data_size); 1595 return ret; 1596 } 1597 1598 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) 1599 { 1600 struct mdp_superblock_1 *sb = page_address(rdev->sb_page); 1601 __u64 ev1 = le64_to_cpu(sb->events); 1602 1603 rdev->raid_disk = -1; 1604 clear_bit(Faulty, &rdev->flags); 1605 clear_bit(In_sync, &rdev->flags); 1606 clear_bit(WriteMostly, &rdev->flags); 1607 1608 if (mddev->raid_disks == 0) { 1609 mddev->major_version = 1; 1610 mddev->patch_version = 0; 1611 mddev->external = 0; 1612 mddev->chunk_sectors = le32_to_cpu(sb->chunksize); 1613 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1614 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1615 mddev->level = le32_to_cpu(sb->level); 1616 mddev->clevel[0] = 0; 1617 mddev->layout = le32_to_cpu(sb->layout); 1618 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1619 mddev->dev_sectors = le64_to_cpu(sb->size); 1620 mddev->events = ev1; 1621 mddev->bitmap_info.offset = 0; 1622 mddev->bitmap_info.space = 0; 1623 /* Default location for bitmap is 1K after superblock 1624 * using 3K - total of 4K 1625 */ 1626 mddev->bitmap_info.default_offset = 1024 >> 9; 1627 mddev->bitmap_info.default_space = (4096-1024) >> 9; 1628 mddev->reshape_backwards = 0; 1629 1630 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1631 memcpy(mddev->uuid, sb->set_uuid, 16); 1632 1633 mddev->max_disks = (4096-256)/2; 1634 1635 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1636 mddev->bitmap_info.file == NULL) { 1637 mddev->bitmap_info.offset = 1638 (__s32)le32_to_cpu(sb->bitmap_offset); 1639 /* Metadata doesn't record how much space is available. 1640 * For 1.0, we assume we can use up to the superblock 1641 * if before, else to 4K beyond superblock. 1642 * For others, assume no change is possible. 1643 */ 1644 if (mddev->minor_version > 0) 1645 mddev->bitmap_info.space = 0; 1646 else if (mddev->bitmap_info.offset > 0) 1647 mddev->bitmap_info.space = 1648 8 - mddev->bitmap_info.offset; 1649 else 1650 mddev->bitmap_info.space = 1651 -mddev->bitmap_info.offset; 1652 } 1653 1654 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { 1655 mddev->reshape_position = le64_to_cpu(sb->reshape_position); 1656 mddev->delta_disks = le32_to_cpu(sb->delta_disks); 1657 mddev->new_level = le32_to_cpu(sb->new_level); 1658 mddev->new_layout = le32_to_cpu(sb->new_layout); 1659 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); 1660 if (mddev->delta_disks < 0 || 1661 (mddev->delta_disks == 0 && 1662 (le32_to_cpu(sb->feature_map) 1663 & MD_FEATURE_RESHAPE_BACKWARDS))) 1664 mddev->reshape_backwards = 1; 1665 } else { 1666 mddev->reshape_position = MaxSector; 1667 mddev->delta_disks = 0; 1668 mddev->new_level = mddev->level; 1669 mddev->new_layout = mddev->layout; 1670 mddev->new_chunk_sectors = mddev->chunk_sectors; 1671 } 1672 1673 } else if (mddev->pers == NULL) { 1674 /* Insist of good event counter while assembling, except for 1675 * spares (which don't need an event count) */ 1676 ++ev1; 1677 if (rdev->desc_nr >= 0 && 1678 rdev->desc_nr < le32_to_cpu(sb->max_dev) && 1679 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) 1680 if (ev1 < mddev->events) 1681 return -EINVAL; 1682 } else if (mddev->bitmap) { 1683 /* If adding to array with a bitmap, then we can accept an 1684 * older device, but not too old. 1685 */ 1686 if (ev1 < mddev->bitmap->events_cleared) 1687 return 0; 1688 } else { 1689 if (ev1 < mddev->events) 1690 /* just a hot-add of a new device, leave raid_disk at -1 */ 1691 return 0; 1692 } 1693 if (mddev->level != LEVEL_MULTIPATH) { 1694 int role; 1695 if (rdev->desc_nr < 0 || 1696 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { 1697 role = 0xffff; 1698 rdev->desc_nr = -1; 1699 } else 1700 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1701 switch(role) { 1702 case 0xffff: /* spare */ 1703 break; 1704 case 0xfffe: /* faulty */ 1705 set_bit(Faulty, &rdev->flags); 1706 break; 1707 default: 1708 if ((le32_to_cpu(sb->feature_map) & 1709 MD_FEATURE_RECOVERY_OFFSET)) 1710 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); 1711 else 1712 set_bit(In_sync, &rdev->flags); 1713 rdev->raid_disk = role; 1714 break; 1715 } 1716 if (sb->devflags & WriteMostly1) 1717 set_bit(WriteMostly, &rdev->flags); 1718 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) 1719 set_bit(Replacement, &rdev->flags); 1720 } else /* MULTIPATH are always insync */ 1721 set_bit(In_sync, &rdev->flags); 1722 1723 return 0; 1724 } 1725 1726 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) 1727 { 1728 struct mdp_superblock_1 *sb; 1729 struct md_rdev *rdev2; 1730 int max_dev, i; 1731 /* make rdev->sb match mddev and rdev data. */ 1732 1733 sb = page_address(rdev->sb_page); 1734 1735 sb->feature_map = 0; 1736 sb->pad0 = 0; 1737 sb->recovery_offset = cpu_to_le64(0); 1738 memset(sb->pad3, 0, sizeof(sb->pad3)); 1739 1740 sb->utime = cpu_to_le64((__u64)mddev->utime); 1741 sb->events = cpu_to_le64(mddev->events); 1742 if (mddev->in_sync) 1743 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1744 else 1745 sb->resync_offset = cpu_to_le64(0); 1746 1747 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); 1748 1749 sb->raid_disks = cpu_to_le32(mddev->raid_disks); 1750 sb->size = cpu_to_le64(mddev->dev_sectors); 1751 sb->chunksize = cpu_to_le32(mddev->chunk_sectors); 1752 sb->level = cpu_to_le32(mddev->level); 1753 sb->layout = cpu_to_le32(mddev->layout); 1754 1755 if (test_bit(WriteMostly, &rdev->flags)) 1756 sb->devflags |= WriteMostly1; 1757 else 1758 sb->devflags &= ~WriteMostly1; 1759 sb->data_offset = cpu_to_le64(rdev->data_offset); 1760 sb->data_size = cpu_to_le64(rdev->sectors); 1761 1762 if (mddev->bitmap && mddev->bitmap_info.file == NULL) { 1763 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); 1764 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); 1765 } 1766 1767 if (rdev->raid_disk >= 0 && 1768 !test_bit(In_sync, &rdev->flags)) { 1769 sb->feature_map |= 1770 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); 1771 sb->recovery_offset = 1772 cpu_to_le64(rdev->recovery_offset); 1773 } 1774 if (test_bit(Replacement, &rdev->flags)) 1775 sb->feature_map |= 1776 cpu_to_le32(MD_FEATURE_REPLACEMENT); 1777 1778 if (mddev->reshape_position != MaxSector) { 1779 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); 1780 sb->reshape_position = cpu_to_le64(mddev->reshape_position); 1781 sb->new_layout = cpu_to_le32(mddev->new_layout); 1782 sb->delta_disks = cpu_to_le32(mddev->delta_disks); 1783 sb->new_level = cpu_to_le32(mddev->new_level); 1784 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); 1785 if (mddev->delta_disks == 0 && 1786 mddev->reshape_backwards) 1787 sb->feature_map 1788 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); 1789 if (rdev->new_data_offset != rdev->data_offset) { 1790 sb->feature_map 1791 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); 1792 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset 1793 - rdev->data_offset)); 1794 } 1795 } 1796 1797 if (rdev->badblocks.count == 0) 1798 /* Nothing to do for bad blocks*/ ; 1799 else if (sb->bblog_offset == 0) 1800 /* Cannot record bad blocks on this device */ 1801 md_error(mddev, rdev); 1802 else { 1803 struct badblocks *bb = &rdev->badblocks; 1804 u64 *bbp = (u64 *)page_address(rdev->bb_page); 1805 u64 *p = bb->page; 1806 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); 1807 if (bb->changed) { 1808 unsigned seq; 1809 1810 retry: 1811 seq = read_seqbegin(&bb->lock); 1812 1813 memset(bbp, 0xff, PAGE_SIZE); 1814 1815 for (i = 0 ; i < bb->count ; i++) { 1816 u64 internal_bb = *p++; 1817 u64 store_bb = ((BB_OFFSET(internal_bb) << 10) 1818 | BB_LEN(internal_bb)); 1819 *bbp++ = cpu_to_le64(store_bb); 1820 } 1821 bb->changed = 0; 1822 if (read_seqretry(&bb->lock, seq)) 1823 goto retry; 1824 1825 bb->sector = (rdev->sb_start + 1826 (int)le32_to_cpu(sb->bblog_offset)); 1827 bb->size = le16_to_cpu(sb->bblog_size); 1828 } 1829 } 1830 1831 max_dev = 0; 1832 rdev_for_each(rdev2, mddev) 1833 if (rdev2->desc_nr+1 > max_dev) 1834 max_dev = rdev2->desc_nr+1; 1835 1836 if (max_dev > le32_to_cpu(sb->max_dev)) { 1837 int bmask; 1838 sb->max_dev = cpu_to_le32(max_dev); 1839 rdev->sb_size = max_dev * 2 + 256; 1840 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1841 if (rdev->sb_size & bmask) 1842 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1843 } else 1844 max_dev = le32_to_cpu(sb->max_dev); 1845 1846 for (i=0; i<max_dev;i++) 1847 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1848 1849 rdev_for_each(rdev2, mddev) { 1850 i = rdev2->desc_nr; 1851 if (test_bit(Faulty, &rdev2->flags)) 1852 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1853 else if (test_bit(In_sync, &rdev2->flags)) 1854 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1855 else if (rdev2->raid_disk >= 0) 1856 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1857 else 1858 sb->dev_roles[i] = cpu_to_le16(0xffff); 1859 } 1860 1861 sb->sb_csum = calc_sb_1_csum(sb); 1862 } 1863 1864 static unsigned long long 1865 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) 1866 { 1867 struct mdp_superblock_1 *sb; 1868 sector_t max_sectors; 1869 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1870 return 0; /* component must fit device */ 1871 if (rdev->data_offset != rdev->new_data_offset) 1872 return 0; /* too confusing */ 1873 if (rdev->sb_start < rdev->data_offset) { 1874 /* minor versions 1 and 2; superblock before data */ 1875 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; 1876 max_sectors -= rdev->data_offset; 1877 if (!num_sectors || num_sectors > max_sectors) 1878 num_sectors = max_sectors; 1879 } else if (rdev->mddev->bitmap_info.offset) { 1880 /* minor version 0 with bitmap we can't move */ 1881 return 0; 1882 } else { 1883 /* minor version 0; superblock after data */ 1884 sector_t sb_start; 1885 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; 1886 sb_start &= ~(sector_t)(4*2 - 1); 1887 max_sectors = rdev->sectors + sb_start - rdev->sb_start; 1888 if (!num_sectors || num_sectors > max_sectors) 1889 num_sectors = max_sectors; 1890 rdev->sb_start = sb_start; 1891 } 1892 sb = page_address(rdev->sb_page); 1893 sb->data_size = cpu_to_le64(num_sectors); 1894 sb->super_offset = rdev->sb_start; 1895 sb->sb_csum = calc_sb_1_csum(sb); 1896 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1897 rdev->sb_page); 1898 md_super_wait(rdev->mddev); 1899 return num_sectors; 1900 1901 } 1902 1903 static int 1904 super_1_allow_new_offset(struct md_rdev *rdev, 1905 unsigned long long new_offset) 1906 { 1907 /* All necessary checks on new >= old have been done */ 1908 struct bitmap *bitmap; 1909 if (new_offset >= rdev->data_offset) 1910 return 1; 1911 1912 /* with 1.0 metadata, there is no metadata to tread on 1913 * so we can always move back */ 1914 if (rdev->mddev->minor_version == 0) 1915 return 1; 1916 1917 /* otherwise we must be sure not to step on 1918 * any metadata, so stay: 1919 * 36K beyond start of superblock 1920 * beyond end of badblocks 1921 * beyond write-intent bitmap 1922 */ 1923 if (rdev->sb_start + (32+4)*2 > new_offset) 1924 return 0; 1925 bitmap = rdev->mddev->bitmap; 1926 if (bitmap && !rdev->mddev->bitmap_info.file && 1927 rdev->sb_start + rdev->mddev->bitmap_info.offset + 1928 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) 1929 return 0; 1930 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) 1931 return 0; 1932 1933 return 1; 1934 } 1935 1936 static struct super_type super_types[] = { 1937 [0] = { 1938 .name = "0.90.0", 1939 .owner = THIS_MODULE, 1940 .load_super = super_90_load, 1941 .validate_super = super_90_validate, 1942 .sync_super = super_90_sync, 1943 .rdev_size_change = super_90_rdev_size_change, 1944 .allow_new_offset = super_90_allow_new_offset, 1945 }, 1946 [1] = { 1947 .name = "md-1", 1948 .owner = THIS_MODULE, 1949 .load_super = super_1_load, 1950 .validate_super = super_1_validate, 1951 .sync_super = super_1_sync, 1952 .rdev_size_change = super_1_rdev_size_change, 1953 .allow_new_offset = super_1_allow_new_offset, 1954 }, 1955 }; 1956 1957 static void sync_super(struct mddev *mddev, struct md_rdev *rdev) 1958 { 1959 if (mddev->sync_super) { 1960 mddev->sync_super(mddev, rdev); 1961 return; 1962 } 1963 1964 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); 1965 1966 super_types[mddev->major_version].sync_super(mddev, rdev); 1967 } 1968 1969 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) 1970 { 1971 struct md_rdev *rdev, *rdev2; 1972 1973 rcu_read_lock(); 1974 rdev_for_each_rcu(rdev, mddev1) 1975 rdev_for_each_rcu(rdev2, mddev2) 1976 if (rdev->bdev->bd_contains == 1977 rdev2->bdev->bd_contains) { 1978 rcu_read_unlock(); 1979 return 1; 1980 } 1981 rcu_read_unlock(); 1982 return 0; 1983 } 1984 1985 static LIST_HEAD(pending_raid_disks); 1986 1987 /* 1988 * Try to register data integrity profile for an mddev 1989 * 1990 * This is called when an array is started and after a disk has been kicked 1991 * from the array. It only succeeds if all working and active component devices 1992 * are integrity capable with matching profiles. 1993 */ 1994 int md_integrity_register(struct mddev *mddev) 1995 { 1996 struct md_rdev *rdev, *reference = NULL; 1997 1998 if (list_empty(&mddev->disks)) 1999 return 0; /* nothing to do */ 2000 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) 2001 return 0; /* shouldn't register, or already is */ 2002 rdev_for_each(rdev, mddev) { 2003 /* skip spares and non-functional disks */ 2004 if (test_bit(Faulty, &rdev->flags)) 2005 continue; 2006 if (rdev->raid_disk < 0) 2007 continue; 2008 if (!reference) { 2009 /* Use the first rdev as the reference */ 2010 reference = rdev; 2011 continue; 2012 } 2013 /* does this rdev's profile match the reference profile? */ 2014 if (blk_integrity_compare(reference->bdev->bd_disk, 2015 rdev->bdev->bd_disk) < 0) 2016 return -EINVAL; 2017 } 2018 if (!reference || !bdev_get_integrity(reference->bdev)) 2019 return 0; 2020 /* 2021 * All component devices are integrity capable and have matching 2022 * profiles, register the common profile for the md device. 2023 */ 2024 if (blk_integrity_register(mddev->gendisk, 2025 bdev_get_integrity(reference->bdev)) != 0) { 2026 printk(KERN_ERR "md: failed to register integrity for %s\n", 2027 mdname(mddev)); 2028 return -EINVAL; 2029 } 2030 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); 2031 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { 2032 printk(KERN_ERR "md: failed to create integrity pool for %s\n", 2033 mdname(mddev)); 2034 return -EINVAL; 2035 } 2036 return 0; 2037 } 2038 EXPORT_SYMBOL(md_integrity_register); 2039 2040 /* Disable data integrity if non-capable/non-matching disk is being added */ 2041 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) 2042 { 2043 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); 2044 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); 2045 2046 if (!bi_mddev) /* nothing to do */ 2047 return; 2048 if (rdev->raid_disk < 0) /* skip spares */ 2049 return; 2050 if (bi_rdev && blk_integrity_compare(mddev->gendisk, 2051 rdev->bdev->bd_disk) >= 0) 2052 return; 2053 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); 2054 blk_integrity_unregister(mddev->gendisk); 2055 } 2056 EXPORT_SYMBOL(md_integrity_add_rdev); 2057 2058 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev) 2059 { 2060 char b[BDEVNAME_SIZE]; 2061 struct kobject *ko; 2062 char *s; 2063 int err; 2064 2065 if (rdev->mddev) { 2066 MD_BUG(); 2067 return -EINVAL; 2068 } 2069 2070 /* prevent duplicates */ 2071 if (find_rdev(mddev, rdev->bdev->bd_dev)) 2072 return -EEXIST; 2073 2074 /* make sure rdev->sectors exceeds mddev->dev_sectors */ 2075 if (rdev->sectors && (mddev->dev_sectors == 0 || 2076 rdev->sectors < mddev->dev_sectors)) { 2077 if (mddev->pers) { 2078 /* Cannot change size, so fail 2079 * If mddev->level <= 0, then we don't care 2080 * about aligning sizes (e.g. linear) 2081 */ 2082 if (mddev->level > 0) 2083 return -ENOSPC; 2084 } else 2085 mddev->dev_sectors = rdev->sectors; 2086 } 2087 2088 /* Verify rdev->desc_nr is unique. 2089 * If it is -1, assign a free number, else 2090 * check number is not in use 2091 */ 2092 if (rdev->desc_nr < 0) { 2093 int choice = 0; 2094 if (mddev->pers) choice = mddev->raid_disks; 2095 while (find_rdev_nr(mddev, choice)) 2096 choice++; 2097 rdev->desc_nr = choice; 2098 } else { 2099 if (find_rdev_nr(mddev, rdev->desc_nr)) 2100 return -EBUSY; 2101 } 2102 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { 2103 printk(KERN_WARNING "md: %s: array is limited to %d devices\n", 2104 mdname(mddev), mddev->max_disks); 2105 return -EBUSY; 2106 } 2107 bdevname(rdev->bdev,b); 2108 while ( (s=strchr(b, '/')) != NULL) 2109 *s = '!'; 2110 2111 rdev->mddev = mddev; 2112 printk(KERN_INFO "md: bind<%s>\n", b); 2113 2114 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) 2115 goto fail; 2116 2117 ko = &part_to_dev(rdev->bdev->bd_part)->kobj; 2118 if (sysfs_create_link(&rdev->kobj, ko, "block")) 2119 /* failure here is OK */; 2120 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); 2121 2122 list_add_rcu(&rdev->same_set, &mddev->disks); 2123 bd_link_disk_holder(rdev->bdev, mddev->gendisk); 2124 2125 /* May as well allow recovery to be retried once */ 2126 mddev->recovery_disabled++; 2127 2128 return 0; 2129 2130 fail: 2131 printk(KERN_WARNING "md: failed to register dev-%s for %s\n", 2132 b, mdname(mddev)); 2133 return err; 2134 } 2135 2136 static void md_delayed_delete(struct work_struct *ws) 2137 { 2138 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); 2139 kobject_del(&rdev->kobj); 2140 kobject_put(&rdev->kobj); 2141 } 2142 2143 static void unbind_rdev_from_array(struct md_rdev * rdev) 2144 { 2145 char b[BDEVNAME_SIZE]; 2146 if (!rdev->mddev) { 2147 MD_BUG(); 2148 return; 2149 } 2150 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); 2151 list_del_rcu(&rdev->same_set); 2152 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 2153 rdev->mddev = NULL; 2154 sysfs_remove_link(&rdev->kobj, "block"); 2155 sysfs_put(rdev->sysfs_state); 2156 rdev->sysfs_state = NULL; 2157 rdev->badblocks.count = 0; 2158 /* We need to delay this, otherwise we can deadlock when 2159 * writing to 'remove' to "dev/state". We also need 2160 * to delay it due to rcu usage. 2161 */ 2162 synchronize_rcu(); 2163 INIT_WORK(&rdev->del_work, md_delayed_delete); 2164 kobject_get(&rdev->kobj); 2165 queue_work(md_misc_wq, &rdev->del_work); 2166 } 2167 2168 /* 2169 * prevent the device from being mounted, repartitioned or 2170 * otherwise reused by a RAID array (or any other kernel 2171 * subsystem), by bd_claiming the device. 2172 */ 2173 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) 2174 { 2175 int err = 0; 2176 struct block_device *bdev; 2177 char b[BDEVNAME_SIZE]; 2178 2179 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, 2180 shared ? (struct md_rdev *)lock_rdev : rdev); 2181 if (IS_ERR(bdev)) { 2182 printk(KERN_ERR "md: could not open %s.\n", 2183 __bdevname(dev, b)); 2184 return PTR_ERR(bdev); 2185 } 2186 rdev->bdev = bdev; 2187 return err; 2188 } 2189 2190 static void unlock_rdev(struct md_rdev *rdev) 2191 { 2192 struct block_device *bdev = rdev->bdev; 2193 rdev->bdev = NULL; 2194 if (!bdev) 2195 MD_BUG(); 2196 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2197 } 2198 2199 void md_autodetect_dev(dev_t dev); 2200 2201 static void export_rdev(struct md_rdev * rdev) 2202 { 2203 char b[BDEVNAME_SIZE]; 2204 printk(KERN_INFO "md: export_rdev(%s)\n", 2205 bdevname(rdev->bdev,b)); 2206 if (rdev->mddev) 2207 MD_BUG(); 2208 md_rdev_clear(rdev); 2209 #ifndef MODULE 2210 if (test_bit(AutoDetected, &rdev->flags)) 2211 md_autodetect_dev(rdev->bdev->bd_dev); 2212 #endif 2213 unlock_rdev(rdev); 2214 kobject_put(&rdev->kobj); 2215 } 2216 2217 static void kick_rdev_from_array(struct md_rdev * rdev) 2218 { 2219 unbind_rdev_from_array(rdev); 2220 export_rdev(rdev); 2221 } 2222 2223 static void export_array(struct mddev *mddev) 2224 { 2225 struct md_rdev *rdev, *tmp; 2226 2227 rdev_for_each_safe(rdev, tmp, mddev) { 2228 if (!rdev->mddev) { 2229 MD_BUG(); 2230 continue; 2231 } 2232 kick_rdev_from_array(rdev); 2233 } 2234 if (!list_empty(&mddev->disks)) 2235 MD_BUG(); 2236 mddev->raid_disks = 0; 2237 mddev->major_version = 0; 2238 } 2239 2240 static void print_desc(mdp_disk_t *desc) 2241 { 2242 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 2243 desc->major,desc->minor,desc->raid_disk,desc->state); 2244 } 2245 2246 static void print_sb_90(mdp_super_t *sb) 2247 { 2248 int i; 2249 2250 printk(KERN_INFO 2251 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 2252 sb->major_version, sb->minor_version, sb->patch_version, 2253 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 2254 sb->ctime); 2255 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 2256 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 2257 sb->md_minor, sb->layout, sb->chunk_size); 2258 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 2259 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 2260 sb->utime, sb->state, sb->active_disks, sb->working_disks, 2261 sb->failed_disks, sb->spare_disks, 2262 sb->sb_csum, (unsigned long)sb->events_lo); 2263 2264 printk(KERN_INFO); 2265 for (i = 0; i < MD_SB_DISKS; i++) { 2266 mdp_disk_t *desc; 2267 2268 desc = sb->disks + i; 2269 if (desc->number || desc->major || desc->minor || 2270 desc->raid_disk || (desc->state && (desc->state != 4))) { 2271 printk(" D %2d: ", i); 2272 print_desc(desc); 2273 } 2274 } 2275 printk(KERN_INFO "md: THIS: "); 2276 print_desc(&sb->this_disk); 2277 } 2278 2279 static void print_sb_1(struct mdp_superblock_1 *sb) 2280 { 2281 __u8 *uuid; 2282 2283 uuid = sb->set_uuid; 2284 printk(KERN_INFO 2285 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" 2286 "md: Name: \"%s\" CT:%llu\n", 2287 le32_to_cpu(sb->major_version), 2288 le32_to_cpu(sb->feature_map), 2289 uuid, 2290 sb->set_name, 2291 (unsigned long long)le64_to_cpu(sb->ctime) 2292 & MD_SUPERBLOCK_1_TIME_SEC_MASK); 2293 2294 uuid = sb->device_uuid; 2295 printk(KERN_INFO 2296 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" 2297 " RO:%llu\n" 2298 "md: Dev:%08x UUID: %pU\n" 2299 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" 2300 "md: (MaxDev:%u) \n", 2301 le32_to_cpu(sb->level), 2302 (unsigned long long)le64_to_cpu(sb->size), 2303 le32_to_cpu(sb->raid_disks), 2304 le32_to_cpu(sb->layout), 2305 le32_to_cpu(sb->chunksize), 2306 (unsigned long long)le64_to_cpu(sb->data_offset), 2307 (unsigned long long)le64_to_cpu(sb->data_size), 2308 (unsigned long long)le64_to_cpu(sb->super_offset), 2309 (unsigned long long)le64_to_cpu(sb->recovery_offset), 2310 le32_to_cpu(sb->dev_number), 2311 uuid, 2312 sb->devflags, 2313 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, 2314 (unsigned long long)le64_to_cpu(sb->events), 2315 (unsigned long long)le64_to_cpu(sb->resync_offset), 2316 le32_to_cpu(sb->sb_csum), 2317 le32_to_cpu(sb->max_dev) 2318 ); 2319 } 2320 2321 static void print_rdev(struct md_rdev *rdev, int major_version) 2322 { 2323 char b[BDEVNAME_SIZE]; 2324 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", 2325 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, 2326 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 2327 rdev->desc_nr); 2328 if (rdev->sb_loaded) { 2329 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); 2330 switch (major_version) { 2331 case 0: 2332 print_sb_90(page_address(rdev->sb_page)); 2333 break; 2334 case 1: 2335 print_sb_1(page_address(rdev->sb_page)); 2336 break; 2337 } 2338 } else 2339 printk(KERN_INFO "md: no rdev superblock!\n"); 2340 } 2341 2342 static void md_print_devices(void) 2343 { 2344 struct list_head *tmp; 2345 struct md_rdev *rdev; 2346 struct mddev *mddev; 2347 char b[BDEVNAME_SIZE]; 2348 2349 printk("\n"); 2350 printk("md: **********************************\n"); 2351 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 2352 printk("md: **********************************\n"); 2353 for_each_mddev(mddev, tmp) { 2354 2355 if (mddev->bitmap) 2356 bitmap_print_sb(mddev->bitmap); 2357 else 2358 printk("%s: ", mdname(mddev)); 2359 rdev_for_each(rdev, mddev) 2360 printk("<%s>", bdevname(rdev->bdev,b)); 2361 printk("\n"); 2362 2363 rdev_for_each(rdev, mddev) 2364 print_rdev(rdev, mddev->major_version); 2365 } 2366 printk("md: **********************************\n"); 2367 printk("\n"); 2368 } 2369 2370 2371 static void sync_sbs(struct mddev * mddev, int nospares) 2372 { 2373 /* Update each superblock (in-memory image), but 2374 * if we are allowed to, skip spares which already 2375 * have the right event counter, or have one earlier 2376 * (which would mean they aren't being marked as dirty 2377 * with the rest of the array) 2378 */ 2379 struct md_rdev *rdev; 2380 rdev_for_each(rdev, mddev) { 2381 if (rdev->sb_events == mddev->events || 2382 (nospares && 2383 rdev->raid_disk < 0 && 2384 rdev->sb_events+1 == mddev->events)) { 2385 /* Don't update this superblock */ 2386 rdev->sb_loaded = 2; 2387 } else { 2388 sync_super(mddev, rdev); 2389 rdev->sb_loaded = 1; 2390 } 2391 } 2392 } 2393 2394 static void md_update_sb(struct mddev * mddev, int force_change) 2395 { 2396 struct md_rdev *rdev; 2397 int sync_req; 2398 int nospares = 0; 2399 int any_badblocks_changed = 0; 2400 2401 repeat: 2402 /* First make sure individual recovery_offsets are correct */ 2403 rdev_for_each(rdev, mddev) { 2404 if (rdev->raid_disk >= 0 && 2405 mddev->delta_disks >= 0 && 2406 !test_bit(In_sync, &rdev->flags) && 2407 mddev->curr_resync_completed > rdev->recovery_offset) 2408 rdev->recovery_offset = mddev->curr_resync_completed; 2409 2410 } 2411 if (!mddev->persistent) { 2412 clear_bit(MD_CHANGE_CLEAN, &mddev->flags); 2413 clear_bit(MD_CHANGE_DEVS, &mddev->flags); 2414 if (!mddev->external) { 2415 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2416 rdev_for_each(rdev, mddev) { 2417 if (rdev->badblocks.changed) { 2418 rdev->badblocks.changed = 0; 2419 md_ack_all_badblocks(&rdev->badblocks); 2420 md_error(mddev, rdev); 2421 } 2422 clear_bit(Blocked, &rdev->flags); 2423 clear_bit(BlockedBadBlocks, &rdev->flags); 2424 wake_up(&rdev->blocked_wait); 2425 } 2426 } 2427 wake_up(&mddev->sb_wait); 2428 return; 2429 } 2430 2431 spin_lock_irq(&mddev->write_lock); 2432 2433 mddev->utime = get_seconds(); 2434 2435 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) 2436 force_change = 1; 2437 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) 2438 /* just a clean<-> dirty transition, possibly leave spares alone, 2439 * though if events isn't the right even/odd, we will have to do 2440 * spares after all 2441 */ 2442 nospares = 1; 2443 if (force_change) 2444 nospares = 0; 2445 if (mddev->degraded) 2446 /* If the array is degraded, then skipping spares is both 2447 * dangerous and fairly pointless. 2448 * Dangerous because a device that was removed from the array 2449 * might have a event_count that still looks up-to-date, 2450 * so it can be re-added without a resync. 2451 * Pointless because if there are any spares to skip, 2452 * then a recovery will happen and soon that array won't 2453 * be degraded any more and the spare can go back to sleep then. 2454 */ 2455 nospares = 0; 2456 2457 sync_req = mddev->in_sync; 2458 2459 /* If this is just a dirty<->clean transition, and the array is clean 2460 * and 'events' is odd, we can roll back to the previous clean state */ 2461 if (nospares 2462 && (mddev->in_sync && mddev->recovery_cp == MaxSector) 2463 && mddev->can_decrease_events 2464 && mddev->events != 1) { 2465 mddev->events--; 2466 mddev->can_decrease_events = 0; 2467 } else { 2468 /* otherwise we have to go forward and ... */ 2469 mddev->events ++; 2470 mddev->can_decrease_events = nospares; 2471 } 2472 2473 if (!mddev->events) { 2474 /* 2475 * oops, this 64-bit counter should never wrap. 2476 * Either we are in around ~1 trillion A.C., assuming 2477 * 1 reboot per second, or we have a bug: 2478 */ 2479 MD_BUG(); 2480 mddev->events --; 2481 } 2482 2483 rdev_for_each(rdev, mddev) { 2484 if (rdev->badblocks.changed) 2485 any_badblocks_changed++; 2486 if (test_bit(Faulty, &rdev->flags)) 2487 set_bit(FaultRecorded, &rdev->flags); 2488 } 2489 2490 sync_sbs(mddev, nospares); 2491 spin_unlock_irq(&mddev->write_lock); 2492 2493 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", 2494 mdname(mddev), mddev->in_sync); 2495 2496 bitmap_update_sb(mddev->bitmap); 2497 rdev_for_each(rdev, mddev) { 2498 char b[BDEVNAME_SIZE]; 2499 2500 if (rdev->sb_loaded != 1) 2501 continue; /* no noise on spare devices */ 2502 2503 if (!test_bit(Faulty, &rdev->flags) && 2504 rdev->saved_raid_disk == -1) { 2505 md_super_write(mddev,rdev, 2506 rdev->sb_start, rdev->sb_size, 2507 rdev->sb_page); 2508 pr_debug("md: (write) %s's sb offset: %llu\n", 2509 bdevname(rdev->bdev, b), 2510 (unsigned long long)rdev->sb_start); 2511 rdev->sb_events = mddev->events; 2512 if (rdev->badblocks.size) { 2513 md_super_write(mddev, rdev, 2514 rdev->badblocks.sector, 2515 rdev->badblocks.size << 9, 2516 rdev->bb_page); 2517 rdev->badblocks.size = 0; 2518 } 2519 2520 } else if (test_bit(Faulty, &rdev->flags)) 2521 pr_debug("md: %s (skipping faulty)\n", 2522 bdevname(rdev->bdev, b)); 2523 else 2524 pr_debug("(skipping incremental s/r "); 2525 2526 if (mddev->level == LEVEL_MULTIPATH) 2527 /* only need to write one superblock... */ 2528 break; 2529 } 2530 md_super_wait(mddev); 2531 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ 2532 2533 spin_lock_irq(&mddev->write_lock); 2534 if (mddev->in_sync != sync_req || 2535 test_bit(MD_CHANGE_DEVS, &mddev->flags)) { 2536 /* have to write it out again */ 2537 spin_unlock_irq(&mddev->write_lock); 2538 goto repeat; 2539 } 2540 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2541 spin_unlock_irq(&mddev->write_lock); 2542 wake_up(&mddev->sb_wait); 2543 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 2544 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 2545 2546 rdev_for_each(rdev, mddev) { 2547 if (test_and_clear_bit(FaultRecorded, &rdev->flags)) 2548 clear_bit(Blocked, &rdev->flags); 2549 2550 if (any_badblocks_changed) 2551 md_ack_all_badblocks(&rdev->badblocks); 2552 clear_bit(BlockedBadBlocks, &rdev->flags); 2553 wake_up(&rdev->blocked_wait); 2554 } 2555 } 2556 2557 /* words written to sysfs files may, or may not, be \n terminated. 2558 * We want to accept with case. For this we use cmd_match. 2559 */ 2560 static int cmd_match(const char *cmd, const char *str) 2561 { 2562 /* See if cmd, written into a sysfs file, matches 2563 * str. They must either be the same, or cmd can 2564 * have a trailing newline 2565 */ 2566 while (*cmd && *str && *cmd == *str) { 2567 cmd++; 2568 str++; 2569 } 2570 if (*cmd == '\n') 2571 cmd++; 2572 if (*str || *cmd) 2573 return 0; 2574 return 1; 2575 } 2576 2577 struct rdev_sysfs_entry { 2578 struct attribute attr; 2579 ssize_t (*show)(struct md_rdev *, char *); 2580 ssize_t (*store)(struct md_rdev *, const char *, size_t); 2581 }; 2582 2583 static ssize_t 2584 state_show(struct md_rdev *rdev, char *page) 2585 { 2586 char *sep = ""; 2587 size_t len = 0; 2588 2589 if (test_bit(Faulty, &rdev->flags) || 2590 rdev->badblocks.unacked_exist) { 2591 len+= sprintf(page+len, "%sfaulty",sep); 2592 sep = ","; 2593 } 2594 if (test_bit(In_sync, &rdev->flags)) { 2595 len += sprintf(page+len, "%sin_sync",sep); 2596 sep = ","; 2597 } 2598 if (test_bit(WriteMostly, &rdev->flags)) { 2599 len += sprintf(page+len, "%swrite_mostly",sep); 2600 sep = ","; 2601 } 2602 if (test_bit(Blocked, &rdev->flags) || 2603 (rdev->badblocks.unacked_exist 2604 && !test_bit(Faulty, &rdev->flags))) { 2605 len += sprintf(page+len, "%sblocked", sep); 2606 sep = ","; 2607 } 2608 if (!test_bit(Faulty, &rdev->flags) && 2609 !test_bit(In_sync, &rdev->flags)) { 2610 len += sprintf(page+len, "%sspare", sep); 2611 sep = ","; 2612 } 2613 if (test_bit(WriteErrorSeen, &rdev->flags)) { 2614 len += sprintf(page+len, "%swrite_error", sep); 2615 sep = ","; 2616 } 2617 if (test_bit(WantReplacement, &rdev->flags)) { 2618 len += sprintf(page+len, "%swant_replacement", sep); 2619 sep = ","; 2620 } 2621 if (test_bit(Replacement, &rdev->flags)) { 2622 len += sprintf(page+len, "%sreplacement", sep); 2623 sep = ","; 2624 } 2625 2626 return len+sprintf(page+len, "\n"); 2627 } 2628 2629 static ssize_t 2630 state_store(struct md_rdev *rdev, const char *buf, size_t len) 2631 { 2632 /* can write 2633 * faulty - simulates an error 2634 * remove - disconnects the device 2635 * writemostly - sets write_mostly 2636 * -writemostly - clears write_mostly 2637 * blocked - sets the Blocked flags 2638 * -blocked - clears the Blocked and possibly simulates an error 2639 * insync - sets Insync providing device isn't active 2640 * write_error - sets WriteErrorSeen 2641 * -write_error - clears WriteErrorSeen 2642 */ 2643 int err = -EINVAL; 2644 if (cmd_match(buf, "faulty") && rdev->mddev->pers) { 2645 md_error(rdev->mddev, rdev); 2646 if (test_bit(Faulty, &rdev->flags)) 2647 err = 0; 2648 else 2649 err = -EBUSY; 2650 } else if (cmd_match(buf, "remove")) { 2651 if (rdev->raid_disk >= 0) 2652 err = -EBUSY; 2653 else { 2654 struct mddev *mddev = rdev->mddev; 2655 kick_rdev_from_array(rdev); 2656 if (mddev->pers) 2657 md_update_sb(mddev, 1); 2658 md_new_event(mddev); 2659 err = 0; 2660 } 2661 } else if (cmd_match(buf, "writemostly")) { 2662 set_bit(WriteMostly, &rdev->flags); 2663 err = 0; 2664 } else if (cmd_match(buf, "-writemostly")) { 2665 clear_bit(WriteMostly, &rdev->flags); 2666 err = 0; 2667 } else if (cmd_match(buf, "blocked")) { 2668 set_bit(Blocked, &rdev->flags); 2669 err = 0; 2670 } else if (cmd_match(buf, "-blocked")) { 2671 if (!test_bit(Faulty, &rdev->flags) && 2672 rdev->badblocks.unacked_exist) { 2673 /* metadata handler doesn't understand badblocks, 2674 * so we need to fail the device 2675 */ 2676 md_error(rdev->mddev, rdev); 2677 } 2678 clear_bit(Blocked, &rdev->flags); 2679 clear_bit(BlockedBadBlocks, &rdev->flags); 2680 wake_up(&rdev->blocked_wait); 2681 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2682 md_wakeup_thread(rdev->mddev->thread); 2683 2684 err = 0; 2685 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { 2686 set_bit(In_sync, &rdev->flags); 2687 err = 0; 2688 } else if (cmd_match(buf, "write_error")) { 2689 set_bit(WriteErrorSeen, &rdev->flags); 2690 err = 0; 2691 } else if (cmd_match(buf, "-write_error")) { 2692 clear_bit(WriteErrorSeen, &rdev->flags); 2693 err = 0; 2694 } else if (cmd_match(buf, "want_replacement")) { 2695 /* Any non-spare device that is not a replacement can 2696 * become want_replacement at any time, but we then need to 2697 * check if recovery is needed. 2698 */ 2699 if (rdev->raid_disk >= 0 && 2700 !test_bit(Replacement, &rdev->flags)) 2701 set_bit(WantReplacement, &rdev->flags); 2702 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2703 md_wakeup_thread(rdev->mddev->thread); 2704 err = 0; 2705 } else if (cmd_match(buf, "-want_replacement")) { 2706 /* Clearing 'want_replacement' is always allowed. 2707 * Once replacements starts it is too late though. 2708 */ 2709 err = 0; 2710 clear_bit(WantReplacement, &rdev->flags); 2711 } else if (cmd_match(buf, "replacement")) { 2712 /* Can only set a device as a replacement when array has not 2713 * yet been started. Once running, replacement is automatic 2714 * from spares, or by assigning 'slot'. 2715 */ 2716 if (rdev->mddev->pers) 2717 err = -EBUSY; 2718 else { 2719 set_bit(Replacement, &rdev->flags); 2720 err = 0; 2721 } 2722 } else if (cmd_match(buf, "-replacement")) { 2723 /* Similarly, can only clear Replacement before start */ 2724 if (rdev->mddev->pers) 2725 err = -EBUSY; 2726 else { 2727 clear_bit(Replacement, &rdev->flags); 2728 err = 0; 2729 } 2730 } 2731 if (!err) 2732 sysfs_notify_dirent_safe(rdev->sysfs_state); 2733 return err ? err : len; 2734 } 2735 static struct rdev_sysfs_entry rdev_state = 2736 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); 2737 2738 static ssize_t 2739 errors_show(struct md_rdev *rdev, char *page) 2740 { 2741 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); 2742 } 2743 2744 static ssize_t 2745 errors_store(struct md_rdev *rdev, const char *buf, size_t len) 2746 { 2747 char *e; 2748 unsigned long n = simple_strtoul(buf, &e, 10); 2749 if (*buf && (*e == 0 || *e == '\n')) { 2750 atomic_set(&rdev->corrected_errors, n); 2751 return len; 2752 } 2753 return -EINVAL; 2754 } 2755 static struct rdev_sysfs_entry rdev_errors = 2756 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); 2757 2758 static ssize_t 2759 slot_show(struct md_rdev *rdev, char *page) 2760 { 2761 if (rdev->raid_disk < 0) 2762 return sprintf(page, "none\n"); 2763 else 2764 return sprintf(page, "%d\n", rdev->raid_disk); 2765 } 2766 2767 static ssize_t 2768 slot_store(struct md_rdev *rdev, const char *buf, size_t len) 2769 { 2770 char *e; 2771 int err; 2772 int slot = simple_strtoul(buf, &e, 10); 2773 if (strncmp(buf, "none", 4)==0) 2774 slot = -1; 2775 else if (e==buf || (*e && *e!= '\n')) 2776 return -EINVAL; 2777 if (rdev->mddev->pers && slot == -1) { 2778 /* Setting 'slot' on an active array requires also 2779 * updating the 'rd%d' link, and communicating 2780 * with the personality with ->hot_*_disk. 2781 * For now we only support removing 2782 * failed/spare devices. This normally happens automatically, 2783 * but not when the metadata is externally managed. 2784 */ 2785 if (rdev->raid_disk == -1) 2786 return -EEXIST; 2787 /* personality does all needed checks */ 2788 if (rdev->mddev->pers->hot_remove_disk == NULL) 2789 return -EINVAL; 2790 err = rdev->mddev->pers-> 2791 hot_remove_disk(rdev->mddev, rdev); 2792 if (err) 2793 return err; 2794 sysfs_unlink_rdev(rdev->mddev, rdev); 2795 rdev->raid_disk = -1; 2796 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2797 md_wakeup_thread(rdev->mddev->thread); 2798 } else if (rdev->mddev->pers) { 2799 /* Activating a spare .. or possibly reactivating 2800 * if we ever get bitmaps working here. 2801 */ 2802 2803 if (rdev->raid_disk != -1) 2804 return -EBUSY; 2805 2806 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) 2807 return -EBUSY; 2808 2809 if (rdev->mddev->pers->hot_add_disk == NULL) 2810 return -EINVAL; 2811 2812 if (slot >= rdev->mddev->raid_disks && 2813 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) 2814 return -ENOSPC; 2815 2816 rdev->raid_disk = slot; 2817 if (test_bit(In_sync, &rdev->flags)) 2818 rdev->saved_raid_disk = slot; 2819 else 2820 rdev->saved_raid_disk = -1; 2821 clear_bit(In_sync, &rdev->flags); 2822 err = rdev->mddev->pers-> 2823 hot_add_disk(rdev->mddev, rdev); 2824 if (err) { 2825 rdev->raid_disk = -1; 2826 return err; 2827 } else 2828 sysfs_notify_dirent_safe(rdev->sysfs_state); 2829 if (sysfs_link_rdev(rdev->mddev, rdev)) 2830 /* failure here is OK */; 2831 /* don't wakeup anyone, leave that to userspace. */ 2832 } else { 2833 if (slot >= rdev->mddev->raid_disks && 2834 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) 2835 return -ENOSPC; 2836 rdev->raid_disk = slot; 2837 /* assume it is working */ 2838 clear_bit(Faulty, &rdev->flags); 2839 clear_bit(WriteMostly, &rdev->flags); 2840 set_bit(In_sync, &rdev->flags); 2841 sysfs_notify_dirent_safe(rdev->sysfs_state); 2842 } 2843 return len; 2844 } 2845 2846 2847 static struct rdev_sysfs_entry rdev_slot = 2848 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); 2849 2850 static ssize_t 2851 offset_show(struct md_rdev *rdev, char *page) 2852 { 2853 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); 2854 } 2855 2856 static ssize_t 2857 offset_store(struct md_rdev *rdev, const char *buf, size_t len) 2858 { 2859 unsigned long long offset; 2860 if (strict_strtoull(buf, 10, &offset) < 0) 2861 return -EINVAL; 2862 if (rdev->mddev->pers && rdev->raid_disk >= 0) 2863 return -EBUSY; 2864 if (rdev->sectors && rdev->mddev->external) 2865 /* Must set offset before size, so overlap checks 2866 * can be sane */ 2867 return -EBUSY; 2868 rdev->data_offset = offset; 2869 rdev->new_data_offset = offset; 2870 return len; 2871 } 2872 2873 static struct rdev_sysfs_entry rdev_offset = 2874 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); 2875 2876 static ssize_t new_offset_show(struct md_rdev *rdev, char *page) 2877 { 2878 return sprintf(page, "%llu\n", 2879 (unsigned long long)rdev->new_data_offset); 2880 } 2881 2882 static ssize_t new_offset_store(struct md_rdev *rdev, 2883 const char *buf, size_t len) 2884 { 2885 unsigned long long new_offset; 2886 struct mddev *mddev = rdev->mddev; 2887 2888 if (strict_strtoull(buf, 10, &new_offset) < 0) 2889 return -EINVAL; 2890 2891 if (mddev->sync_thread) 2892 return -EBUSY; 2893 if (new_offset == rdev->data_offset) 2894 /* reset is always permitted */ 2895 ; 2896 else if (new_offset > rdev->data_offset) { 2897 /* must not push array size beyond rdev_sectors */ 2898 if (new_offset - rdev->data_offset 2899 + mddev->dev_sectors > rdev->sectors) 2900 return -E2BIG; 2901 } 2902 /* Metadata worries about other space details. */ 2903 2904 /* decreasing the offset is inconsistent with a backwards 2905 * reshape. 2906 */ 2907 if (new_offset < rdev->data_offset && 2908 mddev->reshape_backwards) 2909 return -EINVAL; 2910 /* Increasing offset is inconsistent with forwards 2911 * reshape. reshape_direction should be set to 2912 * 'backwards' first. 2913 */ 2914 if (new_offset > rdev->data_offset && 2915 !mddev->reshape_backwards) 2916 return -EINVAL; 2917 2918 if (mddev->pers && mddev->persistent && 2919 !super_types[mddev->major_version] 2920 .allow_new_offset(rdev, new_offset)) 2921 return -E2BIG; 2922 rdev->new_data_offset = new_offset; 2923 if (new_offset > rdev->data_offset) 2924 mddev->reshape_backwards = 1; 2925 else if (new_offset < rdev->data_offset) 2926 mddev->reshape_backwards = 0; 2927 2928 return len; 2929 } 2930 static struct rdev_sysfs_entry rdev_new_offset = 2931 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); 2932 2933 static ssize_t 2934 rdev_size_show(struct md_rdev *rdev, char *page) 2935 { 2936 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); 2937 } 2938 2939 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) 2940 { 2941 /* check if two start/length pairs overlap */ 2942 if (s1+l1 <= s2) 2943 return 0; 2944 if (s2+l2 <= s1) 2945 return 0; 2946 return 1; 2947 } 2948 2949 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) 2950 { 2951 unsigned long long blocks; 2952 sector_t new; 2953 2954 if (strict_strtoull(buf, 10, &blocks) < 0) 2955 return -EINVAL; 2956 2957 if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) 2958 return -EINVAL; /* sector conversion overflow */ 2959 2960 new = blocks * 2; 2961 if (new != blocks * 2) 2962 return -EINVAL; /* unsigned long long to sector_t overflow */ 2963 2964 *sectors = new; 2965 return 0; 2966 } 2967 2968 static ssize_t 2969 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) 2970 { 2971 struct mddev *my_mddev = rdev->mddev; 2972 sector_t oldsectors = rdev->sectors; 2973 sector_t sectors; 2974 2975 if (strict_blocks_to_sectors(buf, §ors) < 0) 2976 return -EINVAL; 2977 if (rdev->data_offset != rdev->new_data_offset) 2978 return -EINVAL; /* too confusing */ 2979 if (my_mddev->pers && rdev->raid_disk >= 0) { 2980 if (my_mddev->persistent) { 2981 sectors = super_types[my_mddev->major_version]. 2982 rdev_size_change(rdev, sectors); 2983 if (!sectors) 2984 return -EBUSY; 2985 } else if (!sectors) 2986 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 2987 rdev->data_offset; 2988 } 2989 if (sectors < my_mddev->dev_sectors) 2990 return -EINVAL; /* component must fit device */ 2991 2992 rdev->sectors = sectors; 2993 if (sectors > oldsectors && my_mddev->external) { 2994 /* need to check that all other rdevs with the same ->bdev 2995 * do not overlap. We need to unlock the mddev to avoid 2996 * a deadlock. We have already changed rdev->sectors, and if 2997 * we have to change it back, we will have the lock again. 2998 */ 2999 struct mddev *mddev; 3000 int overlap = 0; 3001 struct list_head *tmp; 3002 3003 mddev_unlock(my_mddev); 3004 for_each_mddev(mddev, tmp) { 3005 struct md_rdev *rdev2; 3006 3007 mddev_lock(mddev); 3008 rdev_for_each(rdev2, mddev) 3009 if (rdev->bdev == rdev2->bdev && 3010 rdev != rdev2 && 3011 overlaps(rdev->data_offset, rdev->sectors, 3012 rdev2->data_offset, 3013 rdev2->sectors)) { 3014 overlap = 1; 3015 break; 3016 } 3017 mddev_unlock(mddev); 3018 if (overlap) { 3019 mddev_put(mddev); 3020 break; 3021 } 3022 } 3023 mddev_lock(my_mddev); 3024 if (overlap) { 3025 /* Someone else could have slipped in a size 3026 * change here, but doing so is just silly. 3027 * We put oldsectors back because we *know* it is 3028 * safe, and trust userspace not to race with 3029 * itself 3030 */ 3031 rdev->sectors = oldsectors; 3032 return -EBUSY; 3033 } 3034 } 3035 return len; 3036 } 3037 3038 static struct rdev_sysfs_entry rdev_size = 3039 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); 3040 3041 3042 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) 3043 { 3044 unsigned long long recovery_start = rdev->recovery_offset; 3045 3046 if (test_bit(In_sync, &rdev->flags) || 3047 recovery_start == MaxSector) 3048 return sprintf(page, "none\n"); 3049 3050 return sprintf(page, "%llu\n", recovery_start); 3051 } 3052 3053 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) 3054 { 3055 unsigned long long recovery_start; 3056 3057 if (cmd_match(buf, "none")) 3058 recovery_start = MaxSector; 3059 else if (strict_strtoull(buf, 10, &recovery_start)) 3060 return -EINVAL; 3061 3062 if (rdev->mddev->pers && 3063 rdev->raid_disk >= 0) 3064 return -EBUSY; 3065 3066 rdev->recovery_offset = recovery_start; 3067 if (recovery_start == MaxSector) 3068 set_bit(In_sync, &rdev->flags); 3069 else 3070 clear_bit(In_sync, &rdev->flags); 3071 return len; 3072 } 3073 3074 static struct rdev_sysfs_entry rdev_recovery_start = 3075 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); 3076 3077 3078 static ssize_t 3079 badblocks_show(struct badblocks *bb, char *page, int unack); 3080 static ssize_t 3081 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); 3082 3083 static ssize_t bb_show(struct md_rdev *rdev, char *page) 3084 { 3085 return badblocks_show(&rdev->badblocks, page, 0); 3086 } 3087 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) 3088 { 3089 int rv = badblocks_store(&rdev->badblocks, page, len, 0); 3090 /* Maybe that ack was all we needed */ 3091 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) 3092 wake_up(&rdev->blocked_wait); 3093 return rv; 3094 } 3095 static struct rdev_sysfs_entry rdev_bad_blocks = 3096 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); 3097 3098 3099 static ssize_t ubb_show(struct md_rdev *rdev, char *page) 3100 { 3101 return badblocks_show(&rdev->badblocks, page, 1); 3102 } 3103 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) 3104 { 3105 return badblocks_store(&rdev->badblocks, page, len, 1); 3106 } 3107 static struct rdev_sysfs_entry rdev_unack_bad_blocks = 3108 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); 3109 3110 static struct attribute *rdev_default_attrs[] = { 3111 &rdev_state.attr, 3112 &rdev_errors.attr, 3113 &rdev_slot.attr, 3114 &rdev_offset.attr, 3115 &rdev_new_offset.attr, 3116 &rdev_size.attr, 3117 &rdev_recovery_start.attr, 3118 &rdev_bad_blocks.attr, 3119 &rdev_unack_bad_blocks.attr, 3120 NULL, 3121 }; 3122 static ssize_t 3123 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 3124 { 3125 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 3126 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); 3127 struct mddev *mddev = rdev->mddev; 3128 ssize_t rv; 3129 3130 if (!entry->show) 3131 return -EIO; 3132 3133 rv = mddev ? mddev_lock(mddev) : -EBUSY; 3134 if (!rv) { 3135 if (rdev->mddev == NULL) 3136 rv = -EBUSY; 3137 else 3138 rv = entry->show(rdev, page); 3139 mddev_unlock(mddev); 3140 } 3141 return rv; 3142 } 3143 3144 static ssize_t 3145 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 3146 const char *page, size_t length) 3147 { 3148 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 3149 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); 3150 ssize_t rv; 3151 struct mddev *mddev = rdev->mddev; 3152 3153 if (!entry->store) 3154 return -EIO; 3155 if (!capable(CAP_SYS_ADMIN)) 3156 return -EACCES; 3157 rv = mddev ? mddev_lock(mddev): -EBUSY; 3158 if (!rv) { 3159 if (rdev->mddev == NULL) 3160 rv = -EBUSY; 3161 else 3162 rv = entry->store(rdev, page, length); 3163 mddev_unlock(mddev); 3164 } 3165 return rv; 3166 } 3167 3168 static void rdev_free(struct kobject *ko) 3169 { 3170 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); 3171 kfree(rdev); 3172 } 3173 static const struct sysfs_ops rdev_sysfs_ops = { 3174 .show = rdev_attr_show, 3175 .store = rdev_attr_store, 3176 }; 3177 static struct kobj_type rdev_ktype = { 3178 .release = rdev_free, 3179 .sysfs_ops = &rdev_sysfs_ops, 3180 .default_attrs = rdev_default_attrs, 3181 }; 3182 3183 int md_rdev_init(struct md_rdev *rdev) 3184 { 3185 rdev->desc_nr = -1; 3186 rdev->saved_raid_disk = -1; 3187 rdev->raid_disk = -1; 3188 rdev->flags = 0; 3189 rdev->data_offset = 0; 3190 rdev->new_data_offset = 0; 3191 rdev->sb_events = 0; 3192 rdev->last_read_error.tv_sec = 0; 3193 rdev->last_read_error.tv_nsec = 0; 3194 rdev->sb_loaded = 0; 3195 rdev->bb_page = NULL; 3196 atomic_set(&rdev->nr_pending, 0); 3197 atomic_set(&rdev->read_errors, 0); 3198 atomic_set(&rdev->corrected_errors, 0); 3199 3200 INIT_LIST_HEAD(&rdev->same_set); 3201 init_waitqueue_head(&rdev->blocked_wait); 3202 3203 /* Add space to store bad block list. 3204 * This reserves the space even on arrays where it cannot 3205 * be used - I wonder if that matters 3206 */ 3207 rdev->badblocks.count = 0; 3208 rdev->badblocks.shift = 0; 3209 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); 3210 seqlock_init(&rdev->badblocks.lock); 3211 if (rdev->badblocks.page == NULL) 3212 return -ENOMEM; 3213 3214 return 0; 3215 } 3216 EXPORT_SYMBOL_GPL(md_rdev_init); 3217 /* 3218 * Import a device. If 'super_format' >= 0, then sanity check the superblock 3219 * 3220 * mark the device faulty if: 3221 * 3222 * - the device is nonexistent (zero size) 3223 * - the device has no valid superblock 3224 * 3225 * a faulty rdev _never_ has rdev->sb set. 3226 */ 3227 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) 3228 { 3229 char b[BDEVNAME_SIZE]; 3230 int err; 3231 struct md_rdev *rdev; 3232 sector_t size; 3233 3234 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); 3235 if (!rdev) { 3236 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 3237 return ERR_PTR(-ENOMEM); 3238 } 3239 3240 err = md_rdev_init(rdev); 3241 if (err) 3242 goto abort_free; 3243 err = alloc_disk_sb(rdev); 3244 if (err) 3245 goto abort_free; 3246 3247 err = lock_rdev(rdev, newdev, super_format == -2); 3248 if (err) 3249 goto abort_free; 3250 3251 kobject_init(&rdev->kobj, &rdev_ktype); 3252 3253 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; 3254 if (!size) { 3255 printk(KERN_WARNING 3256 "md: %s has zero or unknown size, marking faulty!\n", 3257 bdevname(rdev->bdev,b)); 3258 err = -EINVAL; 3259 goto abort_free; 3260 } 3261 3262 if (super_format >= 0) { 3263 err = super_types[super_format]. 3264 load_super(rdev, NULL, super_minor); 3265 if (err == -EINVAL) { 3266 printk(KERN_WARNING 3267 "md: %s does not have a valid v%d.%d " 3268 "superblock, not importing!\n", 3269 bdevname(rdev->bdev,b), 3270 super_format, super_minor); 3271 goto abort_free; 3272 } 3273 if (err < 0) { 3274 printk(KERN_WARNING 3275 "md: could not read %s's sb, not importing!\n", 3276 bdevname(rdev->bdev,b)); 3277 goto abort_free; 3278 } 3279 } 3280 if (super_format == -1) 3281 /* hot-add for 0.90, or non-persistent: so no badblocks */ 3282 rdev->badblocks.shift = -1; 3283 3284 return rdev; 3285 3286 abort_free: 3287 if (rdev->bdev) 3288 unlock_rdev(rdev); 3289 md_rdev_clear(rdev); 3290 kfree(rdev); 3291 return ERR_PTR(err); 3292 } 3293 3294 /* 3295 * Check a full RAID array for plausibility 3296 */ 3297 3298 3299 static void analyze_sbs(struct mddev * mddev) 3300 { 3301 int i; 3302 struct md_rdev *rdev, *freshest, *tmp; 3303 char b[BDEVNAME_SIZE]; 3304 3305 freshest = NULL; 3306 rdev_for_each_safe(rdev, tmp, mddev) 3307 switch (super_types[mddev->major_version]. 3308 load_super(rdev, freshest, mddev->minor_version)) { 3309 case 1: 3310 freshest = rdev; 3311 break; 3312 case 0: 3313 break; 3314 default: 3315 printk( KERN_ERR \ 3316 "md: fatal superblock inconsistency in %s" 3317 " -- removing from array\n", 3318 bdevname(rdev->bdev,b)); 3319 kick_rdev_from_array(rdev); 3320 } 3321 3322 3323 super_types[mddev->major_version]. 3324 validate_super(mddev, freshest); 3325 3326 i = 0; 3327 rdev_for_each_safe(rdev, tmp, mddev) { 3328 if (mddev->max_disks && 3329 (rdev->desc_nr >= mddev->max_disks || 3330 i > mddev->max_disks)) { 3331 printk(KERN_WARNING 3332 "md: %s: %s: only %d devices permitted\n", 3333 mdname(mddev), bdevname(rdev->bdev, b), 3334 mddev->max_disks); 3335 kick_rdev_from_array(rdev); 3336 continue; 3337 } 3338 if (rdev != freshest) 3339 if (super_types[mddev->major_version]. 3340 validate_super(mddev, rdev)) { 3341 printk(KERN_WARNING "md: kicking non-fresh %s" 3342 " from array!\n", 3343 bdevname(rdev->bdev,b)); 3344 kick_rdev_from_array(rdev); 3345 continue; 3346 } 3347 if (mddev->level == LEVEL_MULTIPATH) { 3348 rdev->desc_nr = i++; 3349 rdev->raid_disk = rdev->desc_nr; 3350 set_bit(In_sync, &rdev->flags); 3351 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { 3352 rdev->raid_disk = -1; 3353 clear_bit(In_sync, &rdev->flags); 3354 } 3355 } 3356 } 3357 3358 /* Read a fixed-point number. 3359 * Numbers in sysfs attributes should be in "standard" units where 3360 * possible, so time should be in seconds. 3361 * However we internally use a a much smaller unit such as 3362 * milliseconds or jiffies. 3363 * This function takes a decimal number with a possible fractional 3364 * component, and produces an integer which is the result of 3365 * multiplying that number by 10^'scale'. 3366 * all without any floating-point arithmetic. 3367 */ 3368 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) 3369 { 3370 unsigned long result = 0; 3371 long decimals = -1; 3372 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { 3373 if (*cp == '.') 3374 decimals = 0; 3375 else if (decimals < scale) { 3376 unsigned int value; 3377 value = *cp - '0'; 3378 result = result * 10 + value; 3379 if (decimals >= 0) 3380 decimals++; 3381 } 3382 cp++; 3383 } 3384 if (*cp == '\n') 3385 cp++; 3386 if (*cp) 3387 return -EINVAL; 3388 if (decimals < 0) 3389 decimals = 0; 3390 while (decimals < scale) { 3391 result *= 10; 3392 decimals ++; 3393 } 3394 *res = result; 3395 return 0; 3396 } 3397 3398 3399 static void md_safemode_timeout(unsigned long data); 3400 3401 static ssize_t 3402 safe_delay_show(struct mddev *mddev, char *page) 3403 { 3404 int msec = (mddev->safemode_delay*1000)/HZ; 3405 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); 3406 } 3407 static ssize_t 3408 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) 3409 { 3410 unsigned long msec; 3411 3412 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) 3413 return -EINVAL; 3414 if (msec == 0) 3415 mddev->safemode_delay = 0; 3416 else { 3417 unsigned long old_delay = mddev->safemode_delay; 3418 mddev->safemode_delay = (msec*HZ)/1000; 3419 if (mddev->safemode_delay == 0) 3420 mddev->safemode_delay = 1; 3421 if (mddev->safemode_delay < old_delay) 3422 md_safemode_timeout((unsigned long)mddev); 3423 } 3424 return len; 3425 } 3426 static struct md_sysfs_entry md_safe_delay = 3427 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); 3428 3429 static ssize_t 3430 level_show(struct mddev *mddev, char *page) 3431 { 3432 struct md_personality *p = mddev->pers; 3433 if (p) 3434 return sprintf(page, "%s\n", p->name); 3435 else if (mddev->clevel[0]) 3436 return sprintf(page, "%s\n", mddev->clevel); 3437 else if (mddev->level != LEVEL_NONE) 3438 return sprintf(page, "%d\n", mddev->level); 3439 else 3440 return 0; 3441 } 3442 3443 static ssize_t 3444 level_store(struct mddev *mddev, const char *buf, size_t len) 3445 { 3446 char clevel[16]; 3447 ssize_t rv = len; 3448 struct md_personality *pers; 3449 long level; 3450 void *priv; 3451 struct md_rdev *rdev; 3452 3453 if (mddev->pers == NULL) { 3454 if (len == 0) 3455 return 0; 3456 if (len >= sizeof(mddev->clevel)) 3457 return -ENOSPC; 3458 strncpy(mddev->clevel, buf, len); 3459 if (mddev->clevel[len-1] == '\n') 3460 len--; 3461 mddev->clevel[len] = 0; 3462 mddev->level = LEVEL_NONE; 3463 return rv; 3464 } 3465 3466 /* request to change the personality. Need to ensure: 3467 * - array is not engaged in resync/recovery/reshape 3468 * - old personality can be suspended 3469 * - new personality will access other array. 3470 */ 3471 3472 if (mddev->sync_thread || 3473 mddev->reshape_position != MaxSector || 3474 mddev->sysfs_active) 3475 return -EBUSY; 3476 3477 if (!mddev->pers->quiesce) { 3478 printk(KERN_WARNING "md: %s: %s does not support online personality change\n", 3479 mdname(mddev), mddev->pers->name); 3480 return -EINVAL; 3481 } 3482 3483 /* Now find the new personality */ 3484 if (len == 0 || len >= sizeof(clevel)) 3485 return -EINVAL; 3486 strncpy(clevel, buf, len); 3487 if (clevel[len-1] == '\n') 3488 len--; 3489 clevel[len] = 0; 3490 if (strict_strtol(clevel, 10, &level)) 3491 level = LEVEL_NONE; 3492 3493 if (request_module("md-%s", clevel) != 0) 3494 request_module("md-level-%s", clevel); 3495 spin_lock(&pers_lock); 3496 pers = find_pers(level, clevel); 3497 if (!pers || !try_module_get(pers->owner)) { 3498 spin_unlock(&pers_lock); 3499 printk(KERN_WARNING "md: personality %s not loaded\n", clevel); 3500 return -EINVAL; 3501 } 3502 spin_unlock(&pers_lock); 3503 3504 if (pers == mddev->pers) { 3505 /* Nothing to do! */ 3506 module_put(pers->owner); 3507 return rv; 3508 } 3509 if (!pers->takeover) { 3510 module_put(pers->owner); 3511 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", 3512 mdname(mddev), clevel); 3513 return -EINVAL; 3514 } 3515 3516 rdev_for_each(rdev, mddev) 3517 rdev->new_raid_disk = rdev->raid_disk; 3518 3519 /* ->takeover must set new_* and/or delta_disks 3520 * if it succeeds, and may set them when it fails. 3521 */ 3522 priv = pers->takeover(mddev); 3523 if (IS_ERR(priv)) { 3524 mddev->new_level = mddev->level; 3525 mddev->new_layout = mddev->layout; 3526 mddev->new_chunk_sectors = mddev->chunk_sectors; 3527 mddev->raid_disks -= mddev->delta_disks; 3528 mddev->delta_disks = 0; 3529 mddev->reshape_backwards = 0; 3530 module_put(pers->owner); 3531 printk(KERN_WARNING "md: %s: %s would not accept array\n", 3532 mdname(mddev), clevel); 3533 return PTR_ERR(priv); 3534 } 3535 3536 /* Looks like we have a winner */ 3537 mddev_suspend(mddev); 3538 mddev->pers->stop(mddev); 3539 3540 if (mddev->pers->sync_request == NULL && 3541 pers->sync_request != NULL) { 3542 /* need to add the md_redundancy_group */ 3543 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 3544 printk(KERN_WARNING 3545 "md: cannot register extra attributes for %s\n", 3546 mdname(mddev)); 3547 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); 3548 } 3549 if (mddev->pers->sync_request != NULL && 3550 pers->sync_request == NULL) { 3551 /* need to remove the md_redundancy_group */ 3552 if (mddev->to_remove == NULL) 3553 mddev->to_remove = &md_redundancy_group; 3554 } 3555 3556 if (mddev->pers->sync_request == NULL && 3557 mddev->external) { 3558 /* We are converting from a no-redundancy array 3559 * to a redundancy array and metadata is managed 3560 * externally so we need to be sure that writes 3561 * won't block due to a need to transition 3562 * clean->dirty 3563 * until external management is started. 3564 */ 3565 mddev->in_sync = 0; 3566 mddev->safemode_delay = 0; 3567 mddev->safemode = 0; 3568 } 3569 3570 rdev_for_each(rdev, mddev) { 3571 if (rdev->raid_disk < 0) 3572 continue; 3573 if (rdev->new_raid_disk >= mddev->raid_disks) 3574 rdev->new_raid_disk = -1; 3575 if (rdev->new_raid_disk == rdev->raid_disk) 3576 continue; 3577 sysfs_unlink_rdev(mddev, rdev); 3578 } 3579 rdev_for_each(rdev, mddev) { 3580 if (rdev->raid_disk < 0) 3581 continue; 3582 if (rdev->new_raid_disk == rdev->raid_disk) 3583 continue; 3584 rdev->raid_disk = rdev->new_raid_disk; 3585 if (rdev->raid_disk < 0) 3586 clear_bit(In_sync, &rdev->flags); 3587 else { 3588 if (sysfs_link_rdev(mddev, rdev)) 3589 printk(KERN_WARNING "md: cannot register rd%d" 3590 " for %s after level change\n", 3591 rdev->raid_disk, mdname(mddev)); 3592 } 3593 } 3594 3595 module_put(mddev->pers->owner); 3596 mddev->pers = pers; 3597 mddev->private = priv; 3598 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 3599 mddev->level = mddev->new_level; 3600 mddev->layout = mddev->new_layout; 3601 mddev->chunk_sectors = mddev->new_chunk_sectors; 3602 mddev->delta_disks = 0; 3603 mddev->reshape_backwards = 0; 3604 mddev->degraded = 0; 3605 if (mddev->pers->sync_request == NULL) { 3606 /* this is now an array without redundancy, so 3607 * it must always be in_sync 3608 */ 3609 mddev->in_sync = 1; 3610 del_timer_sync(&mddev->safemode_timer); 3611 } 3612 pers->run(mddev); 3613 set_bit(MD_CHANGE_DEVS, &mddev->flags); 3614 mddev_resume(mddev); 3615 sysfs_notify(&mddev->kobj, NULL, "level"); 3616 md_new_event(mddev); 3617 return rv; 3618 } 3619 3620 static struct md_sysfs_entry md_level = 3621 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); 3622 3623 3624 static ssize_t 3625 layout_show(struct mddev *mddev, char *page) 3626 { 3627 /* just a number, not meaningful for all levels */ 3628 if (mddev->reshape_position != MaxSector && 3629 mddev->layout != mddev->new_layout) 3630 return sprintf(page, "%d (%d)\n", 3631 mddev->new_layout, mddev->layout); 3632 return sprintf(page, "%d\n", mddev->layout); 3633 } 3634 3635 static ssize_t 3636 layout_store(struct mddev *mddev, const char *buf, size_t len) 3637 { 3638 char *e; 3639 unsigned long n = simple_strtoul(buf, &e, 10); 3640 3641 if (!*buf || (*e && *e != '\n')) 3642 return -EINVAL; 3643 3644 if (mddev->pers) { 3645 int err; 3646 if (mddev->pers->check_reshape == NULL) 3647 return -EBUSY; 3648 mddev->new_layout = n; 3649 err = mddev->pers->check_reshape(mddev); 3650 if (err) { 3651 mddev->new_layout = mddev->layout; 3652 return err; 3653 } 3654 } else { 3655 mddev->new_layout = n; 3656 if (mddev->reshape_position == MaxSector) 3657 mddev->layout = n; 3658 } 3659 return len; 3660 } 3661 static struct md_sysfs_entry md_layout = 3662 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); 3663 3664 3665 static ssize_t 3666 raid_disks_show(struct mddev *mddev, char *page) 3667 { 3668 if (mddev->raid_disks == 0) 3669 return 0; 3670 if (mddev->reshape_position != MaxSector && 3671 mddev->delta_disks != 0) 3672 return sprintf(page, "%d (%d)\n", mddev->raid_disks, 3673 mddev->raid_disks - mddev->delta_disks); 3674 return sprintf(page, "%d\n", mddev->raid_disks); 3675 } 3676 3677 static int update_raid_disks(struct mddev *mddev, int raid_disks); 3678 3679 static ssize_t 3680 raid_disks_store(struct mddev *mddev, const char *buf, size_t len) 3681 { 3682 char *e; 3683 int rv = 0; 3684 unsigned long n = simple_strtoul(buf, &e, 10); 3685 3686 if (!*buf || (*e && *e != '\n')) 3687 return -EINVAL; 3688 3689 if (mddev->pers) 3690 rv = update_raid_disks(mddev, n); 3691 else if (mddev->reshape_position != MaxSector) { 3692 struct md_rdev *rdev; 3693 int olddisks = mddev->raid_disks - mddev->delta_disks; 3694 3695 rdev_for_each(rdev, mddev) { 3696 if (olddisks < n && 3697 rdev->data_offset < rdev->new_data_offset) 3698 return -EINVAL; 3699 if (olddisks > n && 3700 rdev->data_offset > rdev->new_data_offset) 3701 return -EINVAL; 3702 } 3703 mddev->delta_disks = n - olddisks; 3704 mddev->raid_disks = n; 3705 mddev->reshape_backwards = (mddev->delta_disks < 0); 3706 } else 3707 mddev->raid_disks = n; 3708 return rv ? rv : len; 3709 } 3710 static struct md_sysfs_entry md_raid_disks = 3711 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); 3712 3713 static ssize_t 3714 chunk_size_show(struct mddev *mddev, char *page) 3715 { 3716 if (mddev->reshape_position != MaxSector && 3717 mddev->chunk_sectors != mddev->new_chunk_sectors) 3718 return sprintf(page, "%d (%d)\n", 3719 mddev->new_chunk_sectors << 9, 3720 mddev->chunk_sectors << 9); 3721 return sprintf(page, "%d\n", mddev->chunk_sectors << 9); 3722 } 3723 3724 static ssize_t 3725 chunk_size_store(struct mddev *mddev, const char *buf, size_t len) 3726 { 3727 char *e; 3728 unsigned long n = simple_strtoul(buf, &e, 10); 3729 3730 if (!*buf || (*e && *e != '\n')) 3731 return -EINVAL; 3732 3733 if (mddev->pers) { 3734 int err; 3735 if (mddev->pers->check_reshape == NULL) 3736 return -EBUSY; 3737 mddev->new_chunk_sectors = n >> 9; 3738 err = mddev->pers->check_reshape(mddev); 3739 if (err) { 3740 mddev->new_chunk_sectors = mddev->chunk_sectors; 3741 return err; 3742 } 3743 } else { 3744 mddev->new_chunk_sectors = n >> 9; 3745 if (mddev->reshape_position == MaxSector) 3746 mddev->chunk_sectors = n >> 9; 3747 } 3748 return len; 3749 } 3750 static struct md_sysfs_entry md_chunk_size = 3751 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); 3752 3753 static ssize_t 3754 resync_start_show(struct mddev *mddev, char *page) 3755 { 3756 if (mddev->recovery_cp == MaxSector) 3757 return sprintf(page, "none\n"); 3758 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); 3759 } 3760 3761 static ssize_t 3762 resync_start_store(struct mddev *mddev, const char *buf, size_t len) 3763 { 3764 char *e; 3765 unsigned long long n = simple_strtoull(buf, &e, 10); 3766 3767 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 3768 return -EBUSY; 3769 if (cmd_match(buf, "none")) 3770 n = MaxSector; 3771 else if (!*buf || (*e && *e != '\n')) 3772 return -EINVAL; 3773 3774 mddev->recovery_cp = n; 3775 return len; 3776 } 3777 static struct md_sysfs_entry md_resync_start = 3778 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); 3779 3780 /* 3781 * The array state can be: 3782 * 3783 * clear 3784 * No devices, no size, no level 3785 * Equivalent to STOP_ARRAY ioctl 3786 * inactive 3787 * May have some settings, but array is not active 3788 * all IO results in error 3789 * When written, doesn't tear down array, but just stops it 3790 * suspended (not supported yet) 3791 * All IO requests will block. The array can be reconfigured. 3792 * Writing this, if accepted, will block until array is quiescent 3793 * readonly 3794 * no resync can happen. no superblocks get written. 3795 * write requests fail 3796 * read-auto 3797 * like readonly, but behaves like 'clean' on a write request. 3798 * 3799 * clean - no pending writes, but otherwise active. 3800 * When written to inactive array, starts without resync 3801 * If a write request arrives then 3802 * if metadata is known, mark 'dirty' and switch to 'active'. 3803 * if not known, block and switch to write-pending 3804 * If written to an active array that has pending writes, then fails. 3805 * active 3806 * fully active: IO and resync can be happening. 3807 * When written to inactive array, starts with resync 3808 * 3809 * write-pending 3810 * clean, but writes are blocked waiting for 'active' to be written. 3811 * 3812 * active-idle 3813 * like active, but no writes have been seen for a while (100msec). 3814 * 3815 */ 3816 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, 3817 write_pending, active_idle, bad_word}; 3818 static char *array_states[] = { 3819 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", 3820 "write-pending", "active-idle", NULL }; 3821 3822 static int match_word(const char *word, char **list) 3823 { 3824 int n; 3825 for (n=0; list[n]; n++) 3826 if (cmd_match(word, list[n])) 3827 break; 3828 return n; 3829 } 3830 3831 static ssize_t 3832 array_state_show(struct mddev *mddev, char *page) 3833 { 3834 enum array_state st = inactive; 3835 3836 if (mddev->pers) 3837 switch(mddev->ro) { 3838 case 1: 3839 st = readonly; 3840 break; 3841 case 2: 3842 st = read_auto; 3843 break; 3844 case 0: 3845 if (mddev->in_sync) 3846 st = clean; 3847 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 3848 st = write_pending; 3849 else if (mddev->safemode) 3850 st = active_idle; 3851 else 3852 st = active; 3853 } 3854 else { 3855 if (list_empty(&mddev->disks) && 3856 mddev->raid_disks == 0 && 3857 mddev->dev_sectors == 0) 3858 st = clear; 3859 else 3860 st = inactive; 3861 } 3862 return sprintf(page, "%s\n", array_states[st]); 3863 } 3864 3865 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev); 3866 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev); 3867 static int do_md_run(struct mddev * mddev); 3868 static int restart_array(struct mddev *mddev); 3869 3870 static ssize_t 3871 array_state_store(struct mddev *mddev, const char *buf, size_t len) 3872 { 3873 int err = -EINVAL; 3874 enum array_state st = match_word(buf, array_states); 3875 switch(st) { 3876 case bad_word: 3877 break; 3878 case clear: 3879 /* stopping an active array */ 3880 err = do_md_stop(mddev, 0, NULL); 3881 break; 3882 case inactive: 3883 /* stopping an active array */ 3884 if (mddev->pers) 3885 err = do_md_stop(mddev, 2, NULL); 3886 else 3887 err = 0; /* already inactive */ 3888 break; 3889 case suspended: 3890 break; /* not supported yet */ 3891 case readonly: 3892 if (mddev->pers) 3893 err = md_set_readonly(mddev, NULL); 3894 else { 3895 mddev->ro = 1; 3896 set_disk_ro(mddev->gendisk, 1); 3897 err = do_md_run(mddev); 3898 } 3899 break; 3900 case read_auto: 3901 if (mddev->pers) { 3902 if (mddev->ro == 0) 3903 err = md_set_readonly(mddev, NULL); 3904 else if (mddev->ro == 1) 3905 err = restart_array(mddev); 3906 if (err == 0) { 3907 mddev->ro = 2; 3908 set_disk_ro(mddev->gendisk, 0); 3909 } 3910 } else { 3911 mddev->ro = 2; 3912 err = do_md_run(mddev); 3913 } 3914 break; 3915 case clean: 3916 if (mddev->pers) { 3917 restart_array(mddev); 3918 spin_lock_irq(&mddev->write_lock); 3919 if (atomic_read(&mddev->writes_pending) == 0) { 3920 if (mddev->in_sync == 0) { 3921 mddev->in_sync = 1; 3922 if (mddev->safemode == 1) 3923 mddev->safemode = 0; 3924 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 3925 } 3926 err = 0; 3927 } else 3928 err = -EBUSY; 3929 spin_unlock_irq(&mddev->write_lock); 3930 } else 3931 err = -EINVAL; 3932 break; 3933 case active: 3934 if (mddev->pers) { 3935 restart_array(mddev); 3936 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 3937 wake_up(&mddev->sb_wait); 3938 err = 0; 3939 } else { 3940 mddev->ro = 0; 3941 set_disk_ro(mddev->gendisk, 0); 3942 err = do_md_run(mddev); 3943 } 3944 break; 3945 case write_pending: 3946 case active_idle: 3947 /* these cannot be set */ 3948 break; 3949 } 3950 if (err) 3951 return err; 3952 else { 3953 if (mddev->hold_active == UNTIL_IOCTL) 3954 mddev->hold_active = 0; 3955 sysfs_notify_dirent_safe(mddev->sysfs_state); 3956 return len; 3957 } 3958 } 3959 static struct md_sysfs_entry md_array_state = 3960 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); 3961 3962 static ssize_t 3963 max_corrected_read_errors_show(struct mddev *mddev, char *page) { 3964 return sprintf(page, "%d\n", 3965 atomic_read(&mddev->max_corr_read_errors)); 3966 } 3967 3968 static ssize_t 3969 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) 3970 { 3971 char *e; 3972 unsigned long n = simple_strtoul(buf, &e, 10); 3973 3974 if (*buf && (*e == 0 || *e == '\n')) { 3975 atomic_set(&mddev->max_corr_read_errors, n); 3976 return len; 3977 } 3978 return -EINVAL; 3979 } 3980 3981 static struct md_sysfs_entry max_corr_read_errors = 3982 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, 3983 max_corrected_read_errors_store); 3984 3985 static ssize_t 3986 null_show(struct mddev *mddev, char *page) 3987 { 3988 return -EINVAL; 3989 } 3990 3991 static ssize_t 3992 new_dev_store(struct mddev *mddev, const char *buf, size_t len) 3993 { 3994 /* buf must be %d:%d\n? giving major and minor numbers */ 3995 /* The new device is added to the array. 3996 * If the array has a persistent superblock, we read the 3997 * superblock to initialise info and check validity. 3998 * Otherwise, only checking done is that in bind_rdev_to_array, 3999 * which mainly checks size. 4000 */ 4001 char *e; 4002 int major = simple_strtoul(buf, &e, 10); 4003 int minor; 4004 dev_t dev; 4005 struct md_rdev *rdev; 4006 int err; 4007 4008 if (!*buf || *e != ':' || !e[1] || e[1] == '\n') 4009 return -EINVAL; 4010 minor = simple_strtoul(e+1, &e, 10); 4011 if (*e && *e != '\n') 4012 return -EINVAL; 4013 dev = MKDEV(major, minor); 4014 if (major != MAJOR(dev) || 4015 minor != MINOR(dev)) 4016 return -EOVERFLOW; 4017 4018 4019 if (mddev->persistent) { 4020 rdev = md_import_device(dev, mddev->major_version, 4021 mddev->minor_version); 4022 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { 4023 struct md_rdev *rdev0 4024 = list_entry(mddev->disks.next, 4025 struct md_rdev, same_set); 4026 err = super_types[mddev->major_version] 4027 .load_super(rdev, rdev0, mddev->minor_version); 4028 if (err < 0) 4029 goto out; 4030 } 4031 } else if (mddev->external) 4032 rdev = md_import_device(dev, -2, -1); 4033 else 4034 rdev = md_import_device(dev, -1, -1); 4035 4036 if (IS_ERR(rdev)) 4037 return PTR_ERR(rdev); 4038 err = bind_rdev_to_array(rdev, mddev); 4039 out: 4040 if (err) 4041 export_rdev(rdev); 4042 return err ? err : len; 4043 } 4044 4045 static struct md_sysfs_entry md_new_device = 4046 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); 4047 4048 static ssize_t 4049 bitmap_store(struct mddev *mddev, const char *buf, size_t len) 4050 { 4051 char *end; 4052 unsigned long chunk, end_chunk; 4053 4054 if (!mddev->bitmap) 4055 goto out; 4056 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ 4057 while (*buf) { 4058 chunk = end_chunk = simple_strtoul(buf, &end, 0); 4059 if (buf == end) break; 4060 if (*end == '-') { /* range */ 4061 buf = end + 1; 4062 end_chunk = simple_strtoul(buf, &end, 0); 4063 if (buf == end) break; 4064 } 4065 if (*end && !isspace(*end)) break; 4066 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); 4067 buf = skip_spaces(end); 4068 } 4069 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ 4070 out: 4071 return len; 4072 } 4073 4074 static struct md_sysfs_entry md_bitmap = 4075 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); 4076 4077 static ssize_t 4078 size_show(struct mddev *mddev, char *page) 4079 { 4080 return sprintf(page, "%llu\n", 4081 (unsigned long long)mddev->dev_sectors / 2); 4082 } 4083 4084 static int update_size(struct mddev *mddev, sector_t num_sectors); 4085 4086 static ssize_t 4087 size_store(struct mddev *mddev, const char *buf, size_t len) 4088 { 4089 /* If array is inactive, we can reduce the component size, but 4090 * not increase it (except from 0). 4091 * If array is active, we can try an on-line resize 4092 */ 4093 sector_t sectors; 4094 int err = strict_blocks_to_sectors(buf, §ors); 4095 4096 if (err < 0) 4097 return err; 4098 if (mddev->pers) { 4099 err = update_size(mddev, sectors); 4100 md_update_sb(mddev, 1); 4101 } else { 4102 if (mddev->dev_sectors == 0 || 4103 mddev->dev_sectors > sectors) 4104 mddev->dev_sectors = sectors; 4105 else 4106 err = -ENOSPC; 4107 } 4108 return err ? err : len; 4109 } 4110 4111 static struct md_sysfs_entry md_size = 4112 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); 4113 4114 4115 /* Metdata version. 4116 * This is one of 4117 * 'none' for arrays with no metadata (good luck...) 4118 * 'external' for arrays with externally managed metadata, 4119 * or N.M for internally known formats 4120 */ 4121 static ssize_t 4122 metadata_show(struct mddev *mddev, char *page) 4123 { 4124 if (mddev->persistent) 4125 return sprintf(page, "%d.%d\n", 4126 mddev->major_version, mddev->minor_version); 4127 else if (mddev->external) 4128 return sprintf(page, "external:%s\n", mddev->metadata_type); 4129 else 4130 return sprintf(page, "none\n"); 4131 } 4132 4133 static ssize_t 4134 metadata_store(struct mddev *mddev, const char *buf, size_t len) 4135 { 4136 int major, minor; 4137 char *e; 4138 /* Changing the details of 'external' metadata is 4139 * always permitted. Otherwise there must be 4140 * no devices attached to the array. 4141 */ 4142 if (mddev->external && strncmp(buf, "external:", 9) == 0) 4143 ; 4144 else if (!list_empty(&mddev->disks)) 4145 return -EBUSY; 4146 4147 if (cmd_match(buf, "none")) { 4148 mddev->persistent = 0; 4149 mddev->external = 0; 4150 mddev->major_version = 0; 4151 mddev->minor_version = 90; 4152 return len; 4153 } 4154 if (strncmp(buf, "external:", 9) == 0) { 4155 size_t namelen = len-9; 4156 if (namelen >= sizeof(mddev->metadata_type)) 4157 namelen = sizeof(mddev->metadata_type)-1; 4158 strncpy(mddev->metadata_type, buf+9, namelen); 4159 mddev->metadata_type[namelen] = 0; 4160 if (namelen && mddev->metadata_type[namelen-1] == '\n') 4161 mddev->metadata_type[--namelen] = 0; 4162 mddev->persistent = 0; 4163 mddev->external = 1; 4164 mddev->major_version = 0; 4165 mddev->minor_version = 90; 4166 return len; 4167 } 4168 major = simple_strtoul(buf, &e, 10); 4169 if (e==buf || *e != '.') 4170 return -EINVAL; 4171 buf = e+1; 4172 minor = simple_strtoul(buf, &e, 10); 4173 if (e==buf || (*e && *e != '\n') ) 4174 return -EINVAL; 4175 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) 4176 return -ENOENT; 4177 mddev->major_version = major; 4178 mddev->minor_version = minor; 4179 mddev->persistent = 1; 4180 mddev->external = 0; 4181 return len; 4182 } 4183 4184 static struct md_sysfs_entry md_metadata = 4185 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 4186 4187 static ssize_t 4188 action_show(struct mddev *mddev, char *page) 4189 { 4190 char *type = "idle"; 4191 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 4192 type = "frozen"; 4193 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 4194 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { 4195 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 4196 type = "reshape"; 4197 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 4198 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 4199 type = "resync"; 4200 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 4201 type = "check"; 4202 else 4203 type = "repair"; 4204 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) 4205 type = "recover"; 4206 } 4207 return sprintf(page, "%s\n", type); 4208 } 4209 4210 static void reap_sync_thread(struct mddev *mddev); 4211 4212 static ssize_t 4213 action_store(struct mddev *mddev, const char *page, size_t len) 4214 { 4215 if (!mddev->pers || !mddev->pers->sync_request) 4216 return -EINVAL; 4217 4218 if (cmd_match(page, "frozen")) 4219 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4220 else 4221 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4222 4223 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { 4224 if (mddev->sync_thread) { 4225 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4226 reap_sync_thread(mddev); 4227 } 4228 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 4229 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 4230 return -EBUSY; 4231 else if (cmd_match(page, "resync")) 4232 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4233 else if (cmd_match(page, "recover")) { 4234 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 4235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4236 } else if (cmd_match(page, "reshape")) { 4237 int err; 4238 if (mddev->pers->start_reshape == NULL) 4239 return -EINVAL; 4240 err = mddev->pers->start_reshape(mddev); 4241 if (err) 4242 return err; 4243 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4244 } else { 4245 if (cmd_match(page, "check")) 4246 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 4247 else if (!cmd_match(page, "repair")) 4248 return -EINVAL; 4249 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 4250 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 4251 } 4252 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4253 md_wakeup_thread(mddev->thread); 4254 sysfs_notify_dirent_safe(mddev->sysfs_action); 4255 return len; 4256 } 4257 4258 static ssize_t 4259 mismatch_cnt_show(struct mddev *mddev, char *page) 4260 { 4261 return sprintf(page, "%llu\n", 4262 (unsigned long long) mddev->resync_mismatches); 4263 } 4264 4265 static struct md_sysfs_entry md_scan_mode = 4266 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); 4267 4268 4269 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); 4270 4271 static ssize_t 4272 sync_min_show(struct mddev *mddev, char *page) 4273 { 4274 return sprintf(page, "%d (%s)\n", speed_min(mddev), 4275 mddev->sync_speed_min ? "local": "system"); 4276 } 4277 4278 static ssize_t 4279 sync_min_store(struct mddev *mddev, const char *buf, size_t len) 4280 { 4281 int min; 4282 char *e; 4283 if (strncmp(buf, "system", 6)==0) { 4284 mddev->sync_speed_min = 0; 4285 return len; 4286 } 4287 min = simple_strtoul(buf, &e, 10); 4288 if (buf == e || (*e && *e != '\n') || min <= 0) 4289 return -EINVAL; 4290 mddev->sync_speed_min = min; 4291 return len; 4292 } 4293 4294 static struct md_sysfs_entry md_sync_min = 4295 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); 4296 4297 static ssize_t 4298 sync_max_show(struct mddev *mddev, char *page) 4299 { 4300 return sprintf(page, "%d (%s)\n", speed_max(mddev), 4301 mddev->sync_speed_max ? "local": "system"); 4302 } 4303 4304 static ssize_t 4305 sync_max_store(struct mddev *mddev, const char *buf, size_t len) 4306 { 4307 int max; 4308 char *e; 4309 if (strncmp(buf, "system", 6)==0) { 4310 mddev->sync_speed_max = 0; 4311 return len; 4312 } 4313 max = simple_strtoul(buf, &e, 10); 4314 if (buf == e || (*e && *e != '\n') || max <= 0) 4315 return -EINVAL; 4316 mddev->sync_speed_max = max; 4317 return len; 4318 } 4319 4320 static struct md_sysfs_entry md_sync_max = 4321 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); 4322 4323 static ssize_t 4324 degraded_show(struct mddev *mddev, char *page) 4325 { 4326 return sprintf(page, "%d\n", mddev->degraded); 4327 } 4328 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); 4329 4330 static ssize_t 4331 sync_force_parallel_show(struct mddev *mddev, char *page) 4332 { 4333 return sprintf(page, "%d\n", mddev->parallel_resync); 4334 } 4335 4336 static ssize_t 4337 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) 4338 { 4339 long n; 4340 4341 if (strict_strtol(buf, 10, &n)) 4342 return -EINVAL; 4343 4344 if (n != 0 && n != 1) 4345 return -EINVAL; 4346 4347 mddev->parallel_resync = n; 4348 4349 if (mddev->sync_thread) 4350 wake_up(&resync_wait); 4351 4352 return len; 4353 } 4354 4355 /* force parallel resync, even with shared block devices */ 4356 static struct md_sysfs_entry md_sync_force_parallel = 4357 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, 4358 sync_force_parallel_show, sync_force_parallel_store); 4359 4360 static ssize_t 4361 sync_speed_show(struct mddev *mddev, char *page) 4362 { 4363 unsigned long resync, dt, db; 4364 if (mddev->curr_resync == 0) 4365 return sprintf(page, "none\n"); 4366 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); 4367 dt = (jiffies - mddev->resync_mark) / HZ; 4368 if (!dt) dt++; 4369 db = resync - mddev->resync_mark_cnt; 4370 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ 4371 } 4372 4373 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); 4374 4375 static ssize_t 4376 sync_completed_show(struct mddev *mddev, char *page) 4377 { 4378 unsigned long long max_sectors, resync; 4379 4380 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4381 return sprintf(page, "none\n"); 4382 4383 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || 4384 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 4385 max_sectors = mddev->resync_max_sectors; 4386 else 4387 max_sectors = mddev->dev_sectors; 4388 4389 resync = mddev->curr_resync_completed; 4390 return sprintf(page, "%llu / %llu\n", resync, max_sectors); 4391 } 4392 4393 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); 4394 4395 static ssize_t 4396 min_sync_show(struct mddev *mddev, char *page) 4397 { 4398 return sprintf(page, "%llu\n", 4399 (unsigned long long)mddev->resync_min); 4400 } 4401 static ssize_t 4402 min_sync_store(struct mddev *mddev, const char *buf, size_t len) 4403 { 4404 unsigned long long min; 4405 if (strict_strtoull(buf, 10, &min)) 4406 return -EINVAL; 4407 if (min > mddev->resync_max) 4408 return -EINVAL; 4409 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4410 return -EBUSY; 4411 4412 /* Must be a multiple of chunk_size */ 4413 if (mddev->chunk_sectors) { 4414 sector_t temp = min; 4415 if (sector_div(temp, mddev->chunk_sectors)) 4416 return -EINVAL; 4417 } 4418 mddev->resync_min = min; 4419 4420 return len; 4421 } 4422 4423 static struct md_sysfs_entry md_min_sync = 4424 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); 4425 4426 static ssize_t 4427 max_sync_show(struct mddev *mddev, char *page) 4428 { 4429 if (mddev->resync_max == MaxSector) 4430 return sprintf(page, "max\n"); 4431 else 4432 return sprintf(page, "%llu\n", 4433 (unsigned long long)mddev->resync_max); 4434 } 4435 static ssize_t 4436 max_sync_store(struct mddev *mddev, const char *buf, size_t len) 4437 { 4438 if (strncmp(buf, "max", 3) == 0) 4439 mddev->resync_max = MaxSector; 4440 else { 4441 unsigned long long max; 4442 if (strict_strtoull(buf, 10, &max)) 4443 return -EINVAL; 4444 if (max < mddev->resync_min) 4445 return -EINVAL; 4446 if (max < mddev->resync_max && 4447 mddev->ro == 0 && 4448 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4449 return -EBUSY; 4450 4451 /* Must be a multiple of chunk_size */ 4452 if (mddev->chunk_sectors) { 4453 sector_t temp = max; 4454 if (sector_div(temp, mddev->chunk_sectors)) 4455 return -EINVAL; 4456 } 4457 mddev->resync_max = max; 4458 } 4459 wake_up(&mddev->recovery_wait); 4460 return len; 4461 } 4462 4463 static struct md_sysfs_entry md_max_sync = 4464 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); 4465 4466 static ssize_t 4467 suspend_lo_show(struct mddev *mddev, char *page) 4468 { 4469 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); 4470 } 4471 4472 static ssize_t 4473 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) 4474 { 4475 char *e; 4476 unsigned long long new = simple_strtoull(buf, &e, 10); 4477 unsigned long long old = mddev->suspend_lo; 4478 4479 if (mddev->pers == NULL || 4480 mddev->pers->quiesce == NULL) 4481 return -EINVAL; 4482 if (buf == e || (*e && *e != '\n')) 4483 return -EINVAL; 4484 4485 mddev->suspend_lo = new; 4486 if (new >= old) 4487 /* Shrinking suspended region */ 4488 mddev->pers->quiesce(mddev, 2); 4489 else { 4490 /* Expanding suspended region - need to wait */ 4491 mddev->pers->quiesce(mddev, 1); 4492 mddev->pers->quiesce(mddev, 0); 4493 } 4494 return len; 4495 } 4496 static struct md_sysfs_entry md_suspend_lo = 4497 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); 4498 4499 4500 static ssize_t 4501 suspend_hi_show(struct mddev *mddev, char *page) 4502 { 4503 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); 4504 } 4505 4506 static ssize_t 4507 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) 4508 { 4509 char *e; 4510 unsigned long long new = simple_strtoull(buf, &e, 10); 4511 unsigned long long old = mddev->suspend_hi; 4512 4513 if (mddev->pers == NULL || 4514 mddev->pers->quiesce == NULL) 4515 return -EINVAL; 4516 if (buf == e || (*e && *e != '\n')) 4517 return -EINVAL; 4518 4519 mddev->suspend_hi = new; 4520 if (new <= old) 4521 /* Shrinking suspended region */ 4522 mddev->pers->quiesce(mddev, 2); 4523 else { 4524 /* Expanding suspended region - need to wait */ 4525 mddev->pers->quiesce(mddev, 1); 4526 mddev->pers->quiesce(mddev, 0); 4527 } 4528 return len; 4529 } 4530 static struct md_sysfs_entry md_suspend_hi = 4531 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); 4532 4533 static ssize_t 4534 reshape_position_show(struct mddev *mddev, char *page) 4535 { 4536 if (mddev->reshape_position != MaxSector) 4537 return sprintf(page, "%llu\n", 4538 (unsigned long long)mddev->reshape_position); 4539 strcpy(page, "none\n"); 4540 return 5; 4541 } 4542 4543 static ssize_t 4544 reshape_position_store(struct mddev *mddev, const char *buf, size_t len) 4545 { 4546 struct md_rdev *rdev; 4547 char *e; 4548 unsigned long long new = simple_strtoull(buf, &e, 10); 4549 if (mddev->pers) 4550 return -EBUSY; 4551 if (buf == e || (*e && *e != '\n')) 4552 return -EINVAL; 4553 mddev->reshape_position = new; 4554 mddev->delta_disks = 0; 4555 mddev->reshape_backwards = 0; 4556 mddev->new_level = mddev->level; 4557 mddev->new_layout = mddev->layout; 4558 mddev->new_chunk_sectors = mddev->chunk_sectors; 4559 rdev_for_each(rdev, mddev) 4560 rdev->new_data_offset = rdev->data_offset; 4561 return len; 4562 } 4563 4564 static struct md_sysfs_entry md_reshape_position = 4565 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, 4566 reshape_position_store); 4567 4568 static ssize_t 4569 reshape_direction_show(struct mddev *mddev, char *page) 4570 { 4571 return sprintf(page, "%s\n", 4572 mddev->reshape_backwards ? "backwards" : "forwards"); 4573 } 4574 4575 static ssize_t 4576 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) 4577 { 4578 int backwards = 0; 4579 if (cmd_match(buf, "forwards")) 4580 backwards = 0; 4581 else if (cmd_match(buf, "backwards")) 4582 backwards = 1; 4583 else 4584 return -EINVAL; 4585 if (mddev->reshape_backwards == backwards) 4586 return len; 4587 4588 /* check if we are allowed to change */ 4589 if (mddev->delta_disks) 4590 return -EBUSY; 4591 4592 if (mddev->persistent && 4593 mddev->major_version == 0) 4594 return -EINVAL; 4595 4596 mddev->reshape_backwards = backwards; 4597 return len; 4598 } 4599 4600 static struct md_sysfs_entry md_reshape_direction = 4601 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, 4602 reshape_direction_store); 4603 4604 static ssize_t 4605 array_size_show(struct mddev *mddev, char *page) 4606 { 4607 if (mddev->external_size) 4608 return sprintf(page, "%llu\n", 4609 (unsigned long long)mddev->array_sectors/2); 4610 else 4611 return sprintf(page, "default\n"); 4612 } 4613 4614 static ssize_t 4615 array_size_store(struct mddev *mddev, const char *buf, size_t len) 4616 { 4617 sector_t sectors; 4618 4619 if (strncmp(buf, "default", 7) == 0) { 4620 if (mddev->pers) 4621 sectors = mddev->pers->size(mddev, 0, 0); 4622 else 4623 sectors = mddev->array_sectors; 4624 4625 mddev->external_size = 0; 4626 } else { 4627 if (strict_blocks_to_sectors(buf, §ors) < 0) 4628 return -EINVAL; 4629 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) 4630 return -E2BIG; 4631 4632 mddev->external_size = 1; 4633 } 4634 4635 mddev->array_sectors = sectors; 4636 if (mddev->pers) { 4637 set_capacity(mddev->gendisk, mddev->array_sectors); 4638 revalidate_disk(mddev->gendisk); 4639 } 4640 return len; 4641 } 4642 4643 static struct md_sysfs_entry md_array_size = 4644 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, 4645 array_size_store); 4646 4647 static struct attribute *md_default_attrs[] = { 4648 &md_level.attr, 4649 &md_layout.attr, 4650 &md_raid_disks.attr, 4651 &md_chunk_size.attr, 4652 &md_size.attr, 4653 &md_resync_start.attr, 4654 &md_metadata.attr, 4655 &md_new_device.attr, 4656 &md_safe_delay.attr, 4657 &md_array_state.attr, 4658 &md_reshape_position.attr, 4659 &md_reshape_direction.attr, 4660 &md_array_size.attr, 4661 &max_corr_read_errors.attr, 4662 NULL, 4663 }; 4664 4665 static struct attribute *md_redundancy_attrs[] = { 4666 &md_scan_mode.attr, 4667 &md_mismatches.attr, 4668 &md_sync_min.attr, 4669 &md_sync_max.attr, 4670 &md_sync_speed.attr, 4671 &md_sync_force_parallel.attr, 4672 &md_sync_completed.attr, 4673 &md_min_sync.attr, 4674 &md_max_sync.attr, 4675 &md_suspend_lo.attr, 4676 &md_suspend_hi.attr, 4677 &md_bitmap.attr, 4678 &md_degraded.attr, 4679 NULL, 4680 }; 4681 static struct attribute_group md_redundancy_group = { 4682 .name = NULL, 4683 .attrs = md_redundancy_attrs, 4684 }; 4685 4686 4687 static ssize_t 4688 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 4689 { 4690 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4691 struct mddev *mddev = container_of(kobj, struct mddev, kobj); 4692 ssize_t rv; 4693 4694 if (!entry->show) 4695 return -EIO; 4696 spin_lock(&all_mddevs_lock); 4697 if (list_empty(&mddev->all_mddevs)) { 4698 spin_unlock(&all_mddevs_lock); 4699 return -EBUSY; 4700 } 4701 mddev_get(mddev); 4702 spin_unlock(&all_mddevs_lock); 4703 4704 rv = mddev_lock(mddev); 4705 if (!rv) { 4706 rv = entry->show(mddev, page); 4707 mddev_unlock(mddev); 4708 } 4709 mddev_put(mddev); 4710 return rv; 4711 } 4712 4713 static ssize_t 4714 md_attr_store(struct kobject *kobj, struct attribute *attr, 4715 const char *page, size_t length) 4716 { 4717 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4718 struct mddev *mddev = container_of(kobj, struct mddev, kobj); 4719 ssize_t rv; 4720 4721 if (!entry->store) 4722 return -EIO; 4723 if (!capable(CAP_SYS_ADMIN)) 4724 return -EACCES; 4725 spin_lock(&all_mddevs_lock); 4726 if (list_empty(&mddev->all_mddevs)) { 4727 spin_unlock(&all_mddevs_lock); 4728 return -EBUSY; 4729 } 4730 mddev_get(mddev); 4731 spin_unlock(&all_mddevs_lock); 4732 rv = mddev_lock(mddev); 4733 if (!rv) { 4734 rv = entry->store(mddev, page, length); 4735 mddev_unlock(mddev); 4736 } 4737 mddev_put(mddev); 4738 return rv; 4739 } 4740 4741 static void md_free(struct kobject *ko) 4742 { 4743 struct mddev *mddev = container_of(ko, struct mddev, kobj); 4744 4745 if (mddev->sysfs_state) 4746 sysfs_put(mddev->sysfs_state); 4747 4748 if (mddev->gendisk) { 4749 del_gendisk(mddev->gendisk); 4750 put_disk(mddev->gendisk); 4751 } 4752 if (mddev->queue) 4753 blk_cleanup_queue(mddev->queue); 4754 4755 kfree(mddev); 4756 } 4757 4758 static const struct sysfs_ops md_sysfs_ops = { 4759 .show = md_attr_show, 4760 .store = md_attr_store, 4761 }; 4762 static struct kobj_type md_ktype = { 4763 .release = md_free, 4764 .sysfs_ops = &md_sysfs_ops, 4765 .default_attrs = md_default_attrs, 4766 }; 4767 4768 int mdp_major = 0; 4769 4770 static void mddev_delayed_delete(struct work_struct *ws) 4771 { 4772 struct mddev *mddev = container_of(ws, struct mddev, del_work); 4773 4774 sysfs_remove_group(&mddev->kobj, &md_bitmap_group); 4775 kobject_del(&mddev->kobj); 4776 kobject_put(&mddev->kobj); 4777 } 4778 4779 static int md_alloc(dev_t dev, char *name) 4780 { 4781 static DEFINE_MUTEX(disks_mutex); 4782 struct mddev *mddev = mddev_find(dev); 4783 struct gendisk *disk; 4784 int partitioned; 4785 int shift; 4786 int unit; 4787 int error; 4788 4789 if (!mddev) 4790 return -ENODEV; 4791 4792 partitioned = (MAJOR(mddev->unit) != MD_MAJOR); 4793 shift = partitioned ? MdpMinorShift : 0; 4794 unit = MINOR(mddev->unit) >> shift; 4795 4796 /* wait for any previous instance of this device to be 4797 * completely removed (mddev_delayed_delete). 4798 */ 4799 flush_workqueue(md_misc_wq); 4800 4801 mutex_lock(&disks_mutex); 4802 error = -EEXIST; 4803 if (mddev->gendisk) 4804 goto abort; 4805 4806 if (name) { 4807 /* Need to ensure that 'name' is not a duplicate. 4808 */ 4809 struct mddev *mddev2; 4810 spin_lock(&all_mddevs_lock); 4811 4812 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) 4813 if (mddev2->gendisk && 4814 strcmp(mddev2->gendisk->disk_name, name) == 0) { 4815 spin_unlock(&all_mddevs_lock); 4816 goto abort; 4817 } 4818 spin_unlock(&all_mddevs_lock); 4819 } 4820 4821 error = -ENOMEM; 4822 mddev->queue = blk_alloc_queue(GFP_KERNEL); 4823 if (!mddev->queue) 4824 goto abort; 4825 mddev->queue->queuedata = mddev; 4826 4827 blk_queue_make_request(mddev->queue, md_make_request); 4828 blk_set_stacking_limits(&mddev->queue->limits); 4829 4830 disk = alloc_disk(1 << shift); 4831 if (!disk) { 4832 blk_cleanup_queue(mddev->queue); 4833 mddev->queue = NULL; 4834 goto abort; 4835 } 4836 disk->major = MAJOR(mddev->unit); 4837 disk->first_minor = unit << shift; 4838 if (name) 4839 strcpy(disk->disk_name, name); 4840 else if (partitioned) 4841 sprintf(disk->disk_name, "md_d%d", unit); 4842 else 4843 sprintf(disk->disk_name, "md%d", unit); 4844 disk->fops = &md_fops; 4845 disk->private_data = mddev; 4846 disk->queue = mddev->queue; 4847 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); 4848 /* Allow extended partitions. This makes the 4849 * 'mdp' device redundant, but we can't really 4850 * remove it now. 4851 */ 4852 disk->flags |= GENHD_FL_EXT_DEVT; 4853 mddev->gendisk = disk; 4854 /* As soon as we call add_disk(), another thread could get 4855 * through to md_open, so make sure it doesn't get too far 4856 */ 4857 mutex_lock(&mddev->open_mutex); 4858 add_disk(disk); 4859 4860 error = kobject_init_and_add(&mddev->kobj, &md_ktype, 4861 &disk_to_dev(disk)->kobj, "%s", "md"); 4862 if (error) { 4863 /* This isn't possible, but as kobject_init_and_add is marked 4864 * __must_check, we must do something with the result 4865 */ 4866 printk(KERN_WARNING "md: cannot register %s/md - name in use\n", 4867 disk->disk_name); 4868 error = 0; 4869 } 4870 if (mddev->kobj.sd && 4871 sysfs_create_group(&mddev->kobj, &md_bitmap_group)) 4872 printk(KERN_DEBUG "pointless warning\n"); 4873 mutex_unlock(&mddev->open_mutex); 4874 abort: 4875 mutex_unlock(&disks_mutex); 4876 if (!error && mddev->kobj.sd) { 4877 kobject_uevent(&mddev->kobj, KOBJ_ADD); 4878 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); 4879 } 4880 mddev_put(mddev); 4881 return error; 4882 } 4883 4884 static struct kobject *md_probe(dev_t dev, int *part, void *data) 4885 { 4886 md_alloc(dev, NULL); 4887 return NULL; 4888 } 4889 4890 static int add_named_array(const char *val, struct kernel_param *kp) 4891 { 4892 /* val must be "md_*" where * is not all digits. 4893 * We allocate an array with a large free minor number, and 4894 * set the name to val. val must not already be an active name. 4895 */ 4896 int len = strlen(val); 4897 char buf[DISK_NAME_LEN]; 4898 4899 while (len && val[len-1] == '\n') 4900 len--; 4901 if (len >= DISK_NAME_LEN) 4902 return -E2BIG; 4903 strlcpy(buf, val, len+1); 4904 if (strncmp(buf, "md_", 3) != 0) 4905 return -EINVAL; 4906 return md_alloc(0, buf); 4907 } 4908 4909 static void md_safemode_timeout(unsigned long data) 4910 { 4911 struct mddev *mddev = (struct mddev *) data; 4912 4913 if (!atomic_read(&mddev->writes_pending)) { 4914 mddev->safemode = 1; 4915 if (mddev->external) 4916 sysfs_notify_dirent_safe(mddev->sysfs_state); 4917 } 4918 md_wakeup_thread(mddev->thread); 4919 } 4920 4921 static int start_dirty_degraded; 4922 4923 int md_run(struct mddev *mddev) 4924 { 4925 int err; 4926 struct md_rdev *rdev; 4927 struct md_personality *pers; 4928 4929 if (list_empty(&mddev->disks)) 4930 /* cannot run an array with no devices.. */ 4931 return -EINVAL; 4932 4933 if (mddev->pers) 4934 return -EBUSY; 4935 /* Cannot run until previous stop completes properly */ 4936 if (mddev->sysfs_active) 4937 return -EBUSY; 4938 4939 /* 4940 * Analyze all RAID superblock(s) 4941 */ 4942 if (!mddev->raid_disks) { 4943 if (!mddev->persistent) 4944 return -EINVAL; 4945 analyze_sbs(mddev); 4946 } 4947 4948 if (mddev->level != LEVEL_NONE) 4949 request_module("md-level-%d", mddev->level); 4950 else if (mddev->clevel[0]) 4951 request_module("md-%s", mddev->clevel); 4952 4953 /* 4954 * Drop all container device buffers, from now on 4955 * the only valid external interface is through the md 4956 * device. 4957 */ 4958 rdev_for_each(rdev, mddev) { 4959 if (test_bit(Faulty, &rdev->flags)) 4960 continue; 4961 sync_blockdev(rdev->bdev); 4962 invalidate_bdev(rdev->bdev); 4963 4964 /* perform some consistency tests on the device. 4965 * We don't want the data to overlap the metadata, 4966 * Internal Bitmap issues have been handled elsewhere. 4967 */ 4968 if (rdev->meta_bdev) { 4969 /* Nothing to check */; 4970 } else if (rdev->data_offset < rdev->sb_start) { 4971 if (mddev->dev_sectors && 4972 rdev->data_offset + mddev->dev_sectors 4973 > rdev->sb_start) { 4974 printk("md: %s: data overlaps metadata\n", 4975 mdname(mddev)); 4976 return -EINVAL; 4977 } 4978 } else { 4979 if (rdev->sb_start + rdev->sb_size/512 4980 > rdev->data_offset) { 4981 printk("md: %s: metadata overlaps data\n", 4982 mdname(mddev)); 4983 return -EINVAL; 4984 } 4985 } 4986 sysfs_notify_dirent_safe(rdev->sysfs_state); 4987 } 4988 4989 if (mddev->bio_set == NULL) 4990 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); 4991 4992 spin_lock(&pers_lock); 4993 pers = find_pers(mddev->level, mddev->clevel); 4994 if (!pers || !try_module_get(pers->owner)) { 4995 spin_unlock(&pers_lock); 4996 if (mddev->level != LEVEL_NONE) 4997 printk(KERN_WARNING "md: personality for level %d is not loaded!\n", 4998 mddev->level); 4999 else 5000 printk(KERN_WARNING "md: personality for level %s is not loaded!\n", 5001 mddev->clevel); 5002 return -EINVAL; 5003 } 5004 mddev->pers = pers; 5005 spin_unlock(&pers_lock); 5006 if (mddev->level != pers->level) { 5007 mddev->level = pers->level; 5008 mddev->new_level = pers->level; 5009 } 5010 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 5011 5012 if (mddev->reshape_position != MaxSector && 5013 pers->start_reshape == NULL) { 5014 /* This personality cannot handle reshaping... */ 5015 mddev->pers = NULL; 5016 module_put(pers->owner); 5017 return -EINVAL; 5018 } 5019 5020 if (pers->sync_request) { 5021 /* Warn if this is a potentially silly 5022 * configuration. 5023 */ 5024 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 5025 struct md_rdev *rdev2; 5026 int warned = 0; 5027 5028 rdev_for_each(rdev, mddev) 5029 rdev_for_each(rdev2, mddev) { 5030 if (rdev < rdev2 && 5031 rdev->bdev->bd_contains == 5032 rdev2->bdev->bd_contains) { 5033 printk(KERN_WARNING 5034 "%s: WARNING: %s appears to be" 5035 " on the same physical disk as" 5036 " %s.\n", 5037 mdname(mddev), 5038 bdevname(rdev->bdev,b), 5039 bdevname(rdev2->bdev,b2)); 5040 warned = 1; 5041 } 5042 } 5043 5044 if (warned) 5045 printk(KERN_WARNING 5046 "True protection against single-disk" 5047 " failure might be compromised.\n"); 5048 } 5049 5050 mddev->recovery = 0; 5051 /* may be over-ridden by personality */ 5052 mddev->resync_max_sectors = mddev->dev_sectors; 5053 5054 mddev->ok_start_degraded = start_dirty_degraded; 5055 5056 if (start_readonly && mddev->ro == 0) 5057 mddev->ro = 2; /* read-only, but switch on first write */ 5058 5059 err = mddev->pers->run(mddev); 5060 if (err) 5061 printk(KERN_ERR "md: pers->run() failed ...\n"); 5062 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { 5063 WARN_ONCE(!mddev->external_size, "%s: default size too small," 5064 " but 'external_size' not in effect?\n", __func__); 5065 printk(KERN_ERR 5066 "md: invalid array_size %llu > default size %llu\n", 5067 (unsigned long long)mddev->array_sectors / 2, 5068 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); 5069 err = -EINVAL; 5070 mddev->pers->stop(mddev); 5071 } 5072 if (err == 0 && mddev->pers->sync_request && 5073 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { 5074 err = bitmap_create(mddev); 5075 if (err) { 5076 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 5077 mdname(mddev), err); 5078 mddev->pers->stop(mddev); 5079 } 5080 } 5081 if (err) { 5082 module_put(mddev->pers->owner); 5083 mddev->pers = NULL; 5084 bitmap_destroy(mddev); 5085 return err; 5086 } 5087 if (mddev->pers->sync_request) { 5088 if (mddev->kobj.sd && 5089 sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 5090 printk(KERN_WARNING 5091 "md: cannot register extra attributes for %s\n", 5092 mdname(mddev)); 5093 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); 5094 } else if (mddev->ro == 2) /* auto-readonly not meaningful */ 5095 mddev->ro = 0; 5096 5097 atomic_set(&mddev->writes_pending,0); 5098 atomic_set(&mddev->max_corr_read_errors, 5099 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); 5100 mddev->safemode = 0; 5101 mddev->safemode_timer.function = md_safemode_timeout; 5102 mddev->safemode_timer.data = (unsigned long) mddev; 5103 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ 5104 mddev->in_sync = 1; 5105 smp_wmb(); 5106 mddev->ready = 1; 5107 rdev_for_each(rdev, mddev) 5108 if (rdev->raid_disk >= 0) 5109 if (sysfs_link_rdev(mddev, rdev)) 5110 /* failure here is OK */; 5111 5112 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5113 5114 if (mddev->flags) 5115 md_update_sb(mddev, 0); 5116 5117 md_new_event(mddev); 5118 sysfs_notify_dirent_safe(mddev->sysfs_state); 5119 sysfs_notify_dirent_safe(mddev->sysfs_action); 5120 sysfs_notify(&mddev->kobj, NULL, "degraded"); 5121 return 0; 5122 } 5123 EXPORT_SYMBOL_GPL(md_run); 5124 5125 static int do_md_run(struct mddev *mddev) 5126 { 5127 int err; 5128 5129 err = md_run(mddev); 5130 if (err) 5131 goto out; 5132 err = bitmap_load(mddev); 5133 if (err) { 5134 bitmap_destroy(mddev); 5135 goto out; 5136 } 5137 5138 md_wakeup_thread(mddev->thread); 5139 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 5140 5141 set_capacity(mddev->gendisk, mddev->array_sectors); 5142 revalidate_disk(mddev->gendisk); 5143 mddev->changed = 1; 5144 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 5145 out: 5146 return err; 5147 } 5148 5149 static int restart_array(struct mddev *mddev) 5150 { 5151 struct gendisk *disk = mddev->gendisk; 5152 5153 /* Complain if it has no devices */ 5154 if (list_empty(&mddev->disks)) 5155 return -ENXIO; 5156 if (!mddev->pers) 5157 return -EINVAL; 5158 if (!mddev->ro) 5159 return -EBUSY; 5160 mddev->safemode = 0; 5161 mddev->ro = 0; 5162 set_disk_ro(disk, 0); 5163 printk(KERN_INFO "md: %s switched to read-write mode.\n", 5164 mdname(mddev)); 5165 /* Kick recovery or resync if necessary */ 5166 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5167 md_wakeup_thread(mddev->thread); 5168 md_wakeup_thread(mddev->sync_thread); 5169 sysfs_notify_dirent_safe(mddev->sysfs_state); 5170 return 0; 5171 } 5172 5173 /* similar to deny_write_access, but accounts for our holding a reference 5174 * to the file ourselves */ 5175 static int deny_bitmap_write_access(struct file * file) 5176 { 5177 struct inode *inode = file->f_mapping->host; 5178 5179 spin_lock(&inode->i_lock); 5180 if (atomic_read(&inode->i_writecount) > 1) { 5181 spin_unlock(&inode->i_lock); 5182 return -ETXTBSY; 5183 } 5184 atomic_set(&inode->i_writecount, -1); 5185 spin_unlock(&inode->i_lock); 5186 5187 return 0; 5188 } 5189 5190 void restore_bitmap_write_access(struct file *file) 5191 { 5192 struct inode *inode = file->f_mapping->host; 5193 5194 spin_lock(&inode->i_lock); 5195 atomic_set(&inode->i_writecount, 1); 5196 spin_unlock(&inode->i_lock); 5197 } 5198 5199 static void md_clean(struct mddev *mddev) 5200 { 5201 mddev->array_sectors = 0; 5202 mddev->external_size = 0; 5203 mddev->dev_sectors = 0; 5204 mddev->raid_disks = 0; 5205 mddev->recovery_cp = 0; 5206 mddev->resync_min = 0; 5207 mddev->resync_max = MaxSector; 5208 mddev->reshape_position = MaxSector; 5209 mddev->external = 0; 5210 mddev->persistent = 0; 5211 mddev->level = LEVEL_NONE; 5212 mddev->clevel[0] = 0; 5213 mddev->flags = 0; 5214 mddev->ro = 0; 5215 mddev->metadata_type[0] = 0; 5216 mddev->chunk_sectors = 0; 5217 mddev->ctime = mddev->utime = 0; 5218 mddev->layout = 0; 5219 mddev->max_disks = 0; 5220 mddev->events = 0; 5221 mddev->can_decrease_events = 0; 5222 mddev->delta_disks = 0; 5223 mddev->reshape_backwards = 0; 5224 mddev->new_level = LEVEL_NONE; 5225 mddev->new_layout = 0; 5226 mddev->new_chunk_sectors = 0; 5227 mddev->curr_resync = 0; 5228 mddev->resync_mismatches = 0; 5229 mddev->suspend_lo = mddev->suspend_hi = 0; 5230 mddev->sync_speed_min = mddev->sync_speed_max = 0; 5231 mddev->recovery = 0; 5232 mddev->in_sync = 0; 5233 mddev->changed = 0; 5234 mddev->degraded = 0; 5235 mddev->safemode = 0; 5236 mddev->merge_check_needed = 0; 5237 mddev->bitmap_info.offset = 0; 5238 mddev->bitmap_info.default_offset = 0; 5239 mddev->bitmap_info.default_space = 0; 5240 mddev->bitmap_info.chunksize = 0; 5241 mddev->bitmap_info.daemon_sleep = 0; 5242 mddev->bitmap_info.max_write_behind = 0; 5243 } 5244 5245 static void __md_stop_writes(struct mddev *mddev) 5246 { 5247 if (mddev->sync_thread) { 5248 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5249 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 5250 reap_sync_thread(mddev); 5251 } 5252 5253 del_timer_sync(&mddev->safemode_timer); 5254 5255 bitmap_flush(mddev); 5256 md_super_wait(mddev); 5257 5258 if (!mddev->in_sync || mddev->flags) { 5259 /* mark array as shutdown cleanly */ 5260 mddev->in_sync = 1; 5261 md_update_sb(mddev, 1); 5262 } 5263 } 5264 5265 void md_stop_writes(struct mddev *mddev) 5266 { 5267 mddev_lock(mddev); 5268 __md_stop_writes(mddev); 5269 mddev_unlock(mddev); 5270 } 5271 EXPORT_SYMBOL_GPL(md_stop_writes); 5272 5273 void md_stop(struct mddev *mddev) 5274 { 5275 mddev->ready = 0; 5276 mddev->pers->stop(mddev); 5277 if (mddev->pers->sync_request && mddev->to_remove == NULL) 5278 mddev->to_remove = &md_redundancy_group; 5279 module_put(mddev->pers->owner); 5280 mddev->pers = NULL; 5281 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5282 } 5283 EXPORT_SYMBOL_GPL(md_stop); 5284 5285 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) 5286 { 5287 int err = 0; 5288 mutex_lock(&mddev->open_mutex); 5289 if (atomic_read(&mddev->openers) > !!bdev) { 5290 printk("md: %s still in use.\n",mdname(mddev)); 5291 err = -EBUSY; 5292 goto out; 5293 } 5294 if (bdev) 5295 sync_blockdev(bdev); 5296 if (mddev->pers) { 5297 __md_stop_writes(mddev); 5298 5299 err = -ENXIO; 5300 if (mddev->ro==1) 5301 goto out; 5302 mddev->ro = 1; 5303 set_disk_ro(mddev->gendisk, 1); 5304 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5305 sysfs_notify_dirent_safe(mddev->sysfs_state); 5306 err = 0; 5307 } 5308 out: 5309 mutex_unlock(&mddev->open_mutex); 5310 return err; 5311 } 5312 5313 /* mode: 5314 * 0 - completely stop and dis-assemble array 5315 * 2 - stop but do not disassemble array 5316 */ 5317 static int do_md_stop(struct mddev * mddev, int mode, 5318 struct block_device *bdev) 5319 { 5320 struct gendisk *disk = mddev->gendisk; 5321 struct md_rdev *rdev; 5322 5323 mutex_lock(&mddev->open_mutex); 5324 if (atomic_read(&mddev->openers) > !!bdev || 5325 mddev->sysfs_active) { 5326 printk("md: %s still in use.\n",mdname(mddev)); 5327 mutex_unlock(&mddev->open_mutex); 5328 return -EBUSY; 5329 } 5330 if (bdev) 5331 /* It is possible IO was issued on some other 5332 * open file which was closed before we took ->open_mutex. 5333 * As that was not the last close __blkdev_put will not 5334 * have called sync_blockdev, so we must. 5335 */ 5336 sync_blockdev(bdev); 5337 5338 if (mddev->pers) { 5339 if (mddev->ro) 5340 set_disk_ro(disk, 0); 5341 5342 __md_stop_writes(mddev); 5343 md_stop(mddev); 5344 mddev->queue->merge_bvec_fn = NULL; 5345 mddev->queue->backing_dev_info.congested_fn = NULL; 5346 5347 /* tell userspace to handle 'inactive' */ 5348 sysfs_notify_dirent_safe(mddev->sysfs_state); 5349 5350 rdev_for_each(rdev, mddev) 5351 if (rdev->raid_disk >= 0) 5352 sysfs_unlink_rdev(mddev, rdev); 5353 5354 set_capacity(disk, 0); 5355 mutex_unlock(&mddev->open_mutex); 5356 mddev->changed = 1; 5357 revalidate_disk(disk); 5358 5359 if (mddev->ro) 5360 mddev->ro = 0; 5361 } else 5362 mutex_unlock(&mddev->open_mutex); 5363 /* 5364 * Free resources if final stop 5365 */ 5366 if (mode == 0) { 5367 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 5368 5369 bitmap_destroy(mddev); 5370 if (mddev->bitmap_info.file) { 5371 restore_bitmap_write_access(mddev->bitmap_info.file); 5372 fput(mddev->bitmap_info.file); 5373 mddev->bitmap_info.file = NULL; 5374 } 5375 mddev->bitmap_info.offset = 0; 5376 5377 export_array(mddev); 5378 5379 md_clean(mddev); 5380 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 5381 if (mddev->hold_active == UNTIL_STOP) 5382 mddev->hold_active = 0; 5383 } 5384 blk_integrity_unregister(disk); 5385 md_new_event(mddev); 5386 sysfs_notify_dirent_safe(mddev->sysfs_state); 5387 return 0; 5388 } 5389 5390 #ifndef MODULE 5391 static void autorun_array(struct mddev *mddev) 5392 { 5393 struct md_rdev *rdev; 5394 int err; 5395 5396 if (list_empty(&mddev->disks)) 5397 return; 5398 5399 printk(KERN_INFO "md: running: "); 5400 5401 rdev_for_each(rdev, mddev) { 5402 char b[BDEVNAME_SIZE]; 5403 printk("<%s>", bdevname(rdev->bdev,b)); 5404 } 5405 printk("\n"); 5406 5407 err = do_md_run(mddev); 5408 if (err) { 5409 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 5410 do_md_stop(mddev, 0, NULL); 5411 } 5412 } 5413 5414 /* 5415 * lets try to run arrays based on all disks that have arrived 5416 * until now. (those are in pending_raid_disks) 5417 * 5418 * the method: pick the first pending disk, collect all disks with 5419 * the same UUID, remove all from the pending list and put them into 5420 * the 'same_array' list. Then order this list based on superblock 5421 * update time (freshest comes first), kick out 'old' disks and 5422 * compare superblocks. If everything's fine then run it. 5423 * 5424 * If "unit" is allocated, then bump its reference count 5425 */ 5426 static void autorun_devices(int part) 5427 { 5428 struct md_rdev *rdev0, *rdev, *tmp; 5429 struct mddev *mddev; 5430 char b[BDEVNAME_SIZE]; 5431 5432 printk(KERN_INFO "md: autorun ...\n"); 5433 while (!list_empty(&pending_raid_disks)) { 5434 int unit; 5435 dev_t dev; 5436 LIST_HEAD(candidates); 5437 rdev0 = list_entry(pending_raid_disks.next, 5438 struct md_rdev, same_set); 5439 5440 printk(KERN_INFO "md: considering %s ...\n", 5441 bdevname(rdev0->bdev,b)); 5442 INIT_LIST_HEAD(&candidates); 5443 rdev_for_each_list(rdev, tmp, &pending_raid_disks) 5444 if (super_90_load(rdev, rdev0, 0) >= 0) { 5445 printk(KERN_INFO "md: adding %s ...\n", 5446 bdevname(rdev->bdev,b)); 5447 list_move(&rdev->same_set, &candidates); 5448 } 5449 /* 5450 * now we have a set of devices, with all of them having 5451 * mostly sane superblocks. It's time to allocate the 5452 * mddev. 5453 */ 5454 if (part) { 5455 dev = MKDEV(mdp_major, 5456 rdev0->preferred_minor << MdpMinorShift); 5457 unit = MINOR(dev) >> MdpMinorShift; 5458 } else { 5459 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 5460 unit = MINOR(dev); 5461 } 5462 if (rdev0->preferred_minor != unit) { 5463 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 5464 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 5465 break; 5466 } 5467 5468 md_probe(dev, NULL, NULL); 5469 mddev = mddev_find(dev); 5470 if (!mddev || !mddev->gendisk) { 5471 if (mddev) 5472 mddev_put(mddev); 5473 printk(KERN_ERR 5474 "md: cannot allocate memory for md drive.\n"); 5475 break; 5476 } 5477 if (mddev_lock(mddev)) 5478 printk(KERN_WARNING "md: %s locked, cannot run\n", 5479 mdname(mddev)); 5480 else if (mddev->raid_disks || mddev->major_version 5481 || !list_empty(&mddev->disks)) { 5482 printk(KERN_WARNING 5483 "md: %s already running, cannot run %s\n", 5484 mdname(mddev), bdevname(rdev0->bdev,b)); 5485 mddev_unlock(mddev); 5486 } else { 5487 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 5488 mddev->persistent = 1; 5489 rdev_for_each_list(rdev, tmp, &candidates) { 5490 list_del_init(&rdev->same_set); 5491 if (bind_rdev_to_array(rdev, mddev)) 5492 export_rdev(rdev); 5493 } 5494 autorun_array(mddev); 5495 mddev_unlock(mddev); 5496 } 5497 /* on success, candidates will be empty, on error 5498 * it won't... 5499 */ 5500 rdev_for_each_list(rdev, tmp, &candidates) { 5501 list_del_init(&rdev->same_set); 5502 export_rdev(rdev); 5503 } 5504 mddev_put(mddev); 5505 } 5506 printk(KERN_INFO "md: ... autorun DONE.\n"); 5507 } 5508 #endif /* !MODULE */ 5509 5510 static int get_version(void __user * arg) 5511 { 5512 mdu_version_t ver; 5513 5514 ver.major = MD_MAJOR_VERSION; 5515 ver.minor = MD_MINOR_VERSION; 5516 ver.patchlevel = MD_PATCHLEVEL_VERSION; 5517 5518 if (copy_to_user(arg, &ver, sizeof(ver))) 5519 return -EFAULT; 5520 5521 return 0; 5522 } 5523 5524 static int get_array_info(struct mddev * mddev, void __user * arg) 5525 { 5526 mdu_array_info_t info; 5527 int nr,working,insync,failed,spare; 5528 struct md_rdev *rdev; 5529 5530 nr=working=insync=failed=spare=0; 5531 rdev_for_each(rdev, mddev) { 5532 nr++; 5533 if (test_bit(Faulty, &rdev->flags)) 5534 failed++; 5535 else { 5536 working++; 5537 if (test_bit(In_sync, &rdev->flags)) 5538 insync++; 5539 else 5540 spare++; 5541 } 5542 } 5543 5544 info.major_version = mddev->major_version; 5545 info.minor_version = mddev->minor_version; 5546 info.patch_version = MD_PATCHLEVEL_VERSION; 5547 info.ctime = mddev->ctime; 5548 info.level = mddev->level; 5549 info.size = mddev->dev_sectors / 2; 5550 if (info.size != mddev->dev_sectors / 2) /* overflow */ 5551 info.size = -1; 5552 info.nr_disks = nr; 5553 info.raid_disks = mddev->raid_disks; 5554 info.md_minor = mddev->md_minor; 5555 info.not_persistent= !mddev->persistent; 5556 5557 info.utime = mddev->utime; 5558 info.state = 0; 5559 if (mddev->in_sync) 5560 info.state = (1<<MD_SB_CLEAN); 5561 if (mddev->bitmap && mddev->bitmap_info.offset) 5562 info.state = (1<<MD_SB_BITMAP_PRESENT); 5563 info.active_disks = insync; 5564 info.working_disks = working; 5565 info.failed_disks = failed; 5566 info.spare_disks = spare; 5567 5568 info.layout = mddev->layout; 5569 info.chunk_size = mddev->chunk_sectors << 9; 5570 5571 if (copy_to_user(arg, &info, sizeof(info))) 5572 return -EFAULT; 5573 5574 return 0; 5575 } 5576 5577 static int get_bitmap_file(struct mddev * mddev, void __user * arg) 5578 { 5579 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 5580 char *ptr, *buf = NULL; 5581 int err = -ENOMEM; 5582 5583 if (md_allow_write(mddev)) 5584 file = kmalloc(sizeof(*file), GFP_NOIO); 5585 else 5586 file = kmalloc(sizeof(*file), GFP_KERNEL); 5587 5588 if (!file) 5589 goto out; 5590 5591 /* bitmap disabled, zero the first byte and copy out */ 5592 if (!mddev->bitmap || !mddev->bitmap->storage.file) { 5593 file->pathname[0] = '\0'; 5594 goto copy_out; 5595 } 5596 5597 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 5598 if (!buf) 5599 goto out; 5600 5601 ptr = d_path(&mddev->bitmap->storage.file->f_path, 5602 buf, sizeof(file->pathname)); 5603 if (IS_ERR(ptr)) 5604 goto out; 5605 5606 strcpy(file->pathname, ptr); 5607 5608 copy_out: 5609 err = 0; 5610 if (copy_to_user(arg, file, sizeof(*file))) 5611 err = -EFAULT; 5612 out: 5613 kfree(buf); 5614 kfree(file); 5615 return err; 5616 } 5617 5618 static int get_disk_info(struct mddev * mddev, void __user * arg) 5619 { 5620 mdu_disk_info_t info; 5621 struct md_rdev *rdev; 5622 5623 if (copy_from_user(&info, arg, sizeof(info))) 5624 return -EFAULT; 5625 5626 rdev = find_rdev_nr(mddev, info.number); 5627 if (rdev) { 5628 info.major = MAJOR(rdev->bdev->bd_dev); 5629 info.minor = MINOR(rdev->bdev->bd_dev); 5630 info.raid_disk = rdev->raid_disk; 5631 info.state = 0; 5632 if (test_bit(Faulty, &rdev->flags)) 5633 info.state |= (1<<MD_DISK_FAULTY); 5634 else if (test_bit(In_sync, &rdev->flags)) { 5635 info.state |= (1<<MD_DISK_ACTIVE); 5636 info.state |= (1<<MD_DISK_SYNC); 5637 } 5638 if (test_bit(WriteMostly, &rdev->flags)) 5639 info.state |= (1<<MD_DISK_WRITEMOSTLY); 5640 } else { 5641 info.major = info.minor = 0; 5642 info.raid_disk = -1; 5643 info.state = (1<<MD_DISK_REMOVED); 5644 } 5645 5646 if (copy_to_user(arg, &info, sizeof(info))) 5647 return -EFAULT; 5648 5649 return 0; 5650 } 5651 5652 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info) 5653 { 5654 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 5655 struct md_rdev *rdev; 5656 dev_t dev = MKDEV(info->major,info->minor); 5657 5658 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 5659 return -EOVERFLOW; 5660 5661 if (!mddev->raid_disks) { 5662 int err; 5663 /* expecting a device which has a superblock */ 5664 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 5665 if (IS_ERR(rdev)) { 5666 printk(KERN_WARNING 5667 "md: md_import_device returned %ld\n", 5668 PTR_ERR(rdev)); 5669 return PTR_ERR(rdev); 5670 } 5671 if (!list_empty(&mddev->disks)) { 5672 struct md_rdev *rdev0 5673 = list_entry(mddev->disks.next, 5674 struct md_rdev, same_set); 5675 err = super_types[mddev->major_version] 5676 .load_super(rdev, rdev0, mddev->minor_version); 5677 if (err < 0) { 5678 printk(KERN_WARNING 5679 "md: %s has different UUID to %s\n", 5680 bdevname(rdev->bdev,b), 5681 bdevname(rdev0->bdev,b2)); 5682 export_rdev(rdev); 5683 return -EINVAL; 5684 } 5685 } 5686 err = bind_rdev_to_array(rdev, mddev); 5687 if (err) 5688 export_rdev(rdev); 5689 return err; 5690 } 5691 5692 /* 5693 * add_new_disk can be used once the array is assembled 5694 * to add "hot spares". They must already have a superblock 5695 * written 5696 */ 5697 if (mddev->pers) { 5698 int err; 5699 if (!mddev->pers->hot_add_disk) { 5700 printk(KERN_WARNING 5701 "%s: personality does not support diskops!\n", 5702 mdname(mddev)); 5703 return -EINVAL; 5704 } 5705 if (mddev->persistent) 5706 rdev = md_import_device(dev, mddev->major_version, 5707 mddev->minor_version); 5708 else 5709 rdev = md_import_device(dev, -1, -1); 5710 if (IS_ERR(rdev)) { 5711 printk(KERN_WARNING 5712 "md: md_import_device returned %ld\n", 5713 PTR_ERR(rdev)); 5714 return PTR_ERR(rdev); 5715 } 5716 /* set saved_raid_disk if appropriate */ 5717 if (!mddev->persistent) { 5718 if (info->state & (1<<MD_DISK_SYNC) && 5719 info->raid_disk < mddev->raid_disks) { 5720 rdev->raid_disk = info->raid_disk; 5721 set_bit(In_sync, &rdev->flags); 5722 } else 5723 rdev->raid_disk = -1; 5724 } else 5725 super_types[mddev->major_version]. 5726 validate_super(mddev, rdev); 5727 if ((info->state & (1<<MD_DISK_SYNC)) && 5728 rdev->raid_disk != info->raid_disk) { 5729 /* This was a hot-add request, but events doesn't 5730 * match, so reject it. 5731 */ 5732 export_rdev(rdev); 5733 return -EINVAL; 5734 } 5735 5736 if (test_bit(In_sync, &rdev->flags)) 5737 rdev->saved_raid_disk = rdev->raid_disk; 5738 else 5739 rdev->saved_raid_disk = -1; 5740 5741 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 5742 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5743 set_bit(WriteMostly, &rdev->flags); 5744 else 5745 clear_bit(WriteMostly, &rdev->flags); 5746 5747 rdev->raid_disk = -1; 5748 err = bind_rdev_to_array(rdev, mddev); 5749 if (!err && !mddev->pers->hot_remove_disk) { 5750 /* If there is hot_add_disk but no hot_remove_disk 5751 * then added disks for geometry changes, 5752 * and should be added immediately. 5753 */ 5754 super_types[mddev->major_version]. 5755 validate_super(mddev, rdev); 5756 err = mddev->pers->hot_add_disk(mddev, rdev); 5757 if (err) 5758 unbind_rdev_from_array(rdev); 5759 } 5760 if (err) 5761 export_rdev(rdev); 5762 else 5763 sysfs_notify_dirent_safe(rdev->sysfs_state); 5764 5765 md_update_sb(mddev, 1); 5766 if (mddev->degraded) 5767 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 5768 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5769 if (!err) 5770 md_new_event(mddev); 5771 md_wakeup_thread(mddev->thread); 5772 return err; 5773 } 5774 5775 /* otherwise, add_new_disk is only allowed 5776 * for major_version==0 superblocks 5777 */ 5778 if (mddev->major_version != 0) { 5779 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 5780 mdname(mddev)); 5781 return -EINVAL; 5782 } 5783 5784 if (!(info->state & (1<<MD_DISK_FAULTY))) { 5785 int err; 5786 rdev = md_import_device(dev, -1, 0); 5787 if (IS_ERR(rdev)) { 5788 printk(KERN_WARNING 5789 "md: error, md_import_device() returned %ld\n", 5790 PTR_ERR(rdev)); 5791 return PTR_ERR(rdev); 5792 } 5793 rdev->desc_nr = info->number; 5794 if (info->raid_disk < mddev->raid_disks) 5795 rdev->raid_disk = info->raid_disk; 5796 else 5797 rdev->raid_disk = -1; 5798 5799 if (rdev->raid_disk < mddev->raid_disks) 5800 if (info->state & (1<<MD_DISK_SYNC)) 5801 set_bit(In_sync, &rdev->flags); 5802 5803 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5804 set_bit(WriteMostly, &rdev->flags); 5805 5806 if (!mddev->persistent) { 5807 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 5808 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5809 } else 5810 rdev->sb_start = calc_dev_sboffset(rdev); 5811 rdev->sectors = rdev->sb_start; 5812 5813 err = bind_rdev_to_array(rdev, mddev); 5814 if (err) { 5815 export_rdev(rdev); 5816 return err; 5817 } 5818 } 5819 5820 return 0; 5821 } 5822 5823 static int hot_remove_disk(struct mddev * mddev, dev_t dev) 5824 { 5825 char b[BDEVNAME_SIZE]; 5826 struct md_rdev *rdev; 5827 5828 rdev = find_rdev(mddev, dev); 5829 if (!rdev) 5830 return -ENXIO; 5831 5832 if (rdev->raid_disk >= 0) 5833 goto busy; 5834 5835 kick_rdev_from_array(rdev); 5836 md_update_sb(mddev, 1); 5837 md_new_event(mddev); 5838 5839 return 0; 5840 busy: 5841 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", 5842 bdevname(rdev->bdev,b), mdname(mddev)); 5843 return -EBUSY; 5844 } 5845 5846 static int hot_add_disk(struct mddev * mddev, dev_t dev) 5847 { 5848 char b[BDEVNAME_SIZE]; 5849 int err; 5850 struct md_rdev *rdev; 5851 5852 if (!mddev->pers) 5853 return -ENODEV; 5854 5855 if (mddev->major_version != 0) { 5856 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 5857 " version-0 superblocks.\n", 5858 mdname(mddev)); 5859 return -EINVAL; 5860 } 5861 if (!mddev->pers->hot_add_disk) { 5862 printk(KERN_WARNING 5863 "%s: personality does not support diskops!\n", 5864 mdname(mddev)); 5865 return -EINVAL; 5866 } 5867 5868 rdev = md_import_device(dev, -1, 0); 5869 if (IS_ERR(rdev)) { 5870 printk(KERN_WARNING 5871 "md: error, md_import_device() returned %ld\n", 5872 PTR_ERR(rdev)); 5873 return -EINVAL; 5874 } 5875 5876 if (mddev->persistent) 5877 rdev->sb_start = calc_dev_sboffset(rdev); 5878 else 5879 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5880 5881 rdev->sectors = rdev->sb_start; 5882 5883 if (test_bit(Faulty, &rdev->flags)) { 5884 printk(KERN_WARNING 5885 "md: can not hot-add faulty %s disk to %s!\n", 5886 bdevname(rdev->bdev,b), mdname(mddev)); 5887 err = -EINVAL; 5888 goto abort_export; 5889 } 5890 clear_bit(In_sync, &rdev->flags); 5891 rdev->desc_nr = -1; 5892 rdev->saved_raid_disk = -1; 5893 err = bind_rdev_to_array(rdev, mddev); 5894 if (err) 5895 goto abort_export; 5896 5897 /* 5898 * The rest should better be atomic, we can have disk failures 5899 * noticed in interrupt contexts ... 5900 */ 5901 5902 rdev->raid_disk = -1; 5903 5904 md_update_sb(mddev, 1); 5905 5906 /* 5907 * Kick recovery, maybe this spare has to be added to the 5908 * array immediately. 5909 */ 5910 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5911 md_wakeup_thread(mddev->thread); 5912 md_new_event(mddev); 5913 return 0; 5914 5915 abort_export: 5916 export_rdev(rdev); 5917 return err; 5918 } 5919 5920 static int set_bitmap_file(struct mddev *mddev, int fd) 5921 { 5922 int err; 5923 5924 if (mddev->pers) { 5925 if (!mddev->pers->quiesce) 5926 return -EBUSY; 5927 if (mddev->recovery || mddev->sync_thread) 5928 return -EBUSY; 5929 /* we should be able to change the bitmap.. */ 5930 } 5931 5932 5933 if (fd >= 0) { 5934 if (mddev->bitmap) 5935 return -EEXIST; /* cannot add when bitmap is present */ 5936 mddev->bitmap_info.file = fget(fd); 5937 5938 if (mddev->bitmap_info.file == NULL) { 5939 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 5940 mdname(mddev)); 5941 return -EBADF; 5942 } 5943 5944 err = deny_bitmap_write_access(mddev->bitmap_info.file); 5945 if (err) { 5946 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 5947 mdname(mddev)); 5948 fput(mddev->bitmap_info.file); 5949 mddev->bitmap_info.file = NULL; 5950 return err; 5951 } 5952 mddev->bitmap_info.offset = 0; /* file overrides offset */ 5953 } else if (mddev->bitmap == NULL) 5954 return -ENOENT; /* cannot remove what isn't there */ 5955 err = 0; 5956 if (mddev->pers) { 5957 mddev->pers->quiesce(mddev, 1); 5958 if (fd >= 0) { 5959 err = bitmap_create(mddev); 5960 if (!err) 5961 err = bitmap_load(mddev); 5962 } 5963 if (fd < 0 || err) { 5964 bitmap_destroy(mddev); 5965 fd = -1; /* make sure to put the file */ 5966 } 5967 mddev->pers->quiesce(mddev, 0); 5968 } 5969 if (fd < 0) { 5970 if (mddev->bitmap_info.file) { 5971 restore_bitmap_write_access(mddev->bitmap_info.file); 5972 fput(mddev->bitmap_info.file); 5973 } 5974 mddev->bitmap_info.file = NULL; 5975 } 5976 5977 return err; 5978 } 5979 5980 /* 5981 * set_array_info is used two different ways 5982 * The original usage is when creating a new array. 5983 * In this usage, raid_disks is > 0 and it together with 5984 * level, size, not_persistent,layout,chunksize determine the 5985 * shape of the array. 5986 * This will always create an array with a type-0.90.0 superblock. 5987 * The newer usage is when assembling an array. 5988 * In this case raid_disks will be 0, and the major_version field is 5989 * use to determine which style super-blocks are to be found on the devices. 5990 * The minor and patch _version numbers are also kept incase the 5991 * super_block handler wishes to interpret them. 5992 */ 5993 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info) 5994 { 5995 5996 if (info->raid_disks == 0) { 5997 /* just setting version number for superblock loading */ 5998 if (info->major_version < 0 || 5999 info->major_version >= ARRAY_SIZE(super_types) || 6000 super_types[info->major_version].name == NULL) { 6001 /* maybe try to auto-load a module? */ 6002 printk(KERN_INFO 6003 "md: superblock version %d not known\n", 6004 info->major_version); 6005 return -EINVAL; 6006 } 6007 mddev->major_version = info->major_version; 6008 mddev->minor_version = info->minor_version; 6009 mddev->patch_version = info->patch_version; 6010 mddev->persistent = !info->not_persistent; 6011 /* ensure mddev_put doesn't delete this now that there 6012 * is some minimal configuration. 6013 */ 6014 mddev->ctime = get_seconds(); 6015 return 0; 6016 } 6017 mddev->major_version = MD_MAJOR_VERSION; 6018 mddev->minor_version = MD_MINOR_VERSION; 6019 mddev->patch_version = MD_PATCHLEVEL_VERSION; 6020 mddev->ctime = get_seconds(); 6021 6022 mddev->level = info->level; 6023 mddev->clevel[0] = 0; 6024 mddev->dev_sectors = 2 * (sector_t)info->size; 6025 mddev->raid_disks = info->raid_disks; 6026 /* don't set md_minor, it is determined by which /dev/md* was 6027 * openned 6028 */ 6029 if (info->state & (1<<MD_SB_CLEAN)) 6030 mddev->recovery_cp = MaxSector; 6031 else 6032 mddev->recovery_cp = 0; 6033 mddev->persistent = ! info->not_persistent; 6034 mddev->external = 0; 6035 6036 mddev->layout = info->layout; 6037 mddev->chunk_sectors = info->chunk_size >> 9; 6038 6039 mddev->max_disks = MD_SB_DISKS; 6040 6041 if (mddev->persistent) 6042 mddev->flags = 0; 6043 set_bit(MD_CHANGE_DEVS, &mddev->flags); 6044 6045 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 6046 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); 6047 mddev->bitmap_info.offset = 0; 6048 6049 mddev->reshape_position = MaxSector; 6050 6051 /* 6052 * Generate a 128 bit UUID 6053 */ 6054 get_random_bytes(mddev->uuid, 16); 6055 6056 mddev->new_level = mddev->level; 6057 mddev->new_chunk_sectors = mddev->chunk_sectors; 6058 mddev->new_layout = mddev->layout; 6059 mddev->delta_disks = 0; 6060 mddev->reshape_backwards = 0; 6061 6062 return 0; 6063 } 6064 6065 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) 6066 { 6067 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); 6068 6069 if (mddev->external_size) 6070 return; 6071 6072 mddev->array_sectors = array_sectors; 6073 } 6074 EXPORT_SYMBOL(md_set_array_sectors); 6075 6076 static int update_size(struct mddev *mddev, sector_t num_sectors) 6077 { 6078 struct md_rdev *rdev; 6079 int rv; 6080 int fit = (num_sectors == 0); 6081 6082 if (mddev->pers->resize == NULL) 6083 return -EINVAL; 6084 /* The "num_sectors" is the number of sectors of each device that 6085 * is used. This can only make sense for arrays with redundancy. 6086 * linear and raid0 always use whatever space is available. We can only 6087 * consider changing this number if no resync or reconstruction is 6088 * happening, and if the new size is acceptable. It must fit before the 6089 * sb_start or, if that is <data_offset, it must fit before the size 6090 * of each device. If num_sectors is zero, we find the largest size 6091 * that fits. 6092 */ 6093 if (mddev->sync_thread) 6094 return -EBUSY; 6095 6096 rdev_for_each(rdev, mddev) { 6097 sector_t avail = rdev->sectors; 6098 6099 if (fit && (num_sectors == 0 || num_sectors > avail)) 6100 num_sectors = avail; 6101 if (avail < num_sectors) 6102 return -ENOSPC; 6103 } 6104 rv = mddev->pers->resize(mddev, num_sectors); 6105 if (!rv) 6106 revalidate_disk(mddev->gendisk); 6107 return rv; 6108 } 6109 6110 static int update_raid_disks(struct mddev *mddev, int raid_disks) 6111 { 6112 int rv; 6113 struct md_rdev *rdev; 6114 /* change the number of raid disks */ 6115 if (mddev->pers->check_reshape == NULL) 6116 return -EINVAL; 6117 if (raid_disks <= 0 || 6118 (mddev->max_disks && raid_disks >= mddev->max_disks)) 6119 return -EINVAL; 6120 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 6121 return -EBUSY; 6122 6123 rdev_for_each(rdev, mddev) { 6124 if (mddev->raid_disks < raid_disks && 6125 rdev->data_offset < rdev->new_data_offset) 6126 return -EINVAL; 6127 if (mddev->raid_disks > raid_disks && 6128 rdev->data_offset > rdev->new_data_offset) 6129 return -EINVAL; 6130 } 6131 6132 mddev->delta_disks = raid_disks - mddev->raid_disks; 6133 if (mddev->delta_disks < 0) 6134 mddev->reshape_backwards = 1; 6135 else if (mddev->delta_disks > 0) 6136 mddev->reshape_backwards = 0; 6137 6138 rv = mddev->pers->check_reshape(mddev); 6139 if (rv < 0) { 6140 mddev->delta_disks = 0; 6141 mddev->reshape_backwards = 0; 6142 } 6143 return rv; 6144 } 6145 6146 6147 /* 6148 * update_array_info is used to change the configuration of an 6149 * on-line array. 6150 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 6151 * fields in the info are checked against the array. 6152 * Any differences that cannot be handled will cause an error. 6153 * Normally, only one change can be managed at a time. 6154 */ 6155 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) 6156 { 6157 int rv = 0; 6158 int cnt = 0; 6159 int state = 0; 6160 6161 /* calculate expected state,ignoring low bits */ 6162 if (mddev->bitmap && mddev->bitmap_info.offset) 6163 state |= (1 << MD_SB_BITMAP_PRESENT); 6164 6165 if (mddev->major_version != info->major_version || 6166 mddev->minor_version != info->minor_version || 6167 /* mddev->patch_version != info->patch_version || */ 6168 mddev->ctime != info->ctime || 6169 mddev->level != info->level || 6170 /* mddev->layout != info->layout || */ 6171 !mddev->persistent != info->not_persistent|| 6172 mddev->chunk_sectors != info->chunk_size >> 9 || 6173 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 6174 ((state^info->state) & 0xfffffe00) 6175 ) 6176 return -EINVAL; 6177 /* Check there is only one change */ 6178 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 6179 cnt++; 6180 if (mddev->raid_disks != info->raid_disks) 6181 cnt++; 6182 if (mddev->layout != info->layout) 6183 cnt++; 6184 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) 6185 cnt++; 6186 if (cnt == 0) 6187 return 0; 6188 if (cnt > 1) 6189 return -EINVAL; 6190 6191 if (mddev->layout != info->layout) { 6192 /* Change layout 6193 * we don't need to do anything at the md level, the 6194 * personality will take care of it all. 6195 */ 6196 if (mddev->pers->check_reshape == NULL) 6197 return -EINVAL; 6198 else { 6199 mddev->new_layout = info->layout; 6200 rv = mddev->pers->check_reshape(mddev); 6201 if (rv) 6202 mddev->new_layout = mddev->layout; 6203 return rv; 6204 } 6205 } 6206 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 6207 rv = update_size(mddev, (sector_t)info->size * 2); 6208 6209 if (mddev->raid_disks != info->raid_disks) 6210 rv = update_raid_disks(mddev, info->raid_disks); 6211 6212 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 6213 if (mddev->pers->quiesce == NULL) 6214 return -EINVAL; 6215 if (mddev->recovery || mddev->sync_thread) 6216 return -EBUSY; 6217 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 6218 /* add the bitmap */ 6219 if (mddev->bitmap) 6220 return -EEXIST; 6221 if (mddev->bitmap_info.default_offset == 0) 6222 return -EINVAL; 6223 mddev->bitmap_info.offset = 6224 mddev->bitmap_info.default_offset; 6225 mddev->bitmap_info.space = 6226 mddev->bitmap_info.default_space; 6227 mddev->pers->quiesce(mddev, 1); 6228 rv = bitmap_create(mddev); 6229 if (!rv) 6230 rv = bitmap_load(mddev); 6231 if (rv) 6232 bitmap_destroy(mddev); 6233 mddev->pers->quiesce(mddev, 0); 6234 } else { 6235 /* remove the bitmap */ 6236 if (!mddev->bitmap) 6237 return -ENOENT; 6238 if (mddev->bitmap->storage.file) 6239 return -EINVAL; 6240 mddev->pers->quiesce(mddev, 1); 6241 bitmap_destroy(mddev); 6242 mddev->pers->quiesce(mddev, 0); 6243 mddev->bitmap_info.offset = 0; 6244 } 6245 } 6246 md_update_sb(mddev, 1); 6247 return rv; 6248 } 6249 6250 static int set_disk_faulty(struct mddev *mddev, dev_t dev) 6251 { 6252 struct md_rdev *rdev; 6253 6254 if (mddev->pers == NULL) 6255 return -ENODEV; 6256 6257 rdev = find_rdev(mddev, dev); 6258 if (!rdev) 6259 return -ENODEV; 6260 6261 md_error(mddev, rdev); 6262 if (!test_bit(Faulty, &rdev->flags)) 6263 return -EBUSY; 6264 return 0; 6265 } 6266 6267 /* 6268 * We have a problem here : there is no easy way to give a CHS 6269 * virtual geometry. We currently pretend that we have a 2 heads 6270 * 4 sectors (with a BIG number of cylinders...). This drives 6271 * dosfs just mad... ;-) 6272 */ 6273 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) 6274 { 6275 struct mddev *mddev = bdev->bd_disk->private_data; 6276 6277 geo->heads = 2; 6278 geo->sectors = 4; 6279 geo->cylinders = mddev->array_sectors / 8; 6280 return 0; 6281 } 6282 6283 static int md_ioctl(struct block_device *bdev, fmode_t mode, 6284 unsigned int cmd, unsigned long arg) 6285 { 6286 int err = 0; 6287 void __user *argp = (void __user *)arg; 6288 struct mddev *mddev = NULL; 6289 int ro; 6290 6291 switch (cmd) { 6292 case RAID_VERSION: 6293 case GET_ARRAY_INFO: 6294 case GET_DISK_INFO: 6295 break; 6296 default: 6297 if (!capable(CAP_SYS_ADMIN)) 6298 return -EACCES; 6299 } 6300 6301 /* 6302 * Commands dealing with the RAID driver but not any 6303 * particular array: 6304 */ 6305 switch (cmd) 6306 { 6307 case RAID_VERSION: 6308 err = get_version(argp); 6309 goto done; 6310 6311 case PRINT_RAID_DEBUG: 6312 err = 0; 6313 md_print_devices(); 6314 goto done; 6315 6316 #ifndef MODULE 6317 case RAID_AUTORUN: 6318 err = 0; 6319 autostart_arrays(arg); 6320 goto done; 6321 #endif 6322 default:; 6323 } 6324 6325 /* 6326 * Commands creating/starting a new array: 6327 */ 6328 6329 mddev = bdev->bd_disk->private_data; 6330 6331 if (!mddev) { 6332 BUG(); 6333 goto abort; 6334 } 6335 6336 err = mddev_lock(mddev); 6337 if (err) { 6338 printk(KERN_INFO 6339 "md: ioctl lock interrupted, reason %d, cmd %d\n", 6340 err, cmd); 6341 goto abort; 6342 } 6343 6344 switch (cmd) 6345 { 6346 case SET_ARRAY_INFO: 6347 { 6348 mdu_array_info_t info; 6349 if (!arg) 6350 memset(&info, 0, sizeof(info)); 6351 else if (copy_from_user(&info, argp, sizeof(info))) { 6352 err = -EFAULT; 6353 goto abort_unlock; 6354 } 6355 if (mddev->pers) { 6356 err = update_array_info(mddev, &info); 6357 if (err) { 6358 printk(KERN_WARNING "md: couldn't update" 6359 " array info. %d\n", err); 6360 goto abort_unlock; 6361 } 6362 goto done_unlock; 6363 } 6364 if (!list_empty(&mddev->disks)) { 6365 printk(KERN_WARNING 6366 "md: array %s already has disks!\n", 6367 mdname(mddev)); 6368 err = -EBUSY; 6369 goto abort_unlock; 6370 } 6371 if (mddev->raid_disks) { 6372 printk(KERN_WARNING 6373 "md: array %s already initialised!\n", 6374 mdname(mddev)); 6375 err = -EBUSY; 6376 goto abort_unlock; 6377 } 6378 err = set_array_info(mddev, &info); 6379 if (err) { 6380 printk(KERN_WARNING "md: couldn't set" 6381 " array info. %d\n", err); 6382 goto abort_unlock; 6383 } 6384 } 6385 goto done_unlock; 6386 6387 default:; 6388 } 6389 6390 /* 6391 * Commands querying/configuring an existing array: 6392 */ 6393 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 6394 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ 6395 if ((!mddev->raid_disks && !mddev->external) 6396 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 6397 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE 6398 && cmd != GET_BITMAP_FILE) { 6399 err = -ENODEV; 6400 goto abort_unlock; 6401 } 6402 6403 /* 6404 * Commands even a read-only array can execute: 6405 */ 6406 switch (cmd) 6407 { 6408 case GET_ARRAY_INFO: 6409 err = get_array_info(mddev, argp); 6410 goto done_unlock; 6411 6412 case GET_BITMAP_FILE: 6413 err = get_bitmap_file(mddev, argp); 6414 goto done_unlock; 6415 6416 case GET_DISK_INFO: 6417 err = get_disk_info(mddev, argp); 6418 goto done_unlock; 6419 6420 case RESTART_ARRAY_RW: 6421 err = restart_array(mddev); 6422 goto done_unlock; 6423 6424 case STOP_ARRAY: 6425 err = do_md_stop(mddev, 0, bdev); 6426 goto done_unlock; 6427 6428 case STOP_ARRAY_RO: 6429 err = md_set_readonly(mddev, bdev); 6430 goto done_unlock; 6431 6432 case BLKROSET: 6433 if (get_user(ro, (int __user *)(arg))) { 6434 err = -EFAULT; 6435 goto done_unlock; 6436 } 6437 err = -EINVAL; 6438 6439 /* if the bdev is going readonly the value of mddev->ro 6440 * does not matter, no writes are coming 6441 */ 6442 if (ro) 6443 goto done_unlock; 6444 6445 /* are we are already prepared for writes? */ 6446 if (mddev->ro != 1) 6447 goto done_unlock; 6448 6449 /* transitioning to readauto need only happen for 6450 * arrays that call md_write_start 6451 */ 6452 if (mddev->pers) { 6453 err = restart_array(mddev); 6454 if (err == 0) { 6455 mddev->ro = 2; 6456 set_disk_ro(mddev->gendisk, 0); 6457 } 6458 } 6459 goto done_unlock; 6460 } 6461 6462 /* 6463 * The remaining ioctls are changing the state of the 6464 * superblock, so we do not allow them on read-only arrays. 6465 * However non-MD ioctls (e.g. get-size) will still come through 6466 * here and hit the 'default' below, so only disallow 6467 * 'md' ioctls, and switch to rw mode if started auto-readonly. 6468 */ 6469 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { 6470 if (mddev->ro == 2) { 6471 mddev->ro = 0; 6472 sysfs_notify_dirent_safe(mddev->sysfs_state); 6473 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6474 md_wakeup_thread(mddev->thread); 6475 } else { 6476 err = -EROFS; 6477 goto abort_unlock; 6478 } 6479 } 6480 6481 switch (cmd) 6482 { 6483 case ADD_NEW_DISK: 6484 { 6485 mdu_disk_info_t info; 6486 if (copy_from_user(&info, argp, sizeof(info))) 6487 err = -EFAULT; 6488 else 6489 err = add_new_disk(mddev, &info); 6490 goto done_unlock; 6491 } 6492 6493 case HOT_REMOVE_DISK: 6494 err = hot_remove_disk(mddev, new_decode_dev(arg)); 6495 goto done_unlock; 6496 6497 case HOT_ADD_DISK: 6498 err = hot_add_disk(mddev, new_decode_dev(arg)); 6499 goto done_unlock; 6500 6501 case SET_DISK_FAULTY: 6502 err = set_disk_faulty(mddev, new_decode_dev(arg)); 6503 goto done_unlock; 6504 6505 case RUN_ARRAY: 6506 err = do_md_run(mddev); 6507 goto done_unlock; 6508 6509 case SET_BITMAP_FILE: 6510 err = set_bitmap_file(mddev, (int)arg); 6511 goto done_unlock; 6512 6513 default: 6514 err = -EINVAL; 6515 goto abort_unlock; 6516 } 6517 6518 done_unlock: 6519 abort_unlock: 6520 if (mddev->hold_active == UNTIL_IOCTL && 6521 err != -EINVAL) 6522 mddev->hold_active = 0; 6523 mddev_unlock(mddev); 6524 6525 return err; 6526 done: 6527 if (err) 6528 MD_BUG(); 6529 abort: 6530 return err; 6531 } 6532 #ifdef CONFIG_COMPAT 6533 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, 6534 unsigned int cmd, unsigned long arg) 6535 { 6536 switch (cmd) { 6537 case HOT_REMOVE_DISK: 6538 case HOT_ADD_DISK: 6539 case SET_DISK_FAULTY: 6540 case SET_BITMAP_FILE: 6541 /* These take in integer arg, do not convert */ 6542 break; 6543 default: 6544 arg = (unsigned long)compat_ptr(arg); 6545 break; 6546 } 6547 6548 return md_ioctl(bdev, mode, cmd, arg); 6549 } 6550 #endif /* CONFIG_COMPAT */ 6551 6552 static int md_open(struct block_device *bdev, fmode_t mode) 6553 { 6554 /* 6555 * Succeed if we can lock the mddev, which confirms that 6556 * it isn't being stopped right now. 6557 */ 6558 struct mddev *mddev = mddev_find(bdev->bd_dev); 6559 int err; 6560 6561 if (!mddev) 6562 return -ENODEV; 6563 6564 if (mddev->gendisk != bdev->bd_disk) { 6565 /* we are racing with mddev_put which is discarding this 6566 * bd_disk. 6567 */ 6568 mddev_put(mddev); 6569 /* Wait until bdev->bd_disk is definitely gone */ 6570 flush_workqueue(md_misc_wq); 6571 /* Then retry the open from the top */ 6572 return -ERESTARTSYS; 6573 } 6574 BUG_ON(mddev != bdev->bd_disk->private_data); 6575 6576 if ((err = mutex_lock_interruptible(&mddev->open_mutex))) 6577 goto out; 6578 6579 err = 0; 6580 atomic_inc(&mddev->openers); 6581 mutex_unlock(&mddev->open_mutex); 6582 6583 check_disk_change(bdev); 6584 out: 6585 return err; 6586 } 6587 6588 static int md_release(struct gendisk *disk, fmode_t mode) 6589 { 6590 struct mddev *mddev = disk->private_data; 6591 6592 BUG_ON(!mddev); 6593 atomic_dec(&mddev->openers); 6594 mddev_put(mddev); 6595 6596 return 0; 6597 } 6598 6599 static int md_media_changed(struct gendisk *disk) 6600 { 6601 struct mddev *mddev = disk->private_data; 6602 6603 return mddev->changed; 6604 } 6605 6606 static int md_revalidate(struct gendisk *disk) 6607 { 6608 struct mddev *mddev = disk->private_data; 6609 6610 mddev->changed = 0; 6611 return 0; 6612 } 6613 static const struct block_device_operations md_fops = 6614 { 6615 .owner = THIS_MODULE, 6616 .open = md_open, 6617 .release = md_release, 6618 .ioctl = md_ioctl, 6619 #ifdef CONFIG_COMPAT 6620 .compat_ioctl = md_compat_ioctl, 6621 #endif 6622 .getgeo = md_getgeo, 6623 .media_changed = md_media_changed, 6624 .revalidate_disk= md_revalidate, 6625 }; 6626 6627 static int md_thread(void * arg) 6628 { 6629 struct md_thread *thread = arg; 6630 6631 /* 6632 * md_thread is a 'system-thread', it's priority should be very 6633 * high. We avoid resource deadlocks individually in each 6634 * raid personality. (RAID5 does preallocation) We also use RR and 6635 * the very same RT priority as kswapd, thus we will never get 6636 * into a priority inversion deadlock. 6637 * 6638 * we definitely have to have equal or higher priority than 6639 * bdflush, otherwise bdflush will deadlock if there are too 6640 * many dirty RAID5 blocks. 6641 */ 6642 6643 allow_signal(SIGKILL); 6644 while (!kthread_should_stop()) { 6645 6646 /* We need to wait INTERRUPTIBLE so that 6647 * we don't add to the load-average. 6648 * That means we need to be sure no signals are 6649 * pending 6650 */ 6651 if (signal_pending(current)) 6652 flush_signals(current); 6653 6654 wait_event_interruptible_timeout 6655 (thread->wqueue, 6656 test_bit(THREAD_WAKEUP, &thread->flags) 6657 || kthread_should_stop(), 6658 thread->timeout); 6659 6660 clear_bit(THREAD_WAKEUP, &thread->flags); 6661 if (!kthread_should_stop()) 6662 thread->run(thread->mddev); 6663 } 6664 6665 return 0; 6666 } 6667 6668 void md_wakeup_thread(struct md_thread *thread) 6669 { 6670 if (thread) { 6671 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); 6672 set_bit(THREAD_WAKEUP, &thread->flags); 6673 wake_up(&thread->wqueue); 6674 } 6675 } 6676 6677 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev, 6678 const char *name) 6679 { 6680 struct md_thread *thread; 6681 6682 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); 6683 if (!thread) 6684 return NULL; 6685 6686 init_waitqueue_head(&thread->wqueue); 6687 6688 thread->run = run; 6689 thread->mddev = mddev; 6690 thread->timeout = MAX_SCHEDULE_TIMEOUT; 6691 thread->tsk = kthread_run(md_thread, thread, 6692 "%s_%s", 6693 mdname(thread->mddev), 6694 name); 6695 if (IS_ERR(thread->tsk)) { 6696 kfree(thread); 6697 return NULL; 6698 } 6699 return thread; 6700 } 6701 6702 void md_unregister_thread(struct md_thread **threadp) 6703 { 6704 struct md_thread *thread = *threadp; 6705 if (!thread) 6706 return; 6707 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); 6708 /* Locking ensures that mddev_unlock does not wake_up a 6709 * non-existent thread 6710 */ 6711 spin_lock(&pers_lock); 6712 *threadp = NULL; 6713 spin_unlock(&pers_lock); 6714 6715 kthread_stop(thread->tsk); 6716 kfree(thread); 6717 } 6718 6719 void md_error(struct mddev *mddev, struct md_rdev *rdev) 6720 { 6721 if (!mddev) { 6722 MD_BUG(); 6723 return; 6724 } 6725 6726 if (!rdev || test_bit(Faulty, &rdev->flags)) 6727 return; 6728 6729 if (!mddev->pers || !mddev->pers->error_handler) 6730 return; 6731 mddev->pers->error_handler(mddev,rdev); 6732 if (mddev->degraded) 6733 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6734 sysfs_notify_dirent_safe(rdev->sysfs_state); 6735 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6736 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6737 md_wakeup_thread(mddev->thread); 6738 if (mddev->event_work.func) 6739 queue_work(md_misc_wq, &mddev->event_work); 6740 md_new_event_inintr(mddev); 6741 } 6742 6743 /* seq_file implementation /proc/mdstat */ 6744 6745 static void status_unused(struct seq_file *seq) 6746 { 6747 int i = 0; 6748 struct md_rdev *rdev; 6749 6750 seq_printf(seq, "unused devices: "); 6751 6752 list_for_each_entry(rdev, &pending_raid_disks, same_set) { 6753 char b[BDEVNAME_SIZE]; 6754 i++; 6755 seq_printf(seq, "%s ", 6756 bdevname(rdev->bdev,b)); 6757 } 6758 if (!i) 6759 seq_printf(seq, "<none>"); 6760 6761 seq_printf(seq, "\n"); 6762 } 6763 6764 6765 static void status_resync(struct seq_file *seq, struct mddev * mddev) 6766 { 6767 sector_t max_sectors, resync, res; 6768 unsigned long dt, db; 6769 sector_t rt; 6770 int scale; 6771 unsigned int per_milli; 6772 6773 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); 6774 6775 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || 6776 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6777 max_sectors = mddev->resync_max_sectors; 6778 else 6779 max_sectors = mddev->dev_sectors; 6780 6781 /* 6782 * Should not happen. 6783 */ 6784 if (!max_sectors) { 6785 MD_BUG(); 6786 return; 6787 } 6788 /* Pick 'scale' such that (resync>>scale)*1000 will fit 6789 * in a sector_t, and (max_sectors>>scale) will fit in a 6790 * u32, as those are the requirements for sector_div. 6791 * Thus 'scale' must be at least 10 6792 */ 6793 scale = 10; 6794 if (sizeof(sector_t) > sizeof(unsigned long)) { 6795 while ( max_sectors/2 > (1ULL<<(scale+32))) 6796 scale++; 6797 } 6798 res = (resync>>scale)*1000; 6799 sector_div(res, (u32)((max_sectors>>scale)+1)); 6800 6801 per_milli = res; 6802 { 6803 int i, x = per_milli/50, y = 20-x; 6804 seq_printf(seq, "["); 6805 for (i = 0; i < x; i++) 6806 seq_printf(seq, "="); 6807 seq_printf(seq, ">"); 6808 for (i = 0; i < y; i++) 6809 seq_printf(seq, "."); 6810 seq_printf(seq, "] "); 6811 } 6812 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", 6813 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? 6814 "reshape" : 6815 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? 6816 "check" : 6817 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 6818 "resync" : "recovery"))), 6819 per_milli/10, per_milli % 10, 6820 (unsigned long long) resync/2, 6821 (unsigned long long) max_sectors/2); 6822 6823 /* 6824 * dt: time from mark until now 6825 * db: blocks written from mark until now 6826 * rt: remaining time 6827 * 6828 * rt is a sector_t, so could be 32bit or 64bit. 6829 * So we divide before multiply in case it is 32bit and close 6830 * to the limit. 6831 * We scale the divisor (db) by 32 to avoid losing precision 6832 * near the end of resync when the number of remaining sectors 6833 * is close to 'db'. 6834 * We then divide rt by 32 after multiplying by db to compensate. 6835 * The '+1' avoids division by zero if db is very small. 6836 */ 6837 dt = ((jiffies - mddev->resync_mark) / HZ); 6838 if (!dt) dt++; 6839 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) 6840 - mddev->resync_mark_cnt; 6841 6842 rt = max_sectors - resync; /* number of remaining sectors */ 6843 sector_div(rt, db/32+1); 6844 rt *= dt; 6845 rt >>= 5; 6846 6847 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, 6848 ((unsigned long)rt % 60)/6); 6849 6850 seq_printf(seq, " speed=%ldK/sec", db/2/dt); 6851 } 6852 6853 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 6854 { 6855 struct list_head *tmp; 6856 loff_t l = *pos; 6857 struct mddev *mddev; 6858 6859 if (l >= 0x10000) 6860 return NULL; 6861 if (!l--) 6862 /* header */ 6863 return (void*)1; 6864 6865 spin_lock(&all_mddevs_lock); 6866 list_for_each(tmp,&all_mddevs) 6867 if (!l--) { 6868 mddev = list_entry(tmp, struct mddev, all_mddevs); 6869 mddev_get(mddev); 6870 spin_unlock(&all_mddevs_lock); 6871 return mddev; 6872 } 6873 spin_unlock(&all_mddevs_lock); 6874 if (!l--) 6875 return (void*)2;/* tail */ 6876 return NULL; 6877 } 6878 6879 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 6880 { 6881 struct list_head *tmp; 6882 struct mddev *next_mddev, *mddev = v; 6883 6884 ++*pos; 6885 if (v == (void*)2) 6886 return NULL; 6887 6888 spin_lock(&all_mddevs_lock); 6889 if (v == (void*)1) 6890 tmp = all_mddevs.next; 6891 else 6892 tmp = mddev->all_mddevs.next; 6893 if (tmp != &all_mddevs) 6894 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); 6895 else { 6896 next_mddev = (void*)2; 6897 *pos = 0x10000; 6898 } 6899 spin_unlock(&all_mddevs_lock); 6900 6901 if (v != (void*)1) 6902 mddev_put(mddev); 6903 return next_mddev; 6904 6905 } 6906 6907 static void md_seq_stop(struct seq_file *seq, void *v) 6908 { 6909 struct mddev *mddev = v; 6910 6911 if (mddev && v != (void*)1 && v != (void*)2) 6912 mddev_put(mddev); 6913 } 6914 6915 static int md_seq_show(struct seq_file *seq, void *v) 6916 { 6917 struct mddev *mddev = v; 6918 sector_t sectors; 6919 struct md_rdev *rdev; 6920 6921 if (v == (void*)1) { 6922 struct md_personality *pers; 6923 seq_printf(seq, "Personalities : "); 6924 spin_lock(&pers_lock); 6925 list_for_each_entry(pers, &pers_list, list) 6926 seq_printf(seq, "[%s] ", pers->name); 6927 6928 spin_unlock(&pers_lock); 6929 seq_printf(seq, "\n"); 6930 seq->poll_event = atomic_read(&md_event_count); 6931 return 0; 6932 } 6933 if (v == (void*)2) { 6934 status_unused(seq); 6935 return 0; 6936 } 6937 6938 if (mddev_lock(mddev) < 0) 6939 return -EINTR; 6940 6941 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 6942 seq_printf(seq, "%s : %sactive", mdname(mddev), 6943 mddev->pers ? "" : "in"); 6944 if (mddev->pers) { 6945 if (mddev->ro==1) 6946 seq_printf(seq, " (read-only)"); 6947 if (mddev->ro==2) 6948 seq_printf(seq, " (auto-read-only)"); 6949 seq_printf(seq, " %s", mddev->pers->name); 6950 } 6951 6952 sectors = 0; 6953 rdev_for_each(rdev, mddev) { 6954 char b[BDEVNAME_SIZE]; 6955 seq_printf(seq, " %s[%d]", 6956 bdevname(rdev->bdev,b), rdev->desc_nr); 6957 if (test_bit(WriteMostly, &rdev->flags)) 6958 seq_printf(seq, "(W)"); 6959 if (test_bit(Faulty, &rdev->flags)) { 6960 seq_printf(seq, "(F)"); 6961 continue; 6962 } 6963 if (rdev->raid_disk < 0) 6964 seq_printf(seq, "(S)"); /* spare */ 6965 if (test_bit(Replacement, &rdev->flags)) 6966 seq_printf(seq, "(R)"); 6967 sectors += rdev->sectors; 6968 } 6969 6970 if (!list_empty(&mddev->disks)) { 6971 if (mddev->pers) 6972 seq_printf(seq, "\n %llu blocks", 6973 (unsigned long long) 6974 mddev->array_sectors / 2); 6975 else 6976 seq_printf(seq, "\n %llu blocks", 6977 (unsigned long long)sectors / 2); 6978 } 6979 if (mddev->persistent) { 6980 if (mddev->major_version != 0 || 6981 mddev->minor_version != 90) { 6982 seq_printf(seq," super %d.%d", 6983 mddev->major_version, 6984 mddev->minor_version); 6985 } 6986 } else if (mddev->external) 6987 seq_printf(seq, " super external:%s", 6988 mddev->metadata_type); 6989 else 6990 seq_printf(seq, " super non-persistent"); 6991 6992 if (mddev->pers) { 6993 mddev->pers->status(seq, mddev); 6994 seq_printf(seq, "\n "); 6995 if (mddev->pers->sync_request) { 6996 if (mddev->curr_resync > 2) { 6997 status_resync(seq, mddev); 6998 seq_printf(seq, "\n "); 6999 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 7000 seq_printf(seq, "\tresync=DELAYED\n "); 7001 else if (mddev->recovery_cp < MaxSector) 7002 seq_printf(seq, "\tresync=PENDING\n "); 7003 } 7004 } else 7005 seq_printf(seq, "\n "); 7006 7007 bitmap_status(seq, mddev->bitmap); 7008 7009 seq_printf(seq, "\n"); 7010 } 7011 mddev_unlock(mddev); 7012 7013 return 0; 7014 } 7015 7016 static const struct seq_operations md_seq_ops = { 7017 .start = md_seq_start, 7018 .next = md_seq_next, 7019 .stop = md_seq_stop, 7020 .show = md_seq_show, 7021 }; 7022 7023 static int md_seq_open(struct inode *inode, struct file *file) 7024 { 7025 struct seq_file *seq; 7026 int error; 7027 7028 error = seq_open(file, &md_seq_ops); 7029 if (error) 7030 return error; 7031 7032 seq = file->private_data; 7033 seq->poll_event = atomic_read(&md_event_count); 7034 return error; 7035 } 7036 7037 static unsigned int mdstat_poll(struct file *filp, poll_table *wait) 7038 { 7039 struct seq_file *seq = filp->private_data; 7040 int mask; 7041 7042 poll_wait(filp, &md_event_waiters, wait); 7043 7044 /* always allow read */ 7045 mask = POLLIN | POLLRDNORM; 7046 7047 if (seq->poll_event != atomic_read(&md_event_count)) 7048 mask |= POLLERR | POLLPRI; 7049 return mask; 7050 } 7051 7052 static const struct file_operations md_seq_fops = { 7053 .owner = THIS_MODULE, 7054 .open = md_seq_open, 7055 .read = seq_read, 7056 .llseek = seq_lseek, 7057 .release = seq_release_private, 7058 .poll = mdstat_poll, 7059 }; 7060 7061 int register_md_personality(struct md_personality *p) 7062 { 7063 spin_lock(&pers_lock); 7064 list_add_tail(&p->list, &pers_list); 7065 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); 7066 spin_unlock(&pers_lock); 7067 return 0; 7068 } 7069 7070 int unregister_md_personality(struct md_personality *p) 7071 { 7072 printk(KERN_INFO "md: %s personality unregistered\n", p->name); 7073 spin_lock(&pers_lock); 7074 list_del_init(&p->list); 7075 spin_unlock(&pers_lock); 7076 return 0; 7077 } 7078 7079 static int is_mddev_idle(struct mddev *mddev, int init) 7080 { 7081 struct md_rdev * rdev; 7082 int idle; 7083 int curr_events; 7084 7085 idle = 1; 7086 rcu_read_lock(); 7087 rdev_for_each_rcu(rdev, mddev) { 7088 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 7089 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 7090 (int)part_stat_read(&disk->part0, sectors[1]) - 7091 atomic_read(&disk->sync_io); 7092 /* sync IO will cause sync_io to increase before the disk_stats 7093 * as sync_io is counted when a request starts, and 7094 * disk_stats is counted when it completes. 7095 * So resync activity will cause curr_events to be smaller than 7096 * when there was no such activity. 7097 * non-sync IO will cause disk_stat to increase without 7098 * increasing sync_io so curr_events will (eventually) 7099 * be larger than it was before. Once it becomes 7100 * substantially larger, the test below will cause 7101 * the array to appear non-idle, and resync will slow 7102 * down. 7103 * If there is a lot of outstanding resync activity when 7104 * we set last_event to curr_events, then all that activity 7105 * completing might cause the array to appear non-idle 7106 * and resync will be slowed down even though there might 7107 * not have been non-resync activity. This will only 7108 * happen once though. 'last_events' will soon reflect 7109 * the state where there is little or no outstanding 7110 * resync requests, and further resync activity will 7111 * always make curr_events less than last_events. 7112 * 7113 */ 7114 if (init || curr_events - rdev->last_events > 64) { 7115 rdev->last_events = curr_events; 7116 idle = 0; 7117 } 7118 } 7119 rcu_read_unlock(); 7120 return idle; 7121 } 7122 7123 void md_done_sync(struct mddev *mddev, int blocks, int ok) 7124 { 7125 /* another "blocks" (512byte) blocks have been synced */ 7126 atomic_sub(blocks, &mddev->recovery_active); 7127 wake_up(&mddev->recovery_wait); 7128 if (!ok) { 7129 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7130 md_wakeup_thread(mddev->thread); 7131 // stop recovery, signal do_sync .... 7132 } 7133 } 7134 7135 7136 /* md_write_start(mddev, bi) 7137 * If we need to update some array metadata (e.g. 'active' flag 7138 * in superblock) before writing, schedule a superblock update 7139 * and wait for it to complete. 7140 */ 7141 void md_write_start(struct mddev *mddev, struct bio *bi) 7142 { 7143 int did_change = 0; 7144 if (bio_data_dir(bi) != WRITE) 7145 return; 7146 7147 BUG_ON(mddev->ro == 1); 7148 if (mddev->ro == 2) { 7149 /* need to switch to read/write */ 7150 mddev->ro = 0; 7151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7152 md_wakeup_thread(mddev->thread); 7153 md_wakeup_thread(mddev->sync_thread); 7154 did_change = 1; 7155 } 7156 atomic_inc(&mddev->writes_pending); 7157 if (mddev->safemode == 1) 7158 mddev->safemode = 0; 7159 if (mddev->in_sync) { 7160 spin_lock_irq(&mddev->write_lock); 7161 if (mddev->in_sync) { 7162 mddev->in_sync = 0; 7163 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7164 set_bit(MD_CHANGE_PENDING, &mddev->flags); 7165 md_wakeup_thread(mddev->thread); 7166 did_change = 1; 7167 } 7168 spin_unlock_irq(&mddev->write_lock); 7169 } 7170 if (did_change) 7171 sysfs_notify_dirent_safe(mddev->sysfs_state); 7172 wait_event(mddev->sb_wait, 7173 !test_bit(MD_CHANGE_PENDING, &mddev->flags)); 7174 } 7175 7176 void md_write_end(struct mddev *mddev) 7177 { 7178 if (atomic_dec_and_test(&mddev->writes_pending)) { 7179 if (mddev->safemode == 2) 7180 md_wakeup_thread(mddev->thread); 7181 else if (mddev->safemode_delay) 7182 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 7183 } 7184 } 7185 7186 /* md_allow_write(mddev) 7187 * Calling this ensures that the array is marked 'active' so that writes 7188 * may proceed without blocking. It is important to call this before 7189 * attempting a GFP_KERNEL allocation while holding the mddev lock. 7190 * Must be called with mddev_lock held. 7191 * 7192 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock 7193 * is dropped, so return -EAGAIN after notifying userspace. 7194 */ 7195 int md_allow_write(struct mddev *mddev) 7196 { 7197 if (!mddev->pers) 7198 return 0; 7199 if (mddev->ro) 7200 return 0; 7201 if (!mddev->pers->sync_request) 7202 return 0; 7203 7204 spin_lock_irq(&mddev->write_lock); 7205 if (mddev->in_sync) { 7206 mddev->in_sync = 0; 7207 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7208 set_bit(MD_CHANGE_PENDING, &mddev->flags); 7209 if (mddev->safemode_delay && 7210 mddev->safemode == 0) 7211 mddev->safemode = 1; 7212 spin_unlock_irq(&mddev->write_lock); 7213 md_update_sb(mddev, 0); 7214 sysfs_notify_dirent_safe(mddev->sysfs_state); 7215 } else 7216 spin_unlock_irq(&mddev->write_lock); 7217 7218 if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 7219 return -EAGAIN; 7220 else 7221 return 0; 7222 } 7223 EXPORT_SYMBOL_GPL(md_allow_write); 7224 7225 #define SYNC_MARKS 10 7226 #define SYNC_MARK_STEP (3*HZ) 7227 void md_do_sync(struct mddev *mddev) 7228 { 7229 struct mddev *mddev2; 7230 unsigned int currspeed = 0, 7231 window; 7232 sector_t max_sectors,j, io_sectors; 7233 unsigned long mark[SYNC_MARKS]; 7234 sector_t mark_cnt[SYNC_MARKS]; 7235 int last_mark,m; 7236 struct list_head *tmp; 7237 sector_t last_check; 7238 int skipped = 0; 7239 struct md_rdev *rdev; 7240 char *desc; 7241 struct blk_plug plug; 7242 7243 /* just incase thread restarts... */ 7244 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 7245 return; 7246 if (mddev->ro) /* never try to sync a read-only array */ 7247 return; 7248 7249 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7250 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 7251 desc = "data-check"; 7252 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7253 desc = "requested-resync"; 7254 else 7255 desc = "resync"; 7256 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 7257 desc = "reshape"; 7258 else 7259 desc = "recovery"; 7260 7261 /* we overload curr_resync somewhat here. 7262 * 0 == not engaged in resync at all 7263 * 2 == checking that there is no conflict with another sync 7264 * 1 == like 2, but have yielded to allow conflicting resync to 7265 * commense 7266 * other == active in resync - this many blocks 7267 * 7268 * Before starting a resync we must have set curr_resync to 7269 * 2, and then checked that every "conflicting" array has curr_resync 7270 * less than ours. When we find one that is the same or higher 7271 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 7272 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 7273 * This will mean we have to start checking from the beginning again. 7274 * 7275 */ 7276 7277 do { 7278 mddev->curr_resync = 2; 7279 7280 try_again: 7281 if (kthread_should_stop()) 7282 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7283 7284 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7285 goto skip; 7286 for_each_mddev(mddev2, tmp) { 7287 if (mddev2 == mddev) 7288 continue; 7289 if (!mddev->parallel_resync 7290 && mddev2->curr_resync 7291 && match_mddev_units(mddev, mddev2)) { 7292 DEFINE_WAIT(wq); 7293 if (mddev < mddev2 && mddev->curr_resync == 2) { 7294 /* arbitrarily yield */ 7295 mddev->curr_resync = 1; 7296 wake_up(&resync_wait); 7297 } 7298 if (mddev > mddev2 && mddev->curr_resync == 1) 7299 /* no need to wait here, we can wait the next 7300 * time 'round when curr_resync == 2 7301 */ 7302 continue; 7303 /* We need to wait 'interruptible' so as not to 7304 * contribute to the load average, and not to 7305 * be caught by 'softlockup' 7306 */ 7307 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 7308 if (!kthread_should_stop() && 7309 mddev2->curr_resync >= mddev->curr_resync) { 7310 printk(KERN_INFO "md: delaying %s of %s" 7311 " until %s has finished (they" 7312 " share one or more physical units)\n", 7313 desc, mdname(mddev), mdname(mddev2)); 7314 mddev_put(mddev2); 7315 if (signal_pending(current)) 7316 flush_signals(current); 7317 schedule(); 7318 finish_wait(&resync_wait, &wq); 7319 goto try_again; 7320 } 7321 finish_wait(&resync_wait, &wq); 7322 } 7323 } 7324 } while (mddev->curr_resync < 2); 7325 7326 j = 0; 7327 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7328 /* resync follows the size requested by the personality, 7329 * which defaults to physical size, but can be virtual size 7330 */ 7331 max_sectors = mddev->resync_max_sectors; 7332 mddev->resync_mismatches = 0; 7333 /* we don't use the checkpoint if there's a bitmap */ 7334 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7335 j = mddev->resync_min; 7336 else if (!mddev->bitmap) 7337 j = mddev->recovery_cp; 7338 7339 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 7340 max_sectors = mddev->resync_max_sectors; 7341 else { 7342 /* recovery follows the physical size of devices */ 7343 max_sectors = mddev->dev_sectors; 7344 j = MaxSector; 7345 rcu_read_lock(); 7346 rdev_for_each_rcu(rdev, mddev) 7347 if (rdev->raid_disk >= 0 && 7348 !test_bit(Faulty, &rdev->flags) && 7349 !test_bit(In_sync, &rdev->flags) && 7350 rdev->recovery_offset < j) 7351 j = rdev->recovery_offset; 7352 rcu_read_unlock(); 7353 } 7354 7355 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); 7356 printk(KERN_INFO "md: minimum _guaranteed_ speed:" 7357 " %d KB/sec/disk.\n", speed_min(mddev)); 7358 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 7359 "(but not more than %d KB/sec) for %s.\n", 7360 speed_max(mddev), desc); 7361 7362 is_mddev_idle(mddev, 1); /* this initializes IO event counters */ 7363 7364 io_sectors = 0; 7365 for (m = 0; m < SYNC_MARKS; m++) { 7366 mark[m] = jiffies; 7367 mark_cnt[m] = io_sectors; 7368 } 7369 last_mark = 0; 7370 mddev->resync_mark = mark[last_mark]; 7371 mddev->resync_mark_cnt = mark_cnt[last_mark]; 7372 7373 /* 7374 * Tune reconstruction: 7375 */ 7376 window = 32*(PAGE_SIZE/512); 7377 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", 7378 window/2, (unsigned long long)max_sectors/2); 7379 7380 atomic_set(&mddev->recovery_active, 0); 7381 last_check = 0; 7382 7383 if (j>2) { 7384 printk(KERN_INFO 7385 "md: resuming %s of %s from checkpoint.\n", 7386 desc, mdname(mddev)); 7387 mddev->curr_resync = j; 7388 } 7389 mddev->curr_resync_completed = j; 7390 7391 blk_start_plug(&plug); 7392 while (j < max_sectors) { 7393 sector_t sectors; 7394 7395 skipped = 0; 7396 7397 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7398 ((mddev->curr_resync > mddev->curr_resync_completed && 7399 (mddev->curr_resync - mddev->curr_resync_completed) 7400 > (max_sectors >> 4)) || 7401 (j - mddev->curr_resync_completed)*2 7402 >= mddev->resync_max - mddev->curr_resync_completed 7403 )) { 7404 /* time to update curr_resync_completed */ 7405 wait_event(mddev->recovery_wait, 7406 atomic_read(&mddev->recovery_active) == 0); 7407 mddev->curr_resync_completed = j; 7408 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7409 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 7410 } 7411 7412 while (j >= mddev->resync_max && !kthread_should_stop()) { 7413 /* As this condition is controlled by user-space, 7414 * we can block indefinitely, so use '_interruptible' 7415 * to avoid triggering warnings. 7416 */ 7417 flush_signals(current); /* just in case */ 7418 wait_event_interruptible(mddev->recovery_wait, 7419 mddev->resync_max > j 7420 || kthread_should_stop()); 7421 } 7422 7423 if (kthread_should_stop()) 7424 goto interrupted; 7425 7426 sectors = mddev->pers->sync_request(mddev, j, &skipped, 7427 currspeed < speed_min(mddev)); 7428 if (sectors == 0) { 7429 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7430 goto out; 7431 } 7432 7433 if (!skipped) { /* actual IO requested */ 7434 io_sectors += sectors; 7435 atomic_add(sectors, &mddev->recovery_active); 7436 } 7437 7438 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7439 break; 7440 7441 j += sectors; 7442 if (j>1) mddev->curr_resync = j; 7443 mddev->curr_mark_cnt = io_sectors; 7444 if (last_check == 0) 7445 /* this is the earliest that rebuild will be 7446 * visible in /proc/mdstat 7447 */ 7448 md_new_event(mddev); 7449 7450 if (last_check + window > io_sectors || j == max_sectors) 7451 continue; 7452 7453 last_check = io_sectors; 7454 repeat: 7455 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 7456 /* step marks */ 7457 int next = (last_mark+1) % SYNC_MARKS; 7458 7459 mddev->resync_mark = mark[next]; 7460 mddev->resync_mark_cnt = mark_cnt[next]; 7461 mark[next] = jiffies; 7462 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 7463 last_mark = next; 7464 } 7465 7466 7467 if (kthread_should_stop()) 7468 goto interrupted; 7469 7470 7471 /* 7472 * this loop exits only if either when we are slower than 7473 * the 'hard' speed limit, or the system was IO-idle for 7474 * a jiffy. 7475 * the system might be non-idle CPU-wise, but we only care 7476 * about not overloading the IO subsystem. (things like an 7477 * e2fsck being done on the RAID array should execute fast) 7478 */ 7479 cond_resched(); 7480 7481 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 7482 /((jiffies-mddev->resync_mark)/HZ +1) +1; 7483 7484 if (currspeed > speed_min(mddev)) { 7485 if ((currspeed > speed_max(mddev)) || 7486 !is_mddev_idle(mddev, 0)) { 7487 msleep(500); 7488 goto repeat; 7489 } 7490 } 7491 } 7492 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); 7493 /* 7494 * this also signals 'finished resyncing' to md_stop 7495 */ 7496 out: 7497 blk_finish_plug(&plug); 7498 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 7499 7500 /* tell personality that we are finished */ 7501 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 7502 7503 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && 7504 mddev->curr_resync > 2) { 7505 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7506 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 7507 if (mddev->curr_resync >= mddev->recovery_cp) { 7508 printk(KERN_INFO 7509 "md: checkpointing %s of %s.\n", 7510 desc, mdname(mddev)); 7511 mddev->recovery_cp = 7512 mddev->curr_resync_completed; 7513 } 7514 } else 7515 mddev->recovery_cp = MaxSector; 7516 } else { 7517 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7518 mddev->curr_resync = MaxSector; 7519 rcu_read_lock(); 7520 rdev_for_each_rcu(rdev, mddev) 7521 if (rdev->raid_disk >= 0 && 7522 mddev->delta_disks >= 0 && 7523 !test_bit(Faulty, &rdev->flags) && 7524 !test_bit(In_sync, &rdev->flags) && 7525 rdev->recovery_offset < mddev->curr_resync) 7526 rdev->recovery_offset = mddev->curr_resync; 7527 rcu_read_unlock(); 7528 } 7529 } 7530 skip: 7531 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7532 7533 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 7534 /* We completed so min/max setting can be forgotten if used. */ 7535 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7536 mddev->resync_min = 0; 7537 mddev->resync_max = MaxSector; 7538 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7539 mddev->resync_min = mddev->curr_resync_completed; 7540 mddev->curr_resync = 0; 7541 wake_up(&resync_wait); 7542 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 7543 md_wakeup_thread(mddev->thread); 7544 return; 7545 7546 interrupted: 7547 /* 7548 * got a signal, exit. 7549 */ 7550 printk(KERN_INFO 7551 "md: md_do_sync() got signal ... exiting\n"); 7552 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7553 goto out; 7554 7555 } 7556 EXPORT_SYMBOL_GPL(md_do_sync); 7557 7558 static int remove_and_add_spares(struct mddev *mddev) 7559 { 7560 struct md_rdev *rdev; 7561 int spares = 0; 7562 int removed = 0; 7563 7564 mddev->curr_resync_completed = 0; 7565 7566 rdev_for_each(rdev, mddev) 7567 if (rdev->raid_disk >= 0 && 7568 !test_bit(Blocked, &rdev->flags) && 7569 (test_bit(Faulty, &rdev->flags) || 7570 ! test_bit(In_sync, &rdev->flags)) && 7571 atomic_read(&rdev->nr_pending)==0) { 7572 if (mddev->pers->hot_remove_disk( 7573 mddev, rdev) == 0) { 7574 sysfs_unlink_rdev(mddev, rdev); 7575 rdev->raid_disk = -1; 7576 removed++; 7577 } 7578 } 7579 if (removed) 7580 sysfs_notify(&mddev->kobj, NULL, 7581 "degraded"); 7582 7583 7584 rdev_for_each(rdev, mddev) { 7585 if (rdev->raid_disk >= 0 && 7586 !test_bit(In_sync, &rdev->flags) && 7587 !test_bit(Faulty, &rdev->flags)) 7588 spares++; 7589 if (rdev->raid_disk < 0 7590 && !test_bit(Faulty, &rdev->flags)) { 7591 rdev->recovery_offset = 0; 7592 if (mddev->pers-> 7593 hot_add_disk(mddev, rdev) == 0) { 7594 if (sysfs_link_rdev(mddev, rdev)) 7595 /* failure here is OK */; 7596 spares++; 7597 md_new_event(mddev); 7598 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7599 } 7600 } 7601 } 7602 return spares; 7603 } 7604 7605 static void reap_sync_thread(struct mddev *mddev) 7606 { 7607 struct md_rdev *rdev; 7608 7609 /* resync has finished, collect result */ 7610 md_unregister_thread(&mddev->sync_thread); 7611 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && 7612 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 7613 /* success...*/ 7614 /* activate any spares */ 7615 if (mddev->pers->spare_active(mddev)) 7616 sysfs_notify(&mddev->kobj, NULL, 7617 "degraded"); 7618 } 7619 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7620 mddev->pers->finish_reshape) 7621 mddev->pers->finish_reshape(mddev); 7622 7623 /* If array is no-longer degraded, then any saved_raid_disk 7624 * information must be scrapped. Also if any device is now 7625 * In_sync we must scrape the saved_raid_disk for that device 7626 * do the superblock for an incrementally recovered device 7627 * written out. 7628 */ 7629 rdev_for_each(rdev, mddev) 7630 if (!mddev->degraded || 7631 test_bit(In_sync, &rdev->flags)) 7632 rdev->saved_raid_disk = -1; 7633 7634 md_update_sb(mddev, 1); 7635 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7636 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7637 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7638 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7639 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7640 /* flag recovery needed just to double check */ 7641 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7642 sysfs_notify_dirent_safe(mddev->sysfs_action); 7643 md_new_event(mddev); 7644 if (mddev->event_work.func) 7645 queue_work(md_misc_wq, &mddev->event_work); 7646 } 7647 7648 /* 7649 * This routine is regularly called by all per-raid-array threads to 7650 * deal with generic issues like resync and super-block update. 7651 * Raid personalities that don't have a thread (linear/raid0) do not 7652 * need this as they never do any recovery or update the superblock. 7653 * 7654 * It does not do any resync itself, but rather "forks" off other threads 7655 * to do that as needed. 7656 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 7657 * "->recovery" and create a thread at ->sync_thread. 7658 * When the thread finishes it sets MD_RECOVERY_DONE 7659 * and wakeups up this thread which will reap the thread and finish up. 7660 * This thread also removes any faulty devices (with nr_pending == 0). 7661 * 7662 * The overall approach is: 7663 * 1/ if the superblock needs updating, update it. 7664 * 2/ If a recovery thread is running, don't do anything else. 7665 * 3/ If recovery has finished, clean up, possibly marking spares active. 7666 * 4/ If there are any faulty devices, remove them. 7667 * 5/ If array is degraded, try to add spares devices 7668 * 6/ If array has spares or is not in-sync, start a resync thread. 7669 */ 7670 void md_check_recovery(struct mddev *mddev) 7671 { 7672 if (mddev->suspended) 7673 return; 7674 7675 if (mddev->bitmap) 7676 bitmap_daemon_work(mddev); 7677 7678 if (signal_pending(current)) { 7679 if (mddev->pers->sync_request && !mddev->external) { 7680 printk(KERN_INFO "md: %s in immediate safe mode\n", 7681 mdname(mddev)); 7682 mddev->safemode = 2; 7683 } 7684 flush_signals(current); 7685 } 7686 7687 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 7688 return; 7689 if ( ! ( 7690 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || 7691 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 7692 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 7693 (mddev->external == 0 && mddev->safemode == 1) || 7694 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 7695 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 7696 )) 7697 return; 7698 7699 if (mddev_trylock(mddev)) { 7700 int spares = 0; 7701 7702 if (mddev->ro) { 7703 /* Only thing we do on a ro array is remove 7704 * failed devices. 7705 */ 7706 struct md_rdev *rdev; 7707 rdev_for_each(rdev, mddev) 7708 if (rdev->raid_disk >= 0 && 7709 !test_bit(Blocked, &rdev->flags) && 7710 test_bit(Faulty, &rdev->flags) && 7711 atomic_read(&rdev->nr_pending)==0) { 7712 if (mddev->pers->hot_remove_disk( 7713 mddev, rdev) == 0) { 7714 sysfs_unlink_rdev(mddev, rdev); 7715 rdev->raid_disk = -1; 7716 } 7717 } 7718 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7719 goto unlock; 7720 } 7721 7722 if (!mddev->external) { 7723 int did_change = 0; 7724 spin_lock_irq(&mddev->write_lock); 7725 if (mddev->safemode && 7726 !atomic_read(&mddev->writes_pending) && 7727 !mddev->in_sync && 7728 mddev->recovery_cp == MaxSector) { 7729 mddev->in_sync = 1; 7730 did_change = 1; 7731 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7732 } 7733 if (mddev->safemode == 1) 7734 mddev->safemode = 0; 7735 spin_unlock_irq(&mddev->write_lock); 7736 if (did_change) 7737 sysfs_notify_dirent_safe(mddev->sysfs_state); 7738 } 7739 7740 if (mddev->flags) 7741 md_update_sb(mddev, 0); 7742 7743 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 7744 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 7745 /* resync/recovery still happening */ 7746 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7747 goto unlock; 7748 } 7749 if (mddev->sync_thread) { 7750 reap_sync_thread(mddev); 7751 goto unlock; 7752 } 7753 /* Set RUNNING before clearing NEEDED to avoid 7754 * any transients in the value of "sync_action". 7755 */ 7756 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7757 /* Clear some bits that don't mean anything, but 7758 * might be left set 7759 */ 7760 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 7761 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 7762 7763 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 7764 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 7765 goto unlock; 7766 /* no recovery is running. 7767 * remove any failed drives, then 7768 * add spares if possible. 7769 * Spare are also removed and re-added, to allow 7770 * the personality to fail the re-add. 7771 */ 7772 7773 if (mddev->reshape_position != MaxSector) { 7774 if (mddev->pers->check_reshape == NULL || 7775 mddev->pers->check_reshape(mddev) != 0) 7776 /* Cannot proceed */ 7777 goto unlock; 7778 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7779 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7780 } else if ((spares = remove_and_add_spares(mddev))) { 7781 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7782 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7783 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7784 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7785 } else if (mddev->recovery_cp < MaxSector) { 7786 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7787 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7788 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 7789 /* nothing to be done ... */ 7790 goto unlock; 7791 7792 if (mddev->pers->sync_request) { 7793 if (spares) { 7794 /* We are adding a device or devices to an array 7795 * which has the bitmap stored on all devices. 7796 * So make sure all bitmap pages get written 7797 */ 7798 bitmap_write_all(mddev->bitmap); 7799 } 7800 mddev->sync_thread = md_register_thread(md_do_sync, 7801 mddev, 7802 "resync"); 7803 if (!mddev->sync_thread) { 7804 printk(KERN_ERR "%s: could not start resync" 7805 " thread...\n", 7806 mdname(mddev)); 7807 /* leave the spares where they are, it shouldn't hurt */ 7808 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7809 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7810 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7811 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7812 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7813 } else 7814 md_wakeup_thread(mddev->sync_thread); 7815 sysfs_notify_dirent_safe(mddev->sysfs_action); 7816 md_new_event(mddev); 7817 } 7818 unlock: 7819 if (!mddev->sync_thread) { 7820 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7821 if (test_and_clear_bit(MD_RECOVERY_RECOVER, 7822 &mddev->recovery)) 7823 if (mddev->sysfs_action) 7824 sysfs_notify_dirent_safe(mddev->sysfs_action); 7825 } 7826 mddev_unlock(mddev); 7827 } 7828 } 7829 7830 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) 7831 { 7832 sysfs_notify_dirent_safe(rdev->sysfs_state); 7833 wait_event_timeout(rdev->blocked_wait, 7834 !test_bit(Blocked, &rdev->flags) && 7835 !test_bit(BlockedBadBlocks, &rdev->flags), 7836 msecs_to_jiffies(5000)); 7837 rdev_dec_pending(rdev, mddev); 7838 } 7839 EXPORT_SYMBOL(md_wait_for_blocked_rdev); 7840 7841 void md_finish_reshape(struct mddev *mddev) 7842 { 7843 /* called be personality module when reshape completes. */ 7844 struct md_rdev *rdev; 7845 7846 rdev_for_each(rdev, mddev) { 7847 if (rdev->data_offset > rdev->new_data_offset) 7848 rdev->sectors += rdev->data_offset - rdev->new_data_offset; 7849 else 7850 rdev->sectors -= rdev->new_data_offset - rdev->data_offset; 7851 rdev->data_offset = rdev->new_data_offset; 7852 } 7853 } 7854 EXPORT_SYMBOL(md_finish_reshape); 7855 7856 /* Bad block management. 7857 * We can record which blocks on each device are 'bad' and so just 7858 * fail those blocks, or that stripe, rather than the whole device. 7859 * Entries in the bad-block table are 64bits wide. This comprises: 7860 * Length of bad-range, in sectors: 0-511 for lengths 1-512 7861 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) 7862 * A 'shift' can be set so that larger blocks are tracked and 7863 * consequently larger devices can be covered. 7864 * 'Acknowledged' flag - 1 bit. - the most significant bit. 7865 * 7866 * Locking of the bad-block table uses a seqlock so md_is_badblock 7867 * might need to retry if it is very unlucky. 7868 * We will sometimes want to check for bad blocks in a bi_end_io function, 7869 * so we use the write_seqlock_irq variant. 7870 * 7871 * When looking for a bad block we specify a range and want to 7872 * know if any block in the range is bad. So we binary-search 7873 * to the last range that starts at-or-before the given endpoint, 7874 * (or "before the sector after the target range") 7875 * then see if it ends after the given start. 7876 * We return 7877 * 0 if there are no known bad blocks in the range 7878 * 1 if there are known bad block which are all acknowledged 7879 * -1 if there are bad blocks which have not yet been acknowledged in metadata. 7880 * plus the start/length of the first bad section we overlap. 7881 */ 7882 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, 7883 sector_t *first_bad, int *bad_sectors) 7884 { 7885 int hi; 7886 int lo = 0; 7887 u64 *p = bb->page; 7888 int rv = 0; 7889 sector_t target = s + sectors; 7890 unsigned seq; 7891 7892 if (bb->shift > 0) { 7893 /* round the start down, and the end up */ 7894 s >>= bb->shift; 7895 target += (1<<bb->shift) - 1; 7896 target >>= bb->shift; 7897 sectors = target - s; 7898 } 7899 /* 'target' is now the first block after the bad range */ 7900 7901 retry: 7902 seq = read_seqbegin(&bb->lock); 7903 7904 hi = bb->count; 7905 7906 /* Binary search between lo and hi for 'target' 7907 * i.e. for the last range that starts before 'target' 7908 */ 7909 /* INVARIANT: ranges before 'lo' and at-or-after 'hi' 7910 * are known not to be the last range before target. 7911 * VARIANT: hi-lo is the number of possible 7912 * ranges, and decreases until it reaches 1 7913 */ 7914 while (hi - lo > 1) { 7915 int mid = (lo + hi) / 2; 7916 sector_t a = BB_OFFSET(p[mid]); 7917 if (a < target) 7918 /* This could still be the one, earlier ranges 7919 * could not. */ 7920 lo = mid; 7921 else 7922 /* This and later ranges are definitely out. */ 7923 hi = mid; 7924 } 7925 /* 'lo' might be the last that started before target, but 'hi' isn't */ 7926 if (hi > lo) { 7927 /* need to check all range that end after 's' to see if 7928 * any are unacknowledged. 7929 */ 7930 while (lo >= 0 && 7931 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { 7932 if (BB_OFFSET(p[lo]) < target) { 7933 /* starts before the end, and finishes after 7934 * the start, so they must overlap 7935 */ 7936 if (rv != -1 && BB_ACK(p[lo])) 7937 rv = 1; 7938 else 7939 rv = -1; 7940 *first_bad = BB_OFFSET(p[lo]); 7941 *bad_sectors = BB_LEN(p[lo]); 7942 } 7943 lo--; 7944 } 7945 } 7946 7947 if (read_seqretry(&bb->lock, seq)) 7948 goto retry; 7949 7950 return rv; 7951 } 7952 EXPORT_SYMBOL_GPL(md_is_badblock); 7953 7954 /* 7955 * Add a range of bad blocks to the table. 7956 * This might extend the table, or might contract it 7957 * if two adjacent ranges can be merged. 7958 * We binary-search to find the 'insertion' point, then 7959 * decide how best to handle it. 7960 */ 7961 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, 7962 int acknowledged) 7963 { 7964 u64 *p; 7965 int lo, hi; 7966 int rv = 1; 7967 7968 if (bb->shift < 0) 7969 /* badblocks are disabled */ 7970 return 0; 7971 7972 if (bb->shift) { 7973 /* round the start down, and the end up */ 7974 sector_t next = s + sectors; 7975 s >>= bb->shift; 7976 next += (1<<bb->shift) - 1; 7977 next >>= bb->shift; 7978 sectors = next - s; 7979 } 7980 7981 write_seqlock_irq(&bb->lock); 7982 7983 p = bb->page; 7984 lo = 0; 7985 hi = bb->count; 7986 /* Find the last range that starts at-or-before 's' */ 7987 while (hi - lo > 1) { 7988 int mid = (lo + hi) / 2; 7989 sector_t a = BB_OFFSET(p[mid]); 7990 if (a <= s) 7991 lo = mid; 7992 else 7993 hi = mid; 7994 } 7995 if (hi > lo && BB_OFFSET(p[lo]) > s) 7996 hi = lo; 7997 7998 if (hi > lo) { 7999 /* we found a range that might merge with the start 8000 * of our new range 8001 */ 8002 sector_t a = BB_OFFSET(p[lo]); 8003 sector_t e = a + BB_LEN(p[lo]); 8004 int ack = BB_ACK(p[lo]); 8005 if (e >= s) { 8006 /* Yes, we can merge with a previous range */ 8007 if (s == a && s + sectors >= e) 8008 /* new range covers old */ 8009 ack = acknowledged; 8010 else 8011 ack = ack && acknowledged; 8012 8013 if (e < s + sectors) 8014 e = s + sectors; 8015 if (e - a <= BB_MAX_LEN) { 8016 p[lo] = BB_MAKE(a, e-a, ack); 8017 s = e; 8018 } else { 8019 /* does not all fit in one range, 8020 * make p[lo] maximal 8021 */ 8022 if (BB_LEN(p[lo]) != BB_MAX_LEN) 8023 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); 8024 s = a + BB_MAX_LEN; 8025 } 8026 sectors = e - s; 8027 } 8028 } 8029 if (sectors && hi < bb->count) { 8030 /* 'hi' points to the first range that starts after 's'. 8031 * Maybe we can merge with the start of that range */ 8032 sector_t a = BB_OFFSET(p[hi]); 8033 sector_t e = a + BB_LEN(p[hi]); 8034 int ack = BB_ACK(p[hi]); 8035 if (a <= s + sectors) { 8036 /* merging is possible */ 8037 if (e <= s + sectors) { 8038 /* full overlap */ 8039 e = s + sectors; 8040 ack = acknowledged; 8041 } else 8042 ack = ack && acknowledged; 8043 8044 a = s; 8045 if (e - a <= BB_MAX_LEN) { 8046 p[hi] = BB_MAKE(a, e-a, ack); 8047 s = e; 8048 } else { 8049 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); 8050 s = a + BB_MAX_LEN; 8051 } 8052 sectors = e - s; 8053 lo = hi; 8054 hi++; 8055 } 8056 } 8057 if (sectors == 0 && hi < bb->count) { 8058 /* we might be able to combine lo and hi */ 8059 /* Note: 's' is at the end of 'lo' */ 8060 sector_t a = BB_OFFSET(p[hi]); 8061 int lolen = BB_LEN(p[lo]); 8062 int hilen = BB_LEN(p[hi]); 8063 int newlen = lolen + hilen - (s - a); 8064 if (s >= a && newlen < BB_MAX_LEN) { 8065 /* yes, we can combine them */ 8066 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); 8067 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); 8068 memmove(p + hi, p + hi + 1, 8069 (bb->count - hi - 1) * 8); 8070 bb->count--; 8071 } 8072 } 8073 while (sectors) { 8074 /* didn't merge (it all). 8075 * Need to add a range just before 'hi' */ 8076 if (bb->count >= MD_MAX_BADBLOCKS) { 8077 /* No room for more */ 8078 rv = 0; 8079 break; 8080 } else { 8081 int this_sectors = sectors; 8082 memmove(p + hi + 1, p + hi, 8083 (bb->count - hi) * 8); 8084 bb->count++; 8085 8086 if (this_sectors > BB_MAX_LEN) 8087 this_sectors = BB_MAX_LEN; 8088 p[hi] = BB_MAKE(s, this_sectors, acknowledged); 8089 sectors -= this_sectors; 8090 s += this_sectors; 8091 } 8092 } 8093 8094 bb->changed = 1; 8095 if (!acknowledged) 8096 bb->unacked_exist = 1; 8097 write_sequnlock_irq(&bb->lock); 8098 8099 return rv; 8100 } 8101 8102 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, 8103 int is_new) 8104 { 8105 int rv; 8106 if (is_new) 8107 s += rdev->new_data_offset; 8108 else 8109 s += rdev->data_offset; 8110 rv = md_set_badblocks(&rdev->badblocks, 8111 s, sectors, 0); 8112 if (rv) { 8113 /* Make sure they get written out promptly */ 8114 sysfs_notify_dirent_safe(rdev->sysfs_state); 8115 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); 8116 md_wakeup_thread(rdev->mddev->thread); 8117 } 8118 return rv; 8119 } 8120 EXPORT_SYMBOL_GPL(rdev_set_badblocks); 8121 8122 /* 8123 * Remove a range of bad blocks from the table. 8124 * This may involve extending the table if we spilt a region, 8125 * but it must not fail. So if the table becomes full, we just 8126 * drop the remove request. 8127 */ 8128 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) 8129 { 8130 u64 *p; 8131 int lo, hi; 8132 sector_t target = s + sectors; 8133 int rv = 0; 8134 8135 if (bb->shift > 0) { 8136 /* When clearing we round the start up and the end down. 8137 * This should not matter as the shift should align with 8138 * the block size and no rounding should ever be needed. 8139 * However it is better the think a block is bad when it 8140 * isn't than to think a block is not bad when it is. 8141 */ 8142 s += (1<<bb->shift) - 1; 8143 s >>= bb->shift; 8144 target >>= bb->shift; 8145 sectors = target - s; 8146 } 8147 8148 write_seqlock_irq(&bb->lock); 8149 8150 p = bb->page; 8151 lo = 0; 8152 hi = bb->count; 8153 /* Find the last range that starts before 'target' */ 8154 while (hi - lo > 1) { 8155 int mid = (lo + hi) / 2; 8156 sector_t a = BB_OFFSET(p[mid]); 8157 if (a < target) 8158 lo = mid; 8159 else 8160 hi = mid; 8161 } 8162 if (hi > lo) { 8163 /* p[lo] is the last range that could overlap the 8164 * current range. Earlier ranges could also overlap, 8165 * but only this one can overlap the end of the range. 8166 */ 8167 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { 8168 /* Partial overlap, leave the tail of this range */ 8169 int ack = BB_ACK(p[lo]); 8170 sector_t a = BB_OFFSET(p[lo]); 8171 sector_t end = a + BB_LEN(p[lo]); 8172 8173 if (a < s) { 8174 /* we need to split this range */ 8175 if (bb->count >= MD_MAX_BADBLOCKS) { 8176 rv = 0; 8177 goto out; 8178 } 8179 memmove(p+lo+1, p+lo, (bb->count - lo) * 8); 8180 bb->count++; 8181 p[lo] = BB_MAKE(a, s-a, ack); 8182 lo++; 8183 } 8184 p[lo] = BB_MAKE(target, end - target, ack); 8185 /* there is no longer an overlap */ 8186 hi = lo; 8187 lo--; 8188 } 8189 while (lo >= 0 && 8190 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { 8191 /* This range does overlap */ 8192 if (BB_OFFSET(p[lo]) < s) { 8193 /* Keep the early parts of this range. */ 8194 int ack = BB_ACK(p[lo]); 8195 sector_t start = BB_OFFSET(p[lo]); 8196 p[lo] = BB_MAKE(start, s - start, ack); 8197 /* now low doesn't overlap, so.. */ 8198 break; 8199 } 8200 lo--; 8201 } 8202 /* 'lo' is strictly before, 'hi' is strictly after, 8203 * anything between needs to be discarded 8204 */ 8205 if (hi - lo > 1) { 8206 memmove(p+lo+1, p+hi, (bb->count - hi) * 8); 8207 bb->count -= (hi - lo - 1); 8208 } 8209 } 8210 8211 bb->changed = 1; 8212 out: 8213 write_sequnlock_irq(&bb->lock); 8214 return rv; 8215 } 8216 8217 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, 8218 int is_new) 8219 { 8220 if (is_new) 8221 s += rdev->new_data_offset; 8222 else 8223 s += rdev->data_offset; 8224 return md_clear_badblocks(&rdev->badblocks, 8225 s, sectors); 8226 } 8227 EXPORT_SYMBOL_GPL(rdev_clear_badblocks); 8228 8229 /* 8230 * Acknowledge all bad blocks in a list. 8231 * This only succeeds if ->changed is clear. It is used by 8232 * in-kernel metadata updates 8233 */ 8234 void md_ack_all_badblocks(struct badblocks *bb) 8235 { 8236 if (bb->page == NULL || bb->changed) 8237 /* no point even trying */ 8238 return; 8239 write_seqlock_irq(&bb->lock); 8240 8241 if (bb->changed == 0 && bb->unacked_exist) { 8242 u64 *p = bb->page; 8243 int i; 8244 for (i = 0; i < bb->count ; i++) { 8245 if (!BB_ACK(p[i])) { 8246 sector_t start = BB_OFFSET(p[i]); 8247 int len = BB_LEN(p[i]); 8248 p[i] = BB_MAKE(start, len, 1); 8249 } 8250 } 8251 bb->unacked_exist = 0; 8252 } 8253 write_sequnlock_irq(&bb->lock); 8254 } 8255 EXPORT_SYMBOL_GPL(md_ack_all_badblocks); 8256 8257 /* sysfs access to bad-blocks list. 8258 * We present two files. 8259 * 'bad-blocks' lists sector numbers and lengths of ranges that 8260 * are recorded as bad. The list is truncated to fit within 8261 * the one-page limit of sysfs. 8262 * Writing "sector length" to this file adds an acknowledged 8263 * bad block list. 8264 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet 8265 * been acknowledged. Writing to this file adds bad blocks 8266 * without acknowledging them. This is largely for testing. 8267 */ 8268 8269 static ssize_t 8270 badblocks_show(struct badblocks *bb, char *page, int unack) 8271 { 8272 size_t len; 8273 int i; 8274 u64 *p = bb->page; 8275 unsigned seq; 8276 8277 if (bb->shift < 0) 8278 return 0; 8279 8280 retry: 8281 seq = read_seqbegin(&bb->lock); 8282 8283 len = 0; 8284 i = 0; 8285 8286 while (len < PAGE_SIZE && i < bb->count) { 8287 sector_t s = BB_OFFSET(p[i]); 8288 unsigned int length = BB_LEN(p[i]); 8289 int ack = BB_ACK(p[i]); 8290 i++; 8291 8292 if (unack && ack) 8293 continue; 8294 8295 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", 8296 (unsigned long long)s << bb->shift, 8297 length << bb->shift); 8298 } 8299 if (unack && len == 0) 8300 bb->unacked_exist = 0; 8301 8302 if (read_seqretry(&bb->lock, seq)) 8303 goto retry; 8304 8305 return len; 8306 } 8307 8308 #define DO_DEBUG 1 8309 8310 static ssize_t 8311 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) 8312 { 8313 unsigned long long sector; 8314 int length; 8315 char newline; 8316 #ifdef DO_DEBUG 8317 /* Allow clearing via sysfs *only* for testing/debugging. 8318 * Normally only a successful write may clear a badblock 8319 */ 8320 int clear = 0; 8321 if (page[0] == '-') { 8322 clear = 1; 8323 page++; 8324 } 8325 #endif /* DO_DEBUG */ 8326 8327 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { 8328 case 3: 8329 if (newline != '\n') 8330 return -EINVAL; 8331 case 2: 8332 if (length <= 0) 8333 return -EINVAL; 8334 break; 8335 default: 8336 return -EINVAL; 8337 } 8338 8339 #ifdef DO_DEBUG 8340 if (clear) { 8341 md_clear_badblocks(bb, sector, length); 8342 return len; 8343 } 8344 #endif /* DO_DEBUG */ 8345 if (md_set_badblocks(bb, sector, length, !unack)) 8346 return len; 8347 else 8348 return -ENOSPC; 8349 } 8350 8351 static int md_notify_reboot(struct notifier_block *this, 8352 unsigned long code, void *x) 8353 { 8354 struct list_head *tmp; 8355 struct mddev *mddev; 8356 int need_delay = 0; 8357 8358 for_each_mddev(mddev, tmp) { 8359 if (mddev_trylock(mddev)) { 8360 if (mddev->pers) 8361 __md_stop_writes(mddev); 8362 mddev->safemode = 2; 8363 mddev_unlock(mddev); 8364 } 8365 need_delay = 1; 8366 } 8367 /* 8368 * certain more exotic SCSI devices are known to be 8369 * volatile wrt too early system reboots. While the 8370 * right place to handle this issue is the given 8371 * driver, we do want to have a safe RAID driver ... 8372 */ 8373 if (need_delay) 8374 mdelay(1000*1); 8375 8376 return NOTIFY_DONE; 8377 } 8378 8379 static struct notifier_block md_notifier = { 8380 .notifier_call = md_notify_reboot, 8381 .next = NULL, 8382 .priority = INT_MAX, /* before any real devices */ 8383 }; 8384 8385 static void md_geninit(void) 8386 { 8387 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 8388 8389 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); 8390 } 8391 8392 static int __init md_init(void) 8393 { 8394 int ret = -ENOMEM; 8395 8396 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); 8397 if (!md_wq) 8398 goto err_wq; 8399 8400 md_misc_wq = alloc_workqueue("md_misc", 0, 0); 8401 if (!md_misc_wq) 8402 goto err_misc_wq; 8403 8404 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) 8405 goto err_md; 8406 8407 if ((ret = register_blkdev(0, "mdp")) < 0) 8408 goto err_mdp; 8409 mdp_major = ret; 8410 8411 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, 8412 md_probe, NULL, NULL); 8413 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, 8414 md_probe, NULL, NULL); 8415 8416 register_reboot_notifier(&md_notifier); 8417 raid_table_header = register_sysctl_table(raid_root_table); 8418 8419 md_geninit(); 8420 return 0; 8421 8422 err_mdp: 8423 unregister_blkdev(MD_MAJOR, "md"); 8424 err_md: 8425 destroy_workqueue(md_misc_wq); 8426 err_misc_wq: 8427 destroy_workqueue(md_wq); 8428 err_wq: 8429 return ret; 8430 } 8431 8432 #ifndef MODULE 8433 8434 /* 8435 * Searches all registered partitions for autorun RAID arrays 8436 * at boot time. 8437 */ 8438 8439 static LIST_HEAD(all_detected_devices); 8440 struct detected_devices_node { 8441 struct list_head list; 8442 dev_t dev; 8443 }; 8444 8445 void md_autodetect_dev(dev_t dev) 8446 { 8447 struct detected_devices_node *node_detected_dev; 8448 8449 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); 8450 if (node_detected_dev) { 8451 node_detected_dev->dev = dev; 8452 list_add_tail(&node_detected_dev->list, &all_detected_devices); 8453 } else { 8454 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" 8455 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); 8456 } 8457 } 8458 8459 8460 static void autostart_arrays(int part) 8461 { 8462 struct md_rdev *rdev; 8463 struct detected_devices_node *node_detected_dev; 8464 dev_t dev; 8465 int i_scanned, i_passed; 8466 8467 i_scanned = 0; 8468 i_passed = 0; 8469 8470 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 8471 8472 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { 8473 i_scanned++; 8474 node_detected_dev = list_entry(all_detected_devices.next, 8475 struct detected_devices_node, list); 8476 list_del(&node_detected_dev->list); 8477 dev = node_detected_dev->dev; 8478 kfree(node_detected_dev); 8479 rdev = md_import_device(dev,0, 90); 8480 if (IS_ERR(rdev)) 8481 continue; 8482 8483 if (test_bit(Faulty, &rdev->flags)) { 8484 MD_BUG(); 8485 continue; 8486 } 8487 set_bit(AutoDetected, &rdev->flags); 8488 list_add(&rdev->same_set, &pending_raid_disks); 8489 i_passed++; 8490 } 8491 8492 printk(KERN_INFO "md: Scanned %d and added %d devices.\n", 8493 i_scanned, i_passed); 8494 8495 autorun_devices(part); 8496 } 8497 8498 #endif /* !MODULE */ 8499 8500 static __exit void md_exit(void) 8501 { 8502 struct mddev *mddev; 8503 struct list_head *tmp; 8504 8505 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); 8506 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); 8507 8508 unregister_blkdev(MD_MAJOR,"md"); 8509 unregister_blkdev(mdp_major, "mdp"); 8510 unregister_reboot_notifier(&md_notifier); 8511 unregister_sysctl_table(raid_table_header); 8512 remove_proc_entry("mdstat", NULL); 8513 for_each_mddev(mddev, tmp) { 8514 export_array(mddev); 8515 mddev->hold_active = 0; 8516 } 8517 destroy_workqueue(md_misc_wq); 8518 destroy_workqueue(md_wq); 8519 } 8520 8521 subsys_initcall(md_init); 8522 module_exit(md_exit) 8523 8524 static int get_ro(char *buffer, struct kernel_param *kp) 8525 { 8526 return sprintf(buffer, "%d", start_readonly); 8527 } 8528 static int set_ro(const char *val, struct kernel_param *kp) 8529 { 8530 char *e; 8531 int num = simple_strtoul(val, &e, 10); 8532 if (*val && (*e == '\0' || *e == '\n')) { 8533 start_readonly = num; 8534 return 0; 8535 } 8536 return -EINVAL; 8537 } 8538 8539 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); 8540 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); 8541 8542 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); 8543 8544 EXPORT_SYMBOL(register_md_personality); 8545 EXPORT_SYMBOL(unregister_md_personality); 8546 EXPORT_SYMBOL(md_error); 8547 EXPORT_SYMBOL(md_done_sync); 8548 EXPORT_SYMBOL(md_write_start); 8549 EXPORT_SYMBOL(md_write_end); 8550 EXPORT_SYMBOL(md_register_thread); 8551 EXPORT_SYMBOL(md_unregister_thread); 8552 EXPORT_SYMBOL(md_wakeup_thread); 8553 EXPORT_SYMBOL(md_check_recovery); 8554 MODULE_LICENSE("GPL"); 8555 MODULE_DESCRIPTION("MD RAID framework"); 8556 MODULE_ALIAS("md"); 8557 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 8558