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/mutex.h> 40 #include <linux/buffer_head.h> /* for invalidate_bdev */ 41 #include <linux/poll.h> 42 #include <linux/ctype.h> 43 #include <linux/string.h> 44 #include <linux/hdreg.h> 45 #include <linux/proc_fs.h> 46 #include <linux/random.h> 47 #include <linux/module.h> 48 #include <linux/reboot.h> 49 #include <linux/file.h> 50 #include <linux/compat.h> 51 #include <linux/delay.h> 52 #include <linux/raid/md_p.h> 53 #include <linux/raid/md_u.h> 54 #include <linux/slab.h> 55 #include "md.h" 56 #include "bitmap.h" 57 58 #ifndef MODULE 59 static void autostart_arrays(int part); 60 #endif 61 62 /* pers_list is a list of registered personalities protected 63 * by pers_lock. 64 * pers_lock does extra service to protect accesses to 65 * mddev->thread when the mutex cannot be held. 66 */ 67 static LIST_HEAD(pers_list); 68 static DEFINE_SPINLOCK(pers_lock); 69 70 static void md_print_devices(void); 71 72 static DECLARE_WAIT_QUEUE_HEAD(resync_wait); 73 static struct workqueue_struct *md_wq; 74 static struct workqueue_struct *md_misc_wq; 75 76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } 77 78 /* 79 * Default number of read corrections we'll attempt on an rdev 80 * before ejecting it from the array. We divide the read error 81 * count by 2 for every hour elapsed between read errors. 82 */ 83 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 84 /* 85 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' 86 * is 1000 KB/sec, so the extra system load does not show up that much. 87 * Increase it if you want to have more _guaranteed_ speed. Note that 88 * the RAID driver will use the maximum available bandwidth if the IO 89 * subsystem is idle. There is also an 'absolute maximum' reconstruction 90 * speed limit - in case reconstruction slows down your system despite 91 * idle IO detection. 92 * 93 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. 94 * or /sys/block/mdX/md/sync_speed_{min,max} 95 */ 96 97 static int sysctl_speed_limit_min = 1000; 98 static int sysctl_speed_limit_max = 200000; 99 static inline int speed_min(struct mddev *mddev) 100 { 101 return mddev->sync_speed_min ? 102 mddev->sync_speed_min : sysctl_speed_limit_min; 103 } 104 105 static inline int speed_max(struct mddev *mddev) 106 { 107 return mddev->sync_speed_max ? 108 mddev->sync_speed_max : sysctl_speed_limit_max; 109 } 110 111 static struct ctl_table_header *raid_table_header; 112 113 static ctl_table raid_table[] = { 114 { 115 .procname = "speed_limit_min", 116 .data = &sysctl_speed_limit_min, 117 .maxlen = sizeof(int), 118 .mode = S_IRUGO|S_IWUSR, 119 .proc_handler = proc_dointvec, 120 }, 121 { 122 .procname = "speed_limit_max", 123 .data = &sysctl_speed_limit_max, 124 .maxlen = sizeof(int), 125 .mode = S_IRUGO|S_IWUSR, 126 .proc_handler = proc_dointvec, 127 }, 128 { } 129 }; 130 131 static ctl_table raid_dir_table[] = { 132 { 133 .procname = "raid", 134 .maxlen = 0, 135 .mode = S_IRUGO|S_IXUGO, 136 .child = raid_table, 137 }, 138 { } 139 }; 140 141 static ctl_table raid_root_table[] = { 142 { 143 .procname = "dev", 144 .maxlen = 0, 145 .mode = 0555, 146 .child = raid_dir_table, 147 }, 148 { } 149 }; 150 151 static const struct block_device_operations md_fops; 152 153 static int start_readonly; 154 155 /* bio_clone_mddev 156 * like bio_clone, but with a local bio set 157 */ 158 159 static void mddev_bio_destructor(struct bio *bio) 160 { 161 struct mddev *mddev, **mddevp; 162 163 mddevp = (void*)bio; 164 mddev = mddevp[-1]; 165 166 bio_free(bio, mddev->bio_set); 167 } 168 169 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, 170 struct mddev *mddev) 171 { 172 struct bio *b; 173 struct mddev **mddevp; 174 175 if (!mddev || !mddev->bio_set) 176 return bio_alloc(gfp_mask, nr_iovecs); 177 178 b = bio_alloc_bioset(gfp_mask, nr_iovecs, 179 mddev->bio_set); 180 if (!b) 181 return NULL; 182 mddevp = (void*)b; 183 mddevp[-1] = mddev; 184 b->bi_destructor = mddev_bio_destructor; 185 return b; 186 } 187 EXPORT_SYMBOL_GPL(bio_alloc_mddev); 188 189 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, 190 struct mddev *mddev) 191 { 192 struct bio *b; 193 struct mddev **mddevp; 194 195 if (!mddev || !mddev->bio_set) 196 return bio_clone(bio, gfp_mask); 197 198 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, 199 mddev->bio_set); 200 if (!b) 201 return NULL; 202 mddevp = (void*)b; 203 mddevp[-1] = mddev; 204 b->bi_destructor = mddev_bio_destructor; 205 __bio_clone(b, bio); 206 if (bio_integrity(bio)) { 207 int ret; 208 209 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set); 210 211 if (ret < 0) { 212 bio_put(b); 213 return NULL; 214 } 215 } 216 217 return b; 218 } 219 EXPORT_SYMBOL_GPL(bio_clone_mddev); 220 221 void md_trim_bio(struct bio *bio, int offset, int size) 222 { 223 /* 'bio' is a cloned bio which we need to trim to match 224 * the given offset and size. 225 * This requires adjusting bi_sector, bi_size, and bi_io_vec 226 */ 227 int i; 228 struct bio_vec *bvec; 229 int sofar = 0; 230 231 size <<= 9; 232 if (offset == 0 && size == bio->bi_size) 233 return; 234 235 bio->bi_sector += offset; 236 bio->bi_size = size; 237 offset <<= 9; 238 clear_bit(BIO_SEG_VALID, &bio->bi_flags); 239 240 while (bio->bi_idx < bio->bi_vcnt && 241 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) { 242 /* remove this whole bio_vec */ 243 offset -= bio->bi_io_vec[bio->bi_idx].bv_len; 244 bio->bi_idx++; 245 } 246 if (bio->bi_idx < bio->bi_vcnt) { 247 bio->bi_io_vec[bio->bi_idx].bv_offset += offset; 248 bio->bi_io_vec[bio->bi_idx].bv_len -= offset; 249 } 250 /* avoid any complications with bi_idx being non-zero*/ 251 if (bio->bi_idx) { 252 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx, 253 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec)); 254 bio->bi_vcnt -= bio->bi_idx; 255 bio->bi_idx = 0; 256 } 257 /* Make sure vcnt and last bv are not too big */ 258 bio_for_each_segment(bvec, bio, i) { 259 if (sofar + bvec->bv_len > size) 260 bvec->bv_len = size - sofar; 261 if (bvec->bv_len == 0) { 262 bio->bi_vcnt = i; 263 break; 264 } 265 sofar += bvec->bv_len; 266 } 267 } 268 EXPORT_SYMBOL_GPL(md_trim_bio); 269 270 /* 271 * We have a system wide 'event count' that is incremented 272 * on any 'interesting' event, and readers of /proc/mdstat 273 * can use 'poll' or 'select' to find out when the event 274 * count increases. 275 * 276 * Events are: 277 * start array, stop array, error, add device, remove device, 278 * start build, activate spare 279 */ 280 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); 281 static atomic_t md_event_count; 282 void md_new_event(struct mddev *mddev) 283 { 284 atomic_inc(&md_event_count); 285 wake_up(&md_event_waiters); 286 } 287 EXPORT_SYMBOL_GPL(md_new_event); 288 289 /* Alternate version that can be called from interrupts 290 * when calling sysfs_notify isn't needed. 291 */ 292 static void md_new_event_inintr(struct mddev *mddev) 293 { 294 atomic_inc(&md_event_count); 295 wake_up(&md_event_waiters); 296 } 297 298 /* 299 * Enables to iterate over all existing md arrays 300 * all_mddevs_lock protects this list. 301 */ 302 static LIST_HEAD(all_mddevs); 303 static DEFINE_SPINLOCK(all_mddevs_lock); 304 305 306 /* 307 * iterates through all used mddevs in the system. 308 * We take care to grab the all_mddevs_lock whenever navigating 309 * the list, and to always hold a refcount when unlocked. 310 * Any code which breaks out of this loop while own 311 * a reference to the current mddev and must mddev_put it. 312 */ 313 #define for_each_mddev(_mddev,_tmp) \ 314 \ 315 for (({ spin_lock(&all_mddevs_lock); \ 316 _tmp = all_mddevs.next; \ 317 _mddev = NULL;}); \ 318 ({ if (_tmp != &all_mddevs) \ 319 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ 320 spin_unlock(&all_mddevs_lock); \ 321 if (_mddev) mddev_put(_mddev); \ 322 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ 323 _tmp != &all_mddevs;}); \ 324 ({ spin_lock(&all_mddevs_lock); \ 325 _tmp = _tmp->next;}) \ 326 ) 327 328 329 /* Rather than calling directly into the personality make_request function, 330 * IO requests come here first so that we can check if the device is 331 * being suspended pending a reconfiguration. 332 * We hold a refcount over the call to ->make_request. By the time that 333 * call has finished, the bio has been linked into some internal structure 334 * and so is visible to ->quiesce(), so we don't need the refcount any more. 335 */ 336 static int md_make_request(struct request_queue *q, struct bio *bio) 337 { 338 const int rw = bio_data_dir(bio); 339 struct mddev *mddev = q->queuedata; 340 int rv; 341 int cpu; 342 unsigned int sectors; 343 344 if (mddev == NULL || mddev->pers == NULL 345 || !mddev->ready) { 346 bio_io_error(bio); 347 return 0; 348 } 349 smp_rmb(); /* Ensure implications of 'active' are visible */ 350 rcu_read_lock(); 351 if (mddev->suspended) { 352 DEFINE_WAIT(__wait); 353 for (;;) { 354 prepare_to_wait(&mddev->sb_wait, &__wait, 355 TASK_UNINTERRUPTIBLE); 356 if (!mddev->suspended) 357 break; 358 rcu_read_unlock(); 359 schedule(); 360 rcu_read_lock(); 361 } 362 finish_wait(&mddev->sb_wait, &__wait); 363 } 364 atomic_inc(&mddev->active_io); 365 rcu_read_unlock(); 366 367 /* 368 * save the sectors now since our bio can 369 * go away inside make_request 370 */ 371 sectors = bio_sectors(bio); 372 rv = mddev->pers->make_request(mddev, bio); 373 374 cpu = part_stat_lock(); 375 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); 376 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); 377 part_stat_unlock(); 378 379 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) 380 wake_up(&mddev->sb_wait); 381 382 return rv; 383 } 384 385 /* mddev_suspend makes sure no new requests are submitted 386 * to the device, and that any requests that have been submitted 387 * are completely handled. 388 * Once ->stop is called and completes, the module will be completely 389 * unused. 390 */ 391 void mddev_suspend(struct mddev *mddev) 392 { 393 BUG_ON(mddev->suspended); 394 mddev->suspended = 1; 395 synchronize_rcu(); 396 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); 397 mddev->pers->quiesce(mddev, 1); 398 } 399 EXPORT_SYMBOL_GPL(mddev_suspend); 400 401 void mddev_resume(struct mddev *mddev) 402 { 403 mddev->suspended = 0; 404 wake_up(&mddev->sb_wait); 405 mddev->pers->quiesce(mddev, 0); 406 407 md_wakeup_thread(mddev->thread); 408 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 409 } 410 EXPORT_SYMBOL_GPL(mddev_resume); 411 412 int mddev_congested(struct mddev *mddev, int bits) 413 { 414 return mddev->suspended; 415 } 416 EXPORT_SYMBOL(mddev_congested); 417 418 /* 419 * Generic flush handling for md 420 */ 421 422 static void md_end_flush(struct bio *bio, int err) 423 { 424 struct md_rdev *rdev = bio->bi_private; 425 struct mddev *mddev = rdev->mddev; 426 427 rdev_dec_pending(rdev, mddev); 428 429 if (atomic_dec_and_test(&mddev->flush_pending)) { 430 /* The pre-request flush has finished */ 431 queue_work(md_wq, &mddev->flush_work); 432 } 433 bio_put(bio); 434 } 435 436 static void md_submit_flush_data(struct work_struct *ws); 437 438 static void submit_flushes(struct work_struct *ws) 439 { 440 struct mddev *mddev = container_of(ws, struct mddev, flush_work); 441 struct md_rdev *rdev; 442 443 INIT_WORK(&mddev->flush_work, md_submit_flush_data); 444 atomic_set(&mddev->flush_pending, 1); 445 rcu_read_lock(); 446 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 447 if (rdev->raid_disk >= 0 && 448 !test_bit(Faulty, &rdev->flags)) { 449 /* Take two references, one is dropped 450 * when request finishes, one after 451 * we reclaim rcu_read_lock 452 */ 453 struct bio *bi; 454 atomic_inc(&rdev->nr_pending); 455 atomic_inc(&rdev->nr_pending); 456 rcu_read_unlock(); 457 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev); 458 bi->bi_end_io = md_end_flush; 459 bi->bi_private = rdev; 460 bi->bi_bdev = rdev->bdev; 461 atomic_inc(&mddev->flush_pending); 462 submit_bio(WRITE_FLUSH, bi); 463 rcu_read_lock(); 464 rdev_dec_pending(rdev, mddev); 465 } 466 rcu_read_unlock(); 467 if (atomic_dec_and_test(&mddev->flush_pending)) 468 queue_work(md_wq, &mddev->flush_work); 469 } 470 471 static void md_submit_flush_data(struct work_struct *ws) 472 { 473 struct mddev *mddev = container_of(ws, struct mddev, flush_work); 474 struct bio *bio = mddev->flush_bio; 475 476 if (bio->bi_size == 0) 477 /* an empty barrier - all done */ 478 bio_endio(bio, 0); 479 else { 480 bio->bi_rw &= ~REQ_FLUSH; 481 if (mddev->pers->make_request(mddev, bio)) 482 generic_make_request(bio); 483 } 484 485 mddev->flush_bio = NULL; 486 wake_up(&mddev->sb_wait); 487 } 488 489 void md_flush_request(struct mddev *mddev, struct bio *bio) 490 { 491 spin_lock_irq(&mddev->write_lock); 492 wait_event_lock_irq(mddev->sb_wait, 493 !mddev->flush_bio, 494 mddev->write_lock, /*nothing*/); 495 mddev->flush_bio = bio; 496 spin_unlock_irq(&mddev->write_lock); 497 498 INIT_WORK(&mddev->flush_work, submit_flushes); 499 queue_work(md_wq, &mddev->flush_work); 500 } 501 EXPORT_SYMBOL(md_flush_request); 502 503 /* Support for plugging. 504 * This mirrors the plugging support in request_queue, but does not 505 * require having a whole queue or request structures. 506 * We allocate an md_plug_cb for each md device and each thread it gets 507 * plugged on. This links tot the private plug_handle structure in the 508 * personality data where we keep a count of the number of outstanding 509 * plugs so other code can see if a plug is active. 510 */ 511 struct md_plug_cb { 512 struct blk_plug_cb cb; 513 struct mddev *mddev; 514 }; 515 516 static void plugger_unplug(struct blk_plug_cb *cb) 517 { 518 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb); 519 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt)) 520 md_wakeup_thread(mdcb->mddev->thread); 521 kfree(mdcb); 522 } 523 524 /* Check that an unplug wakeup will come shortly. 525 * If not, wakeup the md thread immediately 526 */ 527 int mddev_check_plugged(struct mddev *mddev) 528 { 529 struct blk_plug *plug = current->plug; 530 struct md_plug_cb *mdcb; 531 532 if (!plug) 533 return 0; 534 535 list_for_each_entry(mdcb, &plug->cb_list, cb.list) { 536 if (mdcb->cb.callback == plugger_unplug && 537 mdcb->mddev == mddev) { 538 /* Already on the list, move to top */ 539 if (mdcb != list_first_entry(&plug->cb_list, 540 struct md_plug_cb, 541 cb.list)) 542 list_move(&mdcb->cb.list, &plug->cb_list); 543 return 1; 544 } 545 } 546 /* Not currently on the callback list */ 547 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC); 548 if (!mdcb) 549 return 0; 550 551 mdcb->mddev = mddev; 552 mdcb->cb.callback = plugger_unplug; 553 atomic_inc(&mddev->plug_cnt); 554 list_add(&mdcb->cb.list, &plug->cb_list); 555 return 1; 556 } 557 EXPORT_SYMBOL_GPL(mddev_check_plugged); 558 559 static inline struct mddev *mddev_get(struct mddev *mddev) 560 { 561 atomic_inc(&mddev->active); 562 return mddev; 563 } 564 565 static void mddev_delayed_delete(struct work_struct *ws); 566 567 static void mddev_put(struct mddev *mddev) 568 { 569 struct bio_set *bs = NULL; 570 571 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 572 return; 573 if (!mddev->raid_disks && list_empty(&mddev->disks) && 574 mddev->ctime == 0 && !mddev->hold_active) { 575 /* Array is not configured at all, and not held active, 576 * so destroy it */ 577 list_del(&mddev->all_mddevs); 578 bs = mddev->bio_set; 579 mddev->bio_set = NULL; 580 if (mddev->gendisk) { 581 /* We did a probe so need to clean up. Call 582 * queue_work inside the spinlock so that 583 * flush_workqueue() after mddev_find will 584 * succeed in waiting for the work to be done. 585 */ 586 INIT_WORK(&mddev->del_work, mddev_delayed_delete); 587 queue_work(md_misc_wq, &mddev->del_work); 588 } else 589 kfree(mddev); 590 } 591 spin_unlock(&all_mddevs_lock); 592 if (bs) 593 bioset_free(bs); 594 } 595 596 void mddev_init(struct mddev *mddev) 597 { 598 mutex_init(&mddev->open_mutex); 599 mutex_init(&mddev->reconfig_mutex); 600 mutex_init(&mddev->bitmap_info.mutex); 601 INIT_LIST_HEAD(&mddev->disks); 602 INIT_LIST_HEAD(&mddev->all_mddevs); 603 init_timer(&mddev->safemode_timer); 604 atomic_set(&mddev->active, 1); 605 atomic_set(&mddev->openers, 0); 606 atomic_set(&mddev->active_io, 0); 607 atomic_set(&mddev->plug_cnt, 0); 608 spin_lock_init(&mddev->write_lock); 609 atomic_set(&mddev->flush_pending, 0); 610 init_waitqueue_head(&mddev->sb_wait); 611 init_waitqueue_head(&mddev->recovery_wait); 612 mddev->reshape_position = MaxSector; 613 mddev->resync_min = 0; 614 mddev->resync_max = MaxSector; 615 mddev->level = LEVEL_NONE; 616 } 617 EXPORT_SYMBOL_GPL(mddev_init); 618 619 static struct mddev * mddev_find(dev_t unit) 620 { 621 struct mddev *mddev, *new = NULL; 622 623 if (unit && MAJOR(unit) != MD_MAJOR) 624 unit &= ~((1<<MdpMinorShift)-1); 625 626 retry: 627 spin_lock(&all_mddevs_lock); 628 629 if (unit) { 630 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 631 if (mddev->unit == unit) { 632 mddev_get(mddev); 633 spin_unlock(&all_mddevs_lock); 634 kfree(new); 635 return mddev; 636 } 637 638 if (new) { 639 list_add(&new->all_mddevs, &all_mddevs); 640 spin_unlock(&all_mddevs_lock); 641 new->hold_active = UNTIL_IOCTL; 642 return new; 643 } 644 } else if (new) { 645 /* find an unused unit number */ 646 static int next_minor = 512; 647 int start = next_minor; 648 int is_free = 0; 649 int dev = 0; 650 while (!is_free) { 651 dev = MKDEV(MD_MAJOR, next_minor); 652 next_minor++; 653 if (next_minor > MINORMASK) 654 next_minor = 0; 655 if (next_minor == start) { 656 /* Oh dear, all in use. */ 657 spin_unlock(&all_mddevs_lock); 658 kfree(new); 659 return NULL; 660 } 661 662 is_free = 1; 663 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 664 if (mddev->unit == dev) { 665 is_free = 0; 666 break; 667 } 668 } 669 new->unit = dev; 670 new->md_minor = MINOR(dev); 671 new->hold_active = UNTIL_STOP; 672 list_add(&new->all_mddevs, &all_mddevs); 673 spin_unlock(&all_mddevs_lock); 674 return new; 675 } 676 spin_unlock(&all_mddevs_lock); 677 678 new = kzalloc(sizeof(*new), GFP_KERNEL); 679 if (!new) 680 return NULL; 681 682 new->unit = unit; 683 if (MAJOR(unit) == MD_MAJOR) 684 new->md_minor = MINOR(unit); 685 else 686 new->md_minor = MINOR(unit) >> MdpMinorShift; 687 688 mddev_init(new); 689 690 goto retry; 691 } 692 693 static inline int mddev_lock(struct mddev * mddev) 694 { 695 return mutex_lock_interruptible(&mddev->reconfig_mutex); 696 } 697 698 static inline int mddev_is_locked(struct mddev *mddev) 699 { 700 return mutex_is_locked(&mddev->reconfig_mutex); 701 } 702 703 static inline int mddev_trylock(struct mddev * mddev) 704 { 705 return mutex_trylock(&mddev->reconfig_mutex); 706 } 707 708 static struct attribute_group md_redundancy_group; 709 710 static void mddev_unlock(struct mddev * mddev) 711 { 712 if (mddev->to_remove) { 713 /* These cannot be removed under reconfig_mutex as 714 * an access to the files will try to take reconfig_mutex 715 * while holding the file unremovable, which leads to 716 * a deadlock. 717 * So hold set sysfs_active while the remove in happeing, 718 * and anything else which might set ->to_remove or my 719 * otherwise change the sysfs namespace will fail with 720 * -EBUSY if sysfs_active is still set. 721 * We set sysfs_active under reconfig_mutex and elsewhere 722 * test it under the same mutex to ensure its correct value 723 * is seen. 724 */ 725 struct attribute_group *to_remove = mddev->to_remove; 726 mddev->to_remove = NULL; 727 mddev->sysfs_active = 1; 728 mutex_unlock(&mddev->reconfig_mutex); 729 730 if (mddev->kobj.sd) { 731 if (to_remove != &md_redundancy_group) 732 sysfs_remove_group(&mddev->kobj, to_remove); 733 if (mddev->pers == NULL || 734 mddev->pers->sync_request == NULL) { 735 sysfs_remove_group(&mddev->kobj, &md_redundancy_group); 736 if (mddev->sysfs_action) 737 sysfs_put(mddev->sysfs_action); 738 mddev->sysfs_action = NULL; 739 } 740 } 741 mddev->sysfs_active = 0; 742 } else 743 mutex_unlock(&mddev->reconfig_mutex); 744 745 /* As we've dropped the mutex we need a spinlock to 746 * make sure the thread doesn't disappear 747 */ 748 spin_lock(&pers_lock); 749 md_wakeup_thread(mddev->thread); 750 spin_unlock(&pers_lock); 751 } 752 753 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr) 754 { 755 struct md_rdev *rdev; 756 757 list_for_each_entry(rdev, &mddev->disks, same_set) 758 if (rdev->desc_nr == nr) 759 return rdev; 760 761 return NULL; 762 } 763 764 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev) 765 { 766 struct md_rdev *rdev; 767 768 list_for_each_entry(rdev, &mddev->disks, same_set) 769 if (rdev->bdev->bd_dev == dev) 770 return rdev; 771 772 return NULL; 773 } 774 775 static struct md_personality *find_pers(int level, char *clevel) 776 { 777 struct md_personality *pers; 778 list_for_each_entry(pers, &pers_list, list) { 779 if (level != LEVEL_NONE && pers->level == level) 780 return pers; 781 if (strcmp(pers->name, clevel)==0) 782 return pers; 783 } 784 return NULL; 785 } 786 787 /* return the offset of the super block in 512byte sectors */ 788 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) 789 { 790 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; 791 return MD_NEW_SIZE_SECTORS(num_sectors); 792 } 793 794 static int alloc_disk_sb(struct md_rdev * rdev) 795 { 796 if (rdev->sb_page) 797 MD_BUG(); 798 799 rdev->sb_page = alloc_page(GFP_KERNEL); 800 if (!rdev->sb_page) { 801 printk(KERN_ALERT "md: out of memory.\n"); 802 return -ENOMEM; 803 } 804 805 return 0; 806 } 807 808 static void free_disk_sb(struct md_rdev * rdev) 809 { 810 if (rdev->sb_page) { 811 put_page(rdev->sb_page); 812 rdev->sb_loaded = 0; 813 rdev->sb_page = NULL; 814 rdev->sb_start = 0; 815 rdev->sectors = 0; 816 } 817 if (rdev->bb_page) { 818 put_page(rdev->bb_page); 819 rdev->bb_page = NULL; 820 } 821 } 822 823 824 static void super_written(struct bio *bio, int error) 825 { 826 struct md_rdev *rdev = bio->bi_private; 827 struct mddev *mddev = rdev->mddev; 828 829 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { 830 printk("md: super_written gets error=%d, uptodate=%d\n", 831 error, test_bit(BIO_UPTODATE, &bio->bi_flags)); 832 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); 833 md_error(mddev, rdev); 834 } 835 836 if (atomic_dec_and_test(&mddev->pending_writes)) 837 wake_up(&mddev->sb_wait); 838 bio_put(bio); 839 } 840 841 void md_super_write(struct mddev *mddev, struct md_rdev *rdev, 842 sector_t sector, int size, struct page *page) 843 { 844 /* write first size bytes of page to sector of rdev 845 * Increment mddev->pending_writes before returning 846 * and decrement it on completion, waking up sb_wait 847 * if zero is reached. 848 * If an error occurred, call md_error 849 */ 850 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); 851 852 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; 853 bio->bi_sector = sector; 854 bio_add_page(bio, page, size, 0); 855 bio->bi_private = rdev; 856 bio->bi_end_io = super_written; 857 858 atomic_inc(&mddev->pending_writes); 859 submit_bio(WRITE_FLUSH_FUA, bio); 860 } 861 862 void md_super_wait(struct mddev *mddev) 863 { 864 /* wait for all superblock writes that were scheduled to complete */ 865 DEFINE_WAIT(wq); 866 for(;;) { 867 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 868 if (atomic_read(&mddev->pending_writes)==0) 869 break; 870 schedule(); 871 } 872 finish_wait(&mddev->sb_wait, &wq); 873 } 874 875 static void bi_complete(struct bio *bio, int error) 876 { 877 complete((struct completion*)bio->bi_private); 878 } 879 880 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, 881 struct page *page, int rw, bool metadata_op) 882 { 883 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); 884 struct completion event; 885 int ret; 886 887 rw |= REQ_SYNC; 888 889 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? 890 rdev->meta_bdev : rdev->bdev; 891 if (metadata_op) 892 bio->bi_sector = sector + rdev->sb_start; 893 else 894 bio->bi_sector = sector + rdev->data_offset; 895 bio_add_page(bio, page, size, 0); 896 init_completion(&event); 897 bio->bi_private = &event; 898 bio->bi_end_io = bi_complete; 899 submit_bio(rw, bio); 900 wait_for_completion(&event); 901 902 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 903 bio_put(bio); 904 return ret; 905 } 906 EXPORT_SYMBOL_GPL(sync_page_io); 907 908 static int read_disk_sb(struct md_rdev * rdev, int size) 909 { 910 char b[BDEVNAME_SIZE]; 911 if (!rdev->sb_page) { 912 MD_BUG(); 913 return -EINVAL; 914 } 915 if (rdev->sb_loaded) 916 return 0; 917 918 919 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) 920 goto fail; 921 rdev->sb_loaded = 1; 922 return 0; 923 924 fail: 925 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 926 bdevname(rdev->bdev,b)); 927 return -EINVAL; 928 } 929 930 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 931 { 932 return sb1->set_uuid0 == sb2->set_uuid0 && 933 sb1->set_uuid1 == sb2->set_uuid1 && 934 sb1->set_uuid2 == sb2->set_uuid2 && 935 sb1->set_uuid3 == sb2->set_uuid3; 936 } 937 938 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 939 { 940 int ret; 941 mdp_super_t *tmp1, *tmp2; 942 943 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 944 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 945 946 if (!tmp1 || !tmp2) { 947 ret = 0; 948 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); 949 goto abort; 950 } 951 952 *tmp1 = *sb1; 953 *tmp2 = *sb2; 954 955 /* 956 * nr_disks is not constant 957 */ 958 tmp1->nr_disks = 0; 959 tmp2->nr_disks = 0; 960 961 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); 962 abort: 963 kfree(tmp1); 964 kfree(tmp2); 965 return ret; 966 } 967 968 969 static u32 md_csum_fold(u32 csum) 970 { 971 csum = (csum & 0xffff) + (csum >> 16); 972 return (csum & 0xffff) + (csum >> 16); 973 } 974 975 static unsigned int calc_sb_csum(mdp_super_t * sb) 976 { 977 u64 newcsum = 0; 978 u32 *sb32 = (u32*)sb; 979 int i; 980 unsigned int disk_csum, csum; 981 982 disk_csum = sb->sb_csum; 983 sb->sb_csum = 0; 984 985 for (i = 0; i < MD_SB_BYTES/4 ; i++) 986 newcsum += sb32[i]; 987 csum = (newcsum & 0xffffffff) + (newcsum>>32); 988 989 990 #ifdef CONFIG_ALPHA 991 /* This used to use csum_partial, which was wrong for several 992 * reasons including that different results are returned on 993 * different architectures. It isn't critical that we get exactly 994 * the same return value as before (we always csum_fold before 995 * testing, and that removes any differences). However as we 996 * know that csum_partial always returned a 16bit value on 997 * alphas, do a fold to maximise conformity to previous behaviour. 998 */ 999 sb->sb_csum = md_csum_fold(disk_csum); 1000 #else 1001 sb->sb_csum = disk_csum; 1002 #endif 1003 return csum; 1004 } 1005 1006 1007 /* 1008 * Handle superblock details. 1009 * We want to be able to handle multiple superblock formats 1010 * so we have a common interface to them all, and an array of 1011 * different handlers. 1012 * We rely on user-space to write the initial superblock, and support 1013 * reading and updating of superblocks. 1014 * Interface methods are: 1015 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) 1016 * loads and validates a superblock on dev. 1017 * if refdev != NULL, compare superblocks on both devices 1018 * Return: 1019 * 0 - dev has a superblock that is compatible with refdev 1020 * 1 - dev has a superblock that is compatible and newer than refdev 1021 * so dev should be used as the refdev in future 1022 * -EINVAL superblock incompatible or invalid 1023 * -othererror e.g. -EIO 1024 * 1025 * int validate_super(struct mddev *mddev, struct md_rdev *dev) 1026 * Verify that dev is acceptable into mddev. 1027 * The first time, mddev->raid_disks will be 0, and data from 1028 * dev should be merged in. Subsequent calls check that dev 1029 * is new enough. Return 0 or -EINVAL 1030 * 1031 * void sync_super(struct mddev *mddev, struct md_rdev *dev) 1032 * Update the superblock for rdev with data in mddev 1033 * This does not write to disc. 1034 * 1035 */ 1036 1037 struct super_type { 1038 char *name; 1039 struct module *owner; 1040 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev, 1041 int minor_version); 1042 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev); 1043 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev); 1044 unsigned long long (*rdev_size_change)(struct md_rdev *rdev, 1045 sector_t num_sectors); 1046 }; 1047 1048 /* 1049 * Check that the given mddev has no bitmap. 1050 * 1051 * This function is called from the run method of all personalities that do not 1052 * support bitmaps. It prints an error message and returns non-zero if mddev 1053 * has a bitmap. Otherwise, it returns 0. 1054 * 1055 */ 1056 int md_check_no_bitmap(struct mddev *mddev) 1057 { 1058 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) 1059 return 0; 1060 printk(KERN_ERR "%s: bitmaps are not supported for %s\n", 1061 mdname(mddev), mddev->pers->name); 1062 return 1; 1063 } 1064 EXPORT_SYMBOL(md_check_no_bitmap); 1065 1066 /* 1067 * load_super for 0.90.0 1068 */ 1069 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) 1070 { 1071 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1072 mdp_super_t *sb; 1073 int ret; 1074 1075 /* 1076 * Calculate the position of the superblock (512byte sectors), 1077 * it's at the end of the disk. 1078 * 1079 * It also happens to be a multiple of 4Kb. 1080 */ 1081 rdev->sb_start = calc_dev_sboffset(rdev); 1082 1083 ret = read_disk_sb(rdev, MD_SB_BYTES); 1084 if (ret) return ret; 1085 1086 ret = -EINVAL; 1087 1088 bdevname(rdev->bdev, b); 1089 sb = page_address(rdev->sb_page); 1090 1091 if (sb->md_magic != MD_SB_MAGIC) { 1092 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 1093 b); 1094 goto abort; 1095 } 1096 1097 if (sb->major_version != 0 || 1098 sb->minor_version < 90 || 1099 sb->minor_version > 91) { 1100 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 1101 sb->major_version, sb->minor_version, 1102 b); 1103 goto abort; 1104 } 1105 1106 if (sb->raid_disks <= 0) 1107 goto abort; 1108 1109 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { 1110 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 1111 b); 1112 goto abort; 1113 } 1114 1115 rdev->preferred_minor = sb->md_minor; 1116 rdev->data_offset = 0; 1117 rdev->sb_size = MD_SB_BYTES; 1118 rdev->badblocks.shift = -1; 1119 1120 if (sb->level == LEVEL_MULTIPATH) 1121 rdev->desc_nr = -1; 1122 else 1123 rdev->desc_nr = sb->this_disk.number; 1124 1125 if (!refdev) { 1126 ret = 1; 1127 } else { 1128 __u64 ev1, ev2; 1129 mdp_super_t *refsb = page_address(refdev->sb_page); 1130 if (!uuid_equal(refsb, sb)) { 1131 printk(KERN_WARNING "md: %s has different UUID to %s\n", 1132 b, bdevname(refdev->bdev,b2)); 1133 goto abort; 1134 } 1135 if (!sb_equal(refsb, sb)) { 1136 printk(KERN_WARNING "md: %s has same UUID" 1137 " but different superblock to %s\n", 1138 b, bdevname(refdev->bdev, b2)); 1139 goto abort; 1140 } 1141 ev1 = md_event(sb); 1142 ev2 = md_event(refsb); 1143 if (ev1 > ev2) 1144 ret = 1; 1145 else 1146 ret = 0; 1147 } 1148 rdev->sectors = rdev->sb_start; 1149 /* Limit to 4TB as metadata cannot record more than that */ 1150 if (rdev->sectors >= (2ULL << 32)) 1151 rdev->sectors = (2ULL << 32) - 2; 1152 1153 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) 1154 /* "this cannot possibly happen" ... */ 1155 ret = -EINVAL; 1156 1157 abort: 1158 return ret; 1159 } 1160 1161 /* 1162 * validate_super for 0.90.0 1163 */ 1164 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) 1165 { 1166 mdp_disk_t *desc; 1167 mdp_super_t *sb = page_address(rdev->sb_page); 1168 __u64 ev1 = md_event(sb); 1169 1170 rdev->raid_disk = -1; 1171 clear_bit(Faulty, &rdev->flags); 1172 clear_bit(In_sync, &rdev->flags); 1173 clear_bit(WriteMostly, &rdev->flags); 1174 1175 if (mddev->raid_disks == 0) { 1176 mddev->major_version = 0; 1177 mddev->minor_version = sb->minor_version; 1178 mddev->patch_version = sb->patch_version; 1179 mddev->external = 0; 1180 mddev->chunk_sectors = sb->chunk_size >> 9; 1181 mddev->ctime = sb->ctime; 1182 mddev->utime = sb->utime; 1183 mddev->level = sb->level; 1184 mddev->clevel[0] = 0; 1185 mddev->layout = sb->layout; 1186 mddev->raid_disks = sb->raid_disks; 1187 mddev->dev_sectors = ((sector_t)sb->size) * 2; 1188 mddev->events = ev1; 1189 mddev->bitmap_info.offset = 0; 1190 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 1191 1192 if (mddev->minor_version >= 91) { 1193 mddev->reshape_position = sb->reshape_position; 1194 mddev->delta_disks = sb->delta_disks; 1195 mddev->new_level = sb->new_level; 1196 mddev->new_layout = sb->new_layout; 1197 mddev->new_chunk_sectors = sb->new_chunk >> 9; 1198 } else { 1199 mddev->reshape_position = MaxSector; 1200 mddev->delta_disks = 0; 1201 mddev->new_level = mddev->level; 1202 mddev->new_layout = mddev->layout; 1203 mddev->new_chunk_sectors = mddev->chunk_sectors; 1204 } 1205 1206 if (sb->state & (1<<MD_SB_CLEAN)) 1207 mddev->recovery_cp = MaxSector; 1208 else { 1209 if (sb->events_hi == sb->cp_events_hi && 1210 sb->events_lo == sb->cp_events_lo) { 1211 mddev->recovery_cp = sb->recovery_cp; 1212 } else 1213 mddev->recovery_cp = 0; 1214 } 1215 1216 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 1217 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 1218 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 1219 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 1220 1221 mddev->max_disks = MD_SB_DISKS; 1222 1223 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 1224 mddev->bitmap_info.file == NULL) 1225 mddev->bitmap_info.offset = 1226 mddev->bitmap_info.default_offset; 1227 1228 } else if (mddev->pers == NULL) { 1229 /* Insist on good event counter while assembling, except 1230 * for spares (which don't need an event count) */ 1231 ++ev1; 1232 if (sb->disks[rdev->desc_nr].state & ( 1233 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) 1234 if (ev1 < mddev->events) 1235 return -EINVAL; 1236 } else if (mddev->bitmap) { 1237 /* if adding to array with a bitmap, then we can accept an 1238 * older device ... but not too old. 1239 */ 1240 if (ev1 < mddev->bitmap->events_cleared) 1241 return 0; 1242 } else { 1243 if (ev1 < mddev->events) 1244 /* just a hot-add of a new device, leave raid_disk at -1 */ 1245 return 0; 1246 } 1247 1248 if (mddev->level != LEVEL_MULTIPATH) { 1249 desc = sb->disks + rdev->desc_nr; 1250 1251 if (desc->state & (1<<MD_DISK_FAULTY)) 1252 set_bit(Faulty, &rdev->flags); 1253 else if (desc->state & (1<<MD_DISK_SYNC) /* && 1254 desc->raid_disk < mddev->raid_disks */) { 1255 set_bit(In_sync, &rdev->flags); 1256 rdev->raid_disk = desc->raid_disk; 1257 } else if (desc->state & (1<<MD_DISK_ACTIVE)) { 1258 /* active but not in sync implies recovery up to 1259 * reshape position. We don't know exactly where 1260 * that is, so set to zero for now */ 1261 if (mddev->minor_version >= 91) { 1262 rdev->recovery_offset = 0; 1263 rdev->raid_disk = desc->raid_disk; 1264 } 1265 } 1266 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 1267 set_bit(WriteMostly, &rdev->flags); 1268 } else /* MULTIPATH are always insync */ 1269 set_bit(In_sync, &rdev->flags); 1270 return 0; 1271 } 1272 1273 /* 1274 * sync_super for 0.90.0 1275 */ 1276 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) 1277 { 1278 mdp_super_t *sb; 1279 struct md_rdev *rdev2; 1280 int next_spare = mddev->raid_disks; 1281 1282 1283 /* make rdev->sb match mddev data.. 1284 * 1285 * 1/ zero out disks 1286 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 1287 * 3/ any empty disks < next_spare become removed 1288 * 1289 * disks[0] gets initialised to REMOVED because 1290 * we cannot be sure from other fields if it has 1291 * been initialised or not. 1292 */ 1293 int i; 1294 int active=0, working=0,failed=0,spare=0,nr_disks=0; 1295 1296 rdev->sb_size = MD_SB_BYTES; 1297 1298 sb = page_address(rdev->sb_page); 1299 1300 memset(sb, 0, sizeof(*sb)); 1301 1302 sb->md_magic = MD_SB_MAGIC; 1303 sb->major_version = mddev->major_version; 1304 sb->patch_version = mddev->patch_version; 1305 sb->gvalid_words = 0; /* ignored */ 1306 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 1307 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 1308 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 1309 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 1310 1311 sb->ctime = mddev->ctime; 1312 sb->level = mddev->level; 1313 sb->size = mddev->dev_sectors / 2; 1314 sb->raid_disks = mddev->raid_disks; 1315 sb->md_minor = mddev->md_minor; 1316 sb->not_persistent = 0; 1317 sb->utime = mddev->utime; 1318 sb->state = 0; 1319 sb->events_hi = (mddev->events>>32); 1320 sb->events_lo = (u32)mddev->events; 1321 1322 if (mddev->reshape_position == MaxSector) 1323 sb->minor_version = 90; 1324 else { 1325 sb->minor_version = 91; 1326 sb->reshape_position = mddev->reshape_position; 1327 sb->new_level = mddev->new_level; 1328 sb->delta_disks = mddev->delta_disks; 1329 sb->new_layout = mddev->new_layout; 1330 sb->new_chunk = mddev->new_chunk_sectors << 9; 1331 } 1332 mddev->minor_version = sb->minor_version; 1333 if (mddev->in_sync) 1334 { 1335 sb->recovery_cp = mddev->recovery_cp; 1336 sb->cp_events_hi = (mddev->events>>32); 1337 sb->cp_events_lo = (u32)mddev->events; 1338 if (mddev->recovery_cp == MaxSector) 1339 sb->state = (1<< MD_SB_CLEAN); 1340 } else 1341 sb->recovery_cp = 0; 1342 1343 sb->layout = mddev->layout; 1344 sb->chunk_size = mddev->chunk_sectors << 9; 1345 1346 if (mddev->bitmap && mddev->bitmap_info.file == NULL) 1347 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 1348 1349 sb->disks[0].state = (1<<MD_DISK_REMOVED); 1350 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1351 mdp_disk_t *d; 1352 int desc_nr; 1353 int is_active = test_bit(In_sync, &rdev2->flags); 1354 1355 if (rdev2->raid_disk >= 0 && 1356 sb->minor_version >= 91) 1357 /* we have nowhere to store the recovery_offset, 1358 * but if it is not below the reshape_position, 1359 * we can piggy-back on that. 1360 */ 1361 is_active = 1; 1362 if (rdev2->raid_disk < 0 || 1363 test_bit(Faulty, &rdev2->flags)) 1364 is_active = 0; 1365 if (is_active) 1366 desc_nr = rdev2->raid_disk; 1367 else 1368 desc_nr = next_spare++; 1369 rdev2->desc_nr = desc_nr; 1370 d = &sb->disks[rdev2->desc_nr]; 1371 nr_disks++; 1372 d->number = rdev2->desc_nr; 1373 d->major = MAJOR(rdev2->bdev->bd_dev); 1374 d->minor = MINOR(rdev2->bdev->bd_dev); 1375 if (is_active) 1376 d->raid_disk = rdev2->raid_disk; 1377 else 1378 d->raid_disk = rdev2->desc_nr; /* compatibility */ 1379 if (test_bit(Faulty, &rdev2->flags)) 1380 d->state = (1<<MD_DISK_FAULTY); 1381 else if (is_active) { 1382 d->state = (1<<MD_DISK_ACTIVE); 1383 if (test_bit(In_sync, &rdev2->flags)) 1384 d->state |= (1<<MD_DISK_SYNC); 1385 active++; 1386 working++; 1387 } else { 1388 d->state = 0; 1389 spare++; 1390 working++; 1391 } 1392 if (test_bit(WriteMostly, &rdev2->flags)) 1393 d->state |= (1<<MD_DISK_WRITEMOSTLY); 1394 } 1395 /* now set the "removed" and "faulty" bits on any missing devices */ 1396 for (i=0 ; i < mddev->raid_disks ; i++) { 1397 mdp_disk_t *d = &sb->disks[i]; 1398 if (d->state == 0 && d->number == 0) { 1399 d->number = i; 1400 d->raid_disk = i; 1401 d->state = (1<<MD_DISK_REMOVED); 1402 d->state |= (1<<MD_DISK_FAULTY); 1403 failed++; 1404 } 1405 } 1406 sb->nr_disks = nr_disks; 1407 sb->active_disks = active; 1408 sb->working_disks = working; 1409 sb->failed_disks = failed; 1410 sb->spare_disks = spare; 1411 1412 sb->this_disk = sb->disks[rdev->desc_nr]; 1413 sb->sb_csum = calc_sb_csum(sb); 1414 } 1415 1416 /* 1417 * rdev_size_change for 0.90.0 1418 */ 1419 static unsigned long long 1420 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) 1421 { 1422 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1423 return 0; /* component must fit device */ 1424 if (rdev->mddev->bitmap_info.offset) 1425 return 0; /* can't move bitmap */ 1426 rdev->sb_start = calc_dev_sboffset(rdev); 1427 if (!num_sectors || num_sectors > rdev->sb_start) 1428 num_sectors = rdev->sb_start; 1429 /* Limit to 4TB as metadata cannot record more than that. 1430 * 4TB == 2^32 KB, or 2*2^32 sectors. 1431 */ 1432 if (num_sectors >= (2ULL << 32)) 1433 num_sectors = (2ULL << 32) - 2; 1434 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1435 rdev->sb_page); 1436 md_super_wait(rdev->mddev); 1437 return num_sectors; 1438 } 1439 1440 1441 /* 1442 * version 1 superblock 1443 */ 1444 1445 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) 1446 { 1447 __le32 disk_csum; 1448 u32 csum; 1449 unsigned long long newcsum; 1450 int size = 256 + le32_to_cpu(sb->max_dev)*2; 1451 __le32 *isuper = (__le32*)sb; 1452 int i; 1453 1454 disk_csum = sb->sb_csum; 1455 sb->sb_csum = 0; 1456 newcsum = 0; 1457 for (i=0; size>=4; size -= 4 ) 1458 newcsum += le32_to_cpu(*isuper++); 1459 1460 if (size == 2) 1461 newcsum += le16_to_cpu(*(__le16*) isuper); 1462 1463 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 1464 sb->sb_csum = disk_csum; 1465 return cpu_to_le32(csum); 1466 } 1467 1468 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, 1469 int acknowledged); 1470 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) 1471 { 1472 struct mdp_superblock_1 *sb; 1473 int ret; 1474 sector_t sb_start; 1475 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1476 int bmask; 1477 1478 /* 1479 * Calculate the position of the superblock in 512byte sectors. 1480 * It is always aligned to a 4K boundary and 1481 * depeding on minor_version, it can be: 1482 * 0: At least 8K, but less than 12K, from end of device 1483 * 1: At start of device 1484 * 2: 4K from start of device. 1485 */ 1486 switch(minor_version) { 1487 case 0: 1488 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; 1489 sb_start -= 8*2; 1490 sb_start &= ~(sector_t)(4*2-1); 1491 break; 1492 case 1: 1493 sb_start = 0; 1494 break; 1495 case 2: 1496 sb_start = 8; 1497 break; 1498 default: 1499 return -EINVAL; 1500 } 1501 rdev->sb_start = sb_start; 1502 1503 /* superblock is rarely larger than 1K, but it can be larger, 1504 * and it is safe to read 4k, so we do that 1505 */ 1506 ret = read_disk_sb(rdev, 4096); 1507 if (ret) return ret; 1508 1509 1510 sb = page_address(rdev->sb_page); 1511 1512 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 1513 sb->major_version != cpu_to_le32(1) || 1514 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 1515 le64_to_cpu(sb->super_offset) != rdev->sb_start || 1516 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 1517 return -EINVAL; 1518 1519 if (calc_sb_1_csum(sb) != sb->sb_csum) { 1520 printk("md: invalid superblock checksum on %s\n", 1521 bdevname(rdev->bdev,b)); 1522 return -EINVAL; 1523 } 1524 if (le64_to_cpu(sb->data_size) < 10) { 1525 printk("md: data_size too small on %s\n", 1526 bdevname(rdev->bdev,b)); 1527 return -EINVAL; 1528 } 1529 1530 rdev->preferred_minor = 0xffff; 1531 rdev->data_offset = le64_to_cpu(sb->data_offset); 1532 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); 1533 1534 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 1535 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1536 if (rdev->sb_size & bmask) 1537 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1538 1539 if (minor_version 1540 && rdev->data_offset < sb_start + (rdev->sb_size/512)) 1541 return -EINVAL; 1542 1543 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) 1544 rdev->desc_nr = -1; 1545 else 1546 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1547 1548 if (!rdev->bb_page) { 1549 rdev->bb_page = alloc_page(GFP_KERNEL); 1550 if (!rdev->bb_page) 1551 return -ENOMEM; 1552 } 1553 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && 1554 rdev->badblocks.count == 0) { 1555 /* need to load the bad block list. 1556 * Currently we limit it to one page. 1557 */ 1558 s32 offset; 1559 sector_t bb_sector; 1560 u64 *bbp; 1561 int i; 1562 int sectors = le16_to_cpu(sb->bblog_size); 1563 if (sectors > (PAGE_SIZE / 512)) 1564 return -EINVAL; 1565 offset = le32_to_cpu(sb->bblog_offset); 1566 if (offset == 0) 1567 return -EINVAL; 1568 bb_sector = (long long)offset; 1569 if (!sync_page_io(rdev, bb_sector, sectors << 9, 1570 rdev->bb_page, READ, true)) 1571 return -EIO; 1572 bbp = (u64 *)page_address(rdev->bb_page); 1573 rdev->badblocks.shift = sb->bblog_shift; 1574 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { 1575 u64 bb = le64_to_cpu(*bbp); 1576 int count = bb & (0x3ff); 1577 u64 sector = bb >> 10; 1578 sector <<= sb->bblog_shift; 1579 count <<= sb->bblog_shift; 1580 if (bb + 1 == 0) 1581 break; 1582 if (md_set_badblocks(&rdev->badblocks, 1583 sector, count, 1) == 0) 1584 return -EINVAL; 1585 } 1586 } else if (sb->bblog_offset == 0) 1587 rdev->badblocks.shift = -1; 1588 1589 if (!refdev) { 1590 ret = 1; 1591 } else { 1592 __u64 ev1, ev2; 1593 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); 1594 1595 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 1596 sb->level != refsb->level || 1597 sb->layout != refsb->layout || 1598 sb->chunksize != refsb->chunksize) { 1599 printk(KERN_WARNING "md: %s has strangely different" 1600 " superblock to %s\n", 1601 bdevname(rdev->bdev,b), 1602 bdevname(refdev->bdev,b2)); 1603 return -EINVAL; 1604 } 1605 ev1 = le64_to_cpu(sb->events); 1606 ev2 = le64_to_cpu(refsb->events); 1607 1608 if (ev1 > ev2) 1609 ret = 1; 1610 else 1611 ret = 0; 1612 } 1613 if (minor_version) 1614 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 1615 le64_to_cpu(sb->data_offset); 1616 else 1617 rdev->sectors = rdev->sb_start; 1618 if (rdev->sectors < le64_to_cpu(sb->data_size)) 1619 return -EINVAL; 1620 rdev->sectors = le64_to_cpu(sb->data_size); 1621 if (le64_to_cpu(sb->size) > rdev->sectors) 1622 return -EINVAL; 1623 return ret; 1624 } 1625 1626 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) 1627 { 1628 struct mdp_superblock_1 *sb = page_address(rdev->sb_page); 1629 __u64 ev1 = le64_to_cpu(sb->events); 1630 1631 rdev->raid_disk = -1; 1632 clear_bit(Faulty, &rdev->flags); 1633 clear_bit(In_sync, &rdev->flags); 1634 clear_bit(WriteMostly, &rdev->flags); 1635 1636 if (mddev->raid_disks == 0) { 1637 mddev->major_version = 1; 1638 mddev->patch_version = 0; 1639 mddev->external = 0; 1640 mddev->chunk_sectors = le32_to_cpu(sb->chunksize); 1641 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1642 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1643 mddev->level = le32_to_cpu(sb->level); 1644 mddev->clevel[0] = 0; 1645 mddev->layout = le32_to_cpu(sb->layout); 1646 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1647 mddev->dev_sectors = le64_to_cpu(sb->size); 1648 mddev->events = ev1; 1649 mddev->bitmap_info.offset = 0; 1650 mddev->bitmap_info.default_offset = 1024 >> 9; 1651 1652 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1653 memcpy(mddev->uuid, sb->set_uuid, 16); 1654 1655 mddev->max_disks = (4096-256)/2; 1656 1657 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1658 mddev->bitmap_info.file == NULL ) 1659 mddev->bitmap_info.offset = 1660 (__s32)le32_to_cpu(sb->bitmap_offset); 1661 1662 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { 1663 mddev->reshape_position = le64_to_cpu(sb->reshape_position); 1664 mddev->delta_disks = le32_to_cpu(sb->delta_disks); 1665 mddev->new_level = le32_to_cpu(sb->new_level); 1666 mddev->new_layout = le32_to_cpu(sb->new_layout); 1667 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); 1668 } else { 1669 mddev->reshape_position = MaxSector; 1670 mddev->delta_disks = 0; 1671 mddev->new_level = mddev->level; 1672 mddev->new_layout = mddev->layout; 1673 mddev->new_chunk_sectors = mddev->chunk_sectors; 1674 } 1675 1676 } else if (mddev->pers == NULL) { 1677 /* Insist of good event counter while assembling, except for 1678 * spares (which don't need an event count) */ 1679 ++ev1; 1680 if (rdev->desc_nr >= 0 && 1681 rdev->desc_nr < le32_to_cpu(sb->max_dev) && 1682 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) 1683 if (ev1 < mddev->events) 1684 return -EINVAL; 1685 } else if (mddev->bitmap) { 1686 /* If adding to array with a bitmap, then we can accept an 1687 * older device, but not too old. 1688 */ 1689 if (ev1 < mddev->bitmap->events_cleared) 1690 return 0; 1691 } else { 1692 if (ev1 < mddev->events) 1693 /* just a hot-add of a new device, leave raid_disk at -1 */ 1694 return 0; 1695 } 1696 if (mddev->level != LEVEL_MULTIPATH) { 1697 int role; 1698 if (rdev->desc_nr < 0 || 1699 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { 1700 role = 0xffff; 1701 rdev->desc_nr = -1; 1702 } else 1703 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1704 switch(role) { 1705 case 0xffff: /* spare */ 1706 break; 1707 case 0xfffe: /* faulty */ 1708 set_bit(Faulty, &rdev->flags); 1709 break; 1710 default: 1711 if ((le32_to_cpu(sb->feature_map) & 1712 MD_FEATURE_RECOVERY_OFFSET)) 1713 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); 1714 else 1715 set_bit(In_sync, &rdev->flags); 1716 rdev->raid_disk = role; 1717 break; 1718 } 1719 if (sb->devflags & WriteMostly1) 1720 set_bit(WriteMostly, &rdev->flags); 1721 } else /* MULTIPATH are always insync */ 1722 set_bit(In_sync, &rdev->flags); 1723 1724 return 0; 1725 } 1726 1727 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) 1728 { 1729 struct mdp_superblock_1 *sb; 1730 struct md_rdev *rdev2; 1731 int max_dev, i; 1732 /* make rdev->sb match mddev and rdev data. */ 1733 1734 sb = page_address(rdev->sb_page); 1735 1736 sb->feature_map = 0; 1737 sb->pad0 = 0; 1738 sb->recovery_offset = cpu_to_le64(0); 1739 memset(sb->pad1, 0, sizeof(sb->pad1)); 1740 memset(sb->pad3, 0, sizeof(sb->pad3)); 1741 1742 sb->utime = cpu_to_le64((__u64)mddev->utime); 1743 sb->events = cpu_to_le64(mddev->events); 1744 if (mddev->in_sync) 1745 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1746 else 1747 sb->resync_offset = cpu_to_le64(0); 1748 1749 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); 1750 1751 sb->raid_disks = cpu_to_le32(mddev->raid_disks); 1752 sb->size = cpu_to_le64(mddev->dev_sectors); 1753 sb->chunksize = cpu_to_le32(mddev->chunk_sectors); 1754 sb->level = cpu_to_le32(mddev->level); 1755 sb->layout = cpu_to_le32(mddev->layout); 1756 1757 if (test_bit(WriteMostly, &rdev->flags)) 1758 sb->devflags |= WriteMostly1; 1759 else 1760 sb->devflags &= ~WriteMostly1; 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 1775 if (mddev->reshape_position != MaxSector) { 1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); 1777 sb->reshape_position = cpu_to_le64(mddev->reshape_position); 1778 sb->new_layout = cpu_to_le32(mddev->new_layout); 1779 sb->delta_disks = cpu_to_le32(mddev->delta_disks); 1780 sb->new_level = cpu_to_le32(mddev->new_level); 1781 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); 1782 } 1783 1784 if (rdev->badblocks.count == 0) 1785 /* Nothing to do for bad blocks*/ ; 1786 else if (sb->bblog_offset == 0) 1787 /* Cannot record bad blocks on this device */ 1788 md_error(mddev, rdev); 1789 else { 1790 struct badblocks *bb = &rdev->badblocks; 1791 u64 *bbp = (u64 *)page_address(rdev->bb_page); 1792 u64 *p = bb->page; 1793 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); 1794 if (bb->changed) { 1795 unsigned seq; 1796 1797 retry: 1798 seq = read_seqbegin(&bb->lock); 1799 1800 memset(bbp, 0xff, PAGE_SIZE); 1801 1802 for (i = 0 ; i < bb->count ; i++) { 1803 u64 internal_bb = *p++; 1804 u64 store_bb = ((BB_OFFSET(internal_bb) << 10) 1805 | BB_LEN(internal_bb)); 1806 *bbp++ = cpu_to_le64(store_bb); 1807 } 1808 if (read_seqretry(&bb->lock, seq)) 1809 goto retry; 1810 1811 bb->sector = (rdev->sb_start + 1812 (int)le32_to_cpu(sb->bblog_offset)); 1813 bb->size = le16_to_cpu(sb->bblog_size); 1814 bb->changed = 0; 1815 } 1816 } 1817 1818 max_dev = 0; 1819 list_for_each_entry(rdev2, &mddev->disks, same_set) 1820 if (rdev2->desc_nr+1 > max_dev) 1821 max_dev = rdev2->desc_nr+1; 1822 1823 if (max_dev > le32_to_cpu(sb->max_dev)) { 1824 int bmask; 1825 sb->max_dev = cpu_to_le32(max_dev); 1826 rdev->sb_size = max_dev * 2 + 256; 1827 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1828 if (rdev->sb_size & bmask) 1829 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1830 } else 1831 max_dev = le32_to_cpu(sb->max_dev); 1832 1833 for (i=0; i<max_dev;i++) 1834 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1835 1836 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1837 i = rdev2->desc_nr; 1838 if (test_bit(Faulty, &rdev2->flags)) 1839 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1840 else if (test_bit(In_sync, &rdev2->flags)) 1841 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1842 else if (rdev2->raid_disk >= 0) 1843 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1844 else 1845 sb->dev_roles[i] = cpu_to_le16(0xffff); 1846 } 1847 1848 sb->sb_csum = calc_sb_1_csum(sb); 1849 } 1850 1851 static unsigned long long 1852 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) 1853 { 1854 struct mdp_superblock_1 *sb; 1855 sector_t max_sectors; 1856 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1857 return 0; /* component must fit device */ 1858 if (rdev->sb_start < rdev->data_offset) { 1859 /* minor versions 1 and 2; superblock before data */ 1860 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; 1861 max_sectors -= rdev->data_offset; 1862 if (!num_sectors || num_sectors > max_sectors) 1863 num_sectors = max_sectors; 1864 } else if (rdev->mddev->bitmap_info.offset) { 1865 /* minor version 0 with bitmap we can't move */ 1866 return 0; 1867 } else { 1868 /* minor version 0; superblock after data */ 1869 sector_t sb_start; 1870 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; 1871 sb_start &= ~(sector_t)(4*2 - 1); 1872 max_sectors = rdev->sectors + sb_start - rdev->sb_start; 1873 if (!num_sectors || num_sectors > max_sectors) 1874 num_sectors = max_sectors; 1875 rdev->sb_start = sb_start; 1876 } 1877 sb = page_address(rdev->sb_page); 1878 sb->data_size = cpu_to_le64(num_sectors); 1879 sb->super_offset = rdev->sb_start; 1880 sb->sb_csum = calc_sb_1_csum(sb); 1881 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1882 rdev->sb_page); 1883 md_super_wait(rdev->mddev); 1884 return num_sectors; 1885 } 1886 1887 static struct super_type super_types[] = { 1888 [0] = { 1889 .name = "0.90.0", 1890 .owner = THIS_MODULE, 1891 .load_super = super_90_load, 1892 .validate_super = super_90_validate, 1893 .sync_super = super_90_sync, 1894 .rdev_size_change = super_90_rdev_size_change, 1895 }, 1896 [1] = { 1897 .name = "md-1", 1898 .owner = THIS_MODULE, 1899 .load_super = super_1_load, 1900 .validate_super = super_1_validate, 1901 .sync_super = super_1_sync, 1902 .rdev_size_change = super_1_rdev_size_change, 1903 }, 1904 }; 1905 1906 static void sync_super(struct mddev *mddev, struct md_rdev *rdev) 1907 { 1908 if (mddev->sync_super) { 1909 mddev->sync_super(mddev, rdev); 1910 return; 1911 } 1912 1913 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); 1914 1915 super_types[mddev->major_version].sync_super(mddev, rdev); 1916 } 1917 1918 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) 1919 { 1920 struct md_rdev *rdev, *rdev2; 1921 1922 rcu_read_lock(); 1923 rdev_for_each_rcu(rdev, mddev1) 1924 rdev_for_each_rcu(rdev2, mddev2) 1925 if (rdev->bdev->bd_contains == 1926 rdev2->bdev->bd_contains) { 1927 rcu_read_unlock(); 1928 return 1; 1929 } 1930 rcu_read_unlock(); 1931 return 0; 1932 } 1933 1934 static LIST_HEAD(pending_raid_disks); 1935 1936 /* 1937 * Try to register data integrity profile for an mddev 1938 * 1939 * This is called when an array is started and after a disk has been kicked 1940 * from the array. It only succeeds if all working and active component devices 1941 * are integrity capable with matching profiles. 1942 */ 1943 int md_integrity_register(struct mddev *mddev) 1944 { 1945 struct md_rdev *rdev, *reference = NULL; 1946 1947 if (list_empty(&mddev->disks)) 1948 return 0; /* nothing to do */ 1949 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) 1950 return 0; /* shouldn't register, or already is */ 1951 list_for_each_entry(rdev, &mddev->disks, same_set) { 1952 /* skip spares and non-functional disks */ 1953 if (test_bit(Faulty, &rdev->flags)) 1954 continue; 1955 if (rdev->raid_disk < 0) 1956 continue; 1957 if (!reference) { 1958 /* Use the first rdev as the reference */ 1959 reference = rdev; 1960 continue; 1961 } 1962 /* does this rdev's profile match the reference profile? */ 1963 if (blk_integrity_compare(reference->bdev->bd_disk, 1964 rdev->bdev->bd_disk) < 0) 1965 return -EINVAL; 1966 } 1967 if (!reference || !bdev_get_integrity(reference->bdev)) 1968 return 0; 1969 /* 1970 * All component devices are integrity capable and have matching 1971 * profiles, register the common profile for the md device. 1972 */ 1973 if (blk_integrity_register(mddev->gendisk, 1974 bdev_get_integrity(reference->bdev)) != 0) { 1975 printk(KERN_ERR "md: failed to register integrity for %s\n", 1976 mdname(mddev)); 1977 return -EINVAL; 1978 } 1979 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); 1980 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { 1981 printk(KERN_ERR "md: failed to create integrity pool for %s\n", 1982 mdname(mddev)); 1983 return -EINVAL; 1984 } 1985 return 0; 1986 } 1987 EXPORT_SYMBOL(md_integrity_register); 1988 1989 /* Disable data integrity if non-capable/non-matching disk is being added */ 1990 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) 1991 { 1992 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); 1993 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); 1994 1995 if (!bi_mddev) /* nothing to do */ 1996 return; 1997 if (rdev->raid_disk < 0) /* skip spares */ 1998 return; 1999 if (bi_rdev && blk_integrity_compare(mddev->gendisk, 2000 rdev->bdev->bd_disk) >= 0) 2001 return; 2002 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); 2003 blk_integrity_unregister(mddev->gendisk); 2004 } 2005 EXPORT_SYMBOL(md_integrity_add_rdev); 2006 2007 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev) 2008 { 2009 char b[BDEVNAME_SIZE]; 2010 struct kobject *ko; 2011 char *s; 2012 int err; 2013 2014 if (rdev->mddev) { 2015 MD_BUG(); 2016 return -EINVAL; 2017 } 2018 2019 /* prevent duplicates */ 2020 if (find_rdev(mddev, rdev->bdev->bd_dev)) 2021 return -EEXIST; 2022 2023 /* make sure rdev->sectors exceeds mddev->dev_sectors */ 2024 if (rdev->sectors && (mddev->dev_sectors == 0 || 2025 rdev->sectors < mddev->dev_sectors)) { 2026 if (mddev->pers) { 2027 /* Cannot change size, so fail 2028 * If mddev->level <= 0, then we don't care 2029 * about aligning sizes (e.g. linear) 2030 */ 2031 if (mddev->level > 0) 2032 return -ENOSPC; 2033 } else 2034 mddev->dev_sectors = rdev->sectors; 2035 } 2036 2037 /* Verify rdev->desc_nr is unique. 2038 * If it is -1, assign a free number, else 2039 * check number is not in use 2040 */ 2041 if (rdev->desc_nr < 0) { 2042 int choice = 0; 2043 if (mddev->pers) choice = mddev->raid_disks; 2044 while (find_rdev_nr(mddev, choice)) 2045 choice++; 2046 rdev->desc_nr = choice; 2047 } else { 2048 if (find_rdev_nr(mddev, rdev->desc_nr)) 2049 return -EBUSY; 2050 } 2051 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { 2052 printk(KERN_WARNING "md: %s: array is limited to %d devices\n", 2053 mdname(mddev), mddev->max_disks); 2054 return -EBUSY; 2055 } 2056 bdevname(rdev->bdev,b); 2057 while ( (s=strchr(b, '/')) != NULL) 2058 *s = '!'; 2059 2060 rdev->mddev = mddev; 2061 printk(KERN_INFO "md: bind<%s>\n", b); 2062 2063 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) 2064 goto fail; 2065 2066 ko = &part_to_dev(rdev->bdev->bd_part)->kobj; 2067 if (sysfs_create_link(&rdev->kobj, ko, "block")) 2068 /* failure here is OK */; 2069 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); 2070 2071 list_add_rcu(&rdev->same_set, &mddev->disks); 2072 bd_link_disk_holder(rdev->bdev, mddev->gendisk); 2073 2074 /* May as well allow recovery to be retried once */ 2075 mddev->recovery_disabled++; 2076 2077 return 0; 2078 2079 fail: 2080 printk(KERN_WARNING "md: failed to register dev-%s for %s\n", 2081 b, mdname(mddev)); 2082 return err; 2083 } 2084 2085 static void md_delayed_delete(struct work_struct *ws) 2086 { 2087 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); 2088 kobject_del(&rdev->kobj); 2089 kobject_put(&rdev->kobj); 2090 } 2091 2092 static void unbind_rdev_from_array(struct md_rdev * rdev) 2093 { 2094 char b[BDEVNAME_SIZE]; 2095 if (!rdev->mddev) { 2096 MD_BUG(); 2097 return; 2098 } 2099 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); 2100 list_del_rcu(&rdev->same_set); 2101 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 2102 rdev->mddev = NULL; 2103 sysfs_remove_link(&rdev->kobj, "block"); 2104 sysfs_put(rdev->sysfs_state); 2105 rdev->sysfs_state = NULL; 2106 kfree(rdev->badblocks.page); 2107 rdev->badblocks.count = 0; 2108 rdev->badblocks.page = NULL; 2109 /* We need to delay this, otherwise we can deadlock when 2110 * writing to 'remove' to "dev/state". We also need 2111 * to delay it due to rcu usage. 2112 */ 2113 synchronize_rcu(); 2114 INIT_WORK(&rdev->del_work, md_delayed_delete); 2115 kobject_get(&rdev->kobj); 2116 queue_work(md_misc_wq, &rdev->del_work); 2117 } 2118 2119 /* 2120 * prevent the device from being mounted, repartitioned or 2121 * otherwise reused by a RAID array (or any other kernel 2122 * subsystem), by bd_claiming the device. 2123 */ 2124 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) 2125 { 2126 int err = 0; 2127 struct block_device *bdev; 2128 char b[BDEVNAME_SIZE]; 2129 2130 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, 2131 shared ? (struct md_rdev *)lock_rdev : rdev); 2132 if (IS_ERR(bdev)) { 2133 printk(KERN_ERR "md: could not open %s.\n", 2134 __bdevname(dev, b)); 2135 return PTR_ERR(bdev); 2136 } 2137 rdev->bdev = bdev; 2138 return err; 2139 } 2140 2141 static void unlock_rdev(struct md_rdev *rdev) 2142 { 2143 struct block_device *bdev = rdev->bdev; 2144 rdev->bdev = NULL; 2145 if (!bdev) 2146 MD_BUG(); 2147 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2148 } 2149 2150 void md_autodetect_dev(dev_t dev); 2151 2152 static void export_rdev(struct md_rdev * rdev) 2153 { 2154 char b[BDEVNAME_SIZE]; 2155 printk(KERN_INFO "md: export_rdev(%s)\n", 2156 bdevname(rdev->bdev,b)); 2157 if (rdev->mddev) 2158 MD_BUG(); 2159 free_disk_sb(rdev); 2160 #ifndef MODULE 2161 if (test_bit(AutoDetected, &rdev->flags)) 2162 md_autodetect_dev(rdev->bdev->bd_dev); 2163 #endif 2164 unlock_rdev(rdev); 2165 kobject_put(&rdev->kobj); 2166 } 2167 2168 static void kick_rdev_from_array(struct md_rdev * rdev) 2169 { 2170 unbind_rdev_from_array(rdev); 2171 export_rdev(rdev); 2172 } 2173 2174 static void export_array(struct mddev *mddev) 2175 { 2176 struct md_rdev *rdev, *tmp; 2177 2178 rdev_for_each(rdev, tmp, mddev) { 2179 if (!rdev->mddev) { 2180 MD_BUG(); 2181 continue; 2182 } 2183 kick_rdev_from_array(rdev); 2184 } 2185 if (!list_empty(&mddev->disks)) 2186 MD_BUG(); 2187 mddev->raid_disks = 0; 2188 mddev->major_version = 0; 2189 } 2190 2191 static void print_desc(mdp_disk_t *desc) 2192 { 2193 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 2194 desc->major,desc->minor,desc->raid_disk,desc->state); 2195 } 2196 2197 static void print_sb_90(mdp_super_t *sb) 2198 { 2199 int i; 2200 2201 printk(KERN_INFO 2202 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 2203 sb->major_version, sb->minor_version, sb->patch_version, 2204 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 2205 sb->ctime); 2206 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 2207 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 2208 sb->md_minor, sb->layout, sb->chunk_size); 2209 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 2210 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 2211 sb->utime, sb->state, sb->active_disks, sb->working_disks, 2212 sb->failed_disks, sb->spare_disks, 2213 sb->sb_csum, (unsigned long)sb->events_lo); 2214 2215 printk(KERN_INFO); 2216 for (i = 0; i < MD_SB_DISKS; i++) { 2217 mdp_disk_t *desc; 2218 2219 desc = sb->disks + i; 2220 if (desc->number || desc->major || desc->minor || 2221 desc->raid_disk || (desc->state && (desc->state != 4))) { 2222 printk(" D %2d: ", i); 2223 print_desc(desc); 2224 } 2225 } 2226 printk(KERN_INFO "md: THIS: "); 2227 print_desc(&sb->this_disk); 2228 } 2229 2230 static void print_sb_1(struct mdp_superblock_1 *sb) 2231 { 2232 __u8 *uuid; 2233 2234 uuid = sb->set_uuid; 2235 printk(KERN_INFO 2236 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" 2237 "md: Name: \"%s\" CT:%llu\n", 2238 le32_to_cpu(sb->major_version), 2239 le32_to_cpu(sb->feature_map), 2240 uuid, 2241 sb->set_name, 2242 (unsigned long long)le64_to_cpu(sb->ctime) 2243 & MD_SUPERBLOCK_1_TIME_SEC_MASK); 2244 2245 uuid = sb->device_uuid; 2246 printk(KERN_INFO 2247 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" 2248 " RO:%llu\n" 2249 "md: Dev:%08x UUID: %pU\n" 2250 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" 2251 "md: (MaxDev:%u) \n", 2252 le32_to_cpu(sb->level), 2253 (unsigned long long)le64_to_cpu(sb->size), 2254 le32_to_cpu(sb->raid_disks), 2255 le32_to_cpu(sb->layout), 2256 le32_to_cpu(sb->chunksize), 2257 (unsigned long long)le64_to_cpu(sb->data_offset), 2258 (unsigned long long)le64_to_cpu(sb->data_size), 2259 (unsigned long long)le64_to_cpu(sb->super_offset), 2260 (unsigned long long)le64_to_cpu(sb->recovery_offset), 2261 le32_to_cpu(sb->dev_number), 2262 uuid, 2263 sb->devflags, 2264 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, 2265 (unsigned long long)le64_to_cpu(sb->events), 2266 (unsigned long long)le64_to_cpu(sb->resync_offset), 2267 le32_to_cpu(sb->sb_csum), 2268 le32_to_cpu(sb->max_dev) 2269 ); 2270 } 2271 2272 static void print_rdev(struct md_rdev *rdev, int major_version) 2273 { 2274 char b[BDEVNAME_SIZE]; 2275 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", 2276 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, 2277 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 2278 rdev->desc_nr); 2279 if (rdev->sb_loaded) { 2280 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); 2281 switch (major_version) { 2282 case 0: 2283 print_sb_90(page_address(rdev->sb_page)); 2284 break; 2285 case 1: 2286 print_sb_1(page_address(rdev->sb_page)); 2287 break; 2288 } 2289 } else 2290 printk(KERN_INFO "md: no rdev superblock!\n"); 2291 } 2292 2293 static void md_print_devices(void) 2294 { 2295 struct list_head *tmp; 2296 struct md_rdev *rdev; 2297 struct mddev *mddev; 2298 char b[BDEVNAME_SIZE]; 2299 2300 printk("\n"); 2301 printk("md: **********************************\n"); 2302 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 2303 printk("md: **********************************\n"); 2304 for_each_mddev(mddev, tmp) { 2305 2306 if (mddev->bitmap) 2307 bitmap_print_sb(mddev->bitmap); 2308 else 2309 printk("%s: ", mdname(mddev)); 2310 list_for_each_entry(rdev, &mddev->disks, same_set) 2311 printk("<%s>", bdevname(rdev->bdev,b)); 2312 printk("\n"); 2313 2314 list_for_each_entry(rdev, &mddev->disks, same_set) 2315 print_rdev(rdev, mddev->major_version); 2316 } 2317 printk("md: **********************************\n"); 2318 printk("\n"); 2319 } 2320 2321 2322 static void sync_sbs(struct mddev * mddev, int nospares) 2323 { 2324 /* Update each superblock (in-memory image), but 2325 * if we are allowed to, skip spares which already 2326 * have the right event counter, or have one earlier 2327 * (which would mean they aren't being marked as dirty 2328 * with the rest of the array) 2329 */ 2330 struct md_rdev *rdev; 2331 list_for_each_entry(rdev, &mddev->disks, same_set) { 2332 if (rdev->sb_events == mddev->events || 2333 (nospares && 2334 rdev->raid_disk < 0 && 2335 rdev->sb_events+1 == mddev->events)) { 2336 /* Don't update this superblock */ 2337 rdev->sb_loaded = 2; 2338 } else { 2339 sync_super(mddev, rdev); 2340 rdev->sb_loaded = 1; 2341 } 2342 } 2343 } 2344 2345 static void md_update_sb(struct mddev * mddev, int force_change) 2346 { 2347 struct md_rdev *rdev; 2348 int sync_req; 2349 int nospares = 0; 2350 int any_badblocks_changed = 0; 2351 2352 repeat: 2353 /* First make sure individual recovery_offsets are correct */ 2354 list_for_each_entry(rdev, &mddev->disks, same_set) { 2355 if (rdev->raid_disk >= 0 && 2356 mddev->delta_disks >= 0 && 2357 !test_bit(In_sync, &rdev->flags) && 2358 mddev->curr_resync_completed > rdev->recovery_offset) 2359 rdev->recovery_offset = mddev->curr_resync_completed; 2360 2361 } 2362 if (!mddev->persistent) { 2363 clear_bit(MD_CHANGE_CLEAN, &mddev->flags); 2364 clear_bit(MD_CHANGE_DEVS, &mddev->flags); 2365 if (!mddev->external) { 2366 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2367 list_for_each_entry(rdev, &mddev->disks, same_set) { 2368 if (rdev->badblocks.changed) { 2369 md_ack_all_badblocks(&rdev->badblocks); 2370 md_error(mddev, rdev); 2371 } 2372 clear_bit(Blocked, &rdev->flags); 2373 clear_bit(BlockedBadBlocks, &rdev->flags); 2374 wake_up(&rdev->blocked_wait); 2375 } 2376 } 2377 wake_up(&mddev->sb_wait); 2378 return; 2379 } 2380 2381 spin_lock_irq(&mddev->write_lock); 2382 2383 mddev->utime = get_seconds(); 2384 2385 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) 2386 force_change = 1; 2387 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) 2388 /* just a clean<-> dirty transition, possibly leave spares alone, 2389 * though if events isn't the right even/odd, we will have to do 2390 * spares after all 2391 */ 2392 nospares = 1; 2393 if (force_change) 2394 nospares = 0; 2395 if (mddev->degraded) 2396 /* If the array is degraded, then skipping spares is both 2397 * dangerous and fairly pointless. 2398 * Dangerous because a device that was removed from the array 2399 * might have a event_count that still looks up-to-date, 2400 * so it can be re-added without a resync. 2401 * Pointless because if there are any spares to skip, 2402 * then a recovery will happen and soon that array won't 2403 * be degraded any more and the spare can go back to sleep then. 2404 */ 2405 nospares = 0; 2406 2407 sync_req = mddev->in_sync; 2408 2409 /* If this is just a dirty<->clean transition, and the array is clean 2410 * and 'events' is odd, we can roll back to the previous clean state */ 2411 if (nospares 2412 && (mddev->in_sync && mddev->recovery_cp == MaxSector) 2413 && mddev->can_decrease_events 2414 && mddev->events != 1) { 2415 mddev->events--; 2416 mddev->can_decrease_events = 0; 2417 } else { 2418 /* otherwise we have to go forward and ... */ 2419 mddev->events ++; 2420 mddev->can_decrease_events = nospares; 2421 } 2422 2423 if (!mddev->events) { 2424 /* 2425 * oops, this 64-bit counter should never wrap. 2426 * Either we are in around ~1 trillion A.C., assuming 2427 * 1 reboot per second, or we have a bug: 2428 */ 2429 MD_BUG(); 2430 mddev->events --; 2431 } 2432 2433 list_for_each_entry(rdev, &mddev->disks, same_set) { 2434 if (rdev->badblocks.changed) 2435 any_badblocks_changed++; 2436 if (test_bit(Faulty, &rdev->flags)) 2437 set_bit(FaultRecorded, &rdev->flags); 2438 } 2439 2440 sync_sbs(mddev, nospares); 2441 spin_unlock_irq(&mddev->write_lock); 2442 2443 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", 2444 mdname(mddev), mddev->in_sync); 2445 2446 bitmap_update_sb(mddev->bitmap); 2447 list_for_each_entry(rdev, &mddev->disks, same_set) { 2448 char b[BDEVNAME_SIZE]; 2449 2450 if (rdev->sb_loaded != 1) 2451 continue; /* no noise on spare devices */ 2452 2453 if (!test_bit(Faulty, &rdev->flags) && 2454 rdev->saved_raid_disk == -1) { 2455 md_super_write(mddev,rdev, 2456 rdev->sb_start, rdev->sb_size, 2457 rdev->sb_page); 2458 pr_debug("md: (write) %s's sb offset: %llu\n", 2459 bdevname(rdev->bdev, b), 2460 (unsigned long long)rdev->sb_start); 2461 rdev->sb_events = mddev->events; 2462 if (rdev->badblocks.size) { 2463 md_super_write(mddev, rdev, 2464 rdev->badblocks.sector, 2465 rdev->badblocks.size << 9, 2466 rdev->bb_page); 2467 rdev->badblocks.size = 0; 2468 } 2469 2470 } else if (test_bit(Faulty, &rdev->flags)) 2471 pr_debug("md: %s (skipping faulty)\n", 2472 bdevname(rdev->bdev, b)); 2473 else 2474 pr_debug("(skipping incremental s/r "); 2475 2476 if (mddev->level == LEVEL_MULTIPATH) 2477 /* only need to write one superblock... */ 2478 break; 2479 } 2480 md_super_wait(mddev); 2481 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ 2482 2483 spin_lock_irq(&mddev->write_lock); 2484 if (mddev->in_sync != sync_req || 2485 test_bit(MD_CHANGE_DEVS, &mddev->flags)) { 2486 /* have to write it out again */ 2487 spin_unlock_irq(&mddev->write_lock); 2488 goto repeat; 2489 } 2490 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2491 spin_unlock_irq(&mddev->write_lock); 2492 wake_up(&mddev->sb_wait); 2493 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 2494 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 2495 2496 list_for_each_entry(rdev, &mddev->disks, same_set) { 2497 if (test_and_clear_bit(FaultRecorded, &rdev->flags)) 2498 clear_bit(Blocked, &rdev->flags); 2499 2500 if (any_badblocks_changed) 2501 md_ack_all_badblocks(&rdev->badblocks); 2502 clear_bit(BlockedBadBlocks, &rdev->flags); 2503 wake_up(&rdev->blocked_wait); 2504 } 2505 } 2506 2507 /* words written to sysfs files may, or may not, be \n terminated. 2508 * We want to accept with case. For this we use cmd_match. 2509 */ 2510 static int cmd_match(const char *cmd, const char *str) 2511 { 2512 /* See if cmd, written into a sysfs file, matches 2513 * str. They must either be the same, or cmd can 2514 * have a trailing newline 2515 */ 2516 while (*cmd && *str && *cmd == *str) { 2517 cmd++; 2518 str++; 2519 } 2520 if (*cmd == '\n') 2521 cmd++; 2522 if (*str || *cmd) 2523 return 0; 2524 return 1; 2525 } 2526 2527 struct rdev_sysfs_entry { 2528 struct attribute attr; 2529 ssize_t (*show)(struct md_rdev *, char *); 2530 ssize_t (*store)(struct md_rdev *, const char *, size_t); 2531 }; 2532 2533 static ssize_t 2534 state_show(struct md_rdev *rdev, char *page) 2535 { 2536 char *sep = ""; 2537 size_t len = 0; 2538 2539 if (test_bit(Faulty, &rdev->flags) || 2540 rdev->badblocks.unacked_exist) { 2541 len+= sprintf(page+len, "%sfaulty",sep); 2542 sep = ","; 2543 } 2544 if (test_bit(In_sync, &rdev->flags)) { 2545 len += sprintf(page+len, "%sin_sync",sep); 2546 sep = ","; 2547 } 2548 if (test_bit(WriteMostly, &rdev->flags)) { 2549 len += sprintf(page+len, "%swrite_mostly",sep); 2550 sep = ","; 2551 } 2552 if (test_bit(Blocked, &rdev->flags) || 2553 rdev->badblocks.unacked_exist) { 2554 len += sprintf(page+len, "%sblocked", sep); 2555 sep = ","; 2556 } 2557 if (!test_bit(Faulty, &rdev->flags) && 2558 !test_bit(In_sync, &rdev->flags)) { 2559 len += sprintf(page+len, "%sspare", sep); 2560 sep = ","; 2561 } 2562 if (test_bit(WriteErrorSeen, &rdev->flags)) { 2563 len += sprintf(page+len, "%swrite_error", sep); 2564 sep = ","; 2565 } 2566 return len+sprintf(page+len, "\n"); 2567 } 2568 2569 static ssize_t 2570 state_store(struct md_rdev *rdev, const char *buf, size_t len) 2571 { 2572 /* can write 2573 * faulty - simulates an error 2574 * remove - disconnects the device 2575 * writemostly - sets write_mostly 2576 * -writemostly - clears write_mostly 2577 * blocked - sets the Blocked flags 2578 * -blocked - clears the Blocked and possibly simulates an error 2579 * insync - sets Insync providing device isn't active 2580 * write_error - sets WriteErrorSeen 2581 * -write_error - clears WriteErrorSeen 2582 */ 2583 int err = -EINVAL; 2584 if (cmd_match(buf, "faulty") && rdev->mddev->pers) { 2585 md_error(rdev->mddev, rdev); 2586 if (test_bit(Faulty, &rdev->flags)) 2587 err = 0; 2588 else 2589 err = -EBUSY; 2590 } else if (cmd_match(buf, "remove")) { 2591 if (rdev->raid_disk >= 0) 2592 err = -EBUSY; 2593 else { 2594 struct mddev *mddev = rdev->mddev; 2595 kick_rdev_from_array(rdev); 2596 if (mddev->pers) 2597 md_update_sb(mddev, 1); 2598 md_new_event(mddev); 2599 err = 0; 2600 } 2601 } else if (cmd_match(buf, "writemostly")) { 2602 set_bit(WriteMostly, &rdev->flags); 2603 err = 0; 2604 } else if (cmd_match(buf, "-writemostly")) { 2605 clear_bit(WriteMostly, &rdev->flags); 2606 err = 0; 2607 } else if (cmd_match(buf, "blocked")) { 2608 set_bit(Blocked, &rdev->flags); 2609 err = 0; 2610 } else if (cmd_match(buf, "-blocked")) { 2611 if (!test_bit(Faulty, &rdev->flags) && 2612 rdev->badblocks.unacked_exist) { 2613 /* metadata handler doesn't understand badblocks, 2614 * so we need to fail the device 2615 */ 2616 md_error(rdev->mddev, rdev); 2617 } 2618 clear_bit(Blocked, &rdev->flags); 2619 clear_bit(BlockedBadBlocks, &rdev->flags); 2620 wake_up(&rdev->blocked_wait); 2621 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2622 md_wakeup_thread(rdev->mddev->thread); 2623 2624 err = 0; 2625 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { 2626 set_bit(In_sync, &rdev->flags); 2627 err = 0; 2628 } else if (cmd_match(buf, "write_error")) { 2629 set_bit(WriteErrorSeen, &rdev->flags); 2630 err = 0; 2631 } else if (cmd_match(buf, "-write_error")) { 2632 clear_bit(WriteErrorSeen, &rdev->flags); 2633 err = 0; 2634 } 2635 if (!err) 2636 sysfs_notify_dirent_safe(rdev->sysfs_state); 2637 return err ? err : len; 2638 } 2639 static struct rdev_sysfs_entry rdev_state = 2640 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); 2641 2642 static ssize_t 2643 errors_show(struct md_rdev *rdev, char *page) 2644 { 2645 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); 2646 } 2647 2648 static ssize_t 2649 errors_store(struct md_rdev *rdev, const char *buf, size_t len) 2650 { 2651 char *e; 2652 unsigned long n = simple_strtoul(buf, &e, 10); 2653 if (*buf && (*e == 0 || *e == '\n')) { 2654 atomic_set(&rdev->corrected_errors, n); 2655 return len; 2656 } 2657 return -EINVAL; 2658 } 2659 static struct rdev_sysfs_entry rdev_errors = 2660 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); 2661 2662 static ssize_t 2663 slot_show(struct md_rdev *rdev, char *page) 2664 { 2665 if (rdev->raid_disk < 0) 2666 return sprintf(page, "none\n"); 2667 else 2668 return sprintf(page, "%d\n", rdev->raid_disk); 2669 } 2670 2671 static ssize_t 2672 slot_store(struct md_rdev *rdev, const char *buf, size_t len) 2673 { 2674 char *e; 2675 int err; 2676 int slot = simple_strtoul(buf, &e, 10); 2677 if (strncmp(buf, "none", 4)==0) 2678 slot = -1; 2679 else if (e==buf || (*e && *e!= '\n')) 2680 return -EINVAL; 2681 if (rdev->mddev->pers && slot == -1) { 2682 /* Setting 'slot' on an active array requires also 2683 * updating the 'rd%d' link, and communicating 2684 * with the personality with ->hot_*_disk. 2685 * For now we only support removing 2686 * failed/spare devices. This normally happens automatically, 2687 * but not when the metadata is externally managed. 2688 */ 2689 if (rdev->raid_disk == -1) 2690 return -EEXIST; 2691 /* personality does all needed checks */ 2692 if (rdev->mddev->pers->hot_remove_disk == NULL) 2693 return -EINVAL; 2694 err = rdev->mddev->pers-> 2695 hot_remove_disk(rdev->mddev, rdev->raid_disk); 2696 if (err) 2697 return err; 2698 sysfs_unlink_rdev(rdev->mddev, rdev); 2699 rdev->raid_disk = -1; 2700 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2701 md_wakeup_thread(rdev->mddev->thread); 2702 } else if (rdev->mddev->pers) { 2703 struct md_rdev *rdev2; 2704 /* Activating a spare .. or possibly reactivating 2705 * if we ever get bitmaps working here. 2706 */ 2707 2708 if (rdev->raid_disk != -1) 2709 return -EBUSY; 2710 2711 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) 2712 return -EBUSY; 2713 2714 if (rdev->mddev->pers->hot_add_disk == NULL) 2715 return -EINVAL; 2716 2717 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) 2718 if (rdev2->raid_disk == slot) 2719 return -EEXIST; 2720 2721 if (slot >= rdev->mddev->raid_disks && 2722 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) 2723 return -ENOSPC; 2724 2725 rdev->raid_disk = slot; 2726 if (test_bit(In_sync, &rdev->flags)) 2727 rdev->saved_raid_disk = slot; 2728 else 2729 rdev->saved_raid_disk = -1; 2730 clear_bit(In_sync, &rdev->flags); 2731 err = rdev->mddev->pers-> 2732 hot_add_disk(rdev->mddev, rdev); 2733 if (err) { 2734 rdev->raid_disk = -1; 2735 return err; 2736 } else 2737 sysfs_notify_dirent_safe(rdev->sysfs_state); 2738 if (sysfs_link_rdev(rdev->mddev, rdev)) 2739 /* failure here is OK */; 2740 /* don't wakeup anyone, leave that to userspace. */ 2741 } else { 2742 if (slot >= rdev->mddev->raid_disks && 2743 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) 2744 return -ENOSPC; 2745 rdev->raid_disk = slot; 2746 /* assume it is working */ 2747 clear_bit(Faulty, &rdev->flags); 2748 clear_bit(WriteMostly, &rdev->flags); 2749 set_bit(In_sync, &rdev->flags); 2750 sysfs_notify_dirent_safe(rdev->sysfs_state); 2751 } 2752 return len; 2753 } 2754 2755 2756 static struct rdev_sysfs_entry rdev_slot = 2757 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); 2758 2759 static ssize_t 2760 offset_show(struct md_rdev *rdev, char *page) 2761 { 2762 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); 2763 } 2764 2765 static ssize_t 2766 offset_store(struct md_rdev *rdev, const char *buf, size_t len) 2767 { 2768 char *e; 2769 unsigned long long offset = simple_strtoull(buf, &e, 10); 2770 if (e==buf || (*e && *e != '\n')) 2771 return -EINVAL; 2772 if (rdev->mddev->pers && rdev->raid_disk >= 0) 2773 return -EBUSY; 2774 if (rdev->sectors && rdev->mddev->external) 2775 /* Must set offset before size, so overlap checks 2776 * can be sane */ 2777 return -EBUSY; 2778 rdev->data_offset = offset; 2779 return len; 2780 } 2781 2782 static struct rdev_sysfs_entry rdev_offset = 2783 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); 2784 2785 static ssize_t 2786 rdev_size_show(struct md_rdev *rdev, char *page) 2787 { 2788 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); 2789 } 2790 2791 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) 2792 { 2793 /* check if two start/length pairs overlap */ 2794 if (s1+l1 <= s2) 2795 return 0; 2796 if (s2+l2 <= s1) 2797 return 0; 2798 return 1; 2799 } 2800 2801 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) 2802 { 2803 unsigned long long blocks; 2804 sector_t new; 2805 2806 if (strict_strtoull(buf, 10, &blocks) < 0) 2807 return -EINVAL; 2808 2809 if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) 2810 return -EINVAL; /* sector conversion overflow */ 2811 2812 new = blocks * 2; 2813 if (new != blocks * 2) 2814 return -EINVAL; /* unsigned long long to sector_t overflow */ 2815 2816 *sectors = new; 2817 return 0; 2818 } 2819 2820 static ssize_t 2821 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) 2822 { 2823 struct mddev *my_mddev = rdev->mddev; 2824 sector_t oldsectors = rdev->sectors; 2825 sector_t sectors; 2826 2827 if (strict_blocks_to_sectors(buf, §ors) < 0) 2828 return -EINVAL; 2829 if (my_mddev->pers && rdev->raid_disk >= 0) { 2830 if (my_mddev->persistent) { 2831 sectors = super_types[my_mddev->major_version]. 2832 rdev_size_change(rdev, sectors); 2833 if (!sectors) 2834 return -EBUSY; 2835 } else if (!sectors) 2836 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 2837 rdev->data_offset; 2838 } 2839 if (sectors < my_mddev->dev_sectors) 2840 return -EINVAL; /* component must fit device */ 2841 2842 rdev->sectors = sectors; 2843 if (sectors > oldsectors && my_mddev->external) { 2844 /* need to check that all other rdevs with the same ->bdev 2845 * do not overlap. We need to unlock the mddev to avoid 2846 * a deadlock. We have already changed rdev->sectors, and if 2847 * we have to change it back, we will have the lock again. 2848 */ 2849 struct mddev *mddev; 2850 int overlap = 0; 2851 struct list_head *tmp; 2852 2853 mddev_unlock(my_mddev); 2854 for_each_mddev(mddev, tmp) { 2855 struct md_rdev *rdev2; 2856 2857 mddev_lock(mddev); 2858 list_for_each_entry(rdev2, &mddev->disks, same_set) 2859 if (rdev->bdev == rdev2->bdev && 2860 rdev != rdev2 && 2861 overlaps(rdev->data_offset, rdev->sectors, 2862 rdev2->data_offset, 2863 rdev2->sectors)) { 2864 overlap = 1; 2865 break; 2866 } 2867 mddev_unlock(mddev); 2868 if (overlap) { 2869 mddev_put(mddev); 2870 break; 2871 } 2872 } 2873 mddev_lock(my_mddev); 2874 if (overlap) { 2875 /* Someone else could have slipped in a size 2876 * change here, but doing so is just silly. 2877 * We put oldsectors back because we *know* it is 2878 * safe, and trust userspace not to race with 2879 * itself 2880 */ 2881 rdev->sectors = oldsectors; 2882 return -EBUSY; 2883 } 2884 } 2885 return len; 2886 } 2887 2888 static struct rdev_sysfs_entry rdev_size = 2889 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); 2890 2891 2892 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) 2893 { 2894 unsigned long long recovery_start = rdev->recovery_offset; 2895 2896 if (test_bit(In_sync, &rdev->flags) || 2897 recovery_start == MaxSector) 2898 return sprintf(page, "none\n"); 2899 2900 return sprintf(page, "%llu\n", recovery_start); 2901 } 2902 2903 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) 2904 { 2905 unsigned long long recovery_start; 2906 2907 if (cmd_match(buf, "none")) 2908 recovery_start = MaxSector; 2909 else if (strict_strtoull(buf, 10, &recovery_start)) 2910 return -EINVAL; 2911 2912 if (rdev->mddev->pers && 2913 rdev->raid_disk >= 0) 2914 return -EBUSY; 2915 2916 rdev->recovery_offset = recovery_start; 2917 if (recovery_start == MaxSector) 2918 set_bit(In_sync, &rdev->flags); 2919 else 2920 clear_bit(In_sync, &rdev->flags); 2921 return len; 2922 } 2923 2924 static struct rdev_sysfs_entry rdev_recovery_start = 2925 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); 2926 2927 2928 static ssize_t 2929 badblocks_show(struct badblocks *bb, char *page, int unack); 2930 static ssize_t 2931 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); 2932 2933 static ssize_t bb_show(struct md_rdev *rdev, char *page) 2934 { 2935 return badblocks_show(&rdev->badblocks, page, 0); 2936 } 2937 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) 2938 { 2939 int rv = badblocks_store(&rdev->badblocks, page, len, 0); 2940 /* Maybe that ack was all we needed */ 2941 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) 2942 wake_up(&rdev->blocked_wait); 2943 return rv; 2944 } 2945 static struct rdev_sysfs_entry rdev_bad_blocks = 2946 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); 2947 2948 2949 static ssize_t ubb_show(struct md_rdev *rdev, char *page) 2950 { 2951 return badblocks_show(&rdev->badblocks, page, 1); 2952 } 2953 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) 2954 { 2955 return badblocks_store(&rdev->badblocks, page, len, 1); 2956 } 2957 static struct rdev_sysfs_entry rdev_unack_bad_blocks = 2958 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); 2959 2960 static struct attribute *rdev_default_attrs[] = { 2961 &rdev_state.attr, 2962 &rdev_errors.attr, 2963 &rdev_slot.attr, 2964 &rdev_offset.attr, 2965 &rdev_size.attr, 2966 &rdev_recovery_start.attr, 2967 &rdev_bad_blocks.attr, 2968 &rdev_unack_bad_blocks.attr, 2969 NULL, 2970 }; 2971 static ssize_t 2972 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 2973 { 2974 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2975 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); 2976 struct mddev *mddev = rdev->mddev; 2977 ssize_t rv; 2978 2979 if (!entry->show) 2980 return -EIO; 2981 2982 rv = mddev ? mddev_lock(mddev) : -EBUSY; 2983 if (!rv) { 2984 if (rdev->mddev == NULL) 2985 rv = -EBUSY; 2986 else 2987 rv = entry->show(rdev, page); 2988 mddev_unlock(mddev); 2989 } 2990 return rv; 2991 } 2992 2993 static ssize_t 2994 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 2995 const char *page, size_t length) 2996 { 2997 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2998 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); 2999 ssize_t rv; 3000 struct mddev *mddev = rdev->mddev; 3001 3002 if (!entry->store) 3003 return -EIO; 3004 if (!capable(CAP_SYS_ADMIN)) 3005 return -EACCES; 3006 rv = mddev ? mddev_lock(mddev): -EBUSY; 3007 if (!rv) { 3008 if (rdev->mddev == NULL) 3009 rv = -EBUSY; 3010 else 3011 rv = entry->store(rdev, page, length); 3012 mddev_unlock(mddev); 3013 } 3014 return rv; 3015 } 3016 3017 static void rdev_free(struct kobject *ko) 3018 { 3019 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); 3020 kfree(rdev); 3021 } 3022 static const struct sysfs_ops rdev_sysfs_ops = { 3023 .show = rdev_attr_show, 3024 .store = rdev_attr_store, 3025 }; 3026 static struct kobj_type rdev_ktype = { 3027 .release = rdev_free, 3028 .sysfs_ops = &rdev_sysfs_ops, 3029 .default_attrs = rdev_default_attrs, 3030 }; 3031 3032 int md_rdev_init(struct md_rdev *rdev) 3033 { 3034 rdev->desc_nr = -1; 3035 rdev->saved_raid_disk = -1; 3036 rdev->raid_disk = -1; 3037 rdev->flags = 0; 3038 rdev->data_offset = 0; 3039 rdev->sb_events = 0; 3040 rdev->last_read_error.tv_sec = 0; 3041 rdev->last_read_error.tv_nsec = 0; 3042 rdev->sb_loaded = 0; 3043 rdev->bb_page = NULL; 3044 atomic_set(&rdev->nr_pending, 0); 3045 atomic_set(&rdev->read_errors, 0); 3046 atomic_set(&rdev->corrected_errors, 0); 3047 3048 INIT_LIST_HEAD(&rdev->same_set); 3049 init_waitqueue_head(&rdev->blocked_wait); 3050 3051 /* Add space to store bad block list. 3052 * This reserves the space even on arrays where it cannot 3053 * be used - I wonder if that matters 3054 */ 3055 rdev->badblocks.count = 0; 3056 rdev->badblocks.shift = 0; 3057 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); 3058 seqlock_init(&rdev->badblocks.lock); 3059 if (rdev->badblocks.page == NULL) 3060 return -ENOMEM; 3061 3062 return 0; 3063 } 3064 EXPORT_SYMBOL_GPL(md_rdev_init); 3065 /* 3066 * Import a device. If 'super_format' >= 0, then sanity check the superblock 3067 * 3068 * mark the device faulty if: 3069 * 3070 * - the device is nonexistent (zero size) 3071 * - the device has no valid superblock 3072 * 3073 * a faulty rdev _never_ has rdev->sb set. 3074 */ 3075 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) 3076 { 3077 char b[BDEVNAME_SIZE]; 3078 int err; 3079 struct md_rdev *rdev; 3080 sector_t size; 3081 3082 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); 3083 if (!rdev) { 3084 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 3085 return ERR_PTR(-ENOMEM); 3086 } 3087 3088 err = md_rdev_init(rdev); 3089 if (err) 3090 goto abort_free; 3091 err = alloc_disk_sb(rdev); 3092 if (err) 3093 goto abort_free; 3094 3095 err = lock_rdev(rdev, newdev, super_format == -2); 3096 if (err) 3097 goto abort_free; 3098 3099 kobject_init(&rdev->kobj, &rdev_ktype); 3100 3101 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; 3102 if (!size) { 3103 printk(KERN_WARNING 3104 "md: %s has zero or unknown size, marking faulty!\n", 3105 bdevname(rdev->bdev,b)); 3106 err = -EINVAL; 3107 goto abort_free; 3108 } 3109 3110 if (super_format >= 0) { 3111 err = super_types[super_format]. 3112 load_super(rdev, NULL, super_minor); 3113 if (err == -EINVAL) { 3114 printk(KERN_WARNING 3115 "md: %s does not have a valid v%d.%d " 3116 "superblock, not importing!\n", 3117 bdevname(rdev->bdev,b), 3118 super_format, super_minor); 3119 goto abort_free; 3120 } 3121 if (err < 0) { 3122 printk(KERN_WARNING 3123 "md: could not read %s's sb, not importing!\n", 3124 bdevname(rdev->bdev,b)); 3125 goto abort_free; 3126 } 3127 } 3128 if (super_format == -1) 3129 /* hot-add for 0.90, or non-persistent: so no badblocks */ 3130 rdev->badblocks.shift = -1; 3131 3132 return rdev; 3133 3134 abort_free: 3135 if (rdev->bdev) 3136 unlock_rdev(rdev); 3137 free_disk_sb(rdev); 3138 kfree(rdev->badblocks.page); 3139 kfree(rdev); 3140 return ERR_PTR(err); 3141 } 3142 3143 /* 3144 * Check a full RAID array for plausibility 3145 */ 3146 3147 3148 static void analyze_sbs(struct mddev * mddev) 3149 { 3150 int i; 3151 struct md_rdev *rdev, *freshest, *tmp; 3152 char b[BDEVNAME_SIZE]; 3153 3154 freshest = NULL; 3155 rdev_for_each(rdev, tmp, mddev) 3156 switch (super_types[mddev->major_version]. 3157 load_super(rdev, freshest, mddev->minor_version)) { 3158 case 1: 3159 freshest = rdev; 3160 break; 3161 case 0: 3162 break; 3163 default: 3164 printk( KERN_ERR \ 3165 "md: fatal superblock inconsistency in %s" 3166 " -- removing from array\n", 3167 bdevname(rdev->bdev,b)); 3168 kick_rdev_from_array(rdev); 3169 } 3170 3171 3172 super_types[mddev->major_version]. 3173 validate_super(mddev, freshest); 3174 3175 i = 0; 3176 rdev_for_each(rdev, tmp, mddev) { 3177 if (mddev->max_disks && 3178 (rdev->desc_nr >= mddev->max_disks || 3179 i > mddev->max_disks)) { 3180 printk(KERN_WARNING 3181 "md: %s: %s: only %d devices permitted\n", 3182 mdname(mddev), bdevname(rdev->bdev, b), 3183 mddev->max_disks); 3184 kick_rdev_from_array(rdev); 3185 continue; 3186 } 3187 if (rdev != freshest) 3188 if (super_types[mddev->major_version]. 3189 validate_super(mddev, rdev)) { 3190 printk(KERN_WARNING "md: kicking non-fresh %s" 3191 " from array!\n", 3192 bdevname(rdev->bdev,b)); 3193 kick_rdev_from_array(rdev); 3194 continue; 3195 } 3196 if (mddev->level == LEVEL_MULTIPATH) { 3197 rdev->desc_nr = i++; 3198 rdev->raid_disk = rdev->desc_nr; 3199 set_bit(In_sync, &rdev->flags); 3200 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { 3201 rdev->raid_disk = -1; 3202 clear_bit(In_sync, &rdev->flags); 3203 } 3204 } 3205 } 3206 3207 /* Read a fixed-point number. 3208 * Numbers in sysfs attributes should be in "standard" units where 3209 * possible, so time should be in seconds. 3210 * However we internally use a a much smaller unit such as 3211 * milliseconds or jiffies. 3212 * This function takes a decimal number with a possible fractional 3213 * component, and produces an integer which is the result of 3214 * multiplying that number by 10^'scale'. 3215 * all without any floating-point arithmetic. 3216 */ 3217 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) 3218 { 3219 unsigned long result = 0; 3220 long decimals = -1; 3221 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { 3222 if (*cp == '.') 3223 decimals = 0; 3224 else if (decimals < scale) { 3225 unsigned int value; 3226 value = *cp - '0'; 3227 result = result * 10 + value; 3228 if (decimals >= 0) 3229 decimals++; 3230 } 3231 cp++; 3232 } 3233 if (*cp == '\n') 3234 cp++; 3235 if (*cp) 3236 return -EINVAL; 3237 if (decimals < 0) 3238 decimals = 0; 3239 while (decimals < scale) { 3240 result *= 10; 3241 decimals ++; 3242 } 3243 *res = result; 3244 return 0; 3245 } 3246 3247 3248 static void md_safemode_timeout(unsigned long data); 3249 3250 static ssize_t 3251 safe_delay_show(struct mddev *mddev, char *page) 3252 { 3253 int msec = (mddev->safemode_delay*1000)/HZ; 3254 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); 3255 } 3256 static ssize_t 3257 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) 3258 { 3259 unsigned long msec; 3260 3261 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) 3262 return -EINVAL; 3263 if (msec == 0) 3264 mddev->safemode_delay = 0; 3265 else { 3266 unsigned long old_delay = mddev->safemode_delay; 3267 mddev->safemode_delay = (msec*HZ)/1000; 3268 if (mddev->safemode_delay == 0) 3269 mddev->safemode_delay = 1; 3270 if (mddev->safemode_delay < old_delay) 3271 md_safemode_timeout((unsigned long)mddev); 3272 } 3273 return len; 3274 } 3275 static struct md_sysfs_entry md_safe_delay = 3276 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); 3277 3278 static ssize_t 3279 level_show(struct mddev *mddev, char *page) 3280 { 3281 struct md_personality *p = mddev->pers; 3282 if (p) 3283 return sprintf(page, "%s\n", p->name); 3284 else if (mddev->clevel[0]) 3285 return sprintf(page, "%s\n", mddev->clevel); 3286 else if (mddev->level != LEVEL_NONE) 3287 return sprintf(page, "%d\n", mddev->level); 3288 else 3289 return 0; 3290 } 3291 3292 static ssize_t 3293 level_store(struct mddev *mddev, const char *buf, size_t len) 3294 { 3295 char clevel[16]; 3296 ssize_t rv = len; 3297 struct md_personality *pers; 3298 long level; 3299 void *priv; 3300 struct md_rdev *rdev; 3301 3302 if (mddev->pers == NULL) { 3303 if (len == 0) 3304 return 0; 3305 if (len >= sizeof(mddev->clevel)) 3306 return -ENOSPC; 3307 strncpy(mddev->clevel, buf, len); 3308 if (mddev->clevel[len-1] == '\n') 3309 len--; 3310 mddev->clevel[len] = 0; 3311 mddev->level = LEVEL_NONE; 3312 return rv; 3313 } 3314 3315 /* request to change the personality. Need to ensure: 3316 * - array is not engaged in resync/recovery/reshape 3317 * - old personality can be suspended 3318 * - new personality will access other array. 3319 */ 3320 3321 if (mddev->sync_thread || 3322 mddev->reshape_position != MaxSector || 3323 mddev->sysfs_active) 3324 return -EBUSY; 3325 3326 if (!mddev->pers->quiesce) { 3327 printk(KERN_WARNING "md: %s: %s does not support online personality change\n", 3328 mdname(mddev), mddev->pers->name); 3329 return -EINVAL; 3330 } 3331 3332 /* Now find the new personality */ 3333 if (len == 0 || len >= sizeof(clevel)) 3334 return -EINVAL; 3335 strncpy(clevel, buf, len); 3336 if (clevel[len-1] == '\n') 3337 len--; 3338 clevel[len] = 0; 3339 if (strict_strtol(clevel, 10, &level)) 3340 level = LEVEL_NONE; 3341 3342 if (request_module("md-%s", clevel) != 0) 3343 request_module("md-level-%s", clevel); 3344 spin_lock(&pers_lock); 3345 pers = find_pers(level, clevel); 3346 if (!pers || !try_module_get(pers->owner)) { 3347 spin_unlock(&pers_lock); 3348 printk(KERN_WARNING "md: personality %s not loaded\n", clevel); 3349 return -EINVAL; 3350 } 3351 spin_unlock(&pers_lock); 3352 3353 if (pers == mddev->pers) { 3354 /* Nothing to do! */ 3355 module_put(pers->owner); 3356 return rv; 3357 } 3358 if (!pers->takeover) { 3359 module_put(pers->owner); 3360 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", 3361 mdname(mddev), clevel); 3362 return -EINVAL; 3363 } 3364 3365 list_for_each_entry(rdev, &mddev->disks, same_set) 3366 rdev->new_raid_disk = rdev->raid_disk; 3367 3368 /* ->takeover must set new_* and/or delta_disks 3369 * if it succeeds, and may set them when it fails. 3370 */ 3371 priv = pers->takeover(mddev); 3372 if (IS_ERR(priv)) { 3373 mddev->new_level = mddev->level; 3374 mddev->new_layout = mddev->layout; 3375 mddev->new_chunk_sectors = mddev->chunk_sectors; 3376 mddev->raid_disks -= mddev->delta_disks; 3377 mddev->delta_disks = 0; 3378 module_put(pers->owner); 3379 printk(KERN_WARNING "md: %s: %s would not accept array\n", 3380 mdname(mddev), clevel); 3381 return PTR_ERR(priv); 3382 } 3383 3384 /* Looks like we have a winner */ 3385 mddev_suspend(mddev); 3386 mddev->pers->stop(mddev); 3387 3388 if (mddev->pers->sync_request == NULL && 3389 pers->sync_request != NULL) { 3390 /* need to add the md_redundancy_group */ 3391 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 3392 printk(KERN_WARNING 3393 "md: cannot register extra attributes for %s\n", 3394 mdname(mddev)); 3395 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); 3396 } 3397 if (mddev->pers->sync_request != NULL && 3398 pers->sync_request == NULL) { 3399 /* need to remove the md_redundancy_group */ 3400 if (mddev->to_remove == NULL) 3401 mddev->to_remove = &md_redundancy_group; 3402 } 3403 3404 if (mddev->pers->sync_request == NULL && 3405 mddev->external) { 3406 /* We are converting from a no-redundancy array 3407 * to a redundancy array and metadata is managed 3408 * externally so we need to be sure that writes 3409 * won't block due to a need to transition 3410 * clean->dirty 3411 * until external management is started. 3412 */ 3413 mddev->in_sync = 0; 3414 mddev->safemode_delay = 0; 3415 mddev->safemode = 0; 3416 } 3417 3418 list_for_each_entry(rdev, &mddev->disks, same_set) { 3419 if (rdev->raid_disk < 0) 3420 continue; 3421 if (rdev->new_raid_disk >= mddev->raid_disks) 3422 rdev->new_raid_disk = -1; 3423 if (rdev->new_raid_disk == rdev->raid_disk) 3424 continue; 3425 sysfs_unlink_rdev(mddev, rdev); 3426 } 3427 list_for_each_entry(rdev, &mddev->disks, same_set) { 3428 if (rdev->raid_disk < 0) 3429 continue; 3430 if (rdev->new_raid_disk == rdev->raid_disk) 3431 continue; 3432 rdev->raid_disk = rdev->new_raid_disk; 3433 if (rdev->raid_disk < 0) 3434 clear_bit(In_sync, &rdev->flags); 3435 else { 3436 if (sysfs_link_rdev(mddev, rdev)) 3437 printk(KERN_WARNING "md: cannot register rd%d" 3438 " for %s after level change\n", 3439 rdev->raid_disk, mdname(mddev)); 3440 } 3441 } 3442 3443 module_put(mddev->pers->owner); 3444 mddev->pers = pers; 3445 mddev->private = priv; 3446 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 3447 mddev->level = mddev->new_level; 3448 mddev->layout = mddev->new_layout; 3449 mddev->chunk_sectors = mddev->new_chunk_sectors; 3450 mddev->delta_disks = 0; 3451 mddev->degraded = 0; 3452 if (mddev->pers->sync_request == NULL) { 3453 /* this is now an array without redundancy, so 3454 * it must always be in_sync 3455 */ 3456 mddev->in_sync = 1; 3457 del_timer_sync(&mddev->safemode_timer); 3458 } 3459 pers->run(mddev); 3460 mddev_resume(mddev); 3461 set_bit(MD_CHANGE_DEVS, &mddev->flags); 3462 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3463 md_wakeup_thread(mddev->thread); 3464 sysfs_notify(&mddev->kobj, NULL, "level"); 3465 md_new_event(mddev); 3466 return rv; 3467 } 3468 3469 static struct md_sysfs_entry md_level = 3470 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); 3471 3472 3473 static ssize_t 3474 layout_show(struct mddev *mddev, char *page) 3475 { 3476 /* just a number, not meaningful for all levels */ 3477 if (mddev->reshape_position != MaxSector && 3478 mddev->layout != mddev->new_layout) 3479 return sprintf(page, "%d (%d)\n", 3480 mddev->new_layout, mddev->layout); 3481 return sprintf(page, "%d\n", mddev->layout); 3482 } 3483 3484 static ssize_t 3485 layout_store(struct mddev *mddev, const char *buf, size_t len) 3486 { 3487 char *e; 3488 unsigned long n = simple_strtoul(buf, &e, 10); 3489 3490 if (!*buf || (*e && *e != '\n')) 3491 return -EINVAL; 3492 3493 if (mddev->pers) { 3494 int err; 3495 if (mddev->pers->check_reshape == NULL) 3496 return -EBUSY; 3497 mddev->new_layout = n; 3498 err = mddev->pers->check_reshape(mddev); 3499 if (err) { 3500 mddev->new_layout = mddev->layout; 3501 return err; 3502 } 3503 } else { 3504 mddev->new_layout = n; 3505 if (mddev->reshape_position == MaxSector) 3506 mddev->layout = n; 3507 } 3508 return len; 3509 } 3510 static struct md_sysfs_entry md_layout = 3511 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); 3512 3513 3514 static ssize_t 3515 raid_disks_show(struct mddev *mddev, char *page) 3516 { 3517 if (mddev->raid_disks == 0) 3518 return 0; 3519 if (mddev->reshape_position != MaxSector && 3520 mddev->delta_disks != 0) 3521 return sprintf(page, "%d (%d)\n", mddev->raid_disks, 3522 mddev->raid_disks - mddev->delta_disks); 3523 return sprintf(page, "%d\n", mddev->raid_disks); 3524 } 3525 3526 static int update_raid_disks(struct mddev *mddev, int raid_disks); 3527 3528 static ssize_t 3529 raid_disks_store(struct mddev *mddev, const char *buf, size_t len) 3530 { 3531 char *e; 3532 int rv = 0; 3533 unsigned long n = simple_strtoul(buf, &e, 10); 3534 3535 if (!*buf || (*e && *e != '\n')) 3536 return -EINVAL; 3537 3538 if (mddev->pers) 3539 rv = update_raid_disks(mddev, n); 3540 else if (mddev->reshape_position != MaxSector) { 3541 int olddisks = mddev->raid_disks - mddev->delta_disks; 3542 mddev->delta_disks = n - olddisks; 3543 mddev->raid_disks = n; 3544 } else 3545 mddev->raid_disks = n; 3546 return rv ? rv : len; 3547 } 3548 static struct md_sysfs_entry md_raid_disks = 3549 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); 3550 3551 static ssize_t 3552 chunk_size_show(struct mddev *mddev, char *page) 3553 { 3554 if (mddev->reshape_position != MaxSector && 3555 mddev->chunk_sectors != mddev->new_chunk_sectors) 3556 return sprintf(page, "%d (%d)\n", 3557 mddev->new_chunk_sectors << 9, 3558 mddev->chunk_sectors << 9); 3559 return sprintf(page, "%d\n", mddev->chunk_sectors << 9); 3560 } 3561 3562 static ssize_t 3563 chunk_size_store(struct mddev *mddev, const char *buf, size_t len) 3564 { 3565 char *e; 3566 unsigned long n = simple_strtoul(buf, &e, 10); 3567 3568 if (!*buf || (*e && *e != '\n')) 3569 return -EINVAL; 3570 3571 if (mddev->pers) { 3572 int err; 3573 if (mddev->pers->check_reshape == NULL) 3574 return -EBUSY; 3575 mddev->new_chunk_sectors = n >> 9; 3576 err = mddev->pers->check_reshape(mddev); 3577 if (err) { 3578 mddev->new_chunk_sectors = mddev->chunk_sectors; 3579 return err; 3580 } 3581 } else { 3582 mddev->new_chunk_sectors = n >> 9; 3583 if (mddev->reshape_position == MaxSector) 3584 mddev->chunk_sectors = n >> 9; 3585 } 3586 return len; 3587 } 3588 static struct md_sysfs_entry md_chunk_size = 3589 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); 3590 3591 static ssize_t 3592 resync_start_show(struct mddev *mddev, char *page) 3593 { 3594 if (mddev->recovery_cp == MaxSector) 3595 return sprintf(page, "none\n"); 3596 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); 3597 } 3598 3599 static ssize_t 3600 resync_start_store(struct mddev *mddev, const char *buf, size_t len) 3601 { 3602 char *e; 3603 unsigned long long n = simple_strtoull(buf, &e, 10); 3604 3605 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 3606 return -EBUSY; 3607 if (cmd_match(buf, "none")) 3608 n = MaxSector; 3609 else if (!*buf || (*e && *e != '\n')) 3610 return -EINVAL; 3611 3612 mddev->recovery_cp = n; 3613 return len; 3614 } 3615 static struct md_sysfs_entry md_resync_start = 3616 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); 3617 3618 /* 3619 * The array state can be: 3620 * 3621 * clear 3622 * No devices, no size, no level 3623 * Equivalent to STOP_ARRAY ioctl 3624 * inactive 3625 * May have some settings, but array is not active 3626 * all IO results in error 3627 * When written, doesn't tear down array, but just stops it 3628 * suspended (not supported yet) 3629 * All IO requests will block. The array can be reconfigured. 3630 * Writing this, if accepted, will block until array is quiescent 3631 * readonly 3632 * no resync can happen. no superblocks get written. 3633 * write requests fail 3634 * read-auto 3635 * like readonly, but behaves like 'clean' on a write request. 3636 * 3637 * clean - no pending writes, but otherwise active. 3638 * When written to inactive array, starts without resync 3639 * If a write request arrives then 3640 * if metadata is known, mark 'dirty' and switch to 'active'. 3641 * if not known, block and switch to write-pending 3642 * If written to an active array that has pending writes, then fails. 3643 * active 3644 * fully active: IO and resync can be happening. 3645 * When written to inactive array, starts with resync 3646 * 3647 * write-pending 3648 * clean, but writes are blocked waiting for 'active' to be written. 3649 * 3650 * active-idle 3651 * like active, but no writes have been seen for a while (100msec). 3652 * 3653 */ 3654 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, 3655 write_pending, active_idle, bad_word}; 3656 static char *array_states[] = { 3657 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", 3658 "write-pending", "active-idle", NULL }; 3659 3660 static int match_word(const char *word, char **list) 3661 { 3662 int n; 3663 for (n=0; list[n]; n++) 3664 if (cmd_match(word, list[n])) 3665 break; 3666 return n; 3667 } 3668 3669 static ssize_t 3670 array_state_show(struct mddev *mddev, char *page) 3671 { 3672 enum array_state st = inactive; 3673 3674 if (mddev->pers) 3675 switch(mddev->ro) { 3676 case 1: 3677 st = readonly; 3678 break; 3679 case 2: 3680 st = read_auto; 3681 break; 3682 case 0: 3683 if (mddev->in_sync) 3684 st = clean; 3685 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 3686 st = write_pending; 3687 else if (mddev->safemode) 3688 st = active_idle; 3689 else 3690 st = active; 3691 } 3692 else { 3693 if (list_empty(&mddev->disks) && 3694 mddev->raid_disks == 0 && 3695 mddev->dev_sectors == 0) 3696 st = clear; 3697 else 3698 st = inactive; 3699 } 3700 return sprintf(page, "%s\n", array_states[st]); 3701 } 3702 3703 static int do_md_stop(struct mddev * mddev, int ro, int is_open); 3704 static int md_set_readonly(struct mddev * mddev, int is_open); 3705 static int do_md_run(struct mddev * mddev); 3706 static int restart_array(struct mddev *mddev); 3707 3708 static ssize_t 3709 array_state_store(struct mddev *mddev, const char *buf, size_t len) 3710 { 3711 int err = -EINVAL; 3712 enum array_state st = match_word(buf, array_states); 3713 switch(st) { 3714 case bad_word: 3715 break; 3716 case clear: 3717 /* stopping an active array */ 3718 if (atomic_read(&mddev->openers) > 0) 3719 return -EBUSY; 3720 err = do_md_stop(mddev, 0, 0); 3721 break; 3722 case inactive: 3723 /* stopping an active array */ 3724 if (mddev->pers) { 3725 if (atomic_read(&mddev->openers) > 0) 3726 return -EBUSY; 3727 err = do_md_stop(mddev, 2, 0); 3728 } else 3729 err = 0; /* already inactive */ 3730 break; 3731 case suspended: 3732 break; /* not supported yet */ 3733 case readonly: 3734 if (mddev->pers) 3735 err = md_set_readonly(mddev, 0); 3736 else { 3737 mddev->ro = 1; 3738 set_disk_ro(mddev->gendisk, 1); 3739 err = do_md_run(mddev); 3740 } 3741 break; 3742 case read_auto: 3743 if (mddev->pers) { 3744 if (mddev->ro == 0) 3745 err = md_set_readonly(mddev, 0); 3746 else if (mddev->ro == 1) 3747 err = restart_array(mddev); 3748 if (err == 0) { 3749 mddev->ro = 2; 3750 set_disk_ro(mddev->gendisk, 0); 3751 } 3752 } else { 3753 mddev->ro = 2; 3754 err = do_md_run(mddev); 3755 } 3756 break; 3757 case clean: 3758 if (mddev->pers) { 3759 restart_array(mddev); 3760 spin_lock_irq(&mddev->write_lock); 3761 if (atomic_read(&mddev->writes_pending) == 0) { 3762 if (mddev->in_sync == 0) { 3763 mddev->in_sync = 1; 3764 if (mddev->safemode == 1) 3765 mddev->safemode = 0; 3766 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 3767 } 3768 err = 0; 3769 } else 3770 err = -EBUSY; 3771 spin_unlock_irq(&mddev->write_lock); 3772 } else 3773 err = -EINVAL; 3774 break; 3775 case active: 3776 if (mddev->pers) { 3777 restart_array(mddev); 3778 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 3779 wake_up(&mddev->sb_wait); 3780 err = 0; 3781 } else { 3782 mddev->ro = 0; 3783 set_disk_ro(mddev->gendisk, 0); 3784 err = do_md_run(mddev); 3785 } 3786 break; 3787 case write_pending: 3788 case active_idle: 3789 /* these cannot be set */ 3790 break; 3791 } 3792 if (err) 3793 return err; 3794 else { 3795 sysfs_notify_dirent_safe(mddev->sysfs_state); 3796 return len; 3797 } 3798 } 3799 static struct md_sysfs_entry md_array_state = 3800 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); 3801 3802 static ssize_t 3803 max_corrected_read_errors_show(struct mddev *mddev, char *page) { 3804 return sprintf(page, "%d\n", 3805 atomic_read(&mddev->max_corr_read_errors)); 3806 } 3807 3808 static ssize_t 3809 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) 3810 { 3811 char *e; 3812 unsigned long n = simple_strtoul(buf, &e, 10); 3813 3814 if (*buf && (*e == 0 || *e == '\n')) { 3815 atomic_set(&mddev->max_corr_read_errors, n); 3816 return len; 3817 } 3818 return -EINVAL; 3819 } 3820 3821 static struct md_sysfs_entry max_corr_read_errors = 3822 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, 3823 max_corrected_read_errors_store); 3824 3825 static ssize_t 3826 null_show(struct mddev *mddev, char *page) 3827 { 3828 return -EINVAL; 3829 } 3830 3831 static ssize_t 3832 new_dev_store(struct mddev *mddev, const char *buf, size_t len) 3833 { 3834 /* buf must be %d:%d\n? giving major and minor numbers */ 3835 /* The new device is added to the array. 3836 * If the array has a persistent superblock, we read the 3837 * superblock to initialise info and check validity. 3838 * Otherwise, only checking done is that in bind_rdev_to_array, 3839 * which mainly checks size. 3840 */ 3841 char *e; 3842 int major = simple_strtoul(buf, &e, 10); 3843 int minor; 3844 dev_t dev; 3845 struct md_rdev *rdev; 3846 int err; 3847 3848 if (!*buf || *e != ':' || !e[1] || e[1] == '\n') 3849 return -EINVAL; 3850 minor = simple_strtoul(e+1, &e, 10); 3851 if (*e && *e != '\n') 3852 return -EINVAL; 3853 dev = MKDEV(major, minor); 3854 if (major != MAJOR(dev) || 3855 minor != MINOR(dev)) 3856 return -EOVERFLOW; 3857 3858 3859 if (mddev->persistent) { 3860 rdev = md_import_device(dev, mddev->major_version, 3861 mddev->minor_version); 3862 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { 3863 struct md_rdev *rdev0 3864 = list_entry(mddev->disks.next, 3865 struct md_rdev, same_set); 3866 err = super_types[mddev->major_version] 3867 .load_super(rdev, rdev0, mddev->minor_version); 3868 if (err < 0) 3869 goto out; 3870 } 3871 } else if (mddev->external) 3872 rdev = md_import_device(dev, -2, -1); 3873 else 3874 rdev = md_import_device(dev, -1, -1); 3875 3876 if (IS_ERR(rdev)) 3877 return PTR_ERR(rdev); 3878 err = bind_rdev_to_array(rdev, mddev); 3879 out: 3880 if (err) 3881 export_rdev(rdev); 3882 return err ? err : len; 3883 } 3884 3885 static struct md_sysfs_entry md_new_device = 3886 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); 3887 3888 static ssize_t 3889 bitmap_store(struct mddev *mddev, const char *buf, size_t len) 3890 { 3891 char *end; 3892 unsigned long chunk, end_chunk; 3893 3894 if (!mddev->bitmap) 3895 goto out; 3896 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ 3897 while (*buf) { 3898 chunk = end_chunk = simple_strtoul(buf, &end, 0); 3899 if (buf == end) break; 3900 if (*end == '-') { /* range */ 3901 buf = end + 1; 3902 end_chunk = simple_strtoul(buf, &end, 0); 3903 if (buf == end) break; 3904 } 3905 if (*end && !isspace(*end)) break; 3906 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); 3907 buf = skip_spaces(end); 3908 } 3909 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ 3910 out: 3911 return len; 3912 } 3913 3914 static struct md_sysfs_entry md_bitmap = 3915 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); 3916 3917 static ssize_t 3918 size_show(struct mddev *mddev, char *page) 3919 { 3920 return sprintf(page, "%llu\n", 3921 (unsigned long long)mddev->dev_sectors / 2); 3922 } 3923 3924 static int update_size(struct mddev *mddev, sector_t num_sectors); 3925 3926 static ssize_t 3927 size_store(struct mddev *mddev, const char *buf, size_t len) 3928 { 3929 /* If array is inactive, we can reduce the component size, but 3930 * not increase it (except from 0). 3931 * If array is active, we can try an on-line resize 3932 */ 3933 sector_t sectors; 3934 int err = strict_blocks_to_sectors(buf, §ors); 3935 3936 if (err < 0) 3937 return err; 3938 if (mddev->pers) { 3939 err = update_size(mddev, sectors); 3940 md_update_sb(mddev, 1); 3941 } else { 3942 if (mddev->dev_sectors == 0 || 3943 mddev->dev_sectors > sectors) 3944 mddev->dev_sectors = sectors; 3945 else 3946 err = -ENOSPC; 3947 } 3948 return err ? err : len; 3949 } 3950 3951 static struct md_sysfs_entry md_size = 3952 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); 3953 3954 3955 /* Metdata version. 3956 * This is one of 3957 * 'none' for arrays with no metadata (good luck...) 3958 * 'external' for arrays with externally managed metadata, 3959 * or N.M for internally known formats 3960 */ 3961 static ssize_t 3962 metadata_show(struct mddev *mddev, char *page) 3963 { 3964 if (mddev->persistent) 3965 return sprintf(page, "%d.%d\n", 3966 mddev->major_version, mddev->minor_version); 3967 else if (mddev->external) 3968 return sprintf(page, "external:%s\n", mddev->metadata_type); 3969 else 3970 return sprintf(page, "none\n"); 3971 } 3972 3973 static ssize_t 3974 metadata_store(struct mddev *mddev, const char *buf, size_t len) 3975 { 3976 int major, minor; 3977 char *e; 3978 /* Changing the details of 'external' metadata is 3979 * always permitted. Otherwise there must be 3980 * no devices attached to the array. 3981 */ 3982 if (mddev->external && strncmp(buf, "external:", 9) == 0) 3983 ; 3984 else if (!list_empty(&mddev->disks)) 3985 return -EBUSY; 3986 3987 if (cmd_match(buf, "none")) { 3988 mddev->persistent = 0; 3989 mddev->external = 0; 3990 mddev->major_version = 0; 3991 mddev->minor_version = 90; 3992 return len; 3993 } 3994 if (strncmp(buf, "external:", 9) == 0) { 3995 size_t namelen = len-9; 3996 if (namelen >= sizeof(mddev->metadata_type)) 3997 namelen = sizeof(mddev->metadata_type)-1; 3998 strncpy(mddev->metadata_type, buf+9, namelen); 3999 mddev->metadata_type[namelen] = 0; 4000 if (namelen && mddev->metadata_type[namelen-1] == '\n') 4001 mddev->metadata_type[--namelen] = 0; 4002 mddev->persistent = 0; 4003 mddev->external = 1; 4004 mddev->major_version = 0; 4005 mddev->minor_version = 90; 4006 return len; 4007 } 4008 major = simple_strtoul(buf, &e, 10); 4009 if (e==buf || *e != '.') 4010 return -EINVAL; 4011 buf = e+1; 4012 minor = simple_strtoul(buf, &e, 10); 4013 if (e==buf || (*e && *e != '\n') ) 4014 return -EINVAL; 4015 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) 4016 return -ENOENT; 4017 mddev->major_version = major; 4018 mddev->minor_version = minor; 4019 mddev->persistent = 1; 4020 mddev->external = 0; 4021 return len; 4022 } 4023 4024 static struct md_sysfs_entry md_metadata = 4025 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 4026 4027 static ssize_t 4028 action_show(struct mddev *mddev, char *page) 4029 { 4030 char *type = "idle"; 4031 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 4032 type = "frozen"; 4033 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 4034 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { 4035 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 4036 type = "reshape"; 4037 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 4038 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 4039 type = "resync"; 4040 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 4041 type = "check"; 4042 else 4043 type = "repair"; 4044 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) 4045 type = "recover"; 4046 } 4047 return sprintf(page, "%s\n", type); 4048 } 4049 4050 static void reap_sync_thread(struct mddev *mddev); 4051 4052 static ssize_t 4053 action_store(struct mddev *mddev, const char *page, size_t len) 4054 { 4055 if (!mddev->pers || !mddev->pers->sync_request) 4056 return -EINVAL; 4057 4058 if (cmd_match(page, "frozen")) 4059 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4060 else 4061 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4062 4063 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { 4064 if (mddev->sync_thread) { 4065 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4066 reap_sync_thread(mddev); 4067 } 4068 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 4069 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 4070 return -EBUSY; 4071 else if (cmd_match(page, "resync")) 4072 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4073 else if (cmd_match(page, "recover")) { 4074 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 4075 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4076 } else if (cmd_match(page, "reshape")) { 4077 int err; 4078 if (mddev->pers->start_reshape == NULL) 4079 return -EINVAL; 4080 err = mddev->pers->start_reshape(mddev); 4081 if (err) 4082 return err; 4083 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4084 } else { 4085 if (cmd_match(page, "check")) 4086 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 4087 else if (!cmd_match(page, "repair")) 4088 return -EINVAL; 4089 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 4090 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 4091 } 4092 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4093 md_wakeup_thread(mddev->thread); 4094 sysfs_notify_dirent_safe(mddev->sysfs_action); 4095 return len; 4096 } 4097 4098 static ssize_t 4099 mismatch_cnt_show(struct mddev *mddev, char *page) 4100 { 4101 return sprintf(page, "%llu\n", 4102 (unsigned long long) mddev->resync_mismatches); 4103 } 4104 4105 static struct md_sysfs_entry md_scan_mode = 4106 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); 4107 4108 4109 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); 4110 4111 static ssize_t 4112 sync_min_show(struct mddev *mddev, char *page) 4113 { 4114 return sprintf(page, "%d (%s)\n", speed_min(mddev), 4115 mddev->sync_speed_min ? "local": "system"); 4116 } 4117 4118 static ssize_t 4119 sync_min_store(struct mddev *mddev, const char *buf, size_t len) 4120 { 4121 int min; 4122 char *e; 4123 if (strncmp(buf, "system", 6)==0) { 4124 mddev->sync_speed_min = 0; 4125 return len; 4126 } 4127 min = simple_strtoul(buf, &e, 10); 4128 if (buf == e || (*e && *e != '\n') || min <= 0) 4129 return -EINVAL; 4130 mddev->sync_speed_min = min; 4131 return len; 4132 } 4133 4134 static struct md_sysfs_entry md_sync_min = 4135 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); 4136 4137 static ssize_t 4138 sync_max_show(struct mddev *mddev, char *page) 4139 { 4140 return sprintf(page, "%d (%s)\n", speed_max(mddev), 4141 mddev->sync_speed_max ? "local": "system"); 4142 } 4143 4144 static ssize_t 4145 sync_max_store(struct mddev *mddev, const char *buf, size_t len) 4146 { 4147 int max; 4148 char *e; 4149 if (strncmp(buf, "system", 6)==0) { 4150 mddev->sync_speed_max = 0; 4151 return len; 4152 } 4153 max = simple_strtoul(buf, &e, 10); 4154 if (buf == e || (*e && *e != '\n') || max <= 0) 4155 return -EINVAL; 4156 mddev->sync_speed_max = max; 4157 return len; 4158 } 4159 4160 static struct md_sysfs_entry md_sync_max = 4161 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); 4162 4163 static ssize_t 4164 degraded_show(struct mddev *mddev, char *page) 4165 { 4166 return sprintf(page, "%d\n", mddev->degraded); 4167 } 4168 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); 4169 4170 static ssize_t 4171 sync_force_parallel_show(struct mddev *mddev, char *page) 4172 { 4173 return sprintf(page, "%d\n", mddev->parallel_resync); 4174 } 4175 4176 static ssize_t 4177 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) 4178 { 4179 long n; 4180 4181 if (strict_strtol(buf, 10, &n)) 4182 return -EINVAL; 4183 4184 if (n != 0 && n != 1) 4185 return -EINVAL; 4186 4187 mddev->parallel_resync = n; 4188 4189 if (mddev->sync_thread) 4190 wake_up(&resync_wait); 4191 4192 return len; 4193 } 4194 4195 /* force parallel resync, even with shared block devices */ 4196 static struct md_sysfs_entry md_sync_force_parallel = 4197 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, 4198 sync_force_parallel_show, sync_force_parallel_store); 4199 4200 static ssize_t 4201 sync_speed_show(struct mddev *mddev, char *page) 4202 { 4203 unsigned long resync, dt, db; 4204 if (mddev->curr_resync == 0) 4205 return sprintf(page, "none\n"); 4206 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); 4207 dt = (jiffies - mddev->resync_mark) / HZ; 4208 if (!dt) dt++; 4209 db = resync - mddev->resync_mark_cnt; 4210 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ 4211 } 4212 4213 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); 4214 4215 static ssize_t 4216 sync_completed_show(struct mddev *mddev, char *page) 4217 { 4218 unsigned long long max_sectors, resync; 4219 4220 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4221 return sprintf(page, "none\n"); 4222 4223 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 4224 max_sectors = mddev->resync_max_sectors; 4225 else 4226 max_sectors = mddev->dev_sectors; 4227 4228 resync = mddev->curr_resync_completed; 4229 return sprintf(page, "%llu / %llu\n", resync, max_sectors); 4230 } 4231 4232 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); 4233 4234 static ssize_t 4235 min_sync_show(struct mddev *mddev, char *page) 4236 { 4237 return sprintf(page, "%llu\n", 4238 (unsigned long long)mddev->resync_min); 4239 } 4240 static ssize_t 4241 min_sync_store(struct mddev *mddev, const char *buf, size_t len) 4242 { 4243 unsigned long long min; 4244 if (strict_strtoull(buf, 10, &min)) 4245 return -EINVAL; 4246 if (min > mddev->resync_max) 4247 return -EINVAL; 4248 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4249 return -EBUSY; 4250 4251 /* Must be a multiple of chunk_size */ 4252 if (mddev->chunk_sectors) { 4253 sector_t temp = min; 4254 if (sector_div(temp, mddev->chunk_sectors)) 4255 return -EINVAL; 4256 } 4257 mddev->resync_min = min; 4258 4259 return len; 4260 } 4261 4262 static struct md_sysfs_entry md_min_sync = 4263 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); 4264 4265 static ssize_t 4266 max_sync_show(struct mddev *mddev, char *page) 4267 { 4268 if (mddev->resync_max == MaxSector) 4269 return sprintf(page, "max\n"); 4270 else 4271 return sprintf(page, "%llu\n", 4272 (unsigned long long)mddev->resync_max); 4273 } 4274 static ssize_t 4275 max_sync_store(struct mddev *mddev, const char *buf, size_t len) 4276 { 4277 if (strncmp(buf, "max", 3) == 0) 4278 mddev->resync_max = MaxSector; 4279 else { 4280 unsigned long long max; 4281 if (strict_strtoull(buf, 10, &max)) 4282 return -EINVAL; 4283 if (max < mddev->resync_min) 4284 return -EINVAL; 4285 if (max < mddev->resync_max && 4286 mddev->ro == 0 && 4287 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4288 return -EBUSY; 4289 4290 /* Must be a multiple of chunk_size */ 4291 if (mddev->chunk_sectors) { 4292 sector_t temp = max; 4293 if (sector_div(temp, mddev->chunk_sectors)) 4294 return -EINVAL; 4295 } 4296 mddev->resync_max = max; 4297 } 4298 wake_up(&mddev->recovery_wait); 4299 return len; 4300 } 4301 4302 static struct md_sysfs_entry md_max_sync = 4303 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); 4304 4305 static ssize_t 4306 suspend_lo_show(struct mddev *mddev, char *page) 4307 { 4308 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); 4309 } 4310 4311 static ssize_t 4312 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) 4313 { 4314 char *e; 4315 unsigned long long new = simple_strtoull(buf, &e, 10); 4316 unsigned long long old = mddev->suspend_lo; 4317 4318 if (mddev->pers == NULL || 4319 mddev->pers->quiesce == NULL) 4320 return -EINVAL; 4321 if (buf == e || (*e && *e != '\n')) 4322 return -EINVAL; 4323 4324 mddev->suspend_lo = new; 4325 if (new >= old) 4326 /* Shrinking suspended region */ 4327 mddev->pers->quiesce(mddev, 2); 4328 else { 4329 /* Expanding suspended region - need to wait */ 4330 mddev->pers->quiesce(mddev, 1); 4331 mddev->pers->quiesce(mddev, 0); 4332 } 4333 return len; 4334 } 4335 static struct md_sysfs_entry md_suspend_lo = 4336 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); 4337 4338 4339 static ssize_t 4340 suspend_hi_show(struct mddev *mddev, char *page) 4341 { 4342 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); 4343 } 4344 4345 static ssize_t 4346 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) 4347 { 4348 char *e; 4349 unsigned long long new = simple_strtoull(buf, &e, 10); 4350 unsigned long long old = mddev->suspend_hi; 4351 4352 if (mddev->pers == NULL || 4353 mddev->pers->quiesce == NULL) 4354 return -EINVAL; 4355 if (buf == e || (*e && *e != '\n')) 4356 return -EINVAL; 4357 4358 mddev->suspend_hi = new; 4359 if (new <= old) 4360 /* Shrinking suspended region */ 4361 mddev->pers->quiesce(mddev, 2); 4362 else { 4363 /* Expanding suspended region - need to wait */ 4364 mddev->pers->quiesce(mddev, 1); 4365 mddev->pers->quiesce(mddev, 0); 4366 } 4367 return len; 4368 } 4369 static struct md_sysfs_entry md_suspend_hi = 4370 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); 4371 4372 static ssize_t 4373 reshape_position_show(struct mddev *mddev, char *page) 4374 { 4375 if (mddev->reshape_position != MaxSector) 4376 return sprintf(page, "%llu\n", 4377 (unsigned long long)mddev->reshape_position); 4378 strcpy(page, "none\n"); 4379 return 5; 4380 } 4381 4382 static ssize_t 4383 reshape_position_store(struct mddev *mddev, const char *buf, size_t len) 4384 { 4385 char *e; 4386 unsigned long long new = simple_strtoull(buf, &e, 10); 4387 if (mddev->pers) 4388 return -EBUSY; 4389 if (buf == e || (*e && *e != '\n')) 4390 return -EINVAL; 4391 mddev->reshape_position = new; 4392 mddev->delta_disks = 0; 4393 mddev->new_level = mddev->level; 4394 mddev->new_layout = mddev->layout; 4395 mddev->new_chunk_sectors = mddev->chunk_sectors; 4396 return len; 4397 } 4398 4399 static struct md_sysfs_entry md_reshape_position = 4400 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, 4401 reshape_position_store); 4402 4403 static ssize_t 4404 array_size_show(struct mddev *mddev, char *page) 4405 { 4406 if (mddev->external_size) 4407 return sprintf(page, "%llu\n", 4408 (unsigned long long)mddev->array_sectors/2); 4409 else 4410 return sprintf(page, "default\n"); 4411 } 4412 4413 static ssize_t 4414 array_size_store(struct mddev *mddev, const char *buf, size_t len) 4415 { 4416 sector_t sectors; 4417 4418 if (strncmp(buf, "default", 7) == 0) { 4419 if (mddev->pers) 4420 sectors = mddev->pers->size(mddev, 0, 0); 4421 else 4422 sectors = mddev->array_sectors; 4423 4424 mddev->external_size = 0; 4425 } else { 4426 if (strict_blocks_to_sectors(buf, §ors) < 0) 4427 return -EINVAL; 4428 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) 4429 return -E2BIG; 4430 4431 mddev->external_size = 1; 4432 } 4433 4434 mddev->array_sectors = sectors; 4435 if (mddev->pers) { 4436 set_capacity(mddev->gendisk, mddev->array_sectors); 4437 revalidate_disk(mddev->gendisk); 4438 } 4439 return len; 4440 } 4441 4442 static struct md_sysfs_entry md_array_size = 4443 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, 4444 array_size_store); 4445 4446 static struct attribute *md_default_attrs[] = { 4447 &md_level.attr, 4448 &md_layout.attr, 4449 &md_raid_disks.attr, 4450 &md_chunk_size.attr, 4451 &md_size.attr, 4452 &md_resync_start.attr, 4453 &md_metadata.attr, 4454 &md_new_device.attr, 4455 &md_safe_delay.attr, 4456 &md_array_state.attr, 4457 &md_reshape_position.attr, 4458 &md_array_size.attr, 4459 &max_corr_read_errors.attr, 4460 NULL, 4461 }; 4462 4463 static struct attribute *md_redundancy_attrs[] = { 4464 &md_scan_mode.attr, 4465 &md_mismatches.attr, 4466 &md_sync_min.attr, 4467 &md_sync_max.attr, 4468 &md_sync_speed.attr, 4469 &md_sync_force_parallel.attr, 4470 &md_sync_completed.attr, 4471 &md_min_sync.attr, 4472 &md_max_sync.attr, 4473 &md_suspend_lo.attr, 4474 &md_suspend_hi.attr, 4475 &md_bitmap.attr, 4476 &md_degraded.attr, 4477 NULL, 4478 }; 4479 static struct attribute_group md_redundancy_group = { 4480 .name = NULL, 4481 .attrs = md_redundancy_attrs, 4482 }; 4483 4484 4485 static ssize_t 4486 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 4487 { 4488 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4489 struct mddev *mddev = container_of(kobj, struct mddev, kobj); 4490 ssize_t rv; 4491 4492 if (!entry->show) 4493 return -EIO; 4494 rv = mddev_lock(mddev); 4495 if (!rv) { 4496 rv = entry->show(mddev, page); 4497 mddev_unlock(mddev); 4498 } 4499 return rv; 4500 } 4501 4502 static ssize_t 4503 md_attr_store(struct kobject *kobj, struct attribute *attr, 4504 const char *page, size_t length) 4505 { 4506 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4507 struct mddev *mddev = container_of(kobj, struct mddev, kobj); 4508 ssize_t rv; 4509 4510 if (!entry->store) 4511 return -EIO; 4512 if (!capable(CAP_SYS_ADMIN)) 4513 return -EACCES; 4514 rv = mddev_lock(mddev); 4515 if (mddev->hold_active == UNTIL_IOCTL) 4516 mddev->hold_active = 0; 4517 if (!rv) { 4518 rv = entry->store(mddev, page, length); 4519 mddev_unlock(mddev); 4520 } 4521 return rv; 4522 } 4523 4524 static void md_free(struct kobject *ko) 4525 { 4526 struct mddev *mddev = container_of(ko, struct mddev, kobj); 4527 4528 if (mddev->sysfs_state) 4529 sysfs_put(mddev->sysfs_state); 4530 4531 if (mddev->gendisk) { 4532 del_gendisk(mddev->gendisk); 4533 put_disk(mddev->gendisk); 4534 } 4535 if (mddev->queue) 4536 blk_cleanup_queue(mddev->queue); 4537 4538 kfree(mddev); 4539 } 4540 4541 static const struct sysfs_ops md_sysfs_ops = { 4542 .show = md_attr_show, 4543 .store = md_attr_store, 4544 }; 4545 static struct kobj_type md_ktype = { 4546 .release = md_free, 4547 .sysfs_ops = &md_sysfs_ops, 4548 .default_attrs = md_default_attrs, 4549 }; 4550 4551 int mdp_major = 0; 4552 4553 static void mddev_delayed_delete(struct work_struct *ws) 4554 { 4555 struct mddev *mddev = container_of(ws, struct mddev, del_work); 4556 4557 sysfs_remove_group(&mddev->kobj, &md_bitmap_group); 4558 kobject_del(&mddev->kobj); 4559 kobject_put(&mddev->kobj); 4560 } 4561 4562 static int md_alloc(dev_t dev, char *name) 4563 { 4564 static DEFINE_MUTEX(disks_mutex); 4565 struct mddev *mddev = mddev_find(dev); 4566 struct gendisk *disk; 4567 int partitioned; 4568 int shift; 4569 int unit; 4570 int error; 4571 4572 if (!mddev) 4573 return -ENODEV; 4574 4575 partitioned = (MAJOR(mddev->unit) != MD_MAJOR); 4576 shift = partitioned ? MdpMinorShift : 0; 4577 unit = MINOR(mddev->unit) >> shift; 4578 4579 /* wait for any previous instance of this device to be 4580 * completely removed (mddev_delayed_delete). 4581 */ 4582 flush_workqueue(md_misc_wq); 4583 4584 mutex_lock(&disks_mutex); 4585 error = -EEXIST; 4586 if (mddev->gendisk) 4587 goto abort; 4588 4589 if (name) { 4590 /* Need to ensure that 'name' is not a duplicate. 4591 */ 4592 struct mddev *mddev2; 4593 spin_lock(&all_mddevs_lock); 4594 4595 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) 4596 if (mddev2->gendisk && 4597 strcmp(mddev2->gendisk->disk_name, name) == 0) { 4598 spin_unlock(&all_mddevs_lock); 4599 goto abort; 4600 } 4601 spin_unlock(&all_mddevs_lock); 4602 } 4603 4604 error = -ENOMEM; 4605 mddev->queue = blk_alloc_queue(GFP_KERNEL); 4606 if (!mddev->queue) 4607 goto abort; 4608 mddev->queue->queuedata = mddev; 4609 4610 blk_queue_make_request(mddev->queue, md_make_request); 4611 4612 disk = alloc_disk(1 << shift); 4613 if (!disk) { 4614 blk_cleanup_queue(mddev->queue); 4615 mddev->queue = NULL; 4616 goto abort; 4617 } 4618 disk->major = MAJOR(mddev->unit); 4619 disk->first_minor = unit << shift; 4620 if (name) 4621 strcpy(disk->disk_name, name); 4622 else if (partitioned) 4623 sprintf(disk->disk_name, "md_d%d", unit); 4624 else 4625 sprintf(disk->disk_name, "md%d", unit); 4626 disk->fops = &md_fops; 4627 disk->private_data = mddev; 4628 disk->queue = mddev->queue; 4629 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); 4630 /* Allow extended partitions. This makes the 4631 * 'mdp' device redundant, but we can't really 4632 * remove it now. 4633 */ 4634 disk->flags |= GENHD_FL_EXT_DEVT; 4635 mddev->gendisk = disk; 4636 /* As soon as we call add_disk(), another thread could get 4637 * through to md_open, so make sure it doesn't get too far 4638 */ 4639 mutex_lock(&mddev->open_mutex); 4640 add_disk(disk); 4641 4642 error = kobject_init_and_add(&mddev->kobj, &md_ktype, 4643 &disk_to_dev(disk)->kobj, "%s", "md"); 4644 if (error) { 4645 /* This isn't possible, but as kobject_init_and_add is marked 4646 * __must_check, we must do something with the result 4647 */ 4648 printk(KERN_WARNING "md: cannot register %s/md - name in use\n", 4649 disk->disk_name); 4650 error = 0; 4651 } 4652 if (mddev->kobj.sd && 4653 sysfs_create_group(&mddev->kobj, &md_bitmap_group)) 4654 printk(KERN_DEBUG "pointless warning\n"); 4655 mutex_unlock(&mddev->open_mutex); 4656 abort: 4657 mutex_unlock(&disks_mutex); 4658 if (!error && mddev->kobj.sd) { 4659 kobject_uevent(&mddev->kobj, KOBJ_ADD); 4660 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); 4661 } 4662 mddev_put(mddev); 4663 return error; 4664 } 4665 4666 static struct kobject *md_probe(dev_t dev, int *part, void *data) 4667 { 4668 md_alloc(dev, NULL); 4669 return NULL; 4670 } 4671 4672 static int add_named_array(const char *val, struct kernel_param *kp) 4673 { 4674 /* val must be "md_*" where * is not all digits. 4675 * We allocate an array with a large free minor number, and 4676 * set the name to val. val must not already be an active name. 4677 */ 4678 int len = strlen(val); 4679 char buf[DISK_NAME_LEN]; 4680 4681 while (len && val[len-1] == '\n') 4682 len--; 4683 if (len >= DISK_NAME_LEN) 4684 return -E2BIG; 4685 strlcpy(buf, val, len+1); 4686 if (strncmp(buf, "md_", 3) != 0) 4687 return -EINVAL; 4688 return md_alloc(0, buf); 4689 } 4690 4691 static void md_safemode_timeout(unsigned long data) 4692 { 4693 struct mddev *mddev = (struct mddev *) data; 4694 4695 if (!atomic_read(&mddev->writes_pending)) { 4696 mddev->safemode = 1; 4697 if (mddev->external) 4698 sysfs_notify_dirent_safe(mddev->sysfs_state); 4699 } 4700 md_wakeup_thread(mddev->thread); 4701 } 4702 4703 static int start_dirty_degraded; 4704 4705 int md_run(struct mddev *mddev) 4706 { 4707 int err; 4708 struct md_rdev *rdev; 4709 struct md_personality *pers; 4710 4711 if (list_empty(&mddev->disks)) 4712 /* cannot run an array with no devices.. */ 4713 return -EINVAL; 4714 4715 if (mddev->pers) 4716 return -EBUSY; 4717 /* Cannot run until previous stop completes properly */ 4718 if (mddev->sysfs_active) 4719 return -EBUSY; 4720 4721 /* 4722 * Analyze all RAID superblock(s) 4723 */ 4724 if (!mddev->raid_disks) { 4725 if (!mddev->persistent) 4726 return -EINVAL; 4727 analyze_sbs(mddev); 4728 } 4729 4730 if (mddev->level != LEVEL_NONE) 4731 request_module("md-level-%d", mddev->level); 4732 else if (mddev->clevel[0]) 4733 request_module("md-%s", mddev->clevel); 4734 4735 /* 4736 * Drop all container device buffers, from now on 4737 * the only valid external interface is through the md 4738 * device. 4739 */ 4740 list_for_each_entry(rdev, &mddev->disks, same_set) { 4741 if (test_bit(Faulty, &rdev->flags)) 4742 continue; 4743 sync_blockdev(rdev->bdev); 4744 invalidate_bdev(rdev->bdev); 4745 4746 /* perform some consistency tests on the device. 4747 * We don't want the data to overlap the metadata, 4748 * Internal Bitmap issues have been handled elsewhere. 4749 */ 4750 if (rdev->meta_bdev) { 4751 /* Nothing to check */; 4752 } else if (rdev->data_offset < rdev->sb_start) { 4753 if (mddev->dev_sectors && 4754 rdev->data_offset + mddev->dev_sectors 4755 > rdev->sb_start) { 4756 printk("md: %s: data overlaps metadata\n", 4757 mdname(mddev)); 4758 return -EINVAL; 4759 } 4760 } else { 4761 if (rdev->sb_start + rdev->sb_size/512 4762 > rdev->data_offset) { 4763 printk("md: %s: metadata overlaps data\n", 4764 mdname(mddev)); 4765 return -EINVAL; 4766 } 4767 } 4768 sysfs_notify_dirent_safe(rdev->sysfs_state); 4769 } 4770 4771 if (mddev->bio_set == NULL) 4772 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 4773 sizeof(struct mddev *)); 4774 4775 spin_lock(&pers_lock); 4776 pers = find_pers(mddev->level, mddev->clevel); 4777 if (!pers || !try_module_get(pers->owner)) { 4778 spin_unlock(&pers_lock); 4779 if (mddev->level != LEVEL_NONE) 4780 printk(KERN_WARNING "md: personality for level %d is not loaded!\n", 4781 mddev->level); 4782 else 4783 printk(KERN_WARNING "md: personality for level %s is not loaded!\n", 4784 mddev->clevel); 4785 return -EINVAL; 4786 } 4787 mddev->pers = pers; 4788 spin_unlock(&pers_lock); 4789 if (mddev->level != pers->level) { 4790 mddev->level = pers->level; 4791 mddev->new_level = pers->level; 4792 } 4793 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 4794 4795 if (mddev->reshape_position != MaxSector && 4796 pers->start_reshape == NULL) { 4797 /* This personality cannot handle reshaping... */ 4798 mddev->pers = NULL; 4799 module_put(pers->owner); 4800 return -EINVAL; 4801 } 4802 4803 if (pers->sync_request) { 4804 /* Warn if this is a potentially silly 4805 * configuration. 4806 */ 4807 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 4808 struct md_rdev *rdev2; 4809 int warned = 0; 4810 4811 list_for_each_entry(rdev, &mddev->disks, same_set) 4812 list_for_each_entry(rdev2, &mddev->disks, same_set) { 4813 if (rdev < rdev2 && 4814 rdev->bdev->bd_contains == 4815 rdev2->bdev->bd_contains) { 4816 printk(KERN_WARNING 4817 "%s: WARNING: %s appears to be" 4818 " on the same physical disk as" 4819 " %s.\n", 4820 mdname(mddev), 4821 bdevname(rdev->bdev,b), 4822 bdevname(rdev2->bdev,b2)); 4823 warned = 1; 4824 } 4825 } 4826 4827 if (warned) 4828 printk(KERN_WARNING 4829 "True protection against single-disk" 4830 " failure might be compromised.\n"); 4831 } 4832 4833 mddev->recovery = 0; 4834 /* may be over-ridden by personality */ 4835 mddev->resync_max_sectors = mddev->dev_sectors; 4836 4837 mddev->ok_start_degraded = start_dirty_degraded; 4838 4839 if (start_readonly && mddev->ro == 0) 4840 mddev->ro = 2; /* read-only, but switch on first write */ 4841 4842 err = mddev->pers->run(mddev); 4843 if (err) 4844 printk(KERN_ERR "md: pers->run() failed ...\n"); 4845 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { 4846 WARN_ONCE(!mddev->external_size, "%s: default size too small," 4847 " but 'external_size' not in effect?\n", __func__); 4848 printk(KERN_ERR 4849 "md: invalid array_size %llu > default size %llu\n", 4850 (unsigned long long)mddev->array_sectors / 2, 4851 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); 4852 err = -EINVAL; 4853 mddev->pers->stop(mddev); 4854 } 4855 if (err == 0 && mddev->pers->sync_request) { 4856 err = bitmap_create(mddev); 4857 if (err) { 4858 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 4859 mdname(mddev), err); 4860 mddev->pers->stop(mddev); 4861 } 4862 } 4863 if (err) { 4864 module_put(mddev->pers->owner); 4865 mddev->pers = NULL; 4866 bitmap_destroy(mddev); 4867 return err; 4868 } 4869 if (mddev->pers->sync_request) { 4870 if (mddev->kobj.sd && 4871 sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 4872 printk(KERN_WARNING 4873 "md: cannot register extra attributes for %s\n", 4874 mdname(mddev)); 4875 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); 4876 } else if (mddev->ro == 2) /* auto-readonly not meaningful */ 4877 mddev->ro = 0; 4878 4879 atomic_set(&mddev->writes_pending,0); 4880 atomic_set(&mddev->max_corr_read_errors, 4881 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); 4882 mddev->safemode = 0; 4883 mddev->safemode_timer.function = md_safemode_timeout; 4884 mddev->safemode_timer.data = (unsigned long) mddev; 4885 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ 4886 mddev->in_sync = 1; 4887 smp_wmb(); 4888 mddev->ready = 1; 4889 list_for_each_entry(rdev, &mddev->disks, same_set) 4890 if (rdev->raid_disk >= 0) 4891 if (sysfs_link_rdev(mddev, rdev)) 4892 /* failure here is OK */; 4893 4894 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4895 4896 if (mddev->flags) 4897 md_update_sb(mddev, 0); 4898 4899 md_new_event(mddev); 4900 sysfs_notify_dirent_safe(mddev->sysfs_state); 4901 sysfs_notify_dirent_safe(mddev->sysfs_action); 4902 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4903 return 0; 4904 } 4905 EXPORT_SYMBOL_GPL(md_run); 4906 4907 static int do_md_run(struct mddev *mddev) 4908 { 4909 int err; 4910 4911 err = md_run(mddev); 4912 if (err) 4913 goto out; 4914 err = bitmap_load(mddev); 4915 if (err) { 4916 bitmap_destroy(mddev); 4917 goto out; 4918 } 4919 4920 md_wakeup_thread(mddev->thread); 4921 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 4922 4923 set_capacity(mddev->gendisk, mddev->array_sectors); 4924 revalidate_disk(mddev->gendisk); 4925 mddev->changed = 1; 4926 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4927 out: 4928 return err; 4929 } 4930 4931 static int restart_array(struct mddev *mddev) 4932 { 4933 struct gendisk *disk = mddev->gendisk; 4934 4935 /* Complain if it has no devices */ 4936 if (list_empty(&mddev->disks)) 4937 return -ENXIO; 4938 if (!mddev->pers) 4939 return -EINVAL; 4940 if (!mddev->ro) 4941 return -EBUSY; 4942 mddev->safemode = 0; 4943 mddev->ro = 0; 4944 set_disk_ro(disk, 0); 4945 printk(KERN_INFO "md: %s switched to read-write mode.\n", 4946 mdname(mddev)); 4947 /* Kick recovery or resync if necessary */ 4948 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4949 md_wakeup_thread(mddev->thread); 4950 md_wakeup_thread(mddev->sync_thread); 4951 sysfs_notify_dirent_safe(mddev->sysfs_state); 4952 return 0; 4953 } 4954 4955 /* similar to deny_write_access, but accounts for our holding a reference 4956 * to the file ourselves */ 4957 static int deny_bitmap_write_access(struct file * file) 4958 { 4959 struct inode *inode = file->f_mapping->host; 4960 4961 spin_lock(&inode->i_lock); 4962 if (atomic_read(&inode->i_writecount) > 1) { 4963 spin_unlock(&inode->i_lock); 4964 return -ETXTBSY; 4965 } 4966 atomic_set(&inode->i_writecount, -1); 4967 spin_unlock(&inode->i_lock); 4968 4969 return 0; 4970 } 4971 4972 void restore_bitmap_write_access(struct file *file) 4973 { 4974 struct inode *inode = file->f_mapping->host; 4975 4976 spin_lock(&inode->i_lock); 4977 atomic_set(&inode->i_writecount, 1); 4978 spin_unlock(&inode->i_lock); 4979 } 4980 4981 static void md_clean(struct mddev *mddev) 4982 { 4983 mddev->array_sectors = 0; 4984 mddev->external_size = 0; 4985 mddev->dev_sectors = 0; 4986 mddev->raid_disks = 0; 4987 mddev->recovery_cp = 0; 4988 mddev->resync_min = 0; 4989 mddev->resync_max = MaxSector; 4990 mddev->reshape_position = MaxSector; 4991 mddev->external = 0; 4992 mddev->persistent = 0; 4993 mddev->level = LEVEL_NONE; 4994 mddev->clevel[0] = 0; 4995 mddev->flags = 0; 4996 mddev->ro = 0; 4997 mddev->metadata_type[0] = 0; 4998 mddev->chunk_sectors = 0; 4999 mddev->ctime = mddev->utime = 0; 5000 mddev->layout = 0; 5001 mddev->max_disks = 0; 5002 mddev->events = 0; 5003 mddev->can_decrease_events = 0; 5004 mddev->delta_disks = 0; 5005 mddev->new_level = LEVEL_NONE; 5006 mddev->new_layout = 0; 5007 mddev->new_chunk_sectors = 0; 5008 mddev->curr_resync = 0; 5009 mddev->resync_mismatches = 0; 5010 mddev->suspend_lo = mddev->suspend_hi = 0; 5011 mddev->sync_speed_min = mddev->sync_speed_max = 0; 5012 mddev->recovery = 0; 5013 mddev->in_sync = 0; 5014 mddev->changed = 0; 5015 mddev->degraded = 0; 5016 mddev->safemode = 0; 5017 mddev->bitmap_info.offset = 0; 5018 mddev->bitmap_info.default_offset = 0; 5019 mddev->bitmap_info.chunksize = 0; 5020 mddev->bitmap_info.daemon_sleep = 0; 5021 mddev->bitmap_info.max_write_behind = 0; 5022 } 5023 5024 static void __md_stop_writes(struct mddev *mddev) 5025 { 5026 if (mddev->sync_thread) { 5027 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5028 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 5029 reap_sync_thread(mddev); 5030 } 5031 5032 del_timer_sync(&mddev->safemode_timer); 5033 5034 bitmap_flush(mddev); 5035 md_super_wait(mddev); 5036 5037 if (!mddev->in_sync || mddev->flags) { 5038 /* mark array as shutdown cleanly */ 5039 mddev->in_sync = 1; 5040 md_update_sb(mddev, 1); 5041 } 5042 } 5043 5044 void md_stop_writes(struct mddev *mddev) 5045 { 5046 mddev_lock(mddev); 5047 __md_stop_writes(mddev); 5048 mddev_unlock(mddev); 5049 } 5050 EXPORT_SYMBOL_GPL(md_stop_writes); 5051 5052 void md_stop(struct mddev *mddev) 5053 { 5054 mddev->ready = 0; 5055 mddev->pers->stop(mddev); 5056 if (mddev->pers->sync_request && mddev->to_remove == NULL) 5057 mddev->to_remove = &md_redundancy_group; 5058 module_put(mddev->pers->owner); 5059 mddev->pers = NULL; 5060 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5061 } 5062 EXPORT_SYMBOL_GPL(md_stop); 5063 5064 static int md_set_readonly(struct mddev *mddev, int is_open) 5065 { 5066 int err = 0; 5067 mutex_lock(&mddev->open_mutex); 5068 if (atomic_read(&mddev->openers) > is_open) { 5069 printk("md: %s still in use.\n",mdname(mddev)); 5070 err = -EBUSY; 5071 goto out; 5072 } 5073 if (mddev->pers) { 5074 __md_stop_writes(mddev); 5075 5076 err = -ENXIO; 5077 if (mddev->ro==1) 5078 goto out; 5079 mddev->ro = 1; 5080 set_disk_ro(mddev->gendisk, 1); 5081 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5082 sysfs_notify_dirent_safe(mddev->sysfs_state); 5083 err = 0; 5084 } 5085 out: 5086 mutex_unlock(&mddev->open_mutex); 5087 return err; 5088 } 5089 5090 /* mode: 5091 * 0 - completely stop and dis-assemble array 5092 * 2 - stop but do not disassemble array 5093 */ 5094 static int do_md_stop(struct mddev * mddev, int mode, int is_open) 5095 { 5096 struct gendisk *disk = mddev->gendisk; 5097 struct md_rdev *rdev; 5098 5099 mutex_lock(&mddev->open_mutex); 5100 if (atomic_read(&mddev->openers) > is_open || 5101 mddev->sysfs_active) { 5102 printk("md: %s still in use.\n",mdname(mddev)); 5103 mutex_unlock(&mddev->open_mutex); 5104 return -EBUSY; 5105 } 5106 5107 if (mddev->pers) { 5108 if (mddev->ro) 5109 set_disk_ro(disk, 0); 5110 5111 __md_stop_writes(mddev); 5112 md_stop(mddev); 5113 mddev->queue->merge_bvec_fn = NULL; 5114 mddev->queue->backing_dev_info.congested_fn = NULL; 5115 5116 /* tell userspace to handle 'inactive' */ 5117 sysfs_notify_dirent_safe(mddev->sysfs_state); 5118 5119 list_for_each_entry(rdev, &mddev->disks, same_set) 5120 if (rdev->raid_disk >= 0) 5121 sysfs_unlink_rdev(mddev, rdev); 5122 5123 set_capacity(disk, 0); 5124 mutex_unlock(&mddev->open_mutex); 5125 mddev->changed = 1; 5126 revalidate_disk(disk); 5127 5128 if (mddev->ro) 5129 mddev->ro = 0; 5130 } else 5131 mutex_unlock(&mddev->open_mutex); 5132 /* 5133 * Free resources if final stop 5134 */ 5135 if (mode == 0) { 5136 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 5137 5138 bitmap_destroy(mddev); 5139 if (mddev->bitmap_info.file) { 5140 restore_bitmap_write_access(mddev->bitmap_info.file); 5141 fput(mddev->bitmap_info.file); 5142 mddev->bitmap_info.file = NULL; 5143 } 5144 mddev->bitmap_info.offset = 0; 5145 5146 export_array(mddev); 5147 5148 md_clean(mddev); 5149 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 5150 if (mddev->hold_active == UNTIL_STOP) 5151 mddev->hold_active = 0; 5152 } 5153 blk_integrity_unregister(disk); 5154 md_new_event(mddev); 5155 sysfs_notify_dirent_safe(mddev->sysfs_state); 5156 return 0; 5157 } 5158 5159 #ifndef MODULE 5160 static void autorun_array(struct mddev *mddev) 5161 { 5162 struct md_rdev *rdev; 5163 int err; 5164 5165 if (list_empty(&mddev->disks)) 5166 return; 5167 5168 printk(KERN_INFO "md: running: "); 5169 5170 list_for_each_entry(rdev, &mddev->disks, same_set) { 5171 char b[BDEVNAME_SIZE]; 5172 printk("<%s>", bdevname(rdev->bdev,b)); 5173 } 5174 printk("\n"); 5175 5176 err = do_md_run(mddev); 5177 if (err) { 5178 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 5179 do_md_stop(mddev, 0, 0); 5180 } 5181 } 5182 5183 /* 5184 * lets try to run arrays based on all disks that have arrived 5185 * until now. (those are in pending_raid_disks) 5186 * 5187 * the method: pick the first pending disk, collect all disks with 5188 * the same UUID, remove all from the pending list and put them into 5189 * the 'same_array' list. Then order this list based on superblock 5190 * update time (freshest comes first), kick out 'old' disks and 5191 * compare superblocks. If everything's fine then run it. 5192 * 5193 * If "unit" is allocated, then bump its reference count 5194 */ 5195 static void autorun_devices(int part) 5196 { 5197 struct md_rdev *rdev0, *rdev, *tmp; 5198 struct mddev *mddev; 5199 char b[BDEVNAME_SIZE]; 5200 5201 printk(KERN_INFO "md: autorun ...\n"); 5202 while (!list_empty(&pending_raid_disks)) { 5203 int unit; 5204 dev_t dev; 5205 LIST_HEAD(candidates); 5206 rdev0 = list_entry(pending_raid_disks.next, 5207 struct md_rdev, same_set); 5208 5209 printk(KERN_INFO "md: considering %s ...\n", 5210 bdevname(rdev0->bdev,b)); 5211 INIT_LIST_HEAD(&candidates); 5212 rdev_for_each_list(rdev, tmp, &pending_raid_disks) 5213 if (super_90_load(rdev, rdev0, 0) >= 0) { 5214 printk(KERN_INFO "md: adding %s ...\n", 5215 bdevname(rdev->bdev,b)); 5216 list_move(&rdev->same_set, &candidates); 5217 } 5218 /* 5219 * now we have a set of devices, with all of them having 5220 * mostly sane superblocks. It's time to allocate the 5221 * mddev. 5222 */ 5223 if (part) { 5224 dev = MKDEV(mdp_major, 5225 rdev0->preferred_minor << MdpMinorShift); 5226 unit = MINOR(dev) >> MdpMinorShift; 5227 } else { 5228 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 5229 unit = MINOR(dev); 5230 } 5231 if (rdev0->preferred_minor != unit) { 5232 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 5233 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 5234 break; 5235 } 5236 5237 md_probe(dev, NULL, NULL); 5238 mddev = mddev_find(dev); 5239 if (!mddev || !mddev->gendisk) { 5240 if (mddev) 5241 mddev_put(mddev); 5242 printk(KERN_ERR 5243 "md: cannot allocate memory for md drive.\n"); 5244 break; 5245 } 5246 if (mddev_lock(mddev)) 5247 printk(KERN_WARNING "md: %s locked, cannot run\n", 5248 mdname(mddev)); 5249 else if (mddev->raid_disks || mddev->major_version 5250 || !list_empty(&mddev->disks)) { 5251 printk(KERN_WARNING 5252 "md: %s already running, cannot run %s\n", 5253 mdname(mddev), bdevname(rdev0->bdev,b)); 5254 mddev_unlock(mddev); 5255 } else { 5256 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 5257 mddev->persistent = 1; 5258 rdev_for_each_list(rdev, tmp, &candidates) { 5259 list_del_init(&rdev->same_set); 5260 if (bind_rdev_to_array(rdev, mddev)) 5261 export_rdev(rdev); 5262 } 5263 autorun_array(mddev); 5264 mddev_unlock(mddev); 5265 } 5266 /* on success, candidates will be empty, on error 5267 * it won't... 5268 */ 5269 rdev_for_each_list(rdev, tmp, &candidates) { 5270 list_del_init(&rdev->same_set); 5271 export_rdev(rdev); 5272 } 5273 mddev_put(mddev); 5274 } 5275 printk(KERN_INFO "md: ... autorun DONE.\n"); 5276 } 5277 #endif /* !MODULE */ 5278 5279 static int get_version(void __user * arg) 5280 { 5281 mdu_version_t ver; 5282 5283 ver.major = MD_MAJOR_VERSION; 5284 ver.minor = MD_MINOR_VERSION; 5285 ver.patchlevel = MD_PATCHLEVEL_VERSION; 5286 5287 if (copy_to_user(arg, &ver, sizeof(ver))) 5288 return -EFAULT; 5289 5290 return 0; 5291 } 5292 5293 static int get_array_info(struct mddev * mddev, void __user * arg) 5294 { 5295 mdu_array_info_t info; 5296 int nr,working,insync,failed,spare; 5297 struct md_rdev *rdev; 5298 5299 nr=working=insync=failed=spare=0; 5300 list_for_each_entry(rdev, &mddev->disks, same_set) { 5301 nr++; 5302 if (test_bit(Faulty, &rdev->flags)) 5303 failed++; 5304 else { 5305 working++; 5306 if (test_bit(In_sync, &rdev->flags)) 5307 insync++; 5308 else 5309 spare++; 5310 } 5311 } 5312 5313 info.major_version = mddev->major_version; 5314 info.minor_version = mddev->minor_version; 5315 info.patch_version = MD_PATCHLEVEL_VERSION; 5316 info.ctime = mddev->ctime; 5317 info.level = mddev->level; 5318 info.size = mddev->dev_sectors / 2; 5319 if (info.size != mddev->dev_sectors / 2) /* overflow */ 5320 info.size = -1; 5321 info.nr_disks = nr; 5322 info.raid_disks = mddev->raid_disks; 5323 info.md_minor = mddev->md_minor; 5324 info.not_persistent= !mddev->persistent; 5325 5326 info.utime = mddev->utime; 5327 info.state = 0; 5328 if (mddev->in_sync) 5329 info.state = (1<<MD_SB_CLEAN); 5330 if (mddev->bitmap && mddev->bitmap_info.offset) 5331 info.state = (1<<MD_SB_BITMAP_PRESENT); 5332 info.active_disks = insync; 5333 info.working_disks = working; 5334 info.failed_disks = failed; 5335 info.spare_disks = spare; 5336 5337 info.layout = mddev->layout; 5338 info.chunk_size = mddev->chunk_sectors << 9; 5339 5340 if (copy_to_user(arg, &info, sizeof(info))) 5341 return -EFAULT; 5342 5343 return 0; 5344 } 5345 5346 static int get_bitmap_file(struct mddev * mddev, void __user * arg) 5347 { 5348 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 5349 char *ptr, *buf = NULL; 5350 int err = -ENOMEM; 5351 5352 if (md_allow_write(mddev)) 5353 file = kmalloc(sizeof(*file), GFP_NOIO); 5354 else 5355 file = kmalloc(sizeof(*file), GFP_KERNEL); 5356 5357 if (!file) 5358 goto out; 5359 5360 /* bitmap disabled, zero the first byte and copy out */ 5361 if (!mddev->bitmap || !mddev->bitmap->file) { 5362 file->pathname[0] = '\0'; 5363 goto copy_out; 5364 } 5365 5366 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 5367 if (!buf) 5368 goto out; 5369 5370 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); 5371 if (IS_ERR(ptr)) 5372 goto out; 5373 5374 strcpy(file->pathname, ptr); 5375 5376 copy_out: 5377 err = 0; 5378 if (copy_to_user(arg, file, sizeof(*file))) 5379 err = -EFAULT; 5380 out: 5381 kfree(buf); 5382 kfree(file); 5383 return err; 5384 } 5385 5386 static int get_disk_info(struct mddev * mddev, void __user * arg) 5387 { 5388 mdu_disk_info_t info; 5389 struct md_rdev *rdev; 5390 5391 if (copy_from_user(&info, arg, sizeof(info))) 5392 return -EFAULT; 5393 5394 rdev = find_rdev_nr(mddev, info.number); 5395 if (rdev) { 5396 info.major = MAJOR(rdev->bdev->bd_dev); 5397 info.minor = MINOR(rdev->bdev->bd_dev); 5398 info.raid_disk = rdev->raid_disk; 5399 info.state = 0; 5400 if (test_bit(Faulty, &rdev->flags)) 5401 info.state |= (1<<MD_DISK_FAULTY); 5402 else if (test_bit(In_sync, &rdev->flags)) { 5403 info.state |= (1<<MD_DISK_ACTIVE); 5404 info.state |= (1<<MD_DISK_SYNC); 5405 } 5406 if (test_bit(WriteMostly, &rdev->flags)) 5407 info.state |= (1<<MD_DISK_WRITEMOSTLY); 5408 } else { 5409 info.major = info.minor = 0; 5410 info.raid_disk = -1; 5411 info.state = (1<<MD_DISK_REMOVED); 5412 } 5413 5414 if (copy_to_user(arg, &info, sizeof(info))) 5415 return -EFAULT; 5416 5417 return 0; 5418 } 5419 5420 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info) 5421 { 5422 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 5423 struct md_rdev *rdev; 5424 dev_t dev = MKDEV(info->major,info->minor); 5425 5426 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 5427 return -EOVERFLOW; 5428 5429 if (!mddev->raid_disks) { 5430 int err; 5431 /* expecting a device which has a superblock */ 5432 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 5433 if (IS_ERR(rdev)) { 5434 printk(KERN_WARNING 5435 "md: md_import_device returned %ld\n", 5436 PTR_ERR(rdev)); 5437 return PTR_ERR(rdev); 5438 } 5439 if (!list_empty(&mddev->disks)) { 5440 struct md_rdev *rdev0 5441 = list_entry(mddev->disks.next, 5442 struct md_rdev, same_set); 5443 err = super_types[mddev->major_version] 5444 .load_super(rdev, rdev0, mddev->minor_version); 5445 if (err < 0) { 5446 printk(KERN_WARNING 5447 "md: %s has different UUID to %s\n", 5448 bdevname(rdev->bdev,b), 5449 bdevname(rdev0->bdev,b2)); 5450 export_rdev(rdev); 5451 return -EINVAL; 5452 } 5453 } 5454 err = bind_rdev_to_array(rdev, mddev); 5455 if (err) 5456 export_rdev(rdev); 5457 return err; 5458 } 5459 5460 /* 5461 * add_new_disk can be used once the array is assembled 5462 * to add "hot spares". They must already have a superblock 5463 * written 5464 */ 5465 if (mddev->pers) { 5466 int err; 5467 if (!mddev->pers->hot_add_disk) { 5468 printk(KERN_WARNING 5469 "%s: personality does not support diskops!\n", 5470 mdname(mddev)); 5471 return -EINVAL; 5472 } 5473 if (mddev->persistent) 5474 rdev = md_import_device(dev, mddev->major_version, 5475 mddev->minor_version); 5476 else 5477 rdev = md_import_device(dev, -1, -1); 5478 if (IS_ERR(rdev)) { 5479 printk(KERN_WARNING 5480 "md: md_import_device returned %ld\n", 5481 PTR_ERR(rdev)); 5482 return PTR_ERR(rdev); 5483 } 5484 /* set saved_raid_disk if appropriate */ 5485 if (!mddev->persistent) { 5486 if (info->state & (1<<MD_DISK_SYNC) && 5487 info->raid_disk < mddev->raid_disks) { 5488 rdev->raid_disk = info->raid_disk; 5489 set_bit(In_sync, &rdev->flags); 5490 } else 5491 rdev->raid_disk = -1; 5492 } else 5493 super_types[mddev->major_version]. 5494 validate_super(mddev, rdev); 5495 if ((info->state & (1<<MD_DISK_SYNC)) && 5496 (!test_bit(In_sync, &rdev->flags) || 5497 rdev->raid_disk != info->raid_disk)) { 5498 /* This was a hot-add request, but events doesn't 5499 * match, so reject it. 5500 */ 5501 export_rdev(rdev); 5502 return -EINVAL; 5503 } 5504 5505 if (test_bit(In_sync, &rdev->flags)) 5506 rdev->saved_raid_disk = rdev->raid_disk; 5507 else 5508 rdev->saved_raid_disk = -1; 5509 5510 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 5511 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5512 set_bit(WriteMostly, &rdev->flags); 5513 else 5514 clear_bit(WriteMostly, &rdev->flags); 5515 5516 rdev->raid_disk = -1; 5517 err = bind_rdev_to_array(rdev, mddev); 5518 if (!err && !mddev->pers->hot_remove_disk) { 5519 /* If there is hot_add_disk but no hot_remove_disk 5520 * then added disks for geometry changes, 5521 * and should be added immediately. 5522 */ 5523 super_types[mddev->major_version]. 5524 validate_super(mddev, rdev); 5525 err = mddev->pers->hot_add_disk(mddev, rdev); 5526 if (err) 5527 unbind_rdev_from_array(rdev); 5528 } 5529 if (err) 5530 export_rdev(rdev); 5531 else 5532 sysfs_notify_dirent_safe(rdev->sysfs_state); 5533 5534 md_update_sb(mddev, 1); 5535 if (mddev->degraded) 5536 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 5537 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5538 if (!err) 5539 md_new_event(mddev); 5540 md_wakeup_thread(mddev->thread); 5541 return err; 5542 } 5543 5544 /* otherwise, add_new_disk is only allowed 5545 * for major_version==0 superblocks 5546 */ 5547 if (mddev->major_version != 0) { 5548 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 5549 mdname(mddev)); 5550 return -EINVAL; 5551 } 5552 5553 if (!(info->state & (1<<MD_DISK_FAULTY))) { 5554 int err; 5555 rdev = md_import_device(dev, -1, 0); 5556 if (IS_ERR(rdev)) { 5557 printk(KERN_WARNING 5558 "md: error, md_import_device() returned %ld\n", 5559 PTR_ERR(rdev)); 5560 return PTR_ERR(rdev); 5561 } 5562 rdev->desc_nr = info->number; 5563 if (info->raid_disk < mddev->raid_disks) 5564 rdev->raid_disk = info->raid_disk; 5565 else 5566 rdev->raid_disk = -1; 5567 5568 if (rdev->raid_disk < mddev->raid_disks) 5569 if (info->state & (1<<MD_DISK_SYNC)) 5570 set_bit(In_sync, &rdev->flags); 5571 5572 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5573 set_bit(WriteMostly, &rdev->flags); 5574 5575 if (!mddev->persistent) { 5576 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 5577 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5578 } else 5579 rdev->sb_start = calc_dev_sboffset(rdev); 5580 rdev->sectors = rdev->sb_start; 5581 5582 err = bind_rdev_to_array(rdev, mddev); 5583 if (err) { 5584 export_rdev(rdev); 5585 return err; 5586 } 5587 } 5588 5589 return 0; 5590 } 5591 5592 static int hot_remove_disk(struct mddev * mddev, dev_t dev) 5593 { 5594 char b[BDEVNAME_SIZE]; 5595 struct md_rdev *rdev; 5596 5597 rdev = find_rdev(mddev, dev); 5598 if (!rdev) 5599 return -ENXIO; 5600 5601 if (rdev->raid_disk >= 0) 5602 goto busy; 5603 5604 kick_rdev_from_array(rdev); 5605 md_update_sb(mddev, 1); 5606 md_new_event(mddev); 5607 5608 return 0; 5609 busy: 5610 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", 5611 bdevname(rdev->bdev,b), mdname(mddev)); 5612 return -EBUSY; 5613 } 5614 5615 static int hot_add_disk(struct mddev * mddev, dev_t dev) 5616 { 5617 char b[BDEVNAME_SIZE]; 5618 int err; 5619 struct md_rdev *rdev; 5620 5621 if (!mddev->pers) 5622 return -ENODEV; 5623 5624 if (mddev->major_version != 0) { 5625 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 5626 " version-0 superblocks.\n", 5627 mdname(mddev)); 5628 return -EINVAL; 5629 } 5630 if (!mddev->pers->hot_add_disk) { 5631 printk(KERN_WARNING 5632 "%s: personality does not support diskops!\n", 5633 mdname(mddev)); 5634 return -EINVAL; 5635 } 5636 5637 rdev = md_import_device(dev, -1, 0); 5638 if (IS_ERR(rdev)) { 5639 printk(KERN_WARNING 5640 "md: error, md_import_device() returned %ld\n", 5641 PTR_ERR(rdev)); 5642 return -EINVAL; 5643 } 5644 5645 if (mddev->persistent) 5646 rdev->sb_start = calc_dev_sboffset(rdev); 5647 else 5648 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5649 5650 rdev->sectors = rdev->sb_start; 5651 5652 if (test_bit(Faulty, &rdev->flags)) { 5653 printk(KERN_WARNING 5654 "md: can not hot-add faulty %s disk to %s!\n", 5655 bdevname(rdev->bdev,b), mdname(mddev)); 5656 err = -EINVAL; 5657 goto abort_export; 5658 } 5659 clear_bit(In_sync, &rdev->flags); 5660 rdev->desc_nr = -1; 5661 rdev->saved_raid_disk = -1; 5662 err = bind_rdev_to_array(rdev, mddev); 5663 if (err) 5664 goto abort_export; 5665 5666 /* 5667 * The rest should better be atomic, we can have disk failures 5668 * noticed in interrupt contexts ... 5669 */ 5670 5671 rdev->raid_disk = -1; 5672 5673 md_update_sb(mddev, 1); 5674 5675 /* 5676 * Kick recovery, maybe this spare has to be added to the 5677 * array immediately. 5678 */ 5679 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5680 md_wakeup_thread(mddev->thread); 5681 md_new_event(mddev); 5682 return 0; 5683 5684 abort_export: 5685 export_rdev(rdev); 5686 return err; 5687 } 5688 5689 static int set_bitmap_file(struct mddev *mddev, int fd) 5690 { 5691 int err; 5692 5693 if (mddev->pers) { 5694 if (!mddev->pers->quiesce) 5695 return -EBUSY; 5696 if (mddev->recovery || mddev->sync_thread) 5697 return -EBUSY; 5698 /* we should be able to change the bitmap.. */ 5699 } 5700 5701 5702 if (fd >= 0) { 5703 if (mddev->bitmap) 5704 return -EEXIST; /* cannot add when bitmap is present */ 5705 mddev->bitmap_info.file = fget(fd); 5706 5707 if (mddev->bitmap_info.file == NULL) { 5708 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 5709 mdname(mddev)); 5710 return -EBADF; 5711 } 5712 5713 err = deny_bitmap_write_access(mddev->bitmap_info.file); 5714 if (err) { 5715 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 5716 mdname(mddev)); 5717 fput(mddev->bitmap_info.file); 5718 mddev->bitmap_info.file = NULL; 5719 return err; 5720 } 5721 mddev->bitmap_info.offset = 0; /* file overrides offset */ 5722 } else if (mddev->bitmap == NULL) 5723 return -ENOENT; /* cannot remove what isn't there */ 5724 err = 0; 5725 if (mddev->pers) { 5726 mddev->pers->quiesce(mddev, 1); 5727 if (fd >= 0) { 5728 err = bitmap_create(mddev); 5729 if (!err) 5730 err = bitmap_load(mddev); 5731 } 5732 if (fd < 0 || err) { 5733 bitmap_destroy(mddev); 5734 fd = -1; /* make sure to put the file */ 5735 } 5736 mddev->pers->quiesce(mddev, 0); 5737 } 5738 if (fd < 0) { 5739 if (mddev->bitmap_info.file) { 5740 restore_bitmap_write_access(mddev->bitmap_info.file); 5741 fput(mddev->bitmap_info.file); 5742 } 5743 mddev->bitmap_info.file = NULL; 5744 } 5745 5746 return err; 5747 } 5748 5749 /* 5750 * set_array_info is used two different ways 5751 * The original usage is when creating a new array. 5752 * In this usage, raid_disks is > 0 and it together with 5753 * level, size, not_persistent,layout,chunksize determine the 5754 * shape of the array. 5755 * This will always create an array with a type-0.90.0 superblock. 5756 * The newer usage is when assembling an array. 5757 * In this case raid_disks will be 0, and the major_version field is 5758 * use to determine which style super-blocks are to be found on the devices. 5759 * The minor and patch _version numbers are also kept incase the 5760 * super_block handler wishes to interpret them. 5761 */ 5762 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info) 5763 { 5764 5765 if (info->raid_disks == 0) { 5766 /* just setting version number for superblock loading */ 5767 if (info->major_version < 0 || 5768 info->major_version >= ARRAY_SIZE(super_types) || 5769 super_types[info->major_version].name == NULL) { 5770 /* maybe try to auto-load a module? */ 5771 printk(KERN_INFO 5772 "md: superblock version %d not known\n", 5773 info->major_version); 5774 return -EINVAL; 5775 } 5776 mddev->major_version = info->major_version; 5777 mddev->minor_version = info->minor_version; 5778 mddev->patch_version = info->patch_version; 5779 mddev->persistent = !info->not_persistent; 5780 /* ensure mddev_put doesn't delete this now that there 5781 * is some minimal configuration. 5782 */ 5783 mddev->ctime = get_seconds(); 5784 return 0; 5785 } 5786 mddev->major_version = MD_MAJOR_VERSION; 5787 mddev->minor_version = MD_MINOR_VERSION; 5788 mddev->patch_version = MD_PATCHLEVEL_VERSION; 5789 mddev->ctime = get_seconds(); 5790 5791 mddev->level = info->level; 5792 mddev->clevel[0] = 0; 5793 mddev->dev_sectors = 2 * (sector_t)info->size; 5794 mddev->raid_disks = info->raid_disks; 5795 /* don't set md_minor, it is determined by which /dev/md* was 5796 * openned 5797 */ 5798 if (info->state & (1<<MD_SB_CLEAN)) 5799 mddev->recovery_cp = MaxSector; 5800 else 5801 mddev->recovery_cp = 0; 5802 mddev->persistent = ! info->not_persistent; 5803 mddev->external = 0; 5804 5805 mddev->layout = info->layout; 5806 mddev->chunk_sectors = info->chunk_size >> 9; 5807 5808 mddev->max_disks = MD_SB_DISKS; 5809 5810 if (mddev->persistent) 5811 mddev->flags = 0; 5812 set_bit(MD_CHANGE_DEVS, &mddev->flags); 5813 5814 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 5815 mddev->bitmap_info.offset = 0; 5816 5817 mddev->reshape_position = MaxSector; 5818 5819 /* 5820 * Generate a 128 bit UUID 5821 */ 5822 get_random_bytes(mddev->uuid, 16); 5823 5824 mddev->new_level = mddev->level; 5825 mddev->new_chunk_sectors = mddev->chunk_sectors; 5826 mddev->new_layout = mddev->layout; 5827 mddev->delta_disks = 0; 5828 5829 return 0; 5830 } 5831 5832 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) 5833 { 5834 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); 5835 5836 if (mddev->external_size) 5837 return; 5838 5839 mddev->array_sectors = array_sectors; 5840 } 5841 EXPORT_SYMBOL(md_set_array_sectors); 5842 5843 static int update_size(struct mddev *mddev, sector_t num_sectors) 5844 { 5845 struct md_rdev *rdev; 5846 int rv; 5847 int fit = (num_sectors == 0); 5848 5849 if (mddev->pers->resize == NULL) 5850 return -EINVAL; 5851 /* The "num_sectors" is the number of sectors of each device that 5852 * is used. This can only make sense for arrays with redundancy. 5853 * linear and raid0 always use whatever space is available. We can only 5854 * consider changing this number if no resync or reconstruction is 5855 * happening, and if the new size is acceptable. It must fit before the 5856 * sb_start or, if that is <data_offset, it must fit before the size 5857 * of each device. If num_sectors is zero, we find the largest size 5858 * that fits. 5859 */ 5860 if (mddev->sync_thread) 5861 return -EBUSY; 5862 if (mddev->bitmap) 5863 /* Sorry, cannot grow a bitmap yet, just remove it, 5864 * grow, and re-add. 5865 */ 5866 return -EBUSY; 5867 list_for_each_entry(rdev, &mddev->disks, same_set) { 5868 sector_t avail = rdev->sectors; 5869 5870 if (fit && (num_sectors == 0 || num_sectors > avail)) 5871 num_sectors = avail; 5872 if (avail < num_sectors) 5873 return -ENOSPC; 5874 } 5875 rv = mddev->pers->resize(mddev, num_sectors); 5876 if (!rv) 5877 revalidate_disk(mddev->gendisk); 5878 return rv; 5879 } 5880 5881 static int update_raid_disks(struct mddev *mddev, int raid_disks) 5882 { 5883 int rv; 5884 /* change the number of raid disks */ 5885 if (mddev->pers->check_reshape == NULL) 5886 return -EINVAL; 5887 if (raid_disks <= 0 || 5888 (mddev->max_disks && raid_disks >= mddev->max_disks)) 5889 return -EINVAL; 5890 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 5891 return -EBUSY; 5892 mddev->delta_disks = raid_disks - mddev->raid_disks; 5893 5894 rv = mddev->pers->check_reshape(mddev); 5895 if (rv < 0) 5896 mddev->delta_disks = 0; 5897 return rv; 5898 } 5899 5900 5901 /* 5902 * update_array_info is used to change the configuration of an 5903 * on-line array. 5904 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 5905 * fields in the info are checked against the array. 5906 * Any differences that cannot be handled will cause an error. 5907 * Normally, only one change can be managed at a time. 5908 */ 5909 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) 5910 { 5911 int rv = 0; 5912 int cnt = 0; 5913 int state = 0; 5914 5915 /* calculate expected state,ignoring low bits */ 5916 if (mddev->bitmap && mddev->bitmap_info.offset) 5917 state |= (1 << MD_SB_BITMAP_PRESENT); 5918 5919 if (mddev->major_version != info->major_version || 5920 mddev->minor_version != info->minor_version || 5921 /* mddev->patch_version != info->patch_version || */ 5922 mddev->ctime != info->ctime || 5923 mddev->level != info->level || 5924 /* mddev->layout != info->layout || */ 5925 !mddev->persistent != info->not_persistent|| 5926 mddev->chunk_sectors != info->chunk_size >> 9 || 5927 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 5928 ((state^info->state) & 0xfffffe00) 5929 ) 5930 return -EINVAL; 5931 /* Check there is only one change */ 5932 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5933 cnt++; 5934 if (mddev->raid_disks != info->raid_disks) 5935 cnt++; 5936 if (mddev->layout != info->layout) 5937 cnt++; 5938 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) 5939 cnt++; 5940 if (cnt == 0) 5941 return 0; 5942 if (cnt > 1) 5943 return -EINVAL; 5944 5945 if (mddev->layout != info->layout) { 5946 /* Change layout 5947 * we don't need to do anything at the md level, the 5948 * personality will take care of it all. 5949 */ 5950 if (mddev->pers->check_reshape == NULL) 5951 return -EINVAL; 5952 else { 5953 mddev->new_layout = info->layout; 5954 rv = mddev->pers->check_reshape(mddev); 5955 if (rv) 5956 mddev->new_layout = mddev->layout; 5957 return rv; 5958 } 5959 } 5960 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5961 rv = update_size(mddev, (sector_t)info->size * 2); 5962 5963 if (mddev->raid_disks != info->raid_disks) 5964 rv = update_raid_disks(mddev, info->raid_disks); 5965 5966 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 5967 if (mddev->pers->quiesce == NULL) 5968 return -EINVAL; 5969 if (mddev->recovery || mddev->sync_thread) 5970 return -EBUSY; 5971 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 5972 /* add the bitmap */ 5973 if (mddev->bitmap) 5974 return -EEXIST; 5975 if (mddev->bitmap_info.default_offset == 0) 5976 return -EINVAL; 5977 mddev->bitmap_info.offset = 5978 mddev->bitmap_info.default_offset; 5979 mddev->pers->quiesce(mddev, 1); 5980 rv = bitmap_create(mddev); 5981 if (!rv) 5982 rv = bitmap_load(mddev); 5983 if (rv) 5984 bitmap_destroy(mddev); 5985 mddev->pers->quiesce(mddev, 0); 5986 } else { 5987 /* remove the bitmap */ 5988 if (!mddev->bitmap) 5989 return -ENOENT; 5990 if (mddev->bitmap->file) 5991 return -EINVAL; 5992 mddev->pers->quiesce(mddev, 1); 5993 bitmap_destroy(mddev); 5994 mddev->pers->quiesce(mddev, 0); 5995 mddev->bitmap_info.offset = 0; 5996 } 5997 } 5998 md_update_sb(mddev, 1); 5999 return rv; 6000 } 6001 6002 static int set_disk_faulty(struct mddev *mddev, dev_t dev) 6003 { 6004 struct md_rdev *rdev; 6005 6006 if (mddev->pers == NULL) 6007 return -ENODEV; 6008 6009 rdev = find_rdev(mddev, dev); 6010 if (!rdev) 6011 return -ENODEV; 6012 6013 md_error(mddev, rdev); 6014 if (!test_bit(Faulty, &rdev->flags)) 6015 return -EBUSY; 6016 return 0; 6017 } 6018 6019 /* 6020 * We have a problem here : there is no easy way to give a CHS 6021 * virtual geometry. We currently pretend that we have a 2 heads 6022 * 4 sectors (with a BIG number of cylinders...). This drives 6023 * dosfs just mad... ;-) 6024 */ 6025 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) 6026 { 6027 struct mddev *mddev = bdev->bd_disk->private_data; 6028 6029 geo->heads = 2; 6030 geo->sectors = 4; 6031 geo->cylinders = mddev->array_sectors / 8; 6032 return 0; 6033 } 6034 6035 static int md_ioctl(struct block_device *bdev, fmode_t mode, 6036 unsigned int cmd, unsigned long arg) 6037 { 6038 int err = 0; 6039 void __user *argp = (void __user *)arg; 6040 struct mddev *mddev = NULL; 6041 int ro; 6042 6043 if (!capable(CAP_SYS_ADMIN)) 6044 return -EACCES; 6045 6046 /* 6047 * Commands dealing with the RAID driver but not any 6048 * particular array: 6049 */ 6050 switch (cmd) 6051 { 6052 case RAID_VERSION: 6053 err = get_version(argp); 6054 goto done; 6055 6056 case PRINT_RAID_DEBUG: 6057 err = 0; 6058 md_print_devices(); 6059 goto done; 6060 6061 #ifndef MODULE 6062 case RAID_AUTORUN: 6063 err = 0; 6064 autostart_arrays(arg); 6065 goto done; 6066 #endif 6067 default:; 6068 } 6069 6070 /* 6071 * Commands creating/starting a new array: 6072 */ 6073 6074 mddev = bdev->bd_disk->private_data; 6075 6076 if (!mddev) { 6077 BUG(); 6078 goto abort; 6079 } 6080 6081 err = mddev_lock(mddev); 6082 if (err) { 6083 printk(KERN_INFO 6084 "md: ioctl lock interrupted, reason %d, cmd %d\n", 6085 err, cmd); 6086 goto abort; 6087 } 6088 6089 switch (cmd) 6090 { 6091 case SET_ARRAY_INFO: 6092 { 6093 mdu_array_info_t info; 6094 if (!arg) 6095 memset(&info, 0, sizeof(info)); 6096 else if (copy_from_user(&info, argp, sizeof(info))) { 6097 err = -EFAULT; 6098 goto abort_unlock; 6099 } 6100 if (mddev->pers) { 6101 err = update_array_info(mddev, &info); 6102 if (err) { 6103 printk(KERN_WARNING "md: couldn't update" 6104 " array info. %d\n", err); 6105 goto abort_unlock; 6106 } 6107 goto done_unlock; 6108 } 6109 if (!list_empty(&mddev->disks)) { 6110 printk(KERN_WARNING 6111 "md: array %s already has disks!\n", 6112 mdname(mddev)); 6113 err = -EBUSY; 6114 goto abort_unlock; 6115 } 6116 if (mddev->raid_disks) { 6117 printk(KERN_WARNING 6118 "md: array %s already initialised!\n", 6119 mdname(mddev)); 6120 err = -EBUSY; 6121 goto abort_unlock; 6122 } 6123 err = set_array_info(mddev, &info); 6124 if (err) { 6125 printk(KERN_WARNING "md: couldn't set" 6126 " array info. %d\n", err); 6127 goto abort_unlock; 6128 } 6129 } 6130 goto done_unlock; 6131 6132 default:; 6133 } 6134 6135 /* 6136 * Commands querying/configuring an existing array: 6137 */ 6138 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 6139 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ 6140 if ((!mddev->raid_disks && !mddev->external) 6141 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 6142 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE 6143 && cmd != GET_BITMAP_FILE) { 6144 err = -ENODEV; 6145 goto abort_unlock; 6146 } 6147 6148 /* 6149 * Commands even a read-only array can execute: 6150 */ 6151 switch (cmd) 6152 { 6153 case GET_ARRAY_INFO: 6154 err = get_array_info(mddev, argp); 6155 goto done_unlock; 6156 6157 case GET_BITMAP_FILE: 6158 err = get_bitmap_file(mddev, argp); 6159 goto done_unlock; 6160 6161 case GET_DISK_INFO: 6162 err = get_disk_info(mddev, argp); 6163 goto done_unlock; 6164 6165 case RESTART_ARRAY_RW: 6166 err = restart_array(mddev); 6167 goto done_unlock; 6168 6169 case STOP_ARRAY: 6170 err = do_md_stop(mddev, 0, 1); 6171 goto done_unlock; 6172 6173 case STOP_ARRAY_RO: 6174 err = md_set_readonly(mddev, 1); 6175 goto done_unlock; 6176 6177 case BLKROSET: 6178 if (get_user(ro, (int __user *)(arg))) { 6179 err = -EFAULT; 6180 goto done_unlock; 6181 } 6182 err = -EINVAL; 6183 6184 /* if the bdev is going readonly the value of mddev->ro 6185 * does not matter, no writes are coming 6186 */ 6187 if (ro) 6188 goto done_unlock; 6189 6190 /* are we are already prepared for writes? */ 6191 if (mddev->ro != 1) 6192 goto done_unlock; 6193 6194 /* transitioning to readauto need only happen for 6195 * arrays that call md_write_start 6196 */ 6197 if (mddev->pers) { 6198 err = restart_array(mddev); 6199 if (err == 0) { 6200 mddev->ro = 2; 6201 set_disk_ro(mddev->gendisk, 0); 6202 } 6203 } 6204 goto done_unlock; 6205 } 6206 6207 /* 6208 * The remaining ioctls are changing the state of the 6209 * superblock, so we do not allow them on read-only arrays. 6210 * However non-MD ioctls (e.g. get-size) will still come through 6211 * here and hit the 'default' below, so only disallow 6212 * 'md' ioctls, and switch to rw mode if started auto-readonly. 6213 */ 6214 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { 6215 if (mddev->ro == 2) { 6216 mddev->ro = 0; 6217 sysfs_notify_dirent_safe(mddev->sysfs_state); 6218 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6219 md_wakeup_thread(mddev->thread); 6220 } else { 6221 err = -EROFS; 6222 goto abort_unlock; 6223 } 6224 } 6225 6226 switch (cmd) 6227 { 6228 case ADD_NEW_DISK: 6229 { 6230 mdu_disk_info_t info; 6231 if (copy_from_user(&info, argp, sizeof(info))) 6232 err = -EFAULT; 6233 else 6234 err = add_new_disk(mddev, &info); 6235 goto done_unlock; 6236 } 6237 6238 case HOT_REMOVE_DISK: 6239 err = hot_remove_disk(mddev, new_decode_dev(arg)); 6240 goto done_unlock; 6241 6242 case HOT_ADD_DISK: 6243 err = hot_add_disk(mddev, new_decode_dev(arg)); 6244 goto done_unlock; 6245 6246 case SET_DISK_FAULTY: 6247 err = set_disk_faulty(mddev, new_decode_dev(arg)); 6248 goto done_unlock; 6249 6250 case RUN_ARRAY: 6251 err = do_md_run(mddev); 6252 goto done_unlock; 6253 6254 case SET_BITMAP_FILE: 6255 err = set_bitmap_file(mddev, (int)arg); 6256 goto done_unlock; 6257 6258 default: 6259 err = -EINVAL; 6260 goto abort_unlock; 6261 } 6262 6263 done_unlock: 6264 abort_unlock: 6265 if (mddev->hold_active == UNTIL_IOCTL && 6266 err != -EINVAL) 6267 mddev->hold_active = 0; 6268 mddev_unlock(mddev); 6269 6270 return err; 6271 done: 6272 if (err) 6273 MD_BUG(); 6274 abort: 6275 return err; 6276 } 6277 #ifdef CONFIG_COMPAT 6278 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, 6279 unsigned int cmd, unsigned long arg) 6280 { 6281 switch (cmd) { 6282 case HOT_REMOVE_DISK: 6283 case HOT_ADD_DISK: 6284 case SET_DISK_FAULTY: 6285 case SET_BITMAP_FILE: 6286 /* These take in integer arg, do not convert */ 6287 break; 6288 default: 6289 arg = (unsigned long)compat_ptr(arg); 6290 break; 6291 } 6292 6293 return md_ioctl(bdev, mode, cmd, arg); 6294 } 6295 #endif /* CONFIG_COMPAT */ 6296 6297 static int md_open(struct block_device *bdev, fmode_t mode) 6298 { 6299 /* 6300 * Succeed if we can lock the mddev, which confirms that 6301 * it isn't being stopped right now. 6302 */ 6303 struct mddev *mddev = mddev_find(bdev->bd_dev); 6304 int err; 6305 6306 if (mddev->gendisk != bdev->bd_disk) { 6307 /* we are racing with mddev_put which is discarding this 6308 * bd_disk. 6309 */ 6310 mddev_put(mddev); 6311 /* Wait until bdev->bd_disk is definitely gone */ 6312 flush_workqueue(md_misc_wq); 6313 /* Then retry the open from the top */ 6314 return -ERESTARTSYS; 6315 } 6316 BUG_ON(mddev != bdev->bd_disk->private_data); 6317 6318 if ((err = mutex_lock_interruptible(&mddev->open_mutex))) 6319 goto out; 6320 6321 err = 0; 6322 atomic_inc(&mddev->openers); 6323 mutex_unlock(&mddev->open_mutex); 6324 6325 check_disk_change(bdev); 6326 out: 6327 return err; 6328 } 6329 6330 static int md_release(struct gendisk *disk, fmode_t mode) 6331 { 6332 struct mddev *mddev = disk->private_data; 6333 6334 BUG_ON(!mddev); 6335 atomic_dec(&mddev->openers); 6336 mddev_put(mddev); 6337 6338 return 0; 6339 } 6340 6341 static int md_media_changed(struct gendisk *disk) 6342 { 6343 struct mddev *mddev = disk->private_data; 6344 6345 return mddev->changed; 6346 } 6347 6348 static int md_revalidate(struct gendisk *disk) 6349 { 6350 struct mddev *mddev = disk->private_data; 6351 6352 mddev->changed = 0; 6353 return 0; 6354 } 6355 static const struct block_device_operations md_fops = 6356 { 6357 .owner = THIS_MODULE, 6358 .open = md_open, 6359 .release = md_release, 6360 .ioctl = md_ioctl, 6361 #ifdef CONFIG_COMPAT 6362 .compat_ioctl = md_compat_ioctl, 6363 #endif 6364 .getgeo = md_getgeo, 6365 .media_changed = md_media_changed, 6366 .revalidate_disk= md_revalidate, 6367 }; 6368 6369 static int md_thread(void * arg) 6370 { 6371 struct md_thread *thread = arg; 6372 6373 /* 6374 * md_thread is a 'system-thread', it's priority should be very 6375 * high. We avoid resource deadlocks individually in each 6376 * raid personality. (RAID5 does preallocation) We also use RR and 6377 * the very same RT priority as kswapd, thus we will never get 6378 * into a priority inversion deadlock. 6379 * 6380 * we definitely have to have equal or higher priority than 6381 * bdflush, otherwise bdflush will deadlock if there are too 6382 * many dirty RAID5 blocks. 6383 */ 6384 6385 allow_signal(SIGKILL); 6386 while (!kthread_should_stop()) { 6387 6388 /* We need to wait INTERRUPTIBLE so that 6389 * we don't add to the load-average. 6390 * That means we need to be sure no signals are 6391 * pending 6392 */ 6393 if (signal_pending(current)) 6394 flush_signals(current); 6395 6396 wait_event_interruptible_timeout 6397 (thread->wqueue, 6398 test_bit(THREAD_WAKEUP, &thread->flags) 6399 || kthread_should_stop(), 6400 thread->timeout); 6401 6402 clear_bit(THREAD_WAKEUP, &thread->flags); 6403 if (!kthread_should_stop()) 6404 thread->run(thread->mddev); 6405 } 6406 6407 return 0; 6408 } 6409 6410 void md_wakeup_thread(struct md_thread *thread) 6411 { 6412 if (thread) { 6413 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); 6414 set_bit(THREAD_WAKEUP, &thread->flags); 6415 wake_up(&thread->wqueue); 6416 } 6417 } 6418 6419 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev, 6420 const char *name) 6421 { 6422 struct md_thread *thread; 6423 6424 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); 6425 if (!thread) 6426 return NULL; 6427 6428 init_waitqueue_head(&thread->wqueue); 6429 6430 thread->run = run; 6431 thread->mddev = mddev; 6432 thread->timeout = MAX_SCHEDULE_TIMEOUT; 6433 thread->tsk = kthread_run(md_thread, thread, 6434 "%s_%s", 6435 mdname(thread->mddev), 6436 name ?: mddev->pers->name); 6437 if (IS_ERR(thread->tsk)) { 6438 kfree(thread); 6439 return NULL; 6440 } 6441 return thread; 6442 } 6443 6444 void md_unregister_thread(struct md_thread **threadp) 6445 { 6446 struct md_thread *thread = *threadp; 6447 if (!thread) 6448 return; 6449 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); 6450 /* Locking ensures that mddev_unlock does not wake_up a 6451 * non-existent thread 6452 */ 6453 spin_lock(&pers_lock); 6454 *threadp = NULL; 6455 spin_unlock(&pers_lock); 6456 6457 kthread_stop(thread->tsk); 6458 kfree(thread); 6459 } 6460 6461 void md_error(struct mddev *mddev, struct md_rdev *rdev) 6462 { 6463 if (!mddev) { 6464 MD_BUG(); 6465 return; 6466 } 6467 6468 if (!rdev || test_bit(Faulty, &rdev->flags)) 6469 return; 6470 6471 if (!mddev->pers || !mddev->pers->error_handler) 6472 return; 6473 mddev->pers->error_handler(mddev,rdev); 6474 if (mddev->degraded) 6475 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6476 sysfs_notify_dirent_safe(rdev->sysfs_state); 6477 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6478 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6479 md_wakeup_thread(mddev->thread); 6480 if (mddev->event_work.func) 6481 queue_work(md_misc_wq, &mddev->event_work); 6482 md_new_event_inintr(mddev); 6483 } 6484 6485 /* seq_file implementation /proc/mdstat */ 6486 6487 static void status_unused(struct seq_file *seq) 6488 { 6489 int i = 0; 6490 struct md_rdev *rdev; 6491 6492 seq_printf(seq, "unused devices: "); 6493 6494 list_for_each_entry(rdev, &pending_raid_disks, same_set) { 6495 char b[BDEVNAME_SIZE]; 6496 i++; 6497 seq_printf(seq, "%s ", 6498 bdevname(rdev->bdev,b)); 6499 } 6500 if (!i) 6501 seq_printf(seq, "<none>"); 6502 6503 seq_printf(seq, "\n"); 6504 } 6505 6506 6507 static void status_resync(struct seq_file *seq, struct mddev * mddev) 6508 { 6509 sector_t max_sectors, resync, res; 6510 unsigned long dt, db; 6511 sector_t rt; 6512 int scale; 6513 unsigned int per_milli; 6514 6515 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); 6516 6517 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 6518 max_sectors = mddev->resync_max_sectors; 6519 else 6520 max_sectors = mddev->dev_sectors; 6521 6522 /* 6523 * Should not happen. 6524 */ 6525 if (!max_sectors) { 6526 MD_BUG(); 6527 return; 6528 } 6529 /* Pick 'scale' such that (resync>>scale)*1000 will fit 6530 * in a sector_t, and (max_sectors>>scale) will fit in a 6531 * u32, as those are the requirements for sector_div. 6532 * Thus 'scale' must be at least 10 6533 */ 6534 scale = 10; 6535 if (sizeof(sector_t) > sizeof(unsigned long)) { 6536 while ( max_sectors/2 > (1ULL<<(scale+32))) 6537 scale++; 6538 } 6539 res = (resync>>scale)*1000; 6540 sector_div(res, (u32)((max_sectors>>scale)+1)); 6541 6542 per_milli = res; 6543 { 6544 int i, x = per_milli/50, y = 20-x; 6545 seq_printf(seq, "["); 6546 for (i = 0; i < x; i++) 6547 seq_printf(seq, "="); 6548 seq_printf(seq, ">"); 6549 for (i = 0; i < y; i++) 6550 seq_printf(seq, "."); 6551 seq_printf(seq, "] "); 6552 } 6553 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", 6554 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? 6555 "reshape" : 6556 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? 6557 "check" : 6558 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 6559 "resync" : "recovery"))), 6560 per_milli/10, per_milli % 10, 6561 (unsigned long long) resync/2, 6562 (unsigned long long) max_sectors/2); 6563 6564 /* 6565 * dt: time from mark until now 6566 * db: blocks written from mark until now 6567 * rt: remaining time 6568 * 6569 * rt is a sector_t, so could be 32bit or 64bit. 6570 * So we divide before multiply in case it is 32bit and close 6571 * to the limit. 6572 * We scale the divisor (db) by 32 to avoid losing precision 6573 * near the end of resync when the number of remaining sectors 6574 * is close to 'db'. 6575 * We then divide rt by 32 after multiplying by db to compensate. 6576 * The '+1' avoids division by zero if db is very small. 6577 */ 6578 dt = ((jiffies - mddev->resync_mark) / HZ); 6579 if (!dt) dt++; 6580 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) 6581 - mddev->resync_mark_cnt; 6582 6583 rt = max_sectors - resync; /* number of remaining sectors */ 6584 sector_div(rt, db/32+1); 6585 rt *= dt; 6586 rt >>= 5; 6587 6588 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, 6589 ((unsigned long)rt % 60)/6); 6590 6591 seq_printf(seq, " speed=%ldK/sec", db/2/dt); 6592 } 6593 6594 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 6595 { 6596 struct list_head *tmp; 6597 loff_t l = *pos; 6598 struct mddev *mddev; 6599 6600 if (l >= 0x10000) 6601 return NULL; 6602 if (!l--) 6603 /* header */ 6604 return (void*)1; 6605 6606 spin_lock(&all_mddevs_lock); 6607 list_for_each(tmp,&all_mddevs) 6608 if (!l--) { 6609 mddev = list_entry(tmp, struct mddev, all_mddevs); 6610 mddev_get(mddev); 6611 spin_unlock(&all_mddevs_lock); 6612 return mddev; 6613 } 6614 spin_unlock(&all_mddevs_lock); 6615 if (!l--) 6616 return (void*)2;/* tail */ 6617 return NULL; 6618 } 6619 6620 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 6621 { 6622 struct list_head *tmp; 6623 struct mddev *next_mddev, *mddev = v; 6624 6625 ++*pos; 6626 if (v == (void*)2) 6627 return NULL; 6628 6629 spin_lock(&all_mddevs_lock); 6630 if (v == (void*)1) 6631 tmp = all_mddevs.next; 6632 else 6633 tmp = mddev->all_mddevs.next; 6634 if (tmp != &all_mddevs) 6635 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); 6636 else { 6637 next_mddev = (void*)2; 6638 *pos = 0x10000; 6639 } 6640 spin_unlock(&all_mddevs_lock); 6641 6642 if (v != (void*)1) 6643 mddev_put(mddev); 6644 return next_mddev; 6645 6646 } 6647 6648 static void md_seq_stop(struct seq_file *seq, void *v) 6649 { 6650 struct mddev *mddev = v; 6651 6652 if (mddev && v != (void*)1 && v != (void*)2) 6653 mddev_put(mddev); 6654 } 6655 6656 static int md_seq_show(struct seq_file *seq, void *v) 6657 { 6658 struct mddev *mddev = v; 6659 sector_t sectors; 6660 struct md_rdev *rdev; 6661 struct bitmap *bitmap; 6662 6663 if (v == (void*)1) { 6664 struct md_personality *pers; 6665 seq_printf(seq, "Personalities : "); 6666 spin_lock(&pers_lock); 6667 list_for_each_entry(pers, &pers_list, list) 6668 seq_printf(seq, "[%s] ", pers->name); 6669 6670 spin_unlock(&pers_lock); 6671 seq_printf(seq, "\n"); 6672 seq->poll_event = atomic_read(&md_event_count); 6673 return 0; 6674 } 6675 if (v == (void*)2) { 6676 status_unused(seq); 6677 return 0; 6678 } 6679 6680 if (mddev_lock(mddev) < 0) 6681 return -EINTR; 6682 6683 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 6684 seq_printf(seq, "%s : %sactive", mdname(mddev), 6685 mddev->pers ? "" : "in"); 6686 if (mddev->pers) { 6687 if (mddev->ro==1) 6688 seq_printf(seq, " (read-only)"); 6689 if (mddev->ro==2) 6690 seq_printf(seq, " (auto-read-only)"); 6691 seq_printf(seq, " %s", mddev->pers->name); 6692 } 6693 6694 sectors = 0; 6695 list_for_each_entry(rdev, &mddev->disks, same_set) { 6696 char b[BDEVNAME_SIZE]; 6697 seq_printf(seq, " %s[%d]", 6698 bdevname(rdev->bdev,b), rdev->desc_nr); 6699 if (test_bit(WriteMostly, &rdev->flags)) 6700 seq_printf(seq, "(W)"); 6701 if (test_bit(Faulty, &rdev->flags)) { 6702 seq_printf(seq, "(F)"); 6703 continue; 6704 } else if (rdev->raid_disk < 0) 6705 seq_printf(seq, "(S)"); /* spare */ 6706 sectors += rdev->sectors; 6707 } 6708 6709 if (!list_empty(&mddev->disks)) { 6710 if (mddev->pers) 6711 seq_printf(seq, "\n %llu blocks", 6712 (unsigned long long) 6713 mddev->array_sectors / 2); 6714 else 6715 seq_printf(seq, "\n %llu blocks", 6716 (unsigned long long)sectors / 2); 6717 } 6718 if (mddev->persistent) { 6719 if (mddev->major_version != 0 || 6720 mddev->minor_version != 90) { 6721 seq_printf(seq," super %d.%d", 6722 mddev->major_version, 6723 mddev->minor_version); 6724 } 6725 } else if (mddev->external) 6726 seq_printf(seq, " super external:%s", 6727 mddev->metadata_type); 6728 else 6729 seq_printf(seq, " super non-persistent"); 6730 6731 if (mddev->pers) { 6732 mddev->pers->status(seq, mddev); 6733 seq_printf(seq, "\n "); 6734 if (mddev->pers->sync_request) { 6735 if (mddev->curr_resync > 2) { 6736 status_resync(seq, mddev); 6737 seq_printf(seq, "\n "); 6738 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 6739 seq_printf(seq, "\tresync=DELAYED\n "); 6740 else if (mddev->recovery_cp < MaxSector) 6741 seq_printf(seq, "\tresync=PENDING\n "); 6742 } 6743 } else 6744 seq_printf(seq, "\n "); 6745 6746 if ((bitmap = mddev->bitmap)) { 6747 unsigned long chunk_kb; 6748 unsigned long flags; 6749 spin_lock_irqsave(&bitmap->lock, flags); 6750 chunk_kb = mddev->bitmap_info.chunksize >> 10; 6751 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 6752 "%lu%s chunk", 6753 bitmap->pages - bitmap->missing_pages, 6754 bitmap->pages, 6755 (bitmap->pages - bitmap->missing_pages) 6756 << (PAGE_SHIFT - 10), 6757 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, 6758 chunk_kb ? "KB" : "B"); 6759 if (bitmap->file) { 6760 seq_printf(seq, ", file: "); 6761 seq_path(seq, &bitmap->file->f_path, " \t\n"); 6762 } 6763 6764 seq_printf(seq, "\n"); 6765 spin_unlock_irqrestore(&bitmap->lock, flags); 6766 } 6767 6768 seq_printf(seq, "\n"); 6769 } 6770 mddev_unlock(mddev); 6771 6772 return 0; 6773 } 6774 6775 static const struct seq_operations md_seq_ops = { 6776 .start = md_seq_start, 6777 .next = md_seq_next, 6778 .stop = md_seq_stop, 6779 .show = md_seq_show, 6780 }; 6781 6782 static int md_seq_open(struct inode *inode, struct file *file) 6783 { 6784 struct seq_file *seq; 6785 int error; 6786 6787 error = seq_open(file, &md_seq_ops); 6788 if (error) 6789 return error; 6790 6791 seq = file->private_data; 6792 seq->poll_event = atomic_read(&md_event_count); 6793 return error; 6794 } 6795 6796 static unsigned int mdstat_poll(struct file *filp, poll_table *wait) 6797 { 6798 struct seq_file *seq = filp->private_data; 6799 int mask; 6800 6801 poll_wait(filp, &md_event_waiters, wait); 6802 6803 /* always allow read */ 6804 mask = POLLIN | POLLRDNORM; 6805 6806 if (seq->poll_event != atomic_read(&md_event_count)) 6807 mask |= POLLERR | POLLPRI; 6808 return mask; 6809 } 6810 6811 static const struct file_operations md_seq_fops = { 6812 .owner = THIS_MODULE, 6813 .open = md_seq_open, 6814 .read = seq_read, 6815 .llseek = seq_lseek, 6816 .release = seq_release_private, 6817 .poll = mdstat_poll, 6818 }; 6819 6820 int register_md_personality(struct md_personality *p) 6821 { 6822 spin_lock(&pers_lock); 6823 list_add_tail(&p->list, &pers_list); 6824 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); 6825 spin_unlock(&pers_lock); 6826 return 0; 6827 } 6828 6829 int unregister_md_personality(struct md_personality *p) 6830 { 6831 printk(KERN_INFO "md: %s personality unregistered\n", p->name); 6832 spin_lock(&pers_lock); 6833 list_del_init(&p->list); 6834 spin_unlock(&pers_lock); 6835 return 0; 6836 } 6837 6838 static int is_mddev_idle(struct mddev *mddev, int init) 6839 { 6840 struct md_rdev * rdev; 6841 int idle; 6842 int curr_events; 6843 6844 idle = 1; 6845 rcu_read_lock(); 6846 rdev_for_each_rcu(rdev, mddev) { 6847 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 6848 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 6849 (int)part_stat_read(&disk->part0, sectors[1]) - 6850 atomic_read(&disk->sync_io); 6851 /* sync IO will cause sync_io to increase before the disk_stats 6852 * as sync_io is counted when a request starts, and 6853 * disk_stats is counted when it completes. 6854 * So resync activity will cause curr_events to be smaller than 6855 * when there was no such activity. 6856 * non-sync IO will cause disk_stat to increase without 6857 * increasing sync_io so curr_events will (eventually) 6858 * be larger than it was before. Once it becomes 6859 * substantially larger, the test below will cause 6860 * the array to appear non-idle, and resync will slow 6861 * down. 6862 * If there is a lot of outstanding resync activity when 6863 * we set last_event to curr_events, then all that activity 6864 * completing might cause the array to appear non-idle 6865 * and resync will be slowed down even though there might 6866 * not have been non-resync activity. This will only 6867 * happen once though. 'last_events' will soon reflect 6868 * the state where there is little or no outstanding 6869 * resync requests, and further resync activity will 6870 * always make curr_events less than last_events. 6871 * 6872 */ 6873 if (init || curr_events - rdev->last_events > 64) { 6874 rdev->last_events = curr_events; 6875 idle = 0; 6876 } 6877 } 6878 rcu_read_unlock(); 6879 return idle; 6880 } 6881 6882 void md_done_sync(struct mddev *mddev, int blocks, int ok) 6883 { 6884 /* another "blocks" (512byte) blocks have been synced */ 6885 atomic_sub(blocks, &mddev->recovery_active); 6886 wake_up(&mddev->recovery_wait); 6887 if (!ok) { 6888 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6889 md_wakeup_thread(mddev->thread); 6890 // stop recovery, signal do_sync .... 6891 } 6892 } 6893 6894 6895 /* md_write_start(mddev, bi) 6896 * If we need to update some array metadata (e.g. 'active' flag 6897 * in superblock) before writing, schedule a superblock update 6898 * and wait for it to complete. 6899 */ 6900 void md_write_start(struct mddev *mddev, struct bio *bi) 6901 { 6902 int did_change = 0; 6903 if (bio_data_dir(bi) != WRITE) 6904 return; 6905 6906 BUG_ON(mddev->ro == 1); 6907 if (mddev->ro == 2) { 6908 /* need to switch to read/write */ 6909 mddev->ro = 0; 6910 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6911 md_wakeup_thread(mddev->thread); 6912 md_wakeup_thread(mddev->sync_thread); 6913 did_change = 1; 6914 } 6915 atomic_inc(&mddev->writes_pending); 6916 if (mddev->safemode == 1) 6917 mddev->safemode = 0; 6918 if (mddev->in_sync) { 6919 spin_lock_irq(&mddev->write_lock); 6920 if (mddev->in_sync) { 6921 mddev->in_sync = 0; 6922 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6923 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6924 md_wakeup_thread(mddev->thread); 6925 did_change = 1; 6926 } 6927 spin_unlock_irq(&mddev->write_lock); 6928 } 6929 if (did_change) 6930 sysfs_notify_dirent_safe(mddev->sysfs_state); 6931 wait_event(mddev->sb_wait, 6932 !test_bit(MD_CHANGE_PENDING, &mddev->flags)); 6933 } 6934 6935 void md_write_end(struct mddev *mddev) 6936 { 6937 if (atomic_dec_and_test(&mddev->writes_pending)) { 6938 if (mddev->safemode == 2) 6939 md_wakeup_thread(mddev->thread); 6940 else if (mddev->safemode_delay) 6941 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 6942 } 6943 } 6944 6945 /* md_allow_write(mddev) 6946 * Calling this ensures that the array is marked 'active' so that writes 6947 * may proceed without blocking. It is important to call this before 6948 * attempting a GFP_KERNEL allocation while holding the mddev lock. 6949 * Must be called with mddev_lock held. 6950 * 6951 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock 6952 * is dropped, so return -EAGAIN after notifying userspace. 6953 */ 6954 int md_allow_write(struct mddev *mddev) 6955 { 6956 if (!mddev->pers) 6957 return 0; 6958 if (mddev->ro) 6959 return 0; 6960 if (!mddev->pers->sync_request) 6961 return 0; 6962 6963 spin_lock_irq(&mddev->write_lock); 6964 if (mddev->in_sync) { 6965 mddev->in_sync = 0; 6966 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6967 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6968 if (mddev->safemode_delay && 6969 mddev->safemode == 0) 6970 mddev->safemode = 1; 6971 spin_unlock_irq(&mddev->write_lock); 6972 md_update_sb(mddev, 0); 6973 sysfs_notify_dirent_safe(mddev->sysfs_state); 6974 } else 6975 spin_unlock_irq(&mddev->write_lock); 6976 6977 if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 6978 return -EAGAIN; 6979 else 6980 return 0; 6981 } 6982 EXPORT_SYMBOL_GPL(md_allow_write); 6983 6984 #define SYNC_MARKS 10 6985 #define SYNC_MARK_STEP (3*HZ) 6986 void md_do_sync(struct mddev *mddev) 6987 { 6988 struct mddev *mddev2; 6989 unsigned int currspeed = 0, 6990 window; 6991 sector_t max_sectors,j, io_sectors; 6992 unsigned long mark[SYNC_MARKS]; 6993 sector_t mark_cnt[SYNC_MARKS]; 6994 int last_mark,m; 6995 struct list_head *tmp; 6996 sector_t last_check; 6997 int skipped = 0; 6998 struct md_rdev *rdev; 6999 char *desc; 7000 7001 /* just incase thread restarts... */ 7002 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 7003 return; 7004 if (mddev->ro) /* never try to sync a read-only array */ 7005 return; 7006 7007 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7008 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 7009 desc = "data-check"; 7010 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7011 desc = "requested-resync"; 7012 else 7013 desc = "resync"; 7014 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 7015 desc = "reshape"; 7016 else 7017 desc = "recovery"; 7018 7019 /* we overload curr_resync somewhat here. 7020 * 0 == not engaged in resync at all 7021 * 2 == checking that there is no conflict with another sync 7022 * 1 == like 2, but have yielded to allow conflicting resync to 7023 * commense 7024 * other == active in resync - this many blocks 7025 * 7026 * Before starting a resync we must have set curr_resync to 7027 * 2, and then checked that every "conflicting" array has curr_resync 7028 * less than ours. When we find one that is the same or higher 7029 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 7030 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 7031 * This will mean we have to start checking from the beginning again. 7032 * 7033 */ 7034 7035 do { 7036 mddev->curr_resync = 2; 7037 7038 try_again: 7039 if (kthread_should_stop()) 7040 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7041 7042 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7043 goto skip; 7044 for_each_mddev(mddev2, tmp) { 7045 if (mddev2 == mddev) 7046 continue; 7047 if (!mddev->parallel_resync 7048 && mddev2->curr_resync 7049 && match_mddev_units(mddev, mddev2)) { 7050 DEFINE_WAIT(wq); 7051 if (mddev < mddev2 && mddev->curr_resync == 2) { 7052 /* arbitrarily yield */ 7053 mddev->curr_resync = 1; 7054 wake_up(&resync_wait); 7055 } 7056 if (mddev > mddev2 && mddev->curr_resync == 1) 7057 /* no need to wait here, we can wait the next 7058 * time 'round when curr_resync == 2 7059 */ 7060 continue; 7061 /* We need to wait 'interruptible' so as not to 7062 * contribute to the load average, and not to 7063 * be caught by 'softlockup' 7064 */ 7065 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 7066 if (!kthread_should_stop() && 7067 mddev2->curr_resync >= mddev->curr_resync) { 7068 printk(KERN_INFO "md: delaying %s of %s" 7069 " until %s has finished (they" 7070 " share one or more physical units)\n", 7071 desc, mdname(mddev), mdname(mddev2)); 7072 mddev_put(mddev2); 7073 if (signal_pending(current)) 7074 flush_signals(current); 7075 schedule(); 7076 finish_wait(&resync_wait, &wq); 7077 goto try_again; 7078 } 7079 finish_wait(&resync_wait, &wq); 7080 } 7081 } 7082 } while (mddev->curr_resync < 2); 7083 7084 j = 0; 7085 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7086 /* resync follows the size requested by the personality, 7087 * which defaults to physical size, but can be virtual size 7088 */ 7089 max_sectors = mddev->resync_max_sectors; 7090 mddev->resync_mismatches = 0; 7091 /* we don't use the checkpoint if there's a bitmap */ 7092 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7093 j = mddev->resync_min; 7094 else if (!mddev->bitmap) 7095 j = mddev->recovery_cp; 7096 7097 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 7098 max_sectors = mddev->dev_sectors; 7099 else { 7100 /* recovery follows the physical size of devices */ 7101 max_sectors = mddev->dev_sectors; 7102 j = MaxSector; 7103 rcu_read_lock(); 7104 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 7105 if (rdev->raid_disk >= 0 && 7106 !test_bit(Faulty, &rdev->flags) && 7107 !test_bit(In_sync, &rdev->flags) && 7108 rdev->recovery_offset < j) 7109 j = rdev->recovery_offset; 7110 rcu_read_unlock(); 7111 } 7112 7113 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); 7114 printk(KERN_INFO "md: minimum _guaranteed_ speed:" 7115 " %d KB/sec/disk.\n", speed_min(mddev)); 7116 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 7117 "(but not more than %d KB/sec) for %s.\n", 7118 speed_max(mddev), desc); 7119 7120 is_mddev_idle(mddev, 1); /* this initializes IO event counters */ 7121 7122 io_sectors = 0; 7123 for (m = 0; m < SYNC_MARKS; m++) { 7124 mark[m] = jiffies; 7125 mark_cnt[m] = io_sectors; 7126 } 7127 last_mark = 0; 7128 mddev->resync_mark = mark[last_mark]; 7129 mddev->resync_mark_cnt = mark_cnt[last_mark]; 7130 7131 /* 7132 * Tune reconstruction: 7133 */ 7134 window = 32*(PAGE_SIZE/512); 7135 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", 7136 window/2, (unsigned long long)max_sectors/2); 7137 7138 atomic_set(&mddev->recovery_active, 0); 7139 last_check = 0; 7140 7141 if (j>2) { 7142 printk(KERN_INFO 7143 "md: resuming %s of %s from checkpoint.\n", 7144 desc, mdname(mddev)); 7145 mddev->curr_resync = j; 7146 } 7147 mddev->curr_resync_completed = j; 7148 7149 while (j < max_sectors) { 7150 sector_t sectors; 7151 7152 skipped = 0; 7153 7154 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7155 ((mddev->curr_resync > mddev->curr_resync_completed && 7156 (mddev->curr_resync - mddev->curr_resync_completed) 7157 > (max_sectors >> 4)) || 7158 (j - mddev->curr_resync_completed)*2 7159 >= mddev->resync_max - mddev->curr_resync_completed 7160 )) { 7161 /* time to update curr_resync_completed */ 7162 wait_event(mddev->recovery_wait, 7163 atomic_read(&mddev->recovery_active) == 0); 7164 mddev->curr_resync_completed = j; 7165 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7166 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 7167 } 7168 7169 while (j >= mddev->resync_max && !kthread_should_stop()) { 7170 /* As this condition is controlled by user-space, 7171 * we can block indefinitely, so use '_interruptible' 7172 * to avoid triggering warnings. 7173 */ 7174 flush_signals(current); /* just in case */ 7175 wait_event_interruptible(mddev->recovery_wait, 7176 mddev->resync_max > j 7177 || kthread_should_stop()); 7178 } 7179 7180 if (kthread_should_stop()) 7181 goto interrupted; 7182 7183 sectors = mddev->pers->sync_request(mddev, j, &skipped, 7184 currspeed < speed_min(mddev)); 7185 if (sectors == 0) { 7186 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7187 goto out; 7188 } 7189 7190 if (!skipped) { /* actual IO requested */ 7191 io_sectors += sectors; 7192 atomic_add(sectors, &mddev->recovery_active); 7193 } 7194 7195 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7196 break; 7197 7198 j += sectors; 7199 if (j>1) mddev->curr_resync = j; 7200 mddev->curr_mark_cnt = io_sectors; 7201 if (last_check == 0) 7202 /* this is the earliest that rebuild will be 7203 * visible in /proc/mdstat 7204 */ 7205 md_new_event(mddev); 7206 7207 if (last_check + window > io_sectors || j == max_sectors) 7208 continue; 7209 7210 last_check = io_sectors; 7211 repeat: 7212 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 7213 /* step marks */ 7214 int next = (last_mark+1) % SYNC_MARKS; 7215 7216 mddev->resync_mark = mark[next]; 7217 mddev->resync_mark_cnt = mark_cnt[next]; 7218 mark[next] = jiffies; 7219 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 7220 last_mark = next; 7221 } 7222 7223 7224 if (kthread_should_stop()) 7225 goto interrupted; 7226 7227 7228 /* 7229 * this loop exits only if either when we are slower than 7230 * the 'hard' speed limit, or the system was IO-idle for 7231 * a jiffy. 7232 * the system might be non-idle CPU-wise, but we only care 7233 * about not overloading the IO subsystem. (things like an 7234 * e2fsck being done on the RAID array should execute fast) 7235 */ 7236 cond_resched(); 7237 7238 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 7239 /((jiffies-mddev->resync_mark)/HZ +1) +1; 7240 7241 if (currspeed > speed_min(mddev)) { 7242 if ((currspeed > speed_max(mddev)) || 7243 !is_mddev_idle(mddev, 0)) { 7244 msleep(500); 7245 goto repeat; 7246 } 7247 } 7248 } 7249 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); 7250 /* 7251 * this also signals 'finished resyncing' to md_stop 7252 */ 7253 out: 7254 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 7255 7256 /* tell personality that we are finished */ 7257 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 7258 7259 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && 7260 mddev->curr_resync > 2) { 7261 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7262 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 7263 if (mddev->curr_resync >= mddev->recovery_cp) { 7264 printk(KERN_INFO 7265 "md: checkpointing %s of %s.\n", 7266 desc, mdname(mddev)); 7267 mddev->recovery_cp = mddev->curr_resync; 7268 } 7269 } else 7270 mddev->recovery_cp = MaxSector; 7271 } else { 7272 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7273 mddev->curr_resync = MaxSector; 7274 rcu_read_lock(); 7275 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 7276 if (rdev->raid_disk >= 0 && 7277 mddev->delta_disks >= 0 && 7278 !test_bit(Faulty, &rdev->flags) && 7279 !test_bit(In_sync, &rdev->flags) && 7280 rdev->recovery_offset < mddev->curr_resync) 7281 rdev->recovery_offset = mddev->curr_resync; 7282 rcu_read_unlock(); 7283 } 7284 } 7285 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7286 7287 skip: 7288 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 7289 /* We completed so min/max setting can be forgotten if used. */ 7290 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7291 mddev->resync_min = 0; 7292 mddev->resync_max = MaxSector; 7293 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7294 mddev->resync_min = mddev->curr_resync_completed; 7295 mddev->curr_resync = 0; 7296 wake_up(&resync_wait); 7297 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 7298 md_wakeup_thread(mddev->thread); 7299 return; 7300 7301 interrupted: 7302 /* 7303 * got a signal, exit. 7304 */ 7305 printk(KERN_INFO 7306 "md: md_do_sync() got signal ... exiting\n"); 7307 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7308 goto out; 7309 7310 } 7311 EXPORT_SYMBOL_GPL(md_do_sync); 7312 7313 static int remove_and_add_spares(struct mddev *mddev) 7314 { 7315 struct md_rdev *rdev; 7316 int spares = 0; 7317 7318 mddev->curr_resync_completed = 0; 7319 7320 list_for_each_entry(rdev, &mddev->disks, same_set) 7321 if (rdev->raid_disk >= 0 && 7322 !test_bit(Blocked, &rdev->flags) && 7323 (test_bit(Faulty, &rdev->flags) || 7324 ! test_bit(In_sync, &rdev->flags)) && 7325 atomic_read(&rdev->nr_pending)==0) { 7326 if (mddev->pers->hot_remove_disk( 7327 mddev, rdev->raid_disk)==0) { 7328 sysfs_unlink_rdev(mddev, rdev); 7329 rdev->raid_disk = -1; 7330 } 7331 } 7332 7333 if (mddev->degraded) { 7334 list_for_each_entry(rdev, &mddev->disks, same_set) { 7335 if (rdev->raid_disk >= 0 && 7336 !test_bit(In_sync, &rdev->flags) && 7337 !test_bit(Faulty, &rdev->flags)) 7338 spares++; 7339 if (rdev->raid_disk < 0 7340 && !test_bit(Faulty, &rdev->flags)) { 7341 rdev->recovery_offset = 0; 7342 if (mddev->pers-> 7343 hot_add_disk(mddev, rdev) == 0) { 7344 if (sysfs_link_rdev(mddev, rdev)) 7345 /* failure here is OK */; 7346 spares++; 7347 md_new_event(mddev); 7348 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7349 } else 7350 break; 7351 } 7352 } 7353 } 7354 return spares; 7355 } 7356 7357 static void reap_sync_thread(struct mddev *mddev) 7358 { 7359 struct md_rdev *rdev; 7360 7361 /* resync has finished, collect result */ 7362 md_unregister_thread(&mddev->sync_thread); 7363 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && 7364 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 7365 /* success...*/ 7366 /* activate any spares */ 7367 if (mddev->pers->spare_active(mddev)) 7368 sysfs_notify(&mddev->kobj, NULL, 7369 "degraded"); 7370 } 7371 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7372 mddev->pers->finish_reshape) 7373 mddev->pers->finish_reshape(mddev); 7374 7375 /* If array is no-longer degraded, then any saved_raid_disk 7376 * information must be scrapped. Also if any device is now 7377 * In_sync we must scrape the saved_raid_disk for that device 7378 * do the superblock for an incrementally recovered device 7379 * written out. 7380 */ 7381 list_for_each_entry(rdev, &mddev->disks, same_set) 7382 if (!mddev->degraded || 7383 test_bit(In_sync, &rdev->flags)) 7384 rdev->saved_raid_disk = -1; 7385 7386 md_update_sb(mddev, 1); 7387 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7388 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7389 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7390 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7391 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7392 /* flag recovery needed just to double check */ 7393 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7394 sysfs_notify_dirent_safe(mddev->sysfs_action); 7395 md_new_event(mddev); 7396 if (mddev->event_work.func) 7397 queue_work(md_misc_wq, &mddev->event_work); 7398 } 7399 7400 /* 7401 * This routine is regularly called by all per-raid-array threads to 7402 * deal with generic issues like resync and super-block update. 7403 * Raid personalities that don't have a thread (linear/raid0) do not 7404 * need this as they never do any recovery or update the superblock. 7405 * 7406 * It does not do any resync itself, but rather "forks" off other threads 7407 * to do that as needed. 7408 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 7409 * "->recovery" and create a thread at ->sync_thread. 7410 * When the thread finishes it sets MD_RECOVERY_DONE 7411 * and wakeups up this thread which will reap the thread and finish up. 7412 * This thread also removes any faulty devices (with nr_pending == 0). 7413 * 7414 * The overall approach is: 7415 * 1/ if the superblock needs updating, update it. 7416 * 2/ If a recovery thread is running, don't do anything else. 7417 * 3/ If recovery has finished, clean up, possibly marking spares active. 7418 * 4/ If there are any faulty devices, remove them. 7419 * 5/ If array is degraded, try to add spares devices 7420 * 6/ If array has spares or is not in-sync, start a resync thread. 7421 */ 7422 void md_check_recovery(struct mddev *mddev) 7423 { 7424 if (mddev->suspended) 7425 return; 7426 7427 if (mddev->bitmap) 7428 bitmap_daemon_work(mddev); 7429 7430 if (signal_pending(current)) { 7431 if (mddev->pers->sync_request && !mddev->external) { 7432 printk(KERN_INFO "md: %s in immediate safe mode\n", 7433 mdname(mddev)); 7434 mddev->safemode = 2; 7435 } 7436 flush_signals(current); 7437 } 7438 7439 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 7440 return; 7441 if ( ! ( 7442 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || 7443 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 7444 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 7445 (mddev->external == 0 && mddev->safemode == 1) || 7446 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 7447 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 7448 )) 7449 return; 7450 7451 if (mddev_trylock(mddev)) { 7452 int spares = 0; 7453 7454 if (mddev->ro) { 7455 /* Only thing we do on a ro array is remove 7456 * failed devices. 7457 */ 7458 struct md_rdev *rdev; 7459 list_for_each_entry(rdev, &mddev->disks, same_set) 7460 if (rdev->raid_disk >= 0 && 7461 !test_bit(Blocked, &rdev->flags) && 7462 test_bit(Faulty, &rdev->flags) && 7463 atomic_read(&rdev->nr_pending)==0) { 7464 if (mddev->pers->hot_remove_disk( 7465 mddev, rdev->raid_disk)==0) { 7466 sysfs_unlink_rdev(mddev, rdev); 7467 rdev->raid_disk = -1; 7468 } 7469 } 7470 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7471 goto unlock; 7472 } 7473 7474 if (!mddev->external) { 7475 int did_change = 0; 7476 spin_lock_irq(&mddev->write_lock); 7477 if (mddev->safemode && 7478 !atomic_read(&mddev->writes_pending) && 7479 !mddev->in_sync && 7480 mddev->recovery_cp == MaxSector) { 7481 mddev->in_sync = 1; 7482 did_change = 1; 7483 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7484 } 7485 if (mddev->safemode == 1) 7486 mddev->safemode = 0; 7487 spin_unlock_irq(&mddev->write_lock); 7488 if (did_change) 7489 sysfs_notify_dirent_safe(mddev->sysfs_state); 7490 } 7491 7492 if (mddev->flags) 7493 md_update_sb(mddev, 0); 7494 7495 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 7496 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 7497 /* resync/recovery still happening */ 7498 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7499 goto unlock; 7500 } 7501 if (mddev->sync_thread) { 7502 reap_sync_thread(mddev); 7503 goto unlock; 7504 } 7505 /* Set RUNNING before clearing NEEDED to avoid 7506 * any transients in the value of "sync_action". 7507 */ 7508 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7509 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7510 /* Clear some bits that don't mean anything, but 7511 * might be left set 7512 */ 7513 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 7514 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 7515 7516 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 7517 goto unlock; 7518 /* no recovery is running. 7519 * remove any failed drives, then 7520 * add spares if possible. 7521 * Spare are also removed and re-added, to allow 7522 * the personality to fail the re-add. 7523 */ 7524 7525 if (mddev->reshape_position != MaxSector) { 7526 if (mddev->pers->check_reshape == NULL || 7527 mddev->pers->check_reshape(mddev) != 0) 7528 /* Cannot proceed */ 7529 goto unlock; 7530 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7531 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7532 } else if ((spares = remove_and_add_spares(mddev))) { 7533 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7534 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7535 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7536 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7537 } else if (mddev->recovery_cp < MaxSector) { 7538 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7539 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7540 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 7541 /* nothing to be done ... */ 7542 goto unlock; 7543 7544 if (mddev->pers->sync_request) { 7545 if (spares && mddev->bitmap && ! mddev->bitmap->file) { 7546 /* We are adding a device or devices to an array 7547 * which has the bitmap stored on all devices. 7548 * So make sure all bitmap pages get written 7549 */ 7550 bitmap_write_all(mddev->bitmap); 7551 } 7552 mddev->sync_thread = md_register_thread(md_do_sync, 7553 mddev, 7554 "resync"); 7555 if (!mddev->sync_thread) { 7556 printk(KERN_ERR "%s: could not start resync" 7557 " thread...\n", 7558 mdname(mddev)); 7559 /* leave the spares where they are, it shouldn't hurt */ 7560 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7561 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7562 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7563 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7564 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7565 } else 7566 md_wakeup_thread(mddev->sync_thread); 7567 sysfs_notify_dirent_safe(mddev->sysfs_action); 7568 md_new_event(mddev); 7569 } 7570 unlock: 7571 if (!mddev->sync_thread) { 7572 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7573 if (test_and_clear_bit(MD_RECOVERY_RECOVER, 7574 &mddev->recovery)) 7575 if (mddev->sysfs_action) 7576 sysfs_notify_dirent_safe(mddev->sysfs_action); 7577 } 7578 mddev_unlock(mddev); 7579 } 7580 } 7581 7582 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) 7583 { 7584 sysfs_notify_dirent_safe(rdev->sysfs_state); 7585 wait_event_timeout(rdev->blocked_wait, 7586 !test_bit(Blocked, &rdev->flags) && 7587 !test_bit(BlockedBadBlocks, &rdev->flags), 7588 msecs_to_jiffies(5000)); 7589 rdev_dec_pending(rdev, mddev); 7590 } 7591 EXPORT_SYMBOL(md_wait_for_blocked_rdev); 7592 7593 7594 /* Bad block management. 7595 * We can record which blocks on each device are 'bad' and so just 7596 * fail those blocks, or that stripe, rather than the whole device. 7597 * Entries in the bad-block table are 64bits wide. This comprises: 7598 * Length of bad-range, in sectors: 0-511 for lengths 1-512 7599 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) 7600 * A 'shift' can be set so that larger blocks are tracked and 7601 * consequently larger devices can be covered. 7602 * 'Acknowledged' flag - 1 bit. - the most significant bit. 7603 * 7604 * Locking of the bad-block table uses a seqlock so md_is_badblock 7605 * might need to retry if it is very unlucky. 7606 * We will sometimes want to check for bad blocks in a bi_end_io function, 7607 * so we use the write_seqlock_irq variant. 7608 * 7609 * When looking for a bad block we specify a range and want to 7610 * know if any block in the range is bad. So we binary-search 7611 * to the last range that starts at-or-before the given endpoint, 7612 * (or "before the sector after the target range") 7613 * then see if it ends after the given start. 7614 * We return 7615 * 0 if there are no known bad blocks in the range 7616 * 1 if there are known bad block which are all acknowledged 7617 * -1 if there are bad blocks which have not yet been acknowledged in metadata. 7618 * plus the start/length of the first bad section we overlap. 7619 */ 7620 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, 7621 sector_t *first_bad, int *bad_sectors) 7622 { 7623 int hi; 7624 int lo = 0; 7625 u64 *p = bb->page; 7626 int rv = 0; 7627 sector_t target = s + sectors; 7628 unsigned seq; 7629 7630 if (bb->shift > 0) { 7631 /* round the start down, and the end up */ 7632 s >>= bb->shift; 7633 target += (1<<bb->shift) - 1; 7634 target >>= bb->shift; 7635 sectors = target - s; 7636 } 7637 /* 'target' is now the first block after the bad range */ 7638 7639 retry: 7640 seq = read_seqbegin(&bb->lock); 7641 7642 hi = bb->count; 7643 7644 /* Binary search between lo and hi for 'target' 7645 * i.e. for the last range that starts before 'target' 7646 */ 7647 /* INVARIANT: ranges before 'lo' and at-or-after 'hi' 7648 * are known not to be the last range before target. 7649 * VARIANT: hi-lo is the number of possible 7650 * ranges, and decreases until it reaches 1 7651 */ 7652 while (hi - lo > 1) { 7653 int mid = (lo + hi) / 2; 7654 sector_t a = BB_OFFSET(p[mid]); 7655 if (a < target) 7656 /* This could still be the one, earlier ranges 7657 * could not. */ 7658 lo = mid; 7659 else 7660 /* This and later ranges are definitely out. */ 7661 hi = mid; 7662 } 7663 /* 'lo' might be the last that started before target, but 'hi' isn't */ 7664 if (hi > lo) { 7665 /* need to check all range that end after 's' to see if 7666 * any are unacknowledged. 7667 */ 7668 while (lo >= 0 && 7669 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { 7670 if (BB_OFFSET(p[lo]) < target) { 7671 /* starts before the end, and finishes after 7672 * the start, so they must overlap 7673 */ 7674 if (rv != -1 && BB_ACK(p[lo])) 7675 rv = 1; 7676 else 7677 rv = -1; 7678 *first_bad = BB_OFFSET(p[lo]); 7679 *bad_sectors = BB_LEN(p[lo]); 7680 } 7681 lo--; 7682 } 7683 } 7684 7685 if (read_seqretry(&bb->lock, seq)) 7686 goto retry; 7687 7688 return rv; 7689 } 7690 EXPORT_SYMBOL_GPL(md_is_badblock); 7691 7692 /* 7693 * Add a range of bad blocks to the table. 7694 * This might extend the table, or might contract it 7695 * if two adjacent ranges can be merged. 7696 * We binary-search to find the 'insertion' point, then 7697 * decide how best to handle it. 7698 */ 7699 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, 7700 int acknowledged) 7701 { 7702 u64 *p; 7703 int lo, hi; 7704 int rv = 1; 7705 7706 if (bb->shift < 0) 7707 /* badblocks are disabled */ 7708 return 0; 7709 7710 if (bb->shift) { 7711 /* round the start down, and the end up */ 7712 sector_t next = s + sectors; 7713 s >>= bb->shift; 7714 next += (1<<bb->shift) - 1; 7715 next >>= bb->shift; 7716 sectors = next - s; 7717 } 7718 7719 write_seqlock_irq(&bb->lock); 7720 7721 p = bb->page; 7722 lo = 0; 7723 hi = bb->count; 7724 /* Find the last range that starts at-or-before 's' */ 7725 while (hi - lo > 1) { 7726 int mid = (lo + hi) / 2; 7727 sector_t a = BB_OFFSET(p[mid]); 7728 if (a <= s) 7729 lo = mid; 7730 else 7731 hi = mid; 7732 } 7733 if (hi > lo && BB_OFFSET(p[lo]) > s) 7734 hi = lo; 7735 7736 if (hi > lo) { 7737 /* we found a range that might merge with the start 7738 * of our new range 7739 */ 7740 sector_t a = BB_OFFSET(p[lo]); 7741 sector_t e = a + BB_LEN(p[lo]); 7742 int ack = BB_ACK(p[lo]); 7743 if (e >= s) { 7744 /* Yes, we can merge with a previous range */ 7745 if (s == a && s + sectors >= e) 7746 /* new range covers old */ 7747 ack = acknowledged; 7748 else 7749 ack = ack && acknowledged; 7750 7751 if (e < s + sectors) 7752 e = s + sectors; 7753 if (e - a <= BB_MAX_LEN) { 7754 p[lo] = BB_MAKE(a, e-a, ack); 7755 s = e; 7756 } else { 7757 /* does not all fit in one range, 7758 * make p[lo] maximal 7759 */ 7760 if (BB_LEN(p[lo]) != BB_MAX_LEN) 7761 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); 7762 s = a + BB_MAX_LEN; 7763 } 7764 sectors = e - s; 7765 } 7766 } 7767 if (sectors && hi < bb->count) { 7768 /* 'hi' points to the first range that starts after 's'. 7769 * Maybe we can merge with the start of that range */ 7770 sector_t a = BB_OFFSET(p[hi]); 7771 sector_t e = a + BB_LEN(p[hi]); 7772 int ack = BB_ACK(p[hi]); 7773 if (a <= s + sectors) { 7774 /* merging is possible */ 7775 if (e <= s + sectors) { 7776 /* full overlap */ 7777 e = s + sectors; 7778 ack = acknowledged; 7779 } else 7780 ack = ack && acknowledged; 7781 7782 a = s; 7783 if (e - a <= BB_MAX_LEN) { 7784 p[hi] = BB_MAKE(a, e-a, ack); 7785 s = e; 7786 } else { 7787 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); 7788 s = a + BB_MAX_LEN; 7789 } 7790 sectors = e - s; 7791 lo = hi; 7792 hi++; 7793 } 7794 } 7795 if (sectors == 0 && hi < bb->count) { 7796 /* we might be able to combine lo and hi */ 7797 /* Note: 's' is at the end of 'lo' */ 7798 sector_t a = BB_OFFSET(p[hi]); 7799 int lolen = BB_LEN(p[lo]); 7800 int hilen = BB_LEN(p[hi]); 7801 int newlen = lolen + hilen - (s - a); 7802 if (s >= a && newlen < BB_MAX_LEN) { 7803 /* yes, we can combine them */ 7804 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); 7805 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); 7806 memmove(p + hi, p + hi + 1, 7807 (bb->count - hi - 1) * 8); 7808 bb->count--; 7809 } 7810 } 7811 while (sectors) { 7812 /* didn't merge (it all). 7813 * Need to add a range just before 'hi' */ 7814 if (bb->count >= MD_MAX_BADBLOCKS) { 7815 /* No room for more */ 7816 rv = 0; 7817 break; 7818 } else { 7819 int this_sectors = sectors; 7820 memmove(p + hi + 1, p + hi, 7821 (bb->count - hi) * 8); 7822 bb->count++; 7823 7824 if (this_sectors > BB_MAX_LEN) 7825 this_sectors = BB_MAX_LEN; 7826 p[hi] = BB_MAKE(s, this_sectors, acknowledged); 7827 sectors -= this_sectors; 7828 s += this_sectors; 7829 } 7830 } 7831 7832 bb->changed = 1; 7833 if (!acknowledged) 7834 bb->unacked_exist = 1; 7835 write_sequnlock_irq(&bb->lock); 7836 7837 return rv; 7838 } 7839 7840 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, 7841 int acknowledged) 7842 { 7843 int rv = md_set_badblocks(&rdev->badblocks, 7844 s + rdev->data_offset, sectors, acknowledged); 7845 if (rv) { 7846 /* Make sure they get written out promptly */ 7847 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); 7848 md_wakeup_thread(rdev->mddev->thread); 7849 } 7850 return rv; 7851 } 7852 EXPORT_SYMBOL_GPL(rdev_set_badblocks); 7853 7854 /* 7855 * Remove a range of bad blocks from the table. 7856 * This may involve extending the table if we spilt a region, 7857 * but it must not fail. So if the table becomes full, we just 7858 * drop the remove request. 7859 */ 7860 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) 7861 { 7862 u64 *p; 7863 int lo, hi; 7864 sector_t target = s + sectors; 7865 int rv = 0; 7866 7867 if (bb->shift > 0) { 7868 /* When clearing we round the start up and the end down. 7869 * This should not matter as the shift should align with 7870 * the block size and no rounding should ever be needed. 7871 * However it is better the think a block is bad when it 7872 * isn't than to think a block is not bad when it is. 7873 */ 7874 s += (1<<bb->shift) - 1; 7875 s >>= bb->shift; 7876 target >>= bb->shift; 7877 sectors = target - s; 7878 } 7879 7880 write_seqlock_irq(&bb->lock); 7881 7882 p = bb->page; 7883 lo = 0; 7884 hi = bb->count; 7885 /* Find the last range that starts before 'target' */ 7886 while (hi - lo > 1) { 7887 int mid = (lo + hi) / 2; 7888 sector_t a = BB_OFFSET(p[mid]); 7889 if (a < target) 7890 lo = mid; 7891 else 7892 hi = mid; 7893 } 7894 if (hi > lo) { 7895 /* p[lo] is the last range that could overlap the 7896 * current range. Earlier ranges could also overlap, 7897 * but only this one can overlap the end of the range. 7898 */ 7899 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { 7900 /* Partial overlap, leave the tail of this range */ 7901 int ack = BB_ACK(p[lo]); 7902 sector_t a = BB_OFFSET(p[lo]); 7903 sector_t end = a + BB_LEN(p[lo]); 7904 7905 if (a < s) { 7906 /* we need to split this range */ 7907 if (bb->count >= MD_MAX_BADBLOCKS) { 7908 rv = 0; 7909 goto out; 7910 } 7911 memmove(p+lo+1, p+lo, (bb->count - lo) * 8); 7912 bb->count++; 7913 p[lo] = BB_MAKE(a, s-a, ack); 7914 lo++; 7915 } 7916 p[lo] = BB_MAKE(target, end - target, ack); 7917 /* there is no longer an overlap */ 7918 hi = lo; 7919 lo--; 7920 } 7921 while (lo >= 0 && 7922 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { 7923 /* This range does overlap */ 7924 if (BB_OFFSET(p[lo]) < s) { 7925 /* Keep the early parts of this range. */ 7926 int ack = BB_ACK(p[lo]); 7927 sector_t start = BB_OFFSET(p[lo]); 7928 p[lo] = BB_MAKE(start, s - start, ack); 7929 /* now low doesn't overlap, so.. */ 7930 break; 7931 } 7932 lo--; 7933 } 7934 /* 'lo' is strictly before, 'hi' is strictly after, 7935 * anything between needs to be discarded 7936 */ 7937 if (hi - lo > 1) { 7938 memmove(p+lo+1, p+hi, (bb->count - hi) * 8); 7939 bb->count -= (hi - lo - 1); 7940 } 7941 } 7942 7943 bb->changed = 1; 7944 out: 7945 write_sequnlock_irq(&bb->lock); 7946 return rv; 7947 } 7948 7949 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors) 7950 { 7951 return md_clear_badblocks(&rdev->badblocks, 7952 s + rdev->data_offset, 7953 sectors); 7954 } 7955 EXPORT_SYMBOL_GPL(rdev_clear_badblocks); 7956 7957 /* 7958 * Acknowledge all bad blocks in a list. 7959 * This only succeeds if ->changed is clear. It is used by 7960 * in-kernel metadata updates 7961 */ 7962 void md_ack_all_badblocks(struct badblocks *bb) 7963 { 7964 if (bb->page == NULL || bb->changed) 7965 /* no point even trying */ 7966 return; 7967 write_seqlock_irq(&bb->lock); 7968 7969 if (bb->changed == 0) { 7970 u64 *p = bb->page; 7971 int i; 7972 for (i = 0; i < bb->count ; i++) { 7973 if (!BB_ACK(p[i])) { 7974 sector_t start = BB_OFFSET(p[i]); 7975 int len = BB_LEN(p[i]); 7976 p[i] = BB_MAKE(start, len, 1); 7977 } 7978 } 7979 bb->unacked_exist = 0; 7980 } 7981 write_sequnlock_irq(&bb->lock); 7982 } 7983 EXPORT_SYMBOL_GPL(md_ack_all_badblocks); 7984 7985 /* sysfs access to bad-blocks list. 7986 * We present two files. 7987 * 'bad-blocks' lists sector numbers and lengths of ranges that 7988 * are recorded as bad. The list is truncated to fit within 7989 * the one-page limit of sysfs. 7990 * Writing "sector length" to this file adds an acknowledged 7991 * bad block list. 7992 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet 7993 * been acknowledged. Writing to this file adds bad blocks 7994 * without acknowledging them. This is largely for testing. 7995 */ 7996 7997 static ssize_t 7998 badblocks_show(struct badblocks *bb, char *page, int unack) 7999 { 8000 size_t len; 8001 int i; 8002 u64 *p = bb->page; 8003 unsigned seq; 8004 8005 if (bb->shift < 0) 8006 return 0; 8007 8008 retry: 8009 seq = read_seqbegin(&bb->lock); 8010 8011 len = 0; 8012 i = 0; 8013 8014 while (len < PAGE_SIZE && i < bb->count) { 8015 sector_t s = BB_OFFSET(p[i]); 8016 unsigned int length = BB_LEN(p[i]); 8017 int ack = BB_ACK(p[i]); 8018 i++; 8019 8020 if (unack && ack) 8021 continue; 8022 8023 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", 8024 (unsigned long long)s << bb->shift, 8025 length << bb->shift); 8026 } 8027 if (unack && len == 0) 8028 bb->unacked_exist = 0; 8029 8030 if (read_seqretry(&bb->lock, seq)) 8031 goto retry; 8032 8033 return len; 8034 } 8035 8036 #define DO_DEBUG 1 8037 8038 static ssize_t 8039 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) 8040 { 8041 unsigned long long sector; 8042 int length; 8043 char newline; 8044 #ifdef DO_DEBUG 8045 /* Allow clearing via sysfs *only* for testing/debugging. 8046 * Normally only a successful write may clear a badblock 8047 */ 8048 int clear = 0; 8049 if (page[0] == '-') { 8050 clear = 1; 8051 page++; 8052 } 8053 #endif /* DO_DEBUG */ 8054 8055 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { 8056 case 3: 8057 if (newline != '\n') 8058 return -EINVAL; 8059 case 2: 8060 if (length <= 0) 8061 return -EINVAL; 8062 break; 8063 default: 8064 return -EINVAL; 8065 } 8066 8067 #ifdef DO_DEBUG 8068 if (clear) { 8069 md_clear_badblocks(bb, sector, length); 8070 return len; 8071 } 8072 #endif /* DO_DEBUG */ 8073 if (md_set_badblocks(bb, sector, length, !unack)) 8074 return len; 8075 else 8076 return -ENOSPC; 8077 } 8078 8079 static int md_notify_reboot(struct notifier_block *this, 8080 unsigned long code, void *x) 8081 { 8082 struct list_head *tmp; 8083 struct mddev *mddev; 8084 int need_delay = 0; 8085 8086 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { 8087 8088 printk(KERN_INFO "md: stopping all md devices.\n"); 8089 8090 for_each_mddev(mddev, tmp) { 8091 if (mddev_trylock(mddev)) { 8092 /* Force a switch to readonly even array 8093 * appears to still be in use. Hence 8094 * the '100'. 8095 */ 8096 md_set_readonly(mddev, 100); 8097 mddev_unlock(mddev); 8098 } 8099 need_delay = 1; 8100 } 8101 /* 8102 * certain more exotic SCSI devices are known to be 8103 * volatile wrt too early system reboots. While the 8104 * right place to handle this issue is the given 8105 * driver, we do want to have a safe RAID driver ... 8106 */ 8107 if (need_delay) 8108 mdelay(1000*1); 8109 } 8110 return NOTIFY_DONE; 8111 } 8112 8113 static struct notifier_block md_notifier = { 8114 .notifier_call = md_notify_reboot, 8115 .next = NULL, 8116 .priority = INT_MAX, /* before any real devices */ 8117 }; 8118 8119 static void md_geninit(void) 8120 { 8121 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 8122 8123 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); 8124 } 8125 8126 static int __init md_init(void) 8127 { 8128 int ret = -ENOMEM; 8129 8130 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); 8131 if (!md_wq) 8132 goto err_wq; 8133 8134 md_misc_wq = alloc_workqueue("md_misc", 0, 0); 8135 if (!md_misc_wq) 8136 goto err_misc_wq; 8137 8138 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) 8139 goto err_md; 8140 8141 if ((ret = register_blkdev(0, "mdp")) < 0) 8142 goto err_mdp; 8143 mdp_major = ret; 8144 8145 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, 8146 md_probe, NULL, NULL); 8147 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, 8148 md_probe, NULL, NULL); 8149 8150 register_reboot_notifier(&md_notifier); 8151 raid_table_header = register_sysctl_table(raid_root_table); 8152 8153 md_geninit(); 8154 return 0; 8155 8156 err_mdp: 8157 unregister_blkdev(MD_MAJOR, "md"); 8158 err_md: 8159 destroy_workqueue(md_misc_wq); 8160 err_misc_wq: 8161 destroy_workqueue(md_wq); 8162 err_wq: 8163 return ret; 8164 } 8165 8166 #ifndef MODULE 8167 8168 /* 8169 * Searches all registered partitions for autorun RAID arrays 8170 * at boot time. 8171 */ 8172 8173 static LIST_HEAD(all_detected_devices); 8174 struct detected_devices_node { 8175 struct list_head list; 8176 dev_t dev; 8177 }; 8178 8179 void md_autodetect_dev(dev_t dev) 8180 { 8181 struct detected_devices_node *node_detected_dev; 8182 8183 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); 8184 if (node_detected_dev) { 8185 node_detected_dev->dev = dev; 8186 list_add_tail(&node_detected_dev->list, &all_detected_devices); 8187 } else { 8188 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" 8189 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); 8190 } 8191 } 8192 8193 8194 static void autostart_arrays(int part) 8195 { 8196 struct md_rdev *rdev; 8197 struct detected_devices_node *node_detected_dev; 8198 dev_t dev; 8199 int i_scanned, i_passed; 8200 8201 i_scanned = 0; 8202 i_passed = 0; 8203 8204 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 8205 8206 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { 8207 i_scanned++; 8208 node_detected_dev = list_entry(all_detected_devices.next, 8209 struct detected_devices_node, list); 8210 list_del(&node_detected_dev->list); 8211 dev = node_detected_dev->dev; 8212 kfree(node_detected_dev); 8213 rdev = md_import_device(dev,0, 90); 8214 if (IS_ERR(rdev)) 8215 continue; 8216 8217 if (test_bit(Faulty, &rdev->flags)) { 8218 MD_BUG(); 8219 continue; 8220 } 8221 set_bit(AutoDetected, &rdev->flags); 8222 list_add(&rdev->same_set, &pending_raid_disks); 8223 i_passed++; 8224 } 8225 8226 printk(KERN_INFO "md: Scanned %d and added %d devices.\n", 8227 i_scanned, i_passed); 8228 8229 autorun_devices(part); 8230 } 8231 8232 #endif /* !MODULE */ 8233 8234 static __exit void md_exit(void) 8235 { 8236 struct mddev *mddev; 8237 struct list_head *tmp; 8238 8239 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); 8240 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); 8241 8242 unregister_blkdev(MD_MAJOR,"md"); 8243 unregister_blkdev(mdp_major, "mdp"); 8244 unregister_reboot_notifier(&md_notifier); 8245 unregister_sysctl_table(raid_table_header); 8246 remove_proc_entry("mdstat", NULL); 8247 for_each_mddev(mddev, tmp) { 8248 export_array(mddev); 8249 mddev->hold_active = 0; 8250 } 8251 destroy_workqueue(md_misc_wq); 8252 destroy_workqueue(md_wq); 8253 } 8254 8255 subsys_initcall(md_init); 8256 module_exit(md_exit) 8257 8258 static int get_ro(char *buffer, struct kernel_param *kp) 8259 { 8260 return sprintf(buffer, "%d", start_readonly); 8261 } 8262 static int set_ro(const char *val, struct kernel_param *kp) 8263 { 8264 char *e; 8265 int num = simple_strtoul(val, &e, 10); 8266 if (*val && (*e == '\0' || *e == '\n')) { 8267 start_readonly = num; 8268 return 0; 8269 } 8270 return -EINVAL; 8271 } 8272 8273 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); 8274 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); 8275 8276 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); 8277 8278 EXPORT_SYMBOL(register_md_personality); 8279 EXPORT_SYMBOL(unregister_md_personality); 8280 EXPORT_SYMBOL(md_error); 8281 EXPORT_SYMBOL(md_done_sync); 8282 EXPORT_SYMBOL(md_write_start); 8283 EXPORT_SYMBOL(md_write_end); 8284 EXPORT_SYMBOL(md_register_thread); 8285 EXPORT_SYMBOL(md_unregister_thread); 8286 EXPORT_SYMBOL(md_wakeup_thread); 8287 EXPORT_SYMBOL(md_check_recovery); 8288 MODULE_LICENSE("GPL"); 8289 MODULE_DESCRIPTION("MD RAID framework"); 8290 MODULE_ALIAS("md"); 8291 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 8292