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/reboot.h> 48 #include <linux/file.h> 49 #include <linux/compat.h> 50 #include <linux/delay.h> 51 #include <linux/raid/md_p.h> 52 #include <linux/raid/md_u.h> 53 #include <linux/slab.h> 54 #include "md.h" 55 #include "bitmap.h" 56 57 #define DEBUG 0 58 #define dprintk(x...) ((void)(DEBUG && printk(x))) 59 60 #ifndef MODULE 61 static void autostart_arrays(int part); 62 #endif 63 64 static LIST_HEAD(pers_list); 65 static DEFINE_SPINLOCK(pers_lock); 66 67 static void md_print_devices(void); 68 69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait); 70 static struct workqueue_struct *md_wq; 71 static struct workqueue_struct *md_misc_wq; 72 73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } 74 75 /* 76 * Default number of read corrections we'll attempt on an rdev 77 * before ejecting it from the array. We divide the read error 78 * count by 2 for every hour elapsed between read errors. 79 */ 80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 81 /* 82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' 83 * is 1000 KB/sec, so the extra system load does not show up that much. 84 * Increase it if you want to have more _guaranteed_ speed. Note that 85 * the RAID driver will use the maximum available bandwidth if the IO 86 * subsystem is idle. There is also an 'absolute maximum' reconstruction 87 * speed limit - in case reconstruction slows down your system despite 88 * idle IO detection. 89 * 90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. 91 * or /sys/block/mdX/md/sync_speed_{min,max} 92 */ 93 94 static int sysctl_speed_limit_min = 1000; 95 static int sysctl_speed_limit_max = 200000; 96 static inline int speed_min(mddev_t *mddev) 97 { 98 return mddev->sync_speed_min ? 99 mddev->sync_speed_min : sysctl_speed_limit_min; 100 } 101 102 static inline int speed_max(mddev_t *mddev) 103 { 104 return mddev->sync_speed_max ? 105 mddev->sync_speed_max : sysctl_speed_limit_max; 106 } 107 108 static struct ctl_table_header *raid_table_header; 109 110 static ctl_table raid_table[] = { 111 { 112 .procname = "speed_limit_min", 113 .data = &sysctl_speed_limit_min, 114 .maxlen = sizeof(int), 115 .mode = S_IRUGO|S_IWUSR, 116 .proc_handler = proc_dointvec, 117 }, 118 { 119 .procname = "speed_limit_max", 120 .data = &sysctl_speed_limit_max, 121 .maxlen = sizeof(int), 122 .mode = S_IRUGO|S_IWUSR, 123 .proc_handler = proc_dointvec, 124 }, 125 { } 126 }; 127 128 static ctl_table raid_dir_table[] = { 129 { 130 .procname = "raid", 131 .maxlen = 0, 132 .mode = S_IRUGO|S_IXUGO, 133 .child = raid_table, 134 }, 135 { } 136 }; 137 138 static ctl_table raid_root_table[] = { 139 { 140 .procname = "dev", 141 .maxlen = 0, 142 .mode = 0555, 143 .child = raid_dir_table, 144 }, 145 { } 146 }; 147 148 static const struct block_device_operations md_fops; 149 150 static int start_readonly; 151 152 /* bio_clone_mddev 153 * like bio_clone, but with a local bio set 154 */ 155 156 static void mddev_bio_destructor(struct bio *bio) 157 { 158 mddev_t *mddev, **mddevp; 159 160 mddevp = (void*)bio; 161 mddev = mddevp[-1]; 162 163 bio_free(bio, mddev->bio_set); 164 } 165 166 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, 167 mddev_t *mddev) 168 { 169 struct bio *b; 170 mddev_t **mddevp; 171 172 if (!mddev || !mddev->bio_set) 173 return bio_alloc(gfp_mask, nr_iovecs); 174 175 b = bio_alloc_bioset(gfp_mask, nr_iovecs, 176 mddev->bio_set); 177 if (!b) 178 return NULL; 179 mddevp = (void*)b; 180 mddevp[-1] = mddev; 181 b->bi_destructor = mddev_bio_destructor; 182 return b; 183 } 184 EXPORT_SYMBOL_GPL(bio_alloc_mddev); 185 186 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, 187 mddev_t *mddev) 188 { 189 struct bio *b; 190 mddev_t **mddevp; 191 192 if (!mddev || !mddev->bio_set) 193 return bio_clone(bio, gfp_mask); 194 195 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, 196 mddev->bio_set); 197 if (!b) 198 return NULL; 199 mddevp = (void*)b; 200 mddevp[-1] = mddev; 201 b->bi_destructor = mddev_bio_destructor; 202 __bio_clone(b, bio); 203 if (bio_integrity(bio)) { 204 int ret; 205 206 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set); 207 208 if (ret < 0) { 209 bio_put(b); 210 return NULL; 211 } 212 } 213 214 return b; 215 } 216 EXPORT_SYMBOL_GPL(bio_clone_mddev); 217 218 void md_trim_bio(struct bio *bio, int offset, int size) 219 { 220 /* 'bio' is a cloned bio which we need to trim to match 221 * the given offset and size. 222 * This requires adjusting bi_sector, bi_size, and bi_io_vec 223 */ 224 int i; 225 struct bio_vec *bvec; 226 int sofar = 0; 227 228 size <<= 9; 229 if (offset == 0 && size == bio->bi_size) 230 return; 231 232 bio->bi_sector += offset; 233 bio->bi_size = size; 234 offset <<= 9; 235 clear_bit(BIO_SEG_VALID, &bio->bi_flags); 236 237 while (bio->bi_idx < bio->bi_vcnt && 238 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) { 239 /* remove this whole bio_vec */ 240 offset -= bio->bi_io_vec[bio->bi_idx].bv_len; 241 bio->bi_idx++; 242 } 243 if (bio->bi_idx < bio->bi_vcnt) { 244 bio->bi_io_vec[bio->bi_idx].bv_offset += offset; 245 bio->bi_io_vec[bio->bi_idx].bv_len -= offset; 246 } 247 /* avoid any complications with bi_idx being non-zero*/ 248 if (bio->bi_idx) { 249 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx, 250 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec)); 251 bio->bi_vcnt -= bio->bi_idx; 252 bio->bi_idx = 0; 253 } 254 /* Make sure vcnt and last bv are not too big */ 255 bio_for_each_segment(bvec, bio, i) { 256 if (sofar + bvec->bv_len > size) 257 bvec->bv_len = size - sofar; 258 if (bvec->bv_len == 0) { 259 bio->bi_vcnt = i; 260 break; 261 } 262 sofar += bvec->bv_len; 263 } 264 } 265 EXPORT_SYMBOL_GPL(md_trim_bio); 266 267 /* 268 * We have a system wide 'event count' that is incremented 269 * on any 'interesting' event, and readers of /proc/mdstat 270 * can use 'poll' or 'select' to find out when the event 271 * count increases. 272 * 273 * Events are: 274 * start array, stop array, error, add device, remove device, 275 * start build, activate spare 276 */ 277 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); 278 static atomic_t md_event_count; 279 void md_new_event(mddev_t *mddev) 280 { 281 atomic_inc(&md_event_count); 282 wake_up(&md_event_waiters); 283 } 284 EXPORT_SYMBOL_GPL(md_new_event); 285 286 /* Alternate version that can be called from interrupts 287 * when calling sysfs_notify isn't needed. 288 */ 289 static void md_new_event_inintr(mddev_t *mddev) 290 { 291 atomic_inc(&md_event_count); 292 wake_up(&md_event_waiters); 293 } 294 295 /* 296 * Enables to iterate over all existing md arrays 297 * all_mddevs_lock protects this list. 298 */ 299 static LIST_HEAD(all_mddevs); 300 static DEFINE_SPINLOCK(all_mddevs_lock); 301 302 303 /* 304 * iterates through all used mddevs in the system. 305 * We take care to grab the all_mddevs_lock whenever navigating 306 * the list, and to always hold a refcount when unlocked. 307 * Any code which breaks out of this loop while own 308 * a reference to the current mddev and must mddev_put it. 309 */ 310 #define for_each_mddev(mddev,tmp) \ 311 \ 312 for (({ spin_lock(&all_mddevs_lock); \ 313 tmp = all_mddevs.next; \ 314 mddev = NULL;}); \ 315 ({ if (tmp != &all_mddevs) \ 316 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ 317 spin_unlock(&all_mddevs_lock); \ 318 if (mddev) mddev_put(mddev); \ 319 mddev = list_entry(tmp, mddev_t, all_mddevs); \ 320 tmp != &all_mddevs;}); \ 321 ({ spin_lock(&all_mddevs_lock); \ 322 tmp = tmp->next;}) \ 323 ) 324 325 326 /* Rather than calling directly into the personality make_request function, 327 * IO requests come here first so that we can check if the device is 328 * being suspended pending a reconfiguration. 329 * We hold a refcount over the call to ->make_request. By the time that 330 * call has finished, the bio has been linked into some internal structure 331 * and so is visible to ->quiesce(), so we don't need the refcount any more. 332 */ 333 static int md_make_request(struct request_queue *q, struct bio *bio) 334 { 335 const int rw = bio_data_dir(bio); 336 mddev_t *mddev = q->queuedata; 337 int rv; 338 int cpu; 339 unsigned int sectors; 340 341 if (mddev == NULL || mddev->pers == NULL 342 || !mddev->ready) { 343 bio_io_error(bio); 344 return 0; 345 } 346 smp_rmb(); /* Ensure implications of 'active' are visible */ 347 rcu_read_lock(); 348 if (mddev->suspended) { 349 DEFINE_WAIT(__wait); 350 for (;;) { 351 prepare_to_wait(&mddev->sb_wait, &__wait, 352 TASK_UNINTERRUPTIBLE); 353 if (!mddev->suspended) 354 break; 355 rcu_read_unlock(); 356 schedule(); 357 rcu_read_lock(); 358 } 359 finish_wait(&mddev->sb_wait, &__wait); 360 } 361 atomic_inc(&mddev->active_io); 362 rcu_read_unlock(); 363 364 /* 365 * save the sectors now since our bio can 366 * go away inside make_request 367 */ 368 sectors = bio_sectors(bio); 369 rv = mddev->pers->make_request(mddev, bio); 370 371 cpu = part_stat_lock(); 372 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); 373 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); 374 part_stat_unlock(); 375 376 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) 377 wake_up(&mddev->sb_wait); 378 379 return rv; 380 } 381 382 /* mddev_suspend makes sure no new requests are submitted 383 * to the device, and that any requests that have been submitted 384 * are completely handled. 385 * Once ->stop is called and completes, the module will be completely 386 * unused. 387 */ 388 void mddev_suspend(mddev_t *mddev) 389 { 390 BUG_ON(mddev->suspended); 391 mddev->suspended = 1; 392 synchronize_rcu(); 393 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); 394 mddev->pers->quiesce(mddev, 1); 395 } 396 EXPORT_SYMBOL_GPL(mddev_suspend); 397 398 void mddev_resume(mddev_t *mddev) 399 { 400 mddev->suspended = 0; 401 wake_up(&mddev->sb_wait); 402 mddev->pers->quiesce(mddev, 0); 403 404 md_wakeup_thread(mddev->thread); 405 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 406 } 407 EXPORT_SYMBOL_GPL(mddev_resume); 408 409 int mddev_congested(mddev_t *mddev, int bits) 410 { 411 return mddev->suspended; 412 } 413 EXPORT_SYMBOL(mddev_congested); 414 415 /* 416 * Generic flush handling for md 417 */ 418 419 static void md_end_flush(struct bio *bio, int err) 420 { 421 mdk_rdev_t *rdev = bio->bi_private; 422 mddev_t *mddev = rdev->mddev; 423 424 rdev_dec_pending(rdev, mddev); 425 426 if (atomic_dec_and_test(&mddev->flush_pending)) { 427 /* The pre-request flush has finished */ 428 queue_work(md_wq, &mddev->flush_work); 429 } 430 bio_put(bio); 431 } 432 433 static void md_submit_flush_data(struct work_struct *ws); 434 435 static void submit_flushes(struct work_struct *ws) 436 { 437 mddev_t *mddev = container_of(ws, mddev_t, flush_work); 438 mdk_rdev_t *rdev; 439 440 INIT_WORK(&mddev->flush_work, md_submit_flush_data); 441 atomic_set(&mddev->flush_pending, 1); 442 rcu_read_lock(); 443 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 444 if (rdev->raid_disk >= 0 && 445 !test_bit(Faulty, &rdev->flags)) { 446 /* Take two references, one is dropped 447 * when request finishes, one after 448 * we reclaim rcu_read_lock 449 */ 450 struct bio *bi; 451 atomic_inc(&rdev->nr_pending); 452 atomic_inc(&rdev->nr_pending); 453 rcu_read_unlock(); 454 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev); 455 bi->bi_end_io = md_end_flush; 456 bi->bi_private = rdev; 457 bi->bi_bdev = rdev->bdev; 458 atomic_inc(&mddev->flush_pending); 459 submit_bio(WRITE_FLUSH, bi); 460 rcu_read_lock(); 461 rdev_dec_pending(rdev, mddev); 462 } 463 rcu_read_unlock(); 464 if (atomic_dec_and_test(&mddev->flush_pending)) 465 queue_work(md_wq, &mddev->flush_work); 466 } 467 468 static void md_submit_flush_data(struct work_struct *ws) 469 { 470 mddev_t *mddev = container_of(ws, mddev_t, flush_work); 471 struct bio *bio = mddev->flush_bio; 472 473 if (bio->bi_size == 0) 474 /* an empty barrier - all done */ 475 bio_endio(bio, 0); 476 else { 477 bio->bi_rw &= ~REQ_FLUSH; 478 if (mddev->pers->make_request(mddev, bio)) 479 generic_make_request(bio); 480 } 481 482 mddev->flush_bio = NULL; 483 wake_up(&mddev->sb_wait); 484 } 485 486 void md_flush_request(mddev_t *mddev, struct bio *bio) 487 { 488 spin_lock_irq(&mddev->write_lock); 489 wait_event_lock_irq(mddev->sb_wait, 490 !mddev->flush_bio, 491 mddev->write_lock, /*nothing*/); 492 mddev->flush_bio = bio; 493 spin_unlock_irq(&mddev->write_lock); 494 495 INIT_WORK(&mddev->flush_work, submit_flushes); 496 queue_work(md_wq, &mddev->flush_work); 497 } 498 EXPORT_SYMBOL(md_flush_request); 499 500 /* Support for plugging. 501 * This mirrors the plugging support in request_queue, but does not 502 * require having a whole queue or request structures. 503 * We allocate an md_plug_cb for each md device and each thread it gets 504 * plugged on. This links tot the private plug_handle structure in the 505 * personality data where we keep a count of the number of outstanding 506 * plugs so other code can see if a plug is active. 507 */ 508 struct md_plug_cb { 509 struct blk_plug_cb cb; 510 mddev_t *mddev; 511 }; 512 513 static void plugger_unplug(struct blk_plug_cb *cb) 514 { 515 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb); 516 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt)) 517 md_wakeup_thread(mdcb->mddev->thread); 518 kfree(mdcb); 519 } 520 521 /* Check that an unplug wakeup will come shortly. 522 * If not, wakeup the md thread immediately 523 */ 524 int mddev_check_plugged(mddev_t *mddev) 525 { 526 struct blk_plug *plug = current->plug; 527 struct md_plug_cb *mdcb; 528 529 if (!plug) 530 return 0; 531 532 list_for_each_entry(mdcb, &plug->cb_list, cb.list) { 533 if (mdcb->cb.callback == plugger_unplug && 534 mdcb->mddev == mddev) { 535 /* Already on the list, move to top */ 536 if (mdcb != list_first_entry(&plug->cb_list, 537 struct md_plug_cb, 538 cb.list)) 539 list_move(&mdcb->cb.list, &plug->cb_list); 540 return 1; 541 } 542 } 543 /* Not currently on the callback list */ 544 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC); 545 if (!mdcb) 546 return 0; 547 548 mdcb->mddev = mddev; 549 mdcb->cb.callback = plugger_unplug; 550 atomic_inc(&mddev->plug_cnt); 551 list_add(&mdcb->cb.list, &plug->cb_list); 552 return 1; 553 } 554 EXPORT_SYMBOL_GPL(mddev_check_plugged); 555 556 static inline mddev_t *mddev_get(mddev_t *mddev) 557 { 558 atomic_inc(&mddev->active); 559 return mddev; 560 } 561 562 static void mddev_delayed_delete(struct work_struct *ws); 563 564 static void mddev_put(mddev_t *mddev) 565 { 566 struct bio_set *bs = NULL; 567 568 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 569 return; 570 if (!mddev->raid_disks && list_empty(&mddev->disks) && 571 mddev->ctime == 0 && !mddev->hold_active) { 572 /* Array is not configured at all, and not held active, 573 * so destroy it */ 574 list_del(&mddev->all_mddevs); 575 bs = mddev->bio_set; 576 mddev->bio_set = NULL; 577 if (mddev->gendisk) { 578 /* We did a probe so need to clean up. Call 579 * queue_work inside the spinlock so that 580 * flush_workqueue() after mddev_find will 581 * succeed in waiting for the work to be done. 582 */ 583 INIT_WORK(&mddev->del_work, mddev_delayed_delete); 584 queue_work(md_misc_wq, &mddev->del_work); 585 } else 586 kfree(mddev); 587 } 588 spin_unlock(&all_mddevs_lock); 589 if (bs) 590 bioset_free(bs); 591 } 592 593 void mddev_init(mddev_t *mddev) 594 { 595 mutex_init(&mddev->open_mutex); 596 mutex_init(&mddev->reconfig_mutex); 597 mutex_init(&mddev->bitmap_info.mutex); 598 INIT_LIST_HEAD(&mddev->disks); 599 INIT_LIST_HEAD(&mddev->all_mddevs); 600 init_timer(&mddev->safemode_timer); 601 atomic_set(&mddev->active, 1); 602 atomic_set(&mddev->openers, 0); 603 atomic_set(&mddev->active_io, 0); 604 atomic_set(&mddev->plug_cnt, 0); 605 spin_lock_init(&mddev->write_lock); 606 atomic_set(&mddev->flush_pending, 0); 607 init_waitqueue_head(&mddev->sb_wait); 608 init_waitqueue_head(&mddev->recovery_wait); 609 mddev->reshape_position = MaxSector; 610 mddev->resync_min = 0; 611 mddev->resync_max = MaxSector; 612 mddev->level = LEVEL_NONE; 613 } 614 EXPORT_SYMBOL_GPL(mddev_init); 615 616 static mddev_t * mddev_find(dev_t unit) 617 { 618 mddev_t *mddev, *new = NULL; 619 620 if (unit && MAJOR(unit) != MD_MAJOR) 621 unit &= ~((1<<MdpMinorShift)-1); 622 623 retry: 624 spin_lock(&all_mddevs_lock); 625 626 if (unit) { 627 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 628 if (mddev->unit == unit) { 629 mddev_get(mddev); 630 spin_unlock(&all_mddevs_lock); 631 kfree(new); 632 return mddev; 633 } 634 635 if (new) { 636 list_add(&new->all_mddevs, &all_mddevs); 637 spin_unlock(&all_mddevs_lock); 638 new->hold_active = UNTIL_IOCTL; 639 return new; 640 } 641 } else if (new) { 642 /* find an unused unit number */ 643 static int next_minor = 512; 644 int start = next_minor; 645 int is_free = 0; 646 int dev = 0; 647 while (!is_free) { 648 dev = MKDEV(MD_MAJOR, next_minor); 649 next_minor++; 650 if (next_minor > MINORMASK) 651 next_minor = 0; 652 if (next_minor == start) { 653 /* Oh dear, all in use. */ 654 spin_unlock(&all_mddevs_lock); 655 kfree(new); 656 return NULL; 657 } 658 659 is_free = 1; 660 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 661 if (mddev->unit == dev) { 662 is_free = 0; 663 break; 664 } 665 } 666 new->unit = dev; 667 new->md_minor = MINOR(dev); 668 new->hold_active = UNTIL_STOP; 669 list_add(&new->all_mddevs, &all_mddevs); 670 spin_unlock(&all_mddevs_lock); 671 return new; 672 } 673 spin_unlock(&all_mddevs_lock); 674 675 new = kzalloc(sizeof(*new), GFP_KERNEL); 676 if (!new) 677 return NULL; 678 679 new->unit = unit; 680 if (MAJOR(unit) == MD_MAJOR) 681 new->md_minor = MINOR(unit); 682 else 683 new->md_minor = MINOR(unit) >> MdpMinorShift; 684 685 mddev_init(new); 686 687 goto retry; 688 } 689 690 static inline int mddev_lock(mddev_t * mddev) 691 { 692 return mutex_lock_interruptible(&mddev->reconfig_mutex); 693 } 694 695 static inline int mddev_is_locked(mddev_t *mddev) 696 { 697 return mutex_is_locked(&mddev->reconfig_mutex); 698 } 699 700 static inline int mddev_trylock(mddev_t * mddev) 701 { 702 return mutex_trylock(&mddev->reconfig_mutex); 703 } 704 705 static struct attribute_group md_redundancy_group; 706 707 static void mddev_unlock(mddev_t * mddev) 708 { 709 if (mddev->to_remove) { 710 /* These cannot be removed under reconfig_mutex as 711 * an access to the files will try to take reconfig_mutex 712 * while holding the file unremovable, which leads to 713 * a deadlock. 714 * So hold set sysfs_active while the remove in happeing, 715 * and anything else which might set ->to_remove or my 716 * otherwise change the sysfs namespace will fail with 717 * -EBUSY if sysfs_active is still set. 718 * We set sysfs_active under reconfig_mutex and elsewhere 719 * test it under the same mutex to ensure its correct value 720 * is seen. 721 */ 722 struct attribute_group *to_remove = mddev->to_remove; 723 mddev->to_remove = NULL; 724 mddev->sysfs_active = 1; 725 mutex_unlock(&mddev->reconfig_mutex); 726 727 if (mddev->kobj.sd) { 728 if (to_remove != &md_redundancy_group) 729 sysfs_remove_group(&mddev->kobj, to_remove); 730 if (mddev->pers == NULL || 731 mddev->pers->sync_request == NULL) { 732 sysfs_remove_group(&mddev->kobj, &md_redundancy_group); 733 if (mddev->sysfs_action) 734 sysfs_put(mddev->sysfs_action); 735 mddev->sysfs_action = NULL; 736 } 737 } 738 mddev->sysfs_active = 0; 739 } else 740 mutex_unlock(&mddev->reconfig_mutex); 741 742 md_wakeup_thread(mddev->thread); 743 } 744 745 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) 746 { 747 mdk_rdev_t *rdev; 748 749 list_for_each_entry(rdev, &mddev->disks, same_set) 750 if (rdev->desc_nr == nr) 751 return rdev; 752 753 return NULL; 754 } 755 756 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) 757 { 758 mdk_rdev_t *rdev; 759 760 list_for_each_entry(rdev, &mddev->disks, same_set) 761 if (rdev->bdev->bd_dev == dev) 762 return rdev; 763 764 return NULL; 765 } 766 767 static struct mdk_personality *find_pers(int level, char *clevel) 768 { 769 struct mdk_personality *pers; 770 list_for_each_entry(pers, &pers_list, list) { 771 if (level != LEVEL_NONE && pers->level == level) 772 return pers; 773 if (strcmp(pers->name, clevel)==0) 774 return pers; 775 } 776 return NULL; 777 } 778 779 /* return the offset of the super block in 512byte sectors */ 780 static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev) 781 { 782 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; 783 return MD_NEW_SIZE_SECTORS(num_sectors); 784 } 785 786 static int alloc_disk_sb(mdk_rdev_t * rdev) 787 { 788 if (rdev->sb_page) 789 MD_BUG(); 790 791 rdev->sb_page = alloc_page(GFP_KERNEL); 792 if (!rdev->sb_page) { 793 printk(KERN_ALERT "md: out of memory.\n"); 794 return -ENOMEM; 795 } 796 797 return 0; 798 } 799 800 static void free_disk_sb(mdk_rdev_t * rdev) 801 { 802 if (rdev->sb_page) { 803 put_page(rdev->sb_page); 804 rdev->sb_loaded = 0; 805 rdev->sb_page = NULL; 806 rdev->sb_start = 0; 807 rdev->sectors = 0; 808 } 809 if (rdev->bb_page) { 810 put_page(rdev->bb_page); 811 rdev->bb_page = NULL; 812 } 813 } 814 815 816 static void super_written(struct bio *bio, int error) 817 { 818 mdk_rdev_t *rdev = bio->bi_private; 819 mddev_t *mddev = rdev->mddev; 820 821 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { 822 printk("md: super_written gets error=%d, uptodate=%d\n", 823 error, test_bit(BIO_UPTODATE, &bio->bi_flags)); 824 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); 825 md_error(mddev, rdev); 826 } 827 828 if (atomic_dec_and_test(&mddev->pending_writes)) 829 wake_up(&mddev->sb_wait); 830 bio_put(bio); 831 } 832 833 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, 834 sector_t sector, int size, struct page *page) 835 { 836 /* write first size bytes of page to sector of rdev 837 * Increment mddev->pending_writes before returning 838 * and decrement it on completion, waking up sb_wait 839 * if zero is reached. 840 * If an error occurred, call md_error 841 */ 842 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); 843 844 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; 845 bio->bi_sector = sector; 846 bio_add_page(bio, page, size, 0); 847 bio->bi_private = rdev; 848 bio->bi_end_io = super_written; 849 850 atomic_inc(&mddev->pending_writes); 851 submit_bio(WRITE_FLUSH_FUA, bio); 852 } 853 854 void md_super_wait(mddev_t *mddev) 855 { 856 /* wait for all superblock writes that were scheduled to complete */ 857 DEFINE_WAIT(wq); 858 for(;;) { 859 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 860 if (atomic_read(&mddev->pending_writes)==0) 861 break; 862 schedule(); 863 } 864 finish_wait(&mddev->sb_wait, &wq); 865 } 866 867 static void bi_complete(struct bio *bio, int error) 868 { 869 complete((struct completion*)bio->bi_private); 870 } 871 872 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size, 873 struct page *page, int rw, bool metadata_op) 874 { 875 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); 876 struct completion event; 877 int ret; 878 879 rw |= REQ_SYNC; 880 881 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? 882 rdev->meta_bdev : rdev->bdev; 883 if (metadata_op) 884 bio->bi_sector = sector + rdev->sb_start; 885 else 886 bio->bi_sector = sector + rdev->data_offset; 887 bio_add_page(bio, page, size, 0); 888 init_completion(&event); 889 bio->bi_private = &event; 890 bio->bi_end_io = bi_complete; 891 submit_bio(rw, bio); 892 wait_for_completion(&event); 893 894 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 895 bio_put(bio); 896 return ret; 897 } 898 EXPORT_SYMBOL_GPL(sync_page_io); 899 900 static int read_disk_sb(mdk_rdev_t * rdev, int size) 901 { 902 char b[BDEVNAME_SIZE]; 903 if (!rdev->sb_page) { 904 MD_BUG(); 905 return -EINVAL; 906 } 907 if (rdev->sb_loaded) 908 return 0; 909 910 911 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) 912 goto fail; 913 rdev->sb_loaded = 1; 914 return 0; 915 916 fail: 917 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 918 bdevname(rdev->bdev,b)); 919 return -EINVAL; 920 } 921 922 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 923 { 924 return sb1->set_uuid0 == sb2->set_uuid0 && 925 sb1->set_uuid1 == sb2->set_uuid1 && 926 sb1->set_uuid2 == sb2->set_uuid2 && 927 sb1->set_uuid3 == sb2->set_uuid3; 928 } 929 930 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 931 { 932 int ret; 933 mdp_super_t *tmp1, *tmp2; 934 935 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 936 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 937 938 if (!tmp1 || !tmp2) { 939 ret = 0; 940 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); 941 goto abort; 942 } 943 944 *tmp1 = *sb1; 945 *tmp2 = *sb2; 946 947 /* 948 * nr_disks is not constant 949 */ 950 tmp1->nr_disks = 0; 951 tmp2->nr_disks = 0; 952 953 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); 954 abort: 955 kfree(tmp1); 956 kfree(tmp2); 957 return ret; 958 } 959 960 961 static u32 md_csum_fold(u32 csum) 962 { 963 csum = (csum & 0xffff) + (csum >> 16); 964 return (csum & 0xffff) + (csum >> 16); 965 } 966 967 static unsigned int calc_sb_csum(mdp_super_t * sb) 968 { 969 u64 newcsum = 0; 970 u32 *sb32 = (u32*)sb; 971 int i; 972 unsigned int disk_csum, csum; 973 974 disk_csum = sb->sb_csum; 975 sb->sb_csum = 0; 976 977 for (i = 0; i < MD_SB_BYTES/4 ; i++) 978 newcsum += sb32[i]; 979 csum = (newcsum & 0xffffffff) + (newcsum>>32); 980 981 982 #ifdef CONFIG_ALPHA 983 /* This used to use csum_partial, which was wrong for several 984 * reasons including that different results are returned on 985 * different architectures. It isn't critical that we get exactly 986 * the same return value as before (we always csum_fold before 987 * testing, and that removes any differences). However as we 988 * know that csum_partial always returned a 16bit value on 989 * alphas, do a fold to maximise conformity to previous behaviour. 990 */ 991 sb->sb_csum = md_csum_fold(disk_csum); 992 #else 993 sb->sb_csum = disk_csum; 994 #endif 995 return csum; 996 } 997 998 999 /* 1000 * Handle superblock details. 1001 * We want to be able to handle multiple superblock formats 1002 * so we have a common interface to them all, and an array of 1003 * different handlers. 1004 * We rely on user-space to write the initial superblock, and support 1005 * reading and updating of superblocks. 1006 * Interface methods are: 1007 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) 1008 * loads and validates a superblock on dev. 1009 * if refdev != NULL, compare superblocks on both devices 1010 * Return: 1011 * 0 - dev has a superblock that is compatible with refdev 1012 * 1 - dev has a superblock that is compatible and newer than refdev 1013 * so dev should be used as the refdev in future 1014 * -EINVAL superblock incompatible or invalid 1015 * -othererror e.g. -EIO 1016 * 1017 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) 1018 * Verify that dev is acceptable into mddev. 1019 * The first time, mddev->raid_disks will be 0, and data from 1020 * dev should be merged in. Subsequent calls check that dev 1021 * is new enough. Return 0 or -EINVAL 1022 * 1023 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) 1024 * Update the superblock for rdev with data in mddev 1025 * This does not write to disc. 1026 * 1027 */ 1028 1029 struct super_type { 1030 char *name; 1031 struct module *owner; 1032 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, 1033 int minor_version); 1034 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); 1035 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); 1036 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, 1037 sector_t num_sectors); 1038 }; 1039 1040 /* 1041 * Check that the given mddev has no bitmap. 1042 * 1043 * This function is called from the run method of all personalities that do not 1044 * support bitmaps. It prints an error message and returns non-zero if mddev 1045 * has a bitmap. Otherwise, it returns 0. 1046 * 1047 */ 1048 int md_check_no_bitmap(mddev_t *mddev) 1049 { 1050 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) 1051 return 0; 1052 printk(KERN_ERR "%s: bitmaps are not supported for %s\n", 1053 mdname(mddev), mddev->pers->name); 1054 return 1; 1055 } 1056 EXPORT_SYMBOL(md_check_no_bitmap); 1057 1058 /* 1059 * load_super for 0.90.0 1060 */ 1061 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 1062 { 1063 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1064 mdp_super_t *sb; 1065 int ret; 1066 1067 /* 1068 * Calculate the position of the superblock (512byte sectors), 1069 * it's at the end of the disk. 1070 * 1071 * It also happens to be a multiple of 4Kb. 1072 */ 1073 rdev->sb_start = calc_dev_sboffset(rdev); 1074 1075 ret = read_disk_sb(rdev, MD_SB_BYTES); 1076 if (ret) return ret; 1077 1078 ret = -EINVAL; 1079 1080 bdevname(rdev->bdev, b); 1081 sb = page_address(rdev->sb_page); 1082 1083 if (sb->md_magic != MD_SB_MAGIC) { 1084 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 1085 b); 1086 goto abort; 1087 } 1088 1089 if (sb->major_version != 0 || 1090 sb->minor_version < 90 || 1091 sb->minor_version > 91) { 1092 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 1093 sb->major_version, sb->minor_version, 1094 b); 1095 goto abort; 1096 } 1097 1098 if (sb->raid_disks <= 0) 1099 goto abort; 1100 1101 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { 1102 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 1103 b); 1104 goto abort; 1105 } 1106 1107 rdev->preferred_minor = sb->md_minor; 1108 rdev->data_offset = 0; 1109 rdev->sb_size = MD_SB_BYTES; 1110 rdev->badblocks.shift = -1; 1111 1112 if (sb->level == LEVEL_MULTIPATH) 1113 rdev->desc_nr = -1; 1114 else 1115 rdev->desc_nr = sb->this_disk.number; 1116 1117 if (!refdev) { 1118 ret = 1; 1119 } else { 1120 __u64 ev1, ev2; 1121 mdp_super_t *refsb = page_address(refdev->sb_page); 1122 if (!uuid_equal(refsb, sb)) { 1123 printk(KERN_WARNING "md: %s has different UUID to %s\n", 1124 b, bdevname(refdev->bdev,b2)); 1125 goto abort; 1126 } 1127 if (!sb_equal(refsb, sb)) { 1128 printk(KERN_WARNING "md: %s has same UUID" 1129 " but different superblock to %s\n", 1130 b, bdevname(refdev->bdev, b2)); 1131 goto abort; 1132 } 1133 ev1 = md_event(sb); 1134 ev2 = md_event(refsb); 1135 if (ev1 > ev2) 1136 ret = 1; 1137 else 1138 ret = 0; 1139 } 1140 rdev->sectors = rdev->sb_start; 1141 /* Limit to 4TB as metadata cannot record more than that */ 1142 if (rdev->sectors >= (2ULL << 32)) 1143 rdev->sectors = (2ULL << 32) - 2; 1144 1145 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) 1146 /* "this cannot possibly happen" ... */ 1147 ret = -EINVAL; 1148 1149 abort: 1150 return ret; 1151 } 1152 1153 /* 1154 * validate_super for 0.90.0 1155 */ 1156 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1157 { 1158 mdp_disk_t *desc; 1159 mdp_super_t *sb = page_address(rdev->sb_page); 1160 __u64 ev1 = md_event(sb); 1161 1162 rdev->raid_disk = -1; 1163 clear_bit(Faulty, &rdev->flags); 1164 clear_bit(In_sync, &rdev->flags); 1165 clear_bit(WriteMostly, &rdev->flags); 1166 1167 if (mddev->raid_disks == 0) { 1168 mddev->major_version = 0; 1169 mddev->minor_version = sb->minor_version; 1170 mddev->patch_version = sb->patch_version; 1171 mddev->external = 0; 1172 mddev->chunk_sectors = sb->chunk_size >> 9; 1173 mddev->ctime = sb->ctime; 1174 mddev->utime = sb->utime; 1175 mddev->level = sb->level; 1176 mddev->clevel[0] = 0; 1177 mddev->layout = sb->layout; 1178 mddev->raid_disks = sb->raid_disks; 1179 mddev->dev_sectors = ((sector_t)sb->size) * 2; 1180 mddev->events = ev1; 1181 mddev->bitmap_info.offset = 0; 1182 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 1183 1184 if (mddev->minor_version >= 91) { 1185 mddev->reshape_position = sb->reshape_position; 1186 mddev->delta_disks = sb->delta_disks; 1187 mddev->new_level = sb->new_level; 1188 mddev->new_layout = sb->new_layout; 1189 mddev->new_chunk_sectors = sb->new_chunk >> 9; 1190 } else { 1191 mddev->reshape_position = MaxSector; 1192 mddev->delta_disks = 0; 1193 mddev->new_level = mddev->level; 1194 mddev->new_layout = mddev->layout; 1195 mddev->new_chunk_sectors = mddev->chunk_sectors; 1196 } 1197 1198 if (sb->state & (1<<MD_SB_CLEAN)) 1199 mddev->recovery_cp = MaxSector; 1200 else { 1201 if (sb->events_hi == sb->cp_events_hi && 1202 sb->events_lo == sb->cp_events_lo) { 1203 mddev->recovery_cp = sb->recovery_cp; 1204 } else 1205 mddev->recovery_cp = 0; 1206 } 1207 1208 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 1209 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 1210 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 1211 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 1212 1213 mddev->max_disks = MD_SB_DISKS; 1214 1215 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 1216 mddev->bitmap_info.file == NULL) 1217 mddev->bitmap_info.offset = 1218 mddev->bitmap_info.default_offset; 1219 1220 } else if (mddev->pers == NULL) { 1221 /* Insist on good event counter while assembling, except 1222 * for spares (which don't need an event count) */ 1223 ++ev1; 1224 if (sb->disks[rdev->desc_nr].state & ( 1225 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) 1226 if (ev1 < mddev->events) 1227 return -EINVAL; 1228 } else if (mddev->bitmap) { 1229 /* if adding to array with a bitmap, then we can accept an 1230 * older device ... but not too old. 1231 */ 1232 if (ev1 < mddev->bitmap->events_cleared) 1233 return 0; 1234 } else { 1235 if (ev1 < mddev->events) 1236 /* just a hot-add of a new device, leave raid_disk at -1 */ 1237 return 0; 1238 } 1239 1240 if (mddev->level != LEVEL_MULTIPATH) { 1241 desc = sb->disks + rdev->desc_nr; 1242 1243 if (desc->state & (1<<MD_DISK_FAULTY)) 1244 set_bit(Faulty, &rdev->flags); 1245 else if (desc->state & (1<<MD_DISK_SYNC) /* && 1246 desc->raid_disk < mddev->raid_disks */) { 1247 set_bit(In_sync, &rdev->flags); 1248 rdev->raid_disk = desc->raid_disk; 1249 } else if (desc->state & (1<<MD_DISK_ACTIVE)) { 1250 /* active but not in sync implies recovery up to 1251 * reshape position. We don't know exactly where 1252 * that is, so set to zero for now */ 1253 if (mddev->minor_version >= 91) { 1254 rdev->recovery_offset = 0; 1255 rdev->raid_disk = desc->raid_disk; 1256 } 1257 } 1258 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 1259 set_bit(WriteMostly, &rdev->flags); 1260 } else /* MULTIPATH are always insync */ 1261 set_bit(In_sync, &rdev->flags); 1262 return 0; 1263 } 1264 1265 /* 1266 * sync_super for 0.90.0 1267 */ 1268 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1269 { 1270 mdp_super_t *sb; 1271 mdk_rdev_t *rdev2; 1272 int next_spare = mddev->raid_disks; 1273 1274 1275 /* make rdev->sb match mddev data.. 1276 * 1277 * 1/ zero out disks 1278 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 1279 * 3/ any empty disks < next_spare become removed 1280 * 1281 * disks[0] gets initialised to REMOVED because 1282 * we cannot be sure from other fields if it has 1283 * been initialised or not. 1284 */ 1285 int i; 1286 int active=0, working=0,failed=0,spare=0,nr_disks=0; 1287 1288 rdev->sb_size = MD_SB_BYTES; 1289 1290 sb = page_address(rdev->sb_page); 1291 1292 memset(sb, 0, sizeof(*sb)); 1293 1294 sb->md_magic = MD_SB_MAGIC; 1295 sb->major_version = mddev->major_version; 1296 sb->patch_version = mddev->patch_version; 1297 sb->gvalid_words = 0; /* ignored */ 1298 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 1299 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 1300 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 1301 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 1302 1303 sb->ctime = mddev->ctime; 1304 sb->level = mddev->level; 1305 sb->size = mddev->dev_sectors / 2; 1306 sb->raid_disks = mddev->raid_disks; 1307 sb->md_minor = mddev->md_minor; 1308 sb->not_persistent = 0; 1309 sb->utime = mddev->utime; 1310 sb->state = 0; 1311 sb->events_hi = (mddev->events>>32); 1312 sb->events_lo = (u32)mddev->events; 1313 1314 if (mddev->reshape_position == MaxSector) 1315 sb->minor_version = 90; 1316 else { 1317 sb->minor_version = 91; 1318 sb->reshape_position = mddev->reshape_position; 1319 sb->new_level = mddev->new_level; 1320 sb->delta_disks = mddev->delta_disks; 1321 sb->new_layout = mddev->new_layout; 1322 sb->new_chunk = mddev->new_chunk_sectors << 9; 1323 } 1324 mddev->minor_version = sb->minor_version; 1325 if (mddev->in_sync) 1326 { 1327 sb->recovery_cp = mddev->recovery_cp; 1328 sb->cp_events_hi = (mddev->events>>32); 1329 sb->cp_events_lo = (u32)mddev->events; 1330 if (mddev->recovery_cp == MaxSector) 1331 sb->state = (1<< MD_SB_CLEAN); 1332 } else 1333 sb->recovery_cp = 0; 1334 1335 sb->layout = mddev->layout; 1336 sb->chunk_size = mddev->chunk_sectors << 9; 1337 1338 if (mddev->bitmap && mddev->bitmap_info.file == NULL) 1339 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 1340 1341 sb->disks[0].state = (1<<MD_DISK_REMOVED); 1342 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1343 mdp_disk_t *d; 1344 int desc_nr; 1345 int is_active = test_bit(In_sync, &rdev2->flags); 1346 1347 if (rdev2->raid_disk >= 0 && 1348 sb->minor_version >= 91) 1349 /* we have nowhere to store the recovery_offset, 1350 * but if it is not below the reshape_position, 1351 * we can piggy-back on that. 1352 */ 1353 is_active = 1; 1354 if (rdev2->raid_disk < 0 || 1355 test_bit(Faulty, &rdev2->flags)) 1356 is_active = 0; 1357 if (is_active) 1358 desc_nr = rdev2->raid_disk; 1359 else 1360 desc_nr = next_spare++; 1361 rdev2->desc_nr = desc_nr; 1362 d = &sb->disks[rdev2->desc_nr]; 1363 nr_disks++; 1364 d->number = rdev2->desc_nr; 1365 d->major = MAJOR(rdev2->bdev->bd_dev); 1366 d->minor = MINOR(rdev2->bdev->bd_dev); 1367 if (is_active) 1368 d->raid_disk = rdev2->raid_disk; 1369 else 1370 d->raid_disk = rdev2->desc_nr; /* compatibility */ 1371 if (test_bit(Faulty, &rdev2->flags)) 1372 d->state = (1<<MD_DISK_FAULTY); 1373 else if (is_active) { 1374 d->state = (1<<MD_DISK_ACTIVE); 1375 if (test_bit(In_sync, &rdev2->flags)) 1376 d->state |= (1<<MD_DISK_SYNC); 1377 active++; 1378 working++; 1379 } else { 1380 d->state = 0; 1381 spare++; 1382 working++; 1383 } 1384 if (test_bit(WriteMostly, &rdev2->flags)) 1385 d->state |= (1<<MD_DISK_WRITEMOSTLY); 1386 } 1387 /* now set the "removed" and "faulty" bits on any missing devices */ 1388 for (i=0 ; i < mddev->raid_disks ; i++) { 1389 mdp_disk_t *d = &sb->disks[i]; 1390 if (d->state == 0 && d->number == 0) { 1391 d->number = i; 1392 d->raid_disk = i; 1393 d->state = (1<<MD_DISK_REMOVED); 1394 d->state |= (1<<MD_DISK_FAULTY); 1395 failed++; 1396 } 1397 } 1398 sb->nr_disks = nr_disks; 1399 sb->active_disks = active; 1400 sb->working_disks = working; 1401 sb->failed_disks = failed; 1402 sb->spare_disks = spare; 1403 1404 sb->this_disk = sb->disks[rdev->desc_nr]; 1405 sb->sb_csum = calc_sb_csum(sb); 1406 } 1407 1408 /* 1409 * rdev_size_change for 0.90.0 1410 */ 1411 static unsigned long long 1412 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1413 { 1414 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1415 return 0; /* component must fit device */ 1416 if (rdev->mddev->bitmap_info.offset) 1417 return 0; /* can't move bitmap */ 1418 rdev->sb_start = calc_dev_sboffset(rdev); 1419 if (!num_sectors || num_sectors > rdev->sb_start) 1420 num_sectors = rdev->sb_start; 1421 /* Limit to 4TB as metadata cannot record more than that. 1422 * 4TB == 2^32 KB, or 2*2^32 sectors. 1423 */ 1424 if (num_sectors >= (2ULL << 32)) 1425 num_sectors = (2ULL << 32) - 2; 1426 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1427 rdev->sb_page); 1428 md_super_wait(rdev->mddev); 1429 return num_sectors; 1430 } 1431 1432 1433 /* 1434 * version 1 superblock 1435 */ 1436 1437 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) 1438 { 1439 __le32 disk_csum; 1440 u32 csum; 1441 unsigned long long newcsum; 1442 int size = 256 + le32_to_cpu(sb->max_dev)*2; 1443 __le32 *isuper = (__le32*)sb; 1444 int i; 1445 1446 disk_csum = sb->sb_csum; 1447 sb->sb_csum = 0; 1448 newcsum = 0; 1449 for (i=0; size>=4; size -= 4 ) 1450 newcsum += le32_to_cpu(*isuper++); 1451 1452 if (size == 2) 1453 newcsum += le16_to_cpu(*(__le16*) isuper); 1454 1455 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 1456 sb->sb_csum = disk_csum; 1457 return cpu_to_le32(csum); 1458 } 1459 1460 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, 1461 int acknowledged); 1462 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 1463 { 1464 struct mdp_superblock_1 *sb; 1465 int ret; 1466 sector_t sb_start; 1467 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1468 int bmask; 1469 1470 /* 1471 * Calculate the position of the superblock in 512byte sectors. 1472 * It is always aligned to a 4K boundary and 1473 * depeding on minor_version, it can be: 1474 * 0: At least 8K, but less than 12K, from end of device 1475 * 1: At start of device 1476 * 2: 4K from start of device. 1477 */ 1478 switch(minor_version) { 1479 case 0: 1480 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; 1481 sb_start -= 8*2; 1482 sb_start &= ~(sector_t)(4*2-1); 1483 break; 1484 case 1: 1485 sb_start = 0; 1486 break; 1487 case 2: 1488 sb_start = 8; 1489 break; 1490 default: 1491 return -EINVAL; 1492 } 1493 rdev->sb_start = sb_start; 1494 1495 /* superblock is rarely larger than 1K, but it can be larger, 1496 * and it is safe to read 4k, so we do that 1497 */ 1498 ret = read_disk_sb(rdev, 4096); 1499 if (ret) return ret; 1500 1501 1502 sb = page_address(rdev->sb_page); 1503 1504 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 1505 sb->major_version != cpu_to_le32(1) || 1506 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 1507 le64_to_cpu(sb->super_offset) != rdev->sb_start || 1508 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 1509 return -EINVAL; 1510 1511 if (calc_sb_1_csum(sb) != sb->sb_csum) { 1512 printk("md: invalid superblock checksum on %s\n", 1513 bdevname(rdev->bdev,b)); 1514 return -EINVAL; 1515 } 1516 if (le64_to_cpu(sb->data_size) < 10) { 1517 printk("md: data_size too small on %s\n", 1518 bdevname(rdev->bdev,b)); 1519 return -EINVAL; 1520 } 1521 1522 rdev->preferred_minor = 0xffff; 1523 rdev->data_offset = le64_to_cpu(sb->data_offset); 1524 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); 1525 1526 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 1527 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1528 if (rdev->sb_size & bmask) 1529 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1530 1531 if (minor_version 1532 && rdev->data_offset < sb_start + (rdev->sb_size/512)) 1533 return -EINVAL; 1534 1535 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) 1536 rdev->desc_nr = -1; 1537 else 1538 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1539 1540 if (!rdev->bb_page) { 1541 rdev->bb_page = alloc_page(GFP_KERNEL); 1542 if (!rdev->bb_page) 1543 return -ENOMEM; 1544 } 1545 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && 1546 rdev->badblocks.count == 0) { 1547 /* need to load the bad block list. 1548 * Currently we limit it to one page. 1549 */ 1550 s32 offset; 1551 sector_t bb_sector; 1552 u64 *bbp; 1553 int i; 1554 int sectors = le16_to_cpu(sb->bblog_size); 1555 if (sectors > (PAGE_SIZE / 512)) 1556 return -EINVAL; 1557 offset = le32_to_cpu(sb->bblog_offset); 1558 if (offset == 0) 1559 return -EINVAL; 1560 bb_sector = (long long)offset; 1561 if (!sync_page_io(rdev, bb_sector, sectors << 9, 1562 rdev->bb_page, READ, true)) 1563 return -EIO; 1564 bbp = (u64 *)page_address(rdev->bb_page); 1565 rdev->badblocks.shift = sb->bblog_shift; 1566 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { 1567 u64 bb = le64_to_cpu(*bbp); 1568 int count = bb & (0x3ff); 1569 u64 sector = bb >> 10; 1570 sector <<= sb->bblog_shift; 1571 count <<= sb->bblog_shift; 1572 if (bb + 1 == 0) 1573 break; 1574 if (md_set_badblocks(&rdev->badblocks, 1575 sector, count, 1) == 0) 1576 return -EINVAL; 1577 } 1578 } else if (sb->bblog_offset == 0) 1579 rdev->badblocks.shift = -1; 1580 1581 if (!refdev) { 1582 ret = 1; 1583 } else { 1584 __u64 ev1, ev2; 1585 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); 1586 1587 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 1588 sb->level != refsb->level || 1589 sb->layout != refsb->layout || 1590 sb->chunksize != refsb->chunksize) { 1591 printk(KERN_WARNING "md: %s has strangely different" 1592 " superblock to %s\n", 1593 bdevname(rdev->bdev,b), 1594 bdevname(refdev->bdev,b2)); 1595 return -EINVAL; 1596 } 1597 ev1 = le64_to_cpu(sb->events); 1598 ev2 = le64_to_cpu(refsb->events); 1599 1600 if (ev1 > ev2) 1601 ret = 1; 1602 else 1603 ret = 0; 1604 } 1605 if (minor_version) 1606 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 1607 le64_to_cpu(sb->data_offset); 1608 else 1609 rdev->sectors = rdev->sb_start; 1610 if (rdev->sectors < le64_to_cpu(sb->data_size)) 1611 return -EINVAL; 1612 rdev->sectors = le64_to_cpu(sb->data_size); 1613 if (le64_to_cpu(sb->size) > rdev->sectors) 1614 return -EINVAL; 1615 return ret; 1616 } 1617 1618 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1619 { 1620 struct mdp_superblock_1 *sb = page_address(rdev->sb_page); 1621 __u64 ev1 = le64_to_cpu(sb->events); 1622 1623 rdev->raid_disk = -1; 1624 clear_bit(Faulty, &rdev->flags); 1625 clear_bit(In_sync, &rdev->flags); 1626 clear_bit(WriteMostly, &rdev->flags); 1627 1628 if (mddev->raid_disks == 0) { 1629 mddev->major_version = 1; 1630 mddev->patch_version = 0; 1631 mddev->external = 0; 1632 mddev->chunk_sectors = le32_to_cpu(sb->chunksize); 1633 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1634 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1635 mddev->level = le32_to_cpu(sb->level); 1636 mddev->clevel[0] = 0; 1637 mddev->layout = le32_to_cpu(sb->layout); 1638 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1639 mddev->dev_sectors = le64_to_cpu(sb->size); 1640 mddev->events = ev1; 1641 mddev->bitmap_info.offset = 0; 1642 mddev->bitmap_info.default_offset = 1024 >> 9; 1643 1644 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1645 memcpy(mddev->uuid, sb->set_uuid, 16); 1646 1647 mddev->max_disks = (4096-256)/2; 1648 1649 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1650 mddev->bitmap_info.file == NULL ) 1651 mddev->bitmap_info.offset = 1652 (__s32)le32_to_cpu(sb->bitmap_offset); 1653 1654 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { 1655 mddev->reshape_position = le64_to_cpu(sb->reshape_position); 1656 mddev->delta_disks = le32_to_cpu(sb->delta_disks); 1657 mddev->new_level = le32_to_cpu(sb->new_level); 1658 mddev->new_layout = le32_to_cpu(sb->new_layout); 1659 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); 1660 } else { 1661 mddev->reshape_position = MaxSector; 1662 mddev->delta_disks = 0; 1663 mddev->new_level = mddev->level; 1664 mddev->new_layout = mddev->layout; 1665 mddev->new_chunk_sectors = mddev->chunk_sectors; 1666 } 1667 1668 } else if (mddev->pers == NULL) { 1669 /* Insist of good event counter while assembling, except for 1670 * spares (which don't need an event count) */ 1671 ++ev1; 1672 if (rdev->desc_nr >= 0 && 1673 rdev->desc_nr < le32_to_cpu(sb->max_dev) && 1674 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) 1675 if (ev1 < mddev->events) 1676 return -EINVAL; 1677 } else if (mddev->bitmap) { 1678 /* If adding to array with a bitmap, then we can accept an 1679 * older device, but not too old. 1680 */ 1681 if (ev1 < mddev->bitmap->events_cleared) 1682 return 0; 1683 } else { 1684 if (ev1 < mddev->events) 1685 /* just a hot-add of a new device, leave raid_disk at -1 */ 1686 return 0; 1687 } 1688 if (mddev->level != LEVEL_MULTIPATH) { 1689 int role; 1690 if (rdev->desc_nr < 0 || 1691 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { 1692 role = 0xffff; 1693 rdev->desc_nr = -1; 1694 } else 1695 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1696 switch(role) { 1697 case 0xffff: /* spare */ 1698 break; 1699 case 0xfffe: /* faulty */ 1700 set_bit(Faulty, &rdev->flags); 1701 break; 1702 default: 1703 if ((le32_to_cpu(sb->feature_map) & 1704 MD_FEATURE_RECOVERY_OFFSET)) 1705 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); 1706 else 1707 set_bit(In_sync, &rdev->flags); 1708 rdev->raid_disk = role; 1709 break; 1710 } 1711 if (sb->devflags & WriteMostly1) 1712 set_bit(WriteMostly, &rdev->flags); 1713 } else /* MULTIPATH are always insync */ 1714 set_bit(In_sync, &rdev->flags); 1715 1716 return 0; 1717 } 1718 1719 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1720 { 1721 struct mdp_superblock_1 *sb; 1722 mdk_rdev_t *rdev2; 1723 int max_dev, i; 1724 /* make rdev->sb match mddev and rdev data. */ 1725 1726 sb = page_address(rdev->sb_page); 1727 1728 sb->feature_map = 0; 1729 sb->pad0 = 0; 1730 sb->recovery_offset = cpu_to_le64(0); 1731 memset(sb->pad1, 0, sizeof(sb->pad1)); 1732 memset(sb->pad3, 0, sizeof(sb->pad3)); 1733 1734 sb->utime = cpu_to_le64((__u64)mddev->utime); 1735 sb->events = cpu_to_le64(mddev->events); 1736 if (mddev->in_sync) 1737 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1738 else 1739 sb->resync_offset = cpu_to_le64(0); 1740 1741 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); 1742 1743 sb->raid_disks = cpu_to_le32(mddev->raid_disks); 1744 sb->size = cpu_to_le64(mddev->dev_sectors); 1745 sb->chunksize = cpu_to_le32(mddev->chunk_sectors); 1746 sb->level = cpu_to_le32(mddev->level); 1747 sb->layout = cpu_to_le32(mddev->layout); 1748 1749 if (test_bit(WriteMostly, &rdev->flags)) 1750 sb->devflags |= WriteMostly1; 1751 else 1752 sb->devflags &= ~WriteMostly1; 1753 1754 if (mddev->bitmap && mddev->bitmap_info.file == NULL) { 1755 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); 1756 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); 1757 } 1758 1759 if (rdev->raid_disk >= 0 && 1760 !test_bit(In_sync, &rdev->flags)) { 1761 sb->feature_map |= 1762 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); 1763 sb->recovery_offset = 1764 cpu_to_le64(rdev->recovery_offset); 1765 } 1766 1767 if (mddev->reshape_position != MaxSector) { 1768 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); 1769 sb->reshape_position = cpu_to_le64(mddev->reshape_position); 1770 sb->new_layout = cpu_to_le32(mddev->new_layout); 1771 sb->delta_disks = cpu_to_le32(mddev->delta_disks); 1772 sb->new_level = cpu_to_le32(mddev->new_level); 1773 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); 1774 } 1775 1776 if (rdev->badblocks.count == 0) 1777 /* Nothing to do for bad blocks*/ ; 1778 else if (sb->bblog_offset == 0) 1779 /* Cannot record bad blocks on this device */ 1780 md_error(mddev, rdev); 1781 else { 1782 struct badblocks *bb = &rdev->badblocks; 1783 u64 *bbp = (u64 *)page_address(rdev->bb_page); 1784 u64 *p = bb->page; 1785 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); 1786 if (bb->changed) { 1787 unsigned seq; 1788 1789 retry: 1790 seq = read_seqbegin(&bb->lock); 1791 1792 memset(bbp, 0xff, PAGE_SIZE); 1793 1794 for (i = 0 ; i < bb->count ; i++) { 1795 u64 internal_bb = *p++; 1796 u64 store_bb = ((BB_OFFSET(internal_bb) << 10) 1797 | BB_LEN(internal_bb)); 1798 *bbp++ = cpu_to_le64(store_bb); 1799 } 1800 if (read_seqretry(&bb->lock, seq)) 1801 goto retry; 1802 1803 bb->sector = (rdev->sb_start + 1804 (int)le32_to_cpu(sb->bblog_offset)); 1805 bb->size = le16_to_cpu(sb->bblog_size); 1806 bb->changed = 0; 1807 } 1808 } 1809 1810 max_dev = 0; 1811 list_for_each_entry(rdev2, &mddev->disks, same_set) 1812 if (rdev2->desc_nr+1 > max_dev) 1813 max_dev = rdev2->desc_nr+1; 1814 1815 if (max_dev > le32_to_cpu(sb->max_dev)) { 1816 int bmask; 1817 sb->max_dev = cpu_to_le32(max_dev); 1818 rdev->sb_size = max_dev * 2 + 256; 1819 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1820 if (rdev->sb_size & bmask) 1821 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1822 } else 1823 max_dev = le32_to_cpu(sb->max_dev); 1824 1825 for (i=0; i<max_dev;i++) 1826 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1827 1828 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1829 i = rdev2->desc_nr; 1830 if (test_bit(Faulty, &rdev2->flags)) 1831 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1832 else if (test_bit(In_sync, &rdev2->flags)) 1833 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1834 else if (rdev2->raid_disk >= 0) 1835 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1836 else 1837 sb->dev_roles[i] = cpu_to_le16(0xffff); 1838 } 1839 1840 sb->sb_csum = calc_sb_1_csum(sb); 1841 } 1842 1843 static unsigned long long 1844 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1845 { 1846 struct mdp_superblock_1 *sb; 1847 sector_t max_sectors; 1848 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1849 return 0; /* component must fit device */ 1850 if (rdev->sb_start < rdev->data_offset) { 1851 /* minor versions 1 and 2; superblock before data */ 1852 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; 1853 max_sectors -= rdev->data_offset; 1854 if (!num_sectors || num_sectors > max_sectors) 1855 num_sectors = max_sectors; 1856 } else if (rdev->mddev->bitmap_info.offset) { 1857 /* minor version 0 with bitmap we can't move */ 1858 return 0; 1859 } else { 1860 /* minor version 0; superblock after data */ 1861 sector_t sb_start; 1862 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; 1863 sb_start &= ~(sector_t)(4*2 - 1); 1864 max_sectors = rdev->sectors + sb_start - rdev->sb_start; 1865 if (!num_sectors || num_sectors > max_sectors) 1866 num_sectors = max_sectors; 1867 rdev->sb_start = sb_start; 1868 } 1869 sb = page_address(rdev->sb_page); 1870 sb->data_size = cpu_to_le64(num_sectors); 1871 sb->super_offset = rdev->sb_start; 1872 sb->sb_csum = calc_sb_1_csum(sb); 1873 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1874 rdev->sb_page); 1875 md_super_wait(rdev->mddev); 1876 return num_sectors; 1877 } 1878 1879 static struct super_type super_types[] = { 1880 [0] = { 1881 .name = "0.90.0", 1882 .owner = THIS_MODULE, 1883 .load_super = super_90_load, 1884 .validate_super = super_90_validate, 1885 .sync_super = super_90_sync, 1886 .rdev_size_change = super_90_rdev_size_change, 1887 }, 1888 [1] = { 1889 .name = "md-1", 1890 .owner = THIS_MODULE, 1891 .load_super = super_1_load, 1892 .validate_super = super_1_validate, 1893 .sync_super = super_1_sync, 1894 .rdev_size_change = super_1_rdev_size_change, 1895 }, 1896 }; 1897 1898 static void sync_super(mddev_t *mddev, mdk_rdev_t *rdev) 1899 { 1900 if (mddev->sync_super) { 1901 mddev->sync_super(mddev, rdev); 1902 return; 1903 } 1904 1905 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); 1906 1907 super_types[mddev->major_version].sync_super(mddev, rdev); 1908 } 1909 1910 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) 1911 { 1912 mdk_rdev_t *rdev, *rdev2; 1913 1914 rcu_read_lock(); 1915 rdev_for_each_rcu(rdev, mddev1) 1916 rdev_for_each_rcu(rdev2, mddev2) 1917 if (rdev->bdev->bd_contains == 1918 rdev2->bdev->bd_contains) { 1919 rcu_read_unlock(); 1920 return 1; 1921 } 1922 rcu_read_unlock(); 1923 return 0; 1924 } 1925 1926 static LIST_HEAD(pending_raid_disks); 1927 1928 /* 1929 * Try to register data integrity profile for an mddev 1930 * 1931 * This is called when an array is started and after a disk has been kicked 1932 * from the array. It only succeeds if all working and active component devices 1933 * are integrity capable with matching profiles. 1934 */ 1935 int md_integrity_register(mddev_t *mddev) 1936 { 1937 mdk_rdev_t *rdev, *reference = NULL; 1938 1939 if (list_empty(&mddev->disks)) 1940 return 0; /* nothing to do */ 1941 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) 1942 return 0; /* shouldn't register, or already is */ 1943 list_for_each_entry(rdev, &mddev->disks, same_set) { 1944 /* skip spares and non-functional disks */ 1945 if (test_bit(Faulty, &rdev->flags)) 1946 continue; 1947 if (rdev->raid_disk < 0) 1948 continue; 1949 if (!reference) { 1950 /* Use the first rdev as the reference */ 1951 reference = rdev; 1952 continue; 1953 } 1954 /* does this rdev's profile match the reference profile? */ 1955 if (blk_integrity_compare(reference->bdev->bd_disk, 1956 rdev->bdev->bd_disk) < 0) 1957 return -EINVAL; 1958 } 1959 if (!reference || !bdev_get_integrity(reference->bdev)) 1960 return 0; 1961 /* 1962 * All component devices are integrity capable and have matching 1963 * profiles, register the common profile for the md device. 1964 */ 1965 if (blk_integrity_register(mddev->gendisk, 1966 bdev_get_integrity(reference->bdev)) != 0) { 1967 printk(KERN_ERR "md: failed to register integrity for %s\n", 1968 mdname(mddev)); 1969 return -EINVAL; 1970 } 1971 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); 1972 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { 1973 printk(KERN_ERR "md: failed to create integrity pool for %s\n", 1974 mdname(mddev)); 1975 return -EINVAL; 1976 } 1977 return 0; 1978 } 1979 EXPORT_SYMBOL(md_integrity_register); 1980 1981 /* Disable data integrity if non-capable/non-matching disk is being added */ 1982 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 1983 { 1984 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); 1985 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); 1986 1987 if (!bi_mddev) /* nothing to do */ 1988 return; 1989 if (rdev->raid_disk < 0) /* skip spares */ 1990 return; 1991 if (bi_rdev && blk_integrity_compare(mddev->gendisk, 1992 rdev->bdev->bd_disk) >= 0) 1993 return; 1994 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); 1995 blk_integrity_unregister(mddev->gendisk); 1996 } 1997 EXPORT_SYMBOL(md_integrity_add_rdev); 1998 1999 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) 2000 { 2001 char b[BDEVNAME_SIZE]; 2002 struct kobject *ko; 2003 char *s; 2004 int err; 2005 2006 if (rdev->mddev) { 2007 MD_BUG(); 2008 return -EINVAL; 2009 } 2010 2011 /* prevent duplicates */ 2012 if (find_rdev(mddev, rdev->bdev->bd_dev)) 2013 return -EEXIST; 2014 2015 /* make sure rdev->sectors exceeds mddev->dev_sectors */ 2016 if (rdev->sectors && (mddev->dev_sectors == 0 || 2017 rdev->sectors < mddev->dev_sectors)) { 2018 if (mddev->pers) { 2019 /* Cannot change size, so fail 2020 * If mddev->level <= 0, then we don't care 2021 * about aligning sizes (e.g. linear) 2022 */ 2023 if (mddev->level > 0) 2024 return -ENOSPC; 2025 } else 2026 mddev->dev_sectors = rdev->sectors; 2027 } 2028 2029 /* Verify rdev->desc_nr is unique. 2030 * If it is -1, assign a free number, else 2031 * check number is not in use 2032 */ 2033 if (rdev->desc_nr < 0) { 2034 int choice = 0; 2035 if (mddev->pers) choice = mddev->raid_disks; 2036 while (find_rdev_nr(mddev, choice)) 2037 choice++; 2038 rdev->desc_nr = choice; 2039 } else { 2040 if (find_rdev_nr(mddev, rdev->desc_nr)) 2041 return -EBUSY; 2042 } 2043 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { 2044 printk(KERN_WARNING "md: %s: array is limited to %d devices\n", 2045 mdname(mddev), mddev->max_disks); 2046 return -EBUSY; 2047 } 2048 bdevname(rdev->bdev,b); 2049 while ( (s=strchr(b, '/')) != NULL) 2050 *s = '!'; 2051 2052 rdev->mddev = mddev; 2053 printk(KERN_INFO "md: bind<%s>\n", b); 2054 2055 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) 2056 goto fail; 2057 2058 ko = &part_to_dev(rdev->bdev->bd_part)->kobj; 2059 if (sysfs_create_link(&rdev->kobj, ko, "block")) 2060 /* failure here is OK */; 2061 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); 2062 2063 list_add_rcu(&rdev->same_set, &mddev->disks); 2064 bd_link_disk_holder(rdev->bdev, mddev->gendisk); 2065 2066 /* May as well allow recovery to be retried once */ 2067 mddev->recovery_disabled++; 2068 2069 return 0; 2070 2071 fail: 2072 printk(KERN_WARNING "md: failed to register dev-%s for %s\n", 2073 b, mdname(mddev)); 2074 return err; 2075 } 2076 2077 static void md_delayed_delete(struct work_struct *ws) 2078 { 2079 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); 2080 kobject_del(&rdev->kobj); 2081 kobject_put(&rdev->kobj); 2082 } 2083 2084 static void unbind_rdev_from_array(mdk_rdev_t * rdev) 2085 { 2086 char b[BDEVNAME_SIZE]; 2087 if (!rdev->mddev) { 2088 MD_BUG(); 2089 return; 2090 } 2091 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); 2092 list_del_rcu(&rdev->same_set); 2093 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 2094 rdev->mddev = NULL; 2095 sysfs_remove_link(&rdev->kobj, "block"); 2096 sysfs_put(rdev->sysfs_state); 2097 rdev->sysfs_state = NULL; 2098 kfree(rdev->badblocks.page); 2099 rdev->badblocks.count = 0; 2100 rdev->badblocks.page = NULL; 2101 /* We need to delay this, otherwise we can deadlock when 2102 * writing to 'remove' to "dev/state". We also need 2103 * to delay it due to rcu usage. 2104 */ 2105 synchronize_rcu(); 2106 INIT_WORK(&rdev->del_work, md_delayed_delete); 2107 kobject_get(&rdev->kobj); 2108 queue_work(md_misc_wq, &rdev->del_work); 2109 } 2110 2111 /* 2112 * prevent the device from being mounted, repartitioned or 2113 * otherwise reused by a RAID array (or any other kernel 2114 * subsystem), by bd_claiming the device. 2115 */ 2116 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) 2117 { 2118 int err = 0; 2119 struct block_device *bdev; 2120 char b[BDEVNAME_SIZE]; 2121 2122 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, 2123 shared ? (mdk_rdev_t *)lock_rdev : rdev); 2124 if (IS_ERR(bdev)) { 2125 printk(KERN_ERR "md: could not open %s.\n", 2126 __bdevname(dev, b)); 2127 return PTR_ERR(bdev); 2128 } 2129 rdev->bdev = bdev; 2130 return err; 2131 } 2132 2133 static void unlock_rdev(mdk_rdev_t *rdev) 2134 { 2135 struct block_device *bdev = rdev->bdev; 2136 rdev->bdev = NULL; 2137 if (!bdev) 2138 MD_BUG(); 2139 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2140 } 2141 2142 void md_autodetect_dev(dev_t dev); 2143 2144 static void export_rdev(mdk_rdev_t * rdev) 2145 { 2146 char b[BDEVNAME_SIZE]; 2147 printk(KERN_INFO "md: export_rdev(%s)\n", 2148 bdevname(rdev->bdev,b)); 2149 if (rdev->mddev) 2150 MD_BUG(); 2151 free_disk_sb(rdev); 2152 #ifndef MODULE 2153 if (test_bit(AutoDetected, &rdev->flags)) 2154 md_autodetect_dev(rdev->bdev->bd_dev); 2155 #endif 2156 unlock_rdev(rdev); 2157 kobject_put(&rdev->kobj); 2158 } 2159 2160 static void kick_rdev_from_array(mdk_rdev_t * rdev) 2161 { 2162 unbind_rdev_from_array(rdev); 2163 export_rdev(rdev); 2164 } 2165 2166 static void export_array(mddev_t *mddev) 2167 { 2168 mdk_rdev_t *rdev, *tmp; 2169 2170 rdev_for_each(rdev, tmp, mddev) { 2171 if (!rdev->mddev) { 2172 MD_BUG(); 2173 continue; 2174 } 2175 kick_rdev_from_array(rdev); 2176 } 2177 if (!list_empty(&mddev->disks)) 2178 MD_BUG(); 2179 mddev->raid_disks = 0; 2180 mddev->major_version = 0; 2181 } 2182 2183 static void print_desc(mdp_disk_t *desc) 2184 { 2185 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 2186 desc->major,desc->minor,desc->raid_disk,desc->state); 2187 } 2188 2189 static void print_sb_90(mdp_super_t *sb) 2190 { 2191 int i; 2192 2193 printk(KERN_INFO 2194 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 2195 sb->major_version, sb->minor_version, sb->patch_version, 2196 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 2197 sb->ctime); 2198 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 2199 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 2200 sb->md_minor, sb->layout, sb->chunk_size); 2201 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 2202 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 2203 sb->utime, sb->state, sb->active_disks, sb->working_disks, 2204 sb->failed_disks, sb->spare_disks, 2205 sb->sb_csum, (unsigned long)sb->events_lo); 2206 2207 printk(KERN_INFO); 2208 for (i = 0; i < MD_SB_DISKS; i++) { 2209 mdp_disk_t *desc; 2210 2211 desc = sb->disks + i; 2212 if (desc->number || desc->major || desc->minor || 2213 desc->raid_disk || (desc->state && (desc->state != 4))) { 2214 printk(" D %2d: ", i); 2215 print_desc(desc); 2216 } 2217 } 2218 printk(KERN_INFO "md: THIS: "); 2219 print_desc(&sb->this_disk); 2220 } 2221 2222 static void print_sb_1(struct mdp_superblock_1 *sb) 2223 { 2224 __u8 *uuid; 2225 2226 uuid = sb->set_uuid; 2227 printk(KERN_INFO 2228 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" 2229 "md: Name: \"%s\" CT:%llu\n", 2230 le32_to_cpu(sb->major_version), 2231 le32_to_cpu(sb->feature_map), 2232 uuid, 2233 sb->set_name, 2234 (unsigned long long)le64_to_cpu(sb->ctime) 2235 & MD_SUPERBLOCK_1_TIME_SEC_MASK); 2236 2237 uuid = sb->device_uuid; 2238 printk(KERN_INFO 2239 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" 2240 " RO:%llu\n" 2241 "md: Dev:%08x UUID: %pU\n" 2242 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" 2243 "md: (MaxDev:%u) \n", 2244 le32_to_cpu(sb->level), 2245 (unsigned long long)le64_to_cpu(sb->size), 2246 le32_to_cpu(sb->raid_disks), 2247 le32_to_cpu(sb->layout), 2248 le32_to_cpu(sb->chunksize), 2249 (unsigned long long)le64_to_cpu(sb->data_offset), 2250 (unsigned long long)le64_to_cpu(sb->data_size), 2251 (unsigned long long)le64_to_cpu(sb->super_offset), 2252 (unsigned long long)le64_to_cpu(sb->recovery_offset), 2253 le32_to_cpu(sb->dev_number), 2254 uuid, 2255 sb->devflags, 2256 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, 2257 (unsigned long long)le64_to_cpu(sb->events), 2258 (unsigned long long)le64_to_cpu(sb->resync_offset), 2259 le32_to_cpu(sb->sb_csum), 2260 le32_to_cpu(sb->max_dev) 2261 ); 2262 } 2263 2264 static void print_rdev(mdk_rdev_t *rdev, int major_version) 2265 { 2266 char b[BDEVNAME_SIZE]; 2267 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", 2268 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, 2269 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 2270 rdev->desc_nr); 2271 if (rdev->sb_loaded) { 2272 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); 2273 switch (major_version) { 2274 case 0: 2275 print_sb_90(page_address(rdev->sb_page)); 2276 break; 2277 case 1: 2278 print_sb_1(page_address(rdev->sb_page)); 2279 break; 2280 } 2281 } else 2282 printk(KERN_INFO "md: no rdev superblock!\n"); 2283 } 2284 2285 static void md_print_devices(void) 2286 { 2287 struct list_head *tmp; 2288 mdk_rdev_t *rdev; 2289 mddev_t *mddev; 2290 char b[BDEVNAME_SIZE]; 2291 2292 printk("\n"); 2293 printk("md: **********************************\n"); 2294 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 2295 printk("md: **********************************\n"); 2296 for_each_mddev(mddev, tmp) { 2297 2298 if (mddev->bitmap) 2299 bitmap_print_sb(mddev->bitmap); 2300 else 2301 printk("%s: ", mdname(mddev)); 2302 list_for_each_entry(rdev, &mddev->disks, same_set) 2303 printk("<%s>", bdevname(rdev->bdev,b)); 2304 printk("\n"); 2305 2306 list_for_each_entry(rdev, &mddev->disks, same_set) 2307 print_rdev(rdev, mddev->major_version); 2308 } 2309 printk("md: **********************************\n"); 2310 printk("\n"); 2311 } 2312 2313 2314 static void sync_sbs(mddev_t * mddev, int nospares) 2315 { 2316 /* Update each superblock (in-memory image), but 2317 * if we are allowed to, skip spares which already 2318 * have the right event counter, or have one earlier 2319 * (which would mean they aren't being marked as dirty 2320 * with the rest of the array) 2321 */ 2322 mdk_rdev_t *rdev; 2323 list_for_each_entry(rdev, &mddev->disks, same_set) { 2324 if (rdev->sb_events == mddev->events || 2325 (nospares && 2326 rdev->raid_disk < 0 && 2327 rdev->sb_events+1 == mddev->events)) { 2328 /* Don't update this superblock */ 2329 rdev->sb_loaded = 2; 2330 } else { 2331 sync_super(mddev, rdev); 2332 rdev->sb_loaded = 1; 2333 } 2334 } 2335 } 2336 2337 static void md_update_sb(mddev_t * mddev, int force_change) 2338 { 2339 mdk_rdev_t *rdev; 2340 int sync_req; 2341 int nospares = 0; 2342 int any_badblocks_changed = 0; 2343 2344 repeat: 2345 /* First make sure individual recovery_offsets are correct */ 2346 list_for_each_entry(rdev, &mddev->disks, same_set) { 2347 if (rdev->raid_disk >= 0 && 2348 mddev->delta_disks >= 0 && 2349 !test_bit(In_sync, &rdev->flags) && 2350 mddev->curr_resync_completed > rdev->recovery_offset) 2351 rdev->recovery_offset = mddev->curr_resync_completed; 2352 2353 } 2354 if (!mddev->persistent) { 2355 clear_bit(MD_CHANGE_CLEAN, &mddev->flags); 2356 clear_bit(MD_CHANGE_DEVS, &mddev->flags); 2357 if (!mddev->external) { 2358 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2359 list_for_each_entry(rdev, &mddev->disks, same_set) { 2360 if (rdev->badblocks.changed) { 2361 md_ack_all_badblocks(&rdev->badblocks); 2362 md_error(mddev, rdev); 2363 } 2364 clear_bit(Blocked, &rdev->flags); 2365 clear_bit(BlockedBadBlocks, &rdev->flags); 2366 wake_up(&rdev->blocked_wait); 2367 } 2368 } 2369 wake_up(&mddev->sb_wait); 2370 return; 2371 } 2372 2373 spin_lock_irq(&mddev->write_lock); 2374 2375 mddev->utime = get_seconds(); 2376 2377 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) 2378 force_change = 1; 2379 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) 2380 /* just a clean<-> dirty transition, possibly leave spares alone, 2381 * though if events isn't the right even/odd, we will have to do 2382 * spares after all 2383 */ 2384 nospares = 1; 2385 if (force_change) 2386 nospares = 0; 2387 if (mddev->degraded) 2388 /* If the array is degraded, then skipping spares is both 2389 * dangerous and fairly pointless. 2390 * Dangerous because a device that was removed from the array 2391 * might have a event_count that still looks up-to-date, 2392 * so it can be re-added without a resync. 2393 * Pointless because if there are any spares to skip, 2394 * then a recovery will happen and soon that array won't 2395 * be degraded any more and the spare can go back to sleep then. 2396 */ 2397 nospares = 0; 2398 2399 sync_req = mddev->in_sync; 2400 2401 /* If this is just a dirty<->clean transition, and the array is clean 2402 * and 'events' is odd, we can roll back to the previous clean state */ 2403 if (nospares 2404 && (mddev->in_sync && mddev->recovery_cp == MaxSector) 2405 && mddev->can_decrease_events 2406 && mddev->events != 1) { 2407 mddev->events--; 2408 mddev->can_decrease_events = 0; 2409 } else { 2410 /* otherwise we have to go forward and ... */ 2411 mddev->events ++; 2412 mddev->can_decrease_events = nospares; 2413 } 2414 2415 if (!mddev->events) { 2416 /* 2417 * oops, this 64-bit counter should never wrap. 2418 * Either we are in around ~1 trillion A.C., assuming 2419 * 1 reboot per second, or we have a bug: 2420 */ 2421 MD_BUG(); 2422 mddev->events --; 2423 } 2424 2425 list_for_each_entry(rdev, &mddev->disks, same_set) { 2426 if (rdev->badblocks.changed) 2427 any_badblocks_changed++; 2428 if (test_bit(Faulty, &rdev->flags)) 2429 set_bit(FaultRecorded, &rdev->flags); 2430 } 2431 2432 sync_sbs(mddev, nospares); 2433 spin_unlock_irq(&mddev->write_lock); 2434 2435 dprintk(KERN_INFO 2436 "md: updating %s RAID superblock on device (in sync %d)\n", 2437 mdname(mddev),mddev->in_sync); 2438 2439 bitmap_update_sb(mddev->bitmap); 2440 list_for_each_entry(rdev, &mddev->disks, same_set) { 2441 char b[BDEVNAME_SIZE]; 2442 dprintk(KERN_INFO "md: "); 2443 if (rdev->sb_loaded != 1) 2444 continue; /* no noise on spare devices */ 2445 if (test_bit(Faulty, &rdev->flags)) 2446 dprintk("(skipping faulty "); 2447 2448 dprintk("%s ", bdevname(rdev->bdev,b)); 2449 if (!test_bit(Faulty, &rdev->flags)) { 2450 md_super_write(mddev,rdev, 2451 rdev->sb_start, rdev->sb_size, 2452 rdev->sb_page); 2453 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", 2454 bdevname(rdev->bdev,b), 2455 (unsigned long long)rdev->sb_start); 2456 rdev->sb_events = mddev->events; 2457 if (rdev->badblocks.size) { 2458 md_super_write(mddev, rdev, 2459 rdev->badblocks.sector, 2460 rdev->badblocks.size << 9, 2461 rdev->bb_page); 2462 rdev->badblocks.size = 0; 2463 } 2464 2465 } else 2466 dprintk(")\n"); 2467 if (mddev->level == LEVEL_MULTIPATH) 2468 /* only need to write one superblock... */ 2469 break; 2470 } 2471 md_super_wait(mddev); 2472 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ 2473 2474 spin_lock_irq(&mddev->write_lock); 2475 if (mddev->in_sync != sync_req || 2476 test_bit(MD_CHANGE_DEVS, &mddev->flags)) { 2477 /* have to write it out again */ 2478 spin_unlock_irq(&mddev->write_lock); 2479 goto repeat; 2480 } 2481 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2482 spin_unlock_irq(&mddev->write_lock); 2483 wake_up(&mddev->sb_wait); 2484 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 2485 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 2486 2487 list_for_each_entry(rdev, &mddev->disks, same_set) { 2488 if (test_and_clear_bit(FaultRecorded, &rdev->flags)) 2489 clear_bit(Blocked, &rdev->flags); 2490 2491 if (any_badblocks_changed) 2492 md_ack_all_badblocks(&rdev->badblocks); 2493 clear_bit(BlockedBadBlocks, &rdev->flags); 2494 wake_up(&rdev->blocked_wait); 2495 } 2496 } 2497 2498 /* words written to sysfs files may, or may not, be \n terminated. 2499 * We want to accept with case. For this we use cmd_match. 2500 */ 2501 static int cmd_match(const char *cmd, const char *str) 2502 { 2503 /* See if cmd, written into a sysfs file, matches 2504 * str. They must either be the same, or cmd can 2505 * have a trailing newline 2506 */ 2507 while (*cmd && *str && *cmd == *str) { 2508 cmd++; 2509 str++; 2510 } 2511 if (*cmd == '\n') 2512 cmd++; 2513 if (*str || *cmd) 2514 return 0; 2515 return 1; 2516 } 2517 2518 struct rdev_sysfs_entry { 2519 struct attribute attr; 2520 ssize_t (*show)(mdk_rdev_t *, char *); 2521 ssize_t (*store)(mdk_rdev_t *, const char *, size_t); 2522 }; 2523 2524 static ssize_t 2525 state_show(mdk_rdev_t *rdev, char *page) 2526 { 2527 char *sep = ""; 2528 size_t len = 0; 2529 2530 if (test_bit(Faulty, &rdev->flags) || 2531 rdev->badblocks.unacked_exist) { 2532 len+= sprintf(page+len, "%sfaulty",sep); 2533 sep = ","; 2534 } 2535 if (test_bit(In_sync, &rdev->flags)) { 2536 len += sprintf(page+len, "%sin_sync",sep); 2537 sep = ","; 2538 } 2539 if (test_bit(WriteMostly, &rdev->flags)) { 2540 len += sprintf(page+len, "%swrite_mostly",sep); 2541 sep = ","; 2542 } 2543 if (test_bit(Blocked, &rdev->flags) || 2544 rdev->badblocks.unacked_exist) { 2545 len += sprintf(page+len, "%sblocked", sep); 2546 sep = ","; 2547 } 2548 if (!test_bit(Faulty, &rdev->flags) && 2549 !test_bit(In_sync, &rdev->flags)) { 2550 len += sprintf(page+len, "%sspare", sep); 2551 sep = ","; 2552 } 2553 if (test_bit(WriteErrorSeen, &rdev->flags)) { 2554 len += sprintf(page+len, "%swrite_error", sep); 2555 sep = ","; 2556 } 2557 return len+sprintf(page+len, "\n"); 2558 } 2559 2560 static ssize_t 2561 state_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2562 { 2563 /* can write 2564 * faulty - simulates an error 2565 * remove - disconnects the device 2566 * writemostly - sets write_mostly 2567 * -writemostly - clears write_mostly 2568 * blocked - sets the Blocked flags 2569 * -blocked - clears the Blocked and possibly simulates an error 2570 * insync - sets Insync providing device isn't active 2571 * write_error - sets WriteErrorSeen 2572 * -write_error - clears WriteErrorSeen 2573 */ 2574 int err = -EINVAL; 2575 if (cmd_match(buf, "faulty") && rdev->mddev->pers) { 2576 md_error(rdev->mddev, rdev); 2577 if (test_bit(Faulty, &rdev->flags)) 2578 err = 0; 2579 else 2580 err = -EBUSY; 2581 } else if (cmd_match(buf, "remove")) { 2582 if (rdev->raid_disk >= 0) 2583 err = -EBUSY; 2584 else { 2585 mddev_t *mddev = rdev->mddev; 2586 kick_rdev_from_array(rdev); 2587 if (mddev->pers) 2588 md_update_sb(mddev, 1); 2589 md_new_event(mddev); 2590 err = 0; 2591 } 2592 } else if (cmd_match(buf, "writemostly")) { 2593 set_bit(WriteMostly, &rdev->flags); 2594 err = 0; 2595 } else if (cmd_match(buf, "-writemostly")) { 2596 clear_bit(WriteMostly, &rdev->flags); 2597 err = 0; 2598 } else if (cmd_match(buf, "blocked")) { 2599 set_bit(Blocked, &rdev->flags); 2600 err = 0; 2601 } else if (cmd_match(buf, "-blocked")) { 2602 if (!test_bit(Faulty, &rdev->flags) && 2603 rdev->badblocks.unacked_exist) { 2604 /* metadata handler doesn't understand badblocks, 2605 * so we need to fail the device 2606 */ 2607 md_error(rdev->mddev, rdev); 2608 } 2609 clear_bit(Blocked, &rdev->flags); 2610 clear_bit(BlockedBadBlocks, &rdev->flags); 2611 wake_up(&rdev->blocked_wait); 2612 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2613 md_wakeup_thread(rdev->mddev->thread); 2614 2615 err = 0; 2616 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { 2617 set_bit(In_sync, &rdev->flags); 2618 err = 0; 2619 } else if (cmd_match(buf, "write_error")) { 2620 set_bit(WriteErrorSeen, &rdev->flags); 2621 err = 0; 2622 } else if (cmd_match(buf, "-write_error")) { 2623 clear_bit(WriteErrorSeen, &rdev->flags); 2624 err = 0; 2625 } 2626 if (!err) 2627 sysfs_notify_dirent_safe(rdev->sysfs_state); 2628 return err ? err : len; 2629 } 2630 static struct rdev_sysfs_entry rdev_state = 2631 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); 2632 2633 static ssize_t 2634 errors_show(mdk_rdev_t *rdev, char *page) 2635 { 2636 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); 2637 } 2638 2639 static ssize_t 2640 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2641 { 2642 char *e; 2643 unsigned long n = simple_strtoul(buf, &e, 10); 2644 if (*buf && (*e == 0 || *e == '\n')) { 2645 atomic_set(&rdev->corrected_errors, n); 2646 return len; 2647 } 2648 return -EINVAL; 2649 } 2650 static struct rdev_sysfs_entry rdev_errors = 2651 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); 2652 2653 static ssize_t 2654 slot_show(mdk_rdev_t *rdev, char *page) 2655 { 2656 if (rdev->raid_disk < 0) 2657 return sprintf(page, "none\n"); 2658 else 2659 return sprintf(page, "%d\n", rdev->raid_disk); 2660 } 2661 2662 static ssize_t 2663 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2664 { 2665 char *e; 2666 int err; 2667 int slot = simple_strtoul(buf, &e, 10); 2668 if (strncmp(buf, "none", 4)==0) 2669 slot = -1; 2670 else if (e==buf || (*e && *e!= '\n')) 2671 return -EINVAL; 2672 if (rdev->mddev->pers && slot == -1) { 2673 /* Setting 'slot' on an active array requires also 2674 * updating the 'rd%d' link, and communicating 2675 * with the personality with ->hot_*_disk. 2676 * For now we only support removing 2677 * failed/spare devices. This normally happens automatically, 2678 * but not when the metadata is externally managed. 2679 */ 2680 if (rdev->raid_disk == -1) 2681 return -EEXIST; 2682 /* personality does all needed checks */ 2683 if (rdev->mddev->pers->hot_remove_disk == NULL) 2684 return -EINVAL; 2685 err = rdev->mddev->pers-> 2686 hot_remove_disk(rdev->mddev, rdev->raid_disk); 2687 if (err) 2688 return err; 2689 sysfs_unlink_rdev(rdev->mddev, rdev); 2690 rdev->raid_disk = -1; 2691 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2692 md_wakeup_thread(rdev->mddev->thread); 2693 } else if (rdev->mddev->pers) { 2694 mdk_rdev_t *rdev2; 2695 /* Activating a spare .. or possibly reactivating 2696 * if we ever get bitmaps working here. 2697 */ 2698 2699 if (rdev->raid_disk != -1) 2700 return -EBUSY; 2701 2702 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) 2703 return -EBUSY; 2704 2705 if (rdev->mddev->pers->hot_add_disk == NULL) 2706 return -EINVAL; 2707 2708 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) 2709 if (rdev2->raid_disk == slot) 2710 return -EEXIST; 2711 2712 if (slot >= rdev->mddev->raid_disks && 2713 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) 2714 return -ENOSPC; 2715 2716 rdev->raid_disk = slot; 2717 if (test_bit(In_sync, &rdev->flags)) 2718 rdev->saved_raid_disk = slot; 2719 else 2720 rdev->saved_raid_disk = -1; 2721 err = rdev->mddev->pers-> 2722 hot_add_disk(rdev->mddev, rdev); 2723 if (err) { 2724 rdev->raid_disk = -1; 2725 return err; 2726 } else 2727 sysfs_notify_dirent_safe(rdev->sysfs_state); 2728 if (sysfs_link_rdev(rdev->mddev, rdev)) 2729 /* failure here is OK */; 2730 /* don't wakeup anyone, leave that to userspace. */ 2731 } else { 2732 if (slot >= rdev->mddev->raid_disks && 2733 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) 2734 return -ENOSPC; 2735 rdev->raid_disk = slot; 2736 /* assume it is working */ 2737 clear_bit(Faulty, &rdev->flags); 2738 clear_bit(WriteMostly, &rdev->flags); 2739 set_bit(In_sync, &rdev->flags); 2740 sysfs_notify_dirent_safe(rdev->sysfs_state); 2741 } 2742 return len; 2743 } 2744 2745 2746 static struct rdev_sysfs_entry rdev_slot = 2747 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); 2748 2749 static ssize_t 2750 offset_show(mdk_rdev_t *rdev, char *page) 2751 { 2752 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); 2753 } 2754 2755 static ssize_t 2756 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2757 { 2758 char *e; 2759 unsigned long long offset = simple_strtoull(buf, &e, 10); 2760 if (e==buf || (*e && *e != '\n')) 2761 return -EINVAL; 2762 if (rdev->mddev->pers && rdev->raid_disk >= 0) 2763 return -EBUSY; 2764 if (rdev->sectors && rdev->mddev->external) 2765 /* Must set offset before size, so overlap checks 2766 * can be sane */ 2767 return -EBUSY; 2768 rdev->data_offset = offset; 2769 return len; 2770 } 2771 2772 static struct rdev_sysfs_entry rdev_offset = 2773 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); 2774 2775 static ssize_t 2776 rdev_size_show(mdk_rdev_t *rdev, char *page) 2777 { 2778 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); 2779 } 2780 2781 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) 2782 { 2783 /* check if two start/length pairs overlap */ 2784 if (s1+l1 <= s2) 2785 return 0; 2786 if (s2+l2 <= s1) 2787 return 0; 2788 return 1; 2789 } 2790 2791 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) 2792 { 2793 unsigned long long blocks; 2794 sector_t new; 2795 2796 if (strict_strtoull(buf, 10, &blocks) < 0) 2797 return -EINVAL; 2798 2799 if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) 2800 return -EINVAL; /* sector conversion overflow */ 2801 2802 new = blocks * 2; 2803 if (new != blocks * 2) 2804 return -EINVAL; /* unsigned long long to sector_t overflow */ 2805 2806 *sectors = new; 2807 return 0; 2808 } 2809 2810 static ssize_t 2811 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2812 { 2813 mddev_t *my_mddev = rdev->mddev; 2814 sector_t oldsectors = rdev->sectors; 2815 sector_t sectors; 2816 2817 if (strict_blocks_to_sectors(buf, §ors) < 0) 2818 return -EINVAL; 2819 if (my_mddev->pers && rdev->raid_disk >= 0) { 2820 if (my_mddev->persistent) { 2821 sectors = super_types[my_mddev->major_version]. 2822 rdev_size_change(rdev, sectors); 2823 if (!sectors) 2824 return -EBUSY; 2825 } else if (!sectors) 2826 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 2827 rdev->data_offset; 2828 } 2829 if (sectors < my_mddev->dev_sectors) 2830 return -EINVAL; /* component must fit device */ 2831 2832 rdev->sectors = sectors; 2833 if (sectors > oldsectors && my_mddev->external) { 2834 /* need to check that all other rdevs with the same ->bdev 2835 * do not overlap. We need to unlock the mddev to avoid 2836 * a deadlock. We have already changed rdev->sectors, and if 2837 * we have to change it back, we will have the lock again. 2838 */ 2839 mddev_t *mddev; 2840 int overlap = 0; 2841 struct list_head *tmp; 2842 2843 mddev_unlock(my_mddev); 2844 for_each_mddev(mddev, tmp) { 2845 mdk_rdev_t *rdev2; 2846 2847 mddev_lock(mddev); 2848 list_for_each_entry(rdev2, &mddev->disks, same_set) 2849 if (rdev->bdev == rdev2->bdev && 2850 rdev != rdev2 && 2851 overlaps(rdev->data_offset, rdev->sectors, 2852 rdev2->data_offset, 2853 rdev2->sectors)) { 2854 overlap = 1; 2855 break; 2856 } 2857 mddev_unlock(mddev); 2858 if (overlap) { 2859 mddev_put(mddev); 2860 break; 2861 } 2862 } 2863 mddev_lock(my_mddev); 2864 if (overlap) { 2865 /* Someone else could have slipped in a size 2866 * change here, but doing so is just silly. 2867 * We put oldsectors back because we *know* it is 2868 * safe, and trust userspace not to race with 2869 * itself 2870 */ 2871 rdev->sectors = oldsectors; 2872 return -EBUSY; 2873 } 2874 } 2875 return len; 2876 } 2877 2878 static struct rdev_sysfs_entry rdev_size = 2879 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); 2880 2881 2882 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) 2883 { 2884 unsigned long long recovery_start = rdev->recovery_offset; 2885 2886 if (test_bit(In_sync, &rdev->flags) || 2887 recovery_start == MaxSector) 2888 return sprintf(page, "none\n"); 2889 2890 return sprintf(page, "%llu\n", recovery_start); 2891 } 2892 2893 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2894 { 2895 unsigned long long recovery_start; 2896 2897 if (cmd_match(buf, "none")) 2898 recovery_start = MaxSector; 2899 else if (strict_strtoull(buf, 10, &recovery_start)) 2900 return -EINVAL; 2901 2902 if (rdev->mddev->pers && 2903 rdev->raid_disk >= 0) 2904 return -EBUSY; 2905 2906 rdev->recovery_offset = recovery_start; 2907 if (recovery_start == MaxSector) 2908 set_bit(In_sync, &rdev->flags); 2909 else 2910 clear_bit(In_sync, &rdev->flags); 2911 return len; 2912 } 2913 2914 static struct rdev_sysfs_entry rdev_recovery_start = 2915 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); 2916 2917 2918 static ssize_t 2919 badblocks_show(struct badblocks *bb, char *page, int unack); 2920 static ssize_t 2921 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); 2922 2923 static ssize_t bb_show(mdk_rdev_t *rdev, char *page) 2924 { 2925 return badblocks_show(&rdev->badblocks, page, 0); 2926 } 2927 static ssize_t bb_store(mdk_rdev_t *rdev, const char *page, size_t len) 2928 { 2929 int rv = badblocks_store(&rdev->badblocks, page, len, 0); 2930 /* Maybe that ack was all we needed */ 2931 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) 2932 wake_up(&rdev->blocked_wait); 2933 return rv; 2934 } 2935 static struct rdev_sysfs_entry rdev_bad_blocks = 2936 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); 2937 2938 2939 static ssize_t ubb_show(mdk_rdev_t *rdev, char *page) 2940 { 2941 return badblocks_show(&rdev->badblocks, page, 1); 2942 } 2943 static ssize_t ubb_store(mdk_rdev_t *rdev, const char *page, size_t len) 2944 { 2945 return badblocks_store(&rdev->badblocks, page, len, 1); 2946 } 2947 static struct rdev_sysfs_entry rdev_unack_bad_blocks = 2948 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); 2949 2950 static struct attribute *rdev_default_attrs[] = { 2951 &rdev_state.attr, 2952 &rdev_errors.attr, 2953 &rdev_slot.attr, 2954 &rdev_offset.attr, 2955 &rdev_size.attr, 2956 &rdev_recovery_start.attr, 2957 &rdev_bad_blocks.attr, 2958 &rdev_unack_bad_blocks.attr, 2959 NULL, 2960 }; 2961 static ssize_t 2962 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 2963 { 2964 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2965 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2966 mddev_t *mddev = rdev->mddev; 2967 ssize_t rv; 2968 2969 if (!entry->show) 2970 return -EIO; 2971 2972 rv = mddev ? mddev_lock(mddev) : -EBUSY; 2973 if (!rv) { 2974 if (rdev->mddev == NULL) 2975 rv = -EBUSY; 2976 else 2977 rv = entry->show(rdev, page); 2978 mddev_unlock(mddev); 2979 } 2980 return rv; 2981 } 2982 2983 static ssize_t 2984 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 2985 const char *page, size_t length) 2986 { 2987 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2988 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2989 ssize_t rv; 2990 mddev_t *mddev = rdev->mddev; 2991 2992 if (!entry->store) 2993 return -EIO; 2994 if (!capable(CAP_SYS_ADMIN)) 2995 return -EACCES; 2996 rv = mddev ? mddev_lock(mddev): -EBUSY; 2997 if (!rv) { 2998 if (rdev->mddev == NULL) 2999 rv = -EBUSY; 3000 else 3001 rv = entry->store(rdev, page, length); 3002 mddev_unlock(mddev); 3003 } 3004 return rv; 3005 } 3006 3007 static void rdev_free(struct kobject *ko) 3008 { 3009 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); 3010 kfree(rdev); 3011 } 3012 static const struct sysfs_ops rdev_sysfs_ops = { 3013 .show = rdev_attr_show, 3014 .store = rdev_attr_store, 3015 }; 3016 static struct kobj_type rdev_ktype = { 3017 .release = rdev_free, 3018 .sysfs_ops = &rdev_sysfs_ops, 3019 .default_attrs = rdev_default_attrs, 3020 }; 3021 3022 int md_rdev_init(mdk_rdev_t *rdev) 3023 { 3024 rdev->desc_nr = -1; 3025 rdev->saved_raid_disk = -1; 3026 rdev->raid_disk = -1; 3027 rdev->flags = 0; 3028 rdev->data_offset = 0; 3029 rdev->sb_events = 0; 3030 rdev->last_read_error.tv_sec = 0; 3031 rdev->last_read_error.tv_nsec = 0; 3032 rdev->sb_loaded = 0; 3033 rdev->bb_page = NULL; 3034 atomic_set(&rdev->nr_pending, 0); 3035 atomic_set(&rdev->read_errors, 0); 3036 atomic_set(&rdev->corrected_errors, 0); 3037 3038 INIT_LIST_HEAD(&rdev->same_set); 3039 init_waitqueue_head(&rdev->blocked_wait); 3040 3041 /* Add space to store bad block list. 3042 * This reserves the space even on arrays where it cannot 3043 * be used - I wonder if that matters 3044 */ 3045 rdev->badblocks.count = 0; 3046 rdev->badblocks.shift = 0; 3047 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); 3048 seqlock_init(&rdev->badblocks.lock); 3049 if (rdev->badblocks.page == NULL) 3050 return -ENOMEM; 3051 3052 return 0; 3053 } 3054 EXPORT_SYMBOL_GPL(md_rdev_init); 3055 /* 3056 * Import a device. If 'super_format' >= 0, then sanity check the superblock 3057 * 3058 * mark the device faulty if: 3059 * 3060 * - the device is nonexistent (zero size) 3061 * - the device has no valid superblock 3062 * 3063 * a faulty rdev _never_ has rdev->sb set. 3064 */ 3065 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) 3066 { 3067 char b[BDEVNAME_SIZE]; 3068 int err; 3069 mdk_rdev_t *rdev; 3070 sector_t size; 3071 3072 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); 3073 if (!rdev) { 3074 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 3075 return ERR_PTR(-ENOMEM); 3076 } 3077 3078 err = md_rdev_init(rdev); 3079 if (err) 3080 goto abort_free; 3081 err = alloc_disk_sb(rdev); 3082 if (err) 3083 goto abort_free; 3084 3085 err = lock_rdev(rdev, newdev, super_format == -2); 3086 if (err) 3087 goto abort_free; 3088 3089 kobject_init(&rdev->kobj, &rdev_ktype); 3090 3091 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; 3092 if (!size) { 3093 printk(KERN_WARNING 3094 "md: %s has zero or unknown size, marking faulty!\n", 3095 bdevname(rdev->bdev,b)); 3096 err = -EINVAL; 3097 goto abort_free; 3098 } 3099 3100 if (super_format >= 0) { 3101 err = super_types[super_format]. 3102 load_super(rdev, NULL, super_minor); 3103 if (err == -EINVAL) { 3104 printk(KERN_WARNING 3105 "md: %s does not have a valid v%d.%d " 3106 "superblock, not importing!\n", 3107 bdevname(rdev->bdev,b), 3108 super_format, super_minor); 3109 goto abort_free; 3110 } 3111 if (err < 0) { 3112 printk(KERN_WARNING 3113 "md: could not read %s's sb, not importing!\n", 3114 bdevname(rdev->bdev,b)); 3115 goto abort_free; 3116 } 3117 } 3118 if (super_format == -1) 3119 /* hot-add for 0.90, or non-persistent: so no badblocks */ 3120 rdev->badblocks.shift = -1; 3121 3122 return rdev; 3123 3124 abort_free: 3125 if (rdev->bdev) 3126 unlock_rdev(rdev); 3127 free_disk_sb(rdev); 3128 kfree(rdev->badblocks.page); 3129 kfree(rdev); 3130 return ERR_PTR(err); 3131 } 3132 3133 /* 3134 * Check a full RAID array for plausibility 3135 */ 3136 3137 3138 static void analyze_sbs(mddev_t * mddev) 3139 { 3140 int i; 3141 mdk_rdev_t *rdev, *freshest, *tmp; 3142 char b[BDEVNAME_SIZE]; 3143 3144 freshest = NULL; 3145 rdev_for_each(rdev, tmp, mddev) 3146 switch (super_types[mddev->major_version]. 3147 load_super(rdev, freshest, mddev->minor_version)) { 3148 case 1: 3149 freshest = rdev; 3150 break; 3151 case 0: 3152 break; 3153 default: 3154 printk( KERN_ERR \ 3155 "md: fatal superblock inconsistency in %s" 3156 " -- removing from array\n", 3157 bdevname(rdev->bdev,b)); 3158 kick_rdev_from_array(rdev); 3159 } 3160 3161 3162 super_types[mddev->major_version]. 3163 validate_super(mddev, freshest); 3164 3165 i = 0; 3166 rdev_for_each(rdev, tmp, mddev) { 3167 if (mddev->max_disks && 3168 (rdev->desc_nr >= mddev->max_disks || 3169 i > mddev->max_disks)) { 3170 printk(KERN_WARNING 3171 "md: %s: %s: only %d devices permitted\n", 3172 mdname(mddev), bdevname(rdev->bdev, b), 3173 mddev->max_disks); 3174 kick_rdev_from_array(rdev); 3175 continue; 3176 } 3177 if (rdev != freshest) 3178 if (super_types[mddev->major_version]. 3179 validate_super(mddev, rdev)) { 3180 printk(KERN_WARNING "md: kicking non-fresh %s" 3181 " from array!\n", 3182 bdevname(rdev->bdev,b)); 3183 kick_rdev_from_array(rdev); 3184 continue; 3185 } 3186 if (mddev->level == LEVEL_MULTIPATH) { 3187 rdev->desc_nr = i++; 3188 rdev->raid_disk = rdev->desc_nr; 3189 set_bit(In_sync, &rdev->flags); 3190 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { 3191 rdev->raid_disk = -1; 3192 clear_bit(In_sync, &rdev->flags); 3193 } 3194 } 3195 } 3196 3197 /* Read a fixed-point number. 3198 * Numbers in sysfs attributes should be in "standard" units where 3199 * possible, so time should be in seconds. 3200 * However we internally use a a much smaller unit such as 3201 * milliseconds or jiffies. 3202 * This function takes a decimal number with a possible fractional 3203 * component, and produces an integer which is the result of 3204 * multiplying that number by 10^'scale'. 3205 * all without any floating-point arithmetic. 3206 */ 3207 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) 3208 { 3209 unsigned long result = 0; 3210 long decimals = -1; 3211 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { 3212 if (*cp == '.') 3213 decimals = 0; 3214 else if (decimals < scale) { 3215 unsigned int value; 3216 value = *cp - '0'; 3217 result = result * 10 + value; 3218 if (decimals >= 0) 3219 decimals++; 3220 } 3221 cp++; 3222 } 3223 if (*cp == '\n') 3224 cp++; 3225 if (*cp) 3226 return -EINVAL; 3227 if (decimals < 0) 3228 decimals = 0; 3229 while (decimals < scale) { 3230 result *= 10; 3231 decimals ++; 3232 } 3233 *res = result; 3234 return 0; 3235 } 3236 3237 3238 static void md_safemode_timeout(unsigned long data); 3239 3240 static ssize_t 3241 safe_delay_show(mddev_t *mddev, char *page) 3242 { 3243 int msec = (mddev->safemode_delay*1000)/HZ; 3244 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); 3245 } 3246 static ssize_t 3247 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) 3248 { 3249 unsigned long msec; 3250 3251 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) 3252 return -EINVAL; 3253 if (msec == 0) 3254 mddev->safemode_delay = 0; 3255 else { 3256 unsigned long old_delay = mddev->safemode_delay; 3257 mddev->safemode_delay = (msec*HZ)/1000; 3258 if (mddev->safemode_delay == 0) 3259 mddev->safemode_delay = 1; 3260 if (mddev->safemode_delay < old_delay) 3261 md_safemode_timeout((unsigned long)mddev); 3262 } 3263 return len; 3264 } 3265 static struct md_sysfs_entry md_safe_delay = 3266 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); 3267 3268 static ssize_t 3269 level_show(mddev_t *mddev, char *page) 3270 { 3271 struct mdk_personality *p = mddev->pers; 3272 if (p) 3273 return sprintf(page, "%s\n", p->name); 3274 else if (mddev->clevel[0]) 3275 return sprintf(page, "%s\n", mddev->clevel); 3276 else if (mddev->level != LEVEL_NONE) 3277 return sprintf(page, "%d\n", mddev->level); 3278 else 3279 return 0; 3280 } 3281 3282 static ssize_t 3283 level_store(mddev_t *mddev, const char *buf, size_t len) 3284 { 3285 char clevel[16]; 3286 ssize_t rv = len; 3287 struct mdk_personality *pers; 3288 long level; 3289 void *priv; 3290 mdk_rdev_t *rdev; 3291 3292 if (mddev->pers == NULL) { 3293 if (len == 0) 3294 return 0; 3295 if (len >= sizeof(mddev->clevel)) 3296 return -ENOSPC; 3297 strncpy(mddev->clevel, buf, len); 3298 if (mddev->clevel[len-1] == '\n') 3299 len--; 3300 mddev->clevel[len] = 0; 3301 mddev->level = LEVEL_NONE; 3302 return rv; 3303 } 3304 3305 /* request to change the personality. Need to ensure: 3306 * - array is not engaged in resync/recovery/reshape 3307 * - old personality can be suspended 3308 * - new personality will access other array. 3309 */ 3310 3311 if (mddev->sync_thread || 3312 mddev->reshape_position != MaxSector || 3313 mddev->sysfs_active) 3314 return -EBUSY; 3315 3316 if (!mddev->pers->quiesce) { 3317 printk(KERN_WARNING "md: %s: %s does not support online personality change\n", 3318 mdname(mddev), mddev->pers->name); 3319 return -EINVAL; 3320 } 3321 3322 /* Now find the new personality */ 3323 if (len == 0 || len >= sizeof(clevel)) 3324 return -EINVAL; 3325 strncpy(clevel, buf, len); 3326 if (clevel[len-1] == '\n') 3327 len--; 3328 clevel[len] = 0; 3329 if (strict_strtol(clevel, 10, &level)) 3330 level = LEVEL_NONE; 3331 3332 if (request_module("md-%s", clevel) != 0) 3333 request_module("md-level-%s", clevel); 3334 spin_lock(&pers_lock); 3335 pers = find_pers(level, clevel); 3336 if (!pers || !try_module_get(pers->owner)) { 3337 spin_unlock(&pers_lock); 3338 printk(KERN_WARNING "md: personality %s not loaded\n", clevel); 3339 return -EINVAL; 3340 } 3341 spin_unlock(&pers_lock); 3342 3343 if (pers == mddev->pers) { 3344 /* Nothing to do! */ 3345 module_put(pers->owner); 3346 return rv; 3347 } 3348 if (!pers->takeover) { 3349 module_put(pers->owner); 3350 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", 3351 mdname(mddev), clevel); 3352 return -EINVAL; 3353 } 3354 3355 list_for_each_entry(rdev, &mddev->disks, same_set) 3356 rdev->new_raid_disk = rdev->raid_disk; 3357 3358 /* ->takeover must set new_* and/or delta_disks 3359 * if it succeeds, and may set them when it fails. 3360 */ 3361 priv = pers->takeover(mddev); 3362 if (IS_ERR(priv)) { 3363 mddev->new_level = mddev->level; 3364 mddev->new_layout = mddev->layout; 3365 mddev->new_chunk_sectors = mddev->chunk_sectors; 3366 mddev->raid_disks -= mddev->delta_disks; 3367 mddev->delta_disks = 0; 3368 module_put(pers->owner); 3369 printk(KERN_WARNING "md: %s: %s would not accept array\n", 3370 mdname(mddev), clevel); 3371 return PTR_ERR(priv); 3372 } 3373 3374 /* Looks like we have a winner */ 3375 mddev_suspend(mddev); 3376 mddev->pers->stop(mddev); 3377 3378 if (mddev->pers->sync_request == NULL && 3379 pers->sync_request != NULL) { 3380 /* need to add the md_redundancy_group */ 3381 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 3382 printk(KERN_WARNING 3383 "md: cannot register extra attributes for %s\n", 3384 mdname(mddev)); 3385 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); 3386 } 3387 if (mddev->pers->sync_request != NULL && 3388 pers->sync_request == NULL) { 3389 /* need to remove the md_redundancy_group */ 3390 if (mddev->to_remove == NULL) 3391 mddev->to_remove = &md_redundancy_group; 3392 } 3393 3394 if (mddev->pers->sync_request == NULL && 3395 mddev->external) { 3396 /* We are converting from a no-redundancy array 3397 * to a redundancy array and metadata is managed 3398 * externally so we need to be sure that writes 3399 * won't block due to a need to transition 3400 * clean->dirty 3401 * until external management is started. 3402 */ 3403 mddev->in_sync = 0; 3404 mddev->safemode_delay = 0; 3405 mddev->safemode = 0; 3406 } 3407 3408 list_for_each_entry(rdev, &mddev->disks, same_set) { 3409 if (rdev->raid_disk < 0) 3410 continue; 3411 if (rdev->new_raid_disk >= mddev->raid_disks) 3412 rdev->new_raid_disk = -1; 3413 if (rdev->new_raid_disk == rdev->raid_disk) 3414 continue; 3415 sysfs_unlink_rdev(mddev, rdev); 3416 } 3417 list_for_each_entry(rdev, &mddev->disks, same_set) { 3418 if (rdev->raid_disk < 0) 3419 continue; 3420 if (rdev->new_raid_disk == rdev->raid_disk) 3421 continue; 3422 rdev->raid_disk = rdev->new_raid_disk; 3423 if (rdev->raid_disk < 0) 3424 clear_bit(In_sync, &rdev->flags); 3425 else { 3426 if (sysfs_link_rdev(mddev, rdev)) 3427 printk(KERN_WARNING "md: cannot register rd%d" 3428 " for %s after level change\n", 3429 rdev->raid_disk, mdname(mddev)); 3430 } 3431 } 3432 3433 module_put(mddev->pers->owner); 3434 mddev->pers = pers; 3435 mddev->private = priv; 3436 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 3437 mddev->level = mddev->new_level; 3438 mddev->layout = mddev->new_layout; 3439 mddev->chunk_sectors = mddev->new_chunk_sectors; 3440 mddev->delta_disks = 0; 3441 mddev->degraded = 0; 3442 if (mddev->pers->sync_request == NULL) { 3443 /* this is now an array without redundancy, so 3444 * it must always be in_sync 3445 */ 3446 mddev->in_sync = 1; 3447 del_timer_sync(&mddev->safemode_timer); 3448 } 3449 pers->run(mddev); 3450 mddev_resume(mddev); 3451 set_bit(MD_CHANGE_DEVS, &mddev->flags); 3452 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3453 md_wakeup_thread(mddev->thread); 3454 sysfs_notify(&mddev->kobj, NULL, "level"); 3455 md_new_event(mddev); 3456 return rv; 3457 } 3458 3459 static struct md_sysfs_entry md_level = 3460 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); 3461 3462 3463 static ssize_t 3464 layout_show(mddev_t *mddev, char *page) 3465 { 3466 /* just a number, not meaningful for all levels */ 3467 if (mddev->reshape_position != MaxSector && 3468 mddev->layout != mddev->new_layout) 3469 return sprintf(page, "%d (%d)\n", 3470 mddev->new_layout, mddev->layout); 3471 return sprintf(page, "%d\n", mddev->layout); 3472 } 3473 3474 static ssize_t 3475 layout_store(mddev_t *mddev, const char *buf, size_t len) 3476 { 3477 char *e; 3478 unsigned long n = simple_strtoul(buf, &e, 10); 3479 3480 if (!*buf || (*e && *e != '\n')) 3481 return -EINVAL; 3482 3483 if (mddev->pers) { 3484 int err; 3485 if (mddev->pers->check_reshape == NULL) 3486 return -EBUSY; 3487 mddev->new_layout = n; 3488 err = mddev->pers->check_reshape(mddev); 3489 if (err) { 3490 mddev->new_layout = mddev->layout; 3491 return err; 3492 } 3493 } else { 3494 mddev->new_layout = n; 3495 if (mddev->reshape_position == MaxSector) 3496 mddev->layout = n; 3497 } 3498 return len; 3499 } 3500 static struct md_sysfs_entry md_layout = 3501 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); 3502 3503 3504 static ssize_t 3505 raid_disks_show(mddev_t *mddev, char *page) 3506 { 3507 if (mddev->raid_disks == 0) 3508 return 0; 3509 if (mddev->reshape_position != MaxSector && 3510 mddev->delta_disks != 0) 3511 return sprintf(page, "%d (%d)\n", mddev->raid_disks, 3512 mddev->raid_disks - mddev->delta_disks); 3513 return sprintf(page, "%d\n", mddev->raid_disks); 3514 } 3515 3516 static int update_raid_disks(mddev_t *mddev, int raid_disks); 3517 3518 static ssize_t 3519 raid_disks_store(mddev_t *mddev, const char *buf, size_t len) 3520 { 3521 char *e; 3522 int rv = 0; 3523 unsigned long n = simple_strtoul(buf, &e, 10); 3524 3525 if (!*buf || (*e && *e != '\n')) 3526 return -EINVAL; 3527 3528 if (mddev->pers) 3529 rv = update_raid_disks(mddev, n); 3530 else if (mddev->reshape_position != MaxSector) { 3531 int olddisks = mddev->raid_disks - mddev->delta_disks; 3532 mddev->delta_disks = n - olddisks; 3533 mddev->raid_disks = n; 3534 } else 3535 mddev->raid_disks = n; 3536 return rv ? rv : len; 3537 } 3538 static struct md_sysfs_entry md_raid_disks = 3539 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); 3540 3541 static ssize_t 3542 chunk_size_show(mddev_t *mddev, char *page) 3543 { 3544 if (mddev->reshape_position != MaxSector && 3545 mddev->chunk_sectors != mddev->new_chunk_sectors) 3546 return sprintf(page, "%d (%d)\n", 3547 mddev->new_chunk_sectors << 9, 3548 mddev->chunk_sectors << 9); 3549 return sprintf(page, "%d\n", mddev->chunk_sectors << 9); 3550 } 3551 3552 static ssize_t 3553 chunk_size_store(mddev_t *mddev, const char *buf, size_t len) 3554 { 3555 char *e; 3556 unsigned long n = simple_strtoul(buf, &e, 10); 3557 3558 if (!*buf || (*e && *e != '\n')) 3559 return -EINVAL; 3560 3561 if (mddev->pers) { 3562 int err; 3563 if (mddev->pers->check_reshape == NULL) 3564 return -EBUSY; 3565 mddev->new_chunk_sectors = n >> 9; 3566 err = mddev->pers->check_reshape(mddev); 3567 if (err) { 3568 mddev->new_chunk_sectors = mddev->chunk_sectors; 3569 return err; 3570 } 3571 } else { 3572 mddev->new_chunk_sectors = n >> 9; 3573 if (mddev->reshape_position == MaxSector) 3574 mddev->chunk_sectors = n >> 9; 3575 } 3576 return len; 3577 } 3578 static struct md_sysfs_entry md_chunk_size = 3579 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); 3580 3581 static ssize_t 3582 resync_start_show(mddev_t *mddev, char *page) 3583 { 3584 if (mddev->recovery_cp == MaxSector) 3585 return sprintf(page, "none\n"); 3586 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); 3587 } 3588 3589 static ssize_t 3590 resync_start_store(mddev_t *mddev, const char *buf, size_t len) 3591 { 3592 char *e; 3593 unsigned long long n = simple_strtoull(buf, &e, 10); 3594 3595 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 3596 return -EBUSY; 3597 if (cmd_match(buf, "none")) 3598 n = MaxSector; 3599 else if (!*buf || (*e && *e != '\n')) 3600 return -EINVAL; 3601 3602 mddev->recovery_cp = n; 3603 return len; 3604 } 3605 static struct md_sysfs_entry md_resync_start = 3606 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); 3607 3608 /* 3609 * The array state can be: 3610 * 3611 * clear 3612 * No devices, no size, no level 3613 * Equivalent to STOP_ARRAY ioctl 3614 * inactive 3615 * May have some settings, but array is not active 3616 * all IO results in error 3617 * When written, doesn't tear down array, but just stops it 3618 * suspended (not supported yet) 3619 * All IO requests will block. The array can be reconfigured. 3620 * Writing this, if accepted, will block until array is quiescent 3621 * readonly 3622 * no resync can happen. no superblocks get written. 3623 * write requests fail 3624 * read-auto 3625 * like readonly, but behaves like 'clean' on a write request. 3626 * 3627 * clean - no pending writes, but otherwise active. 3628 * When written to inactive array, starts without resync 3629 * If a write request arrives then 3630 * if metadata is known, mark 'dirty' and switch to 'active'. 3631 * if not known, block and switch to write-pending 3632 * If written to an active array that has pending writes, then fails. 3633 * active 3634 * fully active: IO and resync can be happening. 3635 * When written to inactive array, starts with resync 3636 * 3637 * write-pending 3638 * clean, but writes are blocked waiting for 'active' to be written. 3639 * 3640 * active-idle 3641 * like active, but no writes have been seen for a while (100msec). 3642 * 3643 */ 3644 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, 3645 write_pending, active_idle, bad_word}; 3646 static char *array_states[] = { 3647 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", 3648 "write-pending", "active-idle", NULL }; 3649 3650 static int match_word(const char *word, char **list) 3651 { 3652 int n; 3653 for (n=0; list[n]; n++) 3654 if (cmd_match(word, list[n])) 3655 break; 3656 return n; 3657 } 3658 3659 static ssize_t 3660 array_state_show(mddev_t *mddev, char *page) 3661 { 3662 enum array_state st = inactive; 3663 3664 if (mddev->pers) 3665 switch(mddev->ro) { 3666 case 1: 3667 st = readonly; 3668 break; 3669 case 2: 3670 st = read_auto; 3671 break; 3672 case 0: 3673 if (mddev->in_sync) 3674 st = clean; 3675 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 3676 st = write_pending; 3677 else if (mddev->safemode) 3678 st = active_idle; 3679 else 3680 st = active; 3681 } 3682 else { 3683 if (list_empty(&mddev->disks) && 3684 mddev->raid_disks == 0 && 3685 mddev->dev_sectors == 0) 3686 st = clear; 3687 else 3688 st = inactive; 3689 } 3690 return sprintf(page, "%s\n", array_states[st]); 3691 } 3692 3693 static int do_md_stop(mddev_t * mddev, int ro, int is_open); 3694 static int md_set_readonly(mddev_t * mddev, int is_open); 3695 static int do_md_run(mddev_t * mddev); 3696 static int restart_array(mddev_t *mddev); 3697 3698 static ssize_t 3699 array_state_store(mddev_t *mddev, const char *buf, size_t len) 3700 { 3701 int err = -EINVAL; 3702 enum array_state st = match_word(buf, array_states); 3703 switch(st) { 3704 case bad_word: 3705 break; 3706 case clear: 3707 /* stopping an active array */ 3708 if (atomic_read(&mddev->openers) > 0) 3709 return -EBUSY; 3710 err = do_md_stop(mddev, 0, 0); 3711 break; 3712 case inactive: 3713 /* stopping an active array */ 3714 if (mddev->pers) { 3715 if (atomic_read(&mddev->openers) > 0) 3716 return -EBUSY; 3717 err = do_md_stop(mddev, 2, 0); 3718 } else 3719 err = 0; /* already inactive */ 3720 break; 3721 case suspended: 3722 break; /* not supported yet */ 3723 case readonly: 3724 if (mddev->pers) 3725 err = md_set_readonly(mddev, 0); 3726 else { 3727 mddev->ro = 1; 3728 set_disk_ro(mddev->gendisk, 1); 3729 err = do_md_run(mddev); 3730 } 3731 break; 3732 case read_auto: 3733 if (mddev->pers) { 3734 if (mddev->ro == 0) 3735 err = md_set_readonly(mddev, 0); 3736 else if (mddev->ro == 1) 3737 err = restart_array(mddev); 3738 if (err == 0) { 3739 mddev->ro = 2; 3740 set_disk_ro(mddev->gendisk, 0); 3741 } 3742 } else { 3743 mddev->ro = 2; 3744 err = do_md_run(mddev); 3745 } 3746 break; 3747 case clean: 3748 if (mddev->pers) { 3749 restart_array(mddev); 3750 spin_lock_irq(&mddev->write_lock); 3751 if (atomic_read(&mddev->writes_pending) == 0) { 3752 if (mddev->in_sync == 0) { 3753 mddev->in_sync = 1; 3754 if (mddev->safemode == 1) 3755 mddev->safemode = 0; 3756 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 3757 } 3758 err = 0; 3759 } else 3760 err = -EBUSY; 3761 spin_unlock_irq(&mddev->write_lock); 3762 } else 3763 err = -EINVAL; 3764 break; 3765 case active: 3766 if (mddev->pers) { 3767 restart_array(mddev); 3768 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 3769 wake_up(&mddev->sb_wait); 3770 err = 0; 3771 } else { 3772 mddev->ro = 0; 3773 set_disk_ro(mddev->gendisk, 0); 3774 err = do_md_run(mddev); 3775 } 3776 break; 3777 case write_pending: 3778 case active_idle: 3779 /* these cannot be set */ 3780 break; 3781 } 3782 if (err) 3783 return err; 3784 else { 3785 sysfs_notify_dirent_safe(mddev->sysfs_state); 3786 return len; 3787 } 3788 } 3789 static struct md_sysfs_entry md_array_state = 3790 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); 3791 3792 static ssize_t 3793 max_corrected_read_errors_show(mddev_t *mddev, char *page) { 3794 return sprintf(page, "%d\n", 3795 atomic_read(&mddev->max_corr_read_errors)); 3796 } 3797 3798 static ssize_t 3799 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) 3800 { 3801 char *e; 3802 unsigned long n = simple_strtoul(buf, &e, 10); 3803 3804 if (*buf && (*e == 0 || *e == '\n')) { 3805 atomic_set(&mddev->max_corr_read_errors, n); 3806 return len; 3807 } 3808 return -EINVAL; 3809 } 3810 3811 static struct md_sysfs_entry max_corr_read_errors = 3812 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, 3813 max_corrected_read_errors_store); 3814 3815 static ssize_t 3816 null_show(mddev_t *mddev, char *page) 3817 { 3818 return -EINVAL; 3819 } 3820 3821 static ssize_t 3822 new_dev_store(mddev_t *mddev, const char *buf, size_t len) 3823 { 3824 /* buf must be %d:%d\n? giving major and minor numbers */ 3825 /* The new device is added to the array. 3826 * If the array has a persistent superblock, we read the 3827 * superblock to initialise info and check validity. 3828 * Otherwise, only checking done is that in bind_rdev_to_array, 3829 * which mainly checks size. 3830 */ 3831 char *e; 3832 int major = simple_strtoul(buf, &e, 10); 3833 int minor; 3834 dev_t dev; 3835 mdk_rdev_t *rdev; 3836 int err; 3837 3838 if (!*buf || *e != ':' || !e[1] || e[1] == '\n') 3839 return -EINVAL; 3840 minor = simple_strtoul(e+1, &e, 10); 3841 if (*e && *e != '\n') 3842 return -EINVAL; 3843 dev = MKDEV(major, minor); 3844 if (major != MAJOR(dev) || 3845 minor != MINOR(dev)) 3846 return -EOVERFLOW; 3847 3848 3849 if (mddev->persistent) { 3850 rdev = md_import_device(dev, mddev->major_version, 3851 mddev->minor_version); 3852 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { 3853 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 3854 mdk_rdev_t, same_set); 3855 err = super_types[mddev->major_version] 3856 .load_super(rdev, rdev0, mddev->minor_version); 3857 if (err < 0) 3858 goto out; 3859 } 3860 } else if (mddev->external) 3861 rdev = md_import_device(dev, -2, -1); 3862 else 3863 rdev = md_import_device(dev, -1, -1); 3864 3865 if (IS_ERR(rdev)) 3866 return PTR_ERR(rdev); 3867 err = bind_rdev_to_array(rdev, mddev); 3868 out: 3869 if (err) 3870 export_rdev(rdev); 3871 return err ? err : len; 3872 } 3873 3874 static struct md_sysfs_entry md_new_device = 3875 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); 3876 3877 static ssize_t 3878 bitmap_store(mddev_t *mddev, const char *buf, size_t len) 3879 { 3880 char *end; 3881 unsigned long chunk, end_chunk; 3882 3883 if (!mddev->bitmap) 3884 goto out; 3885 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ 3886 while (*buf) { 3887 chunk = end_chunk = simple_strtoul(buf, &end, 0); 3888 if (buf == end) break; 3889 if (*end == '-') { /* range */ 3890 buf = end + 1; 3891 end_chunk = simple_strtoul(buf, &end, 0); 3892 if (buf == end) break; 3893 } 3894 if (*end && !isspace(*end)) break; 3895 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); 3896 buf = skip_spaces(end); 3897 } 3898 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ 3899 out: 3900 return len; 3901 } 3902 3903 static struct md_sysfs_entry md_bitmap = 3904 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); 3905 3906 static ssize_t 3907 size_show(mddev_t *mddev, char *page) 3908 { 3909 return sprintf(page, "%llu\n", 3910 (unsigned long long)mddev->dev_sectors / 2); 3911 } 3912 3913 static int update_size(mddev_t *mddev, sector_t num_sectors); 3914 3915 static ssize_t 3916 size_store(mddev_t *mddev, const char *buf, size_t len) 3917 { 3918 /* If array is inactive, we can reduce the component size, but 3919 * not increase it (except from 0). 3920 * If array is active, we can try an on-line resize 3921 */ 3922 sector_t sectors; 3923 int err = strict_blocks_to_sectors(buf, §ors); 3924 3925 if (err < 0) 3926 return err; 3927 if (mddev->pers) { 3928 err = update_size(mddev, sectors); 3929 md_update_sb(mddev, 1); 3930 } else { 3931 if (mddev->dev_sectors == 0 || 3932 mddev->dev_sectors > sectors) 3933 mddev->dev_sectors = sectors; 3934 else 3935 err = -ENOSPC; 3936 } 3937 return err ? err : len; 3938 } 3939 3940 static struct md_sysfs_entry md_size = 3941 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); 3942 3943 3944 /* Metdata version. 3945 * This is one of 3946 * 'none' for arrays with no metadata (good luck...) 3947 * 'external' for arrays with externally managed metadata, 3948 * or N.M for internally known formats 3949 */ 3950 static ssize_t 3951 metadata_show(mddev_t *mddev, char *page) 3952 { 3953 if (mddev->persistent) 3954 return sprintf(page, "%d.%d\n", 3955 mddev->major_version, mddev->minor_version); 3956 else if (mddev->external) 3957 return sprintf(page, "external:%s\n", mddev->metadata_type); 3958 else 3959 return sprintf(page, "none\n"); 3960 } 3961 3962 static ssize_t 3963 metadata_store(mddev_t *mddev, const char *buf, size_t len) 3964 { 3965 int major, minor; 3966 char *e; 3967 /* Changing the details of 'external' metadata is 3968 * always permitted. Otherwise there must be 3969 * no devices attached to the array. 3970 */ 3971 if (mddev->external && strncmp(buf, "external:", 9) == 0) 3972 ; 3973 else if (!list_empty(&mddev->disks)) 3974 return -EBUSY; 3975 3976 if (cmd_match(buf, "none")) { 3977 mddev->persistent = 0; 3978 mddev->external = 0; 3979 mddev->major_version = 0; 3980 mddev->minor_version = 90; 3981 return len; 3982 } 3983 if (strncmp(buf, "external:", 9) == 0) { 3984 size_t namelen = len-9; 3985 if (namelen >= sizeof(mddev->metadata_type)) 3986 namelen = sizeof(mddev->metadata_type)-1; 3987 strncpy(mddev->metadata_type, buf+9, namelen); 3988 mddev->metadata_type[namelen] = 0; 3989 if (namelen && mddev->metadata_type[namelen-1] == '\n') 3990 mddev->metadata_type[--namelen] = 0; 3991 mddev->persistent = 0; 3992 mddev->external = 1; 3993 mddev->major_version = 0; 3994 mddev->minor_version = 90; 3995 return len; 3996 } 3997 major = simple_strtoul(buf, &e, 10); 3998 if (e==buf || *e != '.') 3999 return -EINVAL; 4000 buf = e+1; 4001 minor = simple_strtoul(buf, &e, 10); 4002 if (e==buf || (*e && *e != '\n') ) 4003 return -EINVAL; 4004 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) 4005 return -ENOENT; 4006 mddev->major_version = major; 4007 mddev->minor_version = minor; 4008 mddev->persistent = 1; 4009 mddev->external = 0; 4010 return len; 4011 } 4012 4013 static struct md_sysfs_entry md_metadata = 4014 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 4015 4016 static ssize_t 4017 action_show(mddev_t *mddev, char *page) 4018 { 4019 char *type = "idle"; 4020 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 4021 type = "frozen"; 4022 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 4023 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { 4024 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 4025 type = "reshape"; 4026 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 4027 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 4028 type = "resync"; 4029 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 4030 type = "check"; 4031 else 4032 type = "repair"; 4033 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) 4034 type = "recover"; 4035 } 4036 return sprintf(page, "%s\n", type); 4037 } 4038 4039 static void reap_sync_thread(mddev_t *mddev); 4040 4041 static ssize_t 4042 action_store(mddev_t *mddev, const char *page, size_t len) 4043 { 4044 if (!mddev->pers || !mddev->pers->sync_request) 4045 return -EINVAL; 4046 4047 if (cmd_match(page, "frozen")) 4048 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4049 else 4050 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4051 4052 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { 4053 if (mddev->sync_thread) { 4054 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4055 reap_sync_thread(mddev); 4056 } 4057 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 4058 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 4059 return -EBUSY; 4060 else if (cmd_match(page, "resync")) 4061 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4062 else if (cmd_match(page, "recover")) { 4063 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 4064 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4065 } else if (cmd_match(page, "reshape")) { 4066 int err; 4067 if (mddev->pers->start_reshape == NULL) 4068 return -EINVAL; 4069 err = mddev->pers->start_reshape(mddev); 4070 if (err) 4071 return err; 4072 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4073 } else { 4074 if (cmd_match(page, "check")) 4075 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 4076 else if (!cmd_match(page, "repair")) 4077 return -EINVAL; 4078 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 4079 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 4080 } 4081 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4082 md_wakeup_thread(mddev->thread); 4083 sysfs_notify_dirent_safe(mddev->sysfs_action); 4084 return len; 4085 } 4086 4087 static ssize_t 4088 mismatch_cnt_show(mddev_t *mddev, char *page) 4089 { 4090 return sprintf(page, "%llu\n", 4091 (unsigned long long) mddev->resync_mismatches); 4092 } 4093 4094 static struct md_sysfs_entry md_scan_mode = 4095 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); 4096 4097 4098 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); 4099 4100 static ssize_t 4101 sync_min_show(mddev_t *mddev, char *page) 4102 { 4103 return sprintf(page, "%d (%s)\n", speed_min(mddev), 4104 mddev->sync_speed_min ? "local": "system"); 4105 } 4106 4107 static ssize_t 4108 sync_min_store(mddev_t *mddev, const char *buf, size_t len) 4109 { 4110 int min; 4111 char *e; 4112 if (strncmp(buf, "system", 6)==0) { 4113 mddev->sync_speed_min = 0; 4114 return len; 4115 } 4116 min = simple_strtoul(buf, &e, 10); 4117 if (buf == e || (*e && *e != '\n') || min <= 0) 4118 return -EINVAL; 4119 mddev->sync_speed_min = min; 4120 return len; 4121 } 4122 4123 static struct md_sysfs_entry md_sync_min = 4124 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); 4125 4126 static ssize_t 4127 sync_max_show(mddev_t *mddev, char *page) 4128 { 4129 return sprintf(page, "%d (%s)\n", speed_max(mddev), 4130 mddev->sync_speed_max ? "local": "system"); 4131 } 4132 4133 static ssize_t 4134 sync_max_store(mddev_t *mddev, const char *buf, size_t len) 4135 { 4136 int max; 4137 char *e; 4138 if (strncmp(buf, "system", 6)==0) { 4139 mddev->sync_speed_max = 0; 4140 return len; 4141 } 4142 max = simple_strtoul(buf, &e, 10); 4143 if (buf == e || (*e && *e != '\n') || max <= 0) 4144 return -EINVAL; 4145 mddev->sync_speed_max = max; 4146 return len; 4147 } 4148 4149 static struct md_sysfs_entry md_sync_max = 4150 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); 4151 4152 static ssize_t 4153 degraded_show(mddev_t *mddev, char *page) 4154 { 4155 return sprintf(page, "%d\n", mddev->degraded); 4156 } 4157 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); 4158 4159 static ssize_t 4160 sync_force_parallel_show(mddev_t *mddev, char *page) 4161 { 4162 return sprintf(page, "%d\n", mddev->parallel_resync); 4163 } 4164 4165 static ssize_t 4166 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) 4167 { 4168 long n; 4169 4170 if (strict_strtol(buf, 10, &n)) 4171 return -EINVAL; 4172 4173 if (n != 0 && n != 1) 4174 return -EINVAL; 4175 4176 mddev->parallel_resync = n; 4177 4178 if (mddev->sync_thread) 4179 wake_up(&resync_wait); 4180 4181 return len; 4182 } 4183 4184 /* force parallel resync, even with shared block devices */ 4185 static struct md_sysfs_entry md_sync_force_parallel = 4186 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, 4187 sync_force_parallel_show, sync_force_parallel_store); 4188 4189 static ssize_t 4190 sync_speed_show(mddev_t *mddev, char *page) 4191 { 4192 unsigned long resync, dt, db; 4193 if (mddev->curr_resync == 0) 4194 return sprintf(page, "none\n"); 4195 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); 4196 dt = (jiffies - mddev->resync_mark) / HZ; 4197 if (!dt) dt++; 4198 db = resync - mddev->resync_mark_cnt; 4199 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ 4200 } 4201 4202 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); 4203 4204 static ssize_t 4205 sync_completed_show(mddev_t *mddev, char *page) 4206 { 4207 unsigned long long max_sectors, resync; 4208 4209 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4210 return sprintf(page, "none\n"); 4211 4212 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 4213 max_sectors = mddev->resync_max_sectors; 4214 else 4215 max_sectors = mddev->dev_sectors; 4216 4217 resync = mddev->curr_resync_completed; 4218 return sprintf(page, "%llu / %llu\n", resync, max_sectors); 4219 } 4220 4221 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); 4222 4223 static ssize_t 4224 min_sync_show(mddev_t *mddev, char *page) 4225 { 4226 return sprintf(page, "%llu\n", 4227 (unsigned long long)mddev->resync_min); 4228 } 4229 static ssize_t 4230 min_sync_store(mddev_t *mddev, const char *buf, size_t len) 4231 { 4232 unsigned long long min; 4233 if (strict_strtoull(buf, 10, &min)) 4234 return -EINVAL; 4235 if (min > mddev->resync_max) 4236 return -EINVAL; 4237 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4238 return -EBUSY; 4239 4240 /* Must be a multiple of chunk_size */ 4241 if (mddev->chunk_sectors) { 4242 sector_t temp = min; 4243 if (sector_div(temp, mddev->chunk_sectors)) 4244 return -EINVAL; 4245 } 4246 mddev->resync_min = min; 4247 4248 return len; 4249 } 4250 4251 static struct md_sysfs_entry md_min_sync = 4252 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); 4253 4254 static ssize_t 4255 max_sync_show(mddev_t *mddev, char *page) 4256 { 4257 if (mddev->resync_max == MaxSector) 4258 return sprintf(page, "max\n"); 4259 else 4260 return sprintf(page, "%llu\n", 4261 (unsigned long long)mddev->resync_max); 4262 } 4263 static ssize_t 4264 max_sync_store(mddev_t *mddev, const char *buf, size_t len) 4265 { 4266 if (strncmp(buf, "max", 3) == 0) 4267 mddev->resync_max = MaxSector; 4268 else { 4269 unsigned long long max; 4270 if (strict_strtoull(buf, 10, &max)) 4271 return -EINVAL; 4272 if (max < mddev->resync_min) 4273 return -EINVAL; 4274 if (max < mddev->resync_max && 4275 mddev->ro == 0 && 4276 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 4277 return -EBUSY; 4278 4279 /* Must be a multiple of chunk_size */ 4280 if (mddev->chunk_sectors) { 4281 sector_t temp = max; 4282 if (sector_div(temp, mddev->chunk_sectors)) 4283 return -EINVAL; 4284 } 4285 mddev->resync_max = max; 4286 } 4287 wake_up(&mddev->recovery_wait); 4288 return len; 4289 } 4290 4291 static struct md_sysfs_entry md_max_sync = 4292 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); 4293 4294 static ssize_t 4295 suspend_lo_show(mddev_t *mddev, char *page) 4296 { 4297 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); 4298 } 4299 4300 static ssize_t 4301 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) 4302 { 4303 char *e; 4304 unsigned long long new = simple_strtoull(buf, &e, 10); 4305 unsigned long long old = mddev->suspend_lo; 4306 4307 if (mddev->pers == NULL || 4308 mddev->pers->quiesce == NULL) 4309 return -EINVAL; 4310 if (buf == e || (*e && *e != '\n')) 4311 return -EINVAL; 4312 4313 mddev->suspend_lo = new; 4314 if (new >= old) 4315 /* Shrinking suspended region */ 4316 mddev->pers->quiesce(mddev, 2); 4317 else { 4318 /* Expanding suspended region - need to wait */ 4319 mddev->pers->quiesce(mddev, 1); 4320 mddev->pers->quiesce(mddev, 0); 4321 } 4322 return len; 4323 } 4324 static struct md_sysfs_entry md_suspend_lo = 4325 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); 4326 4327 4328 static ssize_t 4329 suspend_hi_show(mddev_t *mddev, char *page) 4330 { 4331 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); 4332 } 4333 4334 static ssize_t 4335 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) 4336 { 4337 char *e; 4338 unsigned long long new = simple_strtoull(buf, &e, 10); 4339 unsigned long long old = mddev->suspend_hi; 4340 4341 if (mddev->pers == NULL || 4342 mddev->pers->quiesce == NULL) 4343 return -EINVAL; 4344 if (buf == e || (*e && *e != '\n')) 4345 return -EINVAL; 4346 4347 mddev->suspend_hi = new; 4348 if (new <= old) 4349 /* Shrinking suspended region */ 4350 mddev->pers->quiesce(mddev, 2); 4351 else { 4352 /* Expanding suspended region - need to wait */ 4353 mddev->pers->quiesce(mddev, 1); 4354 mddev->pers->quiesce(mddev, 0); 4355 } 4356 return len; 4357 } 4358 static struct md_sysfs_entry md_suspend_hi = 4359 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); 4360 4361 static ssize_t 4362 reshape_position_show(mddev_t *mddev, char *page) 4363 { 4364 if (mddev->reshape_position != MaxSector) 4365 return sprintf(page, "%llu\n", 4366 (unsigned long long)mddev->reshape_position); 4367 strcpy(page, "none\n"); 4368 return 5; 4369 } 4370 4371 static ssize_t 4372 reshape_position_store(mddev_t *mddev, const char *buf, size_t len) 4373 { 4374 char *e; 4375 unsigned long long new = simple_strtoull(buf, &e, 10); 4376 if (mddev->pers) 4377 return -EBUSY; 4378 if (buf == e || (*e && *e != '\n')) 4379 return -EINVAL; 4380 mddev->reshape_position = new; 4381 mddev->delta_disks = 0; 4382 mddev->new_level = mddev->level; 4383 mddev->new_layout = mddev->layout; 4384 mddev->new_chunk_sectors = mddev->chunk_sectors; 4385 return len; 4386 } 4387 4388 static struct md_sysfs_entry md_reshape_position = 4389 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, 4390 reshape_position_store); 4391 4392 static ssize_t 4393 array_size_show(mddev_t *mddev, char *page) 4394 { 4395 if (mddev->external_size) 4396 return sprintf(page, "%llu\n", 4397 (unsigned long long)mddev->array_sectors/2); 4398 else 4399 return sprintf(page, "default\n"); 4400 } 4401 4402 static ssize_t 4403 array_size_store(mddev_t *mddev, const char *buf, size_t len) 4404 { 4405 sector_t sectors; 4406 4407 if (strncmp(buf, "default", 7) == 0) { 4408 if (mddev->pers) 4409 sectors = mddev->pers->size(mddev, 0, 0); 4410 else 4411 sectors = mddev->array_sectors; 4412 4413 mddev->external_size = 0; 4414 } else { 4415 if (strict_blocks_to_sectors(buf, §ors) < 0) 4416 return -EINVAL; 4417 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) 4418 return -E2BIG; 4419 4420 mddev->external_size = 1; 4421 } 4422 4423 mddev->array_sectors = sectors; 4424 if (mddev->pers) { 4425 set_capacity(mddev->gendisk, mddev->array_sectors); 4426 revalidate_disk(mddev->gendisk); 4427 } 4428 return len; 4429 } 4430 4431 static struct md_sysfs_entry md_array_size = 4432 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, 4433 array_size_store); 4434 4435 static struct attribute *md_default_attrs[] = { 4436 &md_level.attr, 4437 &md_layout.attr, 4438 &md_raid_disks.attr, 4439 &md_chunk_size.attr, 4440 &md_size.attr, 4441 &md_resync_start.attr, 4442 &md_metadata.attr, 4443 &md_new_device.attr, 4444 &md_safe_delay.attr, 4445 &md_array_state.attr, 4446 &md_reshape_position.attr, 4447 &md_array_size.attr, 4448 &max_corr_read_errors.attr, 4449 NULL, 4450 }; 4451 4452 static struct attribute *md_redundancy_attrs[] = { 4453 &md_scan_mode.attr, 4454 &md_mismatches.attr, 4455 &md_sync_min.attr, 4456 &md_sync_max.attr, 4457 &md_sync_speed.attr, 4458 &md_sync_force_parallel.attr, 4459 &md_sync_completed.attr, 4460 &md_min_sync.attr, 4461 &md_max_sync.attr, 4462 &md_suspend_lo.attr, 4463 &md_suspend_hi.attr, 4464 &md_bitmap.attr, 4465 &md_degraded.attr, 4466 NULL, 4467 }; 4468 static struct attribute_group md_redundancy_group = { 4469 .name = NULL, 4470 .attrs = md_redundancy_attrs, 4471 }; 4472 4473 4474 static ssize_t 4475 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 4476 { 4477 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4478 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 4479 ssize_t rv; 4480 4481 if (!entry->show) 4482 return -EIO; 4483 rv = mddev_lock(mddev); 4484 if (!rv) { 4485 rv = entry->show(mddev, page); 4486 mddev_unlock(mddev); 4487 } 4488 return rv; 4489 } 4490 4491 static ssize_t 4492 md_attr_store(struct kobject *kobj, struct attribute *attr, 4493 const char *page, size_t length) 4494 { 4495 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4496 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 4497 ssize_t rv; 4498 4499 if (!entry->store) 4500 return -EIO; 4501 if (!capable(CAP_SYS_ADMIN)) 4502 return -EACCES; 4503 rv = mddev_lock(mddev); 4504 if (mddev->hold_active == UNTIL_IOCTL) 4505 mddev->hold_active = 0; 4506 if (!rv) { 4507 rv = entry->store(mddev, page, length); 4508 mddev_unlock(mddev); 4509 } 4510 return rv; 4511 } 4512 4513 static void md_free(struct kobject *ko) 4514 { 4515 mddev_t *mddev = container_of(ko, mddev_t, kobj); 4516 4517 if (mddev->sysfs_state) 4518 sysfs_put(mddev->sysfs_state); 4519 4520 if (mddev->gendisk) { 4521 del_gendisk(mddev->gendisk); 4522 put_disk(mddev->gendisk); 4523 } 4524 if (mddev->queue) 4525 blk_cleanup_queue(mddev->queue); 4526 4527 kfree(mddev); 4528 } 4529 4530 static const struct sysfs_ops md_sysfs_ops = { 4531 .show = md_attr_show, 4532 .store = md_attr_store, 4533 }; 4534 static struct kobj_type md_ktype = { 4535 .release = md_free, 4536 .sysfs_ops = &md_sysfs_ops, 4537 .default_attrs = md_default_attrs, 4538 }; 4539 4540 int mdp_major = 0; 4541 4542 static void mddev_delayed_delete(struct work_struct *ws) 4543 { 4544 mddev_t *mddev = container_of(ws, mddev_t, del_work); 4545 4546 sysfs_remove_group(&mddev->kobj, &md_bitmap_group); 4547 kobject_del(&mddev->kobj); 4548 kobject_put(&mddev->kobj); 4549 } 4550 4551 static int md_alloc(dev_t dev, char *name) 4552 { 4553 static DEFINE_MUTEX(disks_mutex); 4554 mddev_t *mddev = mddev_find(dev); 4555 struct gendisk *disk; 4556 int partitioned; 4557 int shift; 4558 int unit; 4559 int error; 4560 4561 if (!mddev) 4562 return -ENODEV; 4563 4564 partitioned = (MAJOR(mddev->unit) != MD_MAJOR); 4565 shift = partitioned ? MdpMinorShift : 0; 4566 unit = MINOR(mddev->unit) >> shift; 4567 4568 /* wait for any previous instance of this device to be 4569 * completely removed (mddev_delayed_delete). 4570 */ 4571 flush_workqueue(md_misc_wq); 4572 4573 mutex_lock(&disks_mutex); 4574 error = -EEXIST; 4575 if (mddev->gendisk) 4576 goto abort; 4577 4578 if (name) { 4579 /* Need to ensure that 'name' is not a duplicate. 4580 */ 4581 mddev_t *mddev2; 4582 spin_lock(&all_mddevs_lock); 4583 4584 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) 4585 if (mddev2->gendisk && 4586 strcmp(mddev2->gendisk->disk_name, name) == 0) { 4587 spin_unlock(&all_mddevs_lock); 4588 goto abort; 4589 } 4590 spin_unlock(&all_mddevs_lock); 4591 } 4592 4593 error = -ENOMEM; 4594 mddev->queue = blk_alloc_queue(GFP_KERNEL); 4595 if (!mddev->queue) 4596 goto abort; 4597 mddev->queue->queuedata = mddev; 4598 4599 blk_queue_make_request(mddev->queue, md_make_request); 4600 4601 disk = alloc_disk(1 << shift); 4602 if (!disk) { 4603 blk_cleanup_queue(mddev->queue); 4604 mddev->queue = NULL; 4605 goto abort; 4606 } 4607 disk->major = MAJOR(mddev->unit); 4608 disk->first_minor = unit << shift; 4609 if (name) 4610 strcpy(disk->disk_name, name); 4611 else if (partitioned) 4612 sprintf(disk->disk_name, "md_d%d", unit); 4613 else 4614 sprintf(disk->disk_name, "md%d", unit); 4615 disk->fops = &md_fops; 4616 disk->private_data = mddev; 4617 disk->queue = mddev->queue; 4618 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); 4619 /* Allow extended partitions. This makes the 4620 * 'mdp' device redundant, but we can't really 4621 * remove it now. 4622 */ 4623 disk->flags |= GENHD_FL_EXT_DEVT; 4624 mddev->gendisk = disk; 4625 /* As soon as we call add_disk(), another thread could get 4626 * through to md_open, so make sure it doesn't get too far 4627 */ 4628 mutex_lock(&mddev->open_mutex); 4629 add_disk(disk); 4630 4631 error = kobject_init_and_add(&mddev->kobj, &md_ktype, 4632 &disk_to_dev(disk)->kobj, "%s", "md"); 4633 if (error) { 4634 /* This isn't possible, but as kobject_init_and_add is marked 4635 * __must_check, we must do something with the result 4636 */ 4637 printk(KERN_WARNING "md: cannot register %s/md - name in use\n", 4638 disk->disk_name); 4639 error = 0; 4640 } 4641 if (mddev->kobj.sd && 4642 sysfs_create_group(&mddev->kobj, &md_bitmap_group)) 4643 printk(KERN_DEBUG "pointless warning\n"); 4644 mutex_unlock(&mddev->open_mutex); 4645 abort: 4646 mutex_unlock(&disks_mutex); 4647 if (!error && mddev->kobj.sd) { 4648 kobject_uevent(&mddev->kobj, KOBJ_ADD); 4649 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); 4650 } 4651 mddev_put(mddev); 4652 return error; 4653 } 4654 4655 static struct kobject *md_probe(dev_t dev, int *part, void *data) 4656 { 4657 md_alloc(dev, NULL); 4658 return NULL; 4659 } 4660 4661 static int add_named_array(const char *val, struct kernel_param *kp) 4662 { 4663 /* val must be "md_*" where * is not all digits. 4664 * We allocate an array with a large free minor number, and 4665 * set the name to val. val must not already be an active name. 4666 */ 4667 int len = strlen(val); 4668 char buf[DISK_NAME_LEN]; 4669 4670 while (len && val[len-1] == '\n') 4671 len--; 4672 if (len >= DISK_NAME_LEN) 4673 return -E2BIG; 4674 strlcpy(buf, val, len+1); 4675 if (strncmp(buf, "md_", 3) != 0) 4676 return -EINVAL; 4677 return md_alloc(0, buf); 4678 } 4679 4680 static void md_safemode_timeout(unsigned long data) 4681 { 4682 mddev_t *mddev = (mddev_t *) data; 4683 4684 if (!atomic_read(&mddev->writes_pending)) { 4685 mddev->safemode = 1; 4686 if (mddev->external) 4687 sysfs_notify_dirent_safe(mddev->sysfs_state); 4688 } 4689 md_wakeup_thread(mddev->thread); 4690 } 4691 4692 static int start_dirty_degraded; 4693 4694 int md_run(mddev_t *mddev) 4695 { 4696 int err; 4697 mdk_rdev_t *rdev; 4698 struct mdk_personality *pers; 4699 4700 if (list_empty(&mddev->disks)) 4701 /* cannot run an array with no devices.. */ 4702 return -EINVAL; 4703 4704 if (mddev->pers) 4705 return -EBUSY; 4706 /* Cannot run until previous stop completes properly */ 4707 if (mddev->sysfs_active) 4708 return -EBUSY; 4709 4710 /* 4711 * Analyze all RAID superblock(s) 4712 */ 4713 if (!mddev->raid_disks) { 4714 if (!mddev->persistent) 4715 return -EINVAL; 4716 analyze_sbs(mddev); 4717 } 4718 4719 if (mddev->level != LEVEL_NONE) 4720 request_module("md-level-%d", mddev->level); 4721 else if (mddev->clevel[0]) 4722 request_module("md-%s", mddev->clevel); 4723 4724 /* 4725 * Drop all container device buffers, from now on 4726 * the only valid external interface is through the md 4727 * device. 4728 */ 4729 list_for_each_entry(rdev, &mddev->disks, same_set) { 4730 if (test_bit(Faulty, &rdev->flags)) 4731 continue; 4732 sync_blockdev(rdev->bdev); 4733 invalidate_bdev(rdev->bdev); 4734 4735 /* perform some consistency tests on the device. 4736 * We don't want the data to overlap the metadata, 4737 * Internal Bitmap issues have been handled elsewhere. 4738 */ 4739 if (rdev->meta_bdev) { 4740 /* Nothing to check */; 4741 } else if (rdev->data_offset < rdev->sb_start) { 4742 if (mddev->dev_sectors && 4743 rdev->data_offset + mddev->dev_sectors 4744 > rdev->sb_start) { 4745 printk("md: %s: data overlaps metadata\n", 4746 mdname(mddev)); 4747 return -EINVAL; 4748 } 4749 } else { 4750 if (rdev->sb_start + rdev->sb_size/512 4751 > rdev->data_offset) { 4752 printk("md: %s: metadata overlaps data\n", 4753 mdname(mddev)); 4754 return -EINVAL; 4755 } 4756 } 4757 sysfs_notify_dirent_safe(rdev->sysfs_state); 4758 } 4759 4760 if (mddev->bio_set == NULL) 4761 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 4762 sizeof(mddev_t *)); 4763 4764 spin_lock(&pers_lock); 4765 pers = find_pers(mddev->level, mddev->clevel); 4766 if (!pers || !try_module_get(pers->owner)) { 4767 spin_unlock(&pers_lock); 4768 if (mddev->level != LEVEL_NONE) 4769 printk(KERN_WARNING "md: personality for level %d is not loaded!\n", 4770 mddev->level); 4771 else 4772 printk(KERN_WARNING "md: personality for level %s is not loaded!\n", 4773 mddev->clevel); 4774 return -EINVAL; 4775 } 4776 mddev->pers = pers; 4777 spin_unlock(&pers_lock); 4778 if (mddev->level != pers->level) { 4779 mddev->level = pers->level; 4780 mddev->new_level = pers->level; 4781 } 4782 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 4783 4784 if (mddev->reshape_position != MaxSector && 4785 pers->start_reshape == NULL) { 4786 /* This personality cannot handle reshaping... */ 4787 mddev->pers = NULL; 4788 module_put(pers->owner); 4789 return -EINVAL; 4790 } 4791 4792 if (pers->sync_request) { 4793 /* Warn if this is a potentially silly 4794 * configuration. 4795 */ 4796 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 4797 mdk_rdev_t *rdev2; 4798 int warned = 0; 4799 4800 list_for_each_entry(rdev, &mddev->disks, same_set) 4801 list_for_each_entry(rdev2, &mddev->disks, same_set) { 4802 if (rdev < rdev2 && 4803 rdev->bdev->bd_contains == 4804 rdev2->bdev->bd_contains) { 4805 printk(KERN_WARNING 4806 "%s: WARNING: %s appears to be" 4807 " on the same physical disk as" 4808 " %s.\n", 4809 mdname(mddev), 4810 bdevname(rdev->bdev,b), 4811 bdevname(rdev2->bdev,b2)); 4812 warned = 1; 4813 } 4814 } 4815 4816 if (warned) 4817 printk(KERN_WARNING 4818 "True protection against single-disk" 4819 " failure might be compromised.\n"); 4820 } 4821 4822 mddev->recovery = 0; 4823 /* may be over-ridden by personality */ 4824 mddev->resync_max_sectors = mddev->dev_sectors; 4825 4826 mddev->ok_start_degraded = start_dirty_degraded; 4827 4828 if (start_readonly && mddev->ro == 0) 4829 mddev->ro = 2; /* read-only, but switch on first write */ 4830 4831 err = mddev->pers->run(mddev); 4832 if (err) 4833 printk(KERN_ERR "md: pers->run() failed ...\n"); 4834 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { 4835 WARN_ONCE(!mddev->external_size, "%s: default size too small," 4836 " but 'external_size' not in effect?\n", __func__); 4837 printk(KERN_ERR 4838 "md: invalid array_size %llu > default size %llu\n", 4839 (unsigned long long)mddev->array_sectors / 2, 4840 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); 4841 err = -EINVAL; 4842 mddev->pers->stop(mddev); 4843 } 4844 if (err == 0 && mddev->pers->sync_request) { 4845 err = bitmap_create(mddev); 4846 if (err) { 4847 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 4848 mdname(mddev), err); 4849 mddev->pers->stop(mddev); 4850 } 4851 } 4852 if (err) { 4853 module_put(mddev->pers->owner); 4854 mddev->pers = NULL; 4855 bitmap_destroy(mddev); 4856 return err; 4857 } 4858 if (mddev->pers->sync_request) { 4859 if (mddev->kobj.sd && 4860 sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 4861 printk(KERN_WARNING 4862 "md: cannot register extra attributes for %s\n", 4863 mdname(mddev)); 4864 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); 4865 } else if (mddev->ro == 2) /* auto-readonly not meaningful */ 4866 mddev->ro = 0; 4867 4868 atomic_set(&mddev->writes_pending,0); 4869 atomic_set(&mddev->max_corr_read_errors, 4870 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); 4871 mddev->safemode = 0; 4872 mddev->safemode_timer.function = md_safemode_timeout; 4873 mddev->safemode_timer.data = (unsigned long) mddev; 4874 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ 4875 mddev->in_sync = 1; 4876 smp_wmb(); 4877 mddev->ready = 1; 4878 list_for_each_entry(rdev, &mddev->disks, same_set) 4879 if (rdev->raid_disk >= 0) 4880 if (sysfs_link_rdev(mddev, rdev)) 4881 /* failure here is OK */; 4882 4883 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4884 4885 if (mddev->flags) 4886 md_update_sb(mddev, 0); 4887 4888 md_new_event(mddev); 4889 sysfs_notify_dirent_safe(mddev->sysfs_state); 4890 sysfs_notify_dirent_safe(mddev->sysfs_action); 4891 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4892 return 0; 4893 } 4894 EXPORT_SYMBOL_GPL(md_run); 4895 4896 static int do_md_run(mddev_t *mddev) 4897 { 4898 int err; 4899 4900 err = md_run(mddev); 4901 if (err) 4902 goto out; 4903 err = bitmap_load(mddev); 4904 if (err) { 4905 bitmap_destroy(mddev); 4906 goto out; 4907 } 4908 4909 md_wakeup_thread(mddev->thread); 4910 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 4911 4912 set_capacity(mddev->gendisk, mddev->array_sectors); 4913 revalidate_disk(mddev->gendisk); 4914 mddev->changed = 1; 4915 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4916 out: 4917 return err; 4918 } 4919 4920 static int restart_array(mddev_t *mddev) 4921 { 4922 struct gendisk *disk = mddev->gendisk; 4923 4924 /* Complain if it has no devices */ 4925 if (list_empty(&mddev->disks)) 4926 return -ENXIO; 4927 if (!mddev->pers) 4928 return -EINVAL; 4929 if (!mddev->ro) 4930 return -EBUSY; 4931 mddev->safemode = 0; 4932 mddev->ro = 0; 4933 set_disk_ro(disk, 0); 4934 printk(KERN_INFO "md: %s switched to read-write mode.\n", 4935 mdname(mddev)); 4936 /* Kick recovery or resync if necessary */ 4937 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4938 md_wakeup_thread(mddev->thread); 4939 md_wakeup_thread(mddev->sync_thread); 4940 sysfs_notify_dirent_safe(mddev->sysfs_state); 4941 return 0; 4942 } 4943 4944 /* similar to deny_write_access, but accounts for our holding a reference 4945 * to the file ourselves */ 4946 static int deny_bitmap_write_access(struct file * file) 4947 { 4948 struct inode *inode = file->f_mapping->host; 4949 4950 spin_lock(&inode->i_lock); 4951 if (atomic_read(&inode->i_writecount) > 1) { 4952 spin_unlock(&inode->i_lock); 4953 return -ETXTBSY; 4954 } 4955 atomic_set(&inode->i_writecount, -1); 4956 spin_unlock(&inode->i_lock); 4957 4958 return 0; 4959 } 4960 4961 void restore_bitmap_write_access(struct file *file) 4962 { 4963 struct inode *inode = file->f_mapping->host; 4964 4965 spin_lock(&inode->i_lock); 4966 atomic_set(&inode->i_writecount, 1); 4967 spin_unlock(&inode->i_lock); 4968 } 4969 4970 static void md_clean(mddev_t *mddev) 4971 { 4972 mddev->array_sectors = 0; 4973 mddev->external_size = 0; 4974 mddev->dev_sectors = 0; 4975 mddev->raid_disks = 0; 4976 mddev->recovery_cp = 0; 4977 mddev->resync_min = 0; 4978 mddev->resync_max = MaxSector; 4979 mddev->reshape_position = MaxSector; 4980 mddev->external = 0; 4981 mddev->persistent = 0; 4982 mddev->level = LEVEL_NONE; 4983 mddev->clevel[0] = 0; 4984 mddev->flags = 0; 4985 mddev->ro = 0; 4986 mddev->metadata_type[0] = 0; 4987 mddev->chunk_sectors = 0; 4988 mddev->ctime = mddev->utime = 0; 4989 mddev->layout = 0; 4990 mddev->max_disks = 0; 4991 mddev->events = 0; 4992 mddev->can_decrease_events = 0; 4993 mddev->delta_disks = 0; 4994 mddev->new_level = LEVEL_NONE; 4995 mddev->new_layout = 0; 4996 mddev->new_chunk_sectors = 0; 4997 mddev->curr_resync = 0; 4998 mddev->resync_mismatches = 0; 4999 mddev->suspend_lo = mddev->suspend_hi = 0; 5000 mddev->sync_speed_min = mddev->sync_speed_max = 0; 5001 mddev->recovery = 0; 5002 mddev->in_sync = 0; 5003 mddev->changed = 0; 5004 mddev->degraded = 0; 5005 mddev->safemode = 0; 5006 mddev->bitmap_info.offset = 0; 5007 mddev->bitmap_info.default_offset = 0; 5008 mddev->bitmap_info.chunksize = 0; 5009 mddev->bitmap_info.daemon_sleep = 0; 5010 mddev->bitmap_info.max_write_behind = 0; 5011 } 5012 5013 static void __md_stop_writes(mddev_t *mddev) 5014 { 5015 if (mddev->sync_thread) { 5016 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5017 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 5018 reap_sync_thread(mddev); 5019 } 5020 5021 del_timer_sync(&mddev->safemode_timer); 5022 5023 bitmap_flush(mddev); 5024 md_super_wait(mddev); 5025 5026 if (!mddev->in_sync || mddev->flags) { 5027 /* mark array as shutdown cleanly */ 5028 mddev->in_sync = 1; 5029 md_update_sb(mddev, 1); 5030 } 5031 } 5032 5033 void md_stop_writes(mddev_t *mddev) 5034 { 5035 mddev_lock(mddev); 5036 __md_stop_writes(mddev); 5037 mddev_unlock(mddev); 5038 } 5039 EXPORT_SYMBOL_GPL(md_stop_writes); 5040 5041 void md_stop(mddev_t *mddev) 5042 { 5043 mddev->ready = 0; 5044 mddev->pers->stop(mddev); 5045 if (mddev->pers->sync_request && mddev->to_remove == NULL) 5046 mddev->to_remove = &md_redundancy_group; 5047 module_put(mddev->pers->owner); 5048 mddev->pers = NULL; 5049 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5050 } 5051 EXPORT_SYMBOL_GPL(md_stop); 5052 5053 static int md_set_readonly(mddev_t *mddev, int is_open) 5054 { 5055 int err = 0; 5056 mutex_lock(&mddev->open_mutex); 5057 if (atomic_read(&mddev->openers) > is_open) { 5058 printk("md: %s still in use.\n",mdname(mddev)); 5059 err = -EBUSY; 5060 goto out; 5061 } 5062 if (mddev->pers) { 5063 __md_stop_writes(mddev); 5064 5065 err = -ENXIO; 5066 if (mddev->ro==1) 5067 goto out; 5068 mddev->ro = 1; 5069 set_disk_ro(mddev->gendisk, 1); 5070 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 5071 sysfs_notify_dirent_safe(mddev->sysfs_state); 5072 err = 0; 5073 } 5074 out: 5075 mutex_unlock(&mddev->open_mutex); 5076 return err; 5077 } 5078 5079 /* mode: 5080 * 0 - completely stop and dis-assemble array 5081 * 2 - stop but do not disassemble array 5082 */ 5083 static int do_md_stop(mddev_t * mddev, int mode, int is_open) 5084 { 5085 struct gendisk *disk = mddev->gendisk; 5086 mdk_rdev_t *rdev; 5087 5088 mutex_lock(&mddev->open_mutex); 5089 if (atomic_read(&mddev->openers) > is_open || 5090 mddev->sysfs_active) { 5091 printk("md: %s still in use.\n",mdname(mddev)); 5092 mutex_unlock(&mddev->open_mutex); 5093 return -EBUSY; 5094 } 5095 5096 if (mddev->pers) { 5097 if (mddev->ro) 5098 set_disk_ro(disk, 0); 5099 5100 __md_stop_writes(mddev); 5101 md_stop(mddev); 5102 mddev->queue->merge_bvec_fn = NULL; 5103 mddev->queue->backing_dev_info.congested_fn = NULL; 5104 5105 /* tell userspace to handle 'inactive' */ 5106 sysfs_notify_dirent_safe(mddev->sysfs_state); 5107 5108 list_for_each_entry(rdev, &mddev->disks, same_set) 5109 if (rdev->raid_disk >= 0) 5110 sysfs_unlink_rdev(mddev, rdev); 5111 5112 set_capacity(disk, 0); 5113 mutex_unlock(&mddev->open_mutex); 5114 mddev->changed = 1; 5115 revalidate_disk(disk); 5116 5117 if (mddev->ro) 5118 mddev->ro = 0; 5119 } else 5120 mutex_unlock(&mddev->open_mutex); 5121 /* 5122 * Free resources if final stop 5123 */ 5124 if (mode == 0) { 5125 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 5126 5127 bitmap_destroy(mddev); 5128 if (mddev->bitmap_info.file) { 5129 restore_bitmap_write_access(mddev->bitmap_info.file); 5130 fput(mddev->bitmap_info.file); 5131 mddev->bitmap_info.file = NULL; 5132 } 5133 mddev->bitmap_info.offset = 0; 5134 5135 export_array(mddev); 5136 5137 md_clean(mddev); 5138 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 5139 if (mddev->hold_active == UNTIL_STOP) 5140 mddev->hold_active = 0; 5141 } 5142 blk_integrity_unregister(disk); 5143 md_new_event(mddev); 5144 sysfs_notify_dirent_safe(mddev->sysfs_state); 5145 return 0; 5146 } 5147 5148 #ifndef MODULE 5149 static void autorun_array(mddev_t *mddev) 5150 { 5151 mdk_rdev_t *rdev; 5152 int err; 5153 5154 if (list_empty(&mddev->disks)) 5155 return; 5156 5157 printk(KERN_INFO "md: running: "); 5158 5159 list_for_each_entry(rdev, &mddev->disks, same_set) { 5160 char b[BDEVNAME_SIZE]; 5161 printk("<%s>", bdevname(rdev->bdev,b)); 5162 } 5163 printk("\n"); 5164 5165 err = do_md_run(mddev); 5166 if (err) { 5167 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 5168 do_md_stop(mddev, 0, 0); 5169 } 5170 } 5171 5172 /* 5173 * lets try to run arrays based on all disks that have arrived 5174 * until now. (those are in pending_raid_disks) 5175 * 5176 * the method: pick the first pending disk, collect all disks with 5177 * the same UUID, remove all from the pending list and put them into 5178 * the 'same_array' list. Then order this list based on superblock 5179 * update time (freshest comes first), kick out 'old' disks and 5180 * compare superblocks. If everything's fine then run it. 5181 * 5182 * If "unit" is allocated, then bump its reference count 5183 */ 5184 static void autorun_devices(int part) 5185 { 5186 mdk_rdev_t *rdev0, *rdev, *tmp; 5187 mddev_t *mddev; 5188 char b[BDEVNAME_SIZE]; 5189 5190 printk(KERN_INFO "md: autorun ...\n"); 5191 while (!list_empty(&pending_raid_disks)) { 5192 int unit; 5193 dev_t dev; 5194 LIST_HEAD(candidates); 5195 rdev0 = list_entry(pending_raid_disks.next, 5196 mdk_rdev_t, same_set); 5197 5198 printk(KERN_INFO "md: considering %s ...\n", 5199 bdevname(rdev0->bdev,b)); 5200 INIT_LIST_HEAD(&candidates); 5201 rdev_for_each_list(rdev, tmp, &pending_raid_disks) 5202 if (super_90_load(rdev, rdev0, 0) >= 0) { 5203 printk(KERN_INFO "md: adding %s ...\n", 5204 bdevname(rdev->bdev,b)); 5205 list_move(&rdev->same_set, &candidates); 5206 } 5207 /* 5208 * now we have a set of devices, with all of them having 5209 * mostly sane superblocks. It's time to allocate the 5210 * mddev. 5211 */ 5212 if (part) { 5213 dev = MKDEV(mdp_major, 5214 rdev0->preferred_minor << MdpMinorShift); 5215 unit = MINOR(dev) >> MdpMinorShift; 5216 } else { 5217 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 5218 unit = MINOR(dev); 5219 } 5220 if (rdev0->preferred_minor != unit) { 5221 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 5222 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 5223 break; 5224 } 5225 5226 md_probe(dev, NULL, NULL); 5227 mddev = mddev_find(dev); 5228 if (!mddev || !mddev->gendisk) { 5229 if (mddev) 5230 mddev_put(mddev); 5231 printk(KERN_ERR 5232 "md: cannot allocate memory for md drive.\n"); 5233 break; 5234 } 5235 if (mddev_lock(mddev)) 5236 printk(KERN_WARNING "md: %s locked, cannot run\n", 5237 mdname(mddev)); 5238 else if (mddev->raid_disks || mddev->major_version 5239 || !list_empty(&mddev->disks)) { 5240 printk(KERN_WARNING 5241 "md: %s already running, cannot run %s\n", 5242 mdname(mddev), bdevname(rdev0->bdev,b)); 5243 mddev_unlock(mddev); 5244 } else { 5245 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 5246 mddev->persistent = 1; 5247 rdev_for_each_list(rdev, tmp, &candidates) { 5248 list_del_init(&rdev->same_set); 5249 if (bind_rdev_to_array(rdev, mddev)) 5250 export_rdev(rdev); 5251 } 5252 autorun_array(mddev); 5253 mddev_unlock(mddev); 5254 } 5255 /* on success, candidates will be empty, on error 5256 * it won't... 5257 */ 5258 rdev_for_each_list(rdev, tmp, &candidates) { 5259 list_del_init(&rdev->same_set); 5260 export_rdev(rdev); 5261 } 5262 mddev_put(mddev); 5263 } 5264 printk(KERN_INFO "md: ... autorun DONE.\n"); 5265 } 5266 #endif /* !MODULE */ 5267 5268 static int get_version(void __user * arg) 5269 { 5270 mdu_version_t ver; 5271 5272 ver.major = MD_MAJOR_VERSION; 5273 ver.minor = MD_MINOR_VERSION; 5274 ver.patchlevel = MD_PATCHLEVEL_VERSION; 5275 5276 if (copy_to_user(arg, &ver, sizeof(ver))) 5277 return -EFAULT; 5278 5279 return 0; 5280 } 5281 5282 static int get_array_info(mddev_t * mddev, void __user * arg) 5283 { 5284 mdu_array_info_t info; 5285 int nr,working,insync,failed,spare; 5286 mdk_rdev_t *rdev; 5287 5288 nr=working=insync=failed=spare=0; 5289 list_for_each_entry(rdev, &mddev->disks, same_set) { 5290 nr++; 5291 if (test_bit(Faulty, &rdev->flags)) 5292 failed++; 5293 else { 5294 working++; 5295 if (test_bit(In_sync, &rdev->flags)) 5296 insync++; 5297 else 5298 spare++; 5299 } 5300 } 5301 5302 info.major_version = mddev->major_version; 5303 info.minor_version = mddev->minor_version; 5304 info.patch_version = MD_PATCHLEVEL_VERSION; 5305 info.ctime = mddev->ctime; 5306 info.level = mddev->level; 5307 info.size = mddev->dev_sectors / 2; 5308 if (info.size != mddev->dev_sectors / 2) /* overflow */ 5309 info.size = -1; 5310 info.nr_disks = nr; 5311 info.raid_disks = mddev->raid_disks; 5312 info.md_minor = mddev->md_minor; 5313 info.not_persistent= !mddev->persistent; 5314 5315 info.utime = mddev->utime; 5316 info.state = 0; 5317 if (mddev->in_sync) 5318 info.state = (1<<MD_SB_CLEAN); 5319 if (mddev->bitmap && mddev->bitmap_info.offset) 5320 info.state = (1<<MD_SB_BITMAP_PRESENT); 5321 info.active_disks = insync; 5322 info.working_disks = working; 5323 info.failed_disks = failed; 5324 info.spare_disks = spare; 5325 5326 info.layout = mddev->layout; 5327 info.chunk_size = mddev->chunk_sectors << 9; 5328 5329 if (copy_to_user(arg, &info, sizeof(info))) 5330 return -EFAULT; 5331 5332 return 0; 5333 } 5334 5335 static int get_bitmap_file(mddev_t * mddev, void __user * arg) 5336 { 5337 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 5338 char *ptr, *buf = NULL; 5339 int err = -ENOMEM; 5340 5341 if (md_allow_write(mddev)) 5342 file = kmalloc(sizeof(*file), GFP_NOIO); 5343 else 5344 file = kmalloc(sizeof(*file), GFP_KERNEL); 5345 5346 if (!file) 5347 goto out; 5348 5349 /* bitmap disabled, zero the first byte and copy out */ 5350 if (!mddev->bitmap || !mddev->bitmap->file) { 5351 file->pathname[0] = '\0'; 5352 goto copy_out; 5353 } 5354 5355 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 5356 if (!buf) 5357 goto out; 5358 5359 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); 5360 if (IS_ERR(ptr)) 5361 goto out; 5362 5363 strcpy(file->pathname, ptr); 5364 5365 copy_out: 5366 err = 0; 5367 if (copy_to_user(arg, file, sizeof(*file))) 5368 err = -EFAULT; 5369 out: 5370 kfree(buf); 5371 kfree(file); 5372 return err; 5373 } 5374 5375 static int get_disk_info(mddev_t * mddev, void __user * arg) 5376 { 5377 mdu_disk_info_t info; 5378 mdk_rdev_t *rdev; 5379 5380 if (copy_from_user(&info, arg, sizeof(info))) 5381 return -EFAULT; 5382 5383 rdev = find_rdev_nr(mddev, info.number); 5384 if (rdev) { 5385 info.major = MAJOR(rdev->bdev->bd_dev); 5386 info.minor = MINOR(rdev->bdev->bd_dev); 5387 info.raid_disk = rdev->raid_disk; 5388 info.state = 0; 5389 if (test_bit(Faulty, &rdev->flags)) 5390 info.state |= (1<<MD_DISK_FAULTY); 5391 else if (test_bit(In_sync, &rdev->flags)) { 5392 info.state |= (1<<MD_DISK_ACTIVE); 5393 info.state |= (1<<MD_DISK_SYNC); 5394 } 5395 if (test_bit(WriteMostly, &rdev->flags)) 5396 info.state |= (1<<MD_DISK_WRITEMOSTLY); 5397 } else { 5398 info.major = info.minor = 0; 5399 info.raid_disk = -1; 5400 info.state = (1<<MD_DISK_REMOVED); 5401 } 5402 5403 if (copy_to_user(arg, &info, sizeof(info))) 5404 return -EFAULT; 5405 5406 return 0; 5407 } 5408 5409 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) 5410 { 5411 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 5412 mdk_rdev_t *rdev; 5413 dev_t dev = MKDEV(info->major,info->minor); 5414 5415 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 5416 return -EOVERFLOW; 5417 5418 if (!mddev->raid_disks) { 5419 int err; 5420 /* expecting a device which has a superblock */ 5421 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 5422 if (IS_ERR(rdev)) { 5423 printk(KERN_WARNING 5424 "md: md_import_device returned %ld\n", 5425 PTR_ERR(rdev)); 5426 return PTR_ERR(rdev); 5427 } 5428 if (!list_empty(&mddev->disks)) { 5429 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 5430 mdk_rdev_t, same_set); 5431 err = super_types[mddev->major_version] 5432 .load_super(rdev, rdev0, mddev->minor_version); 5433 if (err < 0) { 5434 printk(KERN_WARNING 5435 "md: %s has different UUID to %s\n", 5436 bdevname(rdev->bdev,b), 5437 bdevname(rdev0->bdev,b2)); 5438 export_rdev(rdev); 5439 return -EINVAL; 5440 } 5441 } 5442 err = bind_rdev_to_array(rdev, mddev); 5443 if (err) 5444 export_rdev(rdev); 5445 return err; 5446 } 5447 5448 /* 5449 * add_new_disk can be used once the array is assembled 5450 * to add "hot spares". They must already have a superblock 5451 * written 5452 */ 5453 if (mddev->pers) { 5454 int err; 5455 if (!mddev->pers->hot_add_disk) { 5456 printk(KERN_WARNING 5457 "%s: personality does not support diskops!\n", 5458 mdname(mddev)); 5459 return -EINVAL; 5460 } 5461 if (mddev->persistent) 5462 rdev = md_import_device(dev, mddev->major_version, 5463 mddev->minor_version); 5464 else 5465 rdev = md_import_device(dev, -1, -1); 5466 if (IS_ERR(rdev)) { 5467 printk(KERN_WARNING 5468 "md: md_import_device returned %ld\n", 5469 PTR_ERR(rdev)); 5470 return PTR_ERR(rdev); 5471 } 5472 /* set saved_raid_disk if appropriate */ 5473 if (!mddev->persistent) { 5474 if (info->state & (1<<MD_DISK_SYNC) && 5475 info->raid_disk < mddev->raid_disks) { 5476 rdev->raid_disk = info->raid_disk; 5477 set_bit(In_sync, &rdev->flags); 5478 } else 5479 rdev->raid_disk = -1; 5480 } else 5481 super_types[mddev->major_version]. 5482 validate_super(mddev, rdev); 5483 if ((info->state & (1<<MD_DISK_SYNC)) && 5484 (!test_bit(In_sync, &rdev->flags) || 5485 rdev->raid_disk != info->raid_disk)) { 5486 /* This was a hot-add request, but events doesn't 5487 * match, so reject it. 5488 */ 5489 export_rdev(rdev); 5490 return -EINVAL; 5491 } 5492 5493 if (test_bit(In_sync, &rdev->flags)) 5494 rdev->saved_raid_disk = rdev->raid_disk; 5495 else 5496 rdev->saved_raid_disk = -1; 5497 5498 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 5499 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5500 set_bit(WriteMostly, &rdev->flags); 5501 else 5502 clear_bit(WriteMostly, &rdev->flags); 5503 5504 rdev->raid_disk = -1; 5505 err = bind_rdev_to_array(rdev, mddev); 5506 if (!err && !mddev->pers->hot_remove_disk) { 5507 /* If there is hot_add_disk but no hot_remove_disk 5508 * then added disks for geometry changes, 5509 * and should be added immediately. 5510 */ 5511 super_types[mddev->major_version]. 5512 validate_super(mddev, rdev); 5513 err = mddev->pers->hot_add_disk(mddev, rdev); 5514 if (err) 5515 unbind_rdev_from_array(rdev); 5516 } 5517 if (err) 5518 export_rdev(rdev); 5519 else 5520 sysfs_notify_dirent_safe(rdev->sysfs_state); 5521 5522 md_update_sb(mddev, 1); 5523 if (mddev->degraded) 5524 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 5525 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5526 if (!err) 5527 md_new_event(mddev); 5528 md_wakeup_thread(mddev->thread); 5529 return err; 5530 } 5531 5532 /* otherwise, add_new_disk is only allowed 5533 * for major_version==0 superblocks 5534 */ 5535 if (mddev->major_version != 0) { 5536 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 5537 mdname(mddev)); 5538 return -EINVAL; 5539 } 5540 5541 if (!(info->state & (1<<MD_DISK_FAULTY))) { 5542 int err; 5543 rdev = md_import_device(dev, -1, 0); 5544 if (IS_ERR(rdev)) { 5545 printk(KERN_WARNING 5546 "md: error, md_import_device() returned %ld\n", 5547 PTR_ERR(rdev)); 5548 return PTR_ERR(rdev); 5549 } 5550 rdev->desc_nr = info->number; 5551 if (info->raid_disk < mddev->raid_disks) 5552 rdev->raid_disk = info->raid_disk; 5553 else 5554 rdev->raid_disk = -1; 5555 5556 if (rdev->raid_disk < mddev->raid_disks) 5557 if (info->state & (1<<MD_DISK_SYNC)) 5558 set_bit(In_sync, &rdev->flags); 5559 5560 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5561 set_bit(WriteMostly, &rdev->flags); 5562 5563 if (!mddev->persistent) { 5564 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 5565 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5566 } else 5567 rdev->sb_start = calc_dev_sboffset(rdev); 5568 rdev->sectors = rdev->sb_start; 5569 5570 err = bind_rdev_to_array(rdev, mddev); 5571 if (err) { 5572 export_rdev(rdev); 5573 return err; 5574 } 5575 } 5576 5577 return 0; 5578 } 5579 5580 static int hot_remove_disk(mddev_t * mddev, dev_t dev) 5581 { 5582 char b[BDEVNAME_SIZE]; 5583 mdk_rdev_t *rdev; 5584 5585 rdev = find_rdev(mddev, dev); 5586 if (!rdev) 5587 return -ENXIO; 5588 5589 if (rdev->raid_disk >= 0) 5590 goto busy; 5591 5592 kick_rdev_from_array(rdev); 5593 md_update_sb(mddev, 1); 5594 md_new_event(mddev); 5595 5596 return 0; 5597 busy: 5598 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", 5599 bdevname(rdev->bdev,b), mdname(mddev)); 5600 return -EBUSY; 5601 } 5602 5603 static int hot_add_disk(mddev_t * mddev, dev_t dev) 5604 { 5605 char b[BDEVNAME_SIZE]; 5606 int err; 5607 mdk_rdev_t *rdev; 5608 5609 if (!mddev->pers) 5610 return -ENODEV; 5611 5612 if (mddev->major_version != 0) { 5613 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 5614 " version-0 superblocks.\n", 5615 mdname(mddev)); 5616 return -EINVAL; 5617 } 5618 if (!mddev->pers->hot_add_disk) { 5619 printk(KERN_WARNING 5620 "%s: personality does not support diskops!\n", 5621 mdname(mddev)); 5622 return -EINVAL; 5623 } 5624 5625 rdev = md_import_device(dev, -1, 0); 5626 if (IS_ERR(rdev)) { 5627 printk(KERN_WARNING 5628 "md: error, md_import_device() returned %ld\n", 5629 PTR_ERR(rdev)); 5630 return -EINVAL; 5631 } 5632 5633 if (mddev->persistent) 5634 rdev->sb_start = calc_dev_sboffset(rdev); 5635 else 5636 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5637 5638 rdev->sectors = rdev->sb_start; 5639 5640 if (test_bit(Faulty, &rdev->flags)) { 5641 printk(KERN_WARNING 5642 "md: can not hot-add faulty %s disk to %s!\n", 5643 bdevname(rdev->bdev,b), mdname(mddev)); 5644 err = -EINVAL; 5645 goto abort_export; 5646 } 5647 clear_bit(In_sync, &rdev->flags); 5648 rdev->desc_nr = -1; 5649 rdev->saved_raid_disk = -1; 5650 err = bind_rdev_to_array(rdev, mddev); 5651 if (err) 5652 goto abort_export; 5653 5654 /* 5655 * The rest should better be atomic, we can have disk failures 5656 * noticed in interrupt contexts ... 5657 */ 5658 5659 rdev->raid_disk = -1; 5660 5661 md_update_sb(mddev, 1); 5662 5663 /* 5664 * Kick recovery, maybe this spare has to be added to the 5665 * array immediately. 5666 */ 5667 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5668 md_wakeup_thread(mddev->thread); 5669 md_new_event(mddev); 5670 return 0; 5671 5672 abort_export: 5673 export_rdev(rdev); 5674 return err; 5675 } 5676 5677 static int set_bitmap_file(mddev_t *mddev, int fd) 5678 { 5679 int err; 5680 5681 if (mddev->pers) { 5682 if (!mddev->pers->quiesce) 5683 return -EBUSY; 5684 if (mddev->recovery || mddev->sync_thread) 5685 return -EBUSY; 5686 /* we should be able to change the bitmap.. */ 5687 } 5688 5689 5690 if (fd >= 0) { 5691 if (mddev->bitmap) 5692 return -EEXIST; /* cannot add when bitmap is present */ 5693 mddev->bitmap_info.file = fget(fd); 5694 5695 if (mddev->bitmap_info.file == NULL) { 5696 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 5697 mdname(mddev)); 5698 return -EBADF; 5699 } 5700 5701 err = deny_bitmap_write_access(mddev->bitmap_info.file); 5702 if (err) { 5703 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 5704 mdname(mddev)); 5705 fput(mddev->bitmap_info.file); 5706 mddev->bitmap_info.file = NULL; 5707 return err; 5708 } 5709 mddev->bitmap_info.offset = 0; /* file overrides offset */ 5710 } else if (mddev->bitmap == NULL) 5711 return -ENOENT; /* cannot remove what isn't there */ 5712 err = 0; 5713 if (mddev->pers) { 5714 mddev->pers->quiesce(mddev, 1); 5715 if (fd >= 0) { 5716 err = bitmap_create(mddev); 5717 if (!err) 5718 err = bitmap_load(mddev); 5719 } 5720 if (fd < 0 || err) { 5721 bitmap_destroy(mddev); 5722 fd = -1; /* make sure to put the file */ 5723 } 5724 mddev->pers->quiesce(mddev, 0); 5725 } 5726 if (fd < 0) { 5727 if (mddev->bitmap_info.file) { 5728 restore_bitmap_write_access(mddev->bitmap_info.file); 5729 fput(mddev->bitmap_info.file); 5730 } 5731 mddev->bitmap_info.file = NULL; 5732 } 5733 5734 return err; 5735 } 5736 5737 /* 5738 * set_array_info is used two different ways 5739 * The original usage is when creating a new array. 5740 * In this usage, raid_disks is > 0 and it together with 5741 * level, size, not_persistent,layout,chunksize determine the 5742 * shape of the array. 5743 * This will always create an array with a type-0.90.0 superblock. 5744 * The newer usage is when assembling an array. 5745 * In this case raid_disks will be 0, and the major_version field is 5746 * use to determine which style super-blocks are to be found on the devices. 5747 * The minor and patch _version numbers are also kept incase the 5748 * super_block handler wishes to interpret them. 5749 */ 5750 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) 5751 { 5752 5753 if (info->raid_disks == 0) { 5754 /* just setting version number for superblock loading */ 5755 if (info->major_version < 0 || 5756 info->major_version >= ARRAY_SIZE(super_types) || 5757 super_types[info->major_version].name == NULL) { 5758 /* maybe try to auto-load a module? */ 5759 printk(KERN_INFO 5760 "md: superblock version %d not known\n", 5761 info->major_version); 5762 return -EINVAL; 5763 } 5764 mddev->major_version = info->major_version; 5765 mddev->minor_version = info->minor_version; 5766 mddev->patch_version = info->patch_version; 5767 mddev->persistent = !info->not_persistent; 5768 /* ensure mddev_put doesn't delete this now that there 5769 * is some minimal configuration. 5770 */ 5771 mddev->ctime = get_seconds(); 5772 return 0; 5773 } 5774 mddev->major_version = MD_MAJOR_VERSION; 5775 mddev->minor_version = MD_MINOR_VERSION; 5776 mddev->patch_version = MD_PATCHLEVEL_VERSION; 5777 mddev->ctime = get_seconds(); 5778 5779 mddev->level = info->level; 5780 mddev->clevel[0] = 0; 5781 mddev->dev_sectors = 2 * (sector_t)info->size; 5782 mddev->raid_disks = info->raid_disks; 5783 /* don't set md_minor, it is determined by which /dev/md* was 5784 * openned 5785 */ 5786 if (info->state & (1<<MD_SB_CLEAN)) 5787 mddev->recovery_cp = MaxSector; 5788 else 5789 mddev->recovery_cp = 0; 5790 mddev->persistent = ! info->not_persistent; 5791 mddev->external = 0; 5792 5793 mddev->layout = info->layout; 5794 mddev->chunk_sectors = info->chunk_size >> 9; 5795 5796 mddev->max_disks = MD_SB_DISKS; 5797 5798 if (mddev->persistent) 5799 mddev->flags = 0; 5800 set_bit(MD_CHANGE_DEVS, &mddev->flags); 5801 5802 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 5803 mddev->bitmap_info.offset = 0; 5804 5805 mddev->reshape_position = MaxSector; 5806 5807 /* 5808 * Generate a 128 bit UUID 5809 */ 5810 get_random_bytes(mddev->uuid, 16); 5811 5812 mddev->new_level = mddev->level; 5813 mddev->new_chunk_sectors = mddev->chunk_sectors; 5814 mddev->new_layout = mddev->layout; 5815 mddev->delta_disks = 0; 5816 5817 return 0; 5818 } 5819 5820 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) 5821 { 5822 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); 5823 5824 if (mddev->external_size) 5825 return; 5826 5827 mddev->array_sectors = array_sectors; 5828 } 5829 EXPORT_SYMBOL(md_set_array_sectors); 5830 5831 static int update_size(mddev_t *mddev, sector_t num_sectors) 5832 { 5833 mdk_rdev_t *rdev; 5834 int rv; 5835 int fit = (num_sectors == 0); 5836 5837 if (mddev->pers->resize == NULL) 5838 return -EINVAL; 5839 /* The "num_sectors" is the number of sectors of each device that 5840 * is used. This can only make sense for arrays with redundancy. 5841 * linear and raid0 always use whatever space is available. We can only 5842 * consider changing this number if no resync or reconstruction is 5843 * happening, and if the new size is acceptable. It must fit before the 5844 * sb_start or, if that is <data_offset, it must fit before the size 5845 * of each device. If num_sectors is zero, we find the largest size 5846 * that fits. 5847 */ 5848 if (mddev->sync_thread) 5849 return -EBUSY; 5850 if (mddev->bitmap) 5851 /* Sorry, cannot grow a bitmap yet, just remove it, 5852 * grow, and re-add. 5853 */ 5854 return -EBUSY; 5855 list_for_each_entry(rdev, &mddev->disks, same_set) { 5856 sector_t avail = rdev->sectors; 5857 5858 if (fit && (num_sectors == 0 || num_sectors > avail)) 5859 num_sectors = avail; 5860 if (avail < num_sectors) 5861 return -ENOSPC; 5862 } 5863 rv = mddev->pers->resize(mddev, num_sectors); 5864 if (!rv) 5865 revalidate_disk(mddev->gendisk); 5866 return rv; 5867 } 5868 5869 static int update_raid_disks(mddev_t *mddev, int raid_disks) 5870 { 5871 int rv; 5872 /* change the number of raid disks */ 5873 if (mddev->pers->check_reshape == NULL) 5874 return -EINVAL; 5875 if (raid_disks <= 0 || 5876 (mddev->max_disks && raid_disks >= mddev->max_disks)) 5877 return -EINVAL; 5878 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 5879 return -EBUSY; 5880 mddev->delta_disks = raid_disks - mddev->raid_disks; 5881 5882 rv = mddev->pers->check_reshape(mddev); 5883 if (rv < 0) 5884 mddev->delta_disks = 0; 5885 return rv; 5886 } 5887 5888 5889 /* 5890 * update_array_info is used to change the configuration of an 5891 * on-line array. 5892 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 5893 * fields in the info are checked against the array. 5894 * Any differences that cannot be handled will cause an error. 5895 * Normally, only one change can be managed at a time. 5896 */ 5897 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) 5898 { 5899 int rv = 0; 5900 int cnt = 0; 5901 int state = 0; 5902 5903 /* calculate expected state,ignoring low bits */ 5904 if (mddev->bitmap && mddev->bitmap_info.offset) 5905 state |= (1 << MD_SB_BITMAP_PRESENT); 5906 5907 if (mddev->major_version != info->major_version || 5908 mddev->minor_version != info->minor_version || 5909 /* mddev->patch_version != info->patch_version || */ 5910 mddev->ctime != info->ctime || 5911 mddev->level != info->level || 5912 /* mddev->layout != info->layout || */ 5913 !mddev->persistent != info->not_persistent|| 5914 mddev->chunk_sectors != info->chunk_size >> 9 || 5915 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 5916 ((state^info->state) & 0xfffffe00) 5917 ) 5918 return -EINVAL; 5919 /* Check there is only one change */ 5920 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5921 cnt++; 5922 if (mddev->raid_disks != info->raid_disks) 5923 cnt++; 5924 if (mddev->layout != info->layout) 5925 cnt++; 5926 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) 5927 cnt++; 5928 if (cnt == 0) 5929 return 0; 5930 if (cnt > 1) 5931 return -EINVAL; 5932 5933 if (mddev->layout != info->layout) { 5934 /* Change layout 5935 * we don't need to do anything at the md level, the 5936 * personality will take care of it all. 5937 */ 5938 if (mddev->pers->check_reshape == NULL) 5939 return -EINVAL; 5940 else { 5941 mddev->new_layout = info->layout; 5942 rv = mddev->pers->check_reshape(mddev); 5943 if (rv) 5944 mddev->new_layout = mddev->layout; 5945 return rv; 5946 } 5947 } 5948 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5949 rv = update_size(mddev, (sector_t)info->size * 2); 5950 5951 if (mddev->raid_disks != info->raid_disks) 5952 rv = update_raid_disks(mddev, info->raid_disks); 5953 5954 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 5955 if (mddev->pers->quiesce == NULL) 5956 return -EINVAL; 5957 if (mddev->recovery || mddev->sync_thread) 5958 return -EBUSY; 5959 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 5960 /* add the bitmap */ 5961 if (mddev->bitmap) 5962 return -EEXIST; 5963 if (mddev->bitmap_info.default_offset == 0) 5964 return -EINVAL; 5965 mddev->bitmap_info.offset = 5966 mddev->bitmap_info.default_offset; 5967 mddev->pers->quiesce(mddev, 1); 5968 rv = bitmap_create(mddev); 5969 if (!rv) 5970 rv = bitmap_load(mddev); 5971 if (rv) 5972 bitmap_destroy(mddev); 5973 mddev->pers->quiesce(mddev, 0); 5974 } else { 5975 /* remove the bitmap */ 5976 if (!mddev->bitmap) 5977 return -ENOENT; 5978 if (mddev->bitmap->file) 5979 return -EINVAL; 5980 mddev->pers->quiesce(mddev, 1); 5981 bitmap_destroy(mddev); 5982 mddev->pers->quiesce(mddev, 0); 5983 mddev->bitmap_info.offset = 0; 5984 } 5985 } 5986 md_update_sb(mddev, 1); 5987 return rv; 5988 } 5989 5990 static int set_disk_faulty(mddev_t *mddev, dev_t dev) 5991 { 5992 mdk_rdev_t *rdev; 5993 5994 if (mddev->pers == NULL) 5995 return -ENODEV; 5996 5997 rdev = find_rdev(mddev, dev); 5998 if (!rdev) 5999 return -ENODEV; 6000 6001 md_error(mddev, rdev); 6002 if (!test_bit(Faulty, &rdev->flags)) 6003 return -EBUSY; 6004 return 0; 6005 } 6006 6007 /* 6008 * We have a problem here : there is no easy way to give a CHS 6009 * virtual geometry. We currently pretend that we have a 2 heads 6010 * 4 sectors (with a BIG number of cylinders...). This drives 6011 * dosfs just mad... ;-) 6012 */ 6013 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) 6014 { 6015 mddev_t *mddev = bdev->bd_disk->private_data; 6016 6017 geo->heads = 2; 6018 geo->sectors = 4; 6019 geo->cylinders = mddev->array_sectors / 8; 6020 return 0; 6021 } 6022 6023 static int md_ioctl(struct block_device *bdev, fmode_t mode, 6024 unsigned int cmd, unsigned long arg) 6025 { 6026 int err = 0; 6027 void __user *argp = (void __user *)arg; 6028 mddev_t *mddev = NULL; 6029 int ro; 6030 6031 if (!capable(CAP_SYS_ADMIN)) 6032 return -EACCES; 6033 6034 /* 6035 * Commands dealing with the RAID driver but not any 6036 * particular array: 6037 */ 6038 switch (cmd) 6039 { 6040 case RAID_VERSION: 6041 err = get_version(argp); 6042 goto done; 6043 6044 case PRINT_RAID_DEBUG: 6045 err = 0; 6046 md_print_devices(); 6047 goto done; 6048 6049 #ifndef MODULE 6050 case RAID_AUTORUN: 6051 err = 0; 6052 autostart_arrays(arg); 6053 goto done; 6054 #endif 6055 default:; 6056 } 6057 6058 /* 6059 * Commands creating/starting a new array: 6060 */ 6061 6062 mddev = bdev->bd_disk->private_data; 6063 6064 if (!mddev) { 6065 BUG(); 6066 goto abort; 6067 } 6068 6069 err = mddev_lock(mddev); 6070 if (err) { 6071 printk(KERN_INFO 6072 "md: ioctl lock interrupted, reason %d, cmd %d\n", 6073 err, cmd); 6074 goto abort; 6075 } 6076 6077 switch (cmd) 6078 { 6079 case SET_ARRAY_INFO: 6080 { 6081 mdu_array_info_t info; 6082 if (!arg) 6083 memset(&info, 0, sizeof(info)); 6084 else if (copy_from_user(&info, argp, sizeof(info))) { 6085 err = -EFAULT; 6086 goto abort_unlock; 6087 } 6088 if (mddev->pers) { 6089 err = update_array_info(mddev, &info); 6090 if (err) { 6091 printk(KERN_WARNING "md: couldn't update" 6092 " array info. %d\n", err); 6093 goto abort_unlock; 6094 } 6095 goto done_unlock; 6096 } 6097 if (!list_empty(&mddev->disks)) { 6098 printk(KERN_WARNING 6099 "md: array %s already has disks!\n", 6100 mdname(mddev)); 6101 err = -EBUSY; 6102 goto abort_unlock; 6103 } 6104 if (mddev->raid_disks) { 6105 printk(KERN_WARNING 6106 "md: array %s already initialised!\n", 6107 mdname(mddev)); 6108 err = -EBUSY; 6109 goto abort_unlock; 6110 } 6111 err = set_array_info(mddev, &info); 6112 if (err) { 6113 printk(KERN_WARNING "md: couldn't set" 6114 " array info. %d\n", err); 6115 goto abort_unlock; 6116 } 6117 } 6118 goto done_unlock; 6119 6120 default:; 6121 } 6122 6123 /* 6124 * Commands querying/configuring an existing array: 6125 */ 6126 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 6127 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ 6128 if ((!mddev->raid_disks && !mddev->external) 6129 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 6130 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE 6131 && cmd != GET_BITMAP_FILE) { 6132 err = -ENODEV; 6133 goto abort_unlock; 6134 } 6135 6136 /* 6137 * Commands even a read-only array can execute: 6138 */ 6139 switch (cmd) 6140 { 6141 case GET_ARRAY_INFO: 6142 err = get_array_info(mddev, argp); 6143 goto done_unlock; 6144 6145 case GET_BITMAP_FILE: 6146 err = get_bitmap_file(mddev, argp); 6147 goto done_unlock; 6148 6149 case GET_DISK_INFO: 6150 err = get_disk_info(mddev, argp); 6151 goto done_unlock; 6152 6153 case RESTART_ARRAY_RW: 6154 err = restart_array(mddev); 6155 goto done_unlock; 6156 6157 case STOP_ARRAY: 6158 err = do_md_stop(mddev, 0, 1); 6159 goto done_unlock; 6160 6161 case STOP_ARRAY_RO: 6162 err = md_set_readonly(mddev, 1); 6163 goto done_unlock; 6164 6165 case BLKROSET: 6166 if (get_user(ro, (int __user *)(arg))) { 6167 err = -EFAULT; 6168 goto done_unlock; 6169 } 6170 err = -EINVAL; 6171 6172 /* if the bdev is going readonly the value of mddev->ro 6173 * does not matter, no writes are coming 6174 */ 6175 if (ro) 6176 goto done_unlock; 6177 6178 /* are we are already prepared for writes? */ 6179 if (mddev->ro != 1) 6180 goto done_unlock; 6181 6182 /* transitioning to readauto need only happen for 6183 * arrays that call md_write_start 6184 */ 6185 if (mddev->pers) { 6186 err = restart_array(mddev); 6187 if (err == 0) { 6188 mddev->ro = 2; 6189 set_disk_ro(mddev->gendisk, 0); 6190 } 6191 } 6192 goto done_unlock; 6193 } 6194 6195 /* 6196 * The remaining ioctls are changing the state of the 6197 * superblock, so we do not allow them on read-only arrays. 6198 * However non-MD ioctls (e.g. get-size) will still come through 6199 * here and hit the 'default' below, so only disallow 6200 * 'md' ioctls, and switch to rw mode if started auto-readonly. 6201 */ 6202 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { 6203 if (mddev->ro == 2) { 6204 mddev->ro = 0; 6205 sysfs_notify_dirent_safe(mddev->sysfs_state); 6206 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6207 md_wakeup_thread(mddev->thread); 6208 } else { 6209 err = -EROFS; 6210 goto abort_unlock; 6211 } 6212 } 6213 6214 switch (cmd) 6215 { 6216 case ADD_NEW_DISK: 6217 { 6218 mdu_disk_info_t info; 6219 if (copy_from_user(&info, argp, sizeof(info))) 6220 err = -EFAULT; 6221 else 6222 err = add_new_disk(mddev, &info); 6223 goto done_unlock; 6224 } 6225 6226 case HOT_REMOVE_DISK: 6227 err = hot_remove_disk(mddev, new_decode_dev(arg)); 6228 goto done_unlock; 6229 6230 case HOT_ADD_DISK: 6231 err = hot_add_disk(mddev, new_decode_dev(arg)); 6232 goto done_unlock; 6233 6234 case SET_DISK_FAULTY: 6235 err = set_disk_faulty(mddev, new_decode_dev(arg)); 6236 goto done_unlock; 6237 6238 case RUN_ARRAY: 6239 err = do_md_run(mddev); 6240 goto done_unlock; 6241 6242 case SET_BITMAP_FILE: 6243 err = set_bitmap_file(mddev, (int)arg); 6244 goto done_unlock; 6245 6246 default: 6247 err = -EINVAL; 6248 goto abort_unlock; 6249 } 6250 6251 done_unlock: 6252 abort_unlock: 6253 if (mddev->hold_active == UNTIL_IOCTL && 6254 err != -EINVAL) 6255 mddev->hold_active = 0; 6256 mddev_unlock(mddev); 6257 6258 return err; 6259 done: 6260 if (err) 6261 MD_BUG(); 6262 abort: 6263 return err; 6264 } 6265 #ifdef CONFIG_COMPAT 6266 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, 6267 unsigned int cmd, unsigned long arg) 6268 { 6269 switch (cmd) { 6270 case HOT_REMOVE_DISK: 6271 case HOT_ADD_DISK: 6272 case SET_DISK_FAULTY: 6273 case SET_BITMAP_FILE: 6274 /* These take in integer arg, do not convert */ 6275 break; 6276 default: 6277 arg = (unsigned long)compat_ptr(arg); 6278 break; 6279 } 6280 6281 return md_ioctl(bdev, mode, cmd, arg); 6282 } 6283 #endif /* CONFIG_COMPAT */ 6284 6285 static int md_open(struct block_device *bdev, fmode_t mode) 6286 { 6287 /* 6288 * Succeed if we can lock the mddev, which confirms that 6289 * it isn't being stopped right now. 6290 */ 6291 mddev_t *mddev = mddev_find(bdev->bd_dev); 6292 int err; 6293 6294 if (mddev->gendisk != bdev->bd_disk) { 6295 /* we are racing with mddev_put which is discarding this 6296 * bd_disk. 6297 */ 6298 mddev_put(mddev); 6299 /* Wait until bdev->bd_disk is definitely gone */ 6300 flush_workqueue(md_misc_wq); 6301 /* Then retry the open from the top */ 6302 return -ERESTARTSYS; 6303 } 6304 BUG_ON(mddev != bdev->bd_disk->private_data); 6305 6306 if ((err = mutex_lock_interruptible(&mddev->open_mutex))) 6307 goto out; 6308 6309 err = 0; 6310 atomic_inc(&mddev->openers); 6311 mutex_unlock(&mddev->open_mutex); 6312 6313 check_disk_change(bdev); 6314 out: 6315 return err; 6316 } 6317 6318 static int md_release(struct gendisk *disk, fmode_t mode) 6319 { 6320 mddev_t *mddev = disk->private_data; 6321 6322 BUG_ON(!mddev); 6323 atomic_dec(&mddev->openers); 6324 mddev_put(mddev); 6325 6326 return 0; 6327 } 6328 6329 static int md_media_changed(struct gendisk *disk) 6330 { 6331 mddev_t *mddev = disk->private_data; 6332 6333 return mddev->changed; 6334 } 6335 6336 static int md_revalidate(struct gendisk *disk) 6337 { 6338 mddev_t *mddev = disk->private_data; 6339 6340 mddev->changed = 0; 6341 return 0; 6342 } 6343 static const struct block_device_operations md_fops = 6344 { 6345 .owner = THIS_MODULE, 6346 .open = md_open, 6347 .release = md_release, 6348 .ioctl = md_ioctl, 6349 #ifdef CONFIG_COMPAT 6350 .compat_ioctl = md_compat_ioctl, 6351 #endif 6352 .getgeo = md_getgeo, 6353 .media_changed = md_media_changed, 6354 .revalidate_disk= md_revalidate, 6355 }; 6356 6357 static int md_thread(void * arg) 6358 { 6359 mdk_thread_t *thread = arg; 6360 6361 /* 6362 * md_thread is a 'system-thread', it's priority should be very 6363 * high. We avoid resource deadlocks individually in each 6364 * raid personality. (RAID5 does preallocation) We also use RR and 6365 * the very same RT priority as kswapd, thus we will never get 6366 * into a priority inversion deadlock. 6367 * 6368 * we definitely have to have equal or higher priority than 6369 * bdflush, otherwise bdflush will deadlock if there are too 6370 * many dirty RAID5 blocks. 6371 */ 6372 6373 allow_signal(SIGKILL); 6374 while (!kthread_should_stop()) { 6375 6376 /* We need to wait INTERRUPTIBLE so that 6377 * we don't add to the load-average. 6378 * That means we need to be sure no signals are 6379 * pending 6380 */ 6381 if (signal_pending(current)) 6382 flush_signals(current); 6383 6384 wait_event_interruptible_timeout 6385 (thread->wqueue, 6386 test_bit(THREAD_WAKEUP, &thread->flags) 6387 || kthread_should_stop(), 6388 thread->timeout); 6389 6390 clear_bit(THREAD_WAKEUP, &thread->flags); 6391 if (!kthread_should_stop()) 6392 thread->run(thread->mddev); 6393 } 6394 6395 return 0; 6396 } 6397 6398 void md_wakeup_thread(mdk_thread_t *thread) 6399 { 6400 if (thread) { 6401 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); 6402 set_bit(THREAD_WAKEUP, &thread->flags); 6403 wake_up(&thread->wqueue); 6404 } 6405 } 6406 6407 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, 6408 const char *name) 6409 { 6410 mdk_thread_t *thread; 6411 6412 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); 6413 if (!thread) 6414 return NULL; 6415 6416 init_waitqueue_head(&thread->wqueue); 6417 6418 thread->run = run; 6419 thread->mddev = mddev; 6420 thread->timeout = MAX_SCHEDULE_TIMEOUT; 6421 thread->tsk = kthread_run(md_thread, thread, 6422 "%s_%s", 6423 mdname(thread->mddev), 6424 name ?: mddev->pers->name); 6425 if (IS_ERR(thread->tsk)) { 6426 kfree(thread); 6427 return NULL; 6428 } 6429 return thread; 6430 } 6431 6432 void md_unregister_thread(mdk_thread_t *thread) 6433 { 6434 if (!thread) 6435 return; 6436 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); 6437 6438 kthread_stop(thread->tsk); 6439 kfree(thread); 6440 } 6441 6442 void md_error(mddev_t *mddev, mdk_rdev_t *rdev) 6443 { 6444 if (!mddev) { 6445 MD_BUG(); 6446 return; 6447 } 6448 6449 if (!rdev || test_bit(Faulty, &rdev->flags)) 6450 return; 6451 6452 if (!mddev->pers || !mddev->pers->error_handler) 6453 return; 6454 mddev->pers->error_handler(mddev,rdev); 6455 if (mddev->degraded) 6456 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6457 sysfs_notify_dirent_safe(rdev->sysfs_state); 6458 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6459 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6460 md_wakeup_thread(mddev->thread); 6461 if (mddev->event_work.func) 6462 queue_work(md_misc_wq, &mddev->event_work); 6463 md_new_event_inintr(mddev); 6464 } 6465 6466 /* seq_file implementation /proc/mdstat */ 6467 6468 static void status_unused(struct seq_file *seq) 6469 { 6470 int i = 0; 6471 mdk_rdev_t *rdev; 6472 6473 seq_printf(seq, "unused devices: "); 6474 6475 list_for_each_entry(rdev, &pending_raid_disks, same_set) { 6476 char b[BDEVNAME_SIZE]; 6477 i++; 6478 seq_printf(seq, "%s ", 6479 bdevname(rdev->bdev,b)); 6480 } 6481 if (!i) 6482 seq_printf(seq, "<none>"); 6483 6484 seq_printf(seq, "\n"); 6485 } 6486 6487 6488 static void status_resync(struct seq_file *seq, mddev_t * mddev) 6489 { 6490 sector_t max_sectors, resync, res; 6491 unsigned long dt, db; 6492 sector_t rt; 6493 int scale; 6494 unsigned int per_milli; 6495 6496 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); 6497 6498 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 6499 max_sectors = mddev->resync_max_sectors; 6500 else 6501 max_sectors = mddev->dev_sectors; 6502 6503 /* 6504 * Should not happen. 6505 */ 6506 if (!max_sectors) { 6507 MD_BUG(); 6508 return; 6509 } 6510 /* Pick 'scale' such that (resync>>scale)*1000 will fit 6511 * in a sector_t, and (max_sectors>>scale) will fit in a 6512 * u32, as those are the requirements for sector_div. 6513 * Thus 'scale' must be at least 10 6514 */ 6515 scale = 10; 6516 if (sizeof(sector_t) > sizeof(unsigned long)) { 6517 while ( max_sectors/2 > (1ULL<<(scale+32))) 6518 scale++; 6519 } 6520 res = (resync>>scale)*1000; 6521 sector_div(res, (u32)((max_sectors>>scale)+1)); 6522 6523 per_milli = res; 6524 { 6525 int i, x = per_milli/50, y = 20-x; 6526 seq_printf(seq, "["); 6527 for (i = 0; i < x; i++) 6528 seq_printf(seq, "="); 6529 seq_printf(seq, ">"); 6530 for (i = 0; i < y; i++) 6531 seq_printf(seq, "."); 6532 seq_printf(seq, "] "); 6533 } 6534 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", 6535 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? 6536 "reshape" : 6537 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? 6538 "check" : 6539 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 6540 "resync" : "recovery"))), 6541 per_milli/10, per_milli % 10, 6542 (unsigned long long) resync/2, 6543 (unsigned long long) max_sectors/2); 6544 6545 /* 6546 * dt: time from mark until now 6547 * db: blocks written from mark until now 6548 * rt: remaining time 6549 * 6550 * rt is a sector_t, so could be 32bit or 64bit. 6551 * So we divide before multiply in case it is 32bit and close 6552 * to the limit. 6553 * We scale the divisor (db) by 32 to avoid losing precision 6554 * near the end of resync when the number of remaining sectors 6555 * is close to 'db'. 6556 * We then divide rt by 32 after multiplying by db to compensate. 6557 * The '+1' avoids division by zero if db is very small. 6558 */ 6559 dt = ((jiffies - mddev->resync_mark) / HZ); 6560 if (!dt) dt++; 6561 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) 6562 - mddev->resync_mark_cnt; 6563 6564 rt = max_sectors - resync; /* number of remaining sectors */ 6565 sector_div(rt, db/32+1); 6566 rt *= dt; 6567 rt >>= 5; 6568 6569 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, 6570 ((unsigned long)rt % 60)/6); 6571 6572 seq_printf(seq, " speed=%ldK/sec", db/2/dt); 6573 } 6574 6575 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 6576 { 6577 struct list_head *tmp; 6578 loff_t l = *pos; 6579 mddev_t *mddev; 6580 6581 if (l >= 0x10000) 6582 return NULL; 6583 if (!l--) 6584 /* header */ 6585 return (void*)1; 6586 6587 spin_lock(&all_mddevs_lock); 6588 list_for_each(tmp,&all_mddevs) 6589 if (!l--) { 6590 mddev = list_entry(tmp, mddev_t, all_mddevs); 6591 mddev_get(mddev); 6592 spin_unlock(&all_mddevs_lock); 6593 return mddev; 6594 } 6595 spin_unlock(&all_mddevs_lock); 6596 if (!l--) 6597 return (void*)2;/* tail */ 6598 return NULL; 6599 } 6600 6601 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 6602 { 6603 struct list_head *tmp; 6604 mddev_t *next_mddev, *mddev = v; 6605 6606 ++*pos; 6607 if (v == (void*)2) 6608 return NULL; 6609 6610 spin_lock(&all_mddevs_lock); 6611 if (v == (void*)1) 6612 tmp = all_mddevs.next; 6613 else 6614 tmp = mddev->all_mddevs.next; 6615 if (tmp != &all_mddevs) 6616 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); 6617 else { 6618 next_mddev = (void*)2; 6619 *pos = 0x10000; 6620 } 6621 spin_unlock(&all_mddevs_lock); 6622 6623 if (v != (void*)1) 6624 mddev_put(mddev); 6625 return next_mddev; 6626 6627 } 6628 6629 static void md_seq_stop(struct seq_file *seq, void *v) 6630 { 6631 mddev_t *mddev = v; 6632 6633 if (mddev && v != (void*)1 && v != (void*)2) 6634 mddev_put(mddev); 6635 } 6636 6637 static int md_seq_show(struct seq_file *seq, void *v) 6638 { 6639 mddev_t *mddev = v; 6640 sector_t sectors; 6641 mdk_rdev_t *rdev; 6642 struct bitmap *bitmap; 6643 6644 if (v == (void*)1) { 6645 struct mdk_personality *pers; 6646 seq_printf(seq, "Personalities : "); 6647 spin_lock(&pers_lock); 6648 list_for_each_entry(pers, &pers_list, list) 6649 seq_printf(seq, "[%s] ", pers->name); 6650 6651 spin_unlock(&pers_lock); 6652 seq_printf(seq, "\n"); 6653 seq->poll_event = atomic_read(&md_event_count); 6654 return 0; 6655 } 6656 if (v == (void*)2) { 6657 status_unused(seq); 6658 return 0; 6659 } 6660 6661 if (mddev_lock(mddev) < 0) 6662 return -EINTR; 6663 6664 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 6665 seq_printf(seq, "%s : %sactive", mdname(mddev), 6666 mddev->pers ? "" : "in"); 6667 if (mddev->pers) { 6668 if (mddev->ro==1) 6669 seq_printf(seq, " (read-only)"); 6670 if (mddev->ro==2) 6671 seq_printf(seq, " (auto-read-only)"); 6672 seq_printf(seq, " %s", mddev->pers->name); 6673 } 6674 6675 sectors = 0; 6676 list_for_each_entry(rdev, &mddev->disks, same_set) { 6677 char b[BDEVNAME_SIZE]; 6678 seq_printf(seq, " %s[%d]", 6679 bdevname(rdev->bdev,b), rdev->desc_nr); 6680 if (test_bit(WriteMostly, &rdev->flags)) 6681 seq_printf(seq, "(W)"); 6682 if (test_bit(Faulty, &rdev->flags)) { 6683 seq_printf(seq, "(F)"); 6684 continue; 6685 } else if (rdev->raid_disk < 0) 6686 seq_printf(seq, "(S)"); /* spare */ 6687 sectors += rdev->sectors; 6688 } 6689 6690 if (!list_empty(&mddev->disks)) { 6691 if (mddev->pers) 6692 seq_printf(seq, "\n %llu blocks", 6693 (unsigned long long) 6694 mddev->array_sectors / 2); 6695 else 6696 seq_printf(seq, "\n %llu blocks", 6697 (unsigned long long)sectors / 2); 6698 } 6699 if (mddev->persistent) { 6700 if (mddev->major_version != 0 || 6701 mddev->minor_version != 90) { 6702 seq_printf(seq," super %d.%d", 6703 mddev->major_version, 6704 mddev->minor_version); 6705 } 6706 } else if (mddev->external) 6707 seq_printf(seq, " super external:%s", 6708 mddev->metadata_type); 6709 else 6710 seq_printf(seq, " super non-persistent"); 6711 6712 if (mddev->pers) { 6713 mddev->pers->status(seq, mddev); 6714 seq_printf(seq, "\n "); 6715 if (mddev->pers->sync_request) { 6716 if (mddev->curr_resync > 2) { 6717 status_resync(seq, mddev); 6718 seq_printf(seq, "\n "); 6719 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 6720 seq_printf(seq, "\tresync=DELAYED\n "); 6721 else if (mddev->recovery_cp < MaxSector) 6722 seq_printf(seq, "\tresync=PENDING\n "); 6723 } 6724 } else 6725 seq_printf(seq, "\n "); 6726 6727 if ((bitmap = mddev->bitmap)) { 6728 unsigned long chunk_kb; 6729 unsigned long flags; 6730 spin_lock_irqsave(&bitmap->lock, flags); 6731 chunk_kb = mddev->bitmap_info.chunksize >> 10; 6732 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 6733 "%lu%s chunk", 6734 bitmap->pages - bitmap->missing_pages, 6735 bitmap->pages, 6736 (bitmap->pages - bitmap->missing_pages) 6737 << (PAGE_SHIFT - 10), 6738 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, 6739 chunk_kb ? "KB" : "B"); 6740 if (bitmap->file) { 6741 seq_printf(seq, ", file: "); 6742 seq_path(seq, &bitmap->file->f_path, " \t\n"); 6743 } 6744 6745 seq_printf(seq, "\n"); 6746 spin_unlock_irqrestore(&bitmap->lock, flags); 6747 } 6748 6749 seq_printf(seq, "\n"); 6750 } 6751 mddev_unlock(mddev); 6752 6753 return 0; 6754 } 6755 6756 static const struct seq_operations md_seq_ops = { 6757 .start = md_seq_start, 6758 .next = md_seq_next, 6759 .stop = md_seq_stop, 6760 .show = md_seq_show, 6761 }; 6762 6763 static int md_seq_open(struct inode *inode, struct file *file) 6764 { 6765 struct seq_file *seq; 6766 int error; 6767 6768 error = seq_open(file, &md_seq_ops); 6769 if (error) 6770 return error; 6771 6772 seq = file->private_data; 6773 seq->poll_event = atomic_read(&md_event_count); 6774 return error; 6775 } 6776 6777 static unsigned int mdstat_poll(struct file *filp, poll_table *wait) 6778 { 6779 struct seq_file *seq = filp->private_data; 6780 int mask; 6781 6782 poll_wait(filp, &md_event_waiters, wait); 6783 6784 /* always allow read */ 6785 mask = POLLIN | POLLRDNORM; 6786 6787 if (seq->poll_event != atomic_read(&md_event_count)) 6788 mask |= POLLERR | POLLPRI; 6789 return mask; 6790 } 6791 6792 static const struct file_operations md_seq_fops = { 6793 .owner = THIS_MODULE, 6794 .open = md_seq_open, 6795 .read = seq_read, 6796 .llseek = seq_lseek, 6797 .release = seq_release_private, 6798 .poll = mdstat_poll, 6799 }; 6800 6801 int register_md_personality(struct mdk_personality *p) 6802 { 6803 spin_lock(&pers_lock); 6804 list_add_tail(&p->list, &pers_list); 6805 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); 6806 spin_unlock(&pers_lock); 6807 return 0; 6808 } 6809 6810 int unregister_md_personality(struct mdk_personality *p) 6811 { 6812 printk(KERN_INFO "md: %s personality unregistered\n", p->name); 6813 spin_lock(&pers_lock); 6814 list_del_init(&p->list); 6815 spin_unlock(&pers_lock); 6816 return 0; 6817 } 6818 6819 static int is_mddev_idle(mddev_t *mddev, int init) 6820 { 6821 mdk_rdev_t * rdev; 6822 int idle; 6823 int curr_events; 6824 6825 idle = 1; 6826 rcu_read_lock(); 6827 rdev_for_each_rcu(rdev, mddev) { 6828 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 6829 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 6830 (int)part_stat_read(&disk->part0, sectors[1]) - 6831 atomic_read(&disk->sync_io); 6832 /* sync IO will cause sync_io to increase before the disk_stats 6833 * as sync_io is counted when a request starts, and 6834 * disk_stats is counted when it completes. 6835 * So resync activity will cause curr_events to be smaller than 6836 * when there was no such activity. 6837 * non-sync IO will cause disk_stat to increase without 6838 * increasing sync_io so curr_events will (eventually) 6839 * be larger than it was before. Once it becomes 6840 * substantially larger, the test below will cause 6841 * the array to appear non-idle, and resync will slow 6842 * down. 6843 * If there is a lot of outstanding resync activity when 6844 * we set last_event to curr_events, then all that activity 6845 * completing might cause the array to appear non-idle 6846 * and resync will be slowed down even though there might 6847 * not have been non-resync activity. This will only 6848 * happen once though. 'last_events' will soon reflect 6849 * the state where there is little or no outstanding 6850 * resync requests, and further resync activity will 6851 * always make curr_events less than last_events. 6852 * 6853 */ 6854 if (init || curr_events - rdev->last_events > 64) { 6855 rdev->last_events = curr_events; 6856 idle = 0; 6857 } 6858 } 6859 rcu_read_unlock(); 6860 return idle; 6861 } 6862 6863 void md_done_sync(mddev_t *mddev, int blocks, int ok) 6864 { 6865 /* another "blocks" (512byte) blocks have been synced */ 6866 atomic_sub(blocks, &mddev->recovery_active); 6867 wake_up(&mddev->recovery_wait); 6868 if (!ok) { 6869 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6870 md_wakeup_thread(mddev->thread); 6871 // stop recovery, signal do_sync .... 6872 } 6873 } 6874 6875 6876 /* md_write_start(mddev, bi) 6877 * If we need to update some array metadata (e.g. 'active' flag 6878 * in superblock) before writing, schedule a superblock update 6879 * and wait for it to complete. 6880 */ 6881 void md_write_start(mddev_t *mddev, struct bio *bi) 6882 { 6883 int did_change = 0; 6884 if (bio_data_dir(bi) != WRITE) 6885 return; 6886 6887 BUG_ON(mddev->ro == 1); 6888 if (mddev->ro == 2) { 6889 /* need to switch to read/write */ 6890 mddev->ro = 0; 6891 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6892 md_wakeup_thread(mddev->thread); 6893 md_wakeup_thread(mddev->sync_thread); 6894 did_change = 1; 6895 } 6896 atomic_inc(&mddev->writes_pending); 6897 if (mddev->safemode == 1) 6898 mddev->safemode = 0; 6899 if (mddev->in_sync) { 6900 spin_lock_irq(&mddev->write_lock); 6901 if (mddev->in_sync) { 6902 mddev->in_sync = 0; 6903 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6904 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6905 md_wakeup_thread(mddev->thread); 6906 did_change = 1; 6907 } 6908 spin_unlock_irq(&mddev->write_lock); 6909 } 6910 if (did_change) 6911 sysfs_notify_dirent_safe(mddev->sysfs_state); 6912 wait_event(mddev->sb_wait, 6913 !test_bit(MD_CHANGE_PENDING, &mddev->flags)); 6914 } 6915 6916 void md_write_end(mddev_t *mddev) 6917 { 6918 if (atomic_dec_and_test(&mddev->writes_pending)) { 6919 if (mddev->safemode == 2) 6920 md_wakeup_thread(mddev->thread); 6921 else if (mddev->safemode_delay) 6922 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 6923 } 6924 } 6925 6926 /* md_allow_write(mddev) 6927 * Calling this ensures that the array is marked 'active' so that writes 6928 * may proceed without blocking. It is important to call this before 6929 * attempting a GFP_KERNEL allocation while holding the mddev lock. 6930 * Must be called with mddev_lock held. 6931 * 6932 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock 6933 * is dropped, so return -EAGAIN after notifying userspace. 6934 */ 6935 int md_allow_write(mddev_t *mddev) 6936 { 6937 if (!mddev->pers) 6938 return 0; 6939 if (mddev->ro) 6940 return 0; 6941 if (!mddev->pers->sync_request) 6942 return 0; 6943 6944 spin_lock_irq(&mddev->write_lock); 6945 if (mddev->in_sync) { 6946 mddev->in_sync = 0; 6947 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6948 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6949 if (mddev->safemode_delay && 6950 mddev->safemode == 0) 6951 mddev->safemode = 1; 6952 spin_unlock_irq(&mddev->write_lock); 6953 md_update_sb(mddev, 0); 6954 sysfs_notify_dirent_safe(mddev->sysfs_state); 6955 } else 6956 spin_unlock_irq(&mddev->write_lock); 6957 6958 if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 6959 return -EAGAIN; 6960 else 6961 return 0; 6962 } 6963 EXPORT_SYMBOL_GPL(md_allow_write); 6964 6965 #define SYNC_MARKS 10 6966 #define SYNC_MARK_STEP (3*HZ) 6967 void md_do_sync(mddev_t *mddev) 6968 { 6969 mddev_t *mddev2; 6970 unsigned int currspeed = 0, 6971 window; 6972 sector_t max_sectors,j, io_sectors; 6973 unsigned long mark[SYNC_MARKS]; 6974 sector_t mark_cnt[SYNC_MARKS]; 6975 int last_mark,m; 6976 struct list_head *tmp; 6977 sector_t last_check; 6978 int skipped = 0; 6979 mdk_rdev_t *rdev; 6980 char *desc; 6981 6982 /* just incase thread restarts... */ 6983 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 6984 return; 6985 if (mddev->ro) /* never try to sync a read-only array */ 6986 return; 6987 6988 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6989 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 6990 desc = "data-check"; 6991 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6992 desc = "requested-resync"; 6993 else 6994 desc = "resync"; 6995 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6996 desc = "reshape"; 6997 else 6998 desc = "recovery"; 6999 7000 /* we overload curr_resync somewhat here. 7001 * 0 == not engaged in resync at all 7002 * 2 == checking that there is no conflict with another sync 7003 * 1 == like 2, but have yielded to allow conflicting resync to 7004 * commense 7005 * other == active in resync - this many blocks 7006 * 7007 * Before starting a resync we must have set curr_resync to 7008 * 2, and then checked that every "conflicting" array has curr_resync 7009 * less than ours. When we find one that is the same or higher 7010 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 7011 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 7012 * This will mean we have to start checking from the beginning again. 7013 * 7014 */ 7015 7016 do { 7017 mddev->curr_resync = 2; 7018 7019 try_again: 7020 if (kthread_should_stop()) 7021 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7022 7023 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7024 goto skip; 7025 for_each_mddev(mddev2, tmp) { 7026 if (mddev2 == mddev) 7027 continue; 7028 if (!mddev->parallel_resync 7029 && mddev2->curr_resync 7030 && match_mddev_units(mddev, mddev2)) { 7031 DEFINE_WAIT(wq); 7032 if (mddev < mddev2 && mddev->curr_resync == 2) { 7033 /* arbitrarily yield */ 7034 mddev->curr_resync = 1; 7035 wake_up(&resync_wait); 7036 } 7037 if (mddev > mddev2 && mddev->curr_resync == 1) 7038 /* no need to wait here, we can wait the next 7039 * time 'round when curr_resync == 2 7040 */ 7041 continue; 7042 /* We need to wait 'interruptible' so as not to 7043 * contribute to the load average, and not to 7044 * be caught by 'softlockup' 7045 */ 7046 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 7047 if (!kthread_should_stop() && 7048 mddev2->curr_resync >= mddev->curr_resync) { 7049 printk(KERN_INFO "md: delaying %s of %s" 7050 " until %s has finished (they" 7051 " share one or more physical units)\n", 7052 desc, mdname(mddev), mdname(mddev2)); 7053 mddev_put(mddev2); 7054 if (signal_pending(current)) 7055 flush_signals(current); 7056 schedule(); 7057 finish_wait(&resync_wait, &wq); 7058 goto try_again; 7059 } 7060 finish_wait(&resync_wait, &wq); 7061 } 7062 } 7063 } while (mddev->curr_resync < 2); 7064 7065 j = 0; 7066 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7067 /* resync follows the size requested by the personality, 7068 * which defaults to physical size, but can be virtual size 7069 */ 7070 max_sectors = mddev->resync_max_sectors; 7071 mddev->resync_mismatches = 0; 7072 /* we don't use the checkpoint if there's a bitmap */ 7073 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7074 j = mddev->resync_min; 7075 else if (!mddev->bitmap) 7076 j = mddev->recovery_cp; 7077 7078 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 7079 max_sectors = mddev->dev_sectors; 7080 else { 7081 /* recovery follows the physical size of devices */ 7082 max_sectors = mddev->dev_sectors; 7083 j = MaxSector; 7084 rcu_read_lock(); 7085 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 7086 if (rdev->raid_disk >= 0 && 7087 !test_bit(Faulty, &rdev->flags) && 7088 !test_bit(In_sync, &rdev->flags) && 7089 rdev->recovery_offset < j) 7090 j = rdev->recovery_offset; 7091 rcu_read_unlock(); 7092 } 7093 7094 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); 7095 printk(KERN_INFO "md: minimum _guaranteed_ speed:" 7096 " %d KB/sec/disk.\n", speed_min(mddev)); 7097 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 7098 "(but not more than %d KB/sec) for %s.\n", 7099 speed_max(mddev), desc); 7100 7101 is_mddev_idle(mddev, 1); /* this initializes IO event counters */ 7102 7103 io_sectors = 0; 7104 for (m = 0; m < SYNC_MARKS; m++) { 7105 mark[m] = jiffies; 7106 mark_cnt[m] = io_sectors; 7107 } 7108 last_mark = 0; 7109 mddev->resync_mark = mark[last_mark]; 7110 mddev->resync_mark_cnt = mark_cnt[last_mark]; 7111 7112 /* 7113 * Tune reconstruction: 7114 */ 7115 window = 32*(PAGE_SIZE/512); 7116 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", 7117 window/2, (unsigned long long)max_sectors/2); 7118 7119 atomic_set(&mddev->recovery_active, 0); 7120 last_check = 0; 7121 7122 if (j>2) { 7123 printk(KERN_INFO 7124 "md: resuming %s of %s from checkpoint.\n", 7125 desc, mdname(mddev)); 7126 mddev->curr_resync = j; 7127 } 7128 mddev->curr_resync_completed = j; 7129 7130 while (j < max_sectors) { 7131 sector_t sectors; 7132 7133 skipped = 0; 7134 7135 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7136 ((mddev->curr_resync > mddev->curr_resync_completed && 7137 (mddev->curr_resync - mddev->curr_resync_completed) 7138 > (max_sectors >> 4)) || 7139 (j - mddev->curr_resync_completed)*2 7140 >= mddev->resync_max - mddev->curr_resync_completed 7141 )) { 7142 /* time to update curr_resync_completed */ 7143 wait_event(mddev->recovery_wait, 7144 atomic_read(&mddev->recovery_active) == 0); 7145 mddev->curr_resync_completed = j; 7146 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7147 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 7148 } 7149 7150 while (j >= mddev->resync_max && !kthread_should_stop()) { 7151 /* As this condition is controlled by user-space, 7152 * we can block indefinitely, so use '_interruptible' 7153 * to avoid triggering warnings. 7154 */ 7155 flush_signals(current); /* just in case */ 7156 wait_event_interruptible(mddev->recovery_wait, 7157 mddev->resync_max > j 7158 || kthread_should_stop()); 7159 } 7160 7161 if (kthread_should_stop()) 7162 goto interrupted; 7163 7164 sectors = mddev->pers->sync_request(mddev, j, &skipped, 7165 currspeed < speed_min(mddev)); 7166 if (sectors == 0) { 7167 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7168 goto out; 7169 } 7170 7171 if (!skipped) { /* actual IO requested */ 7172 io_sectors += sectors; 7173 atomic_add(sectors, &mddev->recovery_active); 7174 } 7175 7176 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7177 break; 7178 7179 j += sectors; 7180 if (j>1) mddev->curr_resync = j; 7181 mddev->curr_mark_cnt = io_sectors; 7182 if (last_check == 0) 7183 /* this is the earliest that rebuild will be 7184 * visible in /proc/mdstat 7185 */ 7186 md_new_event(mddev); 7187 7188 if (last_check + window > io_sectors || j == max_sectors) 7189 continue; 7190 7191 last_check = io_sectors; 7192 repeat: 7193 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 7194 /* step marks */ 7195 int next = (last_mark+1) % SYNC_MARKS; 7196 7197 mddev->resync_mark = mark[next]; 7198 mddev->resync_mark_cnt = mark_cnt[next]; 7199 mark[next] = jiffies; 7200 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 7201 last_mark = next; 7202 } 7203 7204 7205 if (kthread_should_stop()) 7206 goto interrupted; 7207 7208 7209 /* 7210 * this loop exits only if either when we are slower than 7211 * the 'hard' speed limit, or the system was IO-idle for 7212 * a jiffy. 7213 * the system might be non-idle CPU-wise, but we only care 7214 * about not overloading the IO subsystem. (things like an 7215 * e2fsck being done on the RAID array should execute fast) 7216 */ 7217 cond_resched(); 7218 7219 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 7220 /((jiffies-mddev->resync_mark)/HZ +1) +1; 7221 7222 if (currspeed > speed_min(mddev)) { 7223 if ((currspeed > speed_max(mddev)) || 7224 !is_mddev_idle(mddev, 0)) { 7225 msleep(500); 7226 goto repeat; 7227 } 7228 } 7229 } 7230 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); 7231 /* 7232 * this also signals 'finished resyncing' to md_stop 7233 */ 7234 out: 7235 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 7236 7237 /* tell personality that we are finished */ 7238 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 7239 7240 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && 7241 mddev->curr_resync > 2) { 7242 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 7243 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 7244 if (mddev->curr_resync >= mddev->recovery_cp) { 7245 printk(KERN_INFO 7246 "md: checkpointing %s of %s.\n", 7247 desc, mdname(mddev)); 7248 mddev->recovery_cp = mddev->curr_resync; 7249 } 7250 } else 7251 mddev->recovery_cp = MaxSector; 7252 } else { 7253 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 7254 mddev->curr_resync = MaxSector; 7255 rcu_read_lock(); 7256 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 7257 if (rdev->raid_disk >= 0 && 7258 mddev->delta_disks >= 0 && 7259 !test_bit(Faulty, &rdev->flags) && 7260 !test_bit(In_sync, &rdev->flags) && 7261 rdev->recovery_offset < mddev->curr_resync) 7262 rdev->recovery_offset = mddev->curr_resync; 7263 rcu_read_unlock(); 7264 } 7265 } 7266 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7267 7268 skip: 7269 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 7270 /* We completed so min/max setting can be forgotten if used. */ 7271 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7272 mddev->resync_min = 0; 7273 mddev->resync_max = MaxSector; 7274 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 7275 mddev->resync_min = mddev->curr_resync_completed; 7276 mddev->curr_resync = 0; 7277 wake_up(&resync_wait); 7278 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 7279 md_wakeup_thread(mddev->thread); 7280 return; 7281 7282 interrupted: 7283 /* 7284 * got a signal, exit. 7285 */ 7286 printk(KERN_INFO 7287 "md: md_do_sync() got signal ... exiting\n"); 7288 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 7289 goto out; 7290 7291 } 7292 EXPORT_SYMBOL_GPL(md_do_sync); 7293 7294 static int remove_and_add_spares(mddev_t *mddev) 7295 { 7296 mdk_rdev_t *rdev; 7297 int spares = 0; 7298 7299 mddev->curr_resync_completed = 0; 7300 7301 list_for_each_entry(rdev, &mddev->disks, same_set) 7302 if (rdev->raid_disk >= 0 && 7303 !test_bit(Blocked, &rdev->flags) && 7304 (test_bit(Faulty, &rdev->flags) || 7305 ! test_bit(In_sync, &rdev->flags)) && 7306 atomic_read(&rdev->nr_pending)==0) { 7307 if (mddev->pers->hot_remove_disk( 7308 mddev, rdev->raid_disk)==0) { 7309 sysfs_unlink_rdev(mddev, rdev); 7310 rdev->raid_disk = -1; 7311 } 7312 } 7313 7314 if (mddev->degraded) { 7315 list_for_each_entry(rdev, &mddev->disks, same_set) { 7316 if (rdev->raid_disk >= 0 && 7317 !test_bit(In_sync, &rdev->flags) && 7318 !test_bit(Faulty, &rdev->flags)) 7319 spares++; 7320 if (rdev->raid_disk < 0 7321 && !test_bit(Faulty, &rdev->flags)) { 7322 rdev->recovery_offset = 0; 7323 if (mddev->pers-> 7324 hot_add_disk(mddev, rdev) == 0) { 7325 if (sysfs_link_rdev(mddev, rdev)) 7326 /* failure here is OK */; 7327 spares++; 7328 md_new_event(mddev); 7329 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7330 } else 7331 break; 7332 } 7333 } 7334 } 7335 return spares; 7336 } 7337 7338 static void reap_sync_thread(mddev_t *mddev) 7339 { 7340 mdk_rdev_t *rdev; 7341 7342 /* resync has finished, collect result */ 7343 md_unregister_thread(mddev->sync_thread); 7344 mddev->sync_thread = NULL; 7345 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && 7346 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 7347 /* success...*/ 7348 /* activate any spares */ 7349 if (mddev->pers->spare_active(mddev)) 7350 sysfs_notify(&mddev->kobj, NULL, 7351 "degraded"); 7352 } 7353 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7354 mddev->pers->finish_reshape) 7355 mddev->pers->finish_reshape(mddev); 7356 md_update_sb(mddev, 1); 7357 7358 /* if array is no-longer degraded, then any saved_raid_disk 7359 * information must be scrapped 7360 */ 7361 if (!mddev->degraded) 7362 list_for_each_entry(rdev, &mddev->disks, same_set) 7363 rdev->saved_raid_disk = -1; 7364 7365 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7366 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7367 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7368 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7369 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7370 /* flag recovery needed just to double check */ 7371 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7372 sysfs_notify_dirent_safe(mddev->sysfs_action); 7373 md_new_event(mddev); 7374 if (mddev->event_work.func) 7375 queue_work(md_misc_wq, &mddev->event_work); 7376 } 7377 7378 /* 7379 * This routine is regularly called by all per-raid-array threads to 7380 * deal with generic issues like resync and super-block update. 7381 * Raid personalities that don't have a thread (linear/raid0) do not 7382 * need this as they never do any recovery or update the superblock. 7383 * 7384 * It does not do any resync itself, but rather "forks" off other threads 7385 * to do that as needed. 7386 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 7387 * "->recovery" and create a thread at ->sync_thread. 7388 * When the thread finishes it sets MD_RECOVERY_DONE 7389 * and wakeups up this thread which will reap the thread and finish up. 7390 * This thread also removes any faulty devices (with nr_pending == 0). 7391 * 7392 * The overall approach is: 7393 * 1/ if the superblock needs updating, update it. 7394 * 2/ If a recovery thread is running, don't do anything else. 7395 * 3/ If recovery has finished, clean up, possibly marking spares active. 7396 * 4/ If there are any faulty devices, remove them. 7397 * 5/ If array is degraded, try to add spares devices 7398 * 6/ If array has spares or is not in-sync, start a resync thread. 7399 */ 7400 void md_check_recovery(mddev_t *mddev) 7401 { 7402 if (mddev->suspended) 7403 return; 7404 7405 if (mddev->bitmap) 7406 bitmap_daemon_work(mddev); 7407 7408 if (signal_pending(current)) { 7409 if (mddev->pers->sync_request && !mddev->external) { 7410 printk(KERN_INFO "md: %s in immediate safe mode\n", 7411 mdname(mddev)); 7412 mddev->safemode = 2; 7413 } 7414 flush_signals(current); 7415 } 7416 7417 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 7418 return; 7419 if ( ! ( 7420 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || 7421 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 7422 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 7423 (mddev->external == 0 && mddev->safemode == 1) || 7424 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 7425 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 7426 )) 7427 return; 7428 7429 if (mddev_trylock(mddev)) { 7430 int spares = 0; 7431 7432 if (mddev->ro) { 7433 /* Only thing we do on a ro array is remove 7434 * failed devices. 7435 */ 7436 mdk_rdev_t *rdev; 7437 list_for_each_entry(rdev, &mddev->disks, same_set) 7438 if (rdev->raid_disk >= 0 && 7439 !test_bit(Blocked, &rdev->flags) && 7440 test_bit(Faulty, &rdev->flags) && 7441 atomic_read(&rdev->nr_pending)==0) { 7442 if (mddev->pers->hot_remove_disk( 7443 mddev, rdev->raid_disk)==0) { 7444 sysfs_unlink_rdev(mddev, rdev); 7445 rdev->raid_disk = -1; 7446 } 7447 } 7448 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7449 goto unlock; 7450 } 7451 7452 if (!mddev->external) { 7453 int did_change = 0; 7454 spin_lock_irq(&mddev->write_lock); 7455 if (mddev->safemode && 7456 !atomic_read(&mddev->writes_pending) && 7457 !mddev->in_sync && 7458 mddev->recovery_cp == MaxSector) { 7459 mddev->in_sync = 1; 7460 did_change = 1; 7461 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7462 } 7463 if (mddev->safemode == 1) 7464 mddev->safemode = 0; 7465 spin_unlock_irq(&mddev->write_lock); 7466 if (did_change) 7467 sysfs_notify_dirent_safe(mddev->sysfs_state); 7468 } 7469 7470 if (mddev->flags) 7471 md_update_sb(mddev, 0); 7472 7473 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 7474 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 7475 /* resync/recovery still happening */ 7476 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7477 goto unlock; 7478 } 7479 if (mddev->sync_thread) { 7480 reap_sync_thread(mddev); 7481 goto unlock; 7482 } 7483 /* Set RUNNING before clearing NEEDED to avoid 7484 * any transients in the value of "sync_action". 7485 */ 7486 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7487 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7488 /* Clear some bits that don't mean anything, but 7489 * might be left set 7490 */ 7491 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 7492 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 7493 7494 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 7495 goto unlock; 7496 /* no recovery is running. 7497 * remove any failed drives, then 7498 * add spares if possible. 7499 * Spare are also removed and re-added, to allow 7500 * the personality to fail the re-add. 7501 */ 7502 7503 if (mddev->reshape_position != MaxSector) { 7504 if (mddev->pers->check_reshape == NULL || 7505 mddev->pers->check_reshape(mddev) != 0) 7506 /* Cannot proceed */ 7507 goto unlock; 7508 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7509 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7510 } else if ((spares = remove_and_add_spares(mddev))) { 7511 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7512 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7513 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7514 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7515 } else if (mddev->recovery_cp < MaxSector) { 7516 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7517 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7518 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 7519 /* nothing to be done ... */ 7520 goto unlock; 7521 7522 if (mddev->pers->sync_request) { 7523 if (spares && mddev->bitmap && ! mddev->bitmap->file) { 7524 /* We are adding a device or devices to an array 7525 * which has the bitmap stored on all devices. 7526 * So make sure all bitmap pages get written 7527 */ 7528 bitmap_write_all(mddev->bitmap); 7529 } 7530 mddev->sync_thread = md_register_thread(md_do_sync, 7531 mddev, 7532 "resync"); 7533 if (!mddev->sync_thread) { 7534 printk(KERN_ERR "%s: could not start resync" 7535 " thread...\n", 7536 mdname(mddev)); 7537 /* leave the spares where they are, it shouldn't hurt */ 7538 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7539 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7540 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7541 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7542 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7543 } else 7544 md_wakeup_thread(mddev->sync_thread); 7545 sysfs_notify_dirent_safe(mddev->sysfs_action); 7546 md_new_event(mddev); 7547 } 7548 unlock: 7549 if (!mddev->sync_thread) { 7550 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7551 if (test_and_clear_bit(MD_RECOVERY_RECOVER, 7552 &mddev->recovery)) 7553 if (mddev->sysfs_action) 7554 sysfs_notify_dirent_safe(mddev->sysfs_action); 7555 } 7556 mddev_unlock(mddev); 7557 } 7558 } 7559 7560 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 7561 { 7562 sysfs_notify_dirent_safe(rdev->sysfs_state); 7563 wait_event_timeout(rdev->blocked_wait, 7564 !test_bit(Blocked, &rdev->flags) && 7565 !test_bit(BlockedBadBlocks, &rdev->flags), 7566 msecs_to_jiffies(5000)); 7567 rdev_dec_pending(rdev, mddev); 7568 } 7569 EXPORT_SYMBOL(md_wait_for_blocked_rdev); 7570 7571 7572 /* Bad block management. 7573 * We can record which blocks on each device are 'bad' and so just 7574 * fail those blocks, or that stripe, rather than the whole device. 7575 * Entries in the bad-block table are 64bits wide. This comprises: 7576 * Length of bad-range, in sectors: 0-511 for lengths 1-512 7577 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) 7578 * A 'shift' can be set so that larger blocks are tracked and 7579 * consequently larger devices can be covered. 7580 * 'Acknowledged' flag - 1 bit. - the most significant bit. 7581 * 7582 * Locking of the bad-block table uses a seqlock so md_is_badblock 7583 * might need to retry if it is very unlucky. 7584 * We will sometimes want to check for bad blocks in a bi_end_io function, 7585 * so we use the write_seqlock_irq variant. 7586 * 7587 * When looking for a bad block we specify a range and want to 7588 * know if any block in the range is bad. So we binary-search 7589 * to the last range that starts at-or-before the given endpoint, 7590 * (or "before the sector after the target range") 7591 * then see if it ends after the given start. 7592 * We return 7593 * 0 if there are no known bad blocks in the range 7594 * 1 if there are known bad block which are all acknowledged 7595 * -1 if there are bad blocks which have not yet been acknowledged in metadata. 7596 * plus the start/length of the first bad section we overlap. 7597 */ 7598 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, 7599 sector_t *first_bad, int *bad_sectors) 7600 { 7601 int hi; 7602 int lo = 0; 7603 u64 *p = bb->page; 7604 int rv = 0; 7605 sector_t target = s + sectors; 7606 unsigned seq; 7607 7608 if (bb->shift > 0) { 7609 /* round the start down, and the end up */ 7610 s >>= bb->shift; 7611 target += (1<<bb->shift) - 1; 7612 target >>= bb->shift; 7613 sectors = target - s; 7614 } 7615 /* 'target' is now the first block after the bad range */ 7616 7617 retry: 7618 seq = read_seqbegin(&bb->lock); 7619 7620 hi = bb->count; 7621 7622 /* Binary search between lo and hi for 'target' 7623 * i.e. for the last range that starts before 'target' 7624 */ 7625 /* INVARIANT: ranges before 'lo' and at-or-after 'hi' 7626 * are known not to be the last range before target. 7627 * VARIANT: hi-lo is the number of possible 7628 * ranges, and decreases until it reaches 1 7629 */ 7630 while (hi - lo > 1) { 7631 int mid = (lo + hi) / 2; 7632 sector_t a = BB_OFFSET(p[mid]); 7633 if (a < target) 7634 /* This could still be the one, earlier ranges 7635 * could not. */ 7636 lo = mid; 7637 else 7638 /* This and later ranges are definitely out. */ 7639 hi = mid; 7640 } 7641 /* 'lo' might be the last that started before target, but 'hi' isn't */ 7642 if (hi > lo) { 7643 /* need to check all range that end after 's' to see if 7644 * any are unacknowledged. 7645 */ 7646 while (lo >= 0 && 7647 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { 7648 if (BB_OFFSET(p[lo]) < target) { 7649 /* starts before the end, and finishes after 7650 * the start, so they must overlap 7651 */ 7652 if (rv != -1 && BB_ACK(p[lo])) 7653 rv = 1; 7654 else 7655 rv = -1; 7656 *first_bad = BB_OFFSET(p[lo]); 7657 *bad_sectors = BB_LEN(p[lo]); 7658 } 7659 lo--; 7660 } 7661 } 7662 7663 if (read_seqretry(&bb->lock, seq)) 7664 goto retry; 7665 7666 return rv; 7667 } 7668 EXPORT_SYMBOL_GPL(md_is_badblock); 7669 7670 /* 7671 * Add a range of bad blocks to the table. 7672 * This might extend the table, or might contract it 7673 * if two adjacent ranges can be merged. 7674 * We binary-search to find the 'insertion' point, then 7675 * decide how best to handle it. 7676 */ 7677 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, 7678 int acknowledged) 7679 { 7680 u64 *p; 7681 int lo, hi; 7682 int rv = 1; 7683 7684 if (bb->shift < 0) 7685 /* badblocks are disabled */ 7686 return 0; 7687 7688 if (bb->shift) { 7689 /* round the start down, and the end up */ 7690 sector_t next = s + sectors; 7691 s >>= bb->shift; 7692 next += (1<<bb->shift) - 1; 7693 next >>= bb->shift; 7694 sectors = next - s; 7695 } 7696 7697 write_seqlock_irq(&bb->lock); 7698 7699 p = bb->page; 7700 lo = 0; 7701 hi = bb->count; 7702 /* Find the last range that starts at-or-before 's' */ 7703 while (hi - lo > 1) { 7704 int mid = (lo + hi) / 2; 7705 sector_t a = BB_OFFSET(p[mid]); 7706 if (a <= s) 7707 lo = mid; 7708 else 7709 hi = mid; 7710 } 7711 if (hi > lo && BB_OFFSET(p[lo]) > s) 7712 hi = lo; 7713 7714 if (hi > lo) { 7715 /* we found a range that might merge with the start 7716 * of our new range 7717 */ 7718 sector_t a = BB_OFFSET(p[lo]); 7719 sector_t e = a + BB_LEN(p[lo]); 7720 int ack = BB_ACK(p[lo]); 7721 if (e >= s) { 7722 /* Yes, we can merge with a previous range */ 7723 if (s == a && s + sectors >= e) 7724 /* new range covers old */ 7725 ack = acknowledged; 7726 else 7727 ack = ack && acknowledged; 7728 7729 if (e < s + sectors) 7730 e = s + sectors; 7731 if (e - a <= BB_MAX_LEN) { 7732 p[lo] = BB_MAKE(a, e-a, ack); 7733 s = e; 7734 } else { 7735 /* does not all fit in one range, 7736 * make p[lo] maximal 7737 */ 7738 if (BB_LEN(p[lo]) != BB_MAX_LEN) 7739 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); 7740 s = a + BB_MAX_LEN; 7741 } 7742 sectors = e - s; 7743 } 7744 } 7745 if (sectors && hi < bb->count) { 7746 /* 'hi' points to the first range that starts after 's'. 7747 * Maybe we can merge with the start of that range */ 7748 sector_t a = BB_OFFSET(p[hi]); 7749 sector_t e = a + BB_LEN(p[hi]); 7750 int ack = BB_ACK(p[hi]); 7751 if (a <= s + sectors) { 7752 /* merging is possible */ 7753 if (e <= s + sectors) { 7754 /* full overlap */ 7755 e = s + sectors; 7756 ack = acknowledged; 7757 } else 7758 ack = ack && acknowledged; 7759 7760 a = s; 7761 if (e - a <= BB_MAX_LEN) { 7762 p[hi] = BB_MAKE(a, e-a, ack); 7763 s = e; 7764 } else { 7765 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); 7766 s = a + BB_MAX_LEN; 7767 } 7768 sectors = e - s; 7769 lo = hi; 7770 hi++; 7771 } 7772 } 7773 if (sectors == 0 && hi < bb->count) { 7774 /* we might be able to combine lo and hi */ 7775 /* Note: 's' is at the end of 'lo' */ 7776 sector_t a = BB_OFFSET(p[hi]); 7777 int lolen = BB_LEN(p[lo]); 7778 int hilen = BB_LEN(p[hi]); 7779 int newlen = lolen + hilen - (s - a); 7780 if (s >= a && newlen < BB_MAX_LEN) { 7781 /* yes, we can combine them */ 7782 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); 7783 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); 7784 memmove(p + hi, p + hi + 1, 7785 (bb->count - hi - 1) * 8); 7786 bb->count--; 7787 } 7788 } 7789 while (sectors) { 7790 /* didn't merge (it all). 7791 * Need to add a range just before 'hi' */ 7792 if (bb->count >= MD_MAX_BADBLOCKS) { 7793 /* No room for more */ 7794 rv = 0; 7795 break; 7796 } else { 7797 int this_sectors = sectors; 7798 memmove(p + hi + 1, p + hi, 7799 (bb->count - hi) * 8); 7800 bb->count++; 7801 7802 if (this_sectors > BB_MAX_LEN) 7803 this_sectors = BB_MAX_LEN; 7804 p[hi] = BB_MAKE(s, this_sectors, acknowledged); 7805 sectors -= this_sectors; 7806 s += this_sectors; 7807 } 7808 } 7809 7810 bb->changed = 1; 7811 if (!acknowledged) 7812 bb->unacked_exist = 1; 7813 write_sequnlock_irq(&bb->lock); 7814 7815 return rv; 7816 } 7817 7818 int rdev_set_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors, 7819 int acknowledged) 7820 { 7821 int rv = md_set_badblocks(&rdev->badblocks, 7822 s + rdev->data_offset, sectors, acknowledged); 7823 if (rv) { 7824 /* Make sure they get written out promptly */ 7825 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); 7826 md_wakeup_thread(rdev->mddev->thread); 7827 } 7828 return rv; 7829 } 7830 EXPORT_SYMBOL_GPL(rdev_set_badblocks); 7831 7832 /* 7833 * Remove a range of bad blocks from the table. 7834 * This may involve extending the table if we spilt a region, 7835 * but it must not fail. So if the table becomes full, we just 7836 * drop the remove request. 7837 */ 7838 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) 7839 { 7840 u64 *p; 7841 int lo, hi; 7842 sector_t target = s + sectors; 7843 int rv = 0; 7844 7845 if (bb->shift > 0) { 7846 /* When clearing we round the start up and the end down. 7847 * This should not matter as the shift should align with 7848 * the block size and no rounding should ever be needed. 7849 * However it is better the think a block is bad when it 7850 * isn't than to think a block is not bad when it is. 7851 */ 7852 s += (1<<bb->shift) - 1; 7853 s >>= bb->shift; 7854 target >>= bb->shift; 7855 sectors = target - s; 7856 } 7857 7858 write_seqlock_irq(&bb->lock); 7859 7860 p = bb->page; 7861 lo = 0; 7862 hi = bb->count; 7863 /* Find the last range that starts before 'target' */ 7864 while (hi - lo > 1) { 7865 int mid = (lo + hi) / 2; 7866 sector_t a = BB_OFFSET(p[mid]); 7867 if (a < target) 7868 lo = mid; 7869 else 7870 hi = mid; 7871 } 7872 if (hi > lo) { 7873 /* p[lo] is the last range that could overlap the 7874 * current range. Earlier ranges could also overlap, 7875 * but only this one can overlap the end of the range. 7876 */ 7877 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { 7878 /* Partial overlap, leave the tail of this range */ 7879 int ack = BB_ACK(p[lo]); 7880 sector_t a = BB_OFFSET(p[lo]); 7881 sector_t end = a + BB_LEN(p[lo]); 7882 7883 if (a < s) { 7884 /* we need to split this range */ 7885 if (bb->count >= MD_MAX_BADBLOCKS) { 7886 rv = 0; 7887 goto out; 7888 } 7889 memmove(p+lo+1, p+lo, (bb->count - lo) * 8); 7890 bb->count++; 7891 p[lo] = BB_MAKE(a, s-a, ack); 7892 lo++; 7893 } 7894 p[lo] = BB_MAKE(target, end - target, ack); 7895 /* there is no longer an overlap */ 7896 hi = lo; 7897 lo--; 7898 } 7899 while (lo >= 0 && 7900 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { 7901 /* This range does overlap */ 7902 if (BB_OFFSET(p[lo]) < s) { 7903 /* Keep the early parts of this range. */ 7904 int ack = BB_ACK(p[lo]); 7905 sector_t start = BB_OFFSET(p[lo]); 7906 p[lo] = BB_MAKE(start, s - start, ack); 7907 /* now low doesn't overlap, so.. */ 7908 break; 7909 } 7910 lo--; 7911 } 7912 /* 'lo' is strictly before, 'hi' is strictly after, 7913 * anything between needs to be discarded 7914 */ 7915 if (hi - lo > 1) { 7916 memmove(p+lo+1, p+hi, (bb->count - hi) * 8); 7917 bb->count -= (hi - lo - 1); 7918 } 7919 } 7920 7921 bb->changed = 1; 7922 out: 7923 write_sequnlock_irq(&bb->lock); 7924 return rv; 7925 } 7926 7927 int rdev_clear_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors) 7928 { 7929 return md_clear_badblocks(&rdev->badblocks, 7930 s + rdev->data_offset, 7931 sectors); 7932 } 7933 EXPORT_SYMBOL_GPL(rdev_clear_badblocks); 7934 7935 /* 7936 * Acknowledge all bad blocks in a list. 7937 * This only succeeds if ->changed is clear. It is used by 7938 * in-kernel metadata updates 7939 */ 7940 void md_ack_all_badblocks(struct badblocks *bb) 7941 { 7942 if (bb->page == NULL || bb->changed) 7943 /* no point even trying */ 7944 return; 7945 write_seqlock_irq(&bb->lock); 7946 7947 if (bb->changed == 0) { 7948 u64 *p = bb->page; 7949 int i; 7950 for (i = 0; i < bb->count ; i++) { 7951 if (!BB_ACK(p[i])) { 7952 sector_t start = BB_OFFSET(p[i]); 7953 int len = BB_LEN(p[i]); 7954 p[i] = BB_MAKE(start, len, 1); 7955 } 7956 } 7957 bb->unacked_exist = 0; 7958 } 7959 write_sequnlock_irq(&bb->lock); 7960 } 7961 EXPORT_SYMBOL_GPL(md_ack_all_badblocks); 7962 7963 /* sysfs access to bad-blocks list. 7964 * We present two files. 7965 * 'bad-blocks' lists sector numbers and lengths of ranges that 7966 * are recorded as bad. The list is truncated to fit within 7967 * the one-page limit of sysfs. 7968 * Writing "sector length" to this file adds an acknowledged 7969 * bad block list. 7970 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet 7971 * been acknowledged. Writing to this file adds bad blocks 7972 * without acknowledging them. This is largely for testing. 7973 */ 7974 7975 static ssize_t 7976 badblocks_show(struct badblocks *bb, char *page, int unack) 7977 { 7978 size_t len; 7979 int i; 7980 u64 *p = bb->page; 7981 unsigned seq; 7982 7983 if (bb->shift < 0) 7984 return 0; 7985 7986 retry: 7987 seq = read_seqbegin(&bb->lock); 7988 7989 len = 0; 7990 i = 0; 7991 7992 while (len < PAGE_SIZE && i < bb->count) { 7993 sector_t s = BB_OFFSET(p[i]); 7994 unsigned int length = BB_LEN(p[i]); 7995 int ack = BB_ACK(p[i]); 7996 i++; 7997 7998 if (unack && ack) 7999 continue; 8000 8001 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", 8002 (unsigned long long)s << bb->shift, 8003 length << bb->shift); 8004 } 8005 if (unack && len == 0) 8006 bb->unacked_exist = 0; 8007 8008 if (read_seqretry(&bb->lock, seq)) 8009 goto retry; 8010 8011 return len; 8012 } 8013 8014 #define DO_DEBUG 1 8015 8016 static ssize_t 8017 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) 8018 { 8019 unsigned long long sector; 8020 int length; 8021 char newline; 8022 #ifdef DO_DEBUG 8023 /* Allow clearing via sysfs *only* for testing/debugging. 8024 * Normally only a successful write may clear a badblock 8025 */ 8026 int clear = 0; 8027 if (page[0] == '-') { 8028 clear = 1; 8029 page++; 8030 } 8031 #endif /* DO_DEBUG */ 8032 8033 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { 8034 case 3: 8035 if (newline != '\n') 8036 return -EINVAL; 8037 case 2: 8038 if (length <= 0) 8039 return -EINVAL; 8040 break; 8041 default: 8042 return -EINVAL; 8043 } 8044 8045 #ifdef DO_DEBUG 8046 if (clear) { 8047 md_clear_badblocks(bb, sector, length); 8048 return len; 8049 } 8050 #endif /* DO_DEBUG */ 8051 if (md_set_badblocks(bb, sector, length, !unack)) 8052 return len; 8053 else 8054 return -ENOSPC; 8055 } 8056 8057 static int md_notify_reboot(struct notifier_block *this, 8058 unsigned long code, void *x) 8059 { 8060 struct list_head *tmp; 8061 mddev_t *mddev; 8062 8063 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { 8064 8065 printk(KERN_INFO "md: stopping all md devices.\n"); 8066 8067 for_each_mddev(mddev, tmp) 8068 if (mddev_trylock(mddev)) { 8069 /* Force a switch to readonly even array 8070 * appears to still be in use. Hence 8071 * the '100'. 8072 */ 8073 md_set_readonly(mddev, 100); 8074 mddev_unlock(mddev); 8075 } 8076 /* 8077 * certain more exotic SCSI devices are known to be 8078 * volatile wrt too early system reboots. While the 8079 * right place to handle this issue is the given 8080 * driver, we do want to have a safe RAID driver ... 8081 */ 8082 mdelay(1000*1); 8083 } 8084 return NOTIFY_DONE; 8085 } 8086 8087 static struct notifier_block md_notifier = { 8088 .notifier_call = md_notify_reboot, 8089 .next = NULL, 8090 .priority = INT_MAX, /* before any real devices */ 8091 }; 8092 8093 static void md_geninit(void) 8094 { 8095 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 8096 8097 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); 8098 } 8099 8100 static int __init md_init(void) 8101 { 8102 int ret = -ENOMEM; 8103 8104 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); 8105 if (!md_wq) 8106 goto err_wq; 8107 8108 md_misc_wq = alloc_workqueue("md_misc", 0, 0); 8109 if (!md_misc_wq) 8110 goto err_misc_wq; 8111 8112 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) 8113 goto err_md; 8114 8115 if ((ret = register_blkdev(0, "mdp")) < 0) 8116 goto err_mdp; 8117 mdp_major = ret; 8118 8119 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, 8120 md_probe, NULL, NULL); 8121 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, 8122 md_probe, NULL, NULL); 8123 8124 register_reboot_notifier(&md_notifier); 8125 raid_table_header = register_sysctl_table(raid_root_table); 8126 8127 md_geninit(); 8128 return 0; 8129 8130 err_mdp: 8131 unregister_blkdev(MD_MAJOR, "md"); 8132 err_md: 8133 destroy_workqueue(md_misc_wq); 8134 err_misc_wq: 8135 destroy_workqueue(md_wq); 8136 err_wq: 8137 return ret; 8138 } 8139 8140 #ifndef MODULE 8141 8142 /* 8143 * Searches all registered partitions for autorun RAID arrays 8144 * at boot time. 8145 */ 8146 8147 static LIST_HEAD(all_detected_devices); 8148 struct detected_devices_node { 8149 struct list_head list; 8150 dev_t dev; 8151 }; 8152 8153 void md_autodetect_dev(dev_t dev) 8154 { 8155 struct detected_devices_node *node_detected_dev; 8156 8157 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); 8158 if (node_detected_dev) { 8159 node_detected_dev->dev = dev; 8160 list_add_tail(&node_detected_dev->list, &all_detected_devices); 8161 } else { 8162 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" 8163 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); 8164 } 8165 } 8166 8167 8168 static void autostart_arrays(int part) 8169 { 8170 mdk_rdev_t *rdev; 8171 struct detected_devices_node *node_detected_dev; 8172 dev_t dev; 8173 int i_scanned, i_passed; 8174 8175 i_scanned = 0; 8176 i_passed = 0; 8177 8178 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 8179 8180 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { 8181 i_scanned++; 8182 node_detected_dev = list_entry(all_detected_devices.next, 8183 struct detected_devices_node, list); 8184 list_del(&node_detected_dev->list); 8185 dev = node_detected_dev->dev; 8186 kfree(node_detected_dev); 8187 rdev = md_import_device(dev,0, 90); 8188 if (IS_ERR(rdev)) 8189 continue; 8190 8191 if (test_bit(Faulty, &rdev->flags)) { 8192 MD_BUG(); 8193 continue; 8194 } 8195 set_bit(AutoDetected, &rdev->flags); 8196 list_add(&rdev->same_set, &pending_raid_disks); 8197 i_passed++; 8198 } 8199 8200 printk(KERN_INFO "md: Scanned %d and added %d devices.\n", 8201 i_scanned, i_passed); 8202 8203 autorun_devices(part); 8204 } 8205 8206 #endif /* !MODULE */ 8207 8208 static __exit void md_exit(void) 8209 { 8210 mddev_t *mddev; 8211 struct list_head *tmp; 8212 8213 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); 8214 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); 8215 8216 unregister_blkdev(MD_MAJOR,"md"); 8217 unregister_blkdev(mdp_major, "mdp"); 8218 unregister_reboot_notifier(&md_notifier); 8219 unregister_sysctl_table(raid_table_header); 8220 remove_proc_entry("mdstat", NULL); 8221 for_each_mddev(mddev, tmp) { 8222 export_array(mddev); 8223 mddev->hold_active = 0; 8224 } 8225 destroy_workqueue(md_misc_wq); 8226 destroy_workqueue(md_wq); 8227 } 8228 8229 subsys_initcall(md_init); 8230 module_exit(md_exit) 8231 8232 static int get_ro(char *buffer, struct kernel_param *kp) 8233 { 8234 return sprintf(buffer, "%d", start_readonly); 8235 } 8236 static int set_ro(const char *val, struct kernel_param *kp) 8237 { 8238 char *e; 8239 int num = simple_strtoul(val, &e, 10); 8240 if (*val && (*e == '\0' || *e == '\n')) { 8241 start_readonly = num; 8242 return 0; 8243 } 8244 return -EINVAL; 8245 } 8246 8247 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); 8248 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); 8249 8250 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); 8251 8252 EXPORT_SYMBOL(register_md_personality); 8253 EXPORT_SYMBOL(unregister_md_personality); 8254 EXPORT_SYMBOL(md_error); 8255 EXPORT_SYMBOL(md_done_sync); 8256 EXPORT_SYMBOL(md_write_start); 8257 EXPORT_SYMBOL(md_write_end); 8258 EXPORT_SYMBOL(md_register_thread); 8259 EXPORT_SYMBOL(md_unregister_thread); 8260 EXPORT_SYMBOL(md_wakeup_thread); 8261 EXPORT_SYMBOL(md_check_recovery); 8262 MODULE_LICENSE("GPL"); 8263 MODULE_DESCRIPTION("MD RAID framework"); 8264 MODULE_ALIAS("md"); 8265 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 8266