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 /* 219 * We have a system wide 'event count' that is incremented 220 * on any 'interesting' event, and readers of /proc/mdstat 221 * can use 'poll' or 'select' to find out when the event 222 * count increases. 223 * 224 * Events are: 225 * start array, stop array, error, add device, remove device, 226 * start build, activate spare 227 */ 228 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); 229 static atomic_t md_event_count; 230 void md_new_event(mddev_t *mddev) 231 { 232 atomic_inc(&md_event_count); 233 wake_up(&md_event_waiters); 234 } 235 EXPORT_SYMBOL_GPL(md_new_event); 236 237 /* Alternate version that can be called from interrupts 238 * when calling sysfs_notify isn't needed. 239 */ 240 static void md_new_event_inintr(mddev_t *mddev) 241 { 242 atomic_inc(&md_event_count); 243 wake_up(&md_event_waiters); 244 } 245 246 /* 247 * Enables to iterate over all existing md arrays 248 * all_mddevs_lock protects this list. 249 */ 250 static LIST_HEAD(all_mddevs); 251 static DEFINE_SPINLOCK(all_mddevs_lock); 252 253 254 /* 255 * iterates through all used mddevs in the system. 256 * We take care to grab the all_mddevs_lock whenever navigating 257 * the list, and to always hold a refcount when unlocked. 258 * Any code which breaks out of this loop while own 259 * a reference to the current mddev and must mddev_put it. 260 */ 261 #define for_each_mddev(mddev,tmp) \ 262 \ 263 for (({ spin_lock(&all_mddevs_lock); \ 264 tmp = all_mddevs.next; \ 265 mddev = NULL;}); \ 266 ({ if (tmp != &all_mddevs) \ 267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ 268 spin_unlock(&all_mddevs_lock); \ 269 if (mddev) mddev_put(mddev); \ 270 mddev = list_entry(tmp, mddev_t, all_mddevs); \ 271 tmp != &all_mddevs;}); \ 272 ({ spin_lock(&all_mddevs_lock); \ 273 tmp = tmp->next;}) \ 274 ) 275 276 277 /* Rather than calling directly into the personality make_request function, 278 * IO requests come here first so that we can check if the device is 279 * being suspended pending a reconfiguration. 280 * We hold a refcount over the call to ->make_request. By the time that 281 * call has finished, the bio has been linked into some internal structure 282 * and so is visible to ->quiesce(), so we don't need the refcount any more. 283 */ 284 static int md_make_request(struct request_queue *q, struct bio *bio) 285 { 286 const int rw = bio_data_dir(bio); 287 mddev_t *mddev = q->queuedata; 288 int rv; 289 int cpu; 290 291 if (mddev == NULL || mddev->pers == NULL) { 292 bio_io_error(bio); 293 return 0; 294 } 295 rcu_read_lock(); 296 if (mddev->suspended) { 297 DEFINE_WAIT(__wait); 298 for (;;) { 299 prepare_to_wait(&mddev->sb_wait, &__wait, 300 TASK_UNINTERRUPTIBLE); 301 if (!mddev->suspended) 302 break; 303 rcu_read_unlock(); 304 schedule(); 305 rcu_read_lock(); 306 } 307 finish_wait(&mddev->sb_wait, &__wait); 308 } 309 atomic_inc(&mddev->active_io); 310 rcu_read_unlock(); 311 312 rv = mddev->pers->make_request(mddev, bio); 313 314 cpu = part_stat_lock(); 315 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); 316 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], 317 bio_sectors(bio)); 318 part_stat_unlock(); 319 320 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) 321 wake_up(&mddev->sb_wait); 322 323 return rv; 324 } 325 326 /* mddev_suspend makes sure no new requests are submitted 327 * to the device, and that any requests that have been submitted 328 * are completely handled. 329 * Once ->stop is called and completes, the module will be completely 330 * unused. 331 */ 332 void mddev_suspend(mddev_t *mddev) 333 { 334 BUG_ON(mddev->suspended); 335 mddev->suspended = 1; 336 synchronize_rcu(); 337 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); 338 mddev->pers->quiesce(mddev, 1); 339 } 340 EXPORT_SYMBOL_GPL(mddev_suspend); 341 342 void mddev_resume(mddev_t *mddev) 343 { 344 mddev->suspended = 0; 345 wake_up(&mddev->sb_wait); 346 mddev->pers->quiesce(mddev, 0); 347 } 348 EXPORT_SYMBOL_GPL(mddev_resume); 349 350 int mddev_congested(mddev_t *mddev, int bits) 351 { 352 return mddev->suspended; 353 } 354 EXPORT_SYMBOL(mddev_congested); 355 356 /* 357 * Generic flush handling for md 358 */ 359 360 static void md_end_flush(struct bio *bio, int err) 361 { 362 mdk_rdev_t *rdev = bio->bi_private; 363 mddev_t *mddev = rdev->mddev; 364 365 rdev_dec_pending(rdev, mddev); 366 367 if (atomic_dec_and_test(&mddev->flush_pending)) { 368 /* The pre-request flush has finished */ 369 queue_work(md_wq, &mddev->flush_work); 370 } 371 bio_put(bio); 372 } 373 374 static void md_submit_flush_data(struct work_struct *ws); 375 376 static void submit_flushes(struct work_struct *ws) 377 { 378 mddev_t *mddev = container_of(ws, mddev_t, flush_work); 379 mdk_rdev_t *rdev; 380 381 INIT_WORK(&mddev->flush_work, md_submit_flush_data); 382 atomic_set(&mddev->flush_pending, 1); 383 rcu_read_lock(); 384 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 385 if (rdev->raid_disk >= 0 && 386 !test_bit(Faulty, &rdev->flags)) { 387 /* Take two references, one is dropped 388 * when request finishes, one after 389 * we reclaim rcu_read_lock 390 */ 391 struct bio *bi; 392 atomic_inc(&rdev->nr_pending); 393 atomic_inc(&rdev->nr_pending); 394 rcu_read_unlock(); 395 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev); 396 bi->bi_end_io = md_end_flush; 397 bi->bi_private = rdev; 398 bi->bi_bdev = rdev->bdev; 399 atomic_inc(&mddev->flush_pending); 400 submit_bio(WRITE_FLUSH, bi); 401 rcu_read_lock(); 402 rdev_dec_pending(rdev, mddev); 403 } 404 rcu_read_unlock(); 405 if (atomic_dec_and_test(&mddev->flush_pending)) 406 queue_work(md_wq, &mddev->flush_work); 407 } 408 409 static void md_submit_flush_data(struct work_struct *ws) 410 { 411 mddev_t *mddev = container_of(ws, mddev_t, flush_work); 412 struct bio *bio = mddev->flush_bio; 413 414 if (bio->bi_size == 0) 415 /* an empty barrier - all done */ 416 bio_endio(bio, 0); 417 else { 418 bio->bi_rw &= ~REQ_FLUSH; 419 if (mddev->pers->make_request(mddev, bio)) 420 generic_make_request(bio); 421 } 422 423 mddev->flush_bio = NULL; 424 wake_up(&mddev->sb_wait); 425 } 426 427 void md_flush_request(mddev_t *mddev, struct bio *bio) 428 { 429 spin_lock_irq(&mddev->write_lock); 430 wait_event_lock_irq(mddev->sb_wait, 431 !mddev->flush_bio, 432 mddev->write_lock, /*nothing*/); 433 mddev->flush_bio = bio; 434 spin_unlock_irq(&mddev->write_lock); 435 436 INIT_WORK(&mddev->flush_work, submit_flushes); 437 queue_work(md_wq, &mddev->flush_work); 438 } 439 EXPORT_SYMBOL(md_flush_request); 440 441 /* Support for plugging. 442 * This mirrors the plugging support in request_queue, but does not 443 * require having a whole queue 444 */ 445 static void plugger_work(struct work_struct *work) 446 { 447 struct plug_handle *plug = 448 container_of(work, struct plug_handle, unplug_work); 449 plug->unplug_fn(plug); 450 } 451 static void plugger_timeout(unsigned long data) 452 { 453 struct plug_handle *plug = (void *)data; 454 kblockd_schedule_work(NULL, &plug->unplug_work); 455 } 456 void plugger_init(struct plug_handle *plug, 457 void (*unplug_fn)(struct plug_handle *)) 458 { 459 plug->unplug_flag = 0; 460 plug->unplug_fn = unplug_fn; 461 init_timer(&plug->unplug_timer); 462 plug->unplug_timer.function = plugger_timeout; 463 plug->unplug_timer.data = (unsigned long)plug; 464 INIT_WORK(&plug->unplug_work, plugger_work); 465 } 466 EXPORT_SYMBOL_GPL(plugger_init); 467 468 void plugger_set_plug(struct plug_handle *plug) 469 { 470 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag)) 471 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1); 472 } 473 EXPORT_SYMBOL_GPL(plugger_set_plug); 474 475 int plugger_remove_plug(struct plug_handle *plug) 476 { 477 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) { 478 del_timer(&plug->unplug_timer); 479 return 1; 480 } else 481 return 0; 482 } 483 EXPORT_SYMBOL_GPL(plugger_remove_plug); 484 485 486 static inline mddev_t *mddev_get(mddev_t *mddev) 487 { 488 atomic_inc(&mddev->active); 489 return mddev; 490 } 491 492 static void mddev_delayed_delete(struct work_struct *ws); 493 494 static void mddev_put(mddev_t *mddev) 495 { 496 struct bio_set *bs = NULL; 497 498 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 499 return; 500 if (!mddev->raid_disks && list_empty(&mddev->disks) && 501 mddev->ctime == 0 && !mddev->hold_active) { 502 /* Array is not configured at all, and not held active, 503 * so destroy it */ 504 list_del(&mddev->all_mddevs); 505 bs = mddev->bio_set; 506 mddev->bio_set = NULL; 507 if (mddev->gendisk) { 508 /* We did a probe so need to clean up. Call 509 * queue_work inside the spinlock so that 510 * flush_workqueue() after mddev_find will 511 * succeed in waiting for the work to be done. 512 */ 513 INIT_WORK(&mddev->del_work, mddev_delayed_delete); 514 queue_work(md_misc_wq, &mddev->del_work); 515 } else 516 kfree(mddev); 517 } 518 spin_unlock(&all_mddevs_lock); 519 if (bs) 520 bioset_free(bs); 521 } 522 523 void mddev_init(mddev_t *mddev) 524 { 525 mutex_init(&mddev->open_mutex); 526 mutex_init(&mddev->reconfig_mutex); 527 mutex_init(&mddev->bitmap_info.mutex); 528 INIT_LIST_HEAD(&mddev->disks); 529 INIT_LIST_HEAD(&mddev->all_mddevs); 530 init_timer(&mddev->safemode_timer); 531 atomic_set(&mddev->active, 1); 532 atomic_set(&mddev->openers, 0); 533 atomic_set(&mddev->active_io, 0); 534 spin_lock_init(&mddev->write_lock); 535 atomic_set(&mddev->flush_pending, 0); 536 init_waitqueue_head(&mddev->sb_wait); 537 init_waitqueue_head(&mddev->recovery_wait); 538 mddev->reshape_position = MaxSector; 539 mddev->resync_min = 0; 540 mddev->resync_max = MaxSector; 541 mddev->level = LEVEL_NONE; 542 } 543 EXPORT_SYMBOL_GPL(mddev_init); 544 545 static mddev_t * mddev_find(dev_t unit) 546 { 547 mddev_t *mddev, *new = NULL; 548 549 retry: 550 spin_lock(&all_mddevs_lock); 551 552 if (unit) { 553 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 554 if (mddev->unit == unit) { 555 mddev_get(mddev); 556 spin_unlock(&all_mddevs_lock); 557 kfree(new); 558 return mddev; 559 } 560 561 if (new) { 562 list_add(&new->all_mddevs, &all_mddevs); 563 spin_unlock(&all_mddevs_lock); 564 new->hold_active = UNTIL_IOCTL; 565 return new; 566 } 567 } else if (new) { 568 /* find an unused unit number */ 569 static int next_minor = 512; 570 int start = next_minor; 571 int is_free = 0; 572 int dev = 0; 573 while (!is_free) { 574 dev = MKDEV(MD_MAJOR, next_minor); 575 next_minor++; 576 if (next_minor > MINORMASK) 577 next_minor = 0; 578 if (next_minor == start) { 579 /* Oh dear, all in use. */ 580 spin_unlock(&all_mddevs_lock); 581 kfree(new); 582 return NULL; 583 } 584 585 is_free = 1; 586 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 587 if (mddev->unit == dev) { 588 is_free = 0; 589 break; 590 } 591 } 592 new->unit = dev; 593 new->md_minor = MINOR(dev); 594 new->hold_active = UNTIL_STOP; 595 list_add(&new->all_mddevs, &all_mddevs); 596 spin_unlock(&all_mddevs_lock); 597 return new; 598 } 599 spin_unlock(&all_mddevs_lock); 600 601 new = kzalloc(sizeof(*new), GFP_KERNEL); 602 if (!new) 603 return NULL; 604 605 new->unit = unit; 606 if (MAJOR(unit) == MD_MAJOR) 607 new->md_minor = MINOR(unit); 608 else 609 new->md_minor = MINOR(unit) >> MdpMinorShift; 610 611 mddev_init(new); 612 613 goto retry; 614 } 615 616 static inline int mddev_lock(mddev_t * mddev) 617 { 618 return mutex_lock_interruptible(&mddev->reconfig_mutex); 619 } 620 621 static inline int mddev_is_locked(mddev_t *mddev) 622 { 623 return mutex_is_locked(&mddev->reconfig_mutex); 624 } 625 626 static inline int mddev_trylock(mddev_t * mddev) 627 { 628 return mutex_trylock(&mddev->reconfig_mutex); 629 } 630 631 static struct attribute_group md_redundancy_group; 632 633 static void mddev_unlock(mddev_t * mddev) 634 { 635 if (mddev->to_remove) { 636 /* These cannot be removed under reconfig_mutex as 637 * an access to the files will try to take reconfig_mutex 638 * while holding the file unremovable, which leads to 639 * a deadlock. 640 * So hold set sysfs_active while the remove in happeing, 641 * and anything else which might set ->to_remove or my 642 * otherwise change the sysfs namespace will fail with 643 * -EBUSY if sysfs_active is still set. 644 * We set sysfs_active under reconfig_mutex and elsewhere 645 * test it under the same mutex to ensure its correct value 646 * is seen. 647 */ 648 struct attribute_group *to_remove = mddev->to_remove; 649 mddev->to_remove = NULL; 650 mddev->sysfs_active = 1; 651 mutex_unlock(&mddev->reconfig_mutex); 652 653 if (mddev->kobj.sd) { 654 if (to_remove != &md_redundancy_group) 655 sysfs_remove_group(&mddev->kobj, to_remove); 656 if (mddev->pers == NULL || 657 mddev->pers->sync_request == NULL) { 658 sysfs_remove_group(&mddev->kobj, &md_redundancy_group); 659 if (mddev->sysfs_action) 660 sysfs_put(mddev->sysfs_action); 661 mddev->sysfs_action = NULL; 662 } 663 } 664 mddev->sysfs_active = 0; 665 } else 666 mutex_unlock(&mddev->reconfig_mutex); 667 668 md_wakeup_thread(mddev->thread); 669 } 670 671 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) 672 { 673 mdk_rdev_t *rdev; 674 675 list_for_each_entry(rdev, &mddev->disks, same_set) 676 if (rdev->desc_nr == nr) 677 return rdev; 678 679 return NULL; 680 } 681 682 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) 683 { 684 mdk_rdev_t *rdev; 685 686 list_for_each_entry(rdev, &mddev->disks, same_set) 687 if (rdev->bdev->bd_dev == dev) 688 return rdev; 689 690 return NULL; 691 } 692 693 static struct mdk_personality *find_pers(int level, char *clevel) 694 { 695 struct mdk_personality *pers; 696 list_for_each_entry(pers, &pers_list, list) { 697 if (level != LEVEL_NONE && pers->level == level) 698 return pers; 699 if (strcmp(pers->name, clevel)==0) 700 return pers; 701 } 702 return NULL; 703 } 704 705 /* return the offset of the super block in 512byte sectors */ 706 static inline sector_t calc_dev_sboffset(struct block_device *bdev) 707 { 708 sector_t num_sectors = i_size_read(bdev->bd_inode) / 512; 709 return MD_NEW_SIZE_SECTORS(num_sectors); 710 } 711 712 static int alloc_disk_sb(mdk_rdev_t * rdev) 713 { 714 if (rdev->sb_page) 715 MD_BUG(); 716 717 rdev->sb_page = alloc_page(GFP_KERNEL); 718 if (!rdev->sb_page) { 719 printk(KERN_ALERT "md: out of memory.\n"); 720 return -ENOMEM; 721 } 722 723 return 0; 724 } 725 726 static void free_disk_sb(mdk_rdev_t * rdev) 727 { 728 if (rdev->sb_page) { 729 put_page(rdev->sb_page); 730 rdev->sb_loaded = 0; 731 rdev->sb_page = NULL; 732 rdev->sb_start = 0; 733 rdev->sectors = 0; 734 } 735 } 736 737 738 static void super_written(struct bio *bio, int error) 739 { 740 mdk_rdev_t *rdev = bio->bi_private; 741 mddev_t *mddev = rdev->mddev; 742 743 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { 744 printk("md: super_written gets error=%d, uptodate=%d\n", 745 error, test_bit(BIO_UPTODATE, &bio->bi_flags)); 746 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); 747 md_error(mddev, rdev); 748 } 749 750 if (atomic_dec_and_test(&mddev->pending_writes)) 751 wake_up(&mddev->sb_wait); 752 bio_put(bio); 753 } 754 755 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, 756 sector_t sector, int size, struct page *page) 757 { 758 /* write first size bytes of page to sector of rdev 759 * Increment mddev->pending_writes before returning 760 * and decrement it on completion, waking up sb_wait 761 * if zero is reached. 762 * If an error occurred, call md_error 763 */ 764 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); 765 766 bio->bi_bdev = rdev->bdev; 767 bio->bi_sector = sector; 768 bio_add_page(bio, page, size, 0); 769 bio->bi_private = rdev; 770 bio->bi_end_io = super_written; 771 772 atomic_inc(&mddev->pending_writes); 773 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA, 774 bio); 775 } 776 777 void md_super_wait(mddev_t *mddev) 778 { 779 /* wait for all superblock writes that were scheduled to complete */ 780 DEFINE_WAIT(wq); 781 for(;;) { 782 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 783 if (atomic_read(&mddev->pending_writes)==0) 784 break; 785 schedule(); 786 } 787 finish_wait(&mddev->sb_wait, &wq); 788 } 789 790 static void bi_complete(struct bio *bio, int error) 791 { 792 complete((struct completion*)bio->bi_private); 793 } 794 795 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size, 796 struct page *page, int rw) 797 { 798 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); 799 struct completion event; 800 int ret; 801 802 rw |= REQ_SYNC | REQ_UNPLUG; 803 804 bio->bi_bdev = rdev->bdev; 805 bio->bi_sector = sector; 806 bio_add_page(bio, page, size, 0); 807 init_completion(&event); 808 bio->bi_private = &event; 809 bio->bi_end_io = bi_complete; 810 submit_bio(rw, bio); 811 wait_for_completion(&event); 812 813 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 814 bio_put(bio); 815 return ret; 816 } 817 EXPORT_SYMBOL_GPL(sync_page_io); 818 819 static int read_disk_sb(mdk_rdev_t * rdev, int size) 820 { 821 char b[BDEVNAME_SIZE]; 822 if (!rdev->sb_page) { 823 MD_BUG(); 824 return -EINVAL; 825 } 826 if (rdev->sb_loaded) 827 return 0; 828 829 830 if (!sync_page_io(rdev, rdev->sb_start, size, rdev->sb_page, READ)) 831 goto fail; 832 rdev->sb_loaded = 1; 833 return 0; 834 835 fail: 836 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 837 bdevname(rdev->bdev,b)); 838 return -EINVAL; 839 } 840 841 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 842 { 843 return sb1->set_uuid0 == sb2->set_uuid0 && 844 sb1->set_uuid1 == sb2->set_uuid1 && 845 sb1->set_uuid2 == sb2->set_uuid2 && 846 sb1->set_uuid3 == sb2->set_uuid3; 847 } 848 849 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 850 { 851 int ret; 852 mdp_super_t *tmp1, *tmp2; 853 854 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 855 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 856 857 if (!tmp1 || !tmp2) { 858 ret = 0; 859 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); 860 goto abort; 861 } 862 863 *tmp1 = *sb1; 864 *tmp2 = *sb2; 865 866 /* 867 * nr_disks is not constant 868 */ 869 tmp1->nr_disks = 0; 870 tmp2->nr_disks = 0; 871 872 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); 873 abort: 874 kfree(tmp1); 875 kfree(tmp2); 876 return ret; 877 } 878 879 880 static u32 md_csum_fold(u32 csum) 881 { 882 csum = (csum & 0xffff) + (csum >> 16); 883 return (csum & 0xffff) + (csum >> 16); 884 } 885 886 static unsigned int calc_sb_csum(mdp_super_t * sb) 887 { 888 u64 newcsum = 0; 889 u32 *sb32 = (u32*)sb; 890 int i; 891 unsigned int disk_csum, csum; 892 893 disk_csum = sb->sb_csum; 894 sb->sb_csum = 0; 895 896 for (i = 0; i < MD_SB_BYTES/4 ; i++) 897 newcsum += sb32[i]; 898 csum = (newcsum & 0xffffffff) + (newcsum>>32); 899 900 901 #ifdef CONFIG_ALPHA 902 /* This used to use csum_partial, which was wrong for several 903 * reasons including that different results are returned on 904 * different architectures. It isn't critical that we get exactly 905 * the same return value as before (we always csum_fold before 906 * testing, and that removes any differences). However as we 907 * know that csum_partial always returned a 16bit value on 908 * alphas, do a fold to maximise conformity to previous behaviour. 909 */ 910 sb->sb_csum = md_csum_fold(disk_csum); 911 #else 912 sb->sb_csum = disk_csum; 913 #endif 914 return csum; 915 } 916 917 918 /* 919 * Handle superblock details. 920 * We want to be able to handle multiple superblock formats 921 * so we have a common interface to them all, and an array of 922 * different handlers. 923 * We rely on user-space to write the initial superblock, and support 924 * reading and updating of superblocks. 925 * Interface methods are: 926 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) 927 * loads and validates a superblock on dev. 928 * if refdev != NULL, compare superblocks on both devices 929 * Return: 930 * 0 - dev has a superblock that is compatible with refdev 931 * 1 - dev has a superblock that is compatible and newer than refdev 932 * so dev should be used as the refdev in future 933 * -EINVAL superblock incompatible or invalid 934 * -othererror e.g. -EIO 935 * 936 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) 937 * Verify that dev is acceptable into mddev. 938 * The first time, mddev->raid_disks will be 0, and data from 939 * dev should be merged in. Subsequent calls check that dev 940 * is new enough. Return 0 or -EINVAL 941 * 942 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) 943 * Update the superblock for rdev with data in mddev 944 * This does not write to disc. 945 * 946 */ 947 948 struct super_type { 949 char *name; 950 struct module *owner; 951 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, 952 int minor_version); 953 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); 954 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); 955 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, 956 sector_t num_sectors); 957 }; 958 959 /* 960 * Check that the given mddev has no bitmap. 961 * 962 * This function is called from the run method of all personalities that do not 963 * support bitmaps. It prints an error message and returns non-zero if mddev 964 * has a bitmap. Otherwise, it returns 0. 965 * 966 */ 967 int md_check_no_bitmap(mddev_t *mddev) 968 { 969 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) 970 return 0; 971 printk(KERN_ERR "%s: bitmaps are not supported for %s\n", 972 mdname(mddev), mddev->pers->name); 973 return 1; 974 } 975 EXPORT_SYMBOL(md_check_no_bitmap); 976 977 /* 978 * load_super for 0.90.0 979 */ 980 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 981 { 982 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 983 mdp_super_t *sb; 984 int ret; 985 986 /* 987 * Calculate the position of the superblock (512byte sectors), 988 * it's at the end of the disk. 989 * 990 * It also happens to be a multiple of 4Kb. 991 */ 992 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 993 994 ret = read_disk_sb(rdev, MD_SB_BYTES); 995 if (ret) return ret; 996 997 ret = -EINVAL; 998 999 bdevname(rdev->bdev, b); 1000 sb = (mdp_super_t*)page_address(rdev->sb_page); 1001 1002 if (sb->md_magic != MD_SB_MAGIC) { 1003 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 1004 b); 1005 goto abort; 1006 } 1007 1008 if (sb->major_version != 0 || 1009 sb->minor_version < 90 || 1010 sb->minor_version > 91) { 1011 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 1012 sb->major_version, sb->minor_version, 1013 b); 1014 goto abort; 1015 } 1016 1017 if (sb->raid_disks <= 0) 1018 goto abort; 1019 1020 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { 1021 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 1022 b); 1023 goto abort; 1024 } 1025 1026 rdev->preferred_minor = sb->md_minor; 1027 rdev->data_offset = 0; 1028 rdev->sb_size = MD_SB_BYTES; 1029 1030 if (sb->level == LEVEL_MULTIPATH) 1031 rdev->desc_nr = -1; 1032 else 1033 rdev->desc_nr = sb->this_disk.number; 1034 1035 if (!refdev) { 1036 ret = 1; 1037 } else { 1038 __u64 ev1, ev2; 1039 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); 1040 if (!uuid_equal(refsb, sb)) { 1041 printk(KERN_WARNING "md: %s has different UUID to %s\n", 1042 b, bdevname(refdev->bdev,b2)); 1043 goto abort; 1044 } 1045 if (!sb_equal(refsb, sb)) { 1046 printk(KERN_WARNING "md: %s has same UUID" 1047 " but different superblock to %s\n", 1048 b, bdevname(refdev->bdev, b2)); 1049 goto abort; 1050 } 1051 ev1 = md_event(sb); 1052 ev2 = md_event(refsb); 1053 if (ev1 > ev2) 1054 ret = 1; 1055 else 1056 ret = 0; 1057 } 1058 rdev->sectors = rdev->sb_start; 1059 1060 if (rdev->sectors < sb->size * 2 && sb->level > 1) 1061 /* "this cannot possibly happen" ... */ 1062 ret = -EINVAL; 1063 1064 abort: 1065 return ret; 1066 } 1067 1068 /* 1069 * validate_super for 0.90.0 1070 */ 1071 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1072 { 1073 mdp_disk_t *desc; 1074 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); 1075 __u64 ev1 = md_event(sb); 1076 1077 rdev->raid_disk = -1; 1078 clear_bit(Faulty, &rdev->flags); 1079 clear_bit(In_sync, &rdev->flags); 1080 clear_bit(WriteMostly, &rdev->flags); 1081 1082 if (mddev->raid_disks == 0) { 1083 mddev->major_version = 0; 1084 mddev->minor_version = sb->minor_version; 1085 mddev->patch_version = sb->patch_version; 1086 mddev->external = 0; 1087 mddev->chunk_sectors = sb->chunk_size >> 9; 1088 mddev->ctime = sb->ctime; 1089 mddev->utime = sb->utime; 1090 mddev->level = sb->level; 1091 mddev->clevel[0] = 0; 1092 mddev->layout = sb->layout; 1093 mddev->raid_disks = sb->raid_disks; 1094 mddev->dev_sectors = sb->size * 2; 1095 mddev->events = ev1; 1096 mddev->bitmap_info.offset = 0; 1097 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 1098 1099 if (mddev->minor_version >= 91) { 1100 mddev->reshape_position = sb->reshape_position; 1101 mddev->delta_disks = sb->delta_disks; 1102 mddev->new_level = sb->new_level; 1103 mddev->new_layout = sb->new_layout; 1104 mddev->new_chunk_sectors = sb->new_chunk >> 9; 1105 } else { 1106 mddev->reshape_position = MaxSector; 1107 mddev->delta_disks = 0; 1108 mddev->new_level = mddev->level; 1109 mddev->new_layout = mddev->layout; 1110 mddev->new_chunk_sectors = mddev->chunk_sectors; 1111 } 1112 1113 if (sb->state & (1<<MD_SB_CLEAN)) 1114 mddev->recovery_cp = MaxSector; 1115 else { 1116 if (sb->events_hi == sb->cp_events_hi && 1117 sb->events_lo == sb->cp_events_lo) { 1118 mddev->recovery_cp = sb->recovery_cp; 1119 } else 1120 mddev->recovery_cp = 0; 1121 } 1122 1123 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 1124 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 1125 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 1126 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 1127 1128 mddev->max_disks = MD_SB_DISKS; 1129 1130 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 1131 mddev->bitmap_info.file == NULL) 1132 mddev->bitmap_info.offset = 1133 mddev->bitmap_info.default_offset; 1134 1135 } else if (mddev->pers == NULL) { 1136 /* Insist on good event counter while assembling, except 1137 * for spares (which don't need an event count) */ 1138 ++ev1; 1139 if (sb->disks[rdev->desc_nr].state & ( 1140 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) 1141 if (ev1 < mddev->events) 1142 return -EINVAL; 1143 } else if (mddev->bitmap) { 1144 /* if adding to array with a bitmap, then we can accept an 1145 * older device ... but not too old. 1146 */ 1147 if (ev1 < mddev->bitmap->events_cleared) 1148 return 0; 1149 } else { 1150 if (ev1 < mddev->events) 1151 /* just a hot-add of a new device, leave raid_disk at -1 */ 1152 return 0; 1153 } 1154 1155 if (mddev->level != LEVEL_MULTIPATH) { 1156 desc = sb->disks + rdev->desc_nr; 1157 1158 if (desc->state & (1<<MD_DISK_FAULTY)) 1159 set_bit(Faulty, &rdev->flags); 1160 else if (desc->state & (1<<MD_DISK_SYNC) /* && 1161 desc->raid_disk < mddev->raid_disks */) { 1162 set_bit(In_sync, &rdev->flags); 1163 rdev->raid_disk = desc->raid_disk; 1164 } else if (desc->state & (1<<MD_DISK_ACTIVE)) { 1165 /* active but not in sync implies recovery up to 1166 * reshape position. We don't know exactly where 1167 * that is, so set to zero for now */ 1168 if (mddev->minor_version >= 91) { 1169 rdev->recovery_offset = 0; 1170 rdev->raid_disk = desc->raid_disk; 1171 } 1172 } 1173 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 1174 set_bit(WriteMostly, &rdev->flags); 1175 } else /* MULTIPATH are always insync */ 1176 set_bit(In_sync, &rdev->flags); 1177 return 0; 1178 } 1179 1180 /* 1181 * sync_super for 0.90.0 1182 */ 1183 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1184 { 1185 mdp_super_t *sb; 1186 mdk_rdev_t *rdev2; 1187 int next_spare = mddev->raid_disks; 1188 1189 1190 /* make rdev->sb match mddev data.. 1191 * 1192 * 1/ zero out disks 1193 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 1194 * 3/ any empty disks < next_spare become removed 1195 * 1196 * disks[0] gets initialised to REMOVED because 1197 * we cannot be sure from other fields if it has 1198 * been initialised or not. 1199 */ 1200 int i; 1201 int active=0, working=0,failed=0,spare=0,nr_disks=0; 1202 1203 rdev->sb_size = MD_SB_BYTES; 1204 1205 sb = (mdp_super_t*)page_address(rdev->sb_page); 1206 1207 memset(sb, 0, sizeof(*sb)); 1208 1209 sb->md_magic = MD_SB_MAGIC; 1210 sb->major_version = mddev->major_version; 1211 sb->patch_version = mddev->patch_version; 1212 sb->gvalid_words = 0; /* ignored */ 1213 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 1214 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 1215 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 1216 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 1217 1218 sb->ctime = mddev->ctime; 1219 sb->level = mddev->level; 1220 sb->size = mddev->dev_sectors / 2; 1221 sb->raid_disks = mddev->raid_disks; 1222 sb->md_minor = mddev->md_minor; 1223 sb->not_persistent = 0; 1224 sb->utime = mddev->utime; 1225 sb->state = 0; 1226 sb->events_hi = (mddev->events>>32); 1227 sb->events_lo = (u32)mddev->events; 1228 1229 if (mddev->reshape_position == MaxSector) 1230 sb->minor_version = 90; 1231 else { 1232 sb->minor_version = 91; 1233 sb->reshape_position = mddev->reshape_position; 1234 sb->new_level = mddev->new_level; 1235 sb->delta_disks = mddev->delta_disks; 1236 sb->new_layout = mddev->new_layout; 1237 sb->new_chunk = mddev->new_chunk_sectors << 9; 1238 } 1239 mddev->minor_version = sb->minor_version; 1240 if (mddev->in_sync) 1241 { 1242 sb->recovery_cp = mddev->recovery_cp; 1243 sb->cp_events_hi = (mddev->events>>32); 1244 sb->cp_events_lo = (u32)mddev->events; 1245 if (mddev->recovery_cp == MaxSector) 1246 sb->state = (1<< MD_SB_CLEAN); 1247 } else 1248 sb->recovery_cp = 0; 1249 1250 sb->layout = mddev->layout; 1251 sb->chunk_size = mddev->chunk_sectors << 9; 1252 1253 if (mddev->bitmap && mddev->bitmap_info.file == NULL) 1254 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 1255 1256 sb->disks[0].state = (1<<MD_DISK_REMOVED); 1257 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1258 mdp_disk_t *d; 1259 int desc_nr; 1260 int is_active = test_bit(In_sync, &rdev2->flags); 1261 1262 if (rdev2->raid_disk >= 0 && 1263 sb->minor_version >= 91) 1264 /* we have nowhere to store the recovery_offset, 1265 * but if it is not below the reshape_position, 1266 * we can piggy-back on that. 1267 */ 1268 is_active = 1; 1269 if (rdev2->raid_disk < 0 || 1270 test_bit(Faulty, &rdev2->flags)) 1271 is_active = 0; 1272 if (is_active) 1273 desc_nr = rdev2->raid_disk; 1274 else 1275 desc_nr = next_spare++; 1276 rdev2->desc_nr = desc_nr; 1277 d = &sb->disks[rdev2->desc_nr]; 1278 nr_disks++; 1279 d->number = rdev2->desc_nr; 1280 d->major = MAJOR(rdev2->bdev->bd_dev); 1281 d->minor = MINOR(rdev2->bdev->bd_dev); 1282 if (is_active) 1283 d->raid_disk = rdev2->raid_disk; 1284 else 1285 d->raid_disk = rdev2->desc_nr; /* compatibility */ 1286 if (test_bit(Faulty, &rdev2->flags)) 1287 d->state = (1<<MD_DISK_FAULTY); 1288 else if (is_active) { 1289 d->state = (1<<MD_DISK_ACTIVE); 1290 if (test_bit(In_sync, &rdev2->flags)) 1291 d->state |= (1<<MD_DISK_SYNC); 1292 active++; 1293 working++; 1294 } else { 1295 d->state = 0; 1296 spare++; 1297 working++; 1298 } 1299 if (test_bit(WriteMostly, &rdev2->flags)) 1300 d->state |= (1<<MD_DISK_WRITEMOSTLY); 1301 } 1302 /* now set the "removed" and "faulty" bits on any missing devices */ 1303 for (i=0 ; i < mddev->raid_disks ; i++) { 1304 mdp_disk_t *d = &sb->disks[i]; 1305 if (d->state == 0 && d->number == 0) { 1306 d->number = i; 1307 d->raid_disk = i; 1308 d->state = (1<<MD_DISK_REMOVED); 1309 d->state |= (1<<MD_DISK_FAULTY); 1310 failed++; 1311 } 1312 } 1313 sb->nr_disks = nr_disks; 1314 sb->active_disks = active; 1315 sb->working_disks = working; 1316 sb->failed_disks = failed; 1317 sb->spare_disks = spare; 1318 1319 sb->this_disk = sb->disks[rdev->desc_nr]; 1320 sb->sb_csum = calc_sb_csum(sb); 1321 } 1322 1323 /* 1324 * rdev_size_change for 0.90.0 1325 */ 1326 static unsigned long long 1327 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1328 { 1329 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1330 return 0; /* component must fit device */ 1331 if (rdev->mddev->bitmap_info.offset) 1332 return 0; /* can't move bitmap */ 1333 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 1334 if (!num_sectors || num_sectors > rdev->sb_start) 1335 num_sectors = rdev->sb_start; 1336 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1337 rdev->sb_page); 1338 md_super_wait(rdev->mddev); 1339 return num_sectors; 1340 } 1341 1342 1343 /* 1344 * version 1 superblock 1345 */ 1346 1347 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) 1348 { 1349 __le32 disk_csum; 1350 u32 csum; 1351 unsigned long long newcsum; 1352 int size = 256 + le32_to_cpu(sb->max_dev)*2; 1353 __le32 *isuper = (__le32*)sb; 1354 int i; 1355 1356 disk_csum = sb->sb_csum; 1357 sb->sb_csum = 0; 1358 newcsum = 0; 1359 for (i=0; size>=4; size -= 4 ) 1360 newcsum += le32_to_cpu(*isuper++); 1361 1362 if (size == 2) 1363 newcsum += le16_to_cpu(*(__le16*) isuper); 1364 1365 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 1366 sb->sb_csum = disk_csum; 1367 return cpu_to_le32(csum); 1368 } 1369 1370 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 1371 { 1372 struct mdp_superblock_1 *sb; 1373 int ret; 1374 sector_t sb_start; 1375 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1376 int bmask; 1377 1378 /* 1379 * Calculate the position of the superblock in 512byte sectors. 1380 * It is always aligned to a 4K boundary and 1381 * depeding on minor_version, it can be: 1382 * 0: At least 8K, but less than 12K, from end of device 1383 * 1: At start of device 1384 * 2: 4K from start of device. 1385 */ 1386 switch(minor_version) { 1387 case 0: 1388 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; 1389 sb_start -= 8*2; 1390 sb_start &= ~(sector_t)(4*2-1); 1391 break; 1392 case 1: 1393 sb_start = 0; 1394 break; 1395 case 2: 1396 sb_start = 8; 1397 break; 1398 default: 1399 return -EINVAL; 1400 } 1401 rdev->sb_start = sb_start; 1402 1403 /* superblock is rarely larger than 1K, but it can be larger, 1404 * and it is safe to read 4k, so we do that 1405 */ 1406 ret = read_disk_sb(rdev, 4096); 1407 if (ret) return ret; 1408 1409 1410 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1411 1412 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 1413 sb->major_version != cpu_to_le32(1) || 1414 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 1415 le64_to_cpu(sb->super_offset) != rdev->sb_start || 1416 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 1417 return -EINVAL; 1418 1419 if (calc_sb_1_csum(sb) != sb->sb_csum) { 1420 printk("md: invalid superblock checksum on %s\n", 1421 bdevname(rdev->bdev,b)); 1422 return -EINVAL; 1423 } 1424 if (le64_to_cpu(sb->data_size) < 10) { 1425 printk("md: data_size too small on %s\n", 1426 bdevname(rdev->bdev,b)); 1427 return -EINVAL; 1428 } 1429 1430 rdev->preferred_minor = 0xffff; 1431 rdev->data_offset = le64_to_cpu(sb->data_offset); 1432 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); 1433 1434 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 1435 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1436 if (rdev->sb_size & bmask) 1437 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1438 1439 if (minor_version 1440 && rdev->data_offset < sb_start + (rdev->sb_size/512)) 1441 return -EINVAL; 1442 1443 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) 1444 rdev->desc_nr = -1; 1445 else 1446 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1447 1448 if (!refdev) { 1449 ret = 1; 1450 } else { 1451 __u64 ev1, ev2; 1452 struct mdp_superblock_1 *refsb = 1453 (struct mdp_superblock_1*)page_address(refdev->sb_page); 1454 1455 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 1456 sb->level != refsb->level || 1457 sb->layout != refsb->layout || 1458 sb->chunksize != refsb->chunksize) { 1459 printk(KERN_WARNING "md: %s has strangely different" 1460 " superblock to %s\n", 1461 bdevname(rdev->bdev,b), 1462 bdevname(refdev->bdev,b2)); 1463 return -EINVAL; 1464 } 1465 ev1 = le64_to_cpu(sb->events); 1466 ev2 = le64_to_cpu(refsb->events); 1467 1468 if (ev1 > ev2) 1469 ret = 1; 1470 else 1471 ret = 0; 1472 } 1473 if (minor_version) 1474 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 1475 le64_to_cpu(sb->data_offset); 1476 else 1477 rdev->sectors = rdev->sb_start; 1478 if (rdev->sectors < le64_to_cpu(sb->data_size)) 1479 return -EINVAL; 1480 rdev->sectors = le64_to_cpu(sb->data_size); 1481 if (le64_to_cpu(sb->size) > rdev->sectors) 1482 return -EINVAL; 1483 return ret; 1484 } 1485 1486 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1487 { 1488 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1489 __u64 ev1 = le64_to_cpu(sb->events); 1490 1491 rdev->raid_disk = -1; 1492 clear_bit(Faulty, &rdev->flags); 1493 clear_bit(In_sync, &rdev->flags); 1494 clear_bit(WriteMostly, &rdev->flags); 1495 1496 if (mddev->raid_disks == 0) { 1497 mddev->major_version = 1; 1498 mddev->patch_version = 0; 1499 mddev->external = 0; 1500 mddev->chunk_sectors = le32_to_cpu(sb->chunksize); 1501 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1502 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1503 mddev->level = le32_to_cpu(sb->level); 1504 mddev->clevel[0] = 0; 1505 mddev->layout = le32_to_cpu(sb->layout); 1506 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1507 mddev->dev_sectors = le64_to_cpu(sb->size); 1508 mddev->events = ev1; 1509 mddev->bitmap_info.offset = 0; 1510 mddev->bitmap_info.default_offset = 1024 >> 9; 1511 1512 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1513 memcpy(mddev->uuid, sb->set_uuid, 16); 1514 1515 mddev->max_disks = (4096-256)/2; 1516 1517 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1518 mddev->bitmap_info.file == NULL ) 1519 mddev->bitmap_info.offset = 1520 (__s32)le32_to_cpu(sb->bitmap_offset); 1521 1522 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { 1523 mddev->reshape_position = le64_to_cpu(sb->reshape_position); 1524 mddev->delta_disks = le32_to_cpu(sb->delta_disks); 1525 mddev->new_level = le32_to_cpu(sb->new_level); 1526 mddev->new_layout = le32_to_cpu(sb->new_layout); 1527 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); 1528 } else { 1529 mddev->reshape_position = MaxSector; 1530 mddev->delta_disks = 0; 1531 mddev->new_level = mddev->level; 1532 mddev->new_layout = mddev->layout; 1533 mddev->new_chunk_sectors = mddev->chunk_sectors; 1534 } 1535 1536 } else if (mddev->pers == NULL) { 1537 /* Insist of good event counter while assembling, except for 1538 * spares (which don't need an event count) */ 1539 ++ev1; 1540 if (rdev->desc_nr >= 0 && 1541 rdev->desc_nr < le32_to_cpu(sb->max_dev) && 1542 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) 1543 if (ev1 < mddev->events) 1544 return -EINVAL; 1545 } else if (mddev->bitmap) { 1546 /* If adding to array with a bitmap, then we can accept an 1547 * older device, but not too old. 1548 */ 1549 if (ev1 < mddev->bitmap->events_cleared) 1550 return 0; 1551 } else { 1552 if (ev1 < mddev->events) 1553 /* just a hot-add of a new device, leave raid_disk at -1 */ 1554 return 0; 1555 } 1556 if (mddev->level != LEVEL_MULTIPATH) { 1557 int role; 1558 if (rdev->desc_nr < 0 || 1559 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { 1560 role = 0xffff; 1561 rdev->desc_nr = -1; 1562 } else 1563 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1564 switch(role) { 1565 case 0xffff: /* spare */ 1566 break; 1567 case 0xfffe: /* faulty */ 1568 set_bit(Faulty, &rdev->flags); 1569 break; 1570 default: 1571 if ((le32_to_cpu(sb->feature_map) & 1572 MD_FEATURE_RECOVERY_OFFSET)) 1573 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); 1574 else 1575 set_bit(In_sync, &rdev->flags); 1576 rdev->raid_disk = role; 1577 break; 1578 } 1579 if (sb->devflags & WriteMostly1) 1580 set_bit(WriteMostly, &rdev->flags); 1581 } else /* MULTIPATH are always insync */ 1582 set_bit(In_sync, &rdev->flags); 1583 1584 return 0; 1585 } 1586 1587 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1588 { 1589 struct mdp_superblock_1 *sb; 1590 mdk_rdev_t *rdev2; 1591 int max_dev, i; 1592 /* make rdev->sb match mddev and rdev data. */ 1593 1594 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1595 1596 sb->feature_map = 0; 1597 sb->pad0 = 0; 1598 sb->recovery_offset = cpu_to_le64(0); 1599 memset(sb->pad1, 0, sizeof(sb->pad1)); 1600 memset(sb->pad2, 0, sizeof(sb->pad2)); 1601 memset(sb->pad3, 0, sizeof(sb->pad3)); 1602 1603 sb->utime = cpu_to_le64((__u64)mddev->utime); 1604 sb->events = cpu_to_le64(mddev->events); 1605 if (mddev->in_sync) 1606 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1607 else 1608 sb->resync_offset = cpu_to_le64(0); 1609 1610 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); 1611 1612 sb->raid_disks = cpu_to_le32(mddev->raid_disks); 1613 sb->size = cpu_to_le64(mddev->dev_sectors); 1614 sb->chunksize = cpu_to_le32(mddev->chunk_sectors); 1615 sb->level = cpu_to_le32(mddev->level); 1616 sb->layout = cpu_to_le32(mddev->layout); 1617 1618 if (mddev->bitmap && mddev->bitmap_info.file == NULL) { 1619 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); 1620 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); 1621 } 1622 1623 if (rdev->raid_disk >= 0 && 1624 !test_bit(In_sync, &rdev->flags)) { 1625 sb->feature_map |= 1626 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); 1627 sb->recovery_offset = 1628 cpu_to_le64(rdev->recovery_offset); 1629 } 1630 1631 if (mddev->reshape_position != MaxSector) { 1632 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); 1633 sb->reshape_position = cpu_to_le64(mddev->reshape_position); 1634 sb->new_layout = cpu_to_le32(mddev->new_layout); 1635 sb->delta_disks = cpu_to_le32(mddev->delta_disks); 1636 sb->new_level = cpu_to_le32(mddev->new_level); 1637 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); 1638 } 1639 1640 max_dev = 0; 1641 list_for_each_entry(rdev2, &mddev->disks, same_set) 1642 if (rdev2->desc_nr+1 > max_dev) 1643 max_dev = rdev2->desc_nr+1; 1644 1645 if (max_dev > le32_to_cpu(sb->max_dev)) { 1646 int bmask; 1647 sb->max_dev = cpu_to_le32(max_dev); 1648 rdev->sb_size = max_dev * 2 + 256; 1649 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1650 if (rdev->sb_size & bmask) 1651 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1652 } else 1653 max_dev = le32_to_cpu(sb->max_dev); 1654 1655 for (i=0; i<max_dev;i++) 1656 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1657 1658 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1659 i = rdev2->desc_nr; 1660 if (test_bit(Faulty, &rdev2->flags)) 1661 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1662 else if (test_bit(In_sync, &rdev2->flags)) 1663 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1664 else if (rdev2->raid_disk >= 0) 1665 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1666 else 1667 sb->dev_roles[i] = cpu_to_le16(0xffff); 1668 } 1669 1670 sb->sb_csum = calc_sb_1_csum(sb); 1671 } 1672 1673 static unsigned long long 1674 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1675 { 1676 struct mdp_superblock_1 *sb; 1677 sector_t max_sectors; 1678 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1679 return 0; /* component must fit device */ 1680 if (rdev->sb_start < rdev->data_offset) { 1681 /* minor versions 1 and 2; superblock before data */ 1682 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; 1683 max_sectors -= rdev->data_offset; 1684 if (!num_sectors || num_sectors > max_sectors) 1685 num_sectors = max_sectors; 1686 } else if (rdev->mddev->bitmap_info.offset) { 1687 /* minor version 0 with bitmap we can't move */ 1688 return 0; 1689 } else { 1690 /* minor version 0; superblock after data */ 1691 sector_t sb_start; 1692 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; 1693 sb_start &= ~(sector_t)(4*2 - 1); 1694 max_sectors = rdev->sectors + sb_start - rdev->sb_start; 1695 if (!num_sectors || num_sectors > max_sectors) 1696 num_sectors = max_sectors; 1697 rdev->sb_start = sb_start; 1698 } 1699 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); 1700 sb->data_size = cpu_to_le64(num_sectors); 1701 sb->super_offset = rdev->sb_start; 1702 sb->sb_csum = calc_sb_1_csum(sb); 1703 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1704 rdev->sb_page); 1705 md_super_wait(rdev->mddev); 1706 return num_sectors; 1707 } 1708 1709 static struct super_type super_types[] = { 1710 [0] = { 1711 .name = "0.90.0", 1712 .owner = THIS_MODULE, 1713 .load_super = super_90_load, 1714 .validate_super = super_90_validate, 1715 .sync_super = super_90_sync, 1716 .rdev_size_change = super_90_rdev_size_change, 1717 }, 1718 [1] = { 1719 .name = "md-1", 1720 .owner = THIS_MODULE, 1721 .load_super = super_1_load, 1722 .validate_super = super_1_validate, 1723 .sync_super = super_1_sync, 1724 .rdev_size_change = super_1_rdev_size_change, 1725 }, 1726 }; 1727 1728 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) 1729 { 1730 mdk_rdev_t *rdev, *rdev2; 1731 1732 rcu_read_lock(); 1733 rdev_for_each_rcu(rdev, mddev1) 1734 rdev_for_each_rcu(rdev2, mddev2) 1735 if (rdev->bdev->bd_contains == 1736 rdev2->bdev->bd_contains) { 1737 rcu_read_unlock(); 1738 return 1; 1739 } 1740 rcu_read_unlock(); 1741 return 0; 1742 } 1743 1744 static LIST_HEAD(pending_raid_disks); 1745 1746 /* 1747 * Try to register data integrity profile for an mddev 1748 * 1749 * This is called when an array is started and after a disk has been kicked 1750 * from the array. It only succeeds if all working and active component devices 1751 * are integrity capable with matching profiles. 1752 */ 1753 int md_integrity_register(mddev_t *mddev) 1754 { 1755 mdk_rdev_t *rdev, *reference = NULL; 1756 1757 if (list_empty(&mddev->disks)) 1758 return 0; /* nothing to do */ 1759 if (blk_get_integrity(mddev->gendisk)) 1760 return 0; /* already registered */ 1761 list_for_each_entry(rdev, &mddev->disks, same_set) { 1762 /* skip spares and non-functional disks */ 1763 if (test_bit(Faulty, &rdev->flags)) 1764 continue; 1765 if (rdev->raid_disk < 0) 1766 continue; 1767 /* 1768 * If at least one rdev is not integrity capable, we can not 1769 * enable data integrity for the md device. 1770 */ 1771 if (!bdev_get_integrity(rdev->bdev)) 1772 return -EINVAL; 1773 if (!reference) { 1774 /* Use the first rdev as the reference */ 1775 reference = rdev; 1776 continue; 1777 } 1778 /* does this rdev's profile match the reference profile? */ 1779 if (blk_integrity_compare(reference->bdev->bd_disk, 1780 rdev->bdev->bd_disk) < 0) 1781 return -EINVAL; 1782 } 1783 /* 1784 * All component devices are integrity capable and have matching 1785 * profiles, register the common profile for the md device. 1786 */ 1787 if (blk_integrity_register(mddev->gendisk, 1788 bdev_get_integrity(reference->bdev)) != 0) { 1789 printk(KERN_ERR "md: failed to register integrity for %s\n", 1790 mdname(mddev)); 1791 return -EINVAL; 1792 } 1793 printk(KERN_NOTICE "md: data integrity on %s enabled\n", 1794 mdname(mddev)); 1795 return 0; 1796 } 1797 EXPORT_SYMBOL(md_integrity_register); 1798 1799 /* Disable data integrity if non-capable/non-matching disk is being added */ 1800 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 1801 { 1802 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); 1803 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); 1804 1805 if (!bi_mddev) /* nothing to do */ 1806 return; 1807 if (rdev->raid_disk < 0) /* skip spares */ 1808 return; 1809 if (bi_rdev && blk_integrity_compare(mddev->gendisk, 1810 rdev->bdev->bd_disk) >= 0) 1811 return; 1812 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); 1813 blk_integrity_unregister(mddev->gendisk); 1814 } 1815 EXPORT_SYMBOL(md_integrity_add_rdev); 1816 1817 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) 1818 { 1819 char b[BDEVNAME_SIZE]; 1820 struct kobject *ko; 1821 char *s; 1822 int err; 1823 1824 if (rdev->mddev) { 1825 MD_BUG(); 1826 return -EINVAL; 1827 } 1828 1829 /* prevent duplicates */ 1830 if (find_rdev(mddev, rdev->bdev->bd_dev)) 1831 return -EEXIST; 1832 1833 /* make sure rdev->sectors exceeds mddev->dev_sectors */ 1834 if (rdev->sectors && (mddev->dev_sectors == 0 || 1835 rdev->sectors < mddev->dev_sectors)) { 1836 if (mddev->pers) { 1837 /* Cannot change size, so fail 1838 * If mddev->level <= 0, then we don't care 1839 * about aligning sizes (e.g. linear) 1840 */ 1841 if (mddev->level > 0) 1842 return -ENOSPC; 1843 } else 1844 mddev->dev_sectors = rdev->sectors; 1845 } 1846 1847 /* Verify rdev->desc_nr is unique. 1848 * If it is -1, assign a free number, else 1849 * check number is not in use 1850 */ 1851 if (rdev->desc_nr < 0) { 1852 int choice = 0; 1853 if (mddev->pers) choice = mddev->raid_disks; 1854 while (find_rdev_nr(mddev, choice)) 1855 choice++; 1856 rdev->desc_nr = choice; 1857 } else { 1858 if (find_rdev_nr(mddev, rdev->desc_nr)) 1859 return -EBUSY; 1860 } 1861 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { 1862 printk(KERN_WARNING "md: %s: array is limited to %d devices\n", 1863 mdname(mddev), mddev->max_disks); 1864 return -EBUSY; 1865 } 1866 bdevname(rdev->bdev,b); 1867 while ( (s=strchr(b, '/')) != NULL) 1868 *s = '!'; 1869 1870 rdev->mddev = mddev; 1871 printk(KERN_INFO "md: bind<%s>\n", b); 1872 1873 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) 1874 goto fail; 1875 1876 ko = &part_to_dev(rdev->bdev->bd_part)->kobj; 1877 if (sysfs_create_link(&rdev->kobj, ko, "block")) 1878 /* failure here is OK */; 1879 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); 1880 1881 list_add_rcu(&rdev->same_set, &mddev->disks); 1882 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); 1883 1884 /* May as well allow recovery to be retried once */ 1885 mddev->recovery_disabled = 0; 1886 1887 return 0; 1888 1889 fail: 1890 printk(KERN_WARNING "md: failed to register dev-%s for %s\n", 1891 b, mdname(mddev)); 1892 return err; 1893 } 1894 1895 static void md_delayed_delete(struct work_struct *ws) 1896 { 1897 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); 1898 kobject_del(&rdev->kobj); 1899 kobject_put(&rdev->kobj); 1900 } 1901 1902 static void unbind_rdev_from_array(mdk_rdev_t * rdev) 1903 { 1904 char b[BDEVNAME_SIZE]; 1905 if (!rdev->mddev) { 1906 MD_BUG(); 1907 return; 1908 } 1909 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); 1910 list_del_rcu(&rdev->same_set); 1911 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 1912 rdev->mddev = NULL; 1913 sysfs_remove_link(&rdev->kobj, "block"); 1914 sysfs_put(rdev->sysfs_state); 1915 rdev->sysfs_state = NULL; 1916 /* We need to delay this, otherwise we can deadlock when 1917 * writing to 'remove' to "dev/state". We also need 1918 * to delay it due to rcu usage. 1919 */ 1920 synchronize_rcu(); 1921 INIT_WORK(&rdev->del_work, md_delayed_delete); 1922 kobject_get(&rdev->kobj); 1923 queue_work(md_misc_wq, &rdev->del_work); 1924 } 1925 1926 /* 1927 * prevent the device from being mounted, repartitioned or 1928 * otherwise reused by a RAID array (or any other kernel 1929 * subsystem), by bd_claiming the device. 1930 */ 1931 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) 1932 { 1933 int err = 0; 1934 struct block_device *bdev; 1935 char b[BDEVNAME_SIZE]; 1936 1937 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); 1938 if (IS_ERR(bdev)) { 1939 printk(KERN_ERR "md: could not open %s.\n", 1940 __bdevname(dev, b)); 1941 return PTR_ERR(bdev); 1942 } 1943 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); 1944 if (err) { 1945 printk(KERN_ERR "md: could not bd_claim %s.\n", 1946 bdevname(bdev, b)); 1947 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1948 return err; 1949 } 1950 if (!shared) 1951 set_bit(AllReserved, &rdev->flags); 1952 rdev->bdev = bdev; 1953 return err; 1954 } 1955 1956 static void unlock_rdev(mdk_rdev_t *rdev) 1957 { 1958 struct block_device *bdev = rdev->bdev; 1959 rdev->bdev = NULL; 1960 if (!bdev) 1961 MD_BUG(); 1962 bd_release(bdev); 1963 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1964 } 1965 1966 void md_autodetect_dev(dev_t dev); 1967 1968 static void export_rdev(mdk_rdev_t * rdev) 1969 { 1970 char b[BDEVNAME_SIZE]; 1971 printk(KERN_INFO "md: export_rdev(%s)\n", 1972 bdevname(rdev->bdev,b)); 1973 if (rdev->mddev) 1974 MD_BUG(); 1975 free_disk_sb(rdev); 1976 #ifndef MODULE 1977 if (test_bit(AutoDetected, &rdev->flags)) 1978 md_autodetect_dev(rdev->bdev->bd_dev); 1979 #endif 1980 unlock_rdev(rdev); 1981 kobject_put(&rdev->kobj); 1982 } 1983 1984 static void kick_rdev_from_array(mdk_rdev_t * rdev) 1985 { 1986 unbind_rdev_from_array(rdev); 1987 export_rdev(rdev); 1988 } 1989 1990 static void export_array(mddev_t *mddev) 1991 { 1992 mdk_rdev_t *rdev, *tmp; 1993 1994 rdev_for_each(rdev, tmp, mddev) { 1995 if (!rdev->mddev) { 1996 MD_BUG(); 1997 continue; 1998 } 1999 kick_rdev_from_array(rdev); 2000 } 2001 if (!list_empty(&mddev->disks)) 2002 MD_BUG(); 2003 mddev->raid_disks = 0; 2004 mddev->major_version = 0; 2005 } 2006 2007 static void print_desc(mdp_disk_t *desc) 2008 { 2009 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 2010 desc->major,desc->minor,desc->raid_disk,desc->state); 2011 } 2012 2013 static void print_sb_90(mdp_super_t *sb) 2014 { 2015 int i; 2016 2017 printk(KERN_INFO 2018 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 2019 sb->major_version, sb->minor_version, sb->patch_version, 2020 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 2021 sb->ctime); 2022 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 2023 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 2024 sb->md_minor, sb->layout, sb->chunk_size); 2025 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 2026 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 2027 sb->utime, sb->state, sb->active_disks, sb->working_disks, 2028 sb->failed_disks, sb->spare_disks, 2029 sb->sb_csum, (unsigned long)sb->events_lo); 2030 2031 printk(KERN_INFO); 2032 for (i = 0; i < MD_SB_DISKS; i++) { 2033 mdp_disk_t *desc; 2034 2035 desc = sb->disks + i; 2036 if (desc->number || desc->major || desc->minor || 2037 desc->raid_disk || (desc->state && (desc->state != 4))) { 2038 printk(" D %2d: ", i); 2039 print_desc(desc); 2040 } 2041 } 2042 printk(KERN_INFO "md: THIS: "); 2043 print_desc(&sb->this_disk); 2044 } 2045 2046 static void print_sb_1(struct mdp_superblock_1 *sb) 2047 { 2048 __u8 *uuid; 2049 2050 uuid = sb->set_uuid; 2051 printk(KERN_INFO 2052 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" 2053 "md: Name: \"%s\" CT:%llu\n", 2054 le32_to_cpu(sb->major_version), 2055 le32_to_cpu(sb->feature_map), 2056 uuid, 2057 sb->set_name, 2058 (unsigned long long)le64_to_cpu(sb->ctime) 2059 & MD_SUPERBLOCK_1_TIME_SEC_MASK); 2060 2061 uuid = sb->device_uuid; 2062 printk(KERN_INFO 2063 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" 2064 " RO:%llu\n" 2065 "md: Dev:%08x UUID: %pU\n" 2066 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" 2067 "md: (MaxDev:%u) \n", 2068 le32_to_cpu(sb->level), 2069 (unsigned long long)le64_to_cpu(sb->size), 2070 le32_to_cpu(sb->raid_disks), 2071 le32_to_cpu(sb->layout), 2072 le32_to_cpu(sb->chunksize), 2073 (unsigned long long)le64_to_cpu(sb->data_offset), 2074 (unsigned long long)le64_to_cpu(sb->data_size), 2075 (unsigned long long)le64_to_cpu(sb->super_offset), 2076 (unsigned long long)le64_to_cpu(sb->recovery_offset), 2077 le32_to_cpu(sb->dev_number), 2078 uuid, 2079 sb->devflags, 2080 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, 2081 (unsigned long long)le64_to_cpu(sb->events), 2082 (unsigned long long)le64_to_cpu(sb->resync_offset), 2083 le32_to_cpu(sb->sb_csum), 2084 le32_to_cpu(sb->max_dev) 2085 ); 2086 } 2087 2088 static void print_rdev(mdk_rdev_t *rdev, int major_version) 2089 { 2090 char b[BDEVNAME_SIZE]; 2091 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", 2092 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, 2093 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 2094 rdev->desc_nr); 2095 if (rdev->sb_loaded) { 2096 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); 2097 switch (major_version) { 2098 case 0: 2099 print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); 2100 break; 2101 case 1: 2102 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); 2103 break; 2104 } 2105 } else 2106 printk(KERN_INFO "md: no rdev superblock!\n"); 2107 } 2108 2109 static void md_print_devices(void) 2110 { 2111 struct list_head *tmp; 2112 mdk_rdev_t *rdev; 2113 mddev_t *mddev; 2114 char b[BDEVNAME_SIZE]; 2115 2116 printk("\n"); 2117 printk("md: **********************************\n"); 2118 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 2119 printk("md: **********************************\n"); 2120 for_each_mddev(mddev, tmp) { 2121 2122 if (mddev->bitmap) 2123 bitmap_print_sb(mddev->bitmap); 2124 else 2125 printk("%s: ", mdname(mddev)); 2126 list_for_each_entry(rdev, &mddev->disks, same_set) 2127 printk("<%s>", bdevname(rdev->bdev,b)); 2128 printk("\n"); 2129 2130 list_for_each_entry(rdev, &mddev->disks, same_set) 2131 print_rdev(rdev, mddev->major_version); 2132 } 2133 printk("md: **********************************\n"); 2134 printk("\n"); 2135 } 2136 2137 2138 static void sync_sbs(mddev_t * mddev, int nospares) 2139 { 2140 /* Update each superblock (in-memory image), but 2141 * if we are allowed to, skip spares which already 2142 * have the right event counter, or have one earlier 2143 * (which would mean they aren't being marked as dirty 2144 * with the rest of the array) 2145 */ 2146 mdk_rdev_t *rdev; 2147 list_for_each_entry(rdev, &mddev->disks, same_set) { 2148 if (rdev->sb_events == mddev->events || 2149 (nospares && 2150 rdev->raid_disk < 0 && 2151 rdev->sb_events+1 == mddev->events)) { 2152 /* Don't update this superblock */ 2153 rdev->sb_loaded = 2; 2154 } else { 2155 super_types[mddev->major_version]. 2156 sync_super(mddev, rdev); 2157 rdev->sb_loaded = 1; 2158 } 2159 } 2160 } 2161 2162 static void md_update_sb(mddev_t * mddev, int force_change) 2163 { 2164 mdk_rdev_t *rdev; 2165 int sync_req; 2166 int nospares = 0; 2167 2168 repeat: 2169 /* First make sure individual recovery_offsets are correct */ 2170 list_for_each_entry(rdev, &mddev->disks, same_set) { 2171 if (rdev->raid_disk >= 0 && 2172 mddev->delta_disks >= 0 && 2173 !test_bit(In_sync, &rdev->flags) && 2174 mddev->curr_resync_completed > rdev->recovery_offset) 2175 rdev->recovery_offset = mddev->curr_resync_completed; 2176 2177 } 2178 if (!mddev->persistent) { 2179 clear_bit(MD_CHANGE_CLEAN, &mddev->flags); 2180 clear_bit(MD_CHANGE_DEVS, &mddev->flags); 2181 if (!mddev->external) 2182 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2183 wake_up(&mddev->sb_wait); 2184 return; 2185 } 2186 2187 spin_lock_irq(&mddev->write_lock); 2188 2189 mddev->utime = get_seconds(); 2190 2191 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) 2192 force_change = 1; 2193 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) 2194 /* just a clean<-> dirty transition, possibly leave spares alone, 2195 * though if events isn't the right even/odd, we will have to do 2196 * spares after all 2197 */ 2198 nospares = 1; 2199 if (force_change) 2200 nospares = 0; 2201 if (mddev->degraded) 2202 /* If the array is degraded, then skipping spares is both 2203 * dangerous and fairly pointless. 2204 * Dangerous because a device that was removed from the array 2205 * might have a event_count that still looks up-to-date, 2206 * so it can be re-added without a resync. 2207 * Pointless because if there are any spares to skip, 2208 * then a recovery will happen and soon that array won't 2209 * be degraded any more and the spare can go back to sleep then. 2210 */ 2211 nospares = 0; 2212 2213 sync_req = mddev->in_sync; 2214 2215 /* If this is just a dirty<->clean transition, and the array is clean 2216 * and 'events' is odd, we can roll back to the previous clean state */ 2217 if (nospares 2218 && (mddev->in_sync && mddev->recovery_cp == MaxSector) 2219 && mddev->can_decrease_events 2220 && mddev->events != 1) { 2221 mddev->events--; 2222 mddev->can_decrease_events = 0; 2223 } else { 2224 /* otherwise we have to go forward and ... */ 2225 mddev->events ++; 2226 mddev->can_decrease_events = nospares; 2227 } 2228 2229 if (!mddev->events) { 2230 /* 2231 * oops, this 64-bit counter should never wrap. 2232 * Either we are in around ~1 trillion A.C., assuming 2233 * 1 reboot per second, or we have a bug: 2234 */ 2235 MD_BUG(); 2236 mddev->events --; 2237 } 2238 sync_sbs(mddev, nospares); 2239 spin_unlock_irq(&mddev->write_lock); 2240 2241 dprintk(KERN_INFO 2242 "md: updating %s RAID superblock on device (in sync %d)\n", 2243 mdname(mddev),mddev->in_sync); 2244 2245 bitmap_update_sb(mddev->bitmap); 2246 list_for_each_entry(rdev, &mddev->disks, same_set) { 2247 char b[BDEVNAME_SIZE]; 2248 dprintk(KERN_INFO "md: "); 2249 if (rdev->sb_loaded != 1) 2250 continue; /* no noise on spare devices */ 2251 if (test_bit(Faulty, &rdev->flags)) 2252 dprintk("(skipping faulty "); 2253 2254 dprintk("%s ", bdevname(rdev->bdev,b)); 2255 if (!test_bit(Faulty, &rdev->flags)) { 2256 md_super_write(mddev,rdev, 2257 rdev->sb_start, rdev->sb_size, 2258 rdev->sb_page); 2259 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", 2260 bdevname(rdev->bdev,b), 2261 (unsigned long long)rdev->sb_start); 2262 rdev->sb_events = mddev->events; 2263 2264 } else 2265 dprintk(")\n"); 2266 if (mddev->level == LEVEL_MULTIPATH) 2267 /* only need to write one superblock... */ 2268 break; 2269 } 2270 md_super_wait(mddev); 2271 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ 2272 2273 spin_lock_irq(&mddev->write_lock); 2274 if (mddev->in_sync != sync_req || 2275 test_bit(MD_CHANGE_DEVS, &mddev->flags)) { 2276 /* have to write it out again */ 2277 spin_unlock_irq(&mddev->write_lock); 2278 goto repeat; 2279 } 2280 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2281 spin_unlock_irq(&mddev->write_lock); 2282 wake_up(&mddev->sb_wait); 2283 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 2284 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 2285 2286 } 2287 2288 /* words written to sysfs files may, or may not, be \n terminated. 2289 * We want to accept with case. For this we use cmd_match. 2290 */ 2291 static int cmd_match(const char *cmd, const char *str) 2292 { 2293 /* See if cmd, written into a sysfs file, matches 2294 * str. They must either be the same, or cmd can 2295 * have a trailing newline 2296 */ 2297 while (*cmd && *str && *cmd == *str) { 2298 cmd++; 2299 str++; 2300 } 2301 if (*cmd == '\n') 2302 cmd++; 2303 if (*str || *cmd) 2304 return 0; 2305 return 1; 2306 } 2307 2308 struct rdev_sysfs_entry { 2309 struct attribute attr; 2310 ssize_t (*show)(mdk_rdev_t *, char *); 2311 ssize_t (*store)(mdk_rdev_t *, const char *, size_t); 2312 }; 2313 2314 static ssize_t 2315 state_show(mdk_rdev_t *rdev, char *page) 2316 { 2317 char *sep = ""; 2318 size_t len = 0; 2319 2320 if (test_bit(Faulty, &rdev->flags)) { 2321 len+= sprintf(page+len, "%sfaulty",sep); 2322 sep = ","; 2323 } 2324 if (test_bit(In_sync, &rdev->flags)) { 2325 len += sprintf(page+len, "%sin_sync",sep); 2326 sep = ","; 2327 } 2328 if (test_bit(WriteMostly, &rdev->flags)) { 2329 len += sprintf(page+len, "%swrite_mostly",sep); 2330 sep = ","; 2331 } 2332 if (test_bit(Blocked, &rdev->flags)) { 2333 len += sprintf(page+len, "%sblocked", sep); 2334 sep = ","; 2335 } 2336 if (!test_bit(Faulty, &rdev->flags) && 2337 !test_bit(In_sync, &rdev->flags)) { 2338 len += sprintf(page+len, "%sspare", sep); 2339 sep = ","; 2340 } 2341 return len+sprintf(page+len, "\n"); 2342 } 2343 2344 static ssize_t 2345 state_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2346 { 2347 /* can write 2348 * faulty - simulates and error 2349 * remove - disconnects the device 2350 * writemostly - sets write_mostly 2351 * -writemostly - clears write_mostly 2352 * blocked - sets the Blocked flag 2353 * -blocked - clears the Blocked flag 2354 * insync - sets Insync providing device isn't active 2355 */ 2356 int err = -EINVAL; 2357 if (cmd_match(buf, "faulty") && rdev->mddev->pers) { 2358 md_error(rdev->mddev, rdev); 2359 err = 0; 2360 } else if (cmd_match(buf, "remove")) { 2361 if (rdev->raid_disk >= 0) 2362 err = -EBUSY; 2363 else { 2364 mddev_t *mddev = rdev->mddev; 2365 kick_rdev_from_array(rdev); 2366 if (mddev->pers) 2367 md_update_sb(mddev, 1); 2368 md_new_event(mddev); 2369 err = 0; 2370 } 2371 } else if (cmd_match(buf, "writemostly")) { 2372 set_bit(WriteMostly, &rdev->flags); 2373 err = 0; 2374 } else if (cmd_match(buf, "-writemostly")) { 2375 clear_bit(WriteMostly, &rdev->flags); 2376 err = 0; 2377 } else if (cmd_match(buf, "blocked")) { 2378 set_bit(Blocked, &rdev->flags); 2379 err = 0; 2380 } else if (cmd_match(buf, "-blocked")) { 2381 clear_bit(Blocked, &rdev->flags); 2382 wake_up(&rdev->blocked_wait); 2383 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2384 md_wakeup_thread(rdev->mddev->thread); 2385 2386 err = 0; 2387 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { 2388 set_bit(In_sync, &rdev->flags); 2389 err = 0; 2390 } 2391 if (!err) 2392 sysfs_notify_dirent_safe(rdev->sysfs_state); 2393 return err ? err : len; 2394 } 2395 static struct rdev_sysfs_entry rdev_state = 2396 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); 2397 2398 static ssize_t 2399 errors_show(mdk_rdev_t *rdev, char *page) 2400 { 2401 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); 2402 } 2403 2404 static ssize_t 2405 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2406 { 2407 char *e; 2408 unsigned long n = simple_strtoul(buf, &e, 10); 2409 if (*buf && (*e == 0 || *e == '\n')) { 2410 atomic_set(&rdev->corrected_errors, n); 2411 return len; 2412 } 2413 return -EINVAL; 2414 } 2415 static struct rdev_sysfs_entry rdev_errors = 2416 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); 2417 2418 static ssize_t 2419 slot_show(mdk_rdev_t *rdev, char *page) 2420 { 2421 if (rdev->raid_disk < 0) 2422 return sprintf(page, "none\n"); 2423 else 2424 return sprintf(page, "%d\n", rdev->raid_disk); 2425 } 2426 2427 static ssize_t 2428 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2429 { 2430 char *e; 2431 int err; 2432 char nm[20]; 2433 int slot = simple_strtoul(buf, &e, 10); 2434 if (strncmp(buf, "none", 4)==0) 2435 slot = -1; 2436 else if (e==buf || (*e && *e!= '\n')) 2437 return -EINVAL; 2438 if (rdev->mddev->pers && slot == -1) { 2439 /* Setting 'slot' on an active array requires also 2440 * updating the 'rd%d' link, and communicating 2441 * with the personality with ->hot_*_disk. 2442 * For now we only support removing 2443 * failed/spare devices. This normally happens automatically, 2444 * but not when the metadata is externally managed. 2445 */ 2446 if (rdev->raid_disk == -1) 2447 return -EEXIST; 2448 /* personality does all needed checks */ 2449 if (rdev->mddev->pers->hot_add_disk == NULL) 2450 return -EINVAL; 2451 err = rdev->mddev->pers-> 2452 hot_remove_disk(rdev->mddev, rdev->raid_disk); 2453 if (err) 2454 return err; 2455 sprintf(nm, "rd%d", rdev->raid_disk); 2456 sysfs_remove_link(&rdev->mddev->kobj, nm); 2457 rdev->raid_disk = -1; 2458 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2459 md_wakeup_thread(rdev->mddev->thread); 2460 } else if (rdev->mddev->pers) { 2461 mdk_rdev_t *rdev2; 2462 /* Activating a spare .. or possibly reactivating 2463 * if we ever get bitmaps working here. 2464 */ 2465 2466 if (rdev->raid_disk != -1) 2467 return -EBUSY; 2468 2469 if (rdev->mddev->pers->hot_add_disk == NULL) 2470 return -EINVAL; 2471 2472 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) 2473 if (rdev2->raid_disk == slot) 2474 return -EEXIST; 2475 2476 rdev->raid_disk = slot; 2477 if (test_bit(In_sync, &rdev->flags)) 2478 rdev->saved_raid_disk = slot; 2479 else 2480 rdev->saved_raid_disk = -1; 2481 err = rdev->mddev->pers-> 2482 hot_add_disk(rdev->mddev, rdev); 2483 if (err) { 2484 rdev->raid_disk = -1; 2485 return err; 2486 } else 2487 sysfs_notify_dirent_safe(rdev->sysfs_state); 2488 sprintf(nm, "rd%d", rdev->raid_disk); 2489 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) 2490 /* failure here is OK */; 2491 /* don't wakeup anyone, leave that to userspace. */ 2492 } else { 2493 if (slot >= rdev->mddev->raid_disks) 2494 return -ENOSPC; 2495 rdev->raid_disk = slot; 2496 /* assume it is working */ 2497 clear_bit(Faulty, &rdev->flags); 2498 clear_bit(WriteMostly, &rdev->flags); 2499 set_bit(In_sync, &rdev->flags); 2500 sysfs_notify_dirent_safe(rdev->sysfs_state); 2501 } 2502 return len; 2503 } 2504 2505 2506 static struct rdev_sysfs_entry rdev_slot = 2507 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); 2508 2509 static ssize_t 2510 offset_show(mdk_rdev_t *rdev, char *page) 2511 { 2512 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); 2513 } 2514 2515 static ssize_t 2516 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2517 { 2518 char *e; 2519 unsigned long long offset = simple_strtoull(buf, &e, 10); 2520 if (e==buf || (*e && *e != '\n')) 2521 return -EINVAL; 2522 if (rdev->mddev->pers && rdev->raid_disk >= 0) 2523 return -EBUSY; 2524 if (rdev->sectors && rdev->mddev->external) 2525 /* Must set offset before size, so overlap checks 2526 * can be sane */ 2527 return -EBUSY; 2528 rdev->data_offset = offset; 2529 return len; 2530 } 2531 2532 static struct rdev_sysfs_entry rdev_offset = 2533 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); 2534 2535 static ssize_t 2536 rdev_size_show(mdk_rdev_t *rdev, char *page) 2537 { 2538 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); 2539 } 2540 2541 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) 2542 { 2543 /* check if two start/length pairs overlap */ 2544 if (s1+l1 <= s2) 2545 return 0; 2546 if (s2+l2 <= s1) 2547 return 0; 2548 return 1; 2549 } 2550 2551 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) 2552 { 2553 unsigned long long blocks; 2554 sector_t new; 2555 2556 if (strict_strtoull(buf, 10, &blocks) < 0) 2557 return -EINVAL; 2558 2559 if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) 2560 return -EINVAL; /* sector conversion overflow */ 2561 2562 new = blocks * 2; 2563 if (new != blocks * 2) 2564 return -EINVAL; /* unsigned long long to sector_t overflow */ 2565 2566 *sectors = new; 2567 return 0; 2568 } 2569 2570 static ssize_t 2571 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2572 { 2573 mddev_t *my_mddev = rdev->mddev; 2574 sector_t oldsectors = rdev->sectors; 2575 sector_t sectors; 2576 2577 if (strict_blocks_to_sectors(buf, §ors) < 0) 2578 return -EINVAL; 2579 if (my_mddev->pers && rdev->raid_disk >= 0) { 2580 if (my_mddev->persistent) { 2581 sectors = super_types[my_mddev->major_version]. 2582 rdev_size_change(rdev, sectors); 2583 if (!sectors) 2584 return -EBUSY; 2585 } else if (!sectors) 2586 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - 2587 rdev->data_offset; 2588 } 2589 if (sectors < my_mddev->dev_sectors) 2590 return -EINVAL; /* component must fit device */ 2591 2592 rdev->sectors = sectors; 2593 if (sectors > oldsectors && my_mddev->external) { 2594 /* need to check that all other rdevs with the same ->bdev 2595 * do not overlap. We need to unlock the mddev to avoid 2596 * a deadlock. We have already changed rdev->sectors, and if 2597 * we have to change it back, we will have the lock again. 2598 */ 2599 mddev_t *mddev; 2600 int overlap = 0; 2601 struct list_head *tmp; 2602 2603 mddev_unlock(my_mddev); 2604 for_each_mddev(mddev, tmp) { 2605 mdk_rdev_t *rdev2; 2606 2607 mddev_lock(mddev); 2608 list_for_each_entry(rdev2, &mddev->disks, same_set) 2609 if (test_bit(AllReserved, &rdev2->flags) || 2610 (rdev->bdev == rdev2->bdev && 2611 rdev != rdev2 && 2612 overlaps(rdev->data_offset, rdev->sectors, 2613 rdev2->data_offset, 2614 rdev2->sectors))) { 2615 overlap = 1; 2616 break; 2617 } 2618 mddev_unlock(mddev); 2619 if (overlap) { 2620 mddev_put(mddev); 2621 break; 2622 } 2623 } 2624 mddev_lock(my_mddev); 2625 if (overlap) { 2626 /* Someone else could have slipped in a size 2627 * change here, but doing so is just silly. 2628 * We put oldsectors back because we *know* it is 2629 * safe, and trust userspace not to race with 2630 * itself 2631 */ 2632 rdev->sectors = oldsectors; 2633 return -EBUSY; 2634 } 2635 } 2636 return len; 2637 } 2638 2639 static struct rdev_sysfs_entry rdev_size = 2640 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); 2641 2642 2643 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) 2644 { 2645 unsigned long long recovery_start = rdev->recovery_offset; 2646 2647 if (test_bit(In_sync, &rdev->flags) || 2648 recovery_start == MaxSector) 2649 return sprintf(page, "none\n"); 2650 2651 return sprintf(page, "%llu\n", recovery_start); 2652 } 2653 2654 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2655 { 2656 unsigned long long recovery_start; 2657 2658 if (cmd_match(buf, "none")) 2659 recovery_start = MaxSector; 2660 else if (strict_strtoull(buf, 10, &recovery_start)) 2661 return -EINVAL; 2662 2663 if (rdev->mddev->pers && 2664 rdev->raid_disk >= 0) 2665 return -EBUSY; 2666 2667 rdev->recovery_offset = recovery_start; 2668 if (recovery_start == MaxSector) 2669 set_bit(In_sync, &rdev->flags); 2670 else 2671 clear_bit(In_sync, &rdev->flags); 2672 return len; 2673 } 2674 2675 static struct rdev_sysfs_entry rdev_recovery_start = 2676 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); 2677 2678 static struct attribute *rdev_default_attrs[] = { 2679 &rdev_state.attr, 2680 &rdev_errors.attr, 2681 &rdev_slot.attr, 2682 &rdev_offset.attr, 2683 &rdev_size.attr, 2684 &rdev_recovery_start.attr, 2685 NULL, 2686 }; 2687 static ssize_t 2688 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 2689 { 2690 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2691 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2692 mddev_t *mddev = rdev->mddev; 2693 ssize_t rv; 2694 2695 if (!entry->show) 2696 return -EIO; 2697 2698 rv = mddev ? mddev_lock(mddev) : -EBUSY; 2699 if (!rv) { 2700 if (rdev->mddev == NULL) 2701 rv = -EBUSY; 2702 else 2703 rv = entry->show(rdev, page); 2704 mddev_unlock(mddev); 2705 } 2706 return rv; 2707 } 2708 2709 static ssize_t 2710 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 2711 const char *page, size_t length) 2712 { 2713 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2714 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2715 ssize_t rv; 2716 mddev_t *mddev = rdev->mddev; 2717 2718 if (!entry->store) 2719 return -EIO; 2720 if (!capable(CAP_SYS_ADMIN)) 2721 return -EACCES; 2722 rv = mddev ? mddev_lock(mddev): -EBUSY; 2723 if (!rv) { 2724 if (rdev->mddev == NULL) 2725 rv = -EBUSY; 2726 else 2727 rv = entry->store(rdev, page, length); 2728 mddev_unlock(mddev); 2729 } 2730 return rv; 2731 } 2732 2733 static void rdev_free(struct kobject *ko) 2734 { 2735 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); 2736 kfree(rdev); 2737 } 2738 static const struct sysfs_ops rdev_sysfs_ops = { 2739 .show = rdev_attr_show, 2740 .store = rdev_attr_store, 2741 }; 2742 static struct kobj_type rdev_ktype = { 2743 .release = rdev_free, 2744 .sysfs_ops = &rdev_sysfs_ops, 2745 .default_attrs = rdev_default_attrs, 2746 }; 2747 2748 void md_rdev_init(mdk_rdev_t *rdev) 2749 { 2750 rdev->desc_nr = -1; 2751 rdev->saved_raid_disk = -1; 2752 rdev->raid_disk = -1; 2753 rdev->flags = 0; 2754 rdev->data_offset = 0; 2755 rdev->sb_events = 0; 2756 rdev->last_read_error.tv_sec = 0; 2757 rdev->last_read_error.tv_nsec = 0; 2758 atomic_set(&rdev->nr_pending, 0); 2759 atomic_set(&rdev->read_errors, 0); 2760 atomic_set(&rdev->corrected_errors, 0); 2761 2762 INIT_LIST_HEAD(&rdev->same_set); 2763 init_waitqueue_head(&rdev->blocked_wait); 2764 } 2765 EXPORT_SYMBOL_GPL(md_rdev_init); 2766 /* 2767 * Import a device. If 'super_format' >= 0, then sanity check the superblock 2768 * 2769 * mark the device faulty if: 2770 * 2771 * - the device is nonexistent (zero size) 2772 * - the device has no valid superblock 2773 * 2774 * a faulty rdev _never_ has rdev->sb set. 2775 */ 2776 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) 2777 { 2778 char b[BDEVNAME_SIZE]; 2779 int err; 2780 mdk_rdev_t *rdev; 2781 sector_t size; 2782 2783 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); 2784 if (!rdev) { 2785 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 2786 return ERR_PTR(-ENOMEM); 2787 } 2788 2789 md_rdev_init(rdev); 2790 if ((err = alloc_disk_sb(rdev))) 2791 goto abort_free; 2792 2793 err = lock_rdev(rdev, newdev, super_format == -2); 2794 if (err) 2795 goto abort_free; 2796 2797 kobject_init(&rdev->kobj, &rdev_ktype); 2798 2799 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; 2800 if (!size) { 2801 printk(KERN_WARNING 2802 "md: %s has zero or unknown size, marking faulty!\n", 2803 bdevname(rdev->bdev,b)); 2804 err = -EINVAL; 2805 goto abort_free; 2806 } 2807 2808 if (super_format >= 0) { 2809 err = super_types[super_format]. 2810 load_super(rdev, NULL, super_minor); 2811 if (err == -EINVAL) { 2812 printk(KERN_WARNING 2813 "md: %s does not have a valid v%d.%d " 2814 "superblock, not importing!\n", 2815 bdevname(rdev->bdev,b), 2816 super_format, super_minor); 2817 goto abort_free; 2818 } 2819 if (err < 0) { 2820 printk(KERN_WARNING 2821 "md: could not read %s's sb, not importing!\n", 2822 bdevname(rdev->bdev,b)); 2823 goto abort_free; 2824 } 2825 } 2826 2827 return rdev; 2828 2829 abort_free: 2830 if (rdev->sb_page) { 2831 if (rdev->bdev) 2832 unlock_rdev(rdev); 2833 free_disk_sb(rdev); 2834 } 2835 kfree(rdev); 2836 return ERR_PTR(err); 2837 } 2838 2839 /* 2840 * Check a full RAID array for plausibility 2841 */ 2842 2843 2844 static void analyze_sbs(mddev_t * mddev) 2845 { 2846 int i; 2847 mdk_rdev_t *rdev, *freshest, *tmp; 2848 char b[BDEVNAME_SIZE]; 2849 2850 freshest = NULL; 2851 rdev_for_each(rdev, tmp, mddev) 2852 switch (super_types[mddev->major_version]. 2853 load_super(rdev, freshest, mddev->minor_version)) { 2854 case 1: 2855 freshest = rdev; 2856 break; 2857 case 0: 2858 break; 2859 default: 2860 printk( KERN_ERR \ 2861 "md: fatal superblock inconsistency in %s" 2862 " -- removing from array\n", 2863 bdevname(rdev->bdev,b)); 2864 kick_rdev_from_array(rdev); 2865 } 2866 2867 2868 super_types[mddev->major_version]. 2869 validate_super(mddev, freshest); 2870 2871 i = 0; 2872 rdev_for_each(rdev, tmp, mddev) { 2873 if (mddev->max_disks && 2874 (rdev->desc_nr >= mddev->max_disks || 2875 i > mddev->max_disks)) { 2876 printk(KERN_WARNING 2877 "md: %s: %s: only %d devices permitted\n", 2878 mdname(mddev), bdevname(rdev->bdev, b), 2879 mddev->max_disks); 2880 kick_rdev_from_array(rdev); 2881 continue; 2882 } 2883 if (rdev != freshest) 2884 if (super_types[mddev->major_version]. 2885 validate_super(mddev, rdev)) { 2886 printk(KERN_WARNING "md: kicking non-fresh %s" 2887 " from array!\n", 2888 bdevname(rdev->bdev,b)); 2889 kick_rdev_from_array(rdev); 2890 continue; 2891 } 2892 if (mddev->level == LEVEL_MULTIPATH) { 2893 rdev->desc_nr = i++; 2894 rdev->raid_disk = rdev->desc_nr; 2895 set_bit(In_sync, &rdev->flags); 2896 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { 2897 rdev->raid_disk = -1; 2898 clear_bit(In_sync, &rdev->flags); 2899 } 2900 } 2901 } 2902 2903 /* Read a fixed-point number. 2904 * Numbers in sysfs attributes should be in "standard" units where 2905 * possible, so time should be in seconds. 2906 * However we internally use a a much smaller unit such as 2907 * milliseconds or jiffies. 2908 * This function takes a decimal number with a possible fractional 2909 * component, and produces an integer which is the result of 2910 * multiplying that number by 10^'scale'. 2911 * all without any floating-point arithmetic. 2912 */ 2913 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) 2914 { 2915 unsigned long result = 0; 2916 long decimals = -1; 2917 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { 2918 if (*cp == '.') 2919 decimals = 0; 2920 else if (decimals < scale) { 2921 unsigned int value; 2922 value = *cp - '0'; 2923 result = result * 10 + value; 2924 if (decimals >= 0) 2925 decimals++; 2926 } 2927 cp++; 2928 } 2929 if (*cp == '\n') 2930 cp++; 2931 if (*cp) 2932 return -EINVAL; 2933 if (decimals < 0) 2934 decimals = 0; 2935 while (decimals < scale) { 2936 result *= 10; 2937 decimals ++; 2938 } 2939 *res = result; 2940 return 0; 2941 } 2942 2943 2944 static void md_safemode_timeout(unsigned long data); 2945 2946 static ssize_t 2947 safe_delay_show(mddev_t *mddev, char *page) 2948 { 2949 int msec = (mddev->safemode_delay*1000)/HZ; 2950 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); 2951 } 2952 static ssize_t 2953 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) 2954 { 2955 unsigned long msec; 2956 2957 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) 2958 return -EINVAL; 2959 if (msec == 0) 2960 mddev->safemode_delay = 0; 2961 else { 2962 unsigned long old_delay = mddev->safemode_delay; 2963 mddev->safemode_delay = (msec*HZ)/1000; 2964 if (mddev->safemode_delay == 0) 2965 mddev->safemode_delay = 1; 2966 if (mddev->safemode_delay < old_delay) 2967 md_safemode_timeout((unsigned long)mddev); 2968 } 2969 return len; 2970 } 2971 static struct md_sysfs_entry md_safe_delay = 2972 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); 2973 2974 static ssize_t 2975 level_show(mddev_t *mddev, char *page) 2976 { 2977 struct mdk_personality *p = mddev->pers; 2978 if (p) 2979 return sprintf(page, "%s\n", p->name); 2980 else if (mddev->clevel[0]) 2981 return sprintf(page, "%s\n", mddev->clevel); 2982 else if (mddev->level != LEVEL_NONE) 2983 return sprintf(page, "%d\n", mddev->level); 2984 else 2985 return 0; 2986 } 2987 2988 static ssize_t 2989 level_store(mddev_t *mddev, const char *buf, size_t len) 2990 { 2991 char clevel[16]; 2992 ssize_t rv = len; 2993 struct mdk_personality *pers; 2994 long level; 2995 void *priv; 2996 mdk_rdev_t *rdev; 2997 2998 if (mddev->pers == NULL) { 2999 if (len == 0) 3000 return 0; 3001 if (len >= sizeof(mddev->clevel)) 3002 return -ENOSPC; 3003 strncpy(mddev->clevel, buf, len); 3004 if (mddev->clevel[len-1] == '\n') 3005 len--; 3006 mddev->clevel[len] = 0; 3007 mddev->level = LEVEL_NONE; 3008 return rv; 3009 } 3010 3011 /* request to change the personality. Need to ensure: 3012 * - array is not engaged in resync/recovery/reshape 3013 * - old personality can be suspended 3014 * - new personality will access other array. 3015 */ 3016 3017 if (mddev->sync_thread || 3018 mddev->reshape_position != MaxSector || 3019 mddev->sysfs_active) 3020 return -EBUSY; 3021 3022 if (!mddev->pers->quiesce) { 3023 printk(KERN_WARNING "md: %s: %s does not support online personality change\n", 3024 mdname(mddev), mddev->pers->name); 3025 return -EINVAL; 3026 } 3027 3028 /* Now find the new personality */ 3029 if (len == 0 || len >= sizeof(clevel)) 3030 return -EINVAL; 3031 strncpy(clevel, buf, len); 3032 if (clevel[len-1] == '\n') 3033 len--; 3034 clevel[len] = 0; 3035 if (strict_strtol(clevel, 10, &level)) 3036 level = LEVEL_NONE; 3037 3038 if (request_module("md-%s", clevel) != 0) 3039 request_module("md-level-%s", clevel); 3040 spin_lock(&pers_lock); 3041 pers = find_pers(level, clevel); 3042 if (!pers || !try_module_get(pers->owner)) { 3043 spin_unlock(&pers_lock); 3044 printk(KERN_WARNING "md: personality %s not loaded\n", clevel); 3045 return -EINVAL; 3046 } 3047 spin_unlock(&pers_lock); 3048 3049 if (pers == mddev->pers) { 3050 /* Nothing to do! */ 3051 module_put(pers->owner); 3052 return rv; 3053 } 3054 if (!pers->takeover) { 3055 module_put(pers->owner); 3056 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", 3057 mdname(mddev), clevel); 3058 return -EINVAL; 3059 } 3060 3061 list_for_each_entry(rdev, &mddev->disks, same_set) 3062 rdev->new_raid_disk = rdev->raid_disk; 3063 3064 /* ->takeover must set new_* and/or delta_disks 3065 * if it succeeds, and may set them when it fails. 3066 */ 3067 priv = pers->takeover(mddev); 3068 if (IS_ERR(priv)) { 3069 mddev->new_level = mddev->level; 3070 mddev->new_layout = mddev->layout; 3071 mddev->new_chunk_sectors = mddev->chunk_sectors; 3072 mddev->raid_disks -= mddev->delta_disks; 3073 mddev->delta_disks = 0; 3074 module_put(pers->owner); 3075 printk(KERN_WARNING "md: %s: %s would not accept array\n", 3076 mdname(mddev), clevel); 3077 return PTR_ERR(priv); 3078 } 3079 3080 /* Looks like we have a winner */ 3081 mddev_suspend(mddev); 3082 mddev->pers->stop(mddev); 3083 3084 if (mddev->pers->sync_request == NULL && 3085 pers->sync_request != NULL) { 3086 /* need to add the md_redundancy_group */ 3087 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 3088 printk(KERN_WARNING 3089 "md: cannot register extra attributes for %s\n", 3090 mdname(mddev)); 3091 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); 3092 } 3093 if (mddev->pers->sync_request != NULL && 3094 pers->sync_request == NULL) { 3095 /* need to remove the md_redundancy_group */ 3096 if (mddev->to_remove == NULL) 3097 mddev->to_remove = &md_redundancy_group; 3098 } 3099 3100 if (mddev->pers->sync_request == NULL && 3101 mddev->external) { 3102 /* We are converting from a no-redundancy array 3103 * to a redundancy array and metadata is managed 3104 * externally so we need to be sure that writes 3105 * won't block due to a need to transition 3106 * clean->dirty 3107 * until external management is started. 3108 */ 3109 mddev->in_sync = 0; 3110 mddev->safemode_delay = 0; 3111 mddev->safemode = 0; 3112 } 3113 3114 list_for_each_entry(rdev, &mddev->disks, same_set) { 3115 char nm[20]; 3116 if (rdev->raid_disk < 0) 3117 continue; 3118 if (rdev->new_raid_disk > mddev->raid_disks) 3119 rdev->new_raid_disk = -1; 3120 if (rdev->new_raid_disk == rdev->raid_disk) 3121 continue; 3122 sprintf(nm, "rd%d", rdev->raid_disk); 3123 sysfs_remove_link(&mddev->kobj, nm); 3124 } 3125 list_for_each_entry(rdev, &mddev->disks, same_set) { 3126 if (rdev->raid_disk < 0) 3127 continue; 3128 if (rdev->new_raid_disk == rdev->raid_disk) 3129 continue; 3130 rdev->raid_disk = rdev->new_raid_disk; 3131 if (rdev->raid_disk < 0) 3132 clear_bit(In_sync, &rdev->flags); 3133 else { 3134 char nm[20]; 3135 sprintf(nm, "rd%d", rdev->raid_disk); 3136 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) 3137 printk("md: cannot register %s for %s after level change\n", 3138 nm, mdname(mddev)); 3139 } 3140 } 3141 3142 module_put(mddev->pers->owner); 3143 mddev->pers = pers; 3144 mddev->private = priv; 3145 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 3146 mddev->level = mddev->new_level; 3147 mddev->layout = mddev->new_layout; 3148 mddev->chunk_sectors = mddev->new_chunk_sectors; 3149 mddev->delta_disks = 0; 3150 if (mddev->pers->sync_request == NULL) { 3151 /* this is now an array without redundancy, so 3152 * it must always be in_sync 3153 */ 3154 mddev->in_sync = 1; 3155 del_timer_sync(&mddev->safemode_timer); 3156 } 3157 pers->run(mddev); 3158 mddev_resume(mddev); 3159 set_bit(MD_CHANGE_DEVS, &mddev->flags); 3160 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3161 md_wakeup_thread(mddev->thread); 3162 sysfs_notify(&mddev->kobj, NULL, "level"); 3163 md_new_event(mddev); 3164 return rv; 3165 } 3166 3167 static struct md_sysfs_entry md_level = 3168 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); 3169 3170 3171 static ssize_t 3172 layout_show(mddev_t *mddev, char *page) 3173 { 3174 /* just a number, not meaningful for all levels */ 3175 if (mddev->reshape_position != MaxSector && 3176 mddev->layout != mddev->new_layout) 3177 return sprintf(page, "%d (%d)\n", 3178 mddev->new_layout, mddev->layout); 3179 return sprintf(page, "%d\n", mddev->layout); 3180 } 3181 3182 static ssize_t 3183 layout_store(mddev_t *mddev, const char *buf, size_t len) 3184 { 3185 char *e; 3186 unsigned long n = simple_strtoul(buf, &e, 10); 3187 3188 if (!*buf || (*e && *e != '\n')) 3189 return -EINVAL; 3190 3191 if (mddev->pers) { 3192 int err; 3193 if (mddev->pers->check_reshape == NULL) 3194 return -EBUSY; 3195 mddev->new_layout = n; 3196 err = mddev->pers->check_reshape(mddev); 3197 if (err) { 3198 mddev->new_layout = mddev->layout; 3199 return err; 3200 } 3201 } else { 3202 mddev->new_layout = n; 3203 if (mddev->reshape_position == MaxSector) 3204 mddev->layout = n; 3205 } 3206 return len; 3207 } 3208 static struct md_sysfs_entry md_layout = 3209 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); 3210 3211 3212 static ssize_t 3213 raid_disks_show(mddev_t *mddev, char *page) 3214 { 3215 if (mddev->raid_disks == 0) 3216 return 0; 3217 if (mddev->reshape_position != MaxSector && 3218 mddev->delta_disks != 0) 3219 return sprintf(page, "%d (%d)\n", mddev->raid_disks, 3220 mddev->raid_disks - mddev->delta_disks); 3221 return sprintf(page, "%d\n", mddev->raid_disks); 3222 } 3223 3224 static int update_raid_disks(mddev_t *mddev, int raid_disks); 3225 3226 static ssize_t 3227 raid_disks_store(mddev_t *mddev, const char *buf, size_t len) 3228 { 3229 char *e; 3230 int rv = 0; 3231 unsigned long n = simple_strtoul(buf, &e, 10); 3232 3233 if (!*buf || (*e && *e != '\n')) 3234 return -EINVAL; 3235 3236 if (mddev->pers) 3237 rv = update_raid_disks(mddev, n); 3238 else if (mddev->reshape_position != MaxSector) { 3239 int olddisks = mddev->raid_disks - mddev->delta_disks; 3240 mddev->delta_disks = n - olddisks; 3241 mddev->raid_disks = n; 3242 } else 3243 mddev->raid_disks = n; 3244 return rv ? rv : len; 3245 } 3246 static struct md_sysfs_entry md_raid_disks = 3247 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); 3248 3249 static ssize_t 3250 chunk_size_show(mddev_t *mddev, char *page) 3251 { 3252 if (mddev->reshape_position != MaxSector && 3253 mddev->chunk_sectors != mddev->new_chunk_sectors) 3254 return sprintf(page, "%d (%d)\n", 3255 mddev->new_chunk_sectors << 9, 3256 mddev->chunk_sectors << 9); 3257 return sprintf(page, "%d\n", mddev->chunk_sectors << 9); 3258 } 3259 3260 static ssize_t 3261 chunk_size_store(mddev_t *mddev, const char *buf, size_t len) 3262 { 3263 char *e; 3264 unsigned long n = simple_strtoul(buf, &e, 10); 3265 3266 if (!*buf || (*e && *e != '\n')) 3267 return -EINVAL; 3268 3269 if (mddev->pers) { 3270 int err; 3271 if (mddev->pers->check_reshape == NULL) 3272 return -EBUSY; 3273 mddev->new_chunk_sectors = n >> 9; 3274 err = mddev->pers->check_reshape(mddev); 3275 if (err) { 3276 mddev->new_chunk_sectors = mddev->chunk_sectors; 3277 return err; 3278 } 3279 } else { 3280 mddev->new_chunk_sectors = n >> 9; 3281 if (mddev->reshape_position == MaxSector) 3282 mddev->chunk_sectors = n >> 9; 3283 } 3284 return len; 3285 } 3286 static struct md_sysfs_entry md_chunk_size = 3287 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); 3288 3289 static ssize_t 3290 resync_start_show(mddev_t *mddev, char *page) 3291 { 3292 if (mddev->recovery_cp == MaxSector) 3293 return sprintf(page, "none\n"); 3294 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); 3295 } 3296 3297 static ssize_t 3298 resync_start_store(mddev_t *mddev, const char *buf, size_t len) 3299 { 3300 char *e; 3301 unsigned long long n = simple_strtoull(buf, &e, 10); 3302 3303 if (mddev->pers) 3304 return -EBUSY; 3305 if (cmd_match(buf, "none")) 3306 n = MaxSector; 3307 else if (!*buf || (*e && *e != '\n')) 3308 return -EINVAL; 3309 3310 mddev->recovery_cp = n; 3311 return len; 3312 } 3313 static struct md_sysfs_entry md_resync_start = 3314 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); 3315 3316 /* 3317 * The array state can be: 3318 * 3319 * clear 3320 * No devices, no size, no level 3321 * Equivalent to STOP_ARRAY ioctl 3322 * inactive 3323 * May have some settings, but array is not active 3324 * all IO results in error 3325 * When written, doesn't tear down array, but just stops it 3326 * suspended (not supported yet) 3327 * All IO requests will block. The array can be reconfigured. 3328 * Writing this, if accepted, will block until array is quiescent 3329 * readonly 3330 * no resync can happen. no superblocks get written. 3331 * write requests fail 3332 * read-auto 3333 * like readonly, but behaves like 'clean' on a write request. 3334 * 3335 * clean - no pending writes, but otherwise active. 3336 * When written to inactive array, starts without resync 3337 * If a write request arrives then 3338 * if metadata is known, mark 'dirty' and switch to 'active'. 3339 * if not known, block and switch to write-pending 3340 * If written to an active array that has pending writes, then fails. 3341 * active 3342 * fully active: IO and resync can be happening. 3343 * When written to inactive array, starts with resync 3344 * 3345 * write-pending 3346 * clean, but writes are blocked waiting for 'active' to be written. 3347 * 3348 * active-idle 3349 * like active, but no writes have been seen for a while (100msec). 3350 * 3351 */ 3352 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, 3353 write_pending, active_idle, bad_word}; 3354 static char *array_states[] = { 3355 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", 3356 "write-pending", "active-idle", NULL }; 3357 3358 static int match_word(const char *word, char **list) 3359 { 3360 int n; 3361 for (n=0; list[n]; n++) 3362 if (cmd_match(word, list[n])) 3363 break; 3364 return n; 3365 } 3366 3367 static ssize_t 3368 array_state_show(mddev_t *mddev, char *page) 3369 { 3370 enum array_state st = inactive; 3371 3372 if (mddev->pers) 3373 switch(mddev->ro) { 3374 case 1: 3375 st = readonly; 3376 break; 3377 case 2: 3378 st = read_auto; 3379 break; 3380 case 0: 3381 if (mddev->in_sync) 3382 st = clean; 3383 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 3384 st = write_pending; 3385 else if (mddev->safemode) 3386 st = active_idle; 3387 else 3388 st = active; 3389 } 3390 else { 3391 if (list_empty(&mddev->disks) && 3392 mddev->raid_disks == 0 && 3393 mddev->dev_sectors == 0) 3394 st = clear; 3395 else 3396 st = inactive; 3397 } 3398 return sprintf(page, "%s\n", array_states[st]); 3399 } 3400 3401 static int do_md_stop(mddev_t * mddev, int ro, int is_open); 3402 static int md_set_readonly(mddev_t * mddev, int is_open); 3403 static int do_md_run(mddev_t * mddev); 3404 static int restart_array(mddev_t *mddev); 3405 3406 static ssize_t 3407 array_state_store(mddev_t *mddev, const char *buf, size_t len) 3408 { 3409 int err = -EINVAL; 3410 enum array_state st = match_word(buf, array_states); 3411 switch(st) { 3412 case bad_word: 3413 break; 3414 case clear: 3415 /* stopping an active array */ 3416 if (atomic_read(&mddev->openers) > 0) 3417 return -EBUSY; 3418 err = do_md_stop(mddev, 0, 0); 3419 break; 3420 case inactive: 3421 /* stopping an active array */ 3422 if (mddev->pers) { 3423 if (atomic_read(&mddev->openers) > 0) 3424 return -EBUSY; 3425 err = do_md_stop(mddev, 2, 0); 3426 } else 3427 err = 0; /* already inactive */ 3428 break; 3429 case suspended: 3430 break; /* not supported yet */ 3431 case readonly: 3432 if (mddev->pers) 3433 err = md_set_readonly(mddev, 0); 3434 else { 3435 mddev->ro = 1; 3436 set_disk_ro(mddev->gendisk, 1); 3437 err = do_md_run(mddev); 3438 } 3439 break; 3440 case read_auto: 3441 if (mddev->pers) { 3442 if (mddev->ro == 0) 3443 err = md_set_readonly(mddev, 0); 3444 else if (mddev->ro == 1) 3445 err = restart_array(mddev); 3446 if (err == 0) { 3447 mddev->ro = 2; 3448 set_disk_ro(mddev->gendisk, 0); 3449 } 3450 } else { 3451 mddev->ro = 2; 3452 err = do_md_run(mddev); 3453 } 3454 break; 3455 case clean: 3456 if (mddev->pers) { 3457 restart_array(mddev); 3458 spin_lock_irq(&mddev->write_lock); 3459 if (atomic_read(&mddev->writes_pending) == 0) { 3460 if (mddev->in_sync == 0) { 3461 mddev->in_sync = 1; 3462 if (mddev->safemode == 1) 3463 mddev->safemode = 0; 3464 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 3465 } 3466 err = 0; 3467 } else 3468 err = -EBUSY; 3469 spin_unlock_irq(&mddev->write_lock); 3470 } else 3471 err = -EINVAL; 3472 break; 3473 case active: 3474 if (mddev->pers) { 3475 restart_array(mddev); 3476 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 3477 wake_up(&mddev->sb_wait); 3478 err = 0; 3479 } else { 3480 mddev->ro = 0; 3481 set_disk_ro(mddev->gendisk, 0); 3482 err = do_md_run(mddev); 3483 } 3484 break; 3485 case write_pending: 3486 case active_idle: 3487 /* these cannot be set */ 3488 break; 3489 } 3490 if (err) 3491 return err; 3492 else { 3493 sysfs_notify_dirent_safe(mddev->sysfs_state); 3494 return len; 3495 } 3496 } 3497 static struct md_sysfs_entry md_array_state = 3498 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); 3499 3500 static ssize_t 3501 max_corrected_read_errors_show(mddev_t *mddev, char *page) { 3502 return sprintf(page, "%d\n", 3503 atomic_read(&mddev->max_corr_read_errors)); 3504 } 3505 3506 static ssize_t 3507 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) 3508 { 3509 char *e; 3510 unsigned long n = simple_strtoul(buf, &e, 10); 3511 3512 if (*buf && (*e == 0 || *e == '\n')) { 3513 atomic_set(&mddev->max_corr_read_errors, n); 3514 return len; 3515 } 3516 return -EINVAL; 3517 } 3518 3519 static struct md_sysfs_entry max_corr_read_errors = 3520 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, 3521 max_corrected_read_errors_store); 3522 3523 static ssize_t 3524 null_show(mddev_t *mddev, char *page) 3525 { 3526 return -EINVAL; 3527 } 3528 3529 static ssize_t 3530 new_dev_store(mddev_t *mddev, const char *buf, size_t len) 3531 { 3532 /* buf must be %d:%d\n? giving major and minor numbers */ 3533 /* The new device is added to the array. 3534 * If the array has a persistent superblock, we read the 3535 * superblock to initialise info and check validity. 3536 * Otherwise, only checking done is that in bind_rdev_to_array, 3537 * which mainly checks size. 3538 */ 3539 char *e; 3540 int major = simple_strtoul(buf, &e, 10); 3541 int minor; 3542 dev_t dev; 3543 mdk_rdev_t *rdev; 3544 int err; 3545 3546 if (!*buf || *e != ':' || !e[1] || e[1] == '\n') 3547 return -EINVAL; 3548 minor = simple_strtoul(e+1, &e, 10); 3549 if (*e && *e != '\n') 3550 return -EINVAL; 3551 dev = MKDEV(major, minor); 3552 if (major != MAJOR(dev) || 3553 minor != MINOR(dev)) 3554 return -EOVERFLOW; 3555 3556 3557 if (mddev->persistent) { 3558 rdev = md_import_device(dev, mddev->major_version, 3559 mddev->minor_version); 3560 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { 3561 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 3562 mdk_rdev_t, same_set); 3563 err = super_types[mddev->major_version] 3564 .load_super(rdev, rdev0, mddev->minor_version); 3565 if (err < 0) 3566 goto out; 3567 } 3568 } else if (mddev->external) 3569 rdev = md_import_device(dev, -2, -1); 3570 else 3571 rdev = md_import_device(dev, -1, -1); 3572 3573 if (IS_ERR(rdev)) 3574 return PTR_ERR(rdev); 3575 err = bind_rdev_to_array(rdev, mddev); 3576 out: 3577 if (err) 3578 export_rdev(rdev); 3579 return err ? err : len; 3580 } 3581 3582 static struct md_sysfs_entry md_new_device = 3583 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); 3584 3585 static ssize_t 3586 bitmap_store(mddev_t *mddev, const char *buf, size_t len) 3587 { 3588 char *end; 3589 unsigned long chunk, end_chunk; 3590 3591 if (!mddev->bitmap) 3592 goto out; 3593 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ 3594 while (*buf) { 3595 chunk = end_chunk = simple_strtoul(buf, &end, 0); 3596 if (buf == end) break; 3597 if (*end == '-') { /* range */ 3598 buf = end + 1; 3599 end_chunk = simple_strtoul(buf, &end, 0); 3600 if (buf == end) break; 3601 } 3602 if (*end && !isspace(*end)) break; 3603 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); 3604 buf = skip_spaces(end); 3605 } 3606 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ 3607 out: 3608 return len; 3609 } 3610 3611 static struct md_sysfs_entry md_bitmap = 3612 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); 3613 3614 static ssize_t 3615 size_show(mddev_t *mddev, char *page) 3616 { 3617 return sprintf(page, "%llu\n", 3618 (unsigned long long)mddev->dev_sectors / 2); 3619 } 3620 3621 static int update_size(mddev_t *mddev, sector_t num_sectors); 3622 3623 static ssize_t 3624 size_store(mddev_t *mddev, const char *buf, size_t len) 3625 { 3626 /* If array is inactive, we can reduce the component size, but 3627 * not increase it (except from 0). 3628 * If array is active, we can try an on-line resize 3629 */ 3630 sector_t sectors; 3631 int err = strict_blocks_to_sectors(buf, §ors); 3632 3633 if (err < 0) 3634 return err; 3635 if (mddev->pers) { 3636 err = update_size(mddev, sectors); 3637 md_update_sb(mddev, 1); 3638 } else { 3639 if (mddev->dev_sectors == 0 || 3640 mddev->dev_sectors > sectors) 3641 mddev->dev_sectors = sectors; 3642 else 3643 err = -ENOSPC; 3644 } 3645 return err ? err : len; 3646 } 3647 3648 static struct md_sysfs_entry md_size = 3649 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); 3650 3651 3652 /* Metdata version. 3653 * This is one of 3654 * 'none' for arrays with no metadata (good luck...) 3655 * 'external' for arrays with externally managed metadata, 3656 * or N.M for internally known formats 3657 */ 3658 static ssize_t 3659 metadata_show(mddev_t *mddev, char *page) 3660 { 3661 if (mddev->persistent) 3662 return sprintf(page, "%d.%d\n", 3663 mddev->major_version, mddev->minor_version); 3664 else if (mddev->external) 3665 return sprintf(page, "external:%s\n", mddev->metadata_type); 3666 else 3667 return sprintf(page, "none\n"); 3668 } 3669 3670 static ssize_t 3671 metadata_store(mddev_t *mddev, const char *buf, size_t len) 3672 { 3673 int major, minor; 3674 char *e; 3675 /* Changing the details of 'external' metadata is 3676 * always permitted. Otherwise there must be 3677 * no devices attached to the array. 3678 */ 3679 if (mddev->external && strncmp(buf, "external:", 9) == 0) 3680 ; 3681 else if (!list_empty(&mddev->disks)) 3682 return -EBUSY; 3683 3684 if (cmd_match(buf, "none")) { 3685 mddev->persistent = 0; 3686 mddev->external = 0; 3687 mddev->major_version = 0; 3688 mddev->minor_version = 90; 3689 return len; 3690 } 3691 if (strncmp(buf, "external:", 9) == 0) { 3692 size_t namelen = len-9; 3693 if (namelen >= sizeof(mddev->metadata_type)) 3694 namelen = sizeof(mddev->metadata_type)-1; 3695 strncpy(mddev->metadata_type, buf+9, namelen); 3696 mddev->metadata_type[namelen] = 0; 3697 if (namelen && mddev->metadata_type[namelen-1] == '\n') 3698 mddev->metadata_type[--namelen] = 0; 3699 mddev->persistent = 0; 3700 mddev->external = 1; 3701 mddev->major_version = 0; 3702 mddev->minor_version = 90; 3703 return len; 3704 } 3705 major = simple_strtoul(buf, &e, 10); 3706 if (e==buf || *e != '.') 3707 return -EINVAL; 3708 buf = e+1; 3709 minor = simple_strtoul(buf, &e, 10); 3710 if (e==buf || (*e && *e != '\n') ) 3711 return -EINVAL; 3712 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) 3713 return -ENOENT; 3714 mddev->major_version = major; 3715 mddev->minor_version = minor; 3716 mddev->persistent = 1; 3717 mddev->external = 0; 3718 return len; 3719 } 3720 3721 static struct md_sysfs_entry md_metadata = 3722 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 3723 3724 static ssize_t 3725 action_show(mddev_t *mddev, char *page) 3726 { 3727 char *type = "idle"; 3728 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 3729 type = "frozen"; 3730 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 3731 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { 3732 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 3733 type = "reshape"; 3734 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 3735 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 3736 type = "resync"; 3737 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 3738 type = "check"; 3739 else 3740 type = "repair"; 3741 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) 3742 type = "recover"; 3743 } 3744 return sprintf(page, "%s\n", type); 3745 } 3746 3747 static ssize_t 3748 action_store(mddev_t *mddev, const char *page, size_t len) 3749 { 3750 if (!mddev->pers || !mddev->pers->sync_request) 3751 return -EINVAL; 3752 3753 if (cmd_match(page, "frozen")) 3754 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 3755 else 3756 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 3757 3758 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { 3759 if (mddev->sync_thread) { 3760 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 3761 md_unregister_thread(mddev->sync_thread); 3762 mddev->sync_thread = NULL; 3763 mddev->recovery = 0; 3764 } 3765 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 3766 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 3767 return -EBUSY; 3768 else if (cmd_match(page, "resync")) 3769 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3770 else if (cmd_match(page, "recover")) { 3771 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 3772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3773 } else if (cmd_match(page, "reshape")) { 3774 int err; 3775 if (mddev->pers->start_reshape == NULL) 3776 return -EINVAL; 3777 err = mddev->pers->start_reshape(mddev); 3778 if (err) 3779 return err; 3780 sysfs_notify(&mddev->kobj, NULL, "degraded"); 3781 } else { 3782 if (cmd_match(page, "check")) 3783 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 3784 else if (!cmd_match(page, "repair")) 3785 return -EINVAL; 3786 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 3787 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 3788 } 3789 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3790 md_wakeup_thread(mddev->thread); 3791 sysfs_notify_dirent_safe(mddev->sysfs_action); 3792 return len; 3793 } 3794 3795 static ssize_t 3796 mismatch_cnt_show(mddev_t *mddev, char *page) 3797 { 3798 return sprintf(page, "%llu\n", 3799 (unsigned long long) mddev->resync_mismatches); 3800 } 3801 3802 static struct md_sysfs_entry md_scan_mode = 3803 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); 3804 3805 3806 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); 3807 3808 static ssize_t 3809 sync_min_show(mddev_t *mddev, char *page) 3810 { 3811 return sprintf(page, "%d (%s)\n", speed_min(mddev), 3812 mddev->sync_speed_min ? "local": "system"); 3813 } 3814 3815 static ssize_t 3816 sync_min_store(mddev_t *mddev, const char *buf, size_t len) 3817 { 3818 int min; 3819 char *e; 3820 if (strncmp(buf, "system", 6)==0) { 3821 mddev->sync_speed_min = 0; 3822 return len; 3823 } 3824 min = simple_strtoul(buf, &e, 10); 3825 if (buf == e || (*e && *e != '\n') || min <= 0) 3826 return -EINVAL; 3827 mddev->sync_speed_min = min; 3828 return len; 3829 } 3830 3831 static struct md_sysfs_entry md_sync_min = 3832 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); 3833 3834 static ssize_t 3835 sync_max_show(mddev_t *mddev, char *page) 3836 { 3837 return sprintf(page, "%d (%s)\n", speed_max(mddev), 3838 mddev->sync_speed_max ? "local": "system"); 3839 } 3840 3841 static ssize_t 3842 sync_max_store(mddev_t *mddev, const char *buf, size_t len) 3843 { 3844 int max; 3845 char *e; 3846 if (strncmp(buf, "system", 6)==0) { 3847 mddev->sync_speed_max = 0; 3848 return len; 3849 } 3850 max = simple_strtoul(buf, &e, 10); 3851 if (buf == e || (*e && *e != '\n') || max <= 0) 3852 return -EINVAL; 3853 mddev->sync_speed_max = max; 3854 return len; 3855 } 3856 3857 static struct md_sysfs_entry md_sync_max = 3858 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); 3859 3860 static ssize_t 3861 degraded_show(mddev_t *mddev, char *page) 3862 { 3863 return sprintf(page, "%d\n", mddev->degraded); 3864 } 3865 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); 3866 3867 static ssize_t 3868 sync_force_parallel_show(mddev_t *mddev, char *page) 3869 { 3870 return sprintf(page, "%d\n", mddev->parallel_resync); 3871 } 3872 3873 static ssize_t 3874 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) 3875 { 3876 long n; 3877 3878 if (strict_strtol(buf, 10, &n)) 3879 return -EINVAL; 3880 3881 if (n != 0 && n != 1) 3882 return -EINVAL; 3883 3884 mddev->parallel_resync = n; 3885 3886 if (mddev->sync_thread) 3887 wake_up(&resync_wait); 3888 3889 return len; 3890 } 3891 3892 /* force parallel resync, even with shared block devices */ 3893 static struct md_sysfs_entry md_sync_force_parallel = 3894 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, 3895 sync_force_parallel_show, sync_force_parallel_store); 3896 3897 static ssize_t 3898 sync_speed_show(mddev_t *mddev, char *page) 3899 { 3900 unsigned long resync, dt, db; 3901 if (mddev->curr_resync == 0) 3902 return sprintf(page, "none\n"); 3903 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); 3904 dt = (jiffies - mddev->resync_mark) / HZ; 3905 if (!dt) dt++; 3906 db = resync - mddev->resync_mark_cnt; 3907 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ 3908 } 3909 3910 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); 3911 3912 static ssize_t 3913 sync_completed_show(mddev_t *mddev, char *page) 3914 { 3915 unsigned long max_sectors, resync; 3916 3917 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3918 return sprintf(page, "none\n"); 3919 3920 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 3921 max_sectors = mddev->resync_max_sectors; 3922 else 3923 max_sectors = mddev->dev_sectors; 3924 3925 resync = mddev->curr_resync_completed; 3926 return sprintf(page, "%lu / %lu\n", resync, max_sectors); 3927 } 3928 3929 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); 3930 3931 static ssize_t 3932 min_sync_show(mddev_t *mddev, char *page) 3933 { 3934 return sprintf(page, "%llu\n", 3935 (unsigned long long)mddev->resync_min); 3936 } 3937 static ssize_t 3938 min_sync_store(mddev_t *mddev, const char *buf, size_t len) 3939 { 3940 unsigned long long min; 3941 if (strict_strtoull(buf, 10, &min)) 3942 return -EINVAL; 3943 if (min > mddev->resync_max) 3944 return -EINVAL; 3945 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3946 return -EBUSY; 3947 3948 /* Must be a multiple of chunk_size */ 3949 if (mddev->chunk_sectors) { 3950 sector_t temp = min; 3951 if (sector_div(temp, mddev->chunk_sectors)) 3952 return -EINVAL; 3953 } 3954 mddev->resync_min = min; 3955 3956 return len; 3957 } 3958 3959 static struct md_sysfs_entry md_min_sync = 3960 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); 3961 3962 static ssize_t 3963 max_sync_show(mddev_t *mddev, char *page) 3964 { 3965 if (mddev->resync_max == MaxSector) 3966 return sprintf(page, "max\n"); 3967 else 3968 return sprintf(page, "%llu\n", 3969 (unsigned long long)mddev->resync_max); 3970 } 3971 static ssize_t 3972 max_sync_store(mddev_t *mddev, const char *buf, size_t len) 3973 { 3974 if (strncmp(buf, "max", 3) == 0) 3975 mddev->resync_max = MaxSector; 3976 else { 3977 unsigned long long max; 3978 if (strict_strtoull(buf, 10, &max)) 3979 return -EINVAL; 3980 if (max < mddev->resync_min) 3981 return -EINVAL; 3982 if (max < mddev->resync_max && 3983 mddev->ro == 0 && 3984 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3985 return -EBUSY; 3986 3987 /* Must be a multiple of chunk_size */ 3988 if (mddev->chunk_sectors) { 3989 sector_t temp = max; 3990 if (sector_div(temp, mddev->chunk_sectors)) 3991 return -EINVAL; 3992 } 3993 mddev->resync_max = max; 3994 } 3995 wake_up(&mddev->recovery_wait); 3996 return len; 3997 } 3998 3999 static struct md_sysfs_entry md_max_sync = 4000 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); 4001 4002 static ssize_t 4003 suspend_lo_show(mddev_t *mddev, char *page) 4004 { 4005 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); 4006 } 4007 4008 static ssize_t 4009 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) 4010 { 4011 char *e; 4012 unsigned long long new = simple_strtoull(buf, &e, 10); 4013 4014 if (mddev->pers == NULL || 4015 mddev->pers->quiesce == NULL) 4016 return -EINVAL; 4017 if (buf == e || (*e && *e != '\n')) 4018 return -EINVAL; 4019 if (new >= mddev->suspend_hi || 4020 (new > mddev->suspend_lo && new < mddev->suspend_hi)) { 4021 mddev->suspend_lo = new; 4022 mddev->pers->quiesce(mddev, 2); 4023 return len; 4024 } else 4025 return -EINVAL; 4026 } 4027 static struct md_sysfs_entry md_suspend_lo = 4028 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); 4029 4030 4031 static ssize_t 4032 suspend_hi_show(mddev_t *mddev, char *page) 4033 { 4034 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); 4035 } 4036 4037 static ssize_t 4038 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) 4039 { 4040 char *e; 4041 unsigned long long new = simple_strtoull(buf, &e, 10); 4042 4043 if (mddev->pers == NULL || 4044 mddev->pers->quiesce == NULL) 4045 return -EINVAL; 4046 if (buf == e || (*e && *e != '\n')) 4047 return -EINVAL; 4048 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || 4049 (new > mddev->suspend_lo && new > mddev->suspend_hi)) { 4050 mddev->suspend_hi = new; 4051 mddev->pers->quiesce(mddev, 1); 4052 mddev->pers->quiesce(mddev, 0); 4053 return len; 4054 } else 4055 return -EINVAL; 4056 } 4057 static struct md_sysfs_entry md_suspend_hi = 4058 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); 4059 4060 static ssize_t 4061 reshape_position_show(mddev_t *mddev, char *page) 4062 { 4063 if (mddev->reshape_position != MaxSector) 4064 return sprintf(page, "%llu\n", 4065 (unsigned long long)mddev->reshape_position); 4066 strcpy(page, "none\n"); 4067 return 5; 4068 } 4069 4070 static ssize_t 4071 reshape_position_store(mddev_t *mddev, const char *buf, size_t len) 4072 { 4073 char *e; 4074 unsigned long long new = simple_strtoull(buf, &e, 10); 4075 if (mddev->pers) 4076 return -EBUSY; 4077 if (buf == e || (*e && *e != '\n')) 4078 return -EINVAL; 4079 mddev->reshape_position = new; 4080 mddev->delta_disks = 0; 4081 mddev->new_level = mddev->level; 4082 mddev->new_layout = mddev->layout; 4083 mddev->new_chunk_sectors = mddev->chunk_sectors; 4084 return len; 4085 } 4086 4087 static struct md_sysfs_entry md_reshape_position = 4088 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, 4089 reshape_position_store); 4090 4091 static ssize_t 4092 array_size_show(mddev_t *mddev, char *page) 4093 { 4094 if (mddev->external_size) 4095 return sprintf(page, "%llu\n", 4096 (unsigned long long)mddev->array_sectors/2); 4097 else 4098 return sprintf(page, "default\n"); 4099 } 4100 4101 static ssize_t 4102 array_size_store(mddev_t *mddev, const char *buf, size_t len) 4103 { 4104 sector_t sectors; 4105 4106 if (strncmp(buf, "default", 7) == 0) { 4107 if (mddev->pers) 4108 sectors = mddev->pers->size(mddev, 0, 0); 4109 else 4110 sectors = mddev->array_sectors; 4111 4112 mddev->external_size = 0; 4113 } else { 4114 if (strict_blocks_to_sectors(buf, §ors) < 0) 4115 return -EINVAL; 4116 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) 4117 return -E2BIG; 4118 4119 mddev->external_size = 1; 4120 } 4121 4122 mddev->array_sectors = sectors; 4123 set_capacity(mddev->gendisk, mddev->array_sectors); 4124 if (mddev->pers) 4125 revalidate_disk(mddev->gendisk); 4126 4127 return len; 4128 } 4129 4130 static struct md_sysfs_entry md_array_size = 4131 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, 4132 array_size_store); 4133 4134 static struct attribute *md_default_attrs[] = { 4135 &md_level.attr, 4136 &md_layout.attr, 4137 &md_raid_disks.attr, 4138 &md_chunk_size.attr, 4139 &md_size.attr, 4140 &md_resync_start.attr, 4141 &md_metadata.attr, 4142 &md_new_device.attr, 4143 &md_safe_delay.attr, 4144 &md_array_state.attr, 4145 &md_reshape_position.attr, 4146 &md_array_size.attr, 4147 &max_corr_read_errors.attr, 4148 NULL, 4149 }; 4150 4151 static struct attribute *md_redundancy_attrs[] = { 4152 &md_scan_mode.attr, 4153 &md_mismatches.attr, 4154 &md_sync_min.attr, 4155 &md_sync_max.attr, 4156 &md_sync_speed.attr, 4157 &md_sync_force_parallel.attr, 4158 &md_sync_completed.attr, 4159 &md_min_sync.attr, 4160 &md_max_sync.attr, 4161 &md_suspend_lo.attr, 4162 &md_suspend_hi.attr, 4163 &md_bitmap.attr, 4164 &md_degraded.attr, 4165 NULL, 4166 }; 4167 static struct attribute_group md_redundancy_group = { 4168 .name = NULL, 4169 .attrs = md_redundancy_attrs, 4170 }; 4171 4172 4173 static ssize_t 4174 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 4175 { 4176 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4177 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 4178 ssize_t rv; 4179 4180 if (!entry->show) 4181 return -EIO; 4182 rv = mddev_lock(mddev); 4183 if (!rv) { 4184 rv = entry->show(mddev, page); 4185 mddev_unlock(mddev); 4186 } 4187 return rv; 4188 } 4189 4190 static ssize_t 4191 md_attr_store(struct kobject *kobj, struct attribute *attr, 4192 const char *page, size_t length) 4193 { 4194 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4195 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 4196 ssize_t rv; 4197 4198 if (!entry->store) 4199 return -EIO; 4200 if (!capable(CAP_SYS_ADMIN)) 4201 return -EACCES; 4202 rv = mddev_lock(mddev); 4203 if (mddev->hold_active == UNTIL_IOCTL) 4204 mddev->hold_active = 0; 4205 if (!rv) { 4206 rv = entry->store(mddev, page, length); 4207 mddev_unlock(mddev); 4208 } 4209 return rv; 4210 } 4211 4212 static void md_free(struct kobject *ko) 4213 { 4214 mddev_t *mddev = container_of(ko, mddev_t, kobj); 4215 4216 if (mddev->sysfs_state) 4217 sysfs_put(mddev->sysfs_state); 4218 4219 if (mddev->gendisk) { 4220 del_gendisk(mddev->gendisk); 4221 put_disk(mddev->gendisk); 4222 } 4223 if (mddev->queue) 4224 blk_cleanup_queue(mddev->queue); 4225 4226 kfree(mddev); 4227 } 4228 4229 static const struct sysfs_ops md_sysfs_ops = { 4230 .show = md_attr_show, 4231 .store = md_attr_store, 4232 }; 4233 static struct kobj_type md_ktype = { 4234 .release = md_free, 4235 .sysfs_ops = &md_sysfs_ops, 4236 .default_attrs = md_default_attrs, 4237 }; 4238 4239 int mdp_major = 0; 4240 4241 static void mddev_delayed_delete(struct work_struct *ws) 4242 { 4243 mddev_t *mddev = container_of(ws, mddev_t, del_work); 4244 4245 sysfs_remove_group(&mddev->kobj, &md_bitmap_group); 4246 kobject_del(&mddev->kobj); 4247 kobject_put(&mddev->kobj); 4248 } 4249 4250 static int md_alloc(dev_t dev, char *name) 4251 { 4252 static DEFINE_MUTEX(disks_mutex); 4253 mddev_t *mddev = mddev_find(dev); 4254 struct gendisk *disk; 4255 int partitioned; 4256 int shift; 4257 int unit; 4258 int error; 4259 4260 if (!mddev) 4261 return -ENODEV; 4262 4263 partitioned = (MAJOR(mddev->unit) != MD_MAJOR); 4264 shift = partitioned ? MdpMinorShift : 0; 4265 unit = MINOR(mddev->unit) >> shift; 4266 4267 /* wait for any previous instance of this device to be 4268 * completely removed (mddev_delayed_delete). 4269 */ 4270 flush_workqueue(md_misc_wq); 4271 4272 mutex_lock(&disks_mutex); 4273 error = -EEXIST; 4274 if (mddev->gendisk) 4275 goto abort; 4276 4277 if (name) { 4278 /* Need to ensure that 'name' is not a duplicate. 4279 */ 4280 mddev_t *mddev2; 4281 spin_lock(&all_mddevs_lock); 4282 4283 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) 4284 if (mddev2->gendisk && 4285 strcmp(mddev2->gendisk->disk_name, name) == 0) { 4286 spin_unlock(&all_mddevs_lock); 4287 goto abort; 4288 } 4289 spin_unlock(&all_mddevs_lock); 4290 } 4291 4292 error = -ENOMEM; 4293 mddev->queue = blk_alloc_queue(GFP_KERNEL); 4294 if (!mddev->queue) 4295 goto abort; 4296 mddev->queue->queuedata = mddev; 4297 4298 blk_queue_make_request(mddev->queue, md_make_request); 4299 4300 disk = alloc_disk(1 << shift); 4301 if (!disk) { 4302 blk_cleanup_queue(mddev->queue); 4303 mddev->queue = NULL; 4304 goto abort; 4305 } 4306 disk->major = MAJOR(mddev->unit); 4307 disk->first_minor = unit << shift; 4308 if (name) 4309 strcpy(disk->disk_name, name); 4310 else if (partitioned) 4311 sprintf(disk->disk_name, "md_d%d", unit); 4312 else 4313 sprintf(disk->disk_name, "md%d", unit); 4314 disk->fops = &md_fops; 4315 disk->private_data = mddev; 4316 disk->queue = mddev->queue; 4317 /* Allow extended partitions. This makes the 4318 * 'mdp' device redundant, but we can't really 4319 * remove it now. 4320 */ 4321 disk->flags |= GENHD_FL_EXT_DEVT; 4322 add_disk(disk); 4323 mddev->gendisk = disk; 4324 error = kobject_init_and_add(&mddev->kobj, &md_ktype, 4325 &disk_to_dev(disk)->kobj, "%s", "md"); 4326 if (error) { 4327 /* This isn't possible, but as kobject_init_and_add is marked 4328 * __must_check, we must do something with the result 4329 */ 4330 printk(KERN_WARNING "md: cannot register %s/md - name in use\n", 4331 disk->disk_name); 4332 error = 0; 4333 } 4334 if (mddev->kobj.sd && 4335 sysfs_create_group(&mddev->kobj, &md_bitmap_group)) 4336 printk(KERN_DEBUG "pointless warning\n"); 4337 4338 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); 4339 abort: 4340 mutex_unlock(&disks_mutex); 4341 if (!error && mddev->kobj.sd) { 4342 kobject_uevent(&mddev->kobj, KOBJ_ADD); 4343 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); 4344 } 4345 mddev_put(mddev); 4346 return error; 4347 } 4348 4349 static struct kobject *md_probe(dev_t dev, int *part, void *data) 4350 { 4351 md_alloc(dev, NULL); 4352 return NULL; 4353 } 4354 4355 static int add_named_array(const char *val, struct kernel_param *kp) 4356 { 4357 /* val must be "md_*" where * is not all digits. 4358 * We allocate an array with a large free minor number, and 4359 * set the name to val. val must not already be an active name. 4360 */ 4361 int len = strlen(val); 4362 char buf[DISK_NAME_LEN]; 4363 4364 while (len && val[len-1] == '\n') 4365 len--; 4366 if (len >= DISK_NAME_LEN) 4367 return -E2BIG; 4368 strlcpy(buf, val, len+1); 4369 if (strncmp(buf, "md_", 3) != 0) 4370 return -EINVAL; 4371 return md_alloc(0, buf); 4372 } 4373 4374 static void md_safemode_timeout(unsigned long data) 4375 { 4376 mddev_t *mddev = (mddev_t *) data; 4377 4378 if (!atomic_read(&mddev->writes_pending)) { 4379 mddev->safemode = 1; 4380 if (mddev->external) 4381 sysfs_notify_dirent_safe(mddev->sysfs_state); 4382 } 4383 md_wakeup_thread(mddev->thread); 4384 } 4385 4386 static int start_dirty_degraded; 4387 4388 int md_run(mddev_t *mddev) 4389 { 4390 int err; 4391 mdk_rdev_t *rdev; 4392 struct mdk_personality *pers; 4393 4394 if (list_empty(&mddev->disks)) 4395 /* cannot run an array with no devices.. */ 4396 return -EINVAL; 4397 4398 if (mddev->pers) 4399 return -EBUSY; 4400 /* Cannot run until previous stop completes properly */ 4401 if (mddev->sysfs_active) 4402 return -EBUSY; 4403 4404 /* 4405 * Analyze all RAID superblock(s) 4406 */ 4407 if (!mddev->raid_disks) { 4408 if (!mddev->persistent) 4409 return -EINVAL; 4410 analyze_sbs(mddev); 4411 } 4412 4413 if (mddev->level != LEVEL_NONE) 4414 request_module("md-level-%d", mddev->level); 4415 else if (mddev->clevel[0]) 4416 request_module("md-%s", mddev->clevel); 4417 4418 /* 4419 * Drop all container device buffers, from now on 4420 * the only valid external interface is through the md 4421 * device. 4422 */ 4423 list_for_each_entry(rdev, &mddev->disks, same_set) { 4424 if (test_bit(Faulty, &rdev->flags)) 4425 continue; 4426 sync_blockdev(rdev->bdev); 4427 invalidate_bdev(rdev->bdev); 4428 4429 /* perform some consistency tests on the device. 4430 * We don't want the data to overlap the metadata, 4431 * Internal Bitmap issues have been handled elsewhere. 4432 */ 4433 if (rdev->data_offset < rdev->sb_start) { 4434 if (mddev->dev_sectors && 4435 rdev->data_offset + mddev->dev_sectors 4436 > rdev->sb_start) { 4437 printk("md: %s: data overlaps metadata\n", 4438 mdname(mddev)); 4439 return -EINVAL; 4440 } 4441 } else { 4442 if (rdev->sb_start + rdev->sb_size/512 4443 > rdev->data_offset) { 4444 printk("md: %s: metadata overlaps data\n", 4445 mdname(mddev)); 4446 return -EINVAL; 4447 } 4448 } 4449 sysfs_notify_dirent_safe(rdev->sysfs_state); 4450 } 4451 4452 if (mddev->bio_set == NULL) 4453 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev)); 4454 4455 spin_lock(&pers_lock); 4456 pers = find_pers(mddev->level, mddev->clevel); 4457 if (!pers || !try_module_get(pers->owner)) { 4458 spin_unlock(&pers_lock); 4459 if (mddev->level != LEVEL_NONE) 4460 printk(KERN_WARNING "md: personality for level %d is not loaded!\n", 4461 mddev->level); 4462 else 4463 printk(KERN_WARNING "md: personality for level %s is not loaded!\n", 4464 mddev->clevel); 4465 return -EINVAL; 4466 } 4467 mddev->pers = pers; 4468 spin_unlock(&pers_lock); 4469 if (mddev->level != pers->level) { 4470 mddev->level = pers->level; 4471 mddev->new_level = pers->level; 4472 } 4473 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 4474 4475 if (mddev->reshape_position != MaxSector && 4476 pers->start_reshape == NULL) { 4477 /* This personality cannot handle reshaping... */ 4478 mddev->pers = NULL; 4479 module_put(pers->owner); 4480 return -EINVAL; 4481 } 4482 4483 if (pers->sync_request) { 4484 /* Warn if this is a potentially silly 4485 * configuration. 4486 */ 4487 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 4488 mdk_rdev_t *rdev2; 4489 int warned = 0; 4490 4491 list_for_each_entry(rdev, &mddev->disks, same_set) 4492 list_for_each_entry(rdev2, &mddev->disks, same_set) { 4493 if (rdev < rdev2 && 4494 rdev->bdev->bd_contains == 4495 rdev2->bdev->bd_contains) { 4496 printk(KERN_WARNING 4497 "%s: WARNING: %s appears to be" 4498 " on the same physical disk as" 4499 " %s.\n", 4500 mdname(mddev), 4501 bdevname(rdev->bdev,b), 4502 bdevname(rdev2->bdev,b2)); 4503 warned = 1; 4504 } 4505 } 4506 4507 if (warned) 4508 printk(KERN_WARNING 4509 "True protection against single-disk" 4510 " failure might be compromised.\n"); 4511 } 4512 4513 mddev->recovery = 0; 4514 /* may be over-ridden by personality */ 4515 mddev->resync_max_sectors = mddev->dev_sectors; 4516 4517 mddev->ok_start_degraded = start_dirty_degraded; 4518 4519 if (start_readonly && mddev->ro == 0) 4520 mddev->ro = 2; /* read-only, but switch on first write */ 4521 4522 err = mddev->pers->run(mddev); 4523 if (err) 4524 printk(KERN_ERR "md: pers->run() failed ...\n"); 4525 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { 4526 WARN_ONCE(!mddev->external_size, "%s: default size too small," 4527 " but 'external_size' not in effect?\n", __func__); 4528 printk(KERN_ERR 4529 "md: invalid array_size %llu > default size %llu\n", 4530 (unsigned long long)mddev->array_sectors / 2, 4531 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); 4532 err = -EINVAL; 4533 mddev->pers->stop(mddev); 4534 } 4535 if (err == 0 && mddev->pers->sync_request) { 4536 err = bitmap_create(mddev); 4537 if (err) { 4538 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 4539 mdname(mddev), err); 4540 mddev->pers->stop(mddev); 4541 } 4542 } 4543 if (err) { 4544 module_put(mddev->pers->owner); 4545 mddev->pers = NULL; 4546 bitmap_destroy(mddev); 4547 return err; 4548 } 4549 if (mddev->pers->sync_request) { 4550 if (mddev->kobj.sd && 4551 sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 4552 printk(KERN_WARNING 4553 "md: cannot register extra attributes for %s\n", 4554 mdname(mddev)); 4555 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); 4556 } else if (mddev->ro == 2) /* auto-readonly not meaningful */ 4557 mddev->ro = 0; 4558 4559 atomic_set(&mddev->writes_pending,0); 4560 atomic_set(&mddev->max_corr_read_errors, 4561 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); 4562 mddev->safemode = 0; 4563 mddev->safemode_timer.function = md_safemode_timeout; 4564 mddev->safemode_timer.data = (unsigned long) mddev; 4565 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ 4566 mddev->in_sync = 1; 4567 4568 list_for_each_entry(rdev, &mddev->disks, same_set) 4569 if (rdev->raid_disk >= 0) { 4570 char nm[20]; 4571 sprintf(nm, "rd%d", rdev->raid_disk); 4572 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) 4573 /* failure here is OK */; 4574 } 4575 4576 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4577 4578 if (mddev->flags) 4579 md_update_sb(mddev, 0); 4580 4581 md_wakeup_thread(mddev->thread); 4582 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 4583 4584 md_new_event(mddev); 4585 sysfs_notify_dirent_safe(mddev->sysfs_state); 4586 sysfs_notify_dirent_safe(mddev->sysfs_action); 4587 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4588 return 0; 4589 } 4590 EXPORT_SYMBOL_GPL(md_run); 4591 4592 static int do_md_run(mddev_t *mddev) 4593 { 4594 int err; 4595 4596 err = md_run(mddev); 4597 if (err) 4598 goto out; 4599 err = bitmap_load(mddev); 4600 if (err) { 4601 bitmap_destroy(mddev); 4602 goto out; 4603 } 4604 set_capacity(mddev->gendisk, mddev->array_sectors); 4605 revalidate_disk(mddev->gendisk); 4606 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4607 out: 4608 return err; 4609 } 4610 4611 static int restart_array(mddev_t *mddev) 4612 { 4613 struct gendisk *disk = mddev->gendisk; 4614 4615 /* Complain if it has no devices */ 4616 if (list_empty(&mddev->disks)) 4617 return -ENXIO; 4618 if (!mddev->pers) 4619 return -EINVAL; 4620 if (!mddev->ro) 4621 return -EBUSY; 4622 mddev->safemode = 0; 4623 mddev->ro = 0; 4624 set_disk_ro(disk, 0); 4625 printk(KERN_INFO "md: %s switched to read-write mode.\n", 4626 mdname(mddev)); 4627 /* Kick recovery or resync if necessary */ 4628 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4629 md_wakeup_thread(mddev->thread); 4630 md_wakeup_thread(mddev->sync_thread); 4631 sysfs_notify_dirent_safe(mddev->sysfs_state); 4632 return 0; 4633 } 4634 4635 /* similar to deny_write_access, but accounts for our holding a reference 4636 * to the file ourselves */ 4637 static int deny_bitmap_write_access(struct file * file) 4638 { 4639 struct inode *inode = file->f_mapping->host; 4640 4641 spin_lock(&inode->i_lock); 4642 if (atomic_read(&inode->i_writecount) > 1) { 4643 spin_unlock(&inode->i_lock); 4644 return -ETXTBSY; 4645 } 4646 atomic_set(&inode->i_writecount, -1); 4647 spin_unlock(&inode->i_lock); 4648 4649 return 0; 4650 } 4651 4652 void restore_bitmap_write_access(struct file *file) 4653 { 4654 struct inode *inode = file->f_mapping->host; 4655 4656 spin_lock(&inode->i_lock); 4657 atomic_set(&inode->i_writecount, 1); 4658 spin_unlock(&inode->i_lock); 4659 } 4660 4661 static void md_clean(mddev_t *mddev) 4662 { 4663 mddev->array_sectors = 0; 4664 mddev->external_size = 0; 4665 mddev->dev_sectors = 0; 4666 mddev->raid_disks = 0; 4667 mddev->recovery_cp = 0; 4668 mddev->resync_min = 0; 4669 mddev->resync_max = MaxSector; 4670 mddev->reshape_position = MaxSector; 4671 mddev->external = 0; 4672 mddev->persistent = 0; 4673 mddev->level = LEVEL_NONE; 4674 mddev->clevel[0] = 0; 4675 mddev->flags = 0; 4676 mddev->ro = 0; 4677 mddev->metadata_type[0] = 0; 4678 mddev->chunk_sectors = 0; 4679 mddev->ctime = mddev->utime = 0; 4680 mddev->layout = 0; 4681 mddev->max_disks = 0; 4682 mddev->events = 0; 4683 mddev->can_decrease_events = 0; 4684 mddev->delta_disks = 0; 4685 mddev->new_level = LEVEL_NONE; 4686 mddev->new_layout = 0; 4687 mddev->new_chunk_sectors = 0; 4688 mddev->curr_resync = 0; 4689 mddev->resync_mismatches = 0; 4690 mddev->suspend_lo = mddev->suspend_hi = 0; 4691 mddev->sync_speed_min = mddev->sync_speed_max = 0; 4692 mddev->recovery = 0; 4693 mddev->in_sync = 0; 4694 mddev->degraded = 0; 4695 mddev->safemode = 0; 4696 mddev->bitmap_info.offset = 0; 4697 mddev->bitmap_info.default_offset = 0; 4698 mddev->bitmap_info.chunksize = 0; 4699 mddev->bitmap_info.daemon_sleep = 0; 4700 mddev->bitmap_info.max_write_behind = 0; 4701 mddev->plug = NULL; 4702 } 4703 4704 void md_stop_writes(mddev_t *mddev) 4705 { 4706 if (mddev->sync_thread) { 4707 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4708 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4709 md_unregister_thread(mddev->sync_thread); 4710 mddev->sync_thread = NULL; 4711 } 4712 4713 del_timer_sync(&mddev->safemode_timer); 4714 4715 bitmap_flush(mddev); 4716 md_super_wait(mddev); 4717 4718 if (!mddev->in_sync || mddev->flags) { 4719 /* mark array as shutdown cleanly */ 4720 mddev->in_sync = 1; 4721 md_update_sb(mddev, 1); 4722 } 4723 } 4724 EXPORT_SYMBOL_GPL(md_stop_writes); 4725 4726 void md_stop(mddev_t *mddev) 4727 { 4728 mddev->pers->stop(mddev); 4729 if (mddev->pers->sync_request && mddev->to_remove == NULL) 4730 mddev->to_remove = &md_redundancy_group; 4731 module_put(mddev->pers->owner); 4732 mddev->pers = NULL; 4733 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4734 } 4735 EXPORT_SYMBOL_GPL(md_stop); 4736 4737 static int md_set_readonly(mddev_t *mddev, int is_open) 4738 { 4739 int err = 0; 4740 mutex_lock(&mddev->open_mutex); 4741 if (atomic_read(&mddev->openers) > is_open) { 4742 printk("md: %s still in use.\n",mdname(mddev)); 4743 err = -EBUSY; 4744 goto out; 4745 } 4746 if (mddev->pers) { 4747 md_stop_writes(mddev); 4748 4749 err = -ENXIO; 4750 if (mddev->ro==1) 4751 goto out; 4752 mddev->ro = 1; 4753 set_disk_ro(mddev->gendisk, 1); 4754 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4755 sysfs_notify_dirent_safe(mddev->sysfs_state); 4756 err = 0; 4757 } 4758 out: 4759 mutex_unlock(&mddev->open_mutex); 4760 return err; 4761 } 4762 4763 /* mode: 4764 * 0 - completely stop and dis-assemble array 4765 * 2 - stop but do not disassemble array 4766 */ 4767 static int do_md_stop(mddev_t * mddev, int mode, int is_open) 4768 { 4769 struct gendisk *disk = mddev->gendisk; 4770 mdk_rdev_t *rdev; 4771 4772 mutex_lock(&mddev->open_mutex); 4773 if (atomic_read(&mddev->openers) > is_open || 4774 mddev->sysfs_active) { 4775 printk("md: %s still in use.\n",mdname(mddev)); 4776 mutex_unlock(&mddev->open_mutex); 4777 return -EBUSY; 4778 } 4779 4780 if (mddev->pers) { 4781 if (mddev->ro) 4782 set_disk_ro(disk, 0); 4783 4784 md_stop_writes(mddev); 4785 md_stop(mddev); 4786 mddev->queue->merge_bvec_fn = NULL; 4787 mddev->queue->unplug_fn = NULL; 4788 mddev->queue->backing_dev_info.congested_fn = NULL; 4789 4790 /* tell userspace to handle 'inactive' */ 4791 sysfs_notify_dirent_safe(mddev->sysfs_state); 4792 4793 list_for_each_entry(rdev, &mddev->disks, same_set) 4794 if (rdev->raid_disk >= 0) { 4795 char nm[20]; 4796 sprintf(nm, "rd%d", rdev->raid_disk); 4797 sysfs_remove_link(&mddev->kobj, nm); 4798 } 4799 4800 set_capacity(disk, 0); 4801 mutex_unlock(&mddev->open_mutex); 4802 revalidate_disk(disk); 4803 4804 if (mddev->ro) 4805 mddev->ro = 0; 4806 } else 4807 mutex_unlock(&mddev->open_mutex); 4808 /* 4809 * Free resources if final stop 4810 */ 4811 if (mode == 0) { 4812 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 4813 4814 bitmap_destroy(mddev); 4815 if (mddev->bitmap_info.file) { 4816 restore_bitmap_write_access(mddev->bitmap_info.file); 4817 fput(mddev->bitmap_info.file); 4818 mddev->bitmap_info.file = NULL; 4819 } 4820 mddev->bitmap_info.offset = 0; 4821 4822 export_array(mddev); 4823 4824 md_clean(mddev); 4825 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4826 if (mddev->hold_active == UNTIL_STOP) 4827 mddev->hold_active = 0; 4828 } 4829 blk_integrity_unregister(disk); 4830 md_new_event(mddev); 4831 sysfs_notify_dirent_safe(mddev->sysfs_state); 4832 return 0; 4833 } 4834 4835 #ifndef MODULE 4836 static void autorun_array(mddev_t *mddev) 4837 { 4838 mdk_rdev_t *rdev; 4839 int err; 4840 4841 if (list_empty(&mddev->disks)) 4842 return; 4843 4844 printk(KERN_INFO "md: running: "); 4845 4846 list_for_each_entry(rdev, &mddev->disks, same_set) { 4847 char b[BDEVNAME_SIZE]; 4848 printk("<%s>", bdevname(rdev->bdev,b)); 4849 } 4850 printk("\n"); 4851 4852 err = do_md_run(mddev); 4853 if (err) { 4854 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 4855 do_md_stop(mddev, 0, 0); 4856 } 4857 } 4858 4859 /* 4860 * lets try to run arrays based on all disks that have arrived 4861 * until now. (those are in pending_raid_disks) 4862 * 4863 * the method: pick the first pending disk, collect all disks with 4864 * the same UUID, remove all from the pending list and put them into 4865 * the 'same_array' list. Then order this list based on superblock 4866 * update time (freshest comes first), kick out 'old' disks and 4867 * compare superblocks. If everything's fine then run it. 4868 * 4869 * If "unit" is allocated, then bump its reference count 4870 */ 4871 static void autorun_devices(int part) 4872 { 4873 mdk_rdev_t *rdev0, *rdev, *tmp; 4874 mddev_t *mddev; 4875 char b[BDEVNAME_SIZE]; 4876 4877 printk(KERN_INFO "md: autorun ...\n"); 4878 while (!list_empty(&pending_raid_disks)) { 4879 int unit; 4880 dev_t dev; 4881 LIST_HEAD(candidates); 4882 rdev0 = list_entry(pending_raid_disks.next, 4883 mdk_rdev_t, same_set); 4884 4885 printk(KERN_INFO "md: considering %s ...\n", 4886 bdevname(rdev0->bdev,b)); 4887 INIT_LIST_HEAD(&candidates); 4888 rdev_for_each_list(rdev, tmp, &pending_raid_disks) 4889 if (super_90_load(rdev, rdev0, 0) >= 0) { 4890 printk(KERN_INFO "md: adding %s ...\n", 4891 bdevname(rdev->bdev,b)); 4892 list_move(&rdev->same_set, &candidates); 4893 } 4894 /* 4895 * now we have a set of devices, with all of them having 4896 * mostly sane superblocks. It's time to allocate the 4897 * mddev. 4898 */ 4899 if (part) { 4900 dev = MKDEV(mdp_major, 4901 rdev0->preferred_minor << MdpMinorShift); 4902 unit = MINOR(dev) >> MdpMinorShift; 4903 } else { 4904 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 4905 unit = MINOR(dev); 4906 } 4907 if (rdev0->preferred_minor != unit) { 4908 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 4909 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 4910 break; 4911 } 4912 4913 md_probe(dev, NULL, NULL); 4914 mddev = mddev_find(dev); 4915 if (!mddev || !mddev->gendisk) { 4916 if (mddev) 4917 mddev_put(mddev); 4918 printk(KERN_ERR 4919 "md: cannot allocate memory for md drive.\n"); 4920 break; 4921 } 4922 if (mddev_lock(mddev)) 4923 printk(KERN_WARNING "md: %s locked, cannot run\n", 4924 mdname(mddev)); 4925 else if (mddev->raid_disks || mddev->major_version 4926 || !list_empty(&mddev->disks)) { 4927 printk(KERN_WARNING 4928 "md: %s already running, cannot run %s\n", 4929 mdname(mddev), bdevname(rdev0->bdev,b)); 4930 mddev_unlock(mddev); 4931 } else { 4932 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 4933 mddev->persistent = 1; 4934 rdev_for_each_list(rdev, tmp, &candidates) { 4935 list_del_init(&rdev->same_set); 4936 if (bind_rdev_to_array(rdev, mddev)) 4937 export_rdev(rdev); 4938 } 4939 autorun_array(mddev); 4940 mddev_unlock(mddev); 4941 } 4942 /* on success, candidates will be empty, on error 4943 * it won't... 4944 */ 4945 rdev_for_each_list(rdev, tmp, &candidates) { 4946 list_del_init(&rdev->same_set); 4947 export_rdev(rdev); 4948 } 4949 mddev_put(mddev); 4950 } 4951 printk(KERN_INFO "md: ... autorun DONE.\n"); 4952 } 4953 #endif /* !MODULE */ 4954 4955 static int get_version(void __user * arg) 4956 { 4957 mdu_version_t ver; 4958 4959 ver.major = MD_MAJOR_VERSION; 4960 ver.minor = MD_MINOR_VERSION; 4961 ver.patchlevel = MD_PATCHLEVEL_VERSION; 4962 4963 if (copy_to_user(arg, &ver, sizeof(ver))) 4964 return -EFAULT; 4965 4966 return 0; 4967 } 4968 4969 static int get_array_info(mddev_t * mddev, void __user * arg) 4970 { 4971 mdu_array_info_t info; 4972 int nr,working,insync,failed,spare; 4973 mdk_rdev_t *rdev; 4974 4975 nr=working=insync=failed=spare=0; 4976 list_for_each_entry(rdev, &mddev->disks, same_set) { 4977 nr++; 4978 if (test_bit(Faulty, &rdev->flags)) 4979 failed++; 4980 else { 4981 working++; 4982 if (test_bit(In_sync, &rdev->flags)) 4983 insync++; 4984 else 4985 spare++; 4986 } 4987 } 4988 4989 info.major_version = mddev->major_version; 4990 info.minor_version = mddev->minor_version; 4991 info.patch_version = MD_PATCHLEVEL_VERSION; 4992 info.ctime = mddev->ctime; 4993 info.level = mddev->level; 4994 info.size = mddev->dev_sectors / 2; 4995 if (info.size != mddev->dev_sectors / 2) /* overflow */ 4996 info.size = -1; 4997 info.nr_disks = nr; 4998 info.raid_disks = mddev->raid_disks; 4999 info.md_minor = mddev->md_minor; 5000 info.not_persistent= !mddev->persistent; 5001 5002 info.utime = mddev->utime; 5003 info.state = 0; 5004 if (mddev->in_sync) 5005 info.state = (1<<MD_SB_CLEAN); 5006 if (mddev->bitmap && mddev->bitmap_info.offset) 5007 info.state = (1<<MD_SB_BITMAP_PRESENT); 5008 info.active_disks = insync; 5009 info.working_disks = working; 5010 info.failed_disks = failed; 5011 info.spare_disks = spare; 5012 5013 info.layout = mddev->layout; 5014 info.chunk_size = mddev->chunk_sectors << 9; 5015 5016 if (copy_to_user(arg, &info, sizeof(info))) 5017 return -EFAULT; 5018 5019 return 0; 5020 } 5021 5022 static int get_bitmap_file(mddev_t * mddev, void __user * arg) 5023 { 5024 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 5025 char *ptr, *buf = NULL; 5026 int err = -ENOMEM; 5027 5028 if (md_allow_write(mddev)) 5029 file = kmalloc(sizeof(*file), GFP_NOIO); 5030 else 5031 file = kmalloc(sizeof(*file), GFP_KERNEL); 5032 5033 if (!file) 5034 goto out; 5035 5036 /* bitmap disabled, zero the first byte and copy out */ 5037 if (!mddev->bitmap || !mddev->bitmap->file) { 5038 file->pathname[0] = '\0'; 5039 goto copy_out; 5040 } 5041 5042 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 5043 if (!buf) 5044 goto out; 5045 5046 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); 5047 if (IS_ERR(ptr)) 5048 goto out; 5049 5050 strcpy(file->pathname, ptr); 5051 5052 copy_out: 5053 err = 0; 5054 if (copy_to_user(arg, file, sizeof(*file))) 5055 err = -EFAULT; 5056 out: 5057 kfree(buf); 5058 kfree(file); 5059 return err; 5060 } 5061 5062 static int get_disk_info(mddev_t * mddev, void __user * arg) 5063 { 5064 mdu_disk_info_t info; 5065 mdk_rdev_t *rdev; 5066 5067 if (copy_from_user(&info, arg, sizeof(info))) 5068 return -EFAULT; 5069 5070 rdev = find_rdev_nr(mddev, info.number); 5071 if (rdev) { 5072 info.major = MAJOR(rdev->bdev->bd_dev); 5073 info.minor = MINOR(rdev->bdev->bd_dev); 5074 info.raid_disk = rdev->raid_disk; 5075 info.state = 0; 5076 if (test_bit(Faulty, &rdev->flags)) 5077 info.state |= (1<<MD_DISK_FAULTY); 5078 else if (test_bit(In_sync, &rdev->flags)) { 5079 info.state |= (1<<MD_DISK_ACTIVE); 5080 info.state |= (1<<MD_DISK_SYNC); 5081 } 5082 if (test_bit(WriteMostly, &rdev->flags)) 5083 info.state |= (1<<MD_DISK_WRITEMOSTLY); 5084 } else { 5085 info.major = info.minor = 0; 5086 info.raid_disk = -1; 5087 info.state = (1<<MD_DISK_REMOVED); 5088 } 5089 5090 if (copy_to_user(arg, &info, sizeof(info))) 5091 return -EFAULT; 5092 5093 return 0; 5094 } 5095 5096 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) 5097 { 5098 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 5099 mdk_rdev_t *rdev; 5100 dev_t dev = MKDEV(info->major,info->minor); 5101 5102 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 5103 return -EOVERFLOW; 5104 5105 if (!mddev->raid_disks) { 5106 int err; 5107 /* expecting a device which has a superblock */ 5108 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 5109 if (IS_ERR(rdev)) { 5110 printk(KERN_WARNING 5111 "md: md_import_device returned %ld\n", 5112 PTR_ERR(rdev)); 5113 return PTR_ERR(rdev); 5114 } 5115 if (!list_empty(&mddev->disks)) { 5116 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 5117 mdk_rdev_t, same_set); 5118 err = super_types[mddev->major_version] 5119 .load_super(rdev, rdev0, mddev->minor_version); 5120 if (err < 0) { 5121 printk(KERN_WARNING 5122 "md: %s has different UUID to %s\n", 5123 bdevname(rdev->bdev,b), 5124 bdevname(rdev0->bdev,b2)); 5125 export_rdev(rdev); 5126 return -EINVAL; 5127 } 5128 } 5129 err = bind_rdev_to_array(rdev, mddev); 5130 if (err) 5131 export_rdev(rdev); 5132 return err; 5133 } 5134 5135 /* 5136 * add_new_disk can be used once the array is assembled 5137 * to add "hot spares". They must already have a superblock 5138 * written 5139 */ 5140 if (mddev->pers) { 5141 int err; 5142 if (!mddev->pers->hot_add_disk) { 5143 printk(KERN_WARNING 5144 "%s: personality does not support diskops!\n", 5145 mdname(mddev)); 5146 return -EINVAL; 5147 } 5148 if (mddev->persistent) 5149 rdev = md_import_device(dev, mddev->major_version, 5150 mddev->minor_version); 5151 else 5152 rdev = md_import_device(dev, -1, -1); 5153 if (IS_ERR(rdev)) { 5154 printk(KERN_WARNING 5155 "md: md_import_device returned %ld\n", 5156 PTR_ERR(rdev)); 5157 return PTR_ERR(rdev); 5158 } 5159 /* set saved_raid_disk if appropriate */ 5160 if (!mddev->persistent) { 5161 if (info->state & (1<<MD_DISK_SYNC) && 5162 info->raid_disk < mddev->raid_disks) 5163 rdev->raid_disk = info->raid_disk; 5164 else 5165 rdev->raid_disk = -1; 5166 } else 5167 super_types[mddev->major_version]. 5168 validate_super(mddev, rdev); 5169 if (test_bit(In_sync, &rdev->flags)) 5170 rdev->saved_raid_disk = rdev->raid_disk; 5171 else 5172 rdev->saved_raid_disk = -1; 5173 5174 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 5175 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5176 set_bit(WriteMostly, &rdev->flags); 5177 else 5178 clear_bit(WriteMostly, &rdev->flags); 5179 5180 rdev->raid_disk = -1; 5181 err = bind_rdev_to_array(rdev, mddev); 5182 if (!err && !mddev->pers->hot_remove_disk) { 5183 /* If there is hot_add_disk but no hot_remove_disk 5184 * then added disks for geometry changes, 5185 * and should be added immediately. 5186 */ 5187 super_types[mddev->major_version]. 5188 validate_super(mddev, rdev); 5189 err = mddev->pers->hot_add_disk(mddev, rdev); 5190 if (err) 5191 unbind_rdev_from_array(rdev); 5192 } 5193 if (err) 5194 export_rdev(rdev); 5195 else 5196 sysfs_notify_dirent_safe(rdev->sysfs_state); 5197 5198 md_update_sb(mddev, 1); 5199 if (mddev->degraded) 5200 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 5201 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5202 md_wakeup_thread(mddev->thread); 5203 return err; 5204 } 5205 5206 /* otherwise, add_new_disk is only allowed 5207 * for major_version==0 superblocks 5208 */ 5209 if (mddev->major_version != 0) { 5210 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 5211 mdname(mddev)); 5212 return -EINVAL; 5213 } 5214 5215 if (!(info->state & (1<<MD_DISK_FAULTY))) { 5216 int err; 5217 rdev = md_import_device(dev, -1, 0); 5218 if (IS_ERR(rdev)) { 5219 printk(KERN_WARNING 5220 "md: error, md_import_device() returned %ld\n", 5221 PTR_ERR(rdev)); 5222 return PTR_ERR(rdev); 5223 } 5224 rdev->desc_nr = info->number; 5225 if (info->raid_disk < mddev->raid_disks) 5226 rdev->raid_disk = info->raid_disk; 5227 else 5228 rdev->raid_disk = -1; 5229 5230 if (rdev->raid_disk < mddev->raid_disks) 5231 if (info->state & (1<<MD_DISK_SYNC)) 5232 set_bit(In_sync, &rdev->flags); 5233 5234 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5235 set_bit(WriteMostly, &rdev->flags); 5236 5237 if (!mddev->persistent) { 5238 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 5239 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5240 } else 5241 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 5242 rdev->sectors = rdev->sb_start; 5243 5244 err = bind_rdev_to_array(rdev, mddev); 5245 if (err) { 5246 export_rdev(rdev); 5247 return err; 5248 } 5249 } 5250 5251 return 0; 5252 } 5253 5254 static int hot_remove_disk(mddev_t * mddev, dev_t dev) 5255 { 5256 char b[BDEVNAME_SIZE]; 5257 mdk_rdev_t *rdev; 5258 5259 rdev = find_rdev(mddev, dev); 5260 if (!rdev) 5261 return -ENXIO; 5262 5263 if (rdev->raid_disk >= 0) 5264 goto busy; 5265 5266 kick_rdev_from_array(rdev); 5267 md_update_sb(mddev, 1); 5268 md_new_event(mddev); 5269 5270 return 0; 5271 busy: 5272 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", 5273 bdevname(rdev->bdev,b), mdname(mddev)); 5274 return -EBUSY; 5275 } 5276 5277 static int hot_add_disk(mddev_t * mddev, dev_t dev) 5278 { 5279 char b[BDEVNAME_SIZE]; 5280 int err; 5281 mdk_rdev_t *rdev; 5282 5283 if (!mddev->pers) 5284 return -ENODEV; 5285 5286 if (mddev->major_version != 0) { 5287 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 5288 " version-0 superblocks.\n", 5289 mdname(mddev)); 5290 return -EINVAL; 5291 } 5292 if (!mddev->pers->hot_add_disk) { 5293 printk(KERN_WARNING 5294 "%s: personality does not support diskops!\n", 5295 mdname(mddev)); 5296 return -EINVAL; 5297 } 5298 5299 rdev = md_import_device(dev, -1, 0); 5300 if (IS_ERR(rdev)) { 5301 printk(KERN_WARNING 5302 "md: error, md_import_device() returned %ld\n", 5303 PTR_ERR(rdev)); 5304 return -EINVAL; 5305 } 5306 5307 if (mddev->persistent) 5308 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 5309 else 5310 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; 5311 5312 rdev->sectors = rdev->sb_start; 5313 5314 if (test_bit(Faulty, &rdev->flags)) { 5315 printk(KERN_WARNING 5316 "md: can not hot-add faulty %s disk to %s!\n", 5317 bdevname(rdev->bdev,b), mdname(mddev)); 5318 err = -EINVAL; 5319 goto abort_export; 5320 } 5321 clear_bit(In_sync, &rdev->flags); 5322 rdev->desc_nr = -1; 5323 rdev->saved_raid_disk = -1; 5324 err = bind_rdev_to_array(rdev, mddev); 5325 if (err) 5326 goto abort_export; 5327 5328 /* 5329 * The rest should better be atomic, we can have disk failures 5330 * noticed in interrupt contexts ... 5331 */ 5332 5333 rdev->raid_disk = -1; 5334 5335 md_update_sb(mddev, 1); 5336 5337 /* 5338 * Kick recovery, maybe this spare has to be added to the 5339 * array immediately. 5340 */ 5341 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5342 md_wakeup_thread(mddev->thread); 5343 md_new_event(mddev); 5344 return 0; 5345 5346 abort_export: 5347 export_rdev(rdev); 5348 return err; 5349 } 5350 5351 static int set_bitmap_file(mddev_t *mddev, int fd) 5352 { 5353 int err; 5354 5355 if (mddev->pers) { 5356 if (!mddev->pers->quiesce) 5357 return -EBUSY; 5358 if (mddev->recovery || mddev->sync_thread) 5359 return -EBUSY; 5360 /* we should be able to change the bitmap.. */ 5361 } 5362 5363 5364 if (fd >= 0) { 5365 if (mddev->bitmap) 5366 return -EEXIST; /* cannot add when bitmap is present */ 5367 mddev->bitmap_info.file = fget(fd); 5368 5369 if (mddev->bitmap_info.file == NULL) { 5370 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 5371 mdname(mddev)); 5372 return -EBADF; 5373 } 5374 5375 err = deny_bitmap_write_access(mddev->bitmap_info.file); 5376 if (err) { 5377 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 5378 mdname(mddev)); 5379 fput(mddev->bitmap_info.file); 5380 mddev->bitmap_info.file = NULL; 5381 return err; 5382 } 5383 mddev->bitmap_info.offset = 0; /* file overrides offset */ 5384 } else if (mddev->bitmap == NULL) 5385 return -ENOENT; /* cannot remove what isn't there */ 5386 err = 0; 5387 if (mddev->pers) { 5388 mddev->pers->quiesce(mddev, 1); 5389 if (fd >= 0) { 5390 err = bitmap_create(mddev); 5391 if (!err) 5392 err = bitmap_load(mddev); 5393 } 5394 if (fd < 0 || err) { 5395 bitmap_destroy(mddev); 5396 fd = -1; /* make sure to put the file */ 5397 } 5398 mddev->pers->quiesce(mddev, 0); 5399 } 5400 if (fd < 0) { 5401 if (mddev->bitmap_info.file) { 5402 restore_bitmap_write_access(mddev->bitmap_info.file); 5403 fput(mddev->bitmap_info.file); 5404 } 5405 mddev->bitmap_info.file = NULL; 5406 } 5407 5408 return err; 5409 } 5410 5411 /* 5412 * set_array_info is used two different ways 5413 * The original usage is when creating a new array. 5414 * In this usage, raid_disks is > 0 and it together with 5415 * level, size, not_persistent,layout,chunksize determine the 5416 * shape of the array. 5417 * This will always create an array with a type-0.90.0 superblock. 5418 * The newer usage is when assembling an array. 5419 * In this case raid_disks will be 0, and the major_version field is 5420 * use to determine which style super-blocks are to be found on the devices. 5421 * The minor and patch _version numbers are also kept incase the 5422 * super_block handler wishes to interpret them. 5423 */ 5424 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) 5425 { 5426 5427 if (info->raid_disks == 0) { 5428 /* just setting version number for superblock loading */ 5429 if (info->major_version < 0 || 5430 info->major_version >= ARRAY_SIZE(super_types) || 5431 super_types[info->major_version].name == NULL) { 5432 /* maybe try to auto-load a module? */ 5433 printk(KERN_INFO 5434 "md: superblock version %d not known\n", 5435 info->major_version); 5436 return -EINVAL; 5437 } 5438 mddev->major_version = info->major_version; 5439 mddev->minor_version = info->minor_version; 5440 mddev->patch_version = info->patch_version; 5441 mddev->persistent = !info->not_persistent; 5442 /* ensure mddev_put doesn't delete this now that there 5443 * is some minimal configuration. 5444 */ 5445 mddev->ctime = get_seconds(); 5446 return 0; 5447 } 5448 mddev->major_version = MD_MAJOR_VERSION; 5449 mddev->minor_version = MD_MINOR_VERSION; 5450 mddev->patch_version = MD_PATCHLEVEL_VERSION; 5451 mddev->ctime = get_seconds(); 5452 5453 mddev->level = info->level; 5454 mddev->clevel[0] = 0; 5455 mddev->dev_sectors = 2 * (sector_t)info->size; 5456 mddev->raid_disks = info->raid_disks; 5457 /* don't set md_minor, it is determined by which /dev/md* was 5458 * openned 5459 */ 5460 if (info->state & (1<<MD_SB_CLEAN)) 5461 mddev->recovery_cp = MaxSector; 5462 else 5463 mddev->recovery_cp = 0; 5464 mddev->persistent = ! info->not_persistent; 5465 mddev->external = 0; 5466 5467 mddev->layout = info->layout; 5468 mddev->chunk_sectors = info->chunk_size >> 9; 5469 5470 mddev->max_disks = MD_SB_DISKS; 5471 5472 if (mddev->persistent) 5473 mddev->flags = 0; 5474 set_bit(MD_CHANGE_DEVS, &mddev->flags); 5475 5476 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 5477 mddev->bitmap_info.offset = 0; 5478 5479 mddev->reshape_position = MaxSector; 5480 5481 /* 5482 * Generate a 128 bit UUID 5483 */ 5484 get_random_bytes(mddev->uuid, 16); 5485 5486 mddev->new_level = mddev->level; 5487 mddev->new_chunk_sectors = mddev->chunk_sectors; 5488 mddev->new_layout = mddev->layout; 5489 mddev->delta_disks = 0; 5490 5491 return 0; 5492 } 5493 5494 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) 5495 { 5496 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); 5497 5498 if (mddev->external_size) 5499 return; 5500 5501 mddev->array_sectors = array_sectors; 5502 } 5503 EXPORT_SYMBOL(md_set_array_sectors); 5504 5505 static int update_size(mddev_t *mddev, sector_t num_sectors) 5506 { 5507 mdk_rdev_t *rdev; 5508 int rv; 5509 int fit = (num_sectors == 0); 5510 5511 if (mddev->pers->resize == NULL) 5512 return -EINVAL; 5513 /* The "num_sectors" is the number of sectors of each device that 5514 * is used. This can only make sense for arrays with redundancy. 5515 * linear and raid0 always use whatever space is available. We can only 5516 * consider changing this number if no resync or reconstruction is 5517 * happening, and if the new size is acceptable. It must fit before the 5518 * sb_start or, if that is <data_offset, it must fit before the size 5519 * of each device. If num_sectors is zero, we find the largest size 5520 * that fits. 5521 5522 */ 5523 if (mddev->sync_thread) 5524 return -EBUSY; 5525 if (mddev->bitmap) 5526 /* Sorry, cannot grow a bitmap yet, just remove it, 5527 * grow, and re-add. 5528 */ 5529 return -EBUSY; 5530 list_for_each_entry(rdev, &mddev->disks, same_set) { 5531 sector_t avail = rdev->sectors; 5532 5533 if (fit && (num_sectors == 0 || num_sectors > avail)) 5534 num_sectors = avail; 5535 if (avail < num_sectors) 5536 return -ENOSPC; 5537 } 5538 rv = mddev->pers->resize(mddev, num_sectors); 5539 if (!rv) 5540 revalidate_disk(mddev->gendisk); 5541 return rv; 5542 } 5543 5544 static int update_raid_disks(mddev_t *mddev, int raid_disks) 5545 { 5546 int rv; 5547 /* change the number of raid disks */ 5548 if (mddev->pers->check_reshape == NULL) 5549 return -EINVAL; 5550 if (raid_disks <= 0 || 5551 (mddev->max_disks && raid_disks >= mddev->max_disks)) 5552 return -EINVAL; 5553 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 5554 return -EBUSY; 5555 mddev->delta_disks = raid_disks - mddev->raid_disks; 5556 5557 rv = mddev->pers->check_reshape(mddev); 5558 return rv; 5559 } 5560 5561 5562 /* 5563 * update_array_info is used to change the configuration of an 5564 * on-line array. 5565 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 5566 * fields in the info are checked against the array. 5567 * Any differences that cannot be handled will cause an error. 5568 * Normally, only one change can be managed at a time. 5569 */ 5570 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) 5571 { 5572 int rv = 0; 5573 int cnt = 0; 5574 int state = 0; 5575 5576 /* calculate expected state,ignoring low bits */ 5577 if (mddev->bitmap && mddev->bitmap_info.offset) 5578 state |= (1 << MD_SB_BITMAP_PRESENT); 5579 5580 if (mddev->major_version != info->major_version || 5581 mddev->minor_version != info->minor_version || 5582 /* mddev->patch_version != info->patch_version || */ 5583 mddev->ctime != info->ctime || 5584 mddev->level != info->level || 5585 /* mddev->layout != info->layout || */ 5586 !mddev->persistent != info->not_persistent|| 5587 mddev->chunk_sectors != info->chunk_size >> 9 || 5588 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 5589 ((state^info->state) & 0xfffffe00) 5590 ) 5591 return -EINVAL; 5592 /* Check there is only one change */ 5593 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5594 cnt++; 5595 if (mddev->raid_disks != info->raid_disks) 5596 cnt++; 5597 if (mddev->layout != info->layout) 5598 cnt++; 5599 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) 5600 cnt++; 5601 if (cnt == 0) 5602 return 0; 5603 if (cnt > 1) 5604 return -EINVAL; 5605 5606 if (mddev->layout != info->layout) { 5607 /* Change layout 5608 * we don't need to do anything at the md level, the 5609 * personality will take care of it all. 5610 */ 5611 if (mddev->pers->check_reshape == NULL) 5612 return -EINVAL; 5613 else { 5614 mddev->new_layout = info->layout; 5615 rv = mddev->pers->check_reshape(mddev); 5616 if (rv) 5617 mddev->new_layout = mddev->layout; 5618 return rv; 5619 } 5620 } 5621 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5622 rv = update_size(mddev, (sector_t)info->size * 2); 5623 5624 if (mddev->raid_disks != info->raid_disks) 5625 rv = update_raid_disks(mddev, info->raid_disks); 5626 5627 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 5628 if (mddev->pers->quiesce == NULL) 5629 return -EINVAL; 5630 if (mddev->recovery || mddev->sync_thread) 5631 return -EBUSY; 5632 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 5633 /* add the bitmap */ 5634 if (mddev->bitmap) 5635 return -EEXIST; 5636 if (mddev->bitmap_info.default_offset == 0) 5637 return -EINVAL; 5638 mddev->bitmap_info.offset = 5639 mddev->bitmap_info.default_offset; 5640 mddev->pers->quiesce(mddev, 1); 5641 rv = bitmap_create(mddev); 5642 if (!rv) 5643 rv = bitmap_load(mddev); 5644 if (rv) 5645 bitmap_destroy(mddev); 5646 mddev->pers->quiesce(mddev, 0); 5647 } else { 5648 /* remove the bitmap */ 5649 if (!mddev->bitmap) 5650 return -ENOENT; 5651 if (mddev->bitmap->file) 5652 return -EINVAL; 5653 mddev->pers->quiesce(mddev, 1); 5654 bitmap_destroy(mddev); 5655 mddev->pers->quiesce(mddev, 0); 5656 mddev->bitmap_info.offset = 0; 5657 } 5658 } 5659 md_update_sb(mddev, 1); 5660 return rv; 5661 } 5662 5663 static int set_disk_faulty(mddev_t *mddev, dev_t dev) 5664 { 5665 mdk_rdev_t *rdev; 5666 5667 if (mddev->pers == NULL) 5668 return -ENODEV; 5669 5670 rdev = find_rdev(mddev, dev); 5671 if (!rdev) 5672 return -ENODEV; 5673 5674 md_error(mddev, rdev); 5675 return 0; 5676 } 5677 5678 /* 5679 * We have a problem here : there is no easy way to give a CHS 5680 * virtual geometry. We currently pretend that we have a 2 heads 5681 * 4 sectors (with a BIG number of cylinders...). This drives 5682 * dosfs just mad... ;-) 5683 */ 5684 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) 5685 { 5686 mddev_t *mddev = bdev->bd_disk->private_data; 5687 5688 geo->heads = 2; 5689 geo->sectors = 4; 5690 geo->cylinders = mddev->array_sectors / 8; 5691 return 0; 5692 } 5693 5694 static int md_ioctl(struct block_device *bdev, fmode_t mode, 5695 unsigned int cmd, unsigned long arg) 5696 { 5697 int err = 0; 5698 void __user *argp = (void __user *)arg; 5699 mddev_t *mddev = NULL; 5700 int ro; 5701 5702 if (!capable(CAP_SYS_ADMIN)) 5703 return -EACCES; 5704 5705 /* 5706 * Commands dealing with the RAID driver but not any 5707 * particular array: 5708 */ 5709 switch (cmd) 5710 { 5711 case RAID_VERSION: 5712 err = get_version(argp); 5713 goto done; 5714 5715 case PRINT_RAID_DEBUG: 5716 err = 0; 5717 md_print_devices(); 5718 goto done; 5719 5720 #ifndef MODULE 5721 case RAID_AUTORUN: 5722 err = 0; 5723 autostart_arrays(arg); 5724 goto done; 5725 #endif 5726 default:; 5727 } 5728 5729 /* 5730 * Commands creating/starting a new array: 5731 */ 5732 5733 mddev = bdev->bd_disk->private_data; 5734 5735 if (!mddev) { 5736 BUG(); 5737 goto abort; 5738 } 5739 5740 err = mddev_lock(mddev); 5741 if (err) { 5742 printk(KERN_INFO 5743 "md: ioctl lock interrupted, reason %d, cmd %d\n", 5744 err, cmd); 5745 goto abort; 5746 } 5747 5748 switch (cmd) 5749 { 5750 case SET_ARRAY_INFO: 5751 { 5752 mdu_array_info_t info; 5753 if (!arg) 5754 memset(&info, 0, sizeof(info)); 5755 else if (copy_from_user(&info, argp, sizeof(info))) { 5756 err = -EFAULT; 5757 goto abort_unlock; 5758 } 5759 if (mddev->pers) { 5760 err = update_array_info(mddev, &info); 5761 if (err) { 5762 printk(KERN_WARNING "md: couldn't update" 5763 " array info. %d\n", err); 5764 goto abort_unlock; 5765 } 5766 goto done_unlock; 5767 } 5768 if (!list_empty(&mddev->disks)) { 5769 printk(KERN_WARNING 5770 "md: array %s already has disks!\n", 5771 mdname(mddev)); 5772 err = -EBUSY; 5773 goto abort_unlock; 5774 } 5775 if (mddev->raid_disks) { 5776 printk(KERN_WARNING 5777 "md: array %s already initialised!\n", 5778 mdname(mddev)); 5779 err = -EBUSY; 5780 goto abort_unlock; 5781 } 5782 err = set_array_info(mddev, &info); 5783 if (err) { 5784 printk(KERN_WARNING "md: couldn't set" 5785 " array info. %d\n", err); 5786 goto abort_unlock; 5787 } 5788 } 5789 goto done_unlock; 5790 5791 default:; 5792 } 5793 5794 /* 5795 * Commands querying/configuring an existing array: 5796 */ 5797 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 5798 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ 5799 if ((!mddev->raid_disks && !mddev->external) 5800 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 5801 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE 5802 && cmd != GET_BITMAP_FILE) { 5803 err = -ENODEV; 5804 goto abort_unlock; 5805 } 5806 5807 /* 5808 * Commands even a read-only array can execute: 5809 */ 5810 switch (cmd) 5811 { 5812 case GET_ARRAY_INFO: 5813 err = get_array_info(mddev, argp); 5814 goto done_unlock; 5815 5816 case GET_BITMAP_FILE: 5817 err = get_bitmap_file(mddev, argp); 5818 goto done_unlock; 5819 5820 case GET_DISK_INFO: 5821 err = get_disk_info(mddev, argp); 5822 goto done_unlock; 5823 5824 case RESTART_ARRAY_RW: 5825 err = restart_array(mddev); 5826 goto done_unlock; 5827 5828 case STOP_ARRAY: 5829 err = do_md_stop(mddev, 0, 1); 5830 goto done_unlock; 5831 5832 case STOP_ARRAY_RO: 5833 err = md_set_readonly(mddev, 1); 5834 goto done_unlock; 5835 5836 case BLKROSET: 5837 if (get_user(ro, (int __user *)(arg))) { 5838 err = -EFAULT; 5839 goto done_unlock; 5840 } 5841 err = -EINVAL; 5842 5843 /* if the bdev is going readonly the value of mddev->ro 5844 * does not matter, no writes are coming 5845 */ 5846 if (ro) 5847 goto done_unlock; 5848 5849 /* are we are already prepared for writes? */ 5850 if (mddev->ro != 1) 5851 goto done_unlock; 5852 5853 /* transitioning to readauto need only happen for 5854 * arrays that call md_write_start 5855 */ 5856 if (mddev->pers) { 5857 err = restart_array(mddev); 5858 if (err == 0) { 5859 mddev->ro = 2; 5860 set_disk_ro(mddev->gendisk, 0); 5861 } 5862 } 5863 goto done_unlock; 5864 } 5865 5866 /* 5867 * The remaining ioctls are changing the state of the 5868 * superblock, so we do not allow them on read-only arrays. 5869 * However non-MD ioctls (e.g. get-size) will still come through 5870 * here and hit the 'default' below, so only disallow 5871 * 'md' ioctls, and switch to rw mode if started auto-readonly. 5872 */ 5873 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { 5874 if (mddev->ro == 2) { 5875 mddev->ro = 0; 5876 sysfs_notify_dirent_safe(mddev->sysfs_state); 5877 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5878 md_wakeup_thread(mddev->thread); 5879 } else { 5880 err = -EROFS; 5881 goto abort_unlock; 5882 } 5883 } 5884 5885 switch (cmd) 5886 { 5887 case ADD_NEW_DISK: 5888 { 5889 mdu_disk_info_t info; 5890 if (copy_from_user(&info, argp, sizeof(info))) 5891 err = -EFAULT; 5892 else 5893 err = add_new_disk(mddev, &info); 5894 goto done_unlock; 5895 } 5896 5897 case HOT_REMOVE_DISK: 5898 err = hot_remove_disk(mddev, new_decode_dev(arg)); 5899 goto done_unlock; 5900 5901 case HOT_ADD_DISK: 5902 err = hot_add_disk(mddev, new_decode_dev(arg)); 5903 goto done_unlock; 5904 5905 case SET_DISK_FAULTY: 5906 err = set_disk_faulty(mddev, new_decode_dev(arg)); 5907 goto done_unlock; 5908 5909 case RUN_ARRAY: 5910 err = do_md_run(mddev); 5911 goto done_unlock; 5912 5913 case SET_BITMAP_FILE: 5914 err = set_bitmap_file(mddev, (int)arg); 5915 goto done_unlock; 5916 5917 default: 5918 err = -EINVAL; 5919 goto abort_unlock; 5920 } 5921 5922 done_unlock: 5923 abort_unlock: 5924 if (mddev->hold_active == UNTIL_IOCTL && 5925 err != -EINVAL) 5926 mddev->hold_active = 0; 5927 mddev_unlock(mddev); 5928 5929 return err; 5930 done: 5931 if (err) 5932 MD_BUG(); 5933 abort: 5934 return err; 5935 } 5936 #ifdef CONFIG_COMPAT 5937 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, 5938 unsigned int cmd, unsigned long arg) 5939 { 5940 switch (cmd) { 5941 case HOT_REMOVE_DISK: 5942 case HOT_ADD_DISK: 5943 case SET_DISK_FAULTY: 5944 case SET_BITMAP_FILE: 5945 /* These take in integer arg, do not convert */ 5946 break; 5947 default: 5948 arg = (unsigned long)compat_ptr(arg); 5949 break; 5950 } 5951 5952 return md_ioctl(bdev, mode, cmd, arg); 5953 } 5954 #endif /* CONFIG_COMPAT */ 5955 5956 static int md_open(struct block_device *bdev, fmode_t mode) 5957 { 5958 /* 5959 * Succeed if we can lock the mddev, which confirms that 5960 * it isn't being stopped right now. 5961 */ 5962 mddev_t *mddev = mddev_find(bdev->bd_dev); 5963 int err; 5964 5965 if (mddev->gendisk != bdev->bd_disk) { 5966 /* we are racing with mddev_put which is discarding this 5967 * bd_disk. 5968 */ 5969 mddev_put(mddev); 5970 /* Wait until bdev->bd_disk is definitely gone */ 5971 flush_workqueue(md_misc_wq); 5972 /* Then retry the open from the top */ 5973 return -ERESTARTSYS; 5974 } 5975 BUG_ON(mddev != bdev->bd_disk->private_data); 5976 5977 if ((err = mutex_lock_interruptible(&mddev->open_mutex))) 5978 goto out; 5979 5980 err = 0; 5981 atomic_inc(&mddev->openers); 5982 mutex_unlock(&mddev->open_mutex); 5983 5984 check_disk_size_change(mddev->gendisk, bdev); 5985 out: 5986 return err; 5987 } 5988 5989 static int md_release(struct gendisk *disk, fmode_t mode) 5990 { 5991 mddev_t *mddev = disk->private_data; 5992 5993 BUG_ON(!mddev); 5994 atomic_dec(&mddev->openers); 5995 mddev_put(mddev); 5996 5997 return 0; 5998 } 5999 static const struct block_device_operations md_fops = 6000 { 6001 .owner = THIS_MODULE, 6002 .open = md_open, 6003 .release = md_release, 6004 .ioctl = md_ioctl, 6005 #ifdef CONFIG_COMPAT 6006 .compat_ioctl = md_compat_ioctl, 6007 #endif 6008 .getgeo = md_getgeo, 6009 }; 6010 6011 static int md_thread(void * arg) 6012 { 6013 mdk_thread_t *thread = arg; 6014 6015 /* 6016 * md_thread is a 'system-thread', it's priority should be very 6017 * high. We avoid resource deadlocks individually in each 6018 * raid personality. (RAID5 does preallocation) We also use RR and 6019 * the very same RT priority as kswapd, thus we will never get 6020 * into a priority inversion deadlock. 6021 * 6022 * we definitely have to have equal or higher priority than 6023 * bdflush, otherwise bdflush will deadlock if there are too 6024 * many dirty RAID5 blocks. 6025 */ 6026 6027 allow_signal(SIGKILL); 6028 while (!kthread_should_stop()) { 6029 6030 /* We need to wait INTERRUPTIBLE so that 6031 * we don't add to the load-average. 6032 * That means we need to be sure no signals are 6033 * pending 6034 */ 6035 if (signal_pending(current)) 6036 flush_signals(current); 6037 6038 wait_event_interruptible_timeout 6039 (thread->wqueue, 6040 test_bit(THREAD_WAKEUP, &thread->flags) 6041 || kthread_should_stop(), 6042 thread->timeout); 6043 6044 if (test_and_clear_bit(THREAD_WAKEUP, &thread->flags)) 6045 thread->run(thread->mddev); 6046 } 6047 6048 return 0; 6049 } 6050 6051 void md_wakeup_thread(mdk_thread_t *thread) 6052 { 6053 if (thread) { 6054 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); 6055 set_bit(THREAD_WAKEUP, &thread->flags); 6056 wake_up(&thread->wqueue); 6057 } 6058 } 6059 6060 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, 6061 const char *name) 6062 { 6063 mdk_thread_t *thread; 6064 6065 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); 6066 if (!thread) 6067 return NULL; 6068 6069 init_waitqueue_head(&thread->wqueue); 6070 6071 thread->run = run; 6072 thread->mddev = mddev; 6073 thread->timeout = MAX_SCHEDULE_TIMEOUT; 6074 thread->tsk = kthread_run(md_thread, thread, 6075 "%s_%s", 6076 mdname(thread->mddev), 6077 name ?: mddev->pers->name); 6078 if (IS_ERR(thread->tsk)) { 6079 kfree(thread); 6080 return NULL; 6081 } 6082 return thread; 6083 } 6084 6085 void md_unregister_thread(mdk_thread_t *thread) 6086 { 6087 if (!thread) 6088 return; 6089 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); 6090 6091 kthread_stop(thread->tsk); 6092 kfree(thread); 6093 } 6094 6095 void md_error(mddev_t *mddev, mdk_rdev_t *rdev) 6096 { 6097 if (!mddev) { 6098 MD_BUG(); 6099 return; 6100 } 6101 6102 if (!rdev || test_bit(Faulty, &rdev->flags)) 6103 return; 6104 6105 if (mddev->external) 6106 set_bit(Blocked, &rdev->flags); 6107 /* 6108 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", 6109 mdname(mddev), 6110 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), 6111 __builtin_return_address(0),__builtin_return_address(1), 6112 __builtin_return_address(2),__builtin_return_address(3)); 6113 */ 6114 if (!mddev->pers) 6115 return; 6116 if (!mddev->pers->error_handler) 6117 return; 6118 mddev->pers->error_handler(mddev,rdev); 6119 if (mddev->degraded) 6120 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6121 sysfs_notify_dirent_safe(rdev->sysfs_state); 6122 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6123 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6124 md_wakeup_thread(mddev->thread); 6125 if (mddev->event_work.func) 6126 queue_work(md_misc_wq, &mddev->event_work); 6127 md_new_event_inintr(mddev); 6128 } 6129 6130 /* seq_file implementation /proc/mdstat */ 6131 6132 static void status_unused(struct seq_file *seq) 6133 { 6134 int i = 0; 6135 mdk_rdev_t *rdev; 6136 6137 seq_printf(seq, "unused devices: "); 6138 6139 list_for_each_entry(rdev, &pending_raid_disks, same_set) { 6140 char b[BDEVNAME_SIZE]; 6141 i++; 6142 seq_printf(seq, "%s ", 6143 bdevname(rdev->bdev,b)); 6144 } 6145 if (!i) 6146 seq_printf(seq, "<none>"); 6147 6148 seq_printf(seq, "\n"); 6149 } 6150 6151 6152 static void status_resync(struct seq_file *seq, mddev_t * mddev) 6153 { 6154 sector_t max_sectors, resync, res; 6155 unsigned long dt, db; 6156 sector_t rt; 6157 int scale; 6158 unsigned int per_milli; 6159 6160 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); 6161 6162 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 6163 max_sectors = mddev->resync_max_sectors; 6164 else 6165 max_sectors = mddev->dev_sectors; 6166 6167 /* 6168 * Should not happen. 6169 */ 6170 if (!max_sectors) { 6171 MD_BUG(); 6172 return; 6173 } 6174 /* Pick 'scale' such that (resync>>scale)*1000 will fit 6175 * in a sector_t, and (max_sectors>>scale) will fit in a 6176 * u32, as those are the requirements for sector_div. 6177 * Thus 'scale' must be at least 10 6178 */ 6179 scale = 10; 6180 if (sizeof(sector_t) > sizeof(unsigned long)) { 6181 while ( max_sectors/2 > (1ULL<<(scale+32))) 6182 scale++; 6183 } 6184 res = (resync>>scale)*1000; 6185 sector_div(res, (u32)((max_sectors>>scale)+1)); 6186 6187 per_milli = res; 6188 { 6189 int i, x = per_milli/50, y = 20-x; 6190 seq_printf(seq, "["); 6191 for (i = 0; i < x; i++) 6192 seq_printf(seq, "="); 6193 seq_printf(seq, ">"); 6194 for (i = 0; i < y; i++) 6195 seq_printf(seq, "."); 6196 seq_printf(seq, "] "); 6197 } 6198 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", 6199 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? 6200 "reshape" : 6201 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? 6202 "check" : 6203 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 6204 "resync" : "recovery"))), 6205 per_milli/10, per_milli % 10, 6206 (unsigned long long) resync/2, 6207 (unsigned long long) max_sectors/2); 6208 6209 /* 6210 * dt: time from mark until now 6211 * db: blocks written from mark until now 6212 * rt: remaining time 6213 * 6214 * rt is a sector_t, so could be 32bit or 64bit. 6215 * So we divide before multiply in case it is 32bit and close 6216 * to the limit. 6217 * We scale the divisor (db) by 32 to avoid loosing precision 6218 * near the end of resync when the number of remaining sectors 6219 * is close to 'db'. 6220 * We then divide rt by 32 after multiplying by db to compensate. 6221 * The '+1' avoids division by zero if db is very small. 6222 */ 6223 dt = ((jiffies - mddev->resync_mark) / HZ); 6224 if (!dt) dt++; 6225 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) 6226 - mddev->resync_mark_cnt; 6227 6228 rt = max_sectors - resync; /* number of remaining sectors */ 6229 sector_div(rt, db/32+1); 6230 rt *= dt; 6231 rt >>= 5; 6232 6233 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, 6234 ((unsigned long)rt % 60)/6); 6235 6236 seq_printf(seq, " speed=%ldK/sec", db/2/dt); 6237 } 6238 6239 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 6240 { 6241 struct list_head *tmp; 6242 loff_t l = *pos; 6243 mddev_t *mddev; 6244 6245 if (l >= 0x10000) 6246 return NULL; 6247 if (!l--) 6248 /* header */ 6249 return (void*)1; 6250 6251 spin_lock(&all_mddevs_lock); 6252 list_for_each(tmp,&all_mddevs) 6253 if (!l--) { 6254 mddev = list_entry(tmp, mddev_t, all_mddevs); 6255 mddev_get(mddev); 6256 spin_unlock(&all_mddevs_lock); 6257 return mddev; 6258 } 6259 spin_unlock(&all_mddevs_lock); 6260 if (!l--) 6261 return (void*)2;/* tail */ 6262 return NULL; 6263 } 6264 6265 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 6266 { 6267 struct list_head *tmp; 6268 mddev_t *next_mddev, *mddev = v; 6269 6270 ++*pos; 6271 if (v == (void*)2) 6272 return NULL; 6273 6274 spin_lock(&all_mddevs_lock); 6275 if (v == (void*)1) 6276 tmp = all_mddevs.next; 6277 else 6278 tmp = mddev->all_mddevs.next; 6279 if (tmp != &all_mddevs) 6280 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); 6281 else { 6282 next_mddev = (void*)2; 6283 *pos = 0x10000; 6284 } 6285 spin_unlock(&all_mddevs_lock); 6286 6287 if (v != (void*)1) 6288 mddev_put(mddev); 6289 return next_mddev; 6290 6291 } 6292 6293 static void md_seq_stop(struct seq_file *seq, void *v) 6294 { 6295 mddev_t *mddev = v; 6296 6297 if (mddev && v != (void*)1 && v != (void*)2) 6298 mddev_put(mddev); 6299 } 6300 6301 struct mdstat_info { 6302 int event; 6303 }; 6304 6305 static int md_seq_show(struct seq_file *seq, void *v) 6306 { 6307 mddev_t *mddev = v; 6308 sector_t sectors; 6309 mdk_rdev_t *rdev; 6310 struct mdstat_info *mi = seq->private; 6311 struct bitmap *bitmap; 6312 6313 if (v == (void*)1) { 6314 struct mdk_personality *pers; 6315 seq_printf(seq, "Personalities : "); 6316 spin_lock(&pers_lock); 6317 list_for_each_entry(pers, &pers_list, list) 6318 seq_printf(seq, "[%s] ", pers->name); 6319 6320 spin_unlock(&pers_lock); 6321 seq_printf(seq, "\n"); 6322 mi->event = atomic_read(&md_event_count); 6323 return 0; 6324 } 6325 if (v == (void*)2) { 6326 status_unused(seq); 6327 return 0; 6328 } 6329 6330 if (mddev_lock(mddev) < 0) 6331 return -EINTR; 6332 6333 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 6334 seq_printf(seq, "%s : %sactive", mdname(mddev), 6335 mddev->pers ? "" : "in"); 6336 if (mddev->pers) { 6337 if (mddev->ro==1) 6338 seq_printf(seq, " (read-only)"); 6339 if (mddev->ro==2) 6340 seq_printf(seq, " (auto-read-only)"); 6341 seq_printf(seq, " %s", mddev->pers->name); 6342 } 6343 6344 sectors = 0; 6345 list_for_each_entry(rdev, &mddev->disks, same_set) { 6346 char b[BDEVNAME_SIZE]; 6347 seq_printf(seq, " %s[%d]", 6348 bdevname(rdev->bdev,b), rdev->desc_nr); 6349 if (test_bit(WriteMostly, &rdev->flags)) 6350 seq_printf(seq, "(W)"); 6351 if (test_bit(Faulty, &rdev->flags)) { 6352 seq_printf(seq, "(F)"); 6353 continue; 6354 } else if (rdev->raid_disk < 0) 6355 seq_printf(seq, "(S)"); /* spare */ 6356 sectors += rdev->sectors; 6357 } 6358 6359 if (!list_empty(&mddev->disks)) { 6360 if (mddev->pers) 6361 seq_printf(seq, "\n %llu blocks", 6362 (unsigned long long) 6363 mddev->array_sectors / 2); 6364 else 6365 seq_printf(seq, "\n %llu blocks", 6366 (unsigned long long)sectors / 2); 6367 } 6368 if (mddev->persistent) { 6369 if (mddev->major_version != 0 || 6370 mddev->minor_version != 90) { 6371 seq_printf(seq," super %d.%d", 6372 mddev->major_version, 6373 mddev->minor_version); 6374 } 6375 } else if (mddev->external) 6376 seq_printf(seq, " super external:%s", 6377 mddev->metadata_type); 6378 else 6379 seq_printf(seq, " super non-persistent"); 6380 6381 if (mddev->pers) { 6382 mddev->pers->status(seq, mddev); 6383 seq_printf(seq, "\n "); 6384 if (mddev->pers->sync_request) { 6385 if (mddev->curr_resync > 2) { 6386 status_resync(seq, mddev); 6387 seq_printf(seq, "\n "); 6388 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 6389 seq_printf(seq, "\tresync=DELAYED\n "); 6390 else if (mddev->recovery_cp < MaxSector) 6391 seq_printf(seq, "\tresync=PENDING\n "); 6392 } 6393 } else 6394 seq_printf(seq, "\n "); 6395 6396 if ((bitmap = mddev->bitmap)) { 6397 unsigned long chunk_kb; 6398 unsigned long flags; 6399 spin_lock_irqsave(&bitmap->lock, flags); 6400 chunk_kb = mddev->bitmap_info.chunksize >> 10; 6401 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 6402 "%lu%s chunk", 6403 bitmap->pages - bitmap->missing_pages, 6404 bitmap->pages, 6405 (bitmap->pages - bitmap->missing_pages) 6406 << (PAGE_SHIFT - 10), 6407 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, 6408 chunk_kb ? "KB" : "B"); 6409 if (bitmap->file) { 6410 seq_printf(seq, ", file: "); 6411 seq_path(seq, &bitmap->file->f_path, " \t\n"); 6412 } 6413 6414 seq_printf(seq, "\n"); 6415 spin_unlock_irqrestore(&bitmap->lock, flags); 6416 } 6417 6418 seq_printf(seq, "\n"); 6419 } 6420 mddev_unlock(mddev); 6421 6422 return 0; 6423 } 6424 6425 static const struct seq_operations md_seq_ops = { 6426 .start = md_seq_start, 6427 .next = md_seq_next, 6428 .stop = md_seq_stop, 6429 .show = md_seq_show, 6430 }; 6431 6432 static int md_seq_open(struct inode *inode, struct file *file) 6433 { 6434 int error; 6435 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); 6436 if (mi == NULL) 6437 return -ENOMEM; 6438 6439 error = seq_open(file, &md_seq_ops); 6440 if (error) 6441 kfree(mi); 6442 else { 6443 struct seq_file *p = file->private_data; 6444 p->private = mi; 6445 mi->event = atomic_read(&md_event_count); 6446 } 6447 return error; 6448 } 6449 6450 static unsigned int mdstat_poll(struct file *filp, poll_table *wait) 6451 { 6452 struct seq_file *m = filp->private_data; 6453 struct mdstat_info *mi = m->private; 6454 int mask; 6455 6456 poll_wait(filp, &md_event_waiters, wait); 6457 6458 /* always allow read */ 6459 mask = POLLIN | POLLRDNORM; 6460 6461 if (mi->event != atomic_read(&md_event_count)) 6462 mask |= POLLERR | POLLPRI; 6463 return mask; 6464 } 6465 6466 static const struct file_operations md_seq_fops = { 6467 .owner = THIS_MODULE, 6468 .open = md_seq_open, 6469 .read = seq_read, 6470 .llseek = seq_lseek, 6471 .release = seq_release_private, 6472 .poll = mdstat_poll, 6473 }; 6474 6475 int register_md_personality(struct mdk_personality *p) 6476 { 6477 spin_lock(&pers_lock); 6478 list_add_tail(&p->list, &pers_list); 6479 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); 6480 spin_unlock(&pers_lock); 6481 return 0; 6482 } 6483 6484 int unregister_md_personality(struct mdk_personality *p) 6485 { 6486 printk(KERN_INFO "md: %s personality unregistered\n", p->name); 6487 spin_lock(&pers_lock); 6488 list_del_init(&p->list); 6489 spin_unlock(&pers_lock); 6490 return 0; 6491 } 6492 6493 static int is_mddev_idle(mddev_t *mddev, int init) 6494 { 6495 mdk_rdev_t * rdev; 6496 int idle; 6497 int curr_events; 6498 6499 idle = 1; 6500 rcu_read_lock(); 6501 rdev_for_each_rcu(rdev, mddev) { 6502 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 6503 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 6504 (int)part_stat_read(&disk->part0, sectors[1]) - 6505 atomic_read(&disk->sync_io); 6506 /* sync IO will cause sync_io to increase before the disk_stats 6507 * as sync_io is counted when a request starts, and 6508 * disk_stats is counted when it completes. 6509 * So resync activity will cause curr_events to be smaller than 6510 * when there was no such activity. 6511 * non-sync IO will cause disk_stat to increase without 6512 * increasing sync_io so curr_events will (eventually) 6513 * be larger than it was before. Once it becomes 6514 * substantially larger, the test below will cause 6515 * the array to appear non-idle, and resync will slow 6516 * down. 6517 * If there is a lot of outstanding resync activity when 6518 * we set last_event to curr_events, then all that activity 6519 * completing might cause the array to appear non-idle 6520 * and resync will be slowed down even though there might 6521 * not have been non-resync activity. This will only 6522 * happen once though. 'last_events' will soon reflect 6523 * the state where there is little or no outstanding 6524 * resync requests, and further resync activity will 6525 * always make curr_events less than last_events. 6526 * 6527 */ 6528 if (init || curr_events - rdev->last_events > 64) { 6529 rdev->last_events = curr_events; 6530 idle = 0; 6531 } 6532 } 6533 rcu_read_unlock(); 6534 return idle; 6535 } 6536 6537 void md_done_sync(mddev_t *mddev, int blocks, int ok) 6538 { 6539 /* another "blocks" (512byte) blocks have been synced */ 6540 atomic_sub(blocks, &mddev->recovery_active); 6541 wake_up(&mddev->recovery_wait); 6542 if (!ok) { 6543 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6544 md_wakeup_thread(mddev->thread); 6545 // stop recovery, signal do_sync .... 6546 } 6547 } 6548 6549 6550 /* md_write_start(mddev, bi) 6551 * If we need to update some array metadata (e.g. 'active' flag 6552 * in superblock) before writing, schedule a superblock update 6553 * and wait for it to complete. 6554 */ 6555 void md_write_start(mddev_t *mddev, struct bio *bi) 6556 { 6557 int did_change = 0; 6558 if (bio_data_dir(bi) != WRITE) 6559 return; 6560 6561 BUG_ON(mddev->ro == 1); 6562 if (mddev->ro == 2) { 6563 /* need to switch to read/write */ 6564 mddev->ro = 0; 6565 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6566 md_wakeup_thread(mddev->thread); 6567 md_wakeup_thread(mddev->sync_thread); 6568 did_change = 1; 6569 } 6570 atomic_inc(&mddev->writes_pending); 6571 if (mddev->safemode == 1) 6572 mddev->safemode = 0; 6573 if (mddev->in_sync) { 6574 spin_lock_irq(&mddev->write_lock); 6575 if (mddev->in_sync) { 6576 mddev->in_sync = 0; 6577 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6578 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6579 md_wakeup_thread(mddev->thread); 6580 did_change = 1; 6581 } 6582 spin_unlock_irq(&mddev->write_lock); 6583 } 6584 if (did_change) 6585 sysfs_notify_dirent_safe(mddev->sysfs_state); 6586 wait_event(mddev->sb_wait, 6587 !test_bit(MD_CHANGE_PENDING, &mddev->flags)); 6588 } 6589 6590 void md_write_end(mddev_t *mddev) 6591 { 6592 if (atomic_dec_and_test(&mddev->writes_pending)) { 6593 if (mddev->safemode == 2) 6594 md_wakeup_thread(mddev->thread); 6595 else if (mddev->safemode_delay) 6596 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 6597 } 6598 } 6599 6600 /* md_allow_write(mddev) 6601 * Calling this ensures that the array is marked 'active' so that writes 6602 * may proceed without blocking. It is important to call this before 6603 * attempting a GFP_KERNEL allocation while holding the mddev lock. 6604 * Must be called with mddev_lock held. 6605 * 6606 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock 6607 * is dropped, so return -EAGAIN after notifying userspace. 6608 */ 6609 int md_allow_write(mddev_t *mddev) 6610 { 6611 if (!mddev->pers) 6612 return 0; 6613 if (mddev->ro) 6614 return 0; 6615 if (!mddev->pers->sync_request) 6616 return 0; 6617 6618 spin_lock_irq(&mddev->write_lock); 6619 if (mddev->in_sync) { 6620 mddev->in_sync = 0; 6621 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6622 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6623 if (mddev->safemode_delay && 6624 mddev->safemode == 0) 6625 mddev->safemode = 1; 6626 spin_unlock_irq(&mddev->write_lock); 6627 md_update_sb(mddev, 0); 6628 sysfs_notify_dirent_safe(mddev->sysfs_state); 6629 } else 6630 spin_unlock_irq(&mddev->write_lock); 6631 6632 if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 6633 return -EAGAIN; 6634 else 6635 return 0; 6636 } 6637 EXPORT_SYMBOL_GPL(md_allow_write); 6638 6639 void md_unplug(mddev_t *mddev) 6640 { 6641 if (mddev->queue) 6642 blk_unplug(mddev->queue); 6643 if (mddev->plug) 6644 mddev->plug->unplug_fn(mddev->plug); 6645 } 6646 6647 #define SYNC_MARKS 10 6648 #define SYNC_MARK_STEP (3*HZ) 6649 void md_do_sync(mddev_t *mddev) 6650 { 6651 mddev_t *mddev2; 6652 unsigned int currspeed = 0, 6653 window; 6654 sector_t max_sectors,j, io_sectors; 6655 unsigned long mark[SYNC_MARKS]; 6656 sector_t mark_cnt[SYNC_MARKS]; 6657 int last_mark,m; 6658 struct list_head *tmp; 6659 sector_t last_check; 6660 int skipped = 0; 6661 mdk_rdev_t *rdev; 6662 char *desc; 6663 6664 /* just incase thread restarts... */ 6665 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 6666 return; 6667 if (mddev->ro) /* never try to sync a read-only array */ 6668 return; 6669 6670 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6671 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 6672 desc = "data-check"; 6673 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6674 desc = "requested-resync"; 6675 else 6676 desc = "resync"; 6677 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6678 desc = "reshape"; 6679 else 6680 desc = "recovery"; 6681 6682 /* we overload curr_resync somewhat here. 6683 * 0 == not engaged in resync at all 6684 * 2 == checking that there is no conflict with another sync 6685 * 1 == like 2, but have yielded to allow conflicting resync to 6686 * commense 6687 * other == active in resync - this many blocks 6688 * 6689 * Before starting a resync we must have set curr_resync to 6690 * 2, and then checked that every "conflicting" array has curr_resync 6691 * less than ours. When we find one that is the same or higher 6692 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 6693 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 6694 * This will mean we have to start checking from the beginning again. 6695 * 6696 */ 6697 6698 do { 6699 mddev->curr_resync = 2; 6700 6701 try_again: 6702 if (kthread_should_stop()) 6703 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6704 6705 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6706 goto skip; 6707 for_each_mddev(mddev2, tmp) { 6708 if (mddev2 == mddev) 6709 continue; 6710 if (!mddev->parallel_resync 6711 && mddev2->curr_resync 6712 && match_mddev_units(mddev, mddev2)) { 6713 DEFINE_WAIT(wq); 6714 if (mddev < mddev2 && mddev->curr_resync == 2) { 6715 /* arbitrarily yield */ 6716 mddev->curr_resync = 1; 6717 wake_up(&resync_wait); 6718 } 6719 if (mddev > mddev2 && mddev->curr_resync == 1) 6720 /* no need to wait here, we can wait the next 6721 * time 'round when curr_resync == 2 6722 */ 6723 continue; 6724 /* We need to wait 'interruptible' so as not to 6725 * contribute to the load average, and not to 6726 * be caught by 'softlockup' 6727 */ 6728 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 6729 if (!kthread_should_stop() && 6730 mddev2->curr_resync >= mddev->curr_resync) { 6731 printk(KERN_INFO "md: delaying %s of %s" 6732 " until %s has finished (they" 6733 " share one or more physical units)\n", 6734 desc, mdname(mddev), mdname(mddev2)); 6735 mddev_put(mddev2); 6736 if (signal_pending(current)) 6737 flush_signals(current); 6738 schedule(); 6739 finish_wait(&resync_wait, &wq); 6740 goto try_again; 6741 } 6742 finish_wait(&resync_wait, &wq); 6743 } 6744 } 6745 } while (mddev->curr_resync < 2); 6746 6747 j = 0; 6748 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6749 /* resync follows the size requested by the personality, 6750 * which defaults to physical size, but can be virtual size 6751 */ 6752 max_sectors = mddev->resync_max_sectors; 6753 mddev->resync_mismatches = 0; 6754 /* we don't use the checkpoint if there's a bitmap */ 6755 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6756 j = mddev->resync_min; 6757 else if (!mddev->bitmap) 6758 j = mddev->recovery_cp; 6759 6760 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6761 max_sectors = mddev->dev_sectors; 6762 else { 6763 /* recovery follows the physical size of devices */ 6764 max_sectors = mddev->dev_sectors; 6765 j = MaxSector; 6766 rcu_read_lock(); 6767 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 6768 if (rdev->raid_disk >= 0 && 6769 !test_bit(Faulty, &rdev->flags) && 6770 !test_bit(In_sync, &rdev->flags) && 6771 rdev->recovery_offset < j) 6772 j = rdev->recovery_offset; 6773 rcu_read_unlock(); 6774 } 6775 6776 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); 6777 printk(KERN_INFO "md: minimum _guaranteed_ speed:" 6778 " %d KB/sec/disk.\n", speed_min(mddev)); 6779 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 6780 "(but not more than %d KB/sec) for %s.\n", 6781 speed_max(mddev), desc); 6782 6783 is_mddev_idle(mddev, 1); /* this initializes IO event counters */ 6784 6785 io_sectors = 0; 6786 for (m = 0; m < SYNC_MARKS; m++) { 6787 mark[m] = jiffies; 6788 mark_cnt[m] = io_sectors; 6789 } 6790 last_mark = 0; 6791 mddev->resync_mark = mark[last_mark]; 6792 mddev->resync_mark_cnt = mark_cnt[last_mark]; 6793 6794 /* 6795 * Tune reconstruction: 6796 */ 6797 window = 32*(PAGE_SIZE/512); 6798 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", 6799 window/2,(unsigned long long) max_sectors/2); 6800 6801 atomic_set(&mddev->recovery_active, 0); 6802 last_check = 0; 6803 6804 if (j>2) { 6805 printk(KERN_INFO 6806 "md: resuming %s of %s from checkpoint.\n", 6807 desc, mdname(mddev)); 6808 mddev->curr_resync = j; 6809 } 6810 mddev->curr_resync_completed = mddev->curr_resync; 6811 6812 while (j < max_sectors) { 6813 sector_t sectors; 6814 6815 skipped = 0; 6816 6817 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 6818 ((mddev->curr_resync > mddev->curr_resync_completed && 6819 (mddev->curr_resync - mddev->curr_resync_completed) 6820 > (max_sectors >> 4)) || 6821 (j - mddev->curr_resync_completed)*2 6822 >= mddev->resync_max - mddev->curr_resync_completed 6823 )) { 6824 /* time to update curr_resync_completed */ 6825 md_unplug(mddev); 6826 wait_event(mddev->recovery_wait, 6827 atomic_read(&mddev->recovery_active) == 0); 6828 mddev->curr_resync_completed = 6829 mddev->curr_resync; 6830 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6831 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 6832 } 6833 6834 while (j >= mddev->resync_max && !kthread_should_stop()) { 6835 /* As this condition is controlled by user-space, 6836 * we can block indefinitely, so use '_interruptible' 6837 * to avoid triggering warnings. 6838 */ 6839 flush_signals(current); /* just in case */ 6840 wait_event_interruptible(mddev->recovery_wait, 6841 mddev->resync_max > j 6842 || kthread_should_stop()); 6843 } 6844 6845 if (kthread_should_stop()) 6846 goto interrupted; 6847 6848 sectors = mddev->pers->sync_request(mddev, j, &skipped, 6849 currspeed < speed_min(mddev)); 6850 if (sectors == 0) { 6851 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6852 goto out; 6853 } 6854 6855 if (!skipped) { /* actual IO requested */ 6856 io_sectors += sectors; 6857 atomic_add(sectors, &mddev->recovery_active); 6858 } 6859 6860 j += sectors; 6861 if (j>1) mddev->curr_resync = j; 6862 mddev->curr_mark_cnt = io_sectors; 6863 if (last_check == 0) 6864 /* this is the earliers that rebuilt will be 6865 * visible in /proc/mdstat 6866 */ 6867 md_new_event(mddev); 6868 6869 if (last_check + window > io_sectors || j == max_sectors) 6870 continue; 6871 6872 last_check = io_sectors; 6873 6874 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6875 break; 6876 6877 repeat: 6878 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 6879 /* step marks */ 6880 int next = (last_mark+1) % SYNC_MARKS; 6881 6882 mddev->resync_mark = mark[next]; 6883 mddev->resync_mark_cnt = mark_cnt[next]; 6884 mark[next] = jiffies; 6885 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 6886 last_mark = next; 6887 } 6888 6889 6890 if (kthread_should_stop()) 6891 goto interrupted; 6892 6893 6894 /* 6895 * this loop exits only if either when we are slower than 6896 * the 'hard' speed limit, or the system was IO-idle for 6897 * a jiffy. 6898 * the system might be non-idle CPU-wise, but we only care 6899 * about not overloading the IO subsystem. (things like an 6900 * e2fsck being done on the RAID array should execute fast) 6901 */ 6902 md_unplug(mddev); 6903 cond_resched(); 6904 6905 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 6906 /((jiffies-mddev->resync_mark)/HZ +1) +1; 6907 6908 if (currspeed > speed_min(mddev)) { 6909 if ((currspeed > speed_max(mddev)) || 6910 !is_mddev_idle(mddev, 0)) { 6911 msleep(500); 6912 goto repeat; 6913 } 6914 } 6915 } 6916 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); 6917 /* 6918 * this also signals 'finished resyncing' to md_stop 6919 */ 6920 out: 6921 md_unplug(mddev); 6922 6923 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 6924 6925 /* tell personality that we are finished */ 6926 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 6927 6928 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && 6929 mddev->curr_resync > 2) { 6930 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6931 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 6932 if (mddev->curr_resync >= mddev->recovery_cp) { 6933 printk(KERN_INFO 6934 "md: checkpointing %s of %s.\n", 6935 desc, mdname(mddev)); 6936 mddev->recovery_cp = mddev->curr_resync; 6937 } 6938 } else 6939 mddev->recovery_cp = MaxSector; 6940 } else { 6941 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6942 mddev->curr_resync = MaxSector; 6943 rcu_read_lock(); 6944 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 6945 if (rdev->raid_disk >= 0 && 6946 mddev->delta_disks >= 0 && 6947 !test_bit(Faulty, &rdev->flags) && 6948 !test_bit(In_sync, &rdev->flags) && 6949 rdev->recovery_offset < mddev->curr_resync) 6950 rdev->recovery_offset = mddev->curr_resync; 6951 rcu_read_unlock(); 6952 } 6953 } 6954 set_bit(MD_CHANGE_DEVS, &mddev->flags); 6955 6956 skip: 6957 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 6958 /* We completed so min/max setting can be forgotten if used. */ 6959 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6960 mddev->resync_min = 0; 6961 mddev->resync_max = MaxSector; 6962 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6963 mddev->resync_min = mddev->curr_resync_completed; 6964 mddev->curr_resync = 0; 6965 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6966 mddev->curr_resync_completed = 0; 6967 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 6968 wake_up(&resync_wait); 6969 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 6970 md_wakeup_thread(mddev->thread); 6971 return; 6972 6973 interrupted: 6974 /* 6975 * got a signal, exit. 6976 */ 6977 printk(KERN_INFO 6978 "md: md_do_sync() got signal ... exiting\n"); 6979 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6980 goto out; 6981 6982 } 6983 EXPORT_SYMBOL_GPL(md_do_sync); 6984 6985 6986 static int remove_and_add_spares(mddev_t *mddev) 6987 { 6988 mdk_rdev_t *rdev; 6989 int spares = 0; 6990 6991 mddev->curr_resync_completed = 0; 6992 6993 list_for_each_entry(rdev, &mddev->disks, same_set) 6994 if (rdev->raid_disk >= 0 && 6995 !test_bit(Blocked, &rdev->flags) && 6996 (test_bit(Faulty, &rdev->flags) || 6997 ! test_bit(In_sync, &rdev->flags)) && 6998 atomic_read(&rdev->nr_pending)==0) { 6999 if (mddev->pers->hot_remove_disk( 7000 mddev, rdev->raid_disk)==0) { 7001 char nm[20]; 7002 sprintf(nm,"rd%d", rdev->raid_disk); 7003 sysfs_remove_link(&mddev->kobj, nm); 7004 rdev->raid_disk = -1; 7005 } 7006 } 7007 7008 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { 7009 list_for_each_entry(rdev, &mddev->disks, same_set) { 7010 if (rdev->raid_disk >= 0 && 7011 !test_bit(In_sync, &rdev->flags) && 7012 !test_bit(Blocked, &rdev->flags)) 7013 spares++; 7014 if (rdev->raid_disk < 0 7015 && !test_bit(Faulty, &rdev->flags)) { 7016 rdev->recovery_offset = 0; 7017 if (mddev->pers-> 7018 hot_add_disk(mddev, rdev) == 0) { 7019 char nm[20]; 7020 sprintf(nm, "rd%d", rdev->raid_disk); 7021 if (sysfs_create_link(&mddev->kobj, 7022 &rdev->kobj, nm)) 7023 /* failure here is OK */; 7024 spares++; 7025 md_new_event(mddev); 7026 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7027 } else 7028 break; 7029 } 7030 } 7031 } 7032 return spares; 7033 } 7034 /* 7035 * This routine is regularly called by all per-raid-array threads to 7036 * deal with generic issues like resync and super-block update. 7037 * Raid personalities that don't have a thread (linear/raid0) do not 7038 * need this as they never do any recovery or update the superblock. 7039 * 7040 * It does not do any resync itself, but rather "forks" off other threads 7041 * to do that as needed. 7042 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 7043 * "->recovery" and create a thread at ->sync_thread. 7044 * When the thread finishes it sets MD_RECOVERY_DONE 7045 * and wakeups up this thread which will reap the thread and finish up. 7046 * This thread also removes any faulty devices (with nr_pending == 0). 7047 * 7048 * The overall approach is: 7049 * 1/ if the superblock needs updating, update it. 7050 * 2/ If a recovery thread is running, don't do anything else. 7051 * 3/ If recovery has finished, clean up, possibly marking spares active. 7052 * 4/ If there are any faulty devices, remove them. 7053 * 5/ If array is degraded, try to add spares devices 7054 * 6/ If array has spares or is not in-sync, start a resync thread. 7055 */ 7056 void md_check_recovery(mddev_t *mddev) 7057 { 7058 mdk_rdev_t *rdev; 7059 7060 7061 if (mddev->bitmap) 7062 bitmap_daemon_work(mddev); 7063 7064 if (mddev->ro) 7065 return; 7066 7067 if (signal_pending(current)) { 7068 if (mddev->pers->sync_request && !mddev->external) { 7069 printk(KERN_INFO "md: %s in immediate safe mode\n", 7070 mdname(mddev)); 7071 mddev->safemode = 2; 7072 } 7073 flush_signals(current); 7074 } 7075 7076 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 7077 return; 7078 if ( ! ( 7079 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || 7080 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 7081 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 7082 (mddev->external == 0 && mddev->safemode == 1) || 7083 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 7084 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 7085 )) 7086 return; 7087 7088 if (mddev_trylock(mddev)) { 7089 int spares = 0; 7090 7091 if (mddev->ro) { 7092 /* Only thing we do on a ro array is remove 7093 * failed devices. 7094 */ 7095 remove_and_add_spares(mddev); 7096 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7097 goto unlock; 7098 } 7099 7100 if (!mddev->external) { 7101 int did_change = 0; 7102 spin_lock_irq(&mddev->write_lock); 7103 if (mddev->safemode && 7104 !atomic_read(&mddev->writes_pending) && 7105 !mddev->in_sync && 7106 mddev->recovery_cp == MaxSector) { 7107 mddev->in_sync = 1; 7108 did_change = 1; 7109 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7110 } 7111 if (mddev->safemode == 1) 7112 mddev->safemode = 0; 7113 spin_unlock_irq(&mddev->write_lock); 7114 if (did_change) 7115 sysfs_notify_dirent_safe(mddev->sysfs_state); 7116 } 7117 7118 if (mddev->flags) 7119 md_update_sb(mddev, 0); 7120 7121 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 7122 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 7123 /* resync/recovery still happening */ 7124 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7125 goto unlock; 7126 } 7127 if (mddev->sync_thread) { 7128 /* resync has finished, collect result */ 7129 md_unregister_thread(mddev->sync_thread); 7130 mddev->sync_thread = NULL; 7131 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && 7132 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 7133 /* success...*/ 7134 /* activate any spares */ 7135 if (mddev->pers->spare_active(mddev)) 7136 sysfs_notify(&mddev->kobj, NULL, 7137 "degraded"); 7138 } 7139 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7140 mddev->pers->finish_reshape) 7141 mddev->pers->finish_reshape(mddev); 7142 md_update_sb(mddev, 1); 7143 7144 /* if array is no-longer degraded, then any saved_raid_disk 7145 * information must be scrapped 7146 */ 7147 if (!mddev->degraded) 7148 list_for_each_entry(rdev, &mddev->disks, same_set) 7149 rdev->saved_raid_disk = -1; 7150 7151 mddev->recovery = 0; 7152 /* flag recovery needed just to double check */ 7153 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7154 sysfs_notify_dirent_safe(mddev->sysfs_action); 7155 md_new_event(mddev); 7156 goto unlock; 7157 } 7158 /* Set RUNNING before clearing NEEDED to avoid 7159 * any transients in the value of "sync_action". 7160 */ 7161 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7162 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7163 /* Clear some bits that don't mean anything, but 7164 * might be left set 7165 */ 7166 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 7167 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 7168 7169 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 7170 goto unlock; 7171 /* no recovery is running. 7172 * remove any failed drives, then 7173 * add spares if possible. 7174 * Spare are also removed and re-added, to allow 7175 * the personality to fail the re-add. 7176 */ 7177 7178 if (mddev->reshape_position != MaxSector) { 7179 if (mddev->pers->check_reshape == NULL || 7180 mddev->pers->check_reshape(mddev) != 0) 7181 /* Cannot proceed */ 7182 goto unlock; 7183 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7184 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7185 } else if ((spares = remove_and_add_spares(mddev))) { 7186 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7187 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7188 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7189 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7190 } else if (mddev->recovery_cp < MaxSector) { 7191 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7192 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7193 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 7194 /* nothing to be done ... */ 7195 goto unlock; 7196 7197 if (mddev->pers->sync_request) { 7198 if (spares && mddev->bitmap && ! mddev->bitmap->file) { 7199 /* We are adding a device or devices to an array 7200 * which has the bitmap stored on all devices. 7201 * So make sure all bitmap pages get written 7202 */ 7203 bitmap_write_all(mddev->bitmap); 7204 } 7205 mddev->sync_thread = md_register_thread(md_do_sync, 7206 mddev, 7207 "resync"); 7208 if (!mddev->sync_thread) { 7209 printk(KERN_ERR "%s: could not start resync" 7210 " thread...\n", 7211 mdname(mddev)); 7212 /* leave the spares where they are, it shouldn't hurt */ 7213 mddev->recovery = 0; 7214 } else 7215 md_wakeup_thread(mddev->sync_thread); 7216 sysfs_notify_dirent_safe(mddev->sysfs_action); 7217 md_new_event(mddev); 7218 } 7219 unlock: 7220 if (!mddev->sync_thread) { 7221 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7222 if (test_and_clear_bit(MD_RECOVERY_RECOVER, 7223 &mddev->recovery)) 7224 if (mddev->sysfs_action) 7225 sysfs_notify_dirent_safe(mddev->sysfs_action); 7226 } 7227 mddev_unlock(mddev); 7228 } 7229 } 7230 7231 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 7232 { 7233 sysfs_notify_dirent_safe(rdev->sysfs_state); 7234 wait_event_timeout(rdev->blocked_wait, 7235 !test_bit(Blocked, &rdev->flags), 7236 msecs_to_jiffies(5000)); 7237 rdev_dec_pending(rdev, mddev); 7238 } 7239 EXPORT_SYMBOL(md_wait_for_blocked_rdev); 7240 7241 static int md_notify_reboot(struct notifier_block *this, 7242 unsigned long code, void *x) 7243 { 7244 struct list_head *tmp; 7245 mddev_t *mddev; 7246 7247 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { 7248 7249 printk(KERN_INFO "md: stopping all md devices.\n"); 7250 7251 for_each_mddev(mddev, tmp) 7252 if (mddev_trylock(mddev)) { 7253 /* Force a switch to readonly even array 7254 * appears to still be in use. Hence 7255 * the '100'. 7256 */ 7257 md_set_readonly(mddev, 100); 7258 mddev_unlock(mddev); 7259 } 7260 /* 7261 * certain more exotic SCSI devices are known to be 7262 * volatile wrt too early system reboots. While the 7263 * right place to handle this issue is the given 7264 * driver, we do want to have a safe RAID driver ... 7265 */ 7266 mdelay(1000*1); 7267 } 7268 return NOTIFY_DONE; 7269 } 7270 7271 static struct notifier_block md_notifier = { 7272 .notifier_call = md_notify_reboot, 7273 .next = NULL, 7274 .priority = INT_MAX, /* before any real devices */ 7275 }; 7276 7277 static void md_geninit(void) 7278 { 7279 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 7280 7281 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); 7282 } 7283 7284 static int __init md_init(void) 7285 { 7286 int ret = -ENOMEM; 7287 7288 md_wq = alloc_workqueue("md", WQ_RESCUER, 0); 7289 if (!md_wq) 7290 goto err_wq; 7291 7292 md_misc_wq = alloc_workqueue("md_misc", 0, 0); 7293 if (!md_misc_wq) 7294 goto err_misc_wq; 7295 7296 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) 7297 goto err_md; 7298 7299 if ((ret = register_blkdev(0, "mdp")) < 0) 7300 goto err_mdp; 7301 mdp_major = ret; 7302 7303 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, 7304 md_probe, NULL, NULL); 7305 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, 7306 md_probe, NULL, NULL); 7307 7308 register_reboot_notifier(&md_notifier); 7309 raid_table_header = register_sysctl_table(raid_root_table); 7310 7311 md_geninit(); 7312 return 0; 7313 7314 err_mdp: 7315 unregister_blkdev(MD_MAJOR, "md"); 7316 err_md: 7317 destroy_workqueue(md_misc_wq); 7318 err_misc_wq: 7319 destroy_workqueue(md_wq); 7320 err_wq: 7321 return ret; 7322 } 7323 7324 #ifndef MODULE 7325 7326 /* 7327 * Searches all registered partitions for autorun RAID arrays 7328 * at boot time. 7329 */ 7330 7331 static LIST_HEAD(all_detected_devices); 7332 struct detected_devices_node { 7333 struct list_head list; 7334 dev_t dev; 7335 }; 7336 7337 void md_autodetect_dev(dev_t dev) 7338 { 7339 struct detected_devices_node *node_detected_dev; 7340 7341 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); 7342 if (node_detected_dev) { 7343 node_detected_dev->dev = dev; 7344 list_add_tail(&node_detected_dev->list, &all_detected_devices); 7345 } else { 7346 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" 7347 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); 7348 } 7349 } 7350 7351 7352 static void autostart_arrays(int part) 7353 { 7354 mdk_rdev_t *rdev; 7355 struct detected_devices_node *node_detected_dev; 7356 dev_t dev; 7357 int i_scanned, i_passed; 7358 7359 i_scanned = 0; 7360 i_passed = 0; 7361 7362 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 7363 7364 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { 7365 i_scanned++; 7366 node_detected_dev = list_entry(all_detected_devices.next, 7367 struct detected_devices_node, list); 7368 list_del(&node_detected_dev->list); 7369 dev = node_detected_dev->dev; 7370 kfree(node_detected_dev); 7371 rdev = md_import_device(dev,0, 90); 7372 if (IS_ERR(rdev)) 7373 continue; 7374 7375 if (test_bit(Faulty, &rdev->flags)) { 7376 MD_BUG(); 7377 continue; 7378 } 7379 set_bit(AutoDetected, &rdev->flags); 7380 list_add(&rdev->same_set, &pending_raid_disks); 7381 i_passed++; 7382 } 7383 7384 printk(KERN_INFO "md: Scanned %d and added %d devices.\n", 7385 i_scanned, i_passed); 7386 7387 autorun_devices(part); 7388 } 7389 7390 #endif /* !MODULE */ 7391 7392 static __exit void md_exit(void) 7393 { 7394 mddev_t *mddev; 7395 struct list_head *tmp; 7396 7397 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); 7398 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); 7399 7400 unregister_blkdev(MD_MAJOR,"md"); 7401 unregister_blkdev(mdp_major, "mdp"); 7402 unregister_reboot_notifier(&md_notifier); 7403 unregister_sysctl_table(raid_table_header); 7404 remove_proc_entry("mdstat", NULL); 7405 for_each_mddev(mddev, tmp) { 7406 export_array(mddev); 7407 mddev->hold_active = 0; 7408 } 7409 destroy_workqueue(md_misc_wq); 7410 destroy_workqueue(md_wq); 7411 } 7412 7413 subsys_initcall(md_init); 7414 module_exit(md_exit) 7415 7416 static int get_ro(char *buffer, struct kernel_param *kp) 7417 { 7418 return sprintf(buffer, "%d", start_readonly); 7419 } 7420 static int set_ro(const char *val, struct kernel_param *kp) 7421 { 7422 char *e; 7423 int num = simple_strtoul(val, &e, 10); 7424 if (*val && (*e == '\0' || *e == '\n')) { 7425 start_readonly = num; 7426 return 0; 7427 } 7428 return -EINVAL; 7429 } 7430 7431 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); 7432 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); 7433 7434 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); 7435 7436 EXPORT_SYMBOL(register_md_personality); 7437 EXPORT_SYMBOL(unregister_md_personality); 7438 EXPORT_SYMBOL(md_error); 7439 EXPORT_SYMBOL(md_done_sync); 7440 EXPORT_SYMBOL(md_write_start); 7441 EXPORT_SYMBOL(md_write_end); 7442 EXPORT_SYMBOL(md_register_thread); 7443 EXPORT_SYMBOL(md_unregister_thread); 7444 EXPORT_SYMBOL(md_wakeup_thread); 7445 EXPORT_SYMBOL(md_check_recovery); 7446 MODULE_LICENSE("GPL"); 7447 MODULE_DESCRIPTION("MD RAID framework"); 7448 MODULE_ALIAS("md"); 7449 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 7450