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/smp_lock.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 61 #ifndef MODULE 62 static void autostart_arrays(int part); 63 #endif 64 65 static LIST_HEAD(pers_list); 66 static DEFINE_SPINLOCK(pers_lock); 67 68 static void md_print_devices(void); 69 70 static DECLARE_WAIT_QUEUE_HEAD(resync_wait); 71 72 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } 73 74 /* 75 * Default number of read corrections we'll attempt on an rdev 76 * before ejecting it from the array. We divide the read error 77 * count by 2 for every hour elapsed between read errors. 78 */ 79 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 80 /* 81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' 82 * is 1000 KB/sec, so the extra system load does not show up that much. 83 * Increase it if you want to have more _guaranteed_ speed. Note that 84 * the RAID driver will use the maximum available bandwidth if the IO 85 * subsystem is idle. There is also an 'absolute maximum' reconstruction 86 * speed limit - in case reconstruction slows down your system despite 87 * idle IO detection. 88 * 89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. 90 * or /sys/block/mdX/md/sync_speed_{min,max} 91 */ 92 93 static int sysctl_speed_limit_min = 1000; 94 static int sysctl_speed_limit_max = 200000; 95 static inline int speed_min(mddev_t *mddev) 96 { 97 return mddev->sync_speed_min ? 98 mddev->sync_speed_min : sysctl_speed_limit_min; 99 } 100 101 static inline int speed_max(mddev_t *mddev) 102 { 103 return mddev->sync_speed_max ? 104 mddev->sync_speed_max : sysctl_speed_limit_max; 105 } 106 107 static struct ctl_table_header *raid_table_header; 108 109 static ctl_table raid_table[] = { 110 { 111 .procname = "speed_limit_min", 112 .data = &sysctl_speed_limit_min, 113 .maxlen = sizeof(int), 114 .mode = S_IRUGO|S_IWUSR, 115 .proc_handler = proc_dointvec, 116 }, 117 { 118 .procname = "speed_limit_max", 119 .data = &sysctl_speed_limit_max, 120 .maxlen = sizeof(int), 121 .mode = S_IRUGO|S_IWUSR, 122 .proc_handler = proc_dointvec, 123 }, 124 { } 125 }; 126 127 static ctl_table raid_dir_table[] = { 128 { 129 .procname = "raid", 130 .maxlen = 0, 131 .mode = S_IRUGO|S_IXUGO, 132 .child = raid_table, 133 }, 134 { } 135 }; 136 137 static ctl_table raid_root_table[] = { 138 { 139 .procname = "dev", 140 .maxlen = 0, 141 .mode = 0555, 142 .child = raid_dir_table, 143 }, 144 { } 145 }; 146 147 static const struct block_device_operations md_fops; 148 149 static int start_readonly; 150 151 /* 152 * We have a system wide 'event count' that is incremented 153 * on any 'interesting' event, and readers of /proc/mdstat 154 * can use 'poll' or 'select' to find out when the event 155 * count increases. 156 * 157 * Events are: 158 * start array, stop array, error, add device, remove device, 159 * start build, activate spare 160 */ 161 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); 162 static atomic_t md_event_count; 163 void md_new_event(mddev_t *mddev) 164 { 165 atomic_inc(&md_event_count); 166 wake_up(&md_event_waiters); 167 } 168 EXPORT_SYMBOL_GPL(md_new_event); 169 170 /* Alternate version that can be called from interrupts 171 * when calling sysfs_notify isn't needed. 172 */ 173 static void md_new_event_inintr(mddev_t *mddev) 174 { 175 atomic_inc(&md_event_count); 176 wake_up(&md_event_waiters); 177 } 178 179 /* 180 * Enables to iterate over all existing md arrays 181 * all_mddevs_lock protects this list. 182 */ 183 static LIST_HEAD(all_mddevs); 184 static DEFINE_SPINLOCK(all_mddevs_lock); 185 186 187 /* 188 * iterates through all used mddevs in the system. 189 * We take care to grab the all_mddevs_lock whenever navigating 190 * the list, and to always hold a refcount when unlocked. 191 * Any code which breaks out of this loop while own 192 * a reference to the current mddev and must mddev_put it. 193 */ 194 #define for_each_mddev(mddev,tmp) \ 195 \ 196 for (({ spin_lock(&all_mddevs_lock); \ 197 tmp = all_mddevs.next; \ 198 mddev = NULL;}); \ 199 ({ if (tmp != &all_mddevs) \ 200 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ 201 spin_unlock(&all_mddevs_lock); \ 202 if (mddev) mddev_put(mddev); \ 203 mddev = list_entry(tmp, mddev_t, all_mddevs); \ 204 tmp != &all_mddevs;}); \ 205 ({ spin_lock(&all_mddevs_lock); \ 206 tmp = tmp->next;}) \ 207 ) 208 209 210 /* Rather than calling directly into the personality make_request function, 211 * IO requests come here first so that we can check if the device is 212 * being suspended pending a reconfiguration. 213 * We hold a refcount over the call to ->make_request. By the time that 214 * call has finished, the bio has been linked into some internal structure 215 * and so is visible to ->quiesce(), so we don't need the refcount any more. 216 */ 217 static int md_make_request(struct request_queue *q, struct bio *bio) 218 { 219 const int rw = bio_data_dir(bio); 220 mddev_t *mddev = q->queuedata; 221 int rv; 222 int cpu; 223 224 if (mddev == NULL || mddev->pers == NULL) { 225 bio_io_error(bio); 226 return 0; 227 } 228 rcu_read_lock(); 229 if (mddev->suspended || mddev->barrier) { 230 DEFINE_WAIT(__wait); 231 for (;;) { 232 prepare_to_wait(&mddev->sb_wait, &__wait, 233 TASK_UNINTERRUPTIBLE); 234 if (!mddev->suspended && !mddev->barrier) 235 break; 236 rcu_read_unlock(); 237 schedule(); 238 rcu_read_lock(); 239 } 240 finish_wait(&mddev->sb_wait, &__wait); 241 } 242 atomic_inc(&mddev->active_io); 243 rcu_read_unlock(); 244 245 rv = mddev->pers->make_request(mddev, bio); 246 247 cpu = part_stat_lock(); 248 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); 249 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], 250 bio_sectors(bio)); 251 part_stat_unlock(); 252 253 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) 254 wake_up(&mddev->sb_wait); 255 256 return rv; 257 } 258 259 /* mddev_suspend makes sure no new requests are submitted 260 * to the device, and that any requests that have been submitted 261 * are completely handled. 262 * Once ->stop is called and completes, the module will be completely 263 * unused. 264 */ 265 void mddev_suspend(mddev_t *mddev) 266 { 267 BUG_ON(mddev->suspended); 268 mddev->suspended = 1; 269 synchronize_rcu(); 270 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); 271 mddev->pers->quiesce(mddev, 1); 272 } 273 EXPORT_SYMBOL_GPL(mddev_suspend); 274 275 void mddev_resume(mddev_t *mddev) 276 { 277 mddev->suspended = 0; 278 wake_up(&mddev->sb_wait); 279 mddev->pers->quiesce(mddev, 0); 280 } 281 EXPORT_SYMBOL_GPL(mddev_resume); 282 283 int mddev_congested(mddev_t *mddev, int bits) 284 { 285 if (mddev->barrier) 286 return 1; 287 return mddev->suspended; 288 } 289 EXPORT_SYMBOL(mddev_congested); 290 291 /* 292 * Generic barrier handling for md 293 */ 294 295 #define POST_REQUEST_BARRIER ((void*)1) 296 297 static void md_end_barrier(struct bio *bio, int err) 298 { 299 mdk_rdev_t *rdev = bio->bi_private; 300 mddev_t *mddev = rdev->mddev; 301 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER) 302 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags); 303 304 rdev_dec_pending(rdev, mddev); 305 306 if (atomic_dec_and_test(&mddev->flush_pending)) { 307 if (mddev->barrier == POST_REQUEST_BARRIER) { 308 /* This was a post-request barrier */ 309 mddev->barrier = NULL; 310 wake_up(&mddev->sb_wait); 311 } else 312 /* The pre-request barrier has finished */ 313 schedule_work(&mddev->barrier_work); 314 } 315 bio_put(bio); 316 } 317 318 static void submit_barriers(mddev_t *mddev) 319 { 320 mdk_rdev_t *rdev; 321 322 rcu_read_lock(); 323 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 324 if (rdev->raid_disk >= 0 && 325 !test_bit(Faulty, &rdev->flags)) { 326 /* Take two references, one is dropped 327 * when request finishes, one after 328 * we reclaim rcu_read_lock 329 */ 330 struct bio *bi; 331 atomic_inc(&rdev->nr_pending); 332 atomic_inc(&rdev->nr_pending); 333 rcu_read_unlock(); 334 bi = bio_alloc(GFP_KERNEL, 0); 335 bi->bi_end_io = md_end_barrier; 336 bi->bi_private = rdev; 337 bi->bi_bdev = rdev->bdev; 338 atomic_inc(&mddev->flush_pending); 339 submit_bio(WRITE_BARRIER, bi); 340 rcu_read_lock(); 341 rdev_dec_pending(rdev, mddev); 342 } 343 rcu_read_unlock(); 344 } 345 346 static void md_submit_barrier(struct work_struct *ws) 347 { 348 mddev_t *mddev = container_of(ws, mddev_t, barrier_work); 349 struct bio *bio = mddev->barrier; 350 351 atomic_set(&mddev->flush_pending, 1); 352 353 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags)) 354 bio_endio(bio, -EOPNOTSUPP); 355 else if (bio->bi_size == 0) 356 /* an empty barrier - all done */ 357 bio_endio(bio, 0); 358 else { 359 bio->bi_rw &= ~REQ_HARDBARRIER; 360 if (mddev->pers->make_request(mddev, bio)) 361 generic_make_request(bio); 362 mddev->barrier = POST_REQUEST_BARRIER; 363 submit_barriers(mddev); 364 } 365 if (atomic_dec_and_test(&mddev->flush_pending)) { 366 mddev->barrier = NULL; 367 wake_up(&mddev->sb_wait); 368 } 369 } 370 371 void md_barrier_request(mddev_t *mddev, struct bio *bio) 372 { 373 spin_lock_irq(&mddev->write_lock); 374 wait_event_lock_irq(mddev->sb_wait, 375 !mddev->barrier, 376 mddev->write_lock, /*nothing*/); 377 mddev->barrier = bio; 378 spin_unlock_irq(&mddev->write_lock); 379 380 atomic_set(&mddev->flush_pending, 1); 381 INIT_WORK(&mddev->barrier_work, md_submit_barrier); 382 383 submit_barriers(mddev); 384 385 if (atomic_dec_and_test(&mddev->flush_pending)) 386 schedule_work(&mddev->barrier_work); 387 } 388 EXPORT_SYMBOL(md_barrier_request); 389 390 /* Support for plugging. 391 * This mirrors the plugging support in request_queue, but does not 392 * require having a whole queue 393 */ 394 static void plugger_work(struct work_struct *work) 395 { 396 struct plug_handle *plug = 397 container_of(work, struct plug_handle, unplug_work); 398 plug->unplug_fn(plug); 399 } 400 static void plugger_timeout(unsigned long data) 401 { 402 struct plug_handle *plug = (void *)data; 403 kblockd_schedule_work(NULL, &plug->unplug_work); 404 } 405 void plugger_init(struct plug_handle *plug, 406 void (*unplug_fn)(struct plug_handle *)) 407 { 408 plug->unplug_flag = 0; 409 plug->unplug_fn = unplug_fn; 410 init_timer(&plug->unplug_timer); 411 plug->unplug_timer.function = plugger_timeout; 412 plug->unplug_timer.data = (unsigned long)plug; 413 INIT_WORK(&plug->unplug_work, plugger_work); 414 } 415 EXPORT_SYMBOL_GPL(plugger_init); 416 417 void plugger_set_plug(struct plug_handle *plug) 418 { 419 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag)) 420 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1); 421 } 422 EXPORT_SYMBOL_GPL(plugger_set_plug); 423 424 int plugger_remove_plug(struct plug_handle *plug) 425 { 426 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) { 427 del_timer(&plug->unplug_timer); 428 return 1; 429 } else 430 return 0; 431 } 432 EXPORT_SYMBOL_GPL(plugger_remove_plug); 433 434 435 static inline mddev_t *mddev_get(mddev_t *mddev) 436 { 437 atomic_inc(&mddev->active); 438 return mddev; 439 } 440 441 static void mddev_delayed_delete(struct work_struct *ws); 442 443 static void mddev_put(mddev_t *mddev) 444 { 445 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) 446 return; 447 if (!mddev->raid_disks && list_empty(&mddev->disks) && 448 mddev->ctime == 0 && !mddev->hold_active) { 449 /* Array is not configured at all, and not held active, 450 * so destroy it */ 451 list_del(&mddev->all_mddevs); 452 if (mddev->gendisk) { 453 /* we did a probe so need to clean up. 454 * Call schedule_work inside the spinlock 455 * so that flush_scheduled_work() after 456 * mddev_find will succeed in waiting for the 457 * work to be done. 458 */ 459 INIT_WORK(&mddev->del_work, mddev_delayed_delete); 460 schedule_work(&mddev->del_work); 461 } else 462 kfree(mddev); 463 } 464 spin_unlock(&all_mddevs_lock); 465 } 466 467 void mddev_init(mddev_t *mddev) 468 { 469 mutex_init(&mddev->open_mutex); 470 mutex_init(&mddev->reconfig_mutex); 471 mutex_init(&mddev->bitmap_info.mutex); 472 INIT_LIST_HEAD(&mddev->disks); 473 INIT_LIST_HEAD(&mddev->all_mddevs); 474 init_timer(&mddev->safemode_timer); 475 atomic_set(&mddev->active, 1); 476 atomic_set(&mddev->openers, 0); 477 atomic_set(&mddev->active_io, 0); 478 spin_lock_init(&mddev->write_lock); 479 atomic_set(&mddev->flush_pending, 0); 480 init_waitqueue_head(&mddev->sb_wait); 481 init_waitqueue_head(&mddev->recovery_wait); 482 mddev->reshape_position = MaxSector; 483 mddev->resync_min = 0; 484 mddev->resync_max = MaxSector; 485 mddev->level = LEVEL_NONE; 486 } 487 EXPORT_SYMBOL_GPL(mddev_init); 488 489 static mddev_t * mddev_find(dev_t unit) 490 { 491 mddev_t *mddev, *new = NULL; 492 493 retry: 494 spin_lock(&all_mddevs_lock); 495 496 if (unit) { 497 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 498 if (mddev->unit == unit) { 499 mddev_get(mddev); 500 spin_unlock(&all_mddevs_lock); 501 kfree(new); 502 return mddev; 503 } 504 505 if (new) { 506 list_add(&new->all_mddevs, &all_mddevs); 507 spin_unlock(&all_mddevs_lock); 508 new->hold_active = UNTIL_IOCTL; 509 return new; 510 } 511 } else if (new) { 512 /* find an unused unit number */ 513 static int next_minor = 512; 514 int start = next_minor; 515 int is_free = 0; 516 int dev = 0; 517 while (!is_free) { 518 dev = MKDEV(MD_MAJOR, next_minor); 519 next_minor++; 520 if (next_minor > MINORMASK) 521 next_minor = 0; 522 if (next_minor == start) { 523 /* Oh dear, all in use. */ 524 spin_unlock(&all_mddevs_lock); 525 kfree(new); 526 return NULL; 527 } 528 529 is_free = 1; 530 list_for_each_entry(mddev, &all_mddevs, all_mddevs) 531 if (mddev->unit == dev) { 532 is_free = 0; 533 break; 534 } 535 } 536 new->unit = dev; 537 new->md_minor = MINOR(dev); 538 new->hold_active = UNTIL_STOP; 539 list_add(&new->all_mddevs, &all_mddevs); 540 spin_unlock(&all_mddevs_lock); 541 return new; 542 } 543 spin_unlock(&all_mddevs_lock); 544 545 new = kzalloc(sizeof(*new), GFP_KERNEL); 546 if (!new) 547 return NULL; 548 549 new->unit = unit; 550 if (MAJOR(unit) == MD_MAJOR) 551 new->md_minor = MINOR(unit); 552 else 553 new->md_minor = MINOR(unit) >> MdpMinorShift; 554 555 mddev_init(new); 556 557 goto retry; 558 } 559 560 static inline int mddev_lock(mddev_t * mddev) 561 { 562 return mutex_lock_interruptible(&mddev->reconfig_mutex); 563 } 564 565 static inline int mddev_is_locked(mddev_t *mddev) 566 { 567 return mutex_is_locked(&mddev->reconfig_mutex); 568 } 569 570 static inline int mddev_trylock(mddev_t * mddev) 571 { 572 return mutex_trylock(&mddev->reconfig_mutex); 573 } 574 575 static struct attribute_group md_redundancy_group; 576 577 static void mddev_unlock(mddev_t * mddev) 578 { 579 if (mddev->to_remove) { 580 /* These cannot be removed under reconfig_mutex as 581 * an access to the files will try to take reconfig_mutex 582 * while holding the file unremovable, which leads to 583 * a deadlock. 584 * So hold set sysfs_active while the remove in happeing, 585 * and anything else which might set ->to_remove or my 586 * otherwise change the sysfs namespace will fail with 587 * -EBUSY if sysfs_active is still set. 588 * We set sysfs_active under reconfig_mutex and elsewhere 589 * test it under the same mutex to ensure its correct value 590 * is seen. 591 */ 592 struct attribute_group *to_remove = mddev->to_remove; 593 mddev->to_remove = NULL; 594 mddev->sysfs_active = 1; 595 mutex_unlock(&mddev->reconfig_mutex); 596 597 if (mddev->kobj.sd) { 598 if (to_remove != &md_redundancy_group) 599 sysfs_remove_group(&mddev->kobj, to_remove); 600 if (mddev->pers == NULL || 601 mddev->pers->sync_request == NULL) { 602 sysfs_remove_group(&mddev->kobj, &md_redundancy_group); 603 if (mddev->sysfs_action) 604 sysfs_put(mddev->sysfs_action); 605 mddev->sysfs_action = NULL; 606 } 607 } 608 mddev->sysfs_active = 0; 609 } else 610 mutex_unlock(&mddev->reconfig_mutex); 611 612 md_wakeup_thread(mddev->thread); 613 } 614 615 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) 616 { 617 mdk_rdev_t *rdev; 618 619 list_for_each_entry(rdev, &mddev->disks, same_set) 620 if (rdev->desc_nr == nr) 621 return rdev; 622 623 return NULL; 624 } 625 626 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) 627 { 628 mdk_rdev_t *rdev; 629 630 list_for_each_entry(rdev, &mddev->disks, same_set) 631 if (rdev->bdev->bd_dev == dev) 632 return rdev; 633 634 return NULL; 635 } 636 637 static struct mdk_personality *find_pers(int level, char *clevel) 638 { 639 struct mdk_personality *pers; 640 list_for_each_entry(pers, &pers_list, list) { 641 if (level != LEVEL_NONE && pers->level == level) 642 return pers; 643 if (strcmp(pers->name, clevel)==0) 644 return pers; 645 } 646 return NULL; 647 } 648 649 /* return the offset of the super block in 512byte sectors */ 650 static inline sector_t calc_dev_sboffset(struct block_device *bdev) 651 { 652 sector_t num_sectors = bdev->bd_inode->i_size / 512; 653 return MD_NEW_SIZE_SECTORS(num_sectors); 654 } 655 656 static int alloc_disk_sb(mdk_rdev_t * rdev) 657 { 658 if (rdev->sb_page) 659 MD_BUG(); 660 661 rdev->sb_page = alloc_page(GFP_KERNEL); 662 if (!rdev->sb_page) { 663 printk(KERN_ALERT "md: out of memory.\n"); 664 return -ENOMEM; 665 } 666 667 return 0; 668 } 669 670 static void free_disk_sb(mdk_rdev_t * rdev) 671 { 672 if (rdev->sb_page) { 673 put_page(rdev->sb_page); 674 rdev->sb_loaded = 0; 675 rdev->sb_page = NULL; 676 rdev->sb_start = 0; 677 rdev->sectors = 0; 678 } 679 } 680 681 682 static void super_written(struct bio *bio, int error) 683 { 684 mdk_rdev_t *rdev = bio->bi_private; 685 mddev_t *mddev = rdev->mddev; 686 687 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { 688 printk("md: super_written gets error=%d, uptodate=%d\n", 689 error, test_bit(BIO_UPTODATE, &bio->bi_flags)); 690 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); 691 md_error(mddev, rdev); 692 } 693 694 if (atomic_dec_and_test(&mddev->pending_writes)) 695 wake_up(&mddev->sb_wait); 696 bio_put(bio); 697 } 698 699 static void super_written_barrier(struct bio *bio, int error) 700 { 701 struct bio *bio2 = bio->bi_private; 702 mdk_rdev_t *rdev = bio2->bi_private; 703 mddev_t *mddev = rdev->mddev; 704 705 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && 706 error == -EOPNOTSUPP) { 707 unsigned long flags; 708 /* barriers don't appear to be supported :-( */ 709 set_bit(BarriersNotsupp, &rdev->flags); 710 mddev->barriers_work = 0; 711 spin_lock_irqsave(&mddev->write_lock, flags); 712 bio2->bi_next = mddev->biolist; 713 mddev->biolist = bio2; 714 spin_unlock_irqrestore(&mddev->write_lock, flags); 715 wake_up(&mddev->sb_wait); 716 bio_put(bio); 717 } else { 718 bio_put(bio2); 719 bio->bi_private = rdev; 720 super_written(bio, error); 721 } 722 } 723 724 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, 725 sector_t sector, int size, struct page *page) 726 { 727 /* write first size bytes of page to sector of rdev 728 * Increment mddev->pending_writes before returning 729 * and decrement it on completion, waking up sb_wait 730 * if zero is reached. 731 * If an error occurred, call md_error 732 * 733 * As we might need to resubmit the request if REQ_HARDBARRIER 734 * causes ENOTSUPP, we allocate a spare bio... 735 */ 736 struct bio *bio = bio_alloc(GFP_NOIO, 1); 737 int rw = REQ_WRITE | REQ_SYNC | REQ_UNPLUG; 738 739 bio->bi_bdev = rdev->bdev; 740 bio->bi_sector = sector; 741 bio_add_page(bio, page, size, 0); 742 bio->bi_private = rdev; 743 bio->bi_end_io = super_written; 744 bio->bi_rw = rw; 745 746 atomic_inc(&mddev->pending_writes); 747 if (!test_bit(BarriersNotsupp, &rdev->flags)) { 748 struct bio *rbio; 749 rw |= REQ_HARDBARRIER; 750 rbio = bio_clone(bio, GFP_NOIO); 751 rbio->bi_private = bio; 752 rbio->bi_end_io = super_written_barrier; 753 submit_bio(rw, rbio); 754 } else 755 submit_bio(rw, bio); 756 } 757 758 void md_super_wait(mddev_t *mddev) 759 { 760 /* wait for all superblock writes that were scheduled to complete. 761 * if any had to be retried (due to BARRIER problems), retry them 762 */ 763 DEFINE_WAIT(wq); 764 for(;;) { 765 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); 766 if (atomic_read(&mddev->pending_writes)==0) 767 break; 768 while (mddev->biolist) { 769 struct bio *bio; 770 spin_lock_irq(&mddev->write_lock); 771 bio = mddev->biolist; 772 mddev->biolist = bio->bi_next ; 773 bio->bi_next = NULL; 774 spin_unlock_irq(&mddev->write_lock); 775 submit_bio(bio->bi_rw, bio); 776 } 777 schedule(); 778 } 779 finish_wait(&mddev->sb_wait, &wq); 780 } 781 782 static void bi_complete(struct bio *bio, int error) 783 { 784 complete((struct completion*)bio->bi_private); 785 } 786 787 int sync_page_io(struct block_device *bdev, sector_t sector, int size, 788 struct page *page, int rw) 789 { 790 struct bio *bio = bio_alloc(GFP_NOIO, 1); 791 struct completion event; 792 int ret; 793 794 rw |= REQ_SYNC | REQ_UNPLUG; 795 796 bio->bi_bdev = bdev; 797 bio->bi_sector = sector; 798 bio_add_page(bio, page, size, 0); 799 init_completion(&event); 800 bio->bi_private = &event; 801 bio->bi_end_io = bi_complete; 802 submit_bio(rw, bio); 803 wait_for_completion(&event); 804 805 ret = test_bit(BIO_UPTODATE, &bio->bi_flags); 806 bio_put(bio); 807 return ret; 808 } 809 EXPORT_SYMBOL_GPL(sync_page_io); 810 811 static int read_disk_sb(mdk_rdev_t * rdev, int size) 812 { 813 char b[BDEVNAME_SIZE]; 814 if (!rdev->sb_page) { 815 MD_BUG(); 816 return -EINVAL; 817 } 818 if (rdev->sb_loaded) 819 return 0; 820 821 822 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) 823 goto fail; 824 rdev->sb_loaded = 1; 825 return 0; 826 827 fail: 828 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", 829 bdevname(rdev->bdev,b)); 830 return -EINVAL; 831 } 832 833 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) 834 { 835 return sb1->set_uuid0 == sb2->set_uuid0 && 836 sb1->set_uuid1 == sb2->set_uuid1 && 837 sb1->set_uuid2 == sb2->set_uuid2 && 838 sb1->set_uuid3 == sb2->set_uuid3; 839 } 840 841 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) 842 { 843 int ret; 844 mdp_super_t *tmp1, *tmp2; 845 846 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); 847 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); 848 849 if (!tmp1 || !tmp2) { 850 ret = 0; 851 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); 852 goto abort; 853 } 854 855 *tmp1 = *sb1; 856 *tmp2 = *sb2; 857 858 /* 859 * nr_disks is not constant 860 */ 861 tmp1->nr_disks = 0; 862 tmp2->nr_disks = 0; 863 864 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); 865 abort: 866 kfree(tmp1); 867 kfree(tmp2); 868 return ret; 869 } 870 871 872 static u32 md_csum_fold(u32 csum) 873 { 874 csum = (csum & 0xffff) + (csum >> 16); 875 return (csum & 0xffff) + (csum >> 16); 876 } 877 878 static unsigned int calc_sb_csum(mdp_super_t * sb) 879 { 880 u64 newcsum = 0; 881 u32 *sb32 = (u32*)sb; 882 int i; 883 unsigned int disk_csum, csum; 884 885 disk_csum = sb->sb_csum; 886 sb->sb_csum = 0; 887 888 for (i = 0; i < MD_SB_BYTES/4 ; i++) 889 newcsum += sb32[i]; 890 csum = (newcsum & 0xffffffff) + (newcsum>>32); 891 892 893 #ifdef CONFIG_ALPHA 894 /* This used to use csum_partial, which was wrong for several 895 * reasons including that different results are returned on 896 * different architectures. It isn't critical that we get exactly 897 * the same return value as before (we always csum_fold before 898 * testing, and that removes any differences). However as we 899 * know that csum_partial always returned a 16bit value on 900 * alphas, do a fold to maximise conformity to previous behaviour. 901 */ 902 sb->sb_csum = md_csum_fold(disk_csum); 903 #else 904 sb->sb_csum = disk_csum; 905 #endif 906 return csum; 907 } 908 909 910 /* 911 * Handle superblock details. 912 * We want to be able to handle multiple superblock formats 913 * so we have a common interface to them all, and an array of 914 * different handlers. 915 * We rely on user-space to write the initial superblock, and support 916 * reading and updating of superblocks. 917 * Interface methods are: 918 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) 919 * loads and validates a superblock on dev. 920 * if refdev != NULL, compare superblocks on both devices 921 * Return: 922 * 0 - dev has a superblock that is compatible with refdev 923 * 1 - dev has a superblock that is compatible and newer than refdev 924 * so dev should be used as the refdev in future 925 * -EINVAL superblock incompatible or invalid 926 * -othererror e.g. -EIO 927 * 928 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) 929 * Verify that dev is acceptable into mddev. 930 * The first time, mddev->raid_disks will be 0, and data from 931 * dev should be merged in. Subsequent calls check that dev 932 * is new enough. Return 0 or -EINVAL 933 * 934 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) 935 * Update the superblock for rdev with data in mddev 936 * This does not write to disc. 937 * 938 */ 939 940 struct super_type { 941 char *name; 942 struct module *owner; 943 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, 944 int minor_version); 945 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); 946 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); 947 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, 948 sector_t num_sectors); 949 }; 950 951 /* 952 * Check that the given mddev has no bitmap. 953 * 954 * This function is called from the run method of all personalities that do not 955 * support bitmaps. It prints an error message and returns non-zero if mddev 956 * has a bitmap. Otherwise, it returns 0. 957 * 958 */ 959 int md_check_no_bitmap(mddev_t *mddev) 960 { 961 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) 962 return 0; 963 printk(KERN_ERR "%s: bitmaps are not supported for %s\n", 964 mdname(mddev), mddev->pers->name); 965 return 1; 966 } 967 EXPORT_SYMBOL(md_check_no_bitmap); 968 969 /* 970 * load_super for 0.90.0 971 */ 972 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 973 { 974 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 975 mdp_super_t *sb; 976 int ret; 977 978 /* 979 * Calculate the position of the superblock (512byte sectors), 980 * it's at the end of the disk. 981 * 982 * It also happens to be a multiple of 4Kb. 983 */ 984 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 985 986 ret = read_disk_sb(rdev, MD_SB_BYTES); 987 if (ret) return ret; 988 989 ret = -EINVAL; 990 991 bdevname(rdev->bdev, b); 992 sb = (mdp_super_t*)page_address(rdev->sb_page); 993 994 if (sb->md_magic != MD_SB_MAGIC) { 995 printk(KERN_ERR "md: invalid raid superblock magic on %s\n", 996 b); 997 goto abort; 998 } 999 1000 if (sb->major_version != 0 || 1001 sb->minor_version < 90 || 1002 sb->minor_version > 91) { 1003 printk(KERN_WARNING "Bad version number %d.%d on %s\n", 1004 sb->major_version, sb->minor_version, 1005 b); 1006 goto abort; 1007 } 1008 1009 if (sb->raid_disks <= 0) 1010 goto abort; 1011 1012 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { 1013 printk(KERN_WARNING "md: invalid superblock checksum on %s\n", 1014 b); 1015 goto abort; 1016 } 1017 1018 rdev->preferred_minor = sb->md_minor; 1019 rdev->data_offset = 0; 1020 rdev->sb_size = MD_SB_BYTES; 1021 1022 if (sb->level == LEVEL_MULTIPATH) 1023 rdev->desc_nr = -1; 1024 else 1025 rdev->desc_nr = sb->this_disk.number; 1026 1027 if (!refdev) { 1028 ret = 1; 1029 } else { 1030 __u64 ev1, ev2; 1031 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); 1032 if (!uuid_equal(refsb, sb)) { 1033 printk(KERN_WARNING "md: %s has different UUID to %s\n", 1034 b, bdevname(refdev->bdev,b2)); 1035 goto abort; 1036 } 1037 if (!sb_equal(refsb, sb)) { 1038 printk(KERN_WARNING "md: %s has same UUID" 1039 " but different superblock to %s\n", 1040 b, bdevname(refdev->bdev, b2)); 1041 goto abort; 1042 } 1043 ev1 = md_event(sb); 1044 ev2 = md_event(refsb); 1045 if (ev1 > ev2) 1046 ret = 1; 1047 else 1048 ret = 0; 1049 } 1050 rdev->sectors = rdev->sb_start; 1051 1052 if (rdev->sectors < sb->size * 2 && sb->level > 1) 1053 /* "this cannot possibly happen" ... */ 1054 ret = -EINVAL; 1055 1056 abort: 1057 return ret; 1058 } 1059 1060 /* 1061 * validate_super for 0.90.0 1062 */ 1063 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1064 { 1065 mdp_disk_t *desc; 1066 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); 1067 __u64 ev1 = md_event(sb); 1068 1069 rdev->raid_disk = -1; 1070 clear_bit(Faulty, &rdev->flags); 1071 clear_bit(In_sync, &rdev->flags); 1072 clear_bit(WriteMostly, &rdev->flags); 1073 clear_bit(BarriersNotsupp, &rdev->flags); 1074 1075 if (mddev->raid_disks == 0) { 1076 mddev->major_version = 0; 1077 mddev->minor_version = sb->minor_version; 1078 mddev->patch_version = sb->patch_version; 1079 mddev->external = 0; 1080 mddev->chunk_sectors = sb->chunk_size >> 9; 1081 mddev->ctime = sb->ctime; 1082 mddev->utime = sb->utime; 1083 mddev->level = sb->level; 1084 mddev->clevel[0] = 0; 1085 mddev->layout = sb->layout; 1086 mddev->raid_disks = sb->raid_disks; 1087 mddev->dev_sectors = sb->size * 2; 1088 mddev->events = ev1; 1089 mddev->bitmap_info.offset = 0; 1090 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 1091 1092 if (mddev->minor_version >= 91) { 1093 mddev->reshape_position = sb->reshape_position; 1094 mddev->delta_disks = sb->delta_disks; 1095 mddev->new_level = sb->new_level; 1096 mddev->new_layout = sb->new_layout; 1097 mddev->new_chunk_sectors = sb->new_chunk >> 9; 1098 } else { 1099 mddev->reshape_position = MaxSector; 1100 mddev->delta_disks = 0; 1101 mddev->new_level = mddev->level; 1102 mddev->new_layout = mddev->layout; 1103 mddev->new_chunk_sectors = mddev->chunk_sectors; 1104 } 1105 1106 if (sb->state & (1<<MD_SB_CLEAN)) 1107 mddev->recovery_cp = MaxSector; 1108 else { 1109 if (sb->events_hi == sb->cp_events_hi && 1110 sb->events_lo == sb->cp_events_lo) { 1111 mddev->recovery_cp = sb->recovery_cp; 1112 } else 1113 mddev->recovery_cp = 0; 1114 } 1115 1116 memcpy(mddev->uuid+0, &sb->set_uuid0, 4); 1117 memcpy(mddev->uuid+4, &sb->set_uuid1, 4); 1118 memcpy(mddev->uuid+8, &sb->set_uuid2, 4); 1119 memcpy(mddev->uuid+12,&sb->set_uuid3, 4); 1120 1121 mddev->max_disks = MD_SB_DISKS; 1122 1123 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && 1124 mddev->bitmap_info.file == NULL) 1125 mddev->bitmap_info.offset = 1126 mddev->bitmap_info.default_offset; 1127 1128 } else if (mddev->pers == NULL) { 1129 /* Insist on good event counter while assembling, except 1130 * for spares (which don't need an event count) */ 1131 ++ev1; 1132 if (sb->disks[rdev->desc_nr].state & ( 1133 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) 1134 if (ev1 < mddev->events) 1135 return -EINVAL; 1136 } else if (mddev->bitmap) { 1137 /* if adding to array with a bitmap, then we can accept an 1138 * older device ... but not too old. 1139 */ 1140 if (ev1 < mddev->bitmap->events_cleared) 1141 return 0; 1142 } else { 1143 if (ev1 < mddev->events) 1144 /* just a hot-add of a new device, leave raid_disk at -1 */ 1145 return 0; 1146 } 1147 1148 if (mddev->level != LEVEL_MULTIPATH) { 1149 desc = sb->disks + rdev->desc_nr; 1150 1151 if (desc->state & (1<<MD_DISK_FAULTY)) 1152 set_bit(Faulty, &rdev->flags); 1153 else if (desc->state & (1<<MD_DISK_SYNC) /* && 1154 desc->raid_disk < mddev->raid_disks */) { 1155 set_bit(In_sync, &rdev->flags); 1156 rdev->raid_disk = desc->raid_disk; 1157 } else if (desc->state & (1<<MD_DISK_ACTIVE)) { 1158 /* active but not in sync implies recovery up to 1159 * reshape position. We don't know exactly where 1160 * that is, so set to zero for now */ 1161 if (mddev->minor_version >= 91) { 1162 rdev->recovery_offset = 0; 1163 rdev->raid_disk = desc->raid_disk; 1164 } 1165 } 1166 if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) 1167 set_bit(WriteMostly, &rdev->flags); 1168 } else /* MULTIPATH are always insync */ 1169 set_bit(In_sync, &rdev->flags); 1170 return 0; 1171 } 1172 1173 /* 1174 * sync_super for 0.90.0 1175 */ 1176 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1177 { 1178 mdp_super_t *sb; 1179 mdk_rdev_t *rdev2; 1180 int next_spare = mddev->raid_disks; 1181 1182 1183 /* make rdev->sb match mddev data.. 1184 * 1185 * 1/ zero out disks 1186 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); 1187 * 3/ any empty disks < next_spare become removed 1188 * 1189 * disks[0] gets initialised to REMOVED because 1190 * we cannot be sure from other fields if it has 1191 * been initialised or not. 1192 */ 1193 int i; 1194 int active=0, working=0,failed=0,spare=0,nr_disks=0; 1195 1196 rdev->sb_size = MD_SB_BYTES; 1197 1198 sb = (mdp_super_t*)page_address(rdev->sb_page); 1199 1200 memset(sb, 0, sizeof(*sb)); 1201 1202 sb->md_magic = MD_SB_MAGIC; 1203 sb->major_version = mddev->major_version; 1204 sb->patch_version = mddev->patch_version; 1205 sb->gvalid_words = 0; /* ignored */ 1206 memcpy(&sb->set_uuid0, mddev->uuid+0, 4); 1207 memcpy(&sb->set_uuid1, mddev->uuid+4, 4); 1208 memcpy(&sb->set_uuid2, mddev->uuid+8, 4); 1209 memcpy(&sb->set_uuid3, mddev->uuid+12,4); 1210 1211 sb->ctime = mddev->ctime; 1212 sb->level = mddev->level; 1213 sb->size = mddev->dev_sectors / 2; 1214 sb->raid_disks = mddev->raid_disks; 1215 sb->md_minor = mddev->md_minor; 1216 sb->not_persistent = 0; 1217 sb->utime = mddev->utime; 1218 sb->state = 0; 1219 sb->events_hi = (mddev->events>>32); 1220 sb->events_lo = (u32)mddev->events; 1221 1222 if (mddev->reshape_position == MaxSector) 1223 sb->minor_version = 90; 1224 else { 1225 sb->minor_version = 91; 1226 sb->reshape_position = mddev->reshape_position; 1227 sb->new_level = mddev->new_level; 1228 sb->delta_disks = mddev->delta_disks; 1229 sb->new_layout = mddev->new_layout; 1230 sb->new_chunk = mddev->new_chunk_sectors << 9; 1231 } 1232 mddev->minor_version = sb->minor_version; 1233 if (mddev->in_sync) 1234 { 1235 sb->recovery_cp = mddev->recovery_cp; 1236 sb->cp_events_hi = (mddev->events>>32); 1237 sb->cp_events_lo = (u32)mddev->events; 1238 if (mddev->recovery_cp == MaxSector) 1239 sb->state = (1<< MD_SB_CLEAN); 1240 } else 1241 sb->recovery_cp = 0; 1242 1243 sb->layout = mddev->layout; 1244 sb->chunk_size = mddev->chunk_sectors << 9; 1245 1246 if (mddev->bitmap && mddev->bitmap_info.file == NULL) 1247 sb->state |= (1<<MD_SB_BITMAP_PRESENT); 1248 1249 sb->disks[0].state = (1<<MD_DISK_REMOVED); 1250 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1251 mdp_disk_t *d; 1252 int desc_nr; 1253 int is_active = test_bit(In_sync, &rdev2->flags); 1254 1255 if (rdev2->raid_disk >= 0 && 1256 sb->minor_version >= 91) 1257 /* we have nowhere to store the recovery_offset, 1258 * but if it is not below the reshape_position, 1259 * we can piggy-back on that. 1260 */ 1261 is_active = 1; 1262 if (rdev2->raid_disk < 0 || 1263 test_bit(Faulty, &rdev2->flags)) 1264 is_active = 0; 1265 if (is_active) 1266 desc_nr = rdev2->raid_disk; 1267 else 1268 desc_nr = next_spare++; 1269 rdev2->desc_nr = desc_nr; 1270 d = &sb->disks[rdev2->desc_nr]; 1271 nr_disks++; 1272 d->number = rdev2->desc_nr; 1273 d->major = MAJOR(rdev2->bdev->bd_dev); 1274 d->minor = MINOR(rdev2->bdev->bd_dev); 1275 if (is_active) 1276 d->raid_disk = rdev2->raid_disk; 1277 else 1278 d->raid_disk = rdev2->desc_nr; /* compatibility */ 1279 if (test_bit(Faulty, &rdev2->flags)) 1280 d->state = (1<<MD_DISK_FAULTY); 1281 else if (is_active) { 1282 d->state = (1<<MD_DISK_ACTIVE); 1283 if (test_bit(In_sync, &rdev2->flags)) 1284 d->state |= (1<<MD_DISK_SYNC); 1285 active++; 1286 working++; 1287 } else { 1288 d->state = 0; 1289 spare++; 1290 working++; 1291 } 1292 if (test_bit(WriteMostly, &rdev2->flags)) 1293 d->state |= (1<<MD_DISK_WRITEMOSTLY); 1294 } 1295 /* now set the "removed" and "faulty" bits on any missing devices */ 1296 for (i=0 ; i < mddev->raid_disks ; i++) { 1297 mdp_disk_t *d = &sb->disks[i]; 1298 if (d->state == 0 && d->number == 0) { 1299 d->number = i; 1300 d->raid_disk = i; 1301 d->state = (1<<MD_DISK_REMOVED); 1302 d->state |= (1<<MD_DISK_FAULTY); 1303 failed++; 1304 } 1305 } 1306 sb->nr_disks = nr_disks; 1307 sb->active_disks = active; 1308 sb->working_disks = working; 1309 sb->failed_disks = failed; 1310 sb->spare_disks = spare; 1311 1312 sb->this_disk = sb->disks[rdev->desc_nr]; 1313 sb->sb_csum = calc_sb_csum(sb); 1314 } 1315 1316 /* 1317 * rdev_size_change for 0.90.0 1318 */ 1319 static unsigned long long 1320 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1321 { 1322 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1323 return 0; /* component must fit device */ 1324 if (rdev->mddev->bitmap_info.offset) 1325 return 0; /* can't move bitmap */ 1326 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 1327 if (!num_sectors || num_sectors > rdev->sb_start) 1328 num_sectors = rdev->sb_start; 1329 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1330 rdev->sb_page); 1331 md_super_wait(rdev->mddev); 1332 return num_sectors / 2; /* kB for sysfs */ 1333 } 1334 1335 1336 /* 1337 * version 1 superblock 1338 */ 1339 1340 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) 1341 { 1342 __le32 disk_csum; 1343 u32 csum; 1344 unsigned long long newcsum; 1345 int size = 256 + le32_to_cpu(sb->max_dev)*2; 1346 __le32 *isuper = (__le32*)sb; 1347 int i; 1348 1349 disk_csum = sb->sb_csum; 1350 sb->sb_csum = 0; 1351 newcsum = 0; 1352 for (i=0; size>=4; size -= 4 ) 1353 newcsum += le32_to_cpu(*isuper++); 1354 1355 if (size == 2) 1356 newcsum += le16_to_cpu(*(__le16*) isuper); 1357 1358 csum = (newcsum & 0xffffffff) + (newcsum >> 32); 1359 sb->sb_csum = disk_csum; 1360 return cpu_to_le32(csum); 1361 } 1362 1363 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) 1364 { 1365 struct mdp_superblock_1 *sb; 1366 int ret; 1367 sector_t sb_start; 1368 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 1369 int bmask; 1370 1371 /* 1372 * Calculate the position of the superblock in 512byte sectors. 1373 * It is always aligned to a 4K boundary and 1374 * depeding on minor_version, it can be: 1375 * 0: At least 8K, but less than 12K, from end of device 1376 * 1: At start of device 1377 * 2: 4K from start of device. 1378 */ 1379 switch(minor_version) { 1380 case 0: 1381 sb_start = rdev->bdev->bd_inode->i_size >> 9; 1382 sb_start -= 8*2; 1383 sb_start &= ~(sector_t)(4*2-1); 1384 break; 1385 case 1: 1386 sb_start = 0; 1387 break; 1388 case 2: 1389 sb_start = 8; 1390 break; 1391 default: 1392 return -EINVAL; 1393 } 1394 rdev->sb_start = sb_start; 1395 1396 /* superblock is rarely larger than 1K, but it can be larger, 1397 * and it is safe to read 4k, so we do that 1398 */ 1399 ret = read_disk_sb(rdev, 4096); 1400 if (ret) return ret; 1401 1402 1403 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1404 1405 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || 1406 sb->major_version != cpu_to_le32(1) || 1407 le32_to_cpu(sb->max_dev) > (4096-256)/2 || 1408 le64_to_cpu(sb->super_offset) != rdev->sb_start || 1409 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) 1410 return -EINVAL; 1411 1412 if (calc_sb_1_csum(sb) != sb->sb_csum) { 1413 printk("md: invalid superblock checksum on %s\n", 1414 bdevname(rdev->bdev,b)); 1415 return -EINVAL; 1416 } 1417 if (le64_to_cpu(sb->data_size) < 10) { 1418 printk("md: data_size too small on %s\n", 1419 bdevname(rdev->bdev,b)); 1420 return -EINVAL; 1421 } 1422 1423 rdev->preferred_minor = 0xffff; 1424 rdev->data_offset = le64_to_cpu(sb->data_offset); 1425 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); 1426 1427 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; 1428 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1429 if (rdev->sb_size & bmask) 1430 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1431 1432 if (minor_version 1433 && rdev->data_offset < sb_start + (rdev->sb_size/512)) 1434 return -EINVAL; 1435 1436 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) 1437 rdev->desc_nr = -1; 1438 else 1439 rdev->desc_nr = le32_to_cpu(sb->dev_number); 1440 1441 if (!refdev) { 1442 ret = 1; 1443 } else { 1444 __u64 ev1, ev2; 1445 struct mdp_superblock_1 *refsb = 1446 (struct mdp_superblock_1*)page_address(refdev->sb_page); 1447 1448 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || 1449 sb->level != refsb->level || 1450 sb->layout != refsb->layout || 1451 sb->chunksize != refsb->chunksize) { 1452 printk(KERN_WARNING "md: %s has strangely different" 1453 " superblock to %s\n", 1454 bdevname(rdev->bdev,b), 1455 bdevname(refdev->bdev,b2)); 1456 return -EINVAL; 1457 } 1458 ev1 = le64_to_cpu(sb->events); 1459 ev2 = le64_to_cpu(refsb->events); 1460 1461 if (ev1 > ev2) 1462 ret = 1; 1463 else 1464 ret = 0; 1465 } 1466 if (minor_version) 1467 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - 1468 le64_to_cpu(sb->data_offset); 1469 else 1470 rdev->sectors = rdev->sb_start; 1471 if (rdev->sectors < le64_to_cpu(sb->data_size)) 1472 return -EINVAL; 1473 rdev->sectors = le64_to_cpu(sb->data_size); 1474 if (le64_to_cpu(sb->size) > rdev->sectors) 1475 return -EINVAL; 1476 return ret; 1477 } 1478 1479 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) 1480 { 1481 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1482 __u64 ev1 = le64_to_cpu(sb->events); 1483 1484 rdev->raid_disk = -1; 1485 clear_bit(Faulty, &rdev->flags); 1486 clear_bit(In_sync, &rdev->flags); 1487 clear_bit(WriteMostly, &rdev->flags); 1488 clear_bit(BarriersNotsupp, &rdev->flags); 1489 1490 if (mddev->raid_disks == 0) { 1491 mddev->major_version = 1; 1492 mddev->patch_version = 0; 1493 mddev->external = 0; 1494 mddev->chunk_sectors = le32_to_cpu(sb->chunksize); 1495 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); 1496 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); 1497 mddev->level = le32_to_cpu(sb->level); 1498 mddev->clevel[0] = 0; 1499 mddev->layout = le32_to_cpu(sb->layout); 1500 mddev->raid_disks = le32_to_cpu(sb->raid_disks); 1501 mddev->dev_sectors = le64_to_cpu(sb->size); 1502 mddev->events = ev1; 1503 mddev->bitmap_info.offset = 0; 1504 mddev->bitmap_info.default_offset = 1024 >> 9; 1505 1506 mddev->recovery_cp = le64_to_cpu(sb->resync_offset); 1507 memcpy(mddev->uuid, sb->set_uuid, 16); 1508 1509 mddev->max_disks = (4096-256)/2; 1510 1511 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && 1512 mddev->bitmap_info.file == NULL ) 1513 mddev->bitmap_info.offset = 1514 (__s32)le32_to_cpu(sb->bitmap_offset); 1515 1516 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { 1517 mddev->reshape_position = le64_to_cpu(sb->reshape_position); 1518 mddev->delta_disks = le32_to_cpu(sb->delta_disks); 1519 mddev->new_level = le32_to_cpu(sb->new_level); 1520 mddev->new_layout = le32_to_cpu(sb->new_layout); 1521 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); 1522 } else { 1523 mddev->reshape_position = MaxSector; 1524 mddev->delta_disks = 0; 1525 mddev->new_level = mddev->level; 1526 mddev->new_layout = mddev->layout; 1527 mddev->new_chunk_sectors = mddev->chunk_sectors; 1528 } 1529 1530 } else if (mddev->pers == NULL) { 1531 /* Insist of good event counter while assembling, except for 1532 * spares (which don't need an event count) */ 1533 ++ev1; 1534 if (rdev->desc_nr >= 0 && 1535 rdev->desc_nr < le32_to_cpu(sb->max_dev) && 1536 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) 1537 if (ev1 < mddev->events) 1538 return -EINVAL; 1539 } else if (mddev->bitmap) { 1540 /* If adding to array with a bitmap, then we can accept an 1541 * older device, but not too old. 1542 */ 1543 if (ev1 < mddev->bitmap->events_cleared) 1544 return 0; 1545 } else { 1546 if (ev1 < mddev->events) 1547 /* just a hot-add of a new device, leave raid_disk at -1 */ 1548 return 0; 1549 } 1550 if (mddev->level != LEVEL_MULTIPATH) { 1551 int role; 1552 if (rdev->desc_nr < 0 || 1553 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { 1554 role = 0xffff; 1555 rdev->desc_nr = -1; 1556 } else 1557 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); 1558 switch(role) { 1559 case 0xffff: /* spare */ 1560 break; 1561 case 0xfffe: /* faulty */ 1562 set_bit(Faulty, &rdev->flags); 1563 break; 1564 default: 1565 if ((le32_to_cpu(sb->feature_map) & 1566 MD_FEATURE_RECOVERY_OFFSET)) 1567 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); 1568 else 1569 set_bit(In_sync, &rdev->flags); 1570 rdev->raid_disk = role; 1571 break; 1572 } 1573 if (sb->devflags & WriteMostly1) 1574 set_bit(WriteMostly, &rdev->flags); 1575 } else /* MULTIPATH are always insync */ 1576 set_bit(In_sync, &rdev->flags); 1577 1578 return 0; 1579 } 1580 1581 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) 1582 { 1583 struct mdp_superblock_1 *sb; 1584 mdk_rdev_t *rdev2; 1585 int max_dev, i; 1586 /* make rdev->sb match mddev and rdev data. */ 1587 1588 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); 1589 1590 sb->feature_map = 0; 1591 sb->pad0 = 0; 1592 sb->recovery_offset = cpu_to_le64(0); 1593 memset(sb->pad1, 0, sizeof(sb->pad1)); 1594 memset(sb->pad2, 0, sizeof(sb->pad2)); 1595 memset(sb->pad3, 0, sizeof(sb->pad3)); 1596 1597 sb->utime = cpu_to_le64((__u64)mddev->utime); 1598 sb->events = cpu_to_le64(mddev->events); 1599 if (mddev->in_sync) 1600 sb->resync_offset = cpu_to_le64(mddev->recovery_cp); 1601 else 1602 sb->resync_offset = cpu_to_le64(0); 1603 1604 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); 1605 1606 sb->raid_disks = cpu_to_le32(mddev->raid_disks); 1607 sb->size = cpu_to_le64(mddev->dev_sectors); 1608 sb->chunksize = cpu_to_le32(mddev->chunk_sectors); 1609 sb->level = cpu_to_le32(mddev->level); 1610 sb->layout = cpu_to_le32(mddev->layout); 1611 1612 if (mddev->bitmap && mddev->bitmap_info.file == NULL) { 1613 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); 1614 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); 1615 } 1616 1617 if (rdev->raid_disk >= 0 && 1618 !test_bit(In_sync, &rdev->flags)) { 1619 sb->feature_map |= 1620 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); 1621 sb->recovery_offset = 1622 cpu_to_le64(rdev->recovery_offset); 1623 } 1624 1625 if (mddev->reshape_position != MaxSector) { 1626 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); 1627 sb->reshape_position = cpu_to_le64(mddev->reshape_position); 1628 sb->new_layout = cpu_to_le32(mddev->new_layout); 1629 sb->delta_disks = cpu_to_le32(mddev->delta_disks); 1630 sb->new_level = cpu_to_le32(mddev->new_level); 1631 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); 1632 } 1633 1634 max_dev = 0; 1635 list_for_each_entry(rdev2, &mddev->disks, same_set) 1636 if (rdev2->desc_nr+1 > max_dev) 1637 max_dev = rdev2->desc_nr+1; 1638 1639 if (max_dev > le32_to_cpu(sb->max_dev)) { 1640 int bmask; 1641 sb->max_dev = cpu_to_le32(max_dev); 1642 rdev->sb_size = max_dev * 2 + 256; 1643 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; 1644 if (rdev->sb_size & bmask) 1645 rdev->sb_size = (rdev->sb_size | bmask) + 1; 1646 } else 1647 max_dev = le32_to_cpu(sb->max_dev); 1648 1649 for (i=0; i<max_dev;i++) 1650 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1651 1652 list_for_each_entry(rdev2, &mddev->disks, same_set) { 1653 i = rdev2->desc_nr; 1654 if (test_bit(Faulty, &rdev2->flags)) 1655 sb->dev_roles[i] = cpu_to_le16(0xfffe); 1656 else if (test_bit(In_sync, &rdev2->flags)) 1657 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1658 else if (rdev2->raid_disk >= 0) 1659 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); 1660 else 1661 sb->dev_roles[i] = cpu_to_le16(0xffff); 1662 } 1663 1664 sb->sb_csum = calc_sb_1_csum(sb); 1665 } 1666 1667 static unsigned long long 1668 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) 1669 { 1670 struct mdp_superblock_1 *sb; 1671 sector_t max_sectors; 1672 if (num_sectors && num_sectors < rdev->mddev->dev_sectors) 1673 return 0; /* component must fit device */ 1674 if (rdev->sb_start < rdev->data_offset) { 1675 /* minor versions 1 and 2; superblock before data */ 1676 max_sectors = rdev->bdev->bd_inode->i_size >> 9; 1677 max_sectors -= rdev->data_offset; 1678 if (!num_sectors || num_sectors > max_sectors) 1679 num_sectors = max_sectors; 1680 } else if (rdev->mddev->bitmap_info.offset) { 1681 /* minor version 0 with bitmap we can't move */ 1682 return 0; 1683 } else { 1684 /* minor version 0; superblock after data */ 1685 sector_t sb_start; 1686 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; 1687 sb_start &= ~(sector_t)(4*2 - 1); 1688 max_sectors = rdev->sectors + sb_start - rdev->sb_start; 1689 if (!num_sectors || num_sectors > max_sectors) 1690 num_sectors = max_sectors; 1691 rdev->sb_start = sb_start; 1692 } 1693 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); 1694 sb->data_size = cpu_to_le64(num_sectors); 1695 sb->super_offset = rdev->sb_start; 1696 sb->sb_csum = calc_sb_1_csum(sb); 1697 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, 1698 rdev->sb_page); 1699 md_super_wait(rdev->mddev); 1700 return num_sectors / 2; /* kB for sysfs */ 1701 } 1702 1703 static struct super_type super_types[] = { 1704 [0] = { 1705 .name = "0.90.0", 1706 .owner = THIS_MODULE, 1707 .load_super = super_90_load, 1708 .validate_super = super_90_validate, 1709 .sync_super = super_90_sync, 1710 .rdev_size_change = super_90_rdev_size_change, 1711 }, 1712 [1] = { 1713 .name = "md-1", 1714 .owner = THIS_MODULE, 1715 .load_super = super_1_load, 1716 .validate_super = super_1_validate, 1717 .sync_super = super_1_sync, 1718 .rdev_size_change = super_1_rdev_size_change, 1719 }, 1720 }; 1721 1722 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) 1723 { 1724 mdk_rdev_t *rdev, *rdev2; 1725 1726 rcu_read_lock(); 1727 rdev_for_each_rcu(rdev, mddev1) 1728 rdev_for_each_rcu(rdev2, mddev2) 1729 if (rdev->bdev->bd_contains == 1730 rdev2->bdev->bd_contains) { 1731 rcu_read_unlock(); 1732 return 1; 1733 } 1734 rcu_read_unlock(); 1735 return 0; 1736 } 1737 1738 static LIST_HEAD(pending_raid_disks); 1739 1740 /* 1741 * Try to register data integrity profile for an mddev 1742 * 1743 * This is called when an array is started and after a disk has been kicked 1744 * from the array. It only succeeds if all working and active component devices 1745 * are integrity capable with matching profiles. 1746 */ 1747 int md_integrity_register(mddev_t *mddev) 1748 { 1749 mdk_rdev_t *rdev, *reference = NULL; 1750 1751 if (list_empty(&mddev->disks)) 1752 return 0; /* nothing to do */ 1753 if (blk_get_integrity(mddev->gendisk)) 1754 return 0; /* already registered */ 1755 list_for_each_entry(rdev, &mddev->disks, same_set) { 1756 /* skip spares and non-functional disks */ 1757 if (test_bit(Faulty, &rdev->flags)) 1758 continue; 1759 if (rdev->raid_disk < 0) 1760 continue; 1761 /* 1762 * If at least one rdev is not integrity capable, we can not 1763 * enable data integrity for the md device. 1764 */ 1765 if (!bdev_get_integrity(rdev->bdev)) 1766 return -EINVAL; 1767 if (!reference) { 1768 /* Use the first rdev as the reference */ 1769 reference = rdev; 1770 continue; 1771 } 1772 /* does this rdev's profile match the reference profile? */ 1773 if (blk_integrity_compare(reference->bdev->bd_disk, 1774 rdev->bdev->bd_disk) < 0) 1775 return -EINVAL; 1776 } 1777 /* 1778 * All component devices are integrity capable and have matching 1779 * profiles, register the common profile for the md device. 1780 */ 1781 if (blk_integrity_register(mddev->gendisk, 1782 bdev_get_integrity(reference->bdev)) != 0) { 1783 printk(KERN_ERR "md: failed to register integrity for %s\n", 1784 mdname(mddev)); 1785 return -EINVAL; 1786 } 1787 printk(KERN_NOTICE "md: data integrity on %s enabled\n", 1788 mdname(mddev)); 1789 return 0; 1790 } 1791 EXPORT_SYMBOL(md_integrity_register); 1792 1793 /* Disable data integrity if non-capable/non-matching disk is being added */ 1794 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 1795 { 1796 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); 1797 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); 1798 1799 if (!bi_mddev) /* nothing to do */ 1800 return; 1801 if (rdev->raid_disk < 0) /* skip spares */ 1802 return; 1803 if (bi_rdev && blk_integrity_compare(mddev->gendisk, 1804 rdev->bdev->bd_disk) >= 0) 1805 return; 1806 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); 1807 blk_integrity_unregister(mddev->gendisk); 1808 } 1809 EXPORT_SYMBOL(md_integrity_add_rdev); 1810 1811 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) 1812 { 1813 char b[BDEVNAME_SIZE]; 1814 struct kobject *ko; 1815 char *s; 1816 int err; 1817 1818 if (rdev->mddev) { 1819 MD_BUG(); 1820 return -EINVAL; 1821 } 1822 1823 /* prevent duplicates */ 1824 if (find_rdev(mddev, rdev->bdev->bd_dev)) 1825 return -EEXIST; 1826 1827 /* make sure rdev->sectors exceeds mddev->dev_sectors */ 1828 if (rdev->sectors && (mddev->dev_sectors == 0 || 1829 rdev->sectors < mddev->dev_sectors)) { 1830 if (mddev->pers) { 1831 /* Cannot change size, so fail 1832 * If mddev->level <= 0, then we don't care 1833 * about aligning sizes (e.g. linear) 1834 */ 1835 if (mddev->level > 0) 1836 return -ENOSPC; 1837 } else 1838 mddev->dev_sectors = rdev->sectors; 1839 } 1840 1841 /* Verify rdev->desc_nr is unique. 1842 * If it is -1, assign a free number, else 1843 * check number is not in use 1844 */ 1845 if (rdev->desc_nr < 0) { 1846 int choice = 0; 1847 if (mddev->pers) choice = mddev->raid_disks; 1848 while (find_rdev_nr(mddev, choice)) 1849 choice++; 1850 rdev->desc_nr = choice; 1851 } else { 1852 if (find_rdev_nr(mddev, rdev->desc_nr)) 1853 return -EBUSY; 1854 } 1855 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { 1856 printk(KERN_WARNING "md: %s: array is limited to %d devices\n", 1857 mdname(mddev), mddev->max_disks); 1858 return -EBUSY; 1859 } 1860 bdevname(rdev->bdev,b); 1861 while ( (s=strchr(b, '/')) != NULL) 1862 *s = '!'; 1863 1864 rdev->mddev = mddev; 1865 printk(KERN_INFO "md: bind<%s>\n", b); 1866 1867 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) 1868 goto fail; 1869 1870 ko = &part_to_dev(rdev->bdev->bd_part)->kobj; 1871 if (sysfs_create_link(&rdev->kobj, ko, "block")) 1872 /* failure here is OK */; 1873 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); 1874 1875 list_add_rcu(&rdev->same_set, &mddev->disks); 1876 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); 1877 1878 /* May as well allow recovery to be retried once */ 1879 mddev->recovery_disabled = 0; 1880 1881 return 0; 1882 1883 fail: 1884 printk(KERN_WARNING "md: failed to register dev-%s for %s\n", 1885 b, mdname(mddev)); 1886 return err; 1887 } 1888 1889 static void md_delayed_delete(struct work_struct *ws) 1890 { 1891 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); 1892 kobject_del(&rdev->kobj); 1893 kobject_put(&rdev->kobj); 1894 } 1895 1896 static void unbind_rdev_from_array(mdk_rdev_t * rdev) 1897 { 1898 char b[BDEVNAME_SIZE]; 1899 if (!rdev->mddev) { 1900 MD_BUG(); 1901 return; 1902 } 1903 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); 1904 list_del_rcu(&rdev->same_set); 1905 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); 1906 rdev->mddev = NULL; 1907 sysfs_remove_link(&rdev->kobj, "block"); 1908 sysfs_put(rdev->sysfs_state); 1909 rdev->sysfs_state = NULL; 1910 /* We need to delay this, otherwise we can deadlock when 1911 * writing to 'remove' to "dev/state". We also need 1912 * to delay it due to rcu usage. 1913 */ 1914 synchronize_rcu(); 1915 INIT_WORK(&rdev->del_work, md_delayed_delete); 1916 kobject_get(&rdev->kobj); 1917 schedule_work(&rdev->del_work); 1918 } 1919 1920 /* 1921 * prevent the device from being mounted, repartitioned or 1922 * otherwise reused by a RAID array (or any other kernel 1923 * subsystem), by bd_claiming the device. 1924 */ 1925 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) 1926 { 1927 int err = 0; 1928 struct block_device *bdev; 1929 char b[BDEVNAME_SIZE]; 1930 1931 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); 1932 if (IS_ERR(bdev)) { 1933 printk(KERN_ERR "md: could not open %s.\n", 1934 __bdevname(dev, b)); 1935 return PTR_ERR(bdev); 1936 } 1937 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); 1938 if (err) { 1939 printk(KERN_ERR "md: could not bd_claim %s.\n", 1940 bdevname(bdev, b)); 1941 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1942 return err; 1943 } 1944 if (!shared) 1945 set_bit(AllReserved, &rdev->flags); 1946 rdev->bdev = bdev; 1947 return err; 1948 } 1949 1950 static void unlock_rdev(mdk_rdev_t *rdev) 1951 { 1952 struct block_device *bdev = rdev->bdev; 1953 rdev->bdev = NULL; 1954 if (!bdev) 1955 MD_BUG(); 1956 bd_release(bdev); 1957 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1958 } 1959 1960 void md_autodetect_dev(dev_t dev); 1961 1962 static void export_rdev(mdk_rdev_t * rdev) 1963 { 1964 char b[BDEVNAME_SIZE]; 1965 printk(KERN_INFO "md: export_rdev(%s)\n", 1966 bdevname(rdev->bdev,b)); 1967 if (rdev->mddev) 1968 MD_BUG(); 1969 free_disk_sb(rdev); 1970 #ifndef MODULE 1971 if (test_bit(AutoDetected, &rdev->flags)) 1972 md_autodetect_dev(rdev->bdev->bd_dev); 1973 #endif 1974 unlock_rdev(rdev); 1975 kobject_put(&rdev->kobj); 1976 } 1977 1978 static void kick_rdev_from_array(mdk_rdev_t * rdev) 1979 { 1980 unbind_rdev_from_array(rdev); 1981 export_rdev(rdev); 1982 } 1983 1984 static void export_array(mddev_t *mddev) 1985 { 1986 mdk_rdev_t *rdev, *tmp; 1987 1988 rdev_for_each(rdev, tmp, mddev) { 1989 if (!rdev->mddev) { 1990 MD_BUG(); 1991 continue; 1992 } 1993 kick_rdev_from_array(rdev); 1994 } 1995 if (!list_empty(&mddev->disks)) 1996 MD_BUG(); 1997 mddev->raid_disks = 0; 1998 mddev->major_version = 0; 1999 } 2000 2001 static void print_desc(mdp_disk_t *desc) 2002 { 2003 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, 2004 desc->major,desc->minor,desc->raid_disk,desc->state); 2005 } 2006 2007 static void print_sb_90(mdp_super_t *sb) 2008 { 2009 int i; 2010 2011 printk(KERN_INFO 2012 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", 2013 sb->major_version, sb->minor_version, sb->patch_version, 2014 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, 2015 sb->ctime); 2016 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", 2017 sb->level, sb->size, sb->nr_disks, sb->raid_disks, 2018 sb->md_minor, sb->layout, sb->chunk_size); 2019 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" 2020 " FD:%d SD:%d CSUM:%08x E:%08lx\n", 2021 sb->utime, sb->state, sb->active_disks, sb->working_disks, 2022 sb->failed_disks, sb->spare_disks, 2023 sb->sb_csum, (unsigned long)sb->events_lo); 2024 2025 printk(KERN_INFO); 2026 for (i = 0; i < MD_SB_DISKS; i++) { 2027 mdp_disk_t *desc; 2028 2029 desc = sb->disks + i; 2030 if (desc->number || desc->major || desc->minor || 2031 desc->raid_disk || (desc->state && (desc->state != 4))) { 2032 printk(" D %2d: ", i); 2033 print_desc(desc); 2034 } 2035 } 2036 printk(KERN_INFO "md: THIS: "); 2037 print_desc(&sb->this_disk); 2038 } 2039 2040 static void print_sb_1(struct mdp_superblock_1 *sb) 2041 { 2042 __u8 *uuid; 2043 2044 uuid = sb->set_uuid; 2045 printk(KERN_INFO 2046 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" 2047 "md: Name: \"%s\" CT:%llu\n", 2048 le32_to_cpu(sb->major_version), 2049 le32_to_cpu(sb->feature_map), 2050 uuid, 2051 sb->set_name, 2052 (unsigned long long)le64_to_cpu(sb->ctime) 2053 & MD_SUPERBLOCK_1_TIME_SEC_MASK); 2054 2055 uuid = sb->device_uuid; 2056 printk(KERN_INFO 2057 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" 2058 " RO:%llu\n" 2059 "md: Dev:%08x UUID: %pU\n" 2060 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" 2061 "md: (MaxDev:%u) \n", 2062 le32_to_cpu(sb->level), 2063 (unsigned long long)le64_to_cpu(sb->size), 2064 le32_to_cpu(sb->raid_disks), 2065 le32_to_cpu(sb->layout), 2066 le32_to_cpu(sb->chunksize), 2067 (unsigned long long)le64_to_cpu(sb->data_offset), 2068 (unsigned long long)le64_to_cpu(sb->data_size), 2069 (unsigned long long)le64_to_cpu(sb->super_offset), 2070 (unsigned long long)le64_to_cpu(sb->recovery_offset), 2071 le32_to_cpu(sb->dev_number), 2072 uuid, 2073 sb->devflags, 2074 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, 2075 (unsigned long long)le64_to_cpu(sb->events), 2076 (unsigned long long)le64_to_cpu(sb->resync_offset), 2077 le32_to_cpu(sb->sb_csum), 2078 le32_to_cpu(sb->max_dev) 2079 ); 2080 } 2081 2082 static void print_rdev(mdk_rdev_t *rdev, int major_version) 2083 { 2084 char b[BDEVNAME_SIZE]; 2085 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", 2086 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, 2087 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), 2088 rdev->desc_nr); 2089 if (rdev->sb_loaded) { 2090 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); 2091 switch (major_version) { 2092 case 0: 2093 print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); 2094 break; 2095 case 1: 2096 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); 2097 break; 2098 } 2099 } else 2100 printk(KERN_INFO "md: no rdev superblock!\n"); 2101 } 2102 2103 static void md_print_devices(void) 2104 { 2105 struct list_head *tmp; 2106 mdk_rdev_t *rdev; 2107 mddev_t *mddev; 2108 char b[BDEVNAME_SIZE]; 2109 2110 printk("\n"); 2111 printk("md: **********************************\n"); 2112 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); 2113 printk("md: **********************************\n"); 2114 for_each_mddev(mddev, tmp) { 2115 2116 if (mddev->bitmap) 2117 bitmap_print_sb(mddev->bitmap); 2118 else 2119 printk("%s: ", mdname(mddev)); 2120 list_for_each_entry(rdev, &mddev->disks, same_set) 2121 printk("<%s>", bdevname(rdev->bdev,b)); 2122 printk("\n"); 2123 2124 list_for_each_entry(rdev, &mddev->disks, same_set) 2125 print_rdev(rdev, mddev->major_version); 2126 } 2127 printk("md: **********************************\n"); 2128 printk("\n"); 2129 } 2130 2131 2132 static void sync_sbs(mddev_t * mddev, int nospares) 2133 { 2134 /* Update each superblock (in-memory image), but 2135 * if we are allowed to, skip spares which already 2136 * have the right event counter, or have one earlier 2137 * (which would mean they aren't being marked as dirty 2138 * with the rest of the array) 2139 */ 2140 mdk_rdev_t *rdev; 2141 list_for_each_entry(rdev, &mddev->disks, same_set) { 2142 if (rdev->sb_events == mddev->events || 2143 (nospares && 2144 rdev->raid_disk < 0 && 2145 rdev->sb_events+1 == mddev->events)) { 2146 /* Don't update this superblock */ 2147 rdev->sb_loaded = 2; 2148 } else { 2149 super_types[mddev->major_version]. 2150 sync_super(mddev, rdev); 2151 rdev->sb_loaded = 1; 2152 } 2153 } 2154 } 2155 2156 static void md_update_sb(mddev_t * mddev, int force_change) 2157 { 2158 mdk_rdev_t *rdev; 2159 int sync_req; 2160 int nospares = 0; 2161 2162 repeat: 2163 /* First make sure individual recovery_offsets are correct */ 2164 list_for_each_entry(rdev, &mddev->disks, same_set) { 2165 if (rdev->raid_disk >= 0 && 2166 mddev->delta_disks >= 0 && 2167 !test_bit(In_sync, &rdev->flags) && 2168 mddev->curr_resync_completed > rdev->recovery_offset) 2169 rdev->recovery_offset = mddev->curr_resync_completed; 2170 2171 } 2172 if (!mddev->persistent) { 2173 clear_bit(MD_CHANGE_CLEAN, &mddev->flags); 2174 clear_bit(MD_CHANGE_DEVS, &mddev->flags); 2175 wake_up(&mddev->sb_wait); 2176 return; 2177 } 2178 2179 spin_lock_irq(&mddev->write_lock); 2180 2181 mddev->utime = get_seconds(); 2182 2183 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) 2184 force_change = 1; 2185 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) 2186 /* just a clean<-> dirty transition, possibly leave spares alone, 2187 * though if events isn't the right even/odd, we will have to do 2188 * spares after all 2189 */ 2190 nospares = 1; 2191 if (force_change) 2192 nospares = 0; 2193 if (mddev->degraded) 2194 /* If the array is degraded, then skipping spares is both 2195 * dangerous and fairly pointless. 2196 * Dangerous because a device that was removed from the array 2197 * might have a event_count that still looks up-to-date, 2198 * so it can be re-added without a resync. 2199 * Pointless because if there are any spares to skip, 2200 * then a recovery will happen and soon that array won't 2201 * be degraded any more and the spare can go back to sleep then. 2202 */ 2203 nospares = 0; 2204 2205 sync_req = mddev->in_sync; 2206 2207 /* If this is just a dirty<->clean transition, and the array is clean 2208 * and 'events' is odd, we can roll back to the previous clean state */ 2209 if (nospares 2210 && (mddev->in_sync && mddev->recovery_cp == MaxSector) 2211 && mddev->can_decrease_events 2212 && mddev->events != 1) { 2213 mddev->events--; 2214 mddev->can_decrease_events = 0; 2215 } else { 2216 /* otherwise we have to go forward and ... */ 2217 mddev->events ++; 2218 mddev->can_decrease_events = nospares; 2219 } 2220 2221 if (!mddev->events) { 2222 /* 2223 * oops, this 64-bit counter should never wrap. 2224 * Either we are in around ~1 trillion A.C., assuming 2225 * 1 reboot per second, or we have a bug: 2226 */ 2227 MD_BUG(); 2228 mddev->events --; 2229 } 2230 sync_sbs(mddev, nospares); 2231 spin_unlock_irq(&mddev->write_lock); 2232 2233 dprintk(KERN_INFO 2234 "md: updating %s RAID superblock on device (in sync %d)\n", 2235 mdname(mddev),mddev->in_sync); 2236 2237 bitmap_update_sb(mddev->bitmap); 2238 list_for_each_entry(rdev, &mddev->disks, same_set) { 2239 char b[BDEVNAME_SIZE]; 2240 dprintk(KERN_INFO "md: "); 2241 if (rdev->sb_loaded != 1) 2242 continue; /* no noise on spare devices */ 2243 if (test_bit(Faulty, &rdev->flags)) 2244 dprintk("(skipping faulty "); 2245 2246 dprintk("%s ", bdevname(rdev->bdev,b)); 2247 if (!test_bit(Faulty, &rdev->flags)) { 2248 md_super_write(mddev,rdev, 2249 rdev->sb_start, rdev->sb_size, 2250 rdev->sb_page); 2251 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", 2252 bdevname(rdev->bdev,b), 2253 (unsigned long long)rdev->sb_start); 2254 rdev->sb_events = mddev->events; 2255 2256 } else 2257 dprintk(")\n"); 2258 if (mddev->level == LEVEL_MULTIPATH) 2259 /* only need to write one superblock... */ 2260 break; 2261 } 2262 md_super_wait(mddev); 2263 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ 2264 2265 spin_lock_irq(&mddev->write_lock); 2266 if (mddev->in_sync != sync_req || 2267 test_bit(MD_CHANGE_DEVS, &mddev->flags)) { 2268 /* have to write it out again */ 2269 spin_unlock_irq(&mddev->write_lock); 2270 goto repeat; 2271 } 2272 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 2273 spin_unlock_irq(&mddev->write_lock); 2274 wake_up(&mddev->sb_wait); 2275 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 2276 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 2277 2278 } 2279 2280 /* words written to sysfs files may, or may not, be \n terminated. 2281 * We want to accept with case. For this we use cmd_match. 2282 */ 2283 static int cmd_match(const char *cmd, const char *str) 2284 { 2285 /* See if cmd, written into a sysfs file, matches 2286 * str. They must either be the same, or cmd can 2287 * have a trailing newline 2288 */ 2289 while (*cmd && *str && *cmd == *str) { 2290 cmd++; 2291 str++; 2292 } 2293 if (*cmd == '\n') 2294 cmd++; 2295 if (*str || *cmd) 2296 return 0; 2297 return 1; 2298 } 2299 2300 struct rdev_sysfs_entry { 2301 struct attribute attr; 2302 ssize_t (*show)(mdk_rdev_t *, char *); 2303 ssize_t (*store)(mdk_rdev_t *, const char *, size_t); 2304 }; 2305 2306 static ssize_t 2307 state_show(mdk_rdev_t *rdev, char *page) 2308 { 2309 char *sep = ""; 2310 size_t len = 0; 2311 2312 if (test_bit(Faulty, &rdev->flags)) { 2313 len+= sprintf(page+len, "%sfaulty",sep); 2314 sep = ","; 2315 } 2316 if (test_bit(In_sync, &rdev->flags)) { 2317 len += sprintf(page+len, "%sin_sync",sep); 2318 sep = ","; 2319 } 2320 if (test_bit(WriteMostly, &rdev->flags)) { 2321 len += sprintf(page+len, "%swrite_mostly",sep); 2322 sep = ","; 2323 } 2324 if (test_bit(Blocked, &rdev->flags)) { 2325 len += sprintf(page+len, "%sblocked", sep); 2326 sep = ","; 2327 } 2328 if (!test_bit(Faulty, &rdev->flags) && 2329 !test_bit(In_sync, &rdev->flags)) { 2330 len += sprintf(page+len, "%sspare", sep); 2331 sep = ","; 2332 } 2333 return len+sprintf(page+len, "\n"); 2334 } 2335 2336 static ssize_t 2337 state_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2338 { 2339 /* can write 2340 * faulty - simulates and error 2341 * remove - disconnects the device 2342 * writemostly - sets write_mostly 2343 * -writemostly - clears write_mostly 2344 * blocked - sets the Blocked flag 2345 * -blocked - clears the Blocked flag 2346 * insync - sets Insync providing device isn't active 2347 */ 2348 int err = -EINVAL; 2349 if (cmd_match(buf, "faulty") && rdev->mddev->pers) { 2350 md_error(rdev->mddev, rdev); 2351 err = 0; 2352 } else if (cmd_match(buf, "remove")) { 2353 if (rdev->raid_disk >= 0) 2354 err = -EBUSY; 2355 else { 2356 mddev_t *mddev = rdev->mddev; 2357 kick_rdev_from_array(rdev); 2358 if (mddev->pers) 2359 md_update_sb(mddev, 1); 2360 md_new_event(mddev); 2361 err = 0; 2362 } 2363 } else if (cmd_match(buf, "writemostly")) { 2364 set_bit(WriteMostly, &rdev->flags); 2365 err = 0; 2366 } else if (cmd_match(buf, "-writemostly")) { 2367 clear_bit(WriteMostly, &rdev->flags); 2368 err = 0; 2369 } else if (cmd_match(buf, "blocked")) { 2370 set_bit(Blocked, &rdev->flags); 2371 err = 0; 2372 } else if (cmd_match(buf, "-blocked")) { 2373 clear_bit(Blocked, &rdev->flags); 2374 wake_up(&rdev->blocked_wait); 2375 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2376 md_wakeup_thread(rdev->mddev->thread); 2377 2378 err = 0; 2379 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { 2380 set_bit(In_sync, &rdev->flags); 2381 err = 0; 2382 } 2383 if (!err) 2384 sysfs_notify_dirent_safe(rdev->sysfs_state); 2385 return err ? err : len; 2386 } 2387 static struct rdev_sysfs_entry rdev_state = 2388 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); 2389 2390 static ssize_t 2391 errors_show(mdk_rdev_t *rdev, char *page) 2392 { 2393 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); 2394 } 2395 2396 static ssize_t 2397 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2398 { 2399 char *e; 2400 unsigned long n = simple_strtoul(buf, &e, 10); 2401 if (*buf && (*e == 0 || *e == '\n')) { 2402 atomic_set(&rdev->corrected_errors, n); 2403 return len; 2404 } 2405 return -EINVAL; 2406 } 2407 static struct rdev_sysfs_entry rdev_errors = 2408 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); 2409 2410 static ssize_t 2411 slot_show(mdk_rdev_t *rdev, char *page) 2412 { 2413 if (rdev->raid_disk < 0) 2414 return sprintf(page, "none\n"); 2415 else 2416 return sprintf(page, "%d\n", rdev->raid_disk); 2417 } 2418 2419 static ssize_t 2420 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2421 { 2422 char *e; 2423 int err; 2424 char nm[20]; 2425 int slot = simple_strtoul(buf, &e, 10); 2426 if (strncmp(buf, "none", 4)==0) 2427 slot = -1; 2428 else if (e==buf || (*e && *e!= '\n')) 2429 return -EINVAL; 2430 if (rdev->mddev->pers && slot == -1) { 2431 /* Setting 'slot' on an active array requires also 2432 * updating the 'rd%d' link, and communicating 2433 * with the personality with ->hot_*_disk. 2434 * For now we only support removing 2435 * failed/spare devices. This normally happens automatically, 2436 * but not when the metadata is externally managed. 2437 */ 2438 if (rdev->raid_disk == -1) 2439 return -EEXIST; 2440 /* personality does all needed checks */ 2441 if (rdev->mddev->pers->hot_add_disk == NULL) 2442 return -EINVAL; 2443 err = rdev->mddev->pers-> 2444 hot_remove_disk(rdev->mddev, rdev->raid_disk); 2445 if (err) 2446 return err; 2447 sprintf(nm, "rd%d", rdev->raid_disk); 2448 sysfs_remove_link(&rdev->mddev->kobj, nm); 2449 rdev->raid_disk = -1; 2450 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); 2451 md_wakeup_thread(rdev->mddev->thread); 2452 } else if (rdev->mddev->pers) { 2453 mdk_rdev_t *rdev2; 2454 /* Activating a spare .. or possibly reactivating 2455 * if we ever get bitmaps working here. 2456 */ 2457 2458 if (rdev->raid_disk != -1) 2459 return -EBUSY; 2460 2461 if (rdev->mddev->pers->hot_add_disk == NULL) 2462 return -EINVAL; 2463 2464 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) 2465 if (rdev2->raid_disk == slot) 2466 return -EEXIST; 2467 2468 rdev->raid_disk = slot; 2469 if (test_bit(In_sync, &rdev->flags)) 2470 rdev->saved_raid_disk = slot; 2471 else 2472 rdev->saved_raid_disk = -1; 2473 err = rdev->mddev->pers-> 2474 hot_add_disk(rdev->mddev, rdev); 2475 if (err) { 2476 rdev->raid_disk = -1; 2477 return err; 2478 } else 2479 sysfs_notify_dirent_safe(rdev->sysfs_state); 2480 sprintf(nm, "rd%d", rdev->raid_disk); 2481 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) 2482 /* failure here is OK */; 2483 /* don't wakeup anyone, leave that to userspace. */ 2484 } else { 2485 if (slot >= rdev->mddev->raid_disks) 2486 return -ENOSPC; 2487 rdev->raid_disk = slot; 2488 /* assume it is working */ 2489 clear_bit(Faulty, &rdev->flags); 2490 clear_bit(WriteMostly, &rdev->flags); 2491 set_bit(In_sync, &rdev->flags); 2492 sysfs_notify_dirent_safe(rdev->sysfs_state); 2493 } 2494 return len; 2495 } 2496 2497 2498 static struct rdev_sysfs_entry rdev_slot = 2499 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); 2500 2501 static ssize_t 2502 offset_show(mdk_rdev_t *rdev, char *page) 2503 { 2504 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); 2505 } 2506 2507 static ssize_t 2508 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2509 { 2510 char *e; 2511 unsigned long long offset = simple_strtoull(buf, &e, 10); 2512 if (e==buf || (*e && *e != '\n')) 2513 return -EINVAL; 2514 if (rdev->mddev->pers && rdev->raid_disk >= 0) 2515 return -EBUSY; 2516 if (rdev->sectors && rdev->mddev->external) 2517 /* Must set offset before size, so overlap checks 2518 * can be sane */ 2519 return -EBUSY; 2520 rdev->data_offset = offset; 2521 return len; 2522 } 2523 2524 static struct rdev_sysfs_entry rdev_offset = 2525 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); 2526 2527 static ssize_t 2528 rdev_size_show(mdk_rdev_t *rdev, char *page) 2529 { 2530 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); 2531 } 2532 2533 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) 2534 { 2535 /* check if two start/length pairs overlap */ 2536 if (s1+l1 <= s2) 2537 return 0; 2538 if (s2+l2 <= s1) 2539 return 0; 2540 return 1; 2541 } 2542 2543 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) 2544 { 2545 unsigned long long blocks; 2546 sector_t new; 2547 2548 if (strict_strtoull(buf, 10, &blocks) < 0) 2549 return -EINVAL; 2550 2551 if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) 2552 return -EINVAL; /* sector conversion overflow */ 2553 2554 new = blocks * 2; 2555 if (new != blocks * 2) 2556 return -EINVAL; /* unsigned long long to sector_t overflow */ 2557 2558 *sectors = new; 2559 return 0; 2560 } 2561 2562 static ssize_t 2563 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2564 { 2565 mddev_t *my_mddev = rdev->mddev; 2566 sector_t oldsectors = rdev->sectors; 2567 sector_t sectors; 2568 2569 if (strict_blocks_to_sectors(buf, §ors) < 0) 2570 return -EINVAL; 2571 if (my_mddev->pers && rdev->raid_disk >= 0) { 2572 if (my_mddev->persistent) { 2573 sectors = super_types[my_mddev->major_version]. 2574 rdev_size_change(rdev, sectors); 2575 if (!sectors) 2576 return -EBUSY; 2577 } else if (!sectors) 2578 sectors = (rdev->bdev->bd_inode->i_size >> 9) - 2579 rdev->data_offset; 2580 } 2581 if (sectors < my_mddev->dev_sectors) 2582 return -EINVAL; /* component must fit device */ 2583 2584 rdev->sectors = sectors; 2585 if (sectors > oldsectors && my_mddev->external) { 2586 /* need to check that all other rdevs with the same ->bdev 2587 * do not overlap. We need to unlock the mddev to avoid 2588 * a deadlock. We have already changed rdev->sectors, and if 2589 * we have to change it back, we will have the lock again. 2590 */ 2591 mddev_t *mddev; 2592 int overlap = 0; 2593 struct list_head *tmp; 2594 2595 mddev_unlock(my_mddev); 2596 for_each_mddev(mddev, tmp) { 2597 mdk_rdev_t *rdev2; 2598 2599 mddev_lock(mddev); 2600 list_for_each_entry(rdev2, &mddev->disks, same_set) 2601 if (test_bit(AllReserved, &rdev2->flags) || 2602 (rdev->bdev == rdev2->bdev && 2603 rdev != rdev2 && 2604 overlaps(rdev->data_offset, rdev->sectors, 2605 rdev2->data_offset, 2606 rdev2->sectors))) { 2607 overlap = 1; 2608 break; 2609 } 2610 mddev_unlock(mddev); 2611 if (overlap) { 2612 mddev_put(mddev); 2613 break; 2614 } 2615 } 2616 mddev_lock(my_mddev); 2617 if (overlap) { 2618 /* Someone else could have slipped in a size 2619 * change here, but doing so is just silly. 2620 * We put oldsectors back because we *know* it is 2621 * safe, and trust userspace not to race with 2622 * itself 2623 */ 2624 rdev->sectors = oldsectors; 2625 return -EBUSY; 2626 } 2627 } 2628 return len; 2629 } 2630 2631 static struct rdev_sysfs_entry rdev_size = 2632 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); 2633 2634 2635 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) 2636 { 2637 unsigned long long recovery_start = rdev->recovery_offset; 2638 2639 if (test_bit(In_sync, &rdev->flags) || 2640 recovery_start == MaxSector) 2641 return sprintf(page, "none\n"); 2642 2643 return sprintf(page, "%llu\n", recovery_start); 2644 } 2645 2646 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) 2647 { 2648 unsigned long long recovery_start; 2649 2650 if (cmd_match(buf, "none")) 2651 recovery_start = MaxSector; 2652 else if (strict_strtoull(buf, 10, &recovery_start)) 2653 return -EINVAL; 2654 2655 if (rdev->mddev->pers && 2656 rdev->raid_disk >= 0) 2657 return -EBUSY; 2658 2659 rdev->recovery_offset = recovery_start; 2660 if (recovery_start == MaxSector) 2661 set_bit(In_sync, &rdev->flags); 2662 else 2663 clear_bit(In_sync, &rdev->flags); 2664 return len; 2665 } 2666 2667 static struct rdev_sysfs_entry rdev_recovery_start = 2668 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); 2669 2670 static struct attribute *rdev_default_attrs[] = { 2671 &rdev_state.attr, 2672 &rdev_errors.attr, 2673 &rdev_slot.attr, 2674 &rdev_offset.attr, 2675 &rdev_size.attr, 2676 &rdev_recovery_start.attr, 2677 NULL, 2678 }; 2679 static ssize_t 2680 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 2681 { 2682 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2683 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2684 mddev_t *mddev = rdev->mddev; 2685 ssize_t rv; 2686 2687 if (!entry->show) 2688 return -EIO; 2689 2690 rv = mddev ? mddev_lock(mddev) : -EBUSY; 2691 if (!rv) { 2692 if (rdev->mddev == NULL) 2693 rv = -EBUSY; 2694 else 2695 rv = entry->show(rdev, page); 2696 mddev_unlock(mddev); 2697 } 2698 return rv; 2699 } 2700 2701 static ssize_t 2702 rdev_attr_store(struct kobject *kobj, struct attribute *attr, 2703 const char *page, size_t length) 2704 { 2705 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); 2706 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); 2707 ssize_t rv; 2708 mddev_t *mddev = rdev->mddev; 2709 2710 if (!entry->store) 2711 return -EIO; 2712 if (!capable(CAP_SYS_ADMIN)) 2713 return -EACCES; 2714 rv = mddev ? mddev_lock(mddev): -EBUSY; 2715 if (!rv) { 2716 if (rdev->mddev == NULL) 2717 rv = -EBUSY; 2718 else 2719 rv = entry->store(rdev, page, length); 2720 mddev_unlock(mddev); 2721 } 2722 return rv; 2723 } 2724 2725 static void rdev_free(struct kobject *ko) 2726 { 2727 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); 2728 kfree(rdev); 2729 } 2730 static const struct sysfs_ops rdev_sysfs_ops = { 2731 .show = rdev_attr_show, 2732 .store = rdev_attr_store, 2733 }; 2734 static struct kobj_type rdev_ktype = { 2735 .release = rdev_free, 2736 .sysfs_ops = &rdev_sysfs_ops, 2737 .default_attrs = rdev_default_attrs, 2738 }; 2739 2740 void md_rdev_init(mdk_rdev_t *rdev) 2741 { 2742 rdev->desc_nr = -1; 2743 rdev->saved_raid_disk = -1; 2744 rdev->raid_disk = -1; 2745 rdev->flags = 0; 2746 rdev->data_offset = 0; 2747 rdev->sb_events = 0; 2748 rdev->last_read_error.tv_sec = 0; 2749 rdev->last_read_error.tv_nsec = 0; 2750 atomic_set(&rdev->nr_pending, 0); 2751 atomic_set(&rdev->read_errors, 0); 2752 atomic_set(&rdev->corrected_errors, 0); 2753 2754 INIT_LIST_HEAD(&rdev->same_set); 2755 init_waitqueue_head(&rdev->blocked_wait); 2756 } 2757 EXPORT_SYMBOL_GPL(md_rdev_init); 2758 /* 2759 * Import a device. If 'super_format' >= 0, then sanity check the superblock 2760 * 2761 * mark the device faulty if: 2762 * 2763 * - the device is nonexistent (zero size) 2764 * - the device has no valid superblock 2765 * 2766 * a faulty rdev _never_ has rdev->sb set. 2767 */ 2768 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) 2769 { 2770 char b[BDEVNAME_SIZE]; 2771 int err; 2772 mdk_rdev_t *rdev; 2773 sector_t size; 2774 2775 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); 2776 if (!rdev) { 2777 printk(KERN_ERR "md: could not alloc mem for new device!\n"); 2778 return ERR_PTR(-ENOMEM); 2779 } 2780 2781 md_rdev_init(rdev); 2782 if ((err = alloc_disk_sb(rdev))) 2783 goto abort_free; 2784 2785 err = lock_rdev(rdev, newdev, super_format == -2); 2786 if (err) 2787 goto abort_free; 2788 2789 kobject_init(&rdev->kobj, &rdev_ktype); 2790 2791 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; 2792 if (!size) { 2793 printk(KERN_WARNING 2794 "md: %s has zero or unknown size, marking faulty!\n", 2795 bdevname(rdev->bdev,b)); 2796 err = -EINVAL; 2797 goto abort_free; 2798 } 2799 2800 if (super_format >= 0) { 2801 err = super_types[super_format]. 2802 load_super(rdev, NULL, super_minor); 2803 if (err == -EINVAL) { 2804 printk(KERN_WARNING 2805 "md: %s does not have a valid v%d.%d " 2806 "superblock, not importing!\n", 2807 bdevname(rdev->bdev,b), 2808 super_format, super_minor); 2809 goto abort_free; 2810 } 2811 if (err < 0) { 2812 printk(KERN_WARNING 2813 "md: could not read %s's sb, not importing!\n", 2814 bdevname(rdev->bdev,b)); 2815 goto abort_free; 2816 } 2817 } 2818 2819 return rdev; 2820 2821 abort_free: 2822 if (rdev->sb_page) { 2823 if (rdev->bdev) 2824 unlock_rdev(rdev); 2825 free_disk_sb(rdev); 2826 } 2827 kfree(rdev); 2828 return ERR_PTR(err); 2829 } 2830 2831 /* 2832 * Check a full RAID array for plausibility 2833 */ 2834 2835 2836 static void analyze_sbs(mddev_t * mddev) 2837 { 2838 int i; 2839 mdk_rdev_t *rdev, *freshest, *tmp; 2840 char b[BDEVNAME_SIZE]; 2841 2842 freshest = NULL; 2843 rdev_for_each(rdev, tmp, mddev) 2844 switch (super_types[mddev->major_version]. 2845 load_super(rdev, freshest, mddev->minor_version)) { 2846 case 1: 2847 freshest = rdev; 2848 break; 2849 case 0: 2850 break; 2851 default: 2852 printk( KERN_ERR \ 2853 "md: fatal superblock inconsistency in %s" 2854 " -- removing from array\n", 2855 bdevname(rdev->bdev,b)); 2856 kick_rdev_from_array(rdev); 2857 } 2858 2859 2860 super_types[mddev->major_version]. 2861 validate_super(mddev, freshest); 2862 2863 i = 0; 2864 rdev_for_each(rdev, tmp, mddev) { 2865 if (mddev->max_disks && 2866 (rdev->desc_nr >= mddev->max_disks || 2867 i > mddev->max_disks)) { 2868 printk(KERN_WARNING 2869 "md: %s: %s: only %d devices permitted\n", 2870 mdname(mddev), bdevname(rdev->bdev, b), 2871 mddev->max_disks); 2872 kick_rdev_from_array(rdev); 2873 continue; 2874 } 2875 if (rdev != freshest) 2876 if (super_types[mddev->major_version]. 2877 validate_super(mddev, rdev)) { 2878 printk(KERN_WARNING "md: kicking non-fresh %s" 2879 " from array!\n", 2880 bdevname(rdev->bdev,b)); 2881 kick_rdev_from_array(rdev); 2882 continue; 2883 } 2884 if (mddev->level == LEVEL_MULTIPATH) { 2885 rdev->desc_nr = i++; 2886 rdev->raid_disk = rdev->desc_nr; 2887 set_bit(In_sync, &rdev->flags); 2888 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { 2889 rdev->raid_disk = -1; 2890 clear_bit(In_sync, &rdev->flags); 2891 } 2892 } 2893 } 2894 2895 /* Read a fixed-point number. 2896 * Numbers in sysfs attributes should be in "standard" units where 2897 * possible, so time should be in seconds. 2898 * However we internally use a a much smaller unit such as 2899 * milliseconds or jiffies. 2900 * This function takes a decimal number with a possible fractional 2901 * component, and produces an integer which is the result of 2902 * multiplying that number by 10^'scale'. 2903 * all without any floating-point arithmetic. 2904 */ 2905 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) 2906 { 2907 unsigned long result = 0; 2908 long decimals = -1; 2909 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { 2910 if (*cp == '.') 2911 decimals = 0; 2912 else if (decimals < scale) { 2913 unsigned int value; 2914 value = *cp - '0'; 2915 result = result * 10 + value; 2916 if (decimals >= 0) 2917 decimals++; 2918 } 2919 cp++; 2920 } 2921 if (*cp == '\n') 2922 cp++; 2923 if (*cp) 2924 return -EINVAL; 2925 if (decimals < 0) 2926 decimals = 0; 2927 while (decimals < scale) { 2928 result *= 10; 2929 decimals ++; 2930 } 2931 *res = result; 2932 return 0; 2933 } 2934 2935 2936 static void md_safemode_timeout(unsigned long data); 2937 2938 static ssize_t 2939 safe_delay_show(mddev_t *mddev, char *page) 2940 { 2941 int msec = (mddev->safemode_delay*1000)/HZ; 2942 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); 2943 } 2944 static ssize_t 2945 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) 2946 { 2947 unsigned long msec; 2948 2949 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) 2950 return -EINVAL; 2951 if (msec == 0) 2952 mddev->safemode_delay = 0; 2953 else { 2954 unsigned long old_delay = mddev->safemode_delay; 2955 mddev->safemode_delay = (msec*HZ)/1000; 2956 if (mddev->safemode_delay == 0) 2957 mddev->safemode_delay = 1; 2958 if (mddev->safemode_delay < old_delay) 2959 md_safemode_timeout((unsigned long)mddev); 2960 } 2961 return len; 2962 } 2963 static struct md_sysfs_entry md_safe_delay = 2964 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); 2965 2966 static ssize_t 2967 level_show(mddev_t *mddev, char *page) 2968 { 2969 struct mdk_personality *p = mddev->pers; 2970 if (p) 2971 return sprintf(page, "%s\n", p->name); 2972 else if (mddev->clevel[0]) 2973 return sprintf(page, "%s\n", mddev->clevel); 2974 else if (mddev->level != LEVEL_NONE) 2975 return sprintf(page, "%d\n", mddev->level); 2976 else 2977 return 0; 2978 } 2979 2980 static ssize_t 2981 level_store(mddev_t *mddev, const char *buf, size_t len) 2982 { 2983 char clevel[16]; 2984 ssize_t rv = len; 2985 struct mdk_personality *pers; 2986 long level; 2987 void *priv; 2988 mdk_rdev_t *rdev; 2989 2990 if (mddev->pers == NULL) { 2991 if (len == 0) 2992 return 0; 2993 if (len >= sizeof(mddev->clevel)) 2994 return -ENOSPC; 2995 strncpy(mddev->clevel, buf, len); 2996 if (mddev->clevel[len-1] == '\n') 2997 len--; 2998 mddev->clevel[len] = 0; 2999 mddev->level = LEVEL_NONE; 3000 return rv; 3001 } 3002 3003 /* request to change the personality. Need to ensure: 3004 * - array is not engaged in resync/recovery/reshape 3005 * - old personality can be suspended 3006 * - new personality will access other array. 3007 */ 3008 3009 if (mddev->sync_thread || 3010 mddev->reshape_position != MaxSector || 3011 mddev->sysfs_active) 3012 return -EBUSY; 3013 3014 if (!mddev->pers->quiesce) { 3015 printk(KERN_WARNING "md: %s: %s does not support online personality change\n", 3016 mdname(mddev), mddev->pers->name); 3017 return -EINVAL; 3018 } 3019 3020 /* Now find the new personality */ 3021 if (len == 0 || len >= sizeof(clevel)) 3022 return -EINVAL; 3023 strncpy(clevel, buf, len); 3024 if (clevel[len-1] == '\n') 3025 len--; 3026 clevel[len] = 0; 3027 if (strict_strtol(clevel, 10, &level)) 3028 level = LEVEL_NONE; 3029 3030 if (request_module("md-%s", clevel) != 0) 3031 request_module("md-level-%s", clevel); 3032 spin_lock(&pers_lock); 3033 pers = find_pers(level, clevel); 3034 if (!pers || !try_module_get(pers->owner)) { 3035 spin_unlock(&pers_lock); 3036 printk(KERN_WARNING "md: personality %s not loaded\n", clevel); 3037 return -EINVAL; 3038 } 3039 spin_unlock(&pers_lock); 3040 3041 if (pers == mddev->pers) { 3042 /* Nothing to do! */ 3043 module_put(pers->owner); 3044 return rv; 3045 } 3046 if (!pers->takeover) { 3047 module_put(pers->owner); 3048 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", 3049 mdname(mddev), clevel); 3050 return -EINVAL; 3051 } 3052 3053 list_for_each_entry(rdev, &mddev->disks, same_set) 3054 rdev->new_raid_disk = rdev->raid_disk; 3055 3056 /* ->takeover must set new_* and/or delta_disks 3057 * if it succeeds, and may set them when it fails. 3058 */ 3059 priv = pers->takeover(mddev); 3060 if (IS_ERR(priv)) { 3061 mddev->new_level = mddev->level; 3062 mddev->new_layout = mddev->layout; 3063 mddev->new_chunk_sectors = mddev->chunk_sectors; 3064 mddev->raid_disks -= mddev->delta_disks; 3065 mddev->delta_disks = 0; 3066 module_put(pers->owner); 3067 printk(KERN_WARNING "md: %s: %s would not accept array\n", 3068 mdname(mddev), clevel); 3069 return PTR_ERR(priv); 3070 } 3071 3072 /* Looks like we have a winner */ 3073 mddev_suspend(mddev); 3074 mddev->pers->stop(mddev); 3075 3076 if (mddev->pers->sync_request == NULL && 3077 pers->sync_request != NULL) { 3078 /* need to add the md_redundancy_group */ 3079 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 3080 printk(KERN_WARNING 3081 "md: cannot register extra attributes for %s\n", 3082 mdname(mddev)); 3083 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); 3084 } 3085 if (mddev->pers->sync_request != NULL && 3086 pers->sync_request == NULL) { 3087 /* need to remove the md_redundancy_group */ 3088 if (mddev->to_remove == NULL) 3089 mddev->to_remove = &md_redundancy_group; 3090 } 3091 3092 if (mddev->pers->sync_request == NULL && 3093 mddev->external) { 3094 /* We are converting from a no-redundancy array 3095 * to a redundancy array and metadata is managed 3096 * externally so we need to be sure that writes 3097 * won't block due to a need to transition 3098 * clean->dirty 3099 * until external management is started. 3100 */ 3101 mddev->in_sync = 0; 3102 mddev->safemode_delay = 0; 3103 mddev->safemode = 0; 3104 } 3105 3106 list_for_each_entry(rdev, &mddev->disks, same_set) { 3107 char nm[20]; 3108 if (rdev->raid_disk < 0) 3109 continue; 3110 if (rdev->new_raid_disk > mddev->raid_disks) 3111 rdev->new_raid_disk = -1; 3112 if (rdev->new_raid_disk == rdev->raid_disk) 3113 continue; 3114 sprintf(nm, "rd%d", rdev->raid_disk); 3115 sysfs_remove_link(&mddev->kobj, nm); 3116 } 3117 list_for_each_entry(rdev, &mddev->disks, same_set) { 3118 if (rdev->raid_disk < 0) 3119 continue; 3120 if (rdev->new_raid_disk == rdev->raid_disk) 3121 continue; 3122 rdev->raid_disk = rdev->new_raid_disk; 3123 if (rdev->raid_disk < 0) 3124 clear_bit(In_sync, &rdev->flags); 3125 else { 3126 char nm[20]; 3127 sprintf(nm, "rd%d", rdev->raid_disk); 3128 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) 3129 printk("md: cannot register %s for %s after level change\n", 3130 nm, mdname(mddev)); 3131 } 3132 } 3133 3134 module_put(mddev->pers->owner); 3135 mddev->pers = pers; 3136 mddev->private = priv; 3137 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 3138 mddev->level = mddev->new_level; 3139 mddev->layout = mddev->new_layout; 3140 mddev->chunk_sectors = mddev->new_chunk_sectors; 3141 mddev->delta_disks = 0; 3142 if (mddev->pers->sync_request == NULL) { 3143 /* this is now an array without redundancy, so 3144 * it must always be in_sync 3145 */ 3146 mddev->in_sync = 1; 3147 del_timer_sync(&mddev->safemode_timer); 3148 } 3149 pers->run(mddev); 3150 mddev_resume(mddev); 3151 set_bit(MD_CHANGE_DEVS, &mddev->flags); 3152 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3153 md_wakeup_thread(mddev->thread); 3154 sysfs_notify(&mddev->kobj, NULL, "level"); 3155 md_new_event(mddev); 3156 return rv; 3157 } 3158 3159 static struct md_sysfs_entry md_level = 3160 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); 3161 3162 3163 static ssize_t 3164 layout_show(mddev_t *mddev, char *page) 3165 { 3166 /* just a number, not meaningful for all levels */ 3167 if (mddev->reshape_position != MaxSector && 3168 mddev->layout != mddev->new_layout) 3169 return sprintf(page, "%d (%d)\n", 3170 mddev->new_layout, mddev->layout); 3171 return sprintf(page, "%d\n", mddev->layout); 3172 } 3173 3174 static ssize_t 3175 layout_store(mddev_t *mddev, const char *buf, size_t len) 3176 { 3177 char *e; 3178 unsigned long n = simple_strtoul(buf, &e, 10); 3179 3180 if (!*buf || (*e && *e != '\n')) 3181 return -EINVAL; 3182 3183 if (mddev->pers) { 3184 int err; 3185 if (mddev->pers->check_reshape == NULL) 3186 return -EBUSY; 3187 mddev->new_layout = n; 3188 err = mddev->pers->check_reshape(mddev); 3189 if (err) { 3190 mddev->new_layout = mddev->layout; 3191 return err; 3192 } 3193 } else { 3194 mddev->new_layout = n; 3195 if (mddev->reshape_position == MaxSector) 3196 mddev->layout = n; 3197 } 3198 return len; 3199 } 3200 static struct md_sysfs_entry md_layout = 3201 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); 3202 3203 3204 static ssize_t 3205 raid_disks_show(mddev_t *mddev, char *page) 3206 { 3207 if (mddev->raid_disks == 0) 3208 return 0; 3209 if (mddev->reshape_position != MaxSector && 3210 mddev->delta_disks != 0) 3211 return sprintf(page, "%d (%d)\n", mddev->raid_disks, 3212 mddev->raid_disks - mddev->delta_disks); 3213 return sprintf(page, "%d\n", mddev->raid_disks); 3214 } 3215 3216 static int update_raid_disks(mddev_t *mddev, int raid_disks); 3217 3218 static ssize_t 3219 raid_disks_store(mddev_t *mddev, const char *buf, size_t len) 3220 { 3221 char *e; 3222 int rv = 0; 3223 unsigned long n = simple_strtoul(buf, &e, 10); 3224 3225 if (!*buf || (*e && *e != '\n')) 3226 return -EINVAL; 3227 3228 if (mddev->pers) 3229 rv = update_raid_disks(mddev, n); 3230 else if (mddev->reshape_position != MaxSector) { 3231 int olddisks = mddev->raid_disks - mddev->delta_disks; 3232 mddev->delta_disks = n - olddisks; 3233 mddev->raid_disks = n; 3234 } else 3235 mddev->raid_disks = n; 3236 return rv ? rv : len; 3237 } 3238 static struct md_sysfs_entry md_raid_disks = 3239 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); 3240 3241 static ssize_t 3242 chunk_size_show(mddev_t *mddev, char *page) 3243 { 3244 if (mddev->reshape_position != MaxSector && 3245 mddev->chunk_sectors != mddev->new_chunk_sectors) 3246 return sprintf(page, "%d (%d)\n", 3247 mddev->new_chunk_sectors << 9, 3248 mddev->chunk_sectors << 9); 3249 return sprintf(page, "%d\n", mddev->chunk_sectors << 9); 3250 } 3251 3252 static ssize_t 3253 chunk_size_store(mddev_t *mddev, const char *buf, size_t len) 3254 { 3255 char *e; 3256 unsigned long n = simple_strtoul(buf, &e, 10); 3257 3258 if (!*buf || (*e && *e != '\n')) 3259 return -EINVAL; 3260 3261 if (mddev->pers) { 3262 int err; 3263 if (mddev->pers->check_reshape == NULL) 3264 return -EBUSY; 3265 mddev->new_chunk_sectors = n >> 9; 3266 err = mddev->pers->check_reshape(mddev); 3267 if (err) { 3268 mddev->new_chunk_sectors = mddev->chunk_sectors; 3269 return err; 3270 } 3271 } else { 3272 mddev->new_chunk_sectors = n >> 9; 3273 if (mddev->reshape_position == MaxSector) 3274 mddev->chunk_sectors = n >> 9; 3275 } 3276 return len; 3277 } 3278 static struct md_sysfs_entry md_chunk_size = 3279 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); 3280 3281 static ssize_t 3282 resync_start_show(mddev_t *mddev, char *page) 3283 { 3284 if (mddev->recovery_cp == MaxSector) 3285 return sprintf(page, "none\n"); 3286 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); 3287 } 3288 3289 static ssize_t 3290 resync_start_store(mddev_t *mddev, const char *buf, size_t len) 3291 { 3292 char *e; 3293 unsigned long long n = simple_strtoull(buf, &e, 10); 3294 3295 if (mddev->pers) 3296 return -EBUSY; 3297 if (cmd_match(buf, "none")) 3298 n = MaxSector; 3299 else if (!*buf || (*e && *e != '\n')) 3300 return -EINVAL; 3301 3302 mddev->recovery_cp = n; 3303 return len; 3304 } 3305 static struct md_sysfs_entry md_resync_start = 3306 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); 3307 3308 /* 3309 * The array state can be: 3310 * 3311 * clear 3312 * No devices, no size, no level 3313 * Equivalent to STOP_ARRAY ioctl 3314 * inactive 3315 * May have some settings, but array is not active 3316 * all IO results in error 3317 * When written, doesn't tear down array, but just stops it 3318 * suspended (not supported yet) 3319 * All IO requests will block. The array can be reconfigured. 3320 * Writing this, if accepted, will block until array is quiescent 3321 * readonly 3322 * no resync can happen. no superblocks get written. 3323 * write requests fail 3324 * read-auto 3325 * like readonly, but behaves like 'clean' on a write request. 3326 * 3327 * clean - no pending writes, but otherwise active. 3328 * When written to inactive array, starts without resync 3329 * If a write request arrives then 3330 * if metadata is known, mark 'dirty' and switch to 'active'. 3331 * if not known, block and switch to write-pending 3332 * If written to an active array that has pending writes, then fails. 3333 * active 3334 * fully active: IO and resync can be happening. 3335 * When written to inactive array, starts with resync 3336 * 3337 * write-pending 3338 * clean, but writes are blocked waiting for 'active' to be written. 3339 * 3340 * active-idle 3341 * like active, but no writes have been seen for a while (100msec). 3342 * 3343 */ 3344 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, 3345 write_pending, active_idle, bad_word}; 3346 static char *array_states[] = { 3347 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", 3348 "write-pending", "active-idle", NULL }; 3349 3350 static int match_word(const char *word, char **list) 3351 { 3352 int n; 3353 for (n=0; list[n]; n++) 3354 if (cmd_match(word, list[n])) 3355 break; 3356 return n; 3357 } 3358 3359 static ssize_t 3360 array_state_show(mddev_t *mddev, char *page) 3361 { 3362 enum array_state st = inactive; 3363 3364 if (mddev->pers) 3365 switch(mddev->ro) { 3366 case 1: 3367 st = readonly; 3368 break; 3369 case 2: 3370 st = read_auto; 3371 break; 3372 case 0: 3373 if (mddev->in_sync) 3374 st = clean; 3375 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 3376 st = write_pending; 3377 else if (mddev->safemode) 3378 st = active_idle; 3379 else 3380 st = active; 3381 } 3382 else { 3383 if (list_empty(&mddev->disks) && 3384 mddev->raid_disks == 0 && 3385 mddev->dev_sectors == 0) 3386 st = clear; 3387 else 3388 st = inactive; 3389 } 3390 return sprintf(page, "%s\n", array_states[st]); 3391 } 3392 3393 static int do_md_stop(mddev_t * mddev, int ro, int is_open); 3394 static int md_set_readonly(mddev_t * mddev, int is_open); 3395 static int do_md_run(mddev_t * mddev); 3396 static int restart_array(mddev_t *mddev); 3397 3398 static ssize_t 3399 array_state_store(mddev_t *mddev, const char *buf, size_t len) 3400 { 3401 int err = -EINVAL; 3402 enum array_state st = match_word(buf, array_states); 3403 switch(st) { 3404 case bad_word: 3405 break; 3406 case clear: 3407 /* stopping an active array */ 3408 if (atomic_read(&mddev->openers) > 0) 3409 return -EBUSY; 3410 err = do_md_stop(mddev, 0, 0); 3411 break; 3412 case inactive: 3413 /* stopping an active array */ 3414 if (mddev->pers) { 3415 if (atomic_read(&mddev->openers) > 0) 3416 return -EBUSY; 3417 err = do_md_stop(mddev, 2, 0); 3418 } else 3419 err = 0; /* already inactive */ 3420 break; 3421 case suspended: 3422 break; /* not supported yet */ 3423 case readonly: 3424 if (mddev->pers) 3425 err = md_set_readonly(mddev, 0); 3426 else { 3427 mddev->ro = 1; 3428 set_disk_ro(mddev->gendisk, 1); 3429 err = do_md_run(mddev); 3430 } 3431 break; 3432 case read_auto: 3433 if (mddev->pers) { 3434 if (mddev->ro == 0) 3435 err = md_set_readonly(mddev, 0); 3436 else if (mddev->ro == 1) 3437 err = restart_array(mddev); 3438 if (err == 0) { 3439 mddev->ro = 2; 3440 set_disk_ro(mddev->gendisk, 0); 3441 } 3442 } else { 3443 mddev->ro = 2; 3444 err = do_md_run(mddev); 3445 } 3446 break; 3447 case clean: 3448 if (mddev->pers) { 3449 restart_array(mddev); 3450 spin_lock_irq(&mddev->write_lock); 3451 if (atomic_read(&mddev->writes_pending) == 0) { 3452 if (mddev->in_sync == 0) { 3453 mddev->in_sync = 1; 3454 if (mddev->safemode == 1) 3455 mddev->safemode = 0; 3456 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 3457 } 3458 err = 0; 3459 } else 3460 err = -EBUSY; 3461 spin_unlock_irq(&mddev->write_lock); 3462 } else 3463 err = -EINVAL; 3464 break; 3465 case active: 3466 if (mddev->pers) { 3467 restart_array(mddev); 3468 clear_bit(MD_CHANGE_PENDING, &mddev->flags); 3469 wake_up(&mddev->sb_wait); 3470 err = 0; 3471 } else { 3472 mddev->ro = 0; 3473 set_disk_ro(mddev->gendisk, 0); 3474 err = do_md_run(mddev); 3475 } 3476 break; 3477 case write_pending: 3478 case active_idle: 3479 /* these cannot be set */ 3480 break; 3481 } 3482 if (err) 3483 return err; 3484 else { 3485 sysfs_notify_dirent_safe(mddev->sysfs_state); 3486 return len; 3487 } 3488 } 3489 static struct md_sysfs_entry md_array_state = 3490 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); 3491 3492 static ssize_t 3493 max_corrected_read_errors_show(mddev_t *mddev, char *page) { 3494 return sprintf(page, "%d\n", 3495 atomic_read(&mddev->max_corr_read_errors)); 3496 } 3497 3498 static ssize_t 3499 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) 3500 { 3501 char *e; 3502 unsigned long n = simple_strtoul(buf, &e, 10); 3503 3504 if (*buf && (*e == 0 || *e == '\n')) { 3505 atomic_set(&mddev->max_corr_read_errors, n); 3506 return len; 3507 } 3508 return -EINVAL; 3509 } 3510 3511 static struct md_sysfs_entry max_corr_read_errors = 3512 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, 3513 max_corrected_read_errors_store); 3514 3515 static ssize_t 3516 null_show(mddev_t *mddev, char *page) 3517 { 3518 return -EINVAL; 3519 } 3520 3521 static ssize_t 3522 new_dev_store(mddev_t *mddev, const char *buf, size_t len) 3523 { 3524 /* buf must be %d:%d\n? giving major and minor numbers */ 3525 /* The new device is added to the array. 3526 * If the array has a persistent superblock, we read the 3527 * superblock to initialise info and check validity. 3528 * Otherwise, only checking done is that in bind_rdev_to_array, 3529 * which mainly checks size. 3530 */ 3531 char *e; 3532 int major = simple_strtoul(buf, &e, 10); 3533 int minor; 3534 dev_t dev; 3535 mdk_rdev_t *rdev; 3536 int err; 3537 3538 if (!*buf || *e != ':' || !e[1] || e[1] == '\n') 3539 return -EINVAL; 3540 minor = simple_strtoul(e+1, &e, 10); 3541 if (*e && *e != '\n') 3542 return -EINVAL; 3543 dev = MKDEV(major, minor); 3544 if (major != MAJOR(dev) || 3545 minor != MINOR(dev)) 3546 return -EOVERFLOW; 3547 3548 3549 if (mddev->persistent) { 3550 rdev = md_import_device(dev, mddev->major_version, 3551 mddev->minor_version); 3552 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { 3553 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 3554 mdk_rdev_t, same_set); 3555 err = super_types[mddev->major_version] 3556 .load_super(rdev, rdev0, mddev->minor_version); 3557 if (err < 0) 3558 goto out; 3559 } 3560 } else if (mddev->external) 3561 rdev = md_import_device(dev, -2, -1); 3562 else 3563 rdev = md_import_device(dev, -1, -1); 3564 3565 if (IS_ERR(rdev)) 3566 return PTR_ERR(rdev); 3567 err = bind_rdev_to_array(rdev, mddev); 3568 out: 3569 if (err) 3570 export_rdev(rdev); 3571 return err ? err : len; 3572 } 3573 3574 static struct md_sysfs_entry md_new_device = 3575 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); 3576 3577 static ssize_t 3578 bitmap_store(mddev_t *mddev, const char *buf, size_t len) 3579 { 3580 char *end; 3581 unsigned long chunk, end_chunk; 3582 3583 if (!mddev->bitmap) 3584 goto out; 3585 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ 3586 while (*buf) { 3587 chunk = end_chunk = simple_strtoul(buf, &end, 0); 3588 if (buf == end) break; 3589 if (*end == '-') { /* range */ 3590 buf = end + 1; 3591 end_chunk = simple_strtoul(buf, &end, 0); 3592 if (buf == end) break; 3593 } 3594 if (*end && !isspace(*end)) break; 3595 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); 3596 buf = skip_spaces(end); 3597 } 3598 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ 3599 out: 3600 return len; 3601 } 3602 3603 static struct md_sysfs_entry md_bitmap = 3604 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); 3605 3606 static ssize_t 3607 size_show(mddev_t *mddev, char *page) 3608 { 3609 return sprintf(page, "%llu\n", 3610 (unsigned long long)mddev->dev_sectors / 2); 3611 } 3612 3613 static int update_size(mddev_t *mddev, sector_t num_sectors); 3614 3615 static ssize_t 3616 size_store(mddev_t *mddev, const char *buf, size_t len) 3617 { 3618 /* If array is inactive, we can reduce the component size, but 3619 * not increase it (except from 0). 3620 * If array is active, we can try an on-line resize 3621 */ 3622 sector_t sectors; 3623 int err = strict_blocks_to_sectors(buf, §ors); 3624 3625 if (err < 0) 3626 return err; 3627 if (mddev->pers) { 3628 err = update_size(mddev, sectors); 3629 md_update_sb(mddev, 1); 3630 } else { 3631 if (mddev->dev_sectors == 0 || 3632 mddev->dev_sectors > sectors) 3633 mddev->dev_sectors = sectors; 3634 else 3635 err = -ENOSPC; 3636 } 3637 return err ? err : len; 3638 } 3639 3640 static struct md_sysfs_entry md_size = 3641 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); 3642 3643 3644 /* Metdata version. 3645 * This is one of 3646 * 'none' for arrays with no metadata (good luck...) 3647 * 'external' for arrays with externally managed metadata, 3648 * or N.M for internally known formats 3649 */ 3650 static ssize_t 3651 metadata_show(mddev_t *mddev, char *page) 3652 { 3653 if (mddev->persistent) 3654 return sprintf(page, "%d.%d\n", 3655 mddev->major_version, mddev->minor_version); 3656 else if (mddev->external) 3657 return sprintf(page, "external:%s\n", mddev->metadata_type); 3658 else 3659 return sprintf(page, "none\n"); 3660 } 3661 3662 static ssize_t 3663 metadata_store(mddev_t *mddev, const char *buf, size_t len) 3664 { 3665 int major, minor; 3666 char *e; 3667 /* Changing the details of 'external' metadata is 3668 * always permitted. Otherwise there must be 3669 * no devices attached to the array. 3670 */ 3671 if (mddev->external && strncmp(buf, "external:", 9) == 0) 3672 ; 3673 else if (!list_empty(&mddev->disks)) 3674 return -EBUSY; 3675 3676 if (cmd_match(buf, "none")) { 3677 mddev->persistent = 0; 3678 mddev->external = 0; 3679 mddev->major_version = 0; 3680 mddev->minor_version = 90; 3681 return len; 3682 } 3683 if (strncmp(buf, "external:", 9) == 0) { 3684 size_t namelen = len-9; 3685 if (namelen >= sizeof(mddev->metadata_type)) 3686 namelen = sizeof(mddev->metadata_type)-1; 3687 strncpy(mddev->metadata_type, buf+9, namelen); 3688 mddev->metadata_type[namelen] = 0; 3689 if (namelen && mddev->metadata_type[namelen-1] == '\n') 3690 mddev->metadata_type[--namelen] = 0; 3691 mddev->persistent = 0; 3692 mddev->external = 1; 3693 mddev->major_version = 0; 3694 mddev->minor_version = 90; 3695 return len; 3696 } 3697 major = simple_strtoul(buf, &e, 10); 3698 if (e==buf || *e != '.') 3699 return -EINVAL; 3700 buf = e+1; 3701 minor = simple_strtoul(buf, &e, 10); 3702 if (e==buf || (*e && *e != '\n') ) 3703 return -EINVAL; 3704 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) 3705 return -ENOENT; 3706 mddev->major_version = major; 3707 mddev->minor_version = minor; 3708 mddev->persistent = 1; 3709 mddev->external = 0; 3710 return len; 3711 } 3712 3713 static struct md_sysfs_entry md_metadata = 3714 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 3715 3716 static ssize_t 3717 action_show(mddev_t *mddev, char *page) 3718 { 3719 char *type = "idle"; 3720 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 3721 type = "frozen"; 3722 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 3723 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { 3724 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 3725 type = "reshape"; 3726 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 3727 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 3728 type = "resync"; 3729 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 3730 type = "check"; 3731 else 3732 type = "repair"; 3733 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) 3734 type = "recover"; 3735 } 3736 return sprintf(page, "%s\n", type); 3737 } 3738 3739 static ssize_t 3740 action_store(mddev_t *mddev, const char *page, size_t len) 3741 { 3742 if (!mddev->pers || !mddev->pers->sync_request) 3743 return -EINVAL; 3744 3745 if (cmd_match(page, "frozen")) 3746 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 3747 else 3748 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 3749 3750 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { 3751 if (mddev->sync_thread) { 3752 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 3753 md_unregister_thread(mddev->sync_thread); 3754 mddev->sync_thread = NULL; 3755 mddev->recovery = 0; 3756 } 3757 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || 3758 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 3759 return -EBUSY; 3760 else if (cmd_match(page, "resync")) 3761 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3762 else if (cmd_match(page, "recover")) { 3763 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 3764 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3765 } else if (cmd_match(page, "reshape")) { 3766 int err; 3767 if (mddev->pers->start_reshape == NULL) 3768 return -EINVAL; 3769 err = mddev->pers->start_reshape(mddev); 3770 if (err) 3771 return err; 3772 sysfs_notify(&mddev->kobj, NULL, "degraded"); 3773 } else { 3774 if (cmd_match(page, "check")) 3775 set_bit(MD_RECOVERY_CHECK, &mddev->recovery); 3776 else if (!cmd_match(page, "repair")) 3777 return -EINVAL; 3778 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 3779 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 3780 } 3781 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3782 md_wakeup_thread(mddev->thread); 3783 sysfs_notify_dirent_safe(mddev->sysfs_action); 3784 return len; 3785 } 3786 3787 static ssize_t 3788 mismatch_cnt_show(mddev_t *mddev, char *page) 3789 { 3790 return sprintf(page, "%llu\n", 3791 (unsigned long long) mddev->resync_mismatches); 3792 } 3793 3794 static struct md_sysfs_entry md_scan_mode = 3795 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); 3796 3797 3798 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); 3799 3800 static ssize_t 3801 sync_min_show(mddev_t *mddev, char *page) 3802 { 3803 return sprintf(page, "%d (%s)\n", speed_min(mddev), 3804 mddev->sync_speed_min ? "local": "system"); 3805 } 3806 3807 static ssize_t 3808 sync_min_store(mddev_t *mddev, const char *buf, size_t len) 3809 { 3810 int min; 3811 char *e; 3812 if (strncmp(buf, "system", 6)==0) { 3813 mddev->sync_speed_min = 0; 3814 return len; 3815 } 3816 min = simple_strtoul(buf, &e, 10); 3817 if (buf == e || (*e && *e != '\n') || min <= 0) 3818 return -EINVAL; 3819 mddev->sync_speed_min = min; 3820 return len; 3821 } 3822 3823 static struct md_sysfs_entry md_sync_min = 3824 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); 3825 3826 static ssize_t 3827 sync_max_show(mddev_t *mddev, char *page) 3828 { 3829 return sprintf(page, "%d (%s)\n", speed_max(mddev), 3830 mddev->sync_speed_max ? "local": "system"); 3831 } 3832 3833 static ssize_t 3834 sync_max_store(mddev_t *mddev, const char *buf, size_t len) 3835 { 3836 int max; 3837 char *e; 3838 if (strncmp(buf, "system", 6)==0) { 3839 mddev->sync_speed_max = 0; 3840 return len; 3841 } 3842 max = simple_strtoul(buf, &e, 10); 3843 if (buf == e || (*e && *e != '\n') || max <= 0) 3844 return -EINVAL; 3845 mddev->sync_speed_max = max; 3846 return len; 3847 } 3848 3849 static struct md_sysfs_entry md_sync_max = 3850 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); 3851 3852 static ssize_t 3853 degraded_show(mddev_t *mddev, char *page) 3854 { 3855 return sprintf(page, "%d\n", mddev->degraded); 3856 } 3857 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); 3858 3859 static ssize_t 3860 sync_force_parallel_show(mddev_t *mddev, char *page) 3861 { 3862 return sprintf(page, "%d\n", mddev->parallel_resync); 3863 } 3864 3865 static ssize_t 3866 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) 3867 { 3868 long n; 3869 3870 if (strict_strtol(buf, 10, &n)) 3871 return -EINVAL; 3872 3873 if (n != 0 && n != 1) 3874 return -EINVAL; 3875 3876 mddev->parallel_resync = n; 3877 3878 if (mddev->sync_thread) 3879 wake_up(&resync_wait); 3880 3881 return len; 3882 } 3883 3884 /* force parallel resync, even with shared block devices */ 3885 static struct md_sysfs_entry md_sync_force_parallel = 3886 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, 3887 sync_force_parallel_show, sync_force_parallel_store); 3888 3889 static ssize_t 3890 sync_speed_show(mddev_t *mddev, char *page) 3891 { 3892 unsigned long resync, dt, db; 3893 if (mddev->curr_resync == 0) 3894 return sprintf(page, "none\n"); 3895 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); 3896 dt = (jiffies - mddev->resync_mark) / HZ; 3897 if (!dt) dt++; 3898 db = resync - mddev->resync_mark_cnt; 3899 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ 3900 } 3901 3902 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); 3903 3904 static ssize_t 3905 sync_completed_show(mddev_t *mddev, char *page) 3906 { 3907 unsigned long max_sectors, resync; 3908 3909 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3910 return sprintf(page, "none\n"); 3911 3912 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 3913 max_sectors = mddev->resync_max_sectors; 3914 else 3915 max_sectors = mddev->dev_sectors; 3916 3917 resync = mddev->curr_resync_completed; 3918 return sprintf(page, "%lu / %lu\n", resync, max_sectors); 3919 } 3920 3921 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); 3922 3923 static ssize_t 3924 min_sync_show(mddev_t *mddev, char *page) 3925 { 3926 return sprintf(page, "%llu\n", 3927 (unsigned long long)mddev->resync_min); 3928 } 3929 static ssize_t 3930 min_sync_store(mddev_t *mddev, const char *buf, size_t len) 3931 { 3932 unsigned long long min; 3933 if (strict_strtoull(buf, 10, &min)) 3934 return -EINVAL; 3935 if (min > mddev->resync_max) 3936 return -EINVAL; 3937 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3938 return -EBUSY; 3939 3940 /* Must be a multiple of chunk_size */ 3941 if (mddev->chunk_sectors) { 3942 sector_t temp = min; 3943 if (sector_div(temp, mddev->chunk_sectors)) 3944 return -EINVAL; 3945 } 3946 mddev->resync_min = min; 3947 3948 return len; 3949 } 3950 3951 static struct md_sysfs_entry md_min_sync = 3952 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); 3953 3954 static ssize_t 3955 max_sync_show(mddev_t *mddev, char *page) 3956 { 3957 if (mddev->resync_max == MaxSector) 3958 return sprintf(page, "max\n"); 3959 else 3960 return sprintf(page, "%llu\n", 3961 (unsigned long long)mddev->resync_max); 3962 } 3963 static ssize_t 3964 max_sync_store(mddev_t *mddev, const char *buf, size_t len) 3965 { 3966 if (strncmp(buf, "max", 3) == 0) 3967 mddev->resync_max = MaxSector; 3968 else { 3969 unsigned long long max; 3970 if (strict_strtoull(buf, 10, &max)) 3971 return -EINVAL; 3972 if (max < mddev->resync_min) 3973 return -EINVAL; 3974 if (max < mddev->resync_max && 3975 mddev->ro == 0 && 3976 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) 3977 return -EBUSY; 3978 3979 /* Must be a multiple of chunk_size */ 3980 if (mddev->chunk_sectors) { 3981 sector_t temp = max; 3982 if (sector_div(temp, mddev->chunk_sectors)) 3983 return -EINVAL; 3984 } 3985 mddev->resync_max = max; 3986 } 3987 wake_up(&mddev->recovery_wait); 3988 return len; 3989 } 3990 3991 static struct md_sysfs_entry md_max_sync = 3992 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); 3993 3994 static ssize_t 3995 suspend_lo_show(mddev_t *mddev, char *page) 3996 { 3997 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); 3998 } 3999 4000 static ssize_t 4001 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) 4002 { 4003 char *e; 4004 unsigned long long new = simple_strtoull(buf, &e, 10); 4005 4006 if (mddev->pers == NULL || 4007 mddev->pers->quiesce == NULL) 4008 return -EINVAL; 4009 if (buf == e || (*e && *e != '\n')) 4010 return -EINVAL; 4011 if (new >= mddev->suspend_hi || 4012 (new > mddev->suspend_lo && new < mddev->suspend_hi)) { 4013 mddev->suspend_lo = new; 4014 mddev->pers->quiesce(mddev, 2); 4015 return len; 4016 } else 4017 return -EINVAL; 4018 } 4019 static struct md_sysfs_entry md_suspend_lo = 4020 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); 4021 4022 4023 static ssize_t 4024 suspend_hi_show(mddev_t *mddev, char *page) 4025 { 4026 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); 4027 } 4028 4029 static ssize_t 4030 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) 4031 { 4032 char *e; 4033 unsigned long long new = simple_strtoull(buf, &e, 10); 4034 4035 if (mddev->pers == NULL || 4036 mddev->pers->quiesce == NULL) 4037 return -EINVAL; 4038 if (buf == e || (*e && *e != '\n')) 4039 return -EINVAL; 4040 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || 4041 (new > mddev->suspend_lo && new > mddev->suspend_hi)) { 4042 mddev->suspend_hi = new; 4043 mddev->pers->quiesce(mddev, 1); 4044 mddev->pers->quiesce(mddev, 0); 4045 return len; 4046 } else 4047 return -EINVAL; 4048 } 4049 static struct md_sysfs_entry md_suspend_hi = 4050 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); 4051 4052 static ssize_t 4053 reshape_position_show(mddev_t *mddev, char *page) 4054 { 4055 if (mddev->reshape_position != MaxSector) 4056 return sprintf(page, "%llu\n", 4057 (unsigned long long)mddev->reshape_position); 4058 strcpy(page, "none\n"); 4059 return 5; 4060 } 4061 4062 static ssize_t 4063 reshape_position_store(mddev_t *mddev, const char *buf, size_t len) 4064 { 4065 char *e; 4066 unsigned long long new = simple_strtoull(buf, &e, 10); 4067 if (mddev->pers) 4068 return -EBUSY; 4069 if (buf == e || (*e && *e != '\n')) 4070 return -EINVAL; 4071 mddev->reshape_position = new; 4072 mddev->delta_disks = 0; 4073 mddev->new_level = mddev->level; 4074 mddev->new_layout = mddev->layout; 4075 mddev->new_chunk_sectors = mddev->chunk_sectors; 4076 return len; 4077 } 4078 4079 static struct md_sysfs_entry md_reshape_position = 4080 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, 4081 reshape_position_store); 4082 4083 static ssize_t 4084 array_size_show(mddev_t *mddev, char *page) 4085 { 4086 if (mddev->external_size) 4087 return sprintf(page, "%llu\n", 4088 (unsigned long long)mddev->array_sectors/2); 4089 else 4090 return sprintf(page, "default\n"); 4091 } 4092 4093 static ssize_t 4094 array_size_store(mddev_t *mddev, const char *buf, size_t len) 4095 { 4096 sector_t sectors; 4097 4098 if (strncmp(buf, "default", 7) == 0) { 4099 if (mddev->pers) 4100 sectors = mddev->pers->size(mddev, 0, 0); 4101 else 4102 sectors = mddev->array_sectors; 4103 4104 mddev->external_size = 0; 4105 } else { 4106 if (strict_blocks_to_sectors(buf, §ors) < 0) 4107 return -EINVAL; 4108 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) 4109 return -E2BIG; 4110 4111 mddev->external_size = 1; 4112 } 4113 4114 mddev->array_sectors = sectors; 4115 set_capacity(mddev->gendisk, mddev->array_sectors); 4116 if (mddev->pers) 4117 revalidate_disk(mddev->gendisk); 4118 4119 return len; 4120 } 4121 4122 static struct md_sysfs_entry md_array_size = 4123 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, 4124 array_size_store); 4125 4126 static struct attribute *md_default_attrs[] = { 4127 &md_level.attr, 4128 &md_layout.attr, 4129 &md_raid_disks.attr, 4130 &md_chunk_size.attr, 4131 &md_size.attr, 4132 &md_resync_start.attr, 4133 &md_metadata.attr, 4134 &md_new_device.attr, 4135 &md_safe_delay.attr, 4136 &md_array_state.attr, 4137 &md_reshape_position.attr, 4138 &md_array_size.attr, 4139 &max_corr_read_errors.attr, 4140 NULL, 4141 }; 4142 4143 static struct attribute *md_redundancy_attrs[] = { 4144 &md_scan_mode.attr, 4145 &md_mismatches.attr, 4146 &md_sync_min.attr, 4147 &md_sync_max.attr, 4148 &md_sync_speed.attr, 4149 &md_sync_force_parallel.attr, 4150 &md_sync_completed.attr, 4151 &md_min_sync.attr, 4152 &md_max_sync.attr, 4153 &md_suspend_lo.attr, 4154 &md_suspend_hi.attr, 4155 &md_bitmap.attr, 4156 &md_degraded.attr, 4157 NULL, 4158 }; 4159 static struct attribute_group md_redundancy_group = { 4160 .name = NULL, 4161 .attrs = md_redundancy_attrs, 4162 }; 4163 4164 4165 static ssize_t 4166 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 4167 { 4168 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4169 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 4170 ssize_t rv; 4171 4172 if (!entry->show) 4173 return -EIO; 4174 rv = mddev_lock(mddev); 4175 if (!rv) { 4176 rv = entry->show(mddev, page); 4177 mddev_unlock(mddev); 4178 } 4179 return rv; 4180 } 4181 4182 static ssize_t 4183 md_attr_store(struct kobject *kobj, struct attribute *attr, 4184 const char *page, size_t length) 4185 { 4186 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); 4187 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); 4188 ssize_t rv; 4189 4190 if (!entry->store) 4191 return -EIO; 4192 if (!capable(CAP_SYS_ADMIN)) 4193 return -EACCES; 4194 rv = mddev_lock(mddev); 4195 if (mddev->hold_active == UNTIL_IOCTL) 4196 mddev->hold_active = 0; 4197 if (!rv) { 4198 rv = entry->store(mddev, page, length); 4199 mddev_unlock(mddev); 4200 } 4201 return rv; 4202 } 4203 4204 static void md_free(struct kobject *ko) 4205 { 4206 mddev_t *mddev = container_of(ko, mddev_t, kobj); 4207 4208 if (mddev->sysfs_state) 4209 sysfs_put(mddev->sysfs_state); 4210 4211 if (mddev->gendisk) { 4212 del_gendisk(mddev->gendisk); 4213 put_disk(mddev->gendisk); 4214 } 4215 if (mddev->queue) 4216 blk_cleanup_queue(mddev->queue); 4217 4218 kfree(mddev); 4219 } 4220 4221 static const struct sysfs_ops md_sysfs_ops = { 4222 .show = md_attr_show, 4223 .store = md_attr_store, 4224 }; 4225 static struct kobj_type md_ktype = { 4226 .release = md_free, 4227 .sysfs_ops = &md_sysfs_ops, 4228 .default_attrs = md_default_attrs, 4229 }; 4230 4231 int mdp_major = 0; 4232 4233 static void mddev_delayed_delete(struct work_struct *ws) 4234 { 4235 mddev_t *mddev = container_of(ws, mddev_t, del_work); 4236 4237 sysfs_remove_group(&mddev->kobj, &md_bitmap_group); 4238 kobject_del(&mddev->kobj); 4239 kobject_put(&mddev->kobj); 4240 } 4241 4242 static int md_alloc(dev_t dev, char *name) 4243 { 4244 static DEFINE_MUTEX(disks_mutex); 4245 mddev_t *mddev = mddev_find(dev); 4246 struct gendisk *disk; 4247 int partitioned; 4248 int shift; 4249 int unit; 4250 int error; 4251 4252 if (!mddev) 4253 return -ENODEV; 4254 4255 partitioned = (MAJOR(mddev->unit) != MD_MAJOR); 4256 shift = partitioned ? MdpMinorShift : 0; 4257 unit = MINOR(mddev->unit) >> shift; 4258 4259 /* wait for any previous instance if this device 4260 * to be completed removed (mddev_delayed_delete). 4261 */ 4262 flush_scheduled_work(); 4263 4264 mutex_lock(&disks_mutex); 4265 error = -EEXIST; 4266 if (mddev->gendisk) 4267 goto abort; 4268 4269 if (name) { 4270 /* Need to ensure that 'name' is not a duplicate. 4271 */ 4272 mddev_t *mddev2; 4273 spin_lock(&all_mddevs_lock); 4274 4275 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) 4276 if (mddev2->gendisk && 4277 strcmp(mddev2->gendisk->disk_name, name) == 0) { 4278 spin_unlock(&all_mddevs_lock); 4279 goto abort; 4280 } 4281 spin_unlock(&all_mddevs_lock); 4282 } 4283 4284 error = -ENOMEM; 4285 mddev->queue = blk_alloc_queue(GFP_KERNEL); 4286 if (!mddev->queue) 4287 goto abort; 4288 mddev->queue->queuedata = mddev; 4289 4290 /* Can be unlocked because the queue is new: no concurrency */ 4291 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); 4292 4293 blk_queue_make_request(mddev->queue, md_make_request); 4294 4295 disk = alloc_disk(1 << shift); 4296 if (!disk) { 4297 blk_cleanup_queue(mddev->queue); 4298 mddev->queue = NULL; 4299 goto abort; 4300 } 4301 disk->major = MAJOR(mddev->unit); 4302 disk->first_minor = unit << shift; 4303 if (name) 4304 strcpy(disk->disk_name, name); 4305 else if (partitioned) 4306 sprintf(disk->disk_name, "md_d%d", unit); 4307 else 4308 sprintf(disk->disk_name, "md%d", unit); 4309 disk->fops = &md_fops; 4310 disk->private_data = mddev; 4311 disk->queue = mddev->queue; 4312 /* Allow extended partitions. This makes the 4313 * 'mdp' device redundant, but we can't really 4314 * remove it now. 4315 */ 4316 disk->flags |= GENHD_FL_EXT_DEVT; 4317 add_disk(disk); 4318 mddev->gendisk = disk; 4319 error = kobject_init_and_add(&mddev->kobj, &md_ktype, 4320 &disk_to_dev(disk)->kobj, "%s", "md"); 4321 if (error) { 4322 /* This isn't possible, but as kobject_init_and_add is marked 4323 * __must_check, we must do something with the result 4324 */ 4325 printk(KERN_WARNING "md: cannot register %s/md - name in use\n", 4326 disk->disk_name); 4327 error = 0; 4328 } 4329 if (mddev->kobj.sd && 4330 sysfs_create_group(&mddev->kobj, &md_bitmap_group)) 4331 printk(KERN_DEBUG "pointless warning\n"); 4332 abort: 4333 mutex_unlock(&disks_mutex); 4334 if (!error && mddev->kobj.sd) { 4335 kobject_uevent(&mddev->kobj, KOBJ_ADD); 4336 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); 4337 } 4338 mddev_put(mddev); 4339 return error; 4340 } 4341 4342 static struct kobject *md_probe(dev_t dev, int *part, void *data) 4343 { 4344 md_alloc(dev, NULL); 4345 return NULL; 4346 } 4347 4348 static int add_named_array(const char *val, struct kernel_param *kp) 4349 { 4350 /* val must be "md_*" where * is not all digits. 4351 * We allocate an array with a large free minor number, and 4352 * set the name to val. val must not already be an active name. 4353 */ 4354 int len = strlen(val); 4355 char buf[DISK_NAME_LEN]; 4356 4357 while (len && val[len-1] == '\n') 4358 len--; 4359 if (len >= DISK_NAME_LEN) 4360 return -E2BIG; 4361 strlcpy(buf, val, len+1); 4362 if (strncmp(buf, "md_", 3) != 0) 4363 return -EINVAL; 4364 return md_alloc(0, buf); 4365 } 4366 4367 static void md_safemode_timeout(unsigned long data) 4368 { 4369 mddev_t *mddev = (mddev_t *) data; 4370 4371 if (!atomic_read(&mddev->writes_pending)) { 4372 mddev->safemode = 1; 4373 if (mddev->external) 4374 sysfs_notify_dirent_safe(mddev->sysfs_state); 4375 } 4376 md_wakeup_thread(mddev->thread); 4377 } 4378 4379 static int start_dirty_degraded; 4380 4381 int md_run(mddev_t *mddev) 4382 { 4383 int err; 4384 mdk_rdev_t *rdev; 4385 struct mdk_personality *pers; 4386 4387 if (list_empty(&mddev->disks)) 4388 /* cannot run an array with no devices.. */ 4389 return -EINVAL; 4390 4391 if (mddev->pers) 4392 return -EBUSY; 4393 /* Cannot run until previous stop completes properly */ 4394 if (mddev->sysfs_active) 4395 return -EBUSY; 4396 4397 /* 4398 * Analyze all RAID superblock(s) 4399 */ 4400 if (!mddev->raid_disks) { 4401 if (!mddev->persistent) 4402 return -EINVAL; 4403 analyze_sbs(mddev); 4404 } 4405 4406 if (mddev->level != LEVEL_NONE) 4407 request_module("md-level-%d", mddev->level); 4408 else if (mddev->clevel[0]) 4409 request_module("md-%s", mddev->clevel); 4410 4411 /* 4412 * Drop all container device buffers, from now on 4413 * the only valid external interface is through the md 4414 * device. 4415 */ 4416 list_for_each_entry(rdev, &mddev->disks, same_set) { 4417 if (test_bit(Faulty, &rdev->flags)) 4418 continue; 4419 sync_blockdev(rdev->bdev); 4420 invalidate_bdev(rdev->bdev); 4421 4422 /* perform some consistency tests on the device. 4423 * We don't want the data to overlap the metadata, 4424 * Internal Bitmap issues have been handled elsewhere. 4425 */ 4426 if (rdev->data_offset < rdev->sb_start) { 4427 if (mddev->dev_sectors && 4428 rdev->data_offset + mddev->dev_sectors 4429 > rdev->sb_start) { 4430 printk("md: %s: data overlaps metadata\n", 4431 mdname(mddev)); 4432 return -EINVAL; 4433 } 4434 } else { 4435 if (rdev->sb_start + rdev->sb_size/512 4436 > rdev->data_offset) { 4437 printk("md: %s: metadata overlaps data\n", 4438 mdname(mddev)); 4439 return -EINVAL; 4440 } 4441 } 4442 sysfs_notify_dirent_safe(rdev->sysfs_state); 4443 } 4444 4445 spin_lock(&pers_lock); 4446 pers = find_pers(mddev->level, mddev->clevel); 4447 if (!pers || !try_module_get(pers->owner)) { 4448 spin_unlock(&pers_lock); 4449 if (mddev->level != LEVEL_NONE) 4450 printk(KERN_WARNING "md: personality for level %d is not loaded!\n", 4451 mddev->level); 4452 else 4453 printk(KERN_WARNING "md: personality for level %s is not loaded!\n", 4454 mddev->clevel); 4455 return -EINVAL; 4456 } 4457 mddev->pers = pers; 4458 spin_unlock(&pers_lock); 4459 if (mddev->level != pers->level) { 4460 mddev->level = pers->level; 4461 mddev->new_level = pers->level; 4462 } 4463 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); 4464 4465 if (mddev->reshape_position != MaxSector && 4466 pers->start_reshape == NULL) { 4467 /* This personality cannot handle reshaping... */ 4468 mddev->pers = NULL; 4469 module_put(pers->owner); 4470 return -EINVAL; 4471 } 4472 4473 if (pers->sync_request) { 4474 /* Warn if this is a potentially silly 4475 * configuration. 4476 */ 4477 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 4478 mdk_rdev_t *rdev2; 4479 int warned = 0; 4480 4481 list_for_each_entry(rdev, &mddev->disks, same_set) 4482 list_for_each_entry(rdev2, &mddev->disks, same_set) { 4483 if (rdev < rdev2 && 4484 rdev->bdev->bd_contains == 4485 rdev2->bdev->bd_contains) { 4486 printk(KERN_WARNING 4487 "%s: WARNING: %s appears to be" 4488 " on the same physical disk as" 4489 " %s.\n", 4490 mdname(mddev), 4491 bdevname(rdev->bdev,b), 4492 bdevname(rdev2->bdev,b2)); 4493 warned = 1; 4494 } 4495 } 4496 4497 if (warned) 4498 printk(KERN_WARNING 4499 "True protection against single-disk" 4500 " failure might be compromised.\n"); 4501 } 4502 4503 mddev->recovery = 0; 4504 /* may be over-ridden by personality */ 4505 mddev->resync_max_sectors = mddev->dev_sectors; 4506 4507 mddev->barriers_work = 1; 4508 mddev->ok_start_degraded = start_dirty_degraded; 4509 4510 if (start_readonly && mddev->ro == 0) 4511 mddev->ro = 2; /* read-only, but switch on first write */ 4512 4513 err = mddev->pers->run(mddev); 4514 if (err) 4515 printk(KERN_ERR "md: pers->run() failed ...\n"); 4516 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { 4517 WARN_ONCE(!mddev->external_size, "%s: default size too small," 4518 " but 'external_size' not in effect?\n", __func__); 4519 printk(KERN_ERR 4520 "md: invalid array_size %llu > default size %llu\n", 4521 (unsigned long long)mddev->array_sectors / 2, 4522 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); 4523 err = -EINVAL; 4524 mddev->pers->stop(mddev); 4525 } 4526 if (err == 0 && mddev->pers->sync_request) { 4527 err = bitmap_create(mddev); 4528 if (err) { 4529 printk(KERN_ERR "%s: failed to create bitmap (%d)\n", 4530 mdname(mddev), err); 4531 mddev->pers->stop(mddev); 4532 } 4533 } 4534 if (err) { 4535 module_put(mddev->pers->owner); 4536 mddev->pers = NULL; 4537 bitmap_destroy(mddev); 4538 return err; 4539 } 4540 if (mddev->pers->sync_request) { 4541 if (mddev->kobj.sd && 4542 sysfs_create_group(&mddev->kobj, &md_redundancy_group)) 4543 printk(KERN_WARNING 4544 "md: cannot register extra attributes for %s\n", 4545 mdname(mddev)); 4546 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); 4547 } else if (mddev->ro == 2) /* auto-readonly not meaningful */ 4548 mddev->ro = 0; 4549 4550 atomic_set(&mddev->writes_pending,0); 4551 atomic_set(&mddev->max_corr_read_errors, 4552 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); 4553 mddev->safemode = 0; 4554 mddev->safemode_timer.function = md_safemode_timeout; 4555 mddev->safemode_timer.data = (unsigned long) mddev; 4556 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ 4557 mddev->in_sync = 1; 4558 4559 list_for_each_entry(rdev, &mddev->disks, same_set) 4560 if (rdev->raid_disk >= 0) { 4561 char nm[20]; 4562 sprintf(nm, "rd%d", rdev->raid_disk); 4563 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) 4564 /* failure here is OK */; 4565 } 4566 4567 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4568 4569 if (mddev->flags) 4570 md_update_sb(mddev, 0); 4571 4572 md_wakeup_thread(mddev->thread); 4573 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ 4574 4575 md_new_event(mddev); 4576 sysfs_notify_dirent_safe(mddev->sysfs_state); 4577 sysfs_notify_dirent_safe(mddev->sysfs_action); 4578 sysfs_notify(&mddev->kobj, NULL, "degraded"); 4579 return 0; 4580 } 4581 EXPORT_SYMBOL_GPL(md_run); 4582 4583 static int do_md_run(mddev_t *mddev) 4584 { 4585 int err; 4586 4587 err = md_run(mddev); 4588 if (err) 4589 goto out; 4590 err = bitmap_load(mddev); 4591 if (err) { 4592 bitmap_destroy(mddev); 4593 goto out; 4594 } 4595 set_capacity(mddev->gendisk, mddev->array_sectors); 4596 revalidate_disk(mddev->gendisk); 4597 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4598 out: 4599 return err; 4600 } 4601 4602 static int restart_array(mddev_t *mddev) 4603 { 4604 struct gendisk *disk = mddev->gendisk; 4605 4606 /* Complain if it has no devices */ 4607 if (list_empty(&mddev->disks)) 4608 return -ENXIO; 4609 if (!mddev->pers) 4610 return -EINVAL; 4611 if (!mddev->ro) 4612 return -EBUSY; 4613 mddev->safemode = 0; 4614 mddev->ro = 0; 4615 set_disk_ro(disk, 0); 4616 printk(KERN_INFO "md: %s switched to read-write mode.\n", 4617 mdname(mddev)); 4618 /* Kick recovery or resync if necessary */ 4619 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 4620 md_wakeup_thread(mddev->thread); 4621 md_wakeup_thread(mddev->sync_thread); 4622 sysfs_notify_dirent_safe(mddev->sysfs_state); 4623 return 0; 4624 } 4625 4626 /* similar to deny_write_access, but accounts for our holding a reference 4627 * to the file ourselves */ 4628 static int deny_bitmap_write_access(struct file * file) 4629 { 4630 struct inode *inode = file->f_mapping->host; 4631 4632 spin_lock(&inode->i_lock); 4633 if (atomic_read(&inode->i_writecount) > 1) { 4634 spin_unlock(&inode->i_lock); 4635 return -ETXTBSY; 4636 } 4637 atomic_set(&inode->i_writecount, -1); 4638 spin_unlock(&inode->i_lock); 4639 4640 return 0; 4641 } 4642 4643 void restore_bitmap_write_access(struct file *file) 4644 { 4645 struct inode *inode = file->f_mapping->host; 4646 4647 spin_lock(&inode->i_lock); 4648 atomic_set(&inode->i_writecount, 1); 4649 spin_unlock(&inode->i_lock); 4650 } 4651 4652 static void md_clean(mddev_t *mddev) 4653 { 4654 mddev->array_sectors = 0; 4655 mddev->external_size = 0; 4656 mddev->dev_sectors = 0; 4657 mddev->raid_disks = 0; 4658 mddev->recovery_cp = 0; 4659 mddev->resync_min = 0; 4660 mddev->resync_max = MaxSector; 4661 mddev->reshape_position = MaxSector; 4662 mddev->external = 0; 4663 mddev->persistent = 0; 4664 mddev->level = LEVEL_NONE; 4665 mddev->clevel[0] = 0; 4666 mddev->flags = 0; 4667 mddev->ro = 0; 4668 mddev->metadata_type[0] = 0; 4669 mddev->chunk_sectors = 0; 4670 mddev->ctime = mddev->utime = 0; 4671 mddev->layout = 0; 4672 mddev->max_disks = 0; 4673 mddev->events = 0; 4674 mddev->can_decrease_events = 0; 4675 mddev->delta_disks = 0; 4676 mddev->new_level = LEVEL_NONE; 4677 mddev->new_layout = 0; 4678 mddev->new_chunk_sectors = 0; 4679 mddev->curr_resync = 0; 4680 mddev->resync_mismatches = 0; 4681 mddev->suspend_lo = mddev->suspend_hi = 0; 4682 mddev->sync_speed_min = mddev->sync_speed_max = 0; 4683 mddev->recovery = 0; 4684 mddev->in_sync = 0; 4685 mddev->degraded = 0; 4686 mddev->barriers_work = 0; 4687 mddev->safemode = 0; 4688 mddev->bitmap_info.offset = 0; 4689 mddev->bitmap_info.default_offset = 0; 4690 mddev->bitmap_info.chunksize = 0; 4691 mddev->bitmap_info.daemon_sleep = 0; 4692 mddev->bitmap_info.max_write_behind = 0; 4693 mddev->plug = NULL; 4694 } 4695 4696 void md_stop_writes(mddev_t *mddev) 4697 { 4698 if (mddev->sync_thread) { 4699 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4700 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 4701 md_unregister_thread(mddev->sync_thread); 4702 mddev->sync_thread = NULL; 4703 } 4704 4705 del_timer_sync(&mddev->safemode_timer); 4706 4707 bitmap_flush(mddev); 4708 md_super_wait(mddev); 4709 4710 if (!mddev->in_sync || mddev->flags) { 4711 /* mark array as shutdown cleanly */ 4712 mddev->in_sync = 1; 4713 md_update_sb(mddev, 1); 4714 } 4715 } 4716 EXPORT_SYMBOL_GPL(md_stop_writes); 4717 4718 void md_stop(mddev_t *mddev) 4719 { 4720 mddev->pers->stop(mddev); 4721 if (mddev->pers->sync_request && mddev->to_remove == NULL) 4722 mddev->to_remove = &md_redundancy_group; 4723 module_put(mddev->pers->owner); 4724 mddev->pers = NULL; 4725 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4726 } 4727 EXPORT_SYMBOL_GPL(md_stop); 4728 4729 static int md_set_readonly(mddev_t *mddev, int is_open) 4730 { 4731 int err = 0; 4732 mutex_lock(&mddev->open_mutex); 4733 if (atomic_read(&mddev->openers) > is_open) { 4734 printk("md: %s still in use.\n",mdname(mddev)); 4735 err = -EBUSY; 4736 goto out; 4737 } 4738 if (mddev->pers) { 4739 md_stop_writes(mddev); 4740 4741 err = -ENXIO; 4742 if (mddev->ro==1) 4743 goto out; 4744 mddev->ro = 1; 4745 set_disk_ro(mddev->gendisk, 1); 4746 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); 4747 sysfs_notify_dirent_safe(mddev->sysfs_state); 4748 err = 0; 4749 } 4750 out: 4751 mutex_unlock(&mddev->open_mutex); 4752 return err; 4753 } 4754 4755 /* mode: 4756 * 0 - completely stop and dis-assemble array 4757 * 2 - stop but do not disassemble array 4758 */ 4759 static int do_md_stop(mddev_t * mddev, int mode, int is_open) 4760 { 4761 struct gendisk *disk = mddev->gendisk; 4762 mdk_rdev_t *rdev; 4763 4764 mutex_lock(&mddev->open_mutex); 4765 if (atomic_read(&mddev->openers) > is_open || 4766 mddev->sysfs_active) { 4767 printk("md: %s still in use.\n",mdname(mddev)); 4768 mutex_unlock(&mddev->open_mutex); 4769 return -EBUSY; 4770 } 4771 4772 if (mddev->pers) { 4773 if (mddev->ro) 4774 set_disk_ro(disk, 0); 4775 4776 md_stop_writes(mddev); 4777 md_stop(mddev); 4778 mddev->queue->merge_bvec_fn = NULL; 4779 mddev->queue->unplug_fn = NULL; 4780 mddev->queue->backing_dev_info.congested_fn = NULL; 4781 4782 /* tell userspace to handle 'inactive' */ 4783 sysfs_notify_dirent_safe(mddev->sysfs_state); 4784 4785 list_for_each_entry(rdev, &mddev->disks, same_set) 4786 if (rdev->raid_disk >= 0) { 4787 char nm[20]; 4788 sprintf(nm, "rd%d", rdev->raid_disk); 4789 sysfs_remove_link(&mddev->kobj, nm); 4790 } 4791 4792 set_capacity(disk, 0); 4793 mutex_unlock(&mddev->open_mutex); 4794 revalidate_disk(disk); 4795 4796 if (mddev->ro) 4797 mddev->ro = 0; 4798 } else 4799 mutex_unlock(&mddev->open_mutex); 4800 /* 4801 * Free resources if final stop 4802 */ 4803 if (mode == 0) { 4804 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); 4805 4806 bitmap_destroy(mddev); 4807 if (mddev->bitmap_info.file) { 4808 restore_bitmap_write_access(mddev->bitmap_info.file); 4809 fput(mddev->bitmap_info.file); 4810 mddev->bitmap_info.file = NULL; 4811 } 4812 mddev->bitmap_info.offset = 0; 4813 4814 export_array(mddev); 4815 4816 md_clean(mddev); 4817 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); 4818 if (mddev->hold_active == UNTIL_STOP) 4819 mddev->hold_active = 0; 4820 } 4821 blk_integrity_unregister(disk); 4822 md_new_event(mddev); 4823 sysfs_notify_dirent_safe(mddev->sysfs_state); 4824 return 0; 4825 } 4826 4827 #ifndef MODULE 4828 static void autorun_array(mddev_t *mddev) 4829 { 4830 mdk_rdev_t *rdev; 4831 int err; 4832 4833 if (list_empty(&mddev->disks)) 4834 return; 4835 4836 printk(KERN_INFO "md: running: "); 4837 4838 list_for_each_entry(rdev, &mddev->disks, same_set) { 4839 char b[BDEVNAME_SIZE]; 4840 printk("<%s>", bdevname(rdev->bdev,b)); 4841 } 4842 printk("\n"); 4843 4844 err = do_md_run(mddev); 4845 if (err) { 4846 printk(KERN_WARNING "md: do_md_run() returned %d\n", err); 4847 do_md_stop(mddev, 0, 0); 4848 } 4849 } 4850 4851 /* 4852 * lets try to run arrays based on all disks that have arrived 4853 * until now. (those are in pending_raid_disks) 4854 * 4855 * the method: pick the first pending disk, collect all disks with 4856 * the same UUID, remove all from the pending list and put them into 4857 * the 'same_array' list. Then order this list based on superblock 4858 * update time (freshest comes first), kick out 'old' disks and 4859 * compare superblocks. If everything's fine then run it. 4860 * 4861 * If "unit" is allocated, then bump its reference count 4862 */ 4863 static void autorun_devices(int part) 4864 { 4865 mdk_rdev_t *rdev0, *rdev, *tmp; 4866 mddev_t *mddev; 4867 char b[BDEVNAME_SIZE]; 4868 4869 printk(KERN_INFO "md: autorun ...\n"); 4870 while (!list_empty(&pending_raid_disks)) { 4871 int unit; 4872 dev_t dev; 4873 LIST_HEAD(candidates); 4874 rdev0 = list_entry(pending_raid_disks.next, 4875 mdk_rdev_t, same_set); 4876 4877 printk(KERN_INFO "md: considering %s ...\n", 4878 bdevname(rdev0->bdev,b)); 4879 INIT_LIST_HEAD(&candidates); 4880 rdev_for_each_list(rdev, tmp, &pending_raid_disks) 4881 if (super_90_load(rdev, rdev0, 0) >= 0) { 4882 printk(KERN_INFO "md: adding %s ...\n", 4883 bdevname(rdev->bdev,b)); 4884 list_move(&rdev->same_set, &candidates); 4885 } 4886 /* 4887 * now we have a set of devices, with all of them having 4888 * mostly sane superblocks. It's time to allocate the 4889 * mddev. 4890 */ 4891 if (part) { 4892 dev = MKDEV(mdp_major, 4893 rdev0->preferred_minor << MdpMinorShift); 4894 unit = MINOR(dev) >> MdpMinorShift; 4895 } else { 4896 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); 4897 unit = MINOR(dev); 4898 } 4899 if (rdev0->preferred_minor != unit) { 4900 printk(KERN_INFO "md: unit number in %s is bad: %d\n", 4901 bdevname(rdev0->bdev, b), rdev0->preferred_minor); 4902 break; 4903 } 4904 4905 md_probe(dev, NULL, NULL); 4906 mddev = mddev_find(dev); 4907 if (!mddev || !mddev->gendisk) { 4908 if (mddev) 4909 mddev_put(mddev); 4910 printk(KERN_ERR 4911 "md: cannot allocate memory for md drive.\n"); 4912 break; 4913 } 4914 if (mddev_lock(mddev)) 4915 printk(KERN_WARNING "md: %s locked, cannot run\n", 4916 mdname(mddev)); 4917 else if (mddev->raid_disks || mddev->major_version 4918 || !list_empty(&mddev->disks)) { 4919 printk(KERN_WARNING 4920 "md: %s already running, cannot run %s\n", 4921 mdname(mddev), bdevname(rdev0->bdev,b)); 4922 mddev_unlock(mddev); 4923 } else { 4924 printk(KERN_INFO "md: created %s\n", mdname(mddev)); 4925 mddev->persistent = 1; 4926 rdev_for_each_list(rdev, tmp, &candidates) { 4927 list_del_init(&rdev->same_set); 4928 if (bind_rdev_to_array(rdev, mddev)) 4929 export_rdev(rdev); 4930 } 4931 autorun_array(mddev); 4932 mddev_unlock(mddev); 4933 } 4934 /* on success, candidates will be empty, on error 4935 * it won't... 4936 */ 4937 rdev_for_each_list(rdev, tmp, &candidates) { 4938 list_del_init(&rdev->same_set); 4939 export_rdev(rdev); 4940 } 4941 mddev_put(mddev); 4942 } 4943 printk(KERN_INFO "md: ... autorun DONE.\n"); 4944 } 4945 #endif /* !MODULE */ 4946 4947 static int get_version(void __user * arg) 4948 { 4949 mdu_version_t ver; 4950 4951 ver.major = MD_MAJOR_VERSION; 4952 ver.minor = MD_MINOR_VERSION; 4953 ver.patchlevel = MD_PATCHLEVEL_VERSION; 4954 4955 if (copy_to_user(arg, &ver, sizeof(ver))) 4956 return -EFAULT; 4957 4958 return 0; 4959 } 4960 4961 static int get_array_info(mddev_t * mddev, void __user * arg) 4962 { 4963 mdu_array_info_t info; 4964 int nr,working,insync,failed,spare; 4965 mdk_rdev_t *rdev; 4966 4967 nr=working=insync=failed=spare=0; 4968 list_for_each_entry(rdev, &mddev->disks, same_set) { 4969 nr++; 4970 if (test_bit(Faulty, &rdev->flags)) 4971 failed++; 4972 else { 4973 working++; 4974 if (test_bit(In_sync, &rdev->flags)) 4975 insync++; 4976 else 4977 spare++; 4978 } 4979 } 4980 4981 info.major_version = mddev->major_version; 4982 info.minor_version = mddev->minor_version; 4983 info.patch_version = MD_PATCHLEVEL_VERSION; 4984 info.ctime = mddev->ctime; 4985 info.level = mddev->level; 4986 info.size = mddev->dev_sectors / 2; 4987 if (info.size != mddev->dev_sectors / 2) /* overflow */ 4988 info.size = -1; 4989 info.nr_disks = nr; 4990 info.raid_disks = mddev->raid_disks; 4991 info.md_minor = mddev->md_minor; 4992 info.not_persistent= !mddev->persistent; 4993 4994 info.utime = mddev->utime; 4995 info.state = 0; 4996 if (mddev->in_sync) 4997 info.state = (1<<MD_SB_CLEAN); 4998 if (mddev->bitmap && mddev->bitmap_info.offset) 4999 info.state = (1<<MD_SB_BITMAP_PRESENT); 5000 info.active_disks = insync; 5001 info.working_disks = working; 5002 info.failed_disks = failed; 5003 info.spare_disks = spare; 5004 5005 info.layout = mddev->layout; 5006 info.chunk_size = mddev->chunk_sectors << 9; 5007 5008 if (copy_to_user(arg, &info, sizeof(info))) 5009 return -EFAULT; 5010 5011 return 0; 5012 } 5013 5014 static int get_bitmap_file(mddev_t * mddev, void __user * arg) 5015 { 5016 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ 5017 char *ptr, *buf = NULL; 5018 int err = -ENOMEM; 5019 5020 if (md_allow_write(mddev)) 5021 file = kmalloc(sizeof(*file), GFP_NOIO); 5022 else 5023 file = kmalloc(sizeof(*file), GFP_KERNEL); 5024 5025 if (!file) 5026 goto out; 5027 5028 /* bitmap disabled, zero the first byte and copy out */ 5029 if (!mddev->bitmap || !mddev->bitmap->file) { 5030 file->pathname[0] = '\0'; 5031 goto copy_out; 5032 } 5033 5034 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); 5035 if (!buf) 5036 goto out; 5037 5038 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); 5039 if (IS_ERR(ptr)) 5040 goto out; 5041 5042 strcpy(file->pathname, ptr); 5043 5044 copy_out: 5045 err = 0; 5046 if (copy_to_user(arg, file, sizeof(*file))) 5047 err = -EFAULT; 5048 out: 5049 kfree(buf); 5050 kfree(file); 5051 return err; 5052 } 5053 5054 static int get_disk_info(mddev_t * mddev, void __user * arg) 5055 { 5056 mdu_disk_info_t info; 5057 mdk_rdev_t *rdev; 5058 5059 if (copy_from_user(&info, arg, sizeof(info))) 5060 return -EFAULT; 5061 5062 rdev = find_rdev_nr(mddev, info.number); 5063 if (rdev) { 5064 info.major = MAJOR(rdev->bdev->bd_dev); 5065 info.minor = MINOR(rdev->bdev->bd_dev); 5066 info.raid_disk = rdev->raid_disk; 5067 info.state = 0; 5068 if (test_bit(Faulty, &rdev->flags)) 5069 info.state |= (1<<MD_DISK_FAULTY); 5070 else if (test_bit(In_sync, &rdev->flags)) { 5071 info.state |= (1<<MD_DISK_ACTIVE); 5072 info.state |= (1<<MD_DISK_SYNC); 5073 } 5074 if (test_bit(WriteMostly, &rdev->flags)) 5075 info.state |= (1<<MD_DISK_WRITEMOSTLY); 5076 } else { 5077 info.major = info.minor = 0; 5078 info.raid_disk = -1; 5079 info.state = (1<<MD_DISK_REMOVED); 5080 } 5081 5082 if (copy_to_user(arg, &info, sizeof(info))) 5083 return -EFAULT; 5084 5085 return 0; 5086 } 5087 5088 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) 5089 { 5090 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; 5091 mdk_rdev_t *rdev; 5092 dev_t dev = MKDEV(info->major,info->minor); 5093 5094 if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) 5095 return -EOVERFLOW; 5096 5097 if (!mddev->raid_disks) { 5098 int err; 5099 /* expecting a device which has a superblock */ 5100 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); 5101 if (IS_ERR(rdev)) { 5102 printk(KERN_WARNING 5103 "md: md_import_device returned %ld\n", 5104 PTR_ERR(rdev)); 5105 return PTR_ERR(rdev); 5106 } 5107 if (!list_empty(&mddev->disks)) { 5108 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, 5109 mdk_rdev_t, same_set); 5110 err = super_types[mddev->major_version] 5111 .load_super(rdev, rdev0, mddev->minor_version); 5112 if (err < 0) { 5113 printk(KERN_WARNING 5114 "md: %s has different UUID to %s\n", 5115 bdevname(rdev->bdev,b), 5116 bdevname(rdev0->bdev,b2)); 5117 export_rdev(rdev); 5118 return -EINVAL; 5119 } 5120 } 5121 err = bind_rdev_to_array(rdev, mddev); 5122 if (err) 5123 export_rdev(rdev); 5124 return err; 5125 } 5126 5127 /* 5128 * add_new_disk can be used once the array is assembled 5129 * to add "hot spares". They must already have a superblock 5130 * written 5131 */ 5132 if (mddev->pers) { 5133 int err; 5134 if (!mddev->pers->hot_add_disk) { 5135 printk(KERN_WARNING 5136 "%s: personality does not support diskops!\n", 5137 mdname(mddev)); 5138 return -EINVAL; 5139 } 5140 if (mddev->persistent) 5141 rdev = md_import_device(dev, mddev->major_version, 5142 mddev->minor_version); 5143 else 5144 rdev = md_import_device(dev, -1, -1); 5145 if (IS_ERR(rdev)) { 5146 printk(KERN_WARNING 5147 "md: md_import_device returned %ld\n", 5148 PTR_ERR(rdev)); 5149 return PTR_ERR(rdev); 5150 } 5151 /* set save_raid_disk if appropriate */ 5152 if (!mddev->persistent) { 5153 if (info->state & (1<<MD_DISK_SYNC) && 5154 info->raid_disk < mddev->raid_disks) 5155 rdev->raid_disk = info->raid_disk; 5156 else 5157 rdev->raid_disk = -1; 5158 } else 5159 super_types[mddev->major_version]. 5160 validate_super(mddev, rdev); 5161 rdev->saved_raid_disk = rdev->raid_disk; 5162 5163 clear_bit(In_sync, &rdev->flags); /* just to be sure */ 5164 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5165 set_bit(WriteMostly, &rdev->flags); 5166 else 5167 clear_bit(WriteMostly, &rdev->flags); 5168 5169 rdev->raid_disk = -1; 5170 err = bind_rdev_to_array(rdev, mddev); 5171 if (!err && !mddev->pers->hot_remove_disk) { 5172 /* If there is hot_add_disk but no hot_remove_disk 5173 * then added disks for geometry changes, 5174 * and should be added immediately. 5175 */ 5176 super_types[mddev->major_version]. 5177 validate_super(mddev, rdev); 5178 err = mddev->pers->hot_add_disk(mddev, rdev); 5179 if (err) 5180 unbind_rdev_from_array(rdev); 5181 } 5182 if (err) 5183 export_rdev(rdev); 5184 else 5185 sysfs_notify_dirent_safe(rdev->sysfs_state); 5186 5187 md_update_sb(mddev, 1); 5188 if (mddev->degraded) 5189 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 5190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5191 md_wakeup_thread(mddev->thread); 5192 return err; 5193 } 5194 5195 /* otherwise, add_new_disk is only allowed 5196 * for major_version==0 superblocks 5197 */ 5198 if (mddev->major_version != 0) { 5199 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", 5200 mdname(mddev)); 5201 return -EINVAL; 5202 } 5203 5204 if (!(info->state & (1<<MD_DISK_FAULTY))) { 5205 int err; 5206 rdev = md_import_device(dev, -1, 0); 5207 if (IS_ERR(rdev)) { 5208 printk(KERN_WARNING 5209 "md: error, md_import_device() returned %ld\n", 5210 PTR_ERR(rdev)); 5211 return PTR_ERR(rdev); 5212 } 5213 rdev->desc_nr = info->number; 5214 if (info->raid_disk < mddev->raid_disks) 5215 rdev->raid_disk = info->raid_disk; 5216 else 5217 rdev->raid_disk = -1; 5218 5219 if (rdev->raid_disk < mddev->raid_disks) 5220 if (info->state & (1<<MD_DISK_SYNC)) 5221 set_bit(In_sync, &rdev->flags); 5222 5223 if (info->state & (1<<MD_DISK_WRITEMOSTLY)) 5224 set_bit(WriteMostly, &rdev->flags); 5225 5226 if (!mddev->persistent) { 5227 printk(KERN_INFO "md: nonpersistent superblock ...\n"); 5228 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; 5229 } else 5230 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 5231 rdev->sectors = rdev->sb_start; 5232 5233 err = bind_rdev_to_array(rdev, mddev); 5234 if (err) { 5235 export_rdev(rdev); 5236 return err; 5237 } 5238 } 5239 5240 return 0; 5241 } 5242 5243 static int hot_remove_disk(mddev_t * mddev, dev_t dev) 5244 { 5245 char b[BDEVNAME_SIZE]; 5246 mdk_rdev_t *rdev; 5247 5248 rdev = find_rdev(mddev, dev); 5249 if (!rdev) 5250 return -ENXIO; 5251 5252 if (rdev->raid_disk >= 0) 5253 goto busy; 5254 5255 kick_rdev_from_array(rdev); 5256 md_update_sb(mddev, 1); 5257 md_new_event(mddev); 5258 5259 return 0; 5260 busy: 5261 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", 5262 bdevname(rdev->bdev,b), mdname(mddev)); 5263 return -EBUSY; 5264 } 5265 5266 static int hot_add_disk(mddev_t * mddev, dev_t dev) 5267 { 5268 char b[BDEVNAME_SIZE]; 5269 int err; 5270 mdk_rdev_t *rdev; 5271 5272 if (!mddev->pers) 5273 return -ENODEV; 5274 5275 if (mddev->major_version != 0) { 5276 printk(KERN_WARNING "%s: HOT_ADD may only be used with" 5277 " version-0 superblocks.\n", 5278 mdname(mddev)); 5279 return -EINVAL; 5280 } 5281 if (!mddev->pers->hot_add_disk) { 5282 printk(KERN_WARNING 5283 "%s: personality does not support diskops!\n", 5284 mdname(mddev)); 5285 return -EINVAL; 5286 } 5287 5288 rdev = md_import_device(dev, -1, 0); 5289 if (IS_ERR(rdev)) { 5290 printk(KERN_WARNING 5291 "md: error, md_import_device() returned %ld\n", 5292 PTR_ERR(rdev)); 5293 return -EINVAL; 5294 } 5295 5296 if (mddev->persistent) 5297 rdev->sb_start = calc_dev_sboffset(rdev->bdev); 5298 else 5299 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; 5300 5301 rdev->sectors = rdev->sb_start; 5302 5303 if (test_bit(Faulty, &rdev->flags)) { 5304 printk(KERN_WARNING 5305 "md: can not hot-add faulty %s disk to %s!\n", 5306 bdevname(rdev->bdev,b), mdname(mddev)); 5307 err = -EINVAL; 5308 goto abort_export; 5309 } 5310 clear_bit(In_sync, &rdev->flags); 5311 rdev->desc_nr = -1; 5312 rdev->saved_raid_disk = -1; 5313 err = bind_rdev_to_array(rdev, mddev); 5314 if (err) 5315 goto abort_export; 5316 5317 /* 5318 * The rest should better be atomic, we can have disk failures 5319 * noticed in interrupt contexts ... 5320 */ 5321 5322 rdev->raid_disk = -1; 5323 5324 md_update_sb(mddev, 1); 5325 5326 /* 5327 * Kick recovery, maybe this spare has to be added to the 5328 * array immediately. 5329 */ 5330 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5331 md_wakeup_thread(mddev->thread); 5332 md_new_event(mddev); 5333 return 0; 5334 5335 abort_export: 5336 export_rdev(rdev); 5337 return err; 5338 } 5339 5340 static int set_bitmap_file(mddev_t *mddev, int fd) 5341 { 5342 int err; 5343 5344 if (mddev->pers) { 5345 if (!mddev->pers->quiesce) 5346 return -EBUSY; 5347 if (mddev->recovery || mddev->sync_thread) 5348 return -EBUSY; 5349 /* we should be able to change the bitmap.. */ 5350 } 5351 5352 5353 if (fd >= 0) { 5354 if (mddev->bitmap) 5355 return -EEXIST; /* cannot add when bitmap is present */ 5356 mddev->bitmap_info.file = fget(fd); 5357 5358 if (mddev->bitmap_info.file == NULL) { 5359 printk(KERN_ERR "%s: error: failed to get bitmap file\n", 5360 mdname(mddev)); 5361 return -EBADF; 5362 } 5363 5364 err = deny_bitmap_write_access(mddev->bitmap_info.file); 5365 if (err) { 5366 printk(KERN_ERR "%s: error: bitmap file is already in use\n", 5367 mdname(mddev)); 5368 fput(mddev->bitmap_info.file); 5369 mddev->bitmap_info.file = NULL; 5370 return err; 5371 } 5372 mddev->bitmap_info.offset = 0; /* file overrides offset */ 5373 } else if (mddev->bitmap == NULL) 5374 return -ENOENT; /* cannot remove what isn't there */ 5375 err = 0; 5376 if (mddev->pers) { 5377 mddev->pers->quiesce(mddev, 1); 5378 if (fd >= 0) { 5379 err = bitmap_create(mddev); 5380 if (!err) 5381 err = bitmap_load(mddev); 5382 } 5383 if (fd < 0 || err) { 5384 bitmap_destroy(mddev); 5385 fd = -1; /* make sure to put the file */ 5386 } 5387 mddev->pers->quiesce(mddev, 0); 5388 } 5389 if (fd < 0) { 5390 if (mddev->bitmap_info.file) { 5391 restore_bitmap_write_access(mddev->bitmap_info.file); 5392 fput(mddev->bitmap_info.file); 5393 } 5394 mddev->bitmap_info.file = NULL; 5395 } 5396 5397 return err; 5398 } 5399 5400 /* 5401 * set_array_info is used two different ways 5402 * The original usage is when creating a new array. 5403 * In this usage, raid_disks is > 0 and it together with 5404 * level, size, not_persistent,layout,chunksize determine the 5405 * shape of the array. 5406 * This will always create an array with a type-0.90.0 superblock. 5407 * The newer usage is when assembling an array. 5408 * In this case raid_disks will be 0, and the major_version field is 5409 * use to determine which style super-blocks are to be found on the devices. 5410 * The minor and patch _version numbers are also kept incase the 5411 * super_block handler wishes to interpret them. 5412 */ 5413 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) 5414 { 5415 5416 if (info->raid_disks == 0) { 5417 /* just setting version number for superblock loading */ 5418 if (info->major_version < 0 || 5419 info->major_version >= ARRAY_SIZE(super_types) || 5420 super_types[info->major_version].name == NULL) { 5421 /* maybe try to auto-load a module? */ 5422 printk(KERN_INFO 5423 "md: superblock version %d not known\n", 5424 info->major_version); 5425 return -EINVAL; 5426 } 5427 mddev->major_version = info->major_version; 5428 mddev->minor_version = info->minor_version; 5429 mddev->patch_version = info->patch_version; 5430 mddev->persistent = !info->not_persistent; 5431 /* ensure mddev_put doesn't delete this now that there 5432 * is some minimal configuration. 5433 */ 5434 mddev->ctime = get_seconds(); 5435 return 0; 5436 } 5437 mddev->major_version = MD_MAJOR_VERSION; 5438 mddev->minor_version = MD_MINOR_VERSION; 5439 mddev->patch_version = MD_PATCHLEVEL_VERSION; 5440 mddev->ctime = get_seconds(); 5441 5442 mddev->level = info->level; 5443 mddev->clevel[0] = 0; 5444 mddev->dev_sectors = 2 * (sector_t)info->size; 5445 mddev->raid_disks = info->raid_disks; 5446 /* don't set md_minor, it is determined by which /dev/md* was 5447 * openned 5448 */ 5449 if (info->state & (1<<MD_SB_CLEAN)) 5450 mddev->recovery_cp = MaxSector; 5451 else 5452 mddev->recovery_cp = 0; 5453 mddev->persistent = ! info->not_persistent; 5454 mddev->external = 0; 5455 5456 mddev->layout = info->layout; 5457 mddev->chunk_sectors = info->chunk_size >> 9; 5458 5459 mddev->max_disks = MD_SB_DISKS; 5460 5461 if (mddev->persistent) 5462 mddev->flags = 0; 5463 set_bit(MD_CHANGE_DEVS, &mddev->flags); 5464 5465 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; 5466 mddev->bitmap_info.offset = 0; 5467 5468 mddev->reshape_position = MaxSector; 5469 5470 /* 5471 * Generate a 128 bit UUID 5472 */ 5473 get_random_bytes(mddev->uuid, 16); 5474 5475 mddev->new_level = mddev->level; 5476 mddev->new_chunk_sectors = mddev->chunk_sectors; 5477 mddev->new_layout = mddev->layout; 5478 mddev->delta_disks = 0; 5479 5480 return 0; 5481 } 5482 5483 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) 5484 { 5485 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); 5486 5487 if (mddev->external_size) 5488 return; 5489 5490 mddev->array_sectors = array_sectors; 5491 } 5492 EXPORT_SYMBOL(md_set_array_sectors); 5493 5494 static int update_size(mddev_t *mddev, sector_t num_sectors) 5495 { 5496 mdk_rdev_t *rdev; 5497 int rv; 5498 int fit = (num_sectors == 0); 5499 5500 if (mddev->pers->resize == NULL) 5501 return -EINVAL; 5502 /* The "num_sectors" is the number of sectors of each device that 5503 * is used. This can only make sense for arrays with redundancy. 5504 * linear and raid0 always use whatever space is available. We can only 5505 * consider changing this number if no resync or reconstruction is 5506 * happening, and if the new size is acceptable. It must fit before the 5507 * sb_start or, if that is <data_offset, it must fit before the size 5508 * of each device. If num_sectors is zero, we find the largest size 5509 * that fits. 5510 5511 */ 5512 if (mddev->sync_thread) 5513 return -EBUSY; 5514 if (mddev->bitmap) 5515 /* Sorry, cannot grow a bitmap yet, just remove it, 5516 * grow, and re-add. 5517 */ 5518 return -EBUSY; 5519 list_for_each_entry(rdev, &mddev->disks, same_set) { 5520 sector_t avail = rdev->sectors; 5521 5522 if (fit && (num_sectors == 0 || num_sectors > avail)) 5523 num_sectors = avail; 5524 if (avail < num_sectors) 5525 return -ENOSPC; 5526 } 5527 rv = mddev->pers->resize(mddev, num_sectors); 5528 if (!rv) 5529 revalidate_disk(mddev->gendisk); 5530 return rv; 5531 } 5532 5533 static int update_raid_disks(mddev_t *mddev, int raid_disks) 5534 { 5535 int rv; 5536 /* change the number of raid disks */ 5537 if (mddev->pers->check_reshape == NULL) 5538 return -EINVAL; 5539 if (raid_disks <= 0 || 5540 (mddev->max_disks && raid_disks >= mddev->max_disks)) 5541 return -EINVAL; 5542 if (mddev->sync_thread || mddev->reshape_position != MaxSector) 5543 return -EBUSY; 5544 mddev->delta_disks = raid_disks - mddev->raid_disks; 5545 5546 rv = mddev->pers->check_reshape(mddev); 5547 return rv; 5548 } 5549 5550 5551 /* 5552 * update_array_info is used to change the configuration of an 5553 * on-line array. 5554 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size 5555 * fields in the info are checked against the array. 5556 * Any differences that cannot be handled will cause an error. 5557 * Normally, only one change can be managed at a time. 5558 */ 5559 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) 5560 { 5561 int rv = 0; 5562 int cnt = 0; 5563 int state = 0; 5564 5565 /* calculate expected state,ignoring low bits */ 5566 if (mddev->bitmap && mddev->bitmap_info.offset) 5567 state |= (1 << MD_SB_BITMAP_PRESENT); 5568 5569 if (mddev->major_version != info->major_version || 5570 mddev->minor_version != info->minor_version || 5571 /* mddev->patch_version != info->patch_version || */ 5572 mddev->ctime != info->ctime || 5573 mddev->level != info->level || 5574 /* mddev->layout != info->layout || */ 5575 !mddev->persistent != info->not_persistent|| 5576 mddev->chunk_sectors != info->chunk_size >> 9 || 5577 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ 5578 ((state^info->state) & 0xfffffe00) 5579 ) 5580 return -EINVAL; 5581 /* Check there is only one change */ 5582 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5583 cnt++; 5584 if (mddev->raid_disks != info->raid_disks) 5585 cnt++; 5586 if (mddev->layout != info->layout) 5587 cnt++; 5588 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) 5589 cnt++; 5590 if (cnt == 0) 5591 return 0; 5592 if (cnt > 1) 5593 return -EINVAL; 5594 5595 if (mddev->layout != info->layout) { 5596 /* Change layout 5597 * we don't need to do anything at the md level, the 5598 * personality will take care of it all. 5599 */ 5600 if (mddev->pers->check_reshape == NULL) 5601 return -EINVAL; 5602 else { 5603 mddev->new_layout = info->layout; 5604 rv = mddev->pers->check_reshape(mddev); 5605 if (rv) 5606 mddev->new_layout = mddev->layout; 5607 return rv; 5608 } 5609 } 5610 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) 5611 rv = update_size(mddev, (sector_t)info->size * 2); 5612 5613 if (mddev->raid_disks != info->raid_disks) 5614 rv = update_raid_disks(mddev, info->raid_disks); 5615 5616 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { 5617 if (mddev->pers->quiesce == NULL) 5618 return -EINVAL; 5619 if (mddev->recovery || mddev->sync_thread) 5620 return -EBUSY; 5621 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { 5622 /* add the bitmap */ 5623 if (mddev->bitmap) 5624 return -EEXIST; 5625 if (mddev->bitmap_info.default_offset == 0) 5626 return -EINVAL; 5627 mddev->bitmap_info.offset = 5628 mddev->bitmap_info.default_offset; 5629 mddev->pers->quiesce(mddev, 1); 5630 rv = bitmap_create(mddev); 5631 if (!rv) 5632 rv = bitmap_load(mddev); 5633 if (rv) 5634 bitmap_destroy(mddev); 5635 mddev->pers->quiesce(mddev, 0); 5636 } else { 5637 /* remove the bitmap */ 5638 if (!mddev->bitmap) 5639 return -ENOENT; 5640 if (mddev->bitmap->file) 5641 return -EINVAL; 5642 mddev->pers->quiesce(mddev, 1); 5643 bitmap_destroy(mddev); 5644 mddev->pers->quiesce(mddev, 0); 5645 mddev->bitmap_info.offset = 0; 5646 } 5647 } 5648 md_update_sb(mddev, 1); 5649 return rv; 5650 } 5651 5652 static int set_disk_faulty(mddev_t *mddev, dev_t dev) 5653 { 5654 mdk_rdev_t *rdev; 5655 5656 if (mddev->pers == NULL) 5657 return -ENODEV; 5658 5659 rdev = find_rdev(mddev, dev); 5660 if (!rdev) 5661 return -ENODEV; 5662 5663 md_error(mddev, rdev); 5664 return 0; 5665 } 5666 5667 /* 5668 * We have a problem here : there is no easy way to give a CHS 5669 * virtual geometry. We currently pretend that we have a 2 heads 5670 * 4 sectors (with a BIG number of cylinders...). This drives 5671 * dosfs just mad... ;-) 5672 */ 5673 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) 5674 { 5675 mddev_t *mddev = bdev->bd_disk->private_data; 5676 5677 geo->heads = 2; 5678 geo->sectors = 4; 5679 geo->cylinders = mddev->array_sectors / 8; 5680 return 0; 5681 } 5682 5683 static int md_ioctl(struct block_device *bdev, fmode_t mode, 5684 unsigned int cmd, unsigned long arg) 5685 { 5686 int err = 0; 5687 void __user *argp = (void __user *)arg; 5688 mddev_t *mddev = NULL; 5689 int ro; 5690 5691 if (!capable(CAP_SYS_ADMIN)) 5692 return -EACCES; 5693 5694 /* 5695 * Commands dealing with the RAID driver but not any 5696 * particular array: 5697 */ 5698 switch (cmd) 5699 { 5700 case RAID_VERSION: 5701 err = get_version(argp); 5702 goto done; 5703 5704 case PRINT_RAID_DEBUG: 5705 err = 0; 5706 md_print_devices(); 5707 goto done; 5708 5709 #ifndef MODULE 5710 case RAID_AUTORUN: 5711 err = 0; 5712 autostart_arrays(arg); 5713 goto done; 5714 #endif 5715 default:; 5716 } 5717 5718 /* 5719 * Commands creating/starting a new array: 5720 */ 5721 5722 mddev = bdev->bd_disk->private_data; 5723 5724 if (!mddev) { 5725 BUG(); 5726 goto abort; 5727 } 5728 5729 err = mddev_lock(mddev); 5730 if (err) { 5731 printk(KERN_INFO 5732 "md: ioctl lock interrupted, reason %d, cmd %d\n", 5733 err, cmd); 5734 goto abort; 5735 } 5736 5737 switch (cmd) 5738 { 5739 case SET_ARRAY_INFO: 5740 { 5741 mdu_array_info_t info; 5742 if (!arg) 5743 memset(&info, 0, sizeof(info)); 5744 else if (copy_from_user(&info, argp, sizeof(info))) { 5745 err = -EFAULT; 5746 goto abort_unlock; 5747 } 5748 if (mddev->pers) { 5749 err = update_array_info(mddev, &info); 5750 if (err) { 5751 printk(KERN_WARNING "md: couldn't update" 5752 " array info. %d\n", err); 5753 goto abort_unlock; 5754 } 5755 goto done_unlock; 5756 } 5757 if (!list_empty(&mddev->disks)) { 5758 printk(KERN_WARNING 5759 "md: array %s already has disks!\n", 5760 mdname(mddev)); 5761 err = -EBUSY; 5762 goto abort_unlock; 5763 } 5764 if (mddev->raid_disks) { 5765 printk(KERN_WARNING 5766 "md: array %s already initialised!\n", 5767 mdname(mddev)); 5768 err = -EBUSY; 5769 goto abort_unlock; 5770 } 5771 err = set_array_info(mddev, &info); 5772 if (err) { 5773 printk(KERN_WARNING "md: couldn't set" 5774 " array info. %d\n", err); 5775 goto abort_unlock; 5776 } 5777 } 5778 goto done_unlock; 5779 5780 default:; 5781 } 5782 5783 /* 5784 * Commands querying/configuring an existing array: 5785 */ 5786 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, 5787 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ 5788 if ((!mddev->raid_disks && !mddev->external) 5789 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY 5790 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE 5791 && cmd != GET_BITMAP_FILE) { 5792 err = -ENODEV; 5793 goto abort_unlock; 5794 } 5795 5796 /* 5797 * Commands even a read-only array can execute: 5798 */ 5799 switch (cmd) 5800 { 5801 case GET_ARRAY_INFO: 5802 err = get_array_info(mddev, argp); 5803 goto done_unlock; 5804 5805 case GET_BITMAP_FILE: 5806 err = get_bitmap_file(mddev, argp); 5807 goto done_unlock; 5808 5809 case GET_DISK_INFO: 5810 err = get_disk_info(mddev, argp); 5811 goto done_unlock; 5812 5813 case RESTART_ARRAY_RW: 5814 err = restart_array(mddev); 5815 goto done_unlock; 5816 5817 case STOP_ARRAY: 5818 err = do_md_stop(mddev, 0, 1); 5819 goto done_unlock; 5820 5821 case STOP_ARRAY_RO: 5822 err = md_set_readonly(mddev, 1); 5823 goto done_unlock; 5824 5825 case BLKROSET: 5826 if (get_user(ro, (int __user *)(arg))) { 5827 err = -EFAULT; 5828 goto done_unlock; 5829 } 5830 err = -EINVAL; 5831 5832 /* if the bdev is going readonly the value of mddev->ro 5833 * does not matter, no writes are coming 5834 */ 5835 if (ro) 5836 goto done_unlock; 5837 5838 /* are we are already prepared for writes? */ 5839 if (mddev->ro != 1) 5840 goto done_unlock; 5841 5842 /* transitioning to readauto need only happen for 5843 * arrays that call md_write_start 5844 */ 5845 if (mddev->pers) { 5846 err = restart_array(mddev); 5847 if (err == 0) { 5848 mddev->ro = 2; 5849 set_disk_ro(mddev->gendisk, 0); 5850 } 5851 } 5852 goto done_unlock; 5853 } 5854 5855 /* 5856 * The remaining ioctls are changing the state of the 5857 * superblock, so we do not allow them on read-only arrays. 5858 * However non-MD ioctls (e.g. get-size) will still come through 5859 * here and hit the 'default' below, so only disallow 5860 * 'md' ioctls, and switch to rw mode if started auto-readonly. 5861 */ 5862 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { 5863 if (mddev->ro == 2) { 5864 mddev->ro = 0; 5865 sysfs_notify_dirent_safe(mddev->sysfs_state); 5866 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 5867 md_wakeup_thread(mddev->thread); 5868 } else { 5869 err = -EROFS; 5870 goto abort_unlock; 5871 } 5872 } 5873 5874 switch (cmd) 5875 { 5876 case ADD_NEW_DISK: 5877 { 5878 mdu_disk_info_t info; 5879 if (copy_from_user(&info, argp, sizeof(info))) 5880 err = -EFAULT; 5881 else 5882 err = add_new_disk(mddev, &info); 5883 goto done_unlock; 5884 } 5885 5886 case HOT_REMOVE_DISK: 5887 err = hot_remove_disk(mddev, new_decode_dev(arg)); 5888 goto done_unlock; 5889 5890 case HOT_ADD_DISK: 5891 err = hot_add_disk(mddev, new_decode_dev(arg)); 5892 goto done_unlock; 5893 5894 case SET_DISK_FAULTY: 5895 err = set_disk_faulty(mddev, new_decode_dev(arg)); 5896 goto done_unlock; 5897 5898 case RUN_ARRAY: 5899 err = do_md_run(mddev); 5900 goto done_unlock; 5901 5902 case SET_BITMAP_FILE: 5903 err = set_bitmap_file(mddev, (int)arg); 5904 goto done_unlock; 5905 5906 default: 5907 err = -EINVAL; 5908 goto abort_unlock; 5909 } 5910 5911 done_unlock: 5912 abort_unlock: 5913 if (mddev->hold_active == UNTIL_IOCTL && 5914 err != -EINVAL) 5915 mddev->hold_active = 0; 5916 mddev_unlock(mddev); 5917 5918 return err; 5919 done: 5920 if (err) 5921 MD_BUG(); 5922 abort: 5923 return err; 5924 } 5925 #ifdef CONFIG_COMPAT 5926 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, 5927 unsigned int cmd, unsigned long arg) 5928 { 5929 switch (cmd) { 5930 case HOT_REMOVE_DISK: 5931 case HOT_ADD_DISK: 5932 case SET_DISK_FAULTY: 5933 case SET_BITMAP_FILE: 5934 /* These take in integer arg, do not convert */ 5935 break; 5936 default: 5937 arg = (unsigned long)compat_ptr(arg); 5938 break; 5939 } 5940 5941 return md_ioctl(bdev, mode, cmd, arg); 5942 } 5943 #endif /* CONFIG_COMPAT */ 5944 5945 static int md_open(struct block_device *bdev, fmode_t mode) 5946 { 5947 /* 5948 * Succeed if we can lock the mddev, which confirms that 5949 * it isn't being stopped right now. 5950 */ 5951 mddev_t *mddev = mddev_find(bdev->bd_dev); 5952 int err; 5953 5954 lock_kernel(); 5955 if (mddev->gendisk != bdev->bd_disk) { 5956 /* we are racing with mddev_put which is discarding this 5957 * bd_disk. 5958 */ 5959 mddev_put(mddev); 5960 /* Wait until bdev->bd_disk is definitely gone */ 5961 flush_scheduled_work(); 5962 /* Then retry the open from the top */ 5963 unlock_kernel(); 5964 return -ERESTARTSYS; 5965 } 5966 BUG_ON(mddev != bdev->bd_disk->private_data); 5967 5968 if ((err = mutex_lock_interruptible(&mddev->open_mutex))) 5969 goto out; 5970 5971 err = 0; 5972 atomic_inc(&mddev->openers); 5973 mutex_unlock(&mddev->open_mutex); 5974 5975 check_disk_size_change(mddev->gendisk, bdev); 5976 out: 5977 unlock_kernel(); 5978 return err; 5979 } 5980 5981 static int md_release(struct gendisk *disk, fmode_t mode) 5982 { 5983 mddev_t *mddev = disk->private_data; 5984 5985 BUG_ON(!mddev); 5986 lock_kernel(); 5987 atomic_dec(&mddev->openers); 5988 mddev_put(mddev); 5989 unlock_kernel(); 5990 5991 return 0; 5992 } 5993 static const struct block_device_operations md_fops = 5994 { 5995 .owner = THIS_MODULE, 5996 .open = md_open, 5997 .release = md_release, 5998 .ioctl = md_ioctl, 5999 #ifdef CONFIG_COMPAT 6000 .compat_ioctl = md_compat_ioctl, 6001 #endif 6002 .getgeo = md_getgeo, 6003 }; 6004 6005 static int md_thread(void * arg) 6006 { 6007 mdk_thread_t *thread = arg; 6008 6009 /* 6010 * md_thread is a 'system-thread', it's priority should be very 6011 * high. We avoid resource deadlocks individually in each 6012 * raid personality. (RAID5 does preallocation) We also use RR and 6013 * the very same RT priority as kswapd, thus we will never get 6014 * into a priority inversion deadlock. 6015 * 6016 * we definitely have to have equal or higher priority than 6017 * bdflush, otherwise bdflush will deadlock if there are too 6018 * many dirty RAID5 blocks. 6019 */ 6020 6021 allow_signal(SIGKILL); 6022 while (!kthread_should_stop()) { 6023 6024 /* We need to wait INTERRUPTIBLE so that 6025 * we don't add to the load-average. 6026 * That means we need to be sure no signals are 6027 * pending 6028 */ 6029 if (signal_pending(current)) 6030 flush_signals(current); 6031 6032 wait_event_interruptible_timeout 6033 (thread->wqueue, 6034 test_bit(THREAD_WAKEUP, &thread->flags) 6035 || kthread_should_stop(), 6036 thread->timeout); 6037 6038 clear_bit(THREAD_WAKEUP, &thread->flags); 6039 6040 thread->run(thread->mddev); 6041 } 6042 6043 return 0; 6044 } 6045 6046 void md_wakeup_thread(mdk_thread_t *thread) 6047 { 6048 if (thread) { 6049 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); 6050 set_bit(THREAD_WAKEUP, &thread->flags); 6051 wake_up(&thread->wqueue); 6052 } 6053 } 6054 6055 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, 6056 const char *name) 6057 { 6058 mdk_thread_t *thread; 6059 6060 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); 6061 if (!thread) 6062 return NULL; 6063 6064 init_waitqueue_head(&thread->wqueue); 6065 6066 thread->run = run; 6067 thread->mddev = mddev; 6068 thread->timeout = MAX_SCHEDULE_TIMEOUT; 6069 thread->tsk = kthread_run(md_thread, thread, 6070 "%s_%s", 6071 mdname(thread->mddev), 6072 name ?: mddev->pers->name); 6073 if (IS_ERR(thread->tsk)) { 6074 kfree(thread); 6075 return NULL; 6076 } 6077 return thread; 6078 } 6079 6080 void md_unregister_thread(mdk_thread_t *thread) 6081 { 6082 if (!thread) 6083 return; 6084 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); 6085 6086 kthread_stop(thread->tsk); 6087 kfree(thread); 6088 } 6089 6090 void md_error(mddev_t *mddev, mdk_rdev_t *rdev) 6091 { 6092 if (!mddev) { 6093 MD_BUG(); 6094 return; 6095 } 6096 6097 if (!rdev || test_bit(Faulty, &rdev->flags)) 6098 return; 6099 6100 if (mddev->external) 6101 set_bit(Blocked, &rdev->flags); 6102 /* 6103 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", 6104 mdname(mddev), 6105 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), 6106 __builtin_return_address(0),__builtin_return_address(1), 6107 __builtin_return_address(2),__builtin_return_address(3)); 6108 */ 6109 if (!mddev->pers) 6110 return; 6111 if (!mddev->pers->error_handler) 6112 return; 6113 mddev->pers->error_handler(mddev,rdev); 6114 if (mddev->degraded) 6115 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 6116 sysfs_notify_dirent_safe(rdev->sysfs_state); 6117 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6118 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6119 md_wakeup_thread(mddev->thread); 6120 if (mddev->event_work.func) 6121 schedule_work(&mddev->event_work); 6122 md_new_event_inintr(mddev); 6123 } 6124 6125 /* seq_file implementation /proc/mdstat */ 6126 6127 static void status_unused(struct seq_file *seq) 6128 { 6129 int i = 0; 6130 mdk_rdev_t *rdev; 6131 6132 seq_printf(seq, "unused devices: "); 6133 6134 list_for_each_entry(rdev, &pending_raid_disks, same_set) { 6135 char b[BDEVNAME_SIZE]; 6136 i++; 6137 seq_printf(seq, "%s ", 6138 bdevname(rdev->bdev,b)); 6139 } 6140 if (!i) 6141 seq_printf(seq, "<none>"); 6142 6143 seq_printf(seq, "\n"); 6144 } 6145 6146 6147 static void status_resync(struct seq_file *seq, mddev_t * mddev) 6148 { 6149 sector_t max_sectors, resync, res; 6150 unsigned long dt, db; 6151 sector_t rt; 6152 int scale; 6153 unsigned int per_milli; 6154 6155 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); 6156 6157 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 6158 max_sectors = mddev->resync_max_sectors; 6159 else 6160 max_sectors = mddev->dev_sectors; 6161 6162 /* 6163 * Should not happen. 6164 */ 6165 if (!max_sectors) { 6166 MD_BUG(); 6167 return; 6168 } 6169 /* Pick 'scale' such that (resync>>scale)*1000 will fit 6170 * in a sector_t, and (max_sectors>>scale) will fit in a 6171 * u32, as those are the requirements for sector_div. 6172 * Thus 'scale' must be at least 10 6173 */ 6174 scale = 10; 6175 if (sizeof(sector_t) > sizeof(unsigned long)) { 6176 while ( max_sectors/2 > (1ULL<<(scale+32))) 6177 scale++; 6178 } 6179 res = (resync>>scale)*1000; 6180 sector_div(res, (u32)((max_sectors>>scale)+1)); 6181 6182 per_milli = res; 6183 { 6184 int i, x = per_milli/50, y = 20-x; 6185 seq_printf(seq, "["); 6186 for (i = 0; i < x; i++) 6187 seq_printf(seq, "="); 6188 seq_printf(seq, ">"); 6189 for (i = 0; i < y; i++) 6190 seq_printf(seq, "."); 6191 seq_printf(seq, "] "); 6192 } 6193 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", 6194 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? 6195 "reshape" : 6196 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? 6197 "check" : 6198 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? 6199 "resync" : "recovery"))), 6200 per_milli/10, per_milli % 10, 6201 (unsigned long long) resync/2, 6202 (unsigned long long) max_sectors/2); 6203 6204 /* 6205 * dt: time from mark until now 6206 * db: blocks written from mark until now 6207 * rt: remaining time 6208 * 6209 * rt is a sector_t, so could be 32bit or 64bit. 6210 * So we divide before multiply in case it is 32bit and close 6211 * to the limit. 6212 * We scale the divisor (db) by 32 to avoid loosing precision 6213 * near the end of resync when the number of remaining sectors 6214 * is close to 'db'. 6215 * We then divide rt by 32 after multiplying by db to compensate. 6216 * The '+1' avoids division by zero if db is very small. 6217 */ 6218 dt = ((jiffies - mddev->resync_mark) / HZ); 6219 if (!dt) dt++; 6220 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) 6221 - mddev->resync_mark_cnt; 6222 6223 rt = max_sectors - resync; /* number of remaining sectors */ 6224 sector_div(rt, db/32+1); 6225 rt *= dt; 6226 rt >>= 5; 6227 6228 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, 6229 ((unsigned long)rt % 60)/6); 6230 6231 seq_printf(seq, " speed=%ldK/sec", db/2/dt); 6232 } 6233 6234 static void *md_seq_start(struct seq_file *seq, loff_t *pos) 6235 { 6236 struct list_head *tmp; 6237 loff_t l = *pos; 6238 mddev_t *mddev; 6239 6240 if (l >= 0x10000) 6241 return NULL; 6242 if (!l--) 6243 /* header */ 6244 return (void*)1; 6245 6246 spin_lock(&all_mddevs_lock); 6247 list_for_each(tmp,&all_mddevs) 6248 if (!l--) { 6249 mddev = list_entry(tmp, mddev_t, all_mddevs); 6250 mddev_get(mddev); 6251 spin_unlock(&all_mddevs_lock); 6252 return mddev; 6253 } 6254 spin_unlock(&all_mddevs_lock); 6255 if (!l--) 6256 return (void*)2;/* tail */ 6257 return NULL; 6258 } 6259 6260 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) 6261 { 6262 struct list_head *tmp; 6263 mddev_t *next_mddev, *mddev = v; 6264 6265 ++*pos; 6266 if (v == (void*)2) 6267 return NULL; 6268 6269 spin_lock(&all_mddevs_lock); 6270 if (v == (void*)1) 6271 tmp = all_mddevs.next; 6272 else 6273 tmp = mddev->all_mddevs.next; 6274 if (tmp != &all_mddevs) 6275 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); 6276 else { 6277 next_mddev = (void*)2; 6278 *pos = 0x10000; 6279 } 6280 spin_unlock(&all_mddevs_lock); 6281 6282 if (v != (void*)1) 6283 mddev_put(mddev); 6284 return next_mddev; 6285 6286 } 6287 6288 static void md_seq_stop(struct seq_file *seq, void *v) 6289 { 6290 mddev_t *mddev = v; 6291 6292 if (mddev && v != (void*)1 && v != (void*)2) 6293 mddev_put(mddev); 6294 } 6295 6296 struct mdstat_info { 6297 int event; 6298 }; 6299 6300 static int md_seq_show(struct seq_file *seq, void *v) 6301 { 6302 mddev_t *mddev = v; 6303 sector_t sectors; 6304 mdk_rdev_t *rdev; 6305 struct mdstat_info *mi = seq->private; 6306 struct bitmap *bitmap; 6307 6308 if (v == (void*)1) { 6309 struct mdk_personality *pers; 6310 seq_printf(seq, "Personalities : "); 6311 spin_lock(&pers_lock); 6312 list_for_each_entry(pers, &pers_list, list) 6313 seq_printf(seq, "[%s] ", pers->name); 6314 6315 spin_unlock(&pers_lock); 6316 seq_printf(seq, "\n"); 6317 mi->event = atomic_read(&md_event_count); 6318 return 0; 6319 } 6320 if (v == (void*)2) { 6321 status_unused(seq); 6322 return 0; 6323 } 6324 6325 if (mddev_lock(mddev) < 0) 6326 return -EINTR; 6327 6328 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { 6329 seq_printf(seq, "%s : %sactive", mdname(mddev), 6330 mddev->pers ? "" : "in"); 6331 if (mddev->pers) { 6332 if (mddev->ro==1) 6333 seq_printf(seq, " (read-only)"); 6334 if (mddev->ro==2) 6335 seq_printf(seq, " (auto-read-only)"); 6336 seq_printf(seq, " %s", mddev->pers->name); 6337 } 6338 6339 sectors = 0; 6340 list_for_each_entry(rdev, &mddev->disks, same_set) { 6341 char b[BDEVNAME_SIZE]; 6342 seq_printf(seq, " %s[%d]", 6343 bdevname(rdev->bdev,b), rdev->desc_nr); 6344 if (test_bit(WriteMostly, &rdev->flags)) 6345 seq_printf(seq, "(W)"); 6346 if (test_bit(Faulty, &rdev->flags)) { 6347 seq_printf(seq, "(F)"); 6348 continue; 6349 } else if (rdev->raid_disk < 0) 6350 seq_printf(seq, "(S)"); /* spare */ 6351 sectors += rdev->sectors; 6352 } 6353 6354 if (!list_empty(&mddev->disks)) { 6355 if (mddev->pers) 6356 seq_printf(seq, "\n %llu blocks", 6357 (unsigned long long) 6358 mddev->array_sectors / 2); 6359 else 6360 seq_printf(seq, "\n %llu blocks", 6361 (unsigned long long)sectors / 2); 6362 } 6363 if (mddev->persistent) { 6364 if (mddev->major_version != 0 || 6365 mddev->minor_version != 90) { 6366 seq_printf(seq," super %d.%d", 6367 mddev->major_version, 6368 mddev->minor_version); 6369 } 6370 } else if (mddev->external) 6371 seq_printf(seq, " super external:%s", 6372 mddev->metadata_type); 6373 else 6374 seq_printf(seq, " super non-persistent"); 6375 6376 if (mddev->pers) { 6377 mddev->pers->status(seq, mddev); 6378 seq_printf(seq, "\n "); 6379 if (mddev->pers->sync_request) { 6380 if (mddev->curr_resync > 2) { 6381 status_resync(seq, mddev); 6382 seq_printf(seq, "\n "); 6383 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) 6384 seq_printf(seq, "\tresync=DELAYED\n "); 6385 else if (mddev->recovery_cp < MaxSector) 6386 seq_printf(seq, "\tresync=PENDING\n "); 6387 } 6388 } else 6389 seq_printf(seq, "\n "); 6390 6391 if ((bitmap = mddev->bitmap)) { 6392 unsigned long chunk_kb; 6393 unsigned long flags; 6394 spin_lock_irqsave(&bitmap->lock, flags); 6395 chunk_kb = mddev->bitmap_info.chunksize >> 10; 6396 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 6397 "%lu%s chunk", 6398 bitmap->pages - bitmap->missing_pages, 6399 bitmap->pages, 6400 (bitmap->pages - bitmap->missing_pages) 6401 << (PAGE_SHIFT - 10), 6402 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, 6403 chunk_kb ? "KB" : "B"); 6404 if (bitmap->file) { 6405 seq_printf(seq, ", file: "); 6406 seq_path(seq, &bitmap->file->f_path, " \t\n"); 6407 } 6408 6409 seq_printf(seq, "\n"); 6410 spin_unlock_irqrestore(&bitmap->lock, flags); 6411 } 6412 6413 seq_printf(seq, "\n"); 6414 } 6415 mddev_unlock(mddev); 6416 6417 return 0; 6418 } 6419 6420 static const struct seq_operations md_seq_ops = { 6421 .start = md_seq_start, 6422 .next = md_seq_next, 6423 .stop = md_seq_stop, 6424 .show = md_seq_show, 6425 }; 6426 6427 static int md_seq_open(struct inode *inode, struct file *file) 6428 { 6429 int error; 6430 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); 6431 if (mi == NULL) 6432 return -ENOMEM; 6433 6434 error = seq_open(file, &md_seq_ops); 6435 if (error) 6436 kfree(mi); 6437 else { 6438 struct seq_file *p = file->private_data; 6439 p->private = mi; 6440 mi->event = atomic_read(&md_event_count); 6441 } 6442 return error; 6443 } 6444 6445 static unsigned int mdstat_poll(struct file *filp, poll_table *wait) 6446 { 6447 struct seq_file *m = filp->private_data; 6448 struct mdstat_info *mi = m->private; 6449 int mask; 6450 6451 poll_wait(filp, &md_event_waiters, wait); 6452 6453 /* always allow read */ 6454 mask = POLLIN | POLLRDNORM; 6455 6456 if (mi->event != atomic_read(&md_event_count)) 6457 mask |= POLLERR | POLLPRI; 6458 return mask; 6459 } 6460 6461 static const struct file_operations md_seq_fops = { 6462 .owner = THIS_MODULE, 6463 .open = md_seq_open, 6464 .read = seq_read, 6465 .llseek = seq_lseek, 6466 .release = seq_release_private, 6467 .poll = mdstat_poll, 6468 }; 6469 6470 int register_md_personality(struct mdk_personality *p) 6471 { 6472 spin_lock(&pers_lock); 6473 list_add_tail(&p->list, &pers_list); 6474 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); 6475 spin_unlock(&pers_lock); 6476 return 0; 6477 } 6478 6479 int unregister_md_personality(struct mdk_personality *p) 6480 { 6481 printk(KERN_INFO "md: %s personality unregistered\n", p->name); 6482 spin_lock(&pers_lock); 6483 list_del_init(&p->list); 6484 spin_unlock(&pers_lock); 6485 return 0; 6486 } 6487 6488 static int is_mddev_idle(mddev_t *mddev, int init) 6489 { 6490 mdk_rdev_t * rdev; 6491 int idle; 6492 int curr_events; 6493 6494 idle = 1; 6495 rcu_read_lock(); 6496 rdev_for_each_rcu(rdev, mddev) { 6497 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; 6498 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 6499 (int)part_stat_read(&disk->part0, sectors[1]) - 6500 atomic_read(&disk->sync_io); 6501 /* sync IO will cause sync_io to increase before the disk_stats 6502 * as sync_io is counted when a request starts, and 6503 * disk_stats is counted when it completes. 6504 * So resync activity will cause curr_events to be smaller than 6505 * when there was no such activity. 6506 * non-sync IO will cause disk_stat to increase without 6507 * increasing sync_io so curr_events will (eventually) 6508 * be larger than it was before. Once it becomes 6509 * substantially larger, the test below will cause 6510 * the array to appear non-idle, and resync will slow 6511 * down. 6512 * If there is a lot of outstanding resync activity when 6513 * we set last_event to curr_events, then all that activity 6514 * completing might cause the array to appear non-idle 6515 * and resync will be slowed down even though there might 6516 * not have been non-resync activity. This will only 6517 * happen once though. 'last_events' will soon reflect 6518 * the state where there is little or no outstanding 6519 * resync requests, and further resync activity will 6520 * always make curr_events less than last_events. 6521 * 6522 */ 6523 if (init || curr_events - rdev->last_events > 64) { 6524 rdev->last_events = curr_events; 6525 idle = 0; 6526 } 6527 } 6528 rcu_read_unlock(); 6529 return idle; 6530 } 6531 6532 void md_done_sync(mddev_t *mddev, int blocks, int ok) 6533 { 6534 /* another "blocks" (512byte) blocks have been synced */ 6535 atomic_sub(blocks, &mddev->recovery_active); 6536 wake_up(&mddev->recovery_wait); 6537 if (!ok) { 6538 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6539 md_wakeup_thread(mddev->thread); 6540 // stop recovery, signal do_sync .... 6541 } 6542 } 6543 6544 6545 /* md_write_start(mddev, bi) 6546 * If we need to update some array metadata (e.g. 'active' flag 6547 * in superblock) before writing, schedule a superblock update 6548 * and wait for it to complete. 6549 */ 6550 void md_write_start(mddev_t *mddev, struct bio *bi) 6551 { 6552 int did_change = 0; 6553 if (bio_data_dir(bi) != WRITE) 6554 return; 6555 6556 BUG_ON(mddev->ro == 1); 6557 if (mddev->ro == 2) { 6558 /* need to switch to read/write */ 6559 mddev->ro = 0; 6560 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 6561 md_wakeup_thread(mddev->thread); 6562 md_wakeup_thread(mddev->sync_thread); 6563 did_change = 1; 6564 } 6565 atomic_inc(&mddev->writes_pending); 6566 if (mddev->safemode == 1) 6567 mddev->safemode = 0; 6568 if (mddev->in_sync) { 6569 spin_lock_irq(&mddev->write_lock); 6570 if (mddev->in_sync) { 6571 mddev->in_sync = 0; 6572 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6573 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6574 md_wakeup_thread(mddev->thread); 6575 did_change = 1; 6576 } 6577 spin_unlock_irq(&mddev->write_lock); 6578 } 6579 if (did_change) 6580 sysfs_notify_dirent_safe(mddev->sysfs_state); 6581 wait_event(mddev->sb_wait, 6582 !test_bit(MD_CHANGE_PENDING, &mddev->flags)); 6583 } 6584 6585 void md_write_end(mddev_t *mddev) 6586 { 6587 if (atomic_dec_and_test(&mddev->writes_pending)) { 6588 if (mddev->safemode == 2) 6589 md_wakeup_thread(mddev->thread); 6590 else if (mddev->safemode_delay) 6591 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); 6592 } 6593 } 6594 6595 /* md_allow_write(mddev) 6596 * Calling this ensures that the array is marked 'active' so that writes 6597 * may proceed without blocking. It is important to call this before 6598 * attempting a GFP_KERNEL allocation while holding the mddev lock. 6599 * Must be called with mddev_lock held. 6600 * 6601 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock 6602 * is dropped, so return -EAGAIN after notifying userspace. 6603 */ 6604 int md_allow_write(mddev_t *mddev) 6605 { 6606 if (!mddev->pers) 6607 return 0; 6608 if (mddev->ro) 6609 return 0; 6610 if (!mddev->pers->sync_request) 6611 return 0; 6612 6613 spin_lock_irq(&mddev->write_lock); 6614 if (mddev->in_sync) { 6615 mddev->in_sync = 0; 6616 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6617 set_bit(MD_CHANGE_PENDING, &mddev->flags); 6618 if (mddev->safemode_delay && 6619 mddev->safemode == 0) 6620 mddev->safemode = 1; 6621 spin_unlock_irq(&mddev->write_lock); 6622 md_update_sb(mddev, 0); 6623 sysfs_notify_dirent_safe(mddev->sysfs_state); 6624 } else 6625 spin_unlock_irq(&mddev->write_lock); 6626 6627 if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) 6628 return -EAGAIN; 6629 else 6630 return 0; 6631 } 6632 EXPORT_SYMBOL_GPL(md_allow_write); 6633 6634 void md_unplug(mddev_t *mddev) 6635 { 6636 if (mddev->queue) 6637 blk_unplug(mddev->queue); 6638 if (mddev->plug) 6639 mddev->plug->unplug_fn(mddev->plug); 6640 } 6641 6642 #define SYNC_MARKS 10 6643 #define SYNC_MARK_STEP (3*HZ) 6644 void md_do_sync(mddev_t *mddev) 6645 { 6646 mddev_t *mddev2; 6647 unsigned int currspeed = 0, 6648 window; 6649 sector_t max_sectors,j, io_sectors; 6650 unsigned long mark[SYNC_MARKS]; 6651 sector_t mark_cnt[SYNC_MARKS]; 6652 int last_mark,m; 6653 struct list_head *tmp; 6654 sector_t last_check; 6655 int skipped = 0; 6656 mdk_rdev_t *rdev; 6657 char *desc; 6658 6659 /* just incase thread restarts... */ 6660 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) 6661 return; 6662 if (mddev->ro) /* never try to sync a read-only array */ 6663 return; 6664 6665 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6666 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) 6667 desc = "data-check"; 6668 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6669 desc = "requested-resync"; 6670 else 6671 desc = "resync"; 6672 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6673 desc = "reshape"; 6674 else 6675 desc = "recovery"; 6676 6677 /* we overload curr_resync somewhat here. 6678 * 0 == not engaged in resync at all 6679 * 2 == checking that there is no conflict with another sync 6680 * 1 == like 2, but have yielded to allow conflicting resync to 6681 * commense 6682 * other == active in resync - this many blocks 6683 * 6684 * Before starting a resync we must have set curr_resync to 6685 * 2, and then checked that every "conflicting" array has curr_resync 6686 * less than ours. When we find one that is the same or higher 6687 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync 6688 * to 1 if we choose to yield (based arbitrarily on address of mddev structure). 6689 * This will mean we have to start checking from the beginning again. 6690 * 6691 */ 6692 6693 do { 6694 mddev->curr_resync = 2; 6695 6696 try_again: 6697 if (kthread_should_stop()) 6698 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6699 6700 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6701 goto skip; 6702 for_each_mddev(mddev2, tmp) { 6703 if (mddev2 == mddev) 6704 continue; 6705 if (!mddev->parallel_resync 6706 && mddev2->curr_resync 6707 && match_mddev_units(mddev, mddev2)) { 6708 DEFINE_WAIT(wq); 6709 if (mddev < mddev2 && mddev->curr_resync == 2) { 6710 /* arbitrarily yield */ 6711 mddev->curr_resync = 1; 6712 wake_up(&resync_wait); 6713 } 6714 if (mddev > mddev2 && mddev->curr_resync == 1) 6715 /* no need to wait here, we can wait the next 6716 * time 'round when curr_resync == 2 6717 */ 6718 continue; 6719 /* We need to wait 'interruptible' so as not to 6720 * contribute to the load average, and not to 6721 * be caught by 'softlockup' 6722 */ 6723 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); 6724 if (!kthread_should_stop() && 6725 mddev2->curr_resync >= mddev->curr_resync) { 6726 printk(KERN_INFO "md: delaying %s of %s" 6727 " until %s has finished (they" 6728 " share one or more physical units)\n", 6729 desc, mdname(mddev), mdname(mddev2)); 6730 mddev_put(mddev2); 6731 if (signal_pending(current)) 6732 flush_signals(current); 6733 schedule(); 6734 finish_wait(&resync_wait, &wq); 6735 goto try_again; 6736 } 6737 finish_wait(&resync_wait, &wq); 6738 } 6739 } 6740 } while (mddev->curr_resync < 2); 6741 6742 j = 0; 6743 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6744 /* resync follows the size requested by the personality, 6745 * which defaults to physical size, but can be virtual size 6746 */ 6747 max_sectors = mddev->resync_max_sectors; 6748 mddev->resync_mismatches = 0; 6749 /* we don't use the checkpoint if there's a bitmap */ 6750 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6751 j = mddev->resync_min; 6752 else if (!mddev->bitmap) 6753 j = mddev->recovery_cp; 6754 6755 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) 6756 max_sectors = mddev->dev_sectors; 6757 else { 6758 /* recovery follows the physical size of devices */ 6759 max_sectors = mddev->dev_sectors; 6760 j = MaxSector; 6761 rcu_read_lock(); 6762 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 6763 if (rdev->raid_disk >= 0 && 6764 !test_bit(Faulty, &rdev->flags) && 6765 !test_bit(In_sync, &rdev->flags) && 6766 rdev->recovery_offset < j) 6767 j = rdev->recovery_offset; 6768 rcu_read_unlock(); 6769 } 6770 6771 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); 6772 printk(KERN_INFO "md: minimum _guaranteed_ speed:" 6773 " %d KB/sec/disk.\n", speed_min(mddev)); 6774 printk(KERN_INFO "md: using maximum available idle IO bandwidth " 6775 "(but not more than %d KB/sec) for %s.\n", 6776 speed_max(mddev), desc); 6777 6778 is_mddev_idle(mddev, 1); /* this initializes IO event counters */ 6779 6780 io_sectors = 0; 6781 for (m = 0; m < SYNC_MARKS; m++) { 6782 mark[m] = jiffies; 6783 mark_cnt[m] = io_sectors; 6784 } 6785 last_mark = 0; 6786 mddev->resync_mark = mark[last_mark]; 6787 mddev->resync_mark_cnt = mark_cnt[last_mark]; 6788 6789 /* 6790 * Tune reconstruction: 6791 */ 6792 window = 32*(PAGE_SIZE/512); 6793 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", 6794 window/2,(unsigned long long) max_sectors/2); 6795 6796 atomic_set(&mddev->recovery_active, 0); 6797 last_check = 0; 6798 6799 if (j>2) { 6800 printk(KERN_INFO 6801 "md: resuming %s of %s from checkpoint.\n", 6802 desc, mdname(mddev)); 6803 mddev->curr_resync = j; 6804 } 6805 mddev->curr_resync_completed = mddev->curr_resync; 6806 6807 while (j < max_sectors) { 6808 sector_t sectors; 6809 6810 skipped = 0; 6811 6812 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 6813 ((mddev->curr_resync > mddev->curr_resync_completed && 6814 (mddev->curr_resync - mddev->curr_resync_completed) 6815 > (max_sectors >> 4)) || 6816 (j - mddev->curr_resync_completed)*2 6817 >= mddev->resync_max - mddev->curr_resync_completed 6818 )) { 6819 /* time to update curr_resync_completed */ 6820 md_unplug(mddev); 6821 wait_event(mddev->recovery_wait, 6822 atomic_read(&mddev->recovery_active) == 0); 6823 mddev->curr_resync_completed = 6824 mddev->curr_resync; 6825 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 6826 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 6827 } 6828 6829 while (j >= mddev->resync_max && !kthread_should_stop()) { 6830 /* As this condition is controlled by user-space, 6831 * we can block indefinitely, so use '_interruptible' 6832 * to avoid triggering warnings. 6833 */ 6834 flush_signals(current); /* just in case */ 6835 wait_event_interruptible(mddev->recovery_wait, 6836 mddev->resync_max > j 6837 || kthread_should_stop()); 6838 } 6839 6840 if (kthread_should_stop()) 6841 goto interrupted; 6842 6843 sectors = mddev->pers->sync_request(mddev, j, &skipped, 6844 currspeed < speed_min(mddev)); 6845 if (sectors == 0) { 6846 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6847 goto out; 6848 } 6849 6850 if (!skipped) { /* actual IO requested */ 6851 io_sectors += sectors; 6852 atomic_add(sectors, &mddev->recovery_active); 6853 } 6854 6855 j += sectors; 6856 if (j>1) mddev->curr_resync = j; 6857 mddev->curr_mark_cnt = io_sectors; 6858 if (last_check == 0) 6859 /* this is the earliers that rebuilt will be 6860 * visible in /proc/mdstat 6861 */ 6862 md_new_event(mddev); 6863 6864 if (last_check + window > io_sectors || j == max_sectors) 6865 continue; 6866 6867 last_check = io_sectors; 6868 6869 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6870 break; 6871 6872 repeat: 6873 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { 6874 /* step marks */ 6875 int next = (last_mark+1) % SYNC_MARKS; 6876 6877 mddev->resync_mark = mark[next]; 6878 mddev->resync_mark_cnt = mark_cnt[next]; 6879 mark[next] = jiffies; 6880 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); 6881 last_mark = next; 6882 } 6883 6884 6885 if (kthread_should_stop()) 6886 goto interrupted; 6887 6888 6889 /* 6890 * this loop exits only if either when we are slower than 6891 * the 'hard' speed limit, or the system was IO-idle for 6892 * a jiffy. 6893 * the system might be non-idle CPU-wise, but we only care 6894 * about not overloading the IO subsystem. (things like an 6895 * e2fsck being done on the RAID array should execute fast) 6896 */ 6897 md_unplug(mddev); 6898 cond_resched(); 6899 6900 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 6901 /((jiffies-mddev->resync_mark)/HZ +1) +1; 6902 6903 if (currspeed > speed_min(mddev)) { 6904 if ((currspeed > speed_max(mddev)) || 6905 !is_mddev_idle(mddev, 0)) { 6906 msleep(500); 6907 goto repeat; 6908 } 6909 } 6910 } 6911 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); 6912 /* 6913 * this also signals 'finished resyncing' to md_stop 6914 */ 6915 out: 6916 md_unplug(mddev); 6917 6918 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); 6919 6920 /* tell personality that we are finished */ 6921 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); 6922 6923 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && 6924 mddev->curr_resync > 2) { 6925 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { 6926 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 6927 if (mddev->curr_resync >= mddev->recovery_cp) { 6928 printk(KERN_INFO 6929 "md: checkpointing %s of %s.\n", 6930 desc, mdname(mddev)); 6931 mddev->recovery_cp = mddev->curr_resync; 6932 } 6933 } else 6934 mddev->recovery_cp = MaxSector; 6935 } else { 6936 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6937 mddev->curr_resync = MaxSector; 6938 rcu_read_lock(); 6939 list_for_each_entry_rcu(rdev, &mddev->disks, same_set) 6940 if (rdev->raid_disk >= 0 && 6941 mddev->delta_disks >= 0 && 6942 !test_bit(Faulty, &rdev->flags) && 6943 !test_bit(In_sync, &rdev->flags) && 6944 rdev->recovery_offset < mddev->curr_resync) 6945 rdev->recovery_offset = mddev->curr_resync; 6946 rcu_read_unlock(); 6947 } 6948 } 6949 set_bit(MD_CHANGE_DEVS, &mddev->flags); 6950 6951 skip: 6952 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { 6953 /* We completed so min/max setting can be forgotten if used. */ 6954 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6955 mddev->resync_min = 0; 6956 mddev->resync_max = MaxSector; 6957 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 6958 mddev->resync_min = mddev->curr_resync_completed; 6959 mddev->curr_resync = 0; 6960 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) 6961 mddev->curr_resync_completed = 0; 6962 sysfs_notify(&mddev->kobj, NULL, "sync_completed"); 6963 wake_up(&resync_wait); 6964 set_bit(MD_RECOVERY_DONE, &mddev->recovery); 6965 md_wakeup_thread(mddev->thread); 6966 return; 6967 6968 interrupted: 6969 /* 6970 * got a signal, exit. 6971 */ 6972 printk(KERN_INFO 6973 "md: md_do_sync() got signal ... exiting\n"); 6974 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 6975 goto out; 6976 6977 } 6978 EXPORT_SYMBOL_GPL(md_do_sync); 6979 6980 6981 static int remove_and_add_spares(mddev_t *mddev) 6982 { 6983 mdk_rdev_t *rdev; 6984 int spares = 0; 6985 6986 mddev->curr_resync_completed = 0; 6987 6988 list_for_each_entry(rdev, &mddev->disks, same_set) 6989 if (rdev->raid_disk >= 0 && 6990 !test_bit(Blocked, &rdev->flags) && 6991 (test_bit(Faulty, &rdev->flags) || 6992 ! test_bit(In_sync, &rdev->flags)) && 6993 atomic_read(&rdev->nr_pending)==0) { 6994 if (mddev->pers->hot_remove_disk( 6995 mddev, rdev->raid_disk)==0) { 6996 char nm[20]; 6997 sprintf(nm,"rd%d", rdev->raid_disk); 6998 sysfs_remove_link(&mddev->kobj, nm); 6999 rdev->raid_disk = -1; 7000 } 7001 } 7002 7003 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { 7004 list_for_each_entry(rdev, &mddev->disks, same_set) { 7005 if (rdev->raid_disk >= 0 && 7006 !test_bit(In_sync, &rdev->flags) && 7007 !test_bit(Blocked, &rdev->flags)) 7008 spares++; 7009 if (rdev->raid_disk < 0 7010 && !test_bit(Faulty, &rdev->flags)) { 7011 rdev->recovery_offset = 0; 7012 if (mddev->pers-> 7013 hot_add_disk(mddev, rdev) == 0) { 7014 char nm[20]; 7015 sprintf(nm, "rd%d", rdev->raid_disk); 7016 if (sysfs_create_link(&mddev->kobj, 7017 &rdev->kobj, nm)) 7018 /* failure here is OK */; 7019 spares++; 7020 md_new_event(mddev); 7021 set_bit(MD_CHANGE_DEVS, &mddev->flags); 7022 } else 7023 break; 7024 } 7025 } 7026 } 7027 return spares; 7028 } 7029 /* 7030 * This routine is regularly called by all per-raid-array threads to 7031 * deal with generic issues like resync and super-block update. 7032 * Raid personalities that don't have a thread (linear/raid0) do not 7033 * need this as they never do any recovery or update the superblock. 7034 * 7035 * It does not do any resync itself, but rather "forks" off other threads 7036 * to do that as needed. 7037 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in 7038 * "->recovery" and create a thread at ->sync_thread. 7039 * When the thread finishes it sets MD_RECOVERY_DONE 7040 * and wakeups up this thread which will reap the thread and finish up. 7041 * This thread also removes any faulty devices (with nr_pending == 0). 7042 * 7043 * The overall approach is: 7044 * 1/ if the superblock needs updating, update it. 7045 * 2/ If a recovery thread is running, don't do anything else. 7046 * 3/ If recovery has finished, clean up, possibly marking spares active. 7047 * 4/ If there are any faulty devices, remove them. 7048 * 5/ If array is degraded, try to add spares devices 7049 * 6/ If array has spares or is not in-sync, start a resync thread. 7050 */ 7051 void md_check_recovery(mddev_t *mddev) 7052 { 7053 mdk_rdev_t *rdev; 7054 7055 7056 if (mddev->bitmap) 7057 bitmap_daemon_work(mddev); 7058 7059 if (mddev->ro) 7060 return; 7061 7062 if (signal_pending(current)) { 7063 if (mddev->pers->sync_request && !mddev->external) { 7064 printk(KERN_INFO "md: %s in immediate safe mode\n", 7065 mdname(mddev)); 7066 mddev->safemode = 2; 7067 } 7068 flush_signals(current); 7069 } 7070 7071 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) 7072 return; 7073 if ( ! ( 7074 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || 7075 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || 7076 test_bit(MD_RECOVERY_DONE, &mddev->recovery) || 7077 (mddev->external == 0 && mddev->safemode == 1) || 7078 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) 7079 && !mddev->in_sync && mddev->recovery_cp == MaxSector) 7080 )) 7081 return; 7082 7083 if (mddev_trylock(mddev)) { 7084 int spares = 0; 7085 7086 if (mddev->ro) { 7087 /* Only thing we do on a ro array is remove 7088 * failed devices. 7089 */ 7090 remove_and_add_spares(mddev); 7091 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7092 goto unlock; 7093 } 7094 7095 if (!mddev->external) { 7096 int did_change = 0; 7097 spin_lock_irq(&mddev->write_lock); 7098 if (mddev->safemode && 7099 !atomic_read(&mddev->writes_pending) && 7100 !mddev->in_sync && 7101 mddev->recovery_cp == MaxSector) { 7102 mddev->in_sync = 1; 7103 did_change = 1; 7104 set_bit(MD_CHANGE_CLEAN, &mddev->flags); 7105 } 7106 if (mddev->safemode == 1) 7107 mddev->safemode = 0; 7108 spin_unlock_irq(&mddev->write_lock); 7109 if (did_change) 7110 sysfs_notify_dirent_safe(mddev->sysfs_state); 7111 } 7112 7113 if (mddev->flags) 7114 md_update_sb(mddev, 0); 7115 7116 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && 7117 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { 7118 /* resync/recovery still happening */ 7119 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7120 goto unlock; 7121 } 7122 if (mddev->sync_thread) { 7123 /* resync has finished, collect result */ 7124 md_unregister_thread(mddev->sync_thread); 7125 mddev->sync_thread = NULL; 7126 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && 7127 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 7128 /* success...*/ 7129 /* activate any spares */ 7130 if (mddev->pers->spare_active(mddev)) 7131 sysfs_notify(&mddev->kobj, NULL, 7132 "degraded"); 7133 } 7134 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && 7135 mddev->pers->finish_reshape) 7136 mddev->pers->finish_reshape(mddev); 7137 md_update_sb(mddev, 1); 7138 7139 /* if array is no-longer degraded, then any saved_raid_disk 7140 * information must be scrapped 7141 */ 7142 if (!mddev->degraded) 7143 list_for_each_entry(rdev, &mddev->disks, same_set) 7144 rdev->saved_raid_disk = -1; 7145 7146 mddev->recovery = 0; 7147 /* flag recovery needed just to double check */ 7148 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7149 sysfs_notify_dirent_safe(mddev->sysfs_action); 7150 md_new_event(mddev); 7151 goto unlock; 7152 } 7153 /* Set RUNNING before clearing NEEDED to avoid 7154 * any transients in the value of "sync_action". 7155 */ 7156 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7157 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 7158 /* Clear some bits that don't mean anything, but 7159 * might be left set 7160 */ 7161 clear_bit(MD_RECOVERY_INTR, &mddev->recovery); 7162 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); 7163 7164 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) 7165 goto unlock; 7166 /* no recovery is running. 7167 * remove any failed drives, then 7168 * add spares if possible. 7169 * Spare are also removed and re-added, to allow 7170 * the personality to fail the re-add. 7171 */ 7172 7173 if (mddev->reshape_position != MaxSector) { 7174 if (mddev->pers->check_reshape == NULL || 7175 mddev->pers->check_reshape(mddev) != 0) 7176 /* Cannot proceed */ 7177 goto unlock; 7178 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); 7179 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7180 } else if ((spares = remove_and_add_spares(mddev))) { 7181 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7182 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); 7183 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); 7184 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7185 } else if (mddev->recovery_cp < MaxSector) { 7186 set_bit(MD_RECOVERY_SYNC, &mddev->recovery); 7187 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 7188 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) 7189 /* nothing to be done ... */ 7190 goto unlock; 7191 7192 if (mddev->pers->sync_request) { 7193 if (spares && mddev->bitmap && ! mddev->bitmap->file) { 7194 /* We are adding a device or devices to an array 7195 * which has the bitmap stored on all devices. 7196 * So make sure all bitmap pages get written 7197 */ 7198 bitmap_write_all(mddev->bitmap); 7199 } 7200 mddev->sync_thread = md_register_thread(md_do_sync, 7201 mddev, 7202 "resync"); 7203 if (!mddev->sync_thread) { 7204 printk(KERN_ERR "%s: could not start resync" 7205 " thread...\n", 7206 mdname(mddev)); 7207 /* leave the spares where they are, it shouldn't hurt */ 7208 mddev->recovery = 0; 7209 } else 7210 md_wakeup_thread(mddev->sync_thread); 7211 sysfs_notify_dirent_safe(mddev->sysfs_action); 7212 md_new_event(mddev); 7213 } 7214 unlock: 7215 if (!mddev->sync_thread) { 7216 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); 7217 if (test_and_clear_bit(MD_RECOVERY_RECOVER, 7218 &mddev->recovery)) 7219 if (mddev->sysfs_action) 7220 sysfs_notify_dirent_safe(mddev->sysfs_action); 7221 } 7222 mddev_unlock(mddev); 7223 } 7224 } 7225 7226 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) 7227 { 7228 sysfs_notify_dirent_safe(rdev->sysfs_state); 7229 wait_event_timeout(rdev->blocked_wait, 7230 !test_bit(Blocked, &rdev->flags), 7231 msecs_to_jiffies(5000)); 7232 rdev_dec_pending(rdev, mddev); 7233 } 7234 EXPORT_SYMBOL(md_wait_for_blocked_rdev); 7235 7236 static int md_notify_reboot(struct notifier_block *this, 7237 unsigned long code, void *x) 7238 { 7239 struct list_head *tmp; 7240 mddev_t *mddev; 7241 7242 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { 7243 7244 printk(KERN_INFO "md: stopping all md devices.\n"); 7245 7246 for_each_mddev(mddev, tmp) 7247 if (mddev_trylock(mddev)) { 7248 /* Force a switch to readonly even array 7249 * appears to still be in use. Hence 7250 * the '100'. 7251 */ 7252 md_set_readonly(mddev, 100); 7253 mddev_unlock(mddev); 7254 } 7255 /* 7256 * certain more exotic SCSI devices are known to be 7257 * volatile wrt too early system reboots. While the 7258 * right place to handle this issue is the given 7259 * driver, we do want to have a safe RAID driver ... 7260 */ 7261 mdelay(1000*1); 7262 } 7263 return NOTIFY_DONE; 7264 } 7265 7266 static struct notifier_block md_notifier = { 7267 .notifier_call = md_notify_reboot, 7268 .next = NULL, 7269 .priority = INT_MAX, /* before any real devices */ 7270 }; 7271 7272 static void md_geninit(void) 7273 { 7274 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); 7275 7276 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); 7277 } 7278 7279 static int __init md_init(void) 7280 { 7281 if (register_blkdev(MD_MAJOR, "md")) 7282 return -1; 7283 if ((mdp_major=register_blkdev(0, "mdp"))<=0) { 7284 unregister_blkdev(MD_MAJOR, "md"); 7285 return -1; 7286 } 7287 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, 7288 md_probe, NULL, NULL); 7289 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, 7290 md_probe, NULL, NULL); 7291 7292 register_reboot_notifier(&md_notifier); 7293 raid_table_header = register_sysctl_table(raid_root_table); 7294 7295 md_geninit(); 7296 return 0; 7297 } 7298 7299 7300 #ifndef MODULE 7301 7302 /* 7303 * Searches all registered partitions for autorun RAID arrays 7304 * at boot time. 7305 */ 7306 7307 static LIST_HEAD(all_detected_devices); 7308 struct detected_devices_node { 7309 struct list_head list; 7310 dev_t dev; 7311 }; 7312 7313 void md_autodetect_dev(dev_t dev) 7314 { 7315 struct detected_devices_node *node_detected_dev; 7316 7317 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); 7318 if (node_detected_dev) { 7319 node_detected_dev->dev = dev; 7320 list_add_tail(&node_detected_dev->list, &all_detected_devices); 7321 } else { 7322 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" 7323 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); 7324 } 7325 } 7326 7327 7328 static void autostart_arrays(int part) 7329 { 7330 mdk_rdev_t *rdev; 7331 struct detected_devices_node *node_detected_dev; 7332 dev_t dev; 7333 int i_scanned, i_passed; 7334 7335 i_scanned = 0; 7336 i_passed = 0; 7337 7338 printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); 7339 7340 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { 7341 i_scanned++; 7342 node_detected_dev = list_entry(all_detected_devices.next, 7343 struct detected_devices_node, list); 7344 list_del(&node_detected_dev->list); 7345 dev = node_detected_dev->dev; 7346 kfree(node_detected_dev); 7347 rdev = md_import_device(dev,0, 90); 7348 if (IS_ERR(rdev)) 7349 continue; 7350 7351 if (test_bit(Faulty, &rdev->flags)) { 7352 MD_BUG(); 7353 continue; 7354 } 7355 set_bit(AutoDetected, &rdev->flags); 7356 list_add(&rdev->same_set, &pending_raid_disks); 7357 i_passed++; 7358 } 7359 7360 printk(KERN_INFO "md: Scanned %d and added %d devices.\n", 7361 i_scanned, i_passed); 7362 7363 autorun_devices(part); 7364 } 7365 7366 #endif /* !MODULE */ 7367 7368 static __exit void md_exit(void) 7369 { 7370 mddev_t *mddev; 7371 struct list_head *tmp; 7372 7373 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); 7374 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); 7375 7376 unregister_blkdev(MD_MAJOR,"md"); 7377 unregister_blkdev(mdp_major, "mdp"); 7378 unregister_reboot_notifier(&md_notifier); 7379 unregister_sysctl_table(raid_table_header); 7380 remove_proc_entry("mdstat", NULL); 7381 for_each_mddev(mddev, tmp) { 7382 export_array(mddev); 7383 mddev->hold_active = 0; 7384 } 7385 } 7386 7387 subsys_initcall(md_init); 7388 module_exit(md_exit) 7389 7390 static int get_ro(char *buffer, struct kernel_param *kp) 7391 { 7392 return sprintf(buffer, "%d", start_readonly); 7393 } 7394 static int set_ro(const char *val, struct kernel_param *kp) 7395 { 7396 char *e; 7397 int num = simple_strtoul(val, &e, 10); 7398 if (*val && (*e == '\0' || *e == '\n')) { 7399 start_readonly = num; 7400 return 0; 7401 } 7402 return -EINVAL; 7403 } 7404 7405 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); 7406 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); 7407 7408 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); 7409 7410 EXPORT_SYMBOL(register_md_personality); 7411 EXPORT_SYMBOL(unregister_md_personality); 7412 EXPORT_SYMBOL(md_error); 7413 EXPORT_SYMBOL(md_done_sync); 7414 EXPORT_SYMBOL(md_write_start); 7415 EXPORT_SYMBOL(md_write_end); 7416 EXPORT_SYMBOL(md_register_thread); 7417 EXPORT_SYMBOL(md_unregister_thread); 7418 EXPORT_SYMBOL(md_wakeup_thread); 7419 EXPORT_SYMBOL(md_check_recovery); 7420 MODULE_LICENSE("GPL"); 7421 MODULE_DESCRIPTION("MD RAID framework"); 7422 MODULE_ALIAS("md"); 7423 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); 7424