1/* 2 * Copyright (C) 2007 Oracle. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18#include <linux/sched.h> 19#include <linux/bio.h> 20#include <linux/slab.h> 21#include <linux/buffer_head.h> 22#include <linux/blkdev.h> 23#include <linux/random.h> 24#include <linux/iocontext.h> 25#include <asm/div64.h> 26#include "compat.h" 27#include "ctree.h" 28#include "extent_map.h" 29#include "disk-io.h" 30#include "transaction.h" 31#include "print-tree.h" 32#include "volumes.h" 33#include "async-thread.h" 34 35struct map_lookup { 36 u64 type; 37 int io_align; 38 int io_width; 39 int stripe_len; 40 int sector_size; 41 int num_stripes; 42 int sub_stripes; 43 struct btrfs_bio_stripe stripes[]; 44}; 45 46static int init_first_rw_device(struct btrfs_trans_handle *trans, 47 struct btrfs_root *root, 48 struct btrfs_device *device); 49static int btrfs_relocate_sys_chunks(struct btrfs_root *root); 50 51#define map_lookup_size(n) (sizeof(struct map_lookup) + \ 52 (sizeof(struct btrfs_bio_stripe) * (n))) 53 54static DEFINE_MUTEX(uuid_mutex); 55static LIST_HEAD(fs_uuids); 56 57void btrfs_lock_volumes(void) 58{ 59 mutex_lock(&uuid_mutex); 60} 61 62void btrfs_unlock_volumes(void) 63{ 64 mutex_unlock(&uuid_mutex); 65} 66 67static void lock_chunks(struct btrfs_root *root) 68{ 69 mutex_lock(&root->fs_info->chunk_mutex); 70} 71 72static void unlock_chunks(struct btrfs_root *root) 73{ 74 mutex_unlock(&root->fs_info->chunk_mutex); 75} 76 77static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 78{ 79 struct btrfs_device *device; 80 WARN_ON(fs_devices->opened); 81 while (!list_empty(&fs_devices->devices)) { 82 device = list_entry(fs_devices->devices.next, 83 struct btrfs_device, dev_list); 84 list_del(&device->dev_list); 85 kfree(device->name); 86 kfree(device); 87 } 88 kfree(fs_devices); 89} 90 91int btrfs_cleanup_fs_uuids(void) 92{ 93 struct btrfs_fs_devices *fs_devices; 94 95 while (!list_empty(&fs_uuids)) { 96 fs_devices = list_entry(fs_uuids.next, 97 struct btrfs_fs_devices, list); 98 list_del(&fs_devices->list); 99 free_fs_devices(fs_devices); 100 } 101 return 0; 102} 103 104static noinline struct btrfs_device *__find_device(struct list_head *head, 105 u64 devid, u8 *uuid) 106{ 107 struct btrfs_device *dev; 108 109 list_for_each_entry(dev, head, dev_list) { 110 if (dev->devid == devid && 111 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { 112 return dev; 113 } 114 } 115 return NULL; 116} 117 118static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) 119{ 120 struct btrfs_fs_devices *fs_devices; 121 122 list_for_each_entry(fs_devices, &fs_uuids, list) { 123 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 124 return fs_devices; 125 } 126 return NULL; 127} 128 129static void requeue_list(struct btrfs_pending_bios *pending_bios, 130 struct bio *head, struct bio *tail) 131{ 132 133 struct bio *old_head; 134 135 old_head = pending_bios->head; 136 pending_bios->head = head; 137 if (pending_bios->tail) 138 tail->bi_next = old_head; 139 else 140 pending_bios->tail = tail; 141} 142 143/* 144 * we try to collect pending bios for a device so we don't get a large 145 * number of procs sending bios down to the same device. This greatly 146 * improves the schedulers ability to collect and merge the bios. 147 * 148 * But, it also turns into a long list of bios to process and that is sure 149 * to eventually make the worker thread block. The solution here is to 150 * make some progress and then put this work struct back at the end of 151 * the list if the block device is congested. This way, multiple devices 152 * can make progress from a single worker thread. 153 */ 154static noinline int run_scheduled_bios(struct btrfs_device *device) 155{ 156 struct bio *pending; 157 struct backing_dev_info *bdi; 158 struct btrfs_fs_info *fs_info; 159 struct btrfs_pending_bios *pending_bios; 160 struct bio *tail; 161 struct bio *cur; 162 int again = 0; 163 unsigned long num_run; 164 unsigned long num_sync_run; 165 unsigned long batch_run = 0; 166 unsigned long limit; 167 unsigned long last_waited = 0; 168 int force_reg = 0; 169 170 bdi = blk_get_backing_dev_info(device->bdev); 171 fs_info = device->dev_root->fs_info; 172 limit = btrfs_async_submit_limit(fs_info); 173 limit = limit * 2 / 3; 174 175 /* we want to make sure that every time we switch from the sync 176 * list to the normal list, we unplug 177 */ 178 num_sync_run = 0; 179 180loop: 181 spin_lock(&device->io_lock); 182 183loop_lock: 184 num_run = 0; 185 186 /* take all the bios off the list at once and process them 187 * later on (without the lock held). But, remember the 188 * tail and other pointers so the bios can be properly reinserted 189 * into the list if we hit congestion 190 */ 191 if (!force_reg && device->pending_sync_bios.head) { 192 pending_bios = &device->pending_sync_bios; 193 force_reg = 1; 194 } else { 195 pending_bios = &device->pending_bios; 196 force_reg = 0; 197 } 198 199 pending = pending_bios->head; 200 tail = pending_bios->tail; 201 WARN_ON(pending && !tail); 202 203 /* 204 * if pending was null this time around, no bios need processing 205 * at all and we can stop. Otherwise it'll loop back up again 206 * and do an additional check so no bios are missed. 207 * 208 * device->running_pending is used to synchronize with the 209 * schedule_bio code. 210 */ 211 if (device->pending_sync_bios.head == NULL && 212 device->pending_bios.head == NULL) { 213 again = 0; 214 device->running_pending = 0; 215 } else { 216 again = 1; 217 device->running_pending = 1; 218 } 219 220 pending_bios->head = NULL; 221 pending_bios->tail = NULL; 222 223 spin_unlock(&device->io_lock); 224 225 /* 226 * if we're doing the regular priority list, make sure we unplug 227 * for any high prio bios we've sent down 228 */ 229 if (pending_bios == &device->pending_bios && num_sync_run > 0) { 230 num_sync_run = 0; 231 blk_run_backing_dev(bdi, NULL); 232 } 233 234 while (pending) { 235 236 rmb(); 237 /* we want to work on both lists, but do more bios on the 238 * sync list than the regular list 239 */ 240 if ((num_run > 32 && 241 pending_bios != &device->pending_sync_bios && 242 device->pending_sync_bios.head) || 243 (num_run > 64 && pending_bios == &device->pending_sync_bios && 244 device->pending_bios.head)) { 245 spin_lock(&device->io_lock); 246 requeue_list(pending_bios, pending, tail); 247 goto loop_lock; 248 } 249 250 cur = pending; 251 pending = pending->bi_next; 252 cur->bi_next = NULL; 253 atomic_dec(&fs_info->nr_async_bios); 254 255 if (atomic_read(&fs_info->nr_async_bios) < limit && 256 waitqueue_active(&fs_info->async_submit_wait)) 257 wake_up(&fs_info->async_submit_wait); 258 259 BUG_ON(atomic_read(&cur->bi_cnt) == 0); 260 261 if (cur->bi_rw & REQ_SYNC) 262 num_sync_run++; 263 264 submit_bio(cur->bi_rw, cur); 265 num_run++; 266 batch_run++; 267 if (need_resched()) { 268 if (num_sync_run) { 269 blk_run_backing_dev(bdi, NULL); 270 num_sync_run = 0; 271 } 272 cond_resched(); 273 } 274 275 /* 276 * we made progress, there is more work to do and the bdi 277 * is now congested. Back off and let other work structs 278 * run instead 279 */ 280 if (pending && bdi_write_congested(bdi) && batch_run > 8 && 281 fs_info->fs_devices->open_devices > 1) { 282 struct io_context *ioc; 283 284 ioc = current->io_context; 285 286 /* 287 * the main goal here is that we don't want to 288 * block if we're going to be able to submit 289 * more requests without blocking. 290 * 291 * This code does two great things, it pokes into 292 * the elevator code from a filesystem _and_ 293 * it makes assumptions about how batching works. 294 */ 295 if (ioc && ioc->nr_batch_requests > 0 && 296 time_before(jiffies, ioc->last_waited + HZ/50UL) && 297 (last_waited == 0 || 298 ioc->last_waited == last_waited)) { 299 /* 300 * we want to go through our batch of 301 * requests and stop. So, we copy out 302 * the ioc->last_waited time and test 303 * against it before looping 304 */ 305 last_waited = ioc->last_waited; 306 if (need_resched()) { 307 if (num_sync_run) { 308 blk_run_backing_dev(bdi, NULL); 309 num_sync_run = 0; 310 } 311 cond_resched(); 312 } 313 continue; 314 } 315 spin_lock(&device->io_lock); 316 requeue_list(pending_bios, pending, tail); 317 device->running_pending = 1; 318 319 spin_unlock(&device->io_lock); 320 btrfs_requeue_work(&device->work); 321 goto done; 322 } 323 } 324 325 if (num_sync_run) { 326 num_sync_run = 0; 327 blk_run_backing_dev(bdi, NULL); 328 } 329 /* 330 * IO has already been through a long path to get here. Checksumming, 331 * async helper threads, perhaps compression. We've done a pretty 332 * good job of collecting a batch of IO and should just unplug 333 * the device right away. 334 * 335 * This will help anyone who is waiting on the IO, they might have 336 * already unplugged, but managed to do so before the bio they 337 * cared about found its way down here. 338 */ 339 blk_run_backing_dev(bdi, NULL); 340 341 cond_resched(); 342 if (again) 343 goto loop; 344 345 spin_lock(&device->io_lock); 346 if (device->pending_bios.head || device->pending_sync_bios.head) 347 goto loop_lock; 348 spin_unlock(&device->io_lock); 349 350done: 351 return 0; 352} 353 354static void pending_bios_fn(struct btrfs_work *work) 355{ 356 struct btrfs_device *device; 357 358 device = container_of(work, struct btrfs_device, work); 359 run_scheduled_bios(device); 360} 361 362static noinline int device_list_add(const char *path, 363 struct btrfs_super_block *disk_super, 364 u64 devid, struct btrfs_fs_devices **fs_devices_ret) 365{ 366 struct btrfs_device *device; 367 struct btrfs_fs_devices *fs_devices; 368 u64 found_transid = btrfs_super_generation(disk_super); 369 char *name; 370 371 fs_devices = find_fsid(disk_super->fsid); 372 if (!fs_devices) { 373 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); 374 if (!fs_devices) 375 return -ENOMEM; 376 INIT_LIST_HEAD(&fs_devices->devices); 377 INIT_LIST_HEAD(&fs_devices->alloc_list); 378 list_add(&fs_devices->list, &fs_uuids); 379 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 380 fs_devices->latest_devid = devid; 381 fs_devices->latest_trans = found_transid; 382 mutex_init(&fs_devices->device_list_mutex); 383 device = NULL; 384 } else { 385 device = __find_device(&fs_devices->devices, devid, 386 disk_super->dev_item.uuid); 387 } 388 if (!device) { 389 if (fs_devices->opened) 390 return -EBUSY; 391 392 device = kzalloc(sizeof(*device), GFP_NOFS); 393 if (!device) { 394 /* we can safely leave the fs_devices entry around */ 395 return -ENOMEM; 396 } 397 device->devid = devid; 398 device->work.func = pending_bios_fn; 399 memcpy(device->uuid, disk_super->dev_item.uuid, 400 BTRFS_UUID_SIZE); 401 device->barriers = 1; 402 spin_lock_init(&device->io_lock); 403 device->name = kstrdup(path, GFP_NOFS); 404 if (!device->name) { 405 kfree(device); 406 return -ENOMEM; 407 } 408 INIT_LIST_HEAD(&device->dev_alloc_list); 409 410 mutex_lock(&fs_devices->device_list_mutex); 411 list_add(&device->dev_list, &fs_devices->devices); 412 mutex_unlock(&fs_devices->device_list_mutex); 413 414 device->fs_devices = fs_devices; 415 fs_devices->num_devices++; 416 } else if (strcmp(device->name, path)) { 417 name = kstrdup(path, GFP_NOFS); 418 if (!name) 419 return -ENOMEM; 420 kfree(device->name); 421 device->name = name; 422 } 423 424 if (found_transid > fs_devices->latest_trans) { 425 fs_devices->latest_devid = devid; 426 fs_devices->latest_trans = found_transid; 427 } 428 *fs_devices_ret = fs_devices; 429 return 0; 430} 431 432static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 433{ 434 struct btrfs_fs_devices *fs_devices; 435 struct btrfs_device *device; 436 struct btrfs_device *orig_dev; 437 438 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); 439 if (!fs_devices) 440 return ERR_PTR(-ENOMEM); 441 442 INIT_LIST_HEAD(&fs_devices->devices); 443 INIT_LIST_HEAD(&fs_devices->alloc_list); 444 INIT_LIST_HEAD(&fs_devices->list); 445 mutex_init(&fs_devices->device_list_mutex); 446 fs_devices->latest_devid = orig->latest_devid; 447 fs_devices->latest_trans = orig->latest_trans; 448 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); 449 450 mutex_lock(&orig->device_list_mutex); 451 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 452 device = kzalloc(sizeof(*device), GFP_NOFS); 453 if (!device) 454 goto error; 455 456 device->name = kstrdup(orig_dev->name, GFP_NOFS); 457 if (!device->name) { 458 kfree(device); 459 goto error; 460 } 461 462 device->devid = orig_dev->devid; 463 device->work.func = pending_bios_fn; 464 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); 465 device->barriers = 1; 466 spin_lock_init(&device->io_lock); 467 INIT_LIST_HEAD(&device->dev_list); 468 INIT_LIST_HEAD(&device->dev_alloc_list); 469 470 list_add(&device->dev_list, &fs_devices->devices); 471 device->fs_devices = fs_devices; 472 fs_devices->num_devices++; 473 } 474 mutex_unlock(&orig->device_list_mutex); 475 return fs_devices; 476error: 477 mutex_unlock(&orig->device_list_mutex); 478 free_fs_devices(fs_devices); 479 return ERR_PTR(-ENOMEM); 480} 481 482int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) 483{ 484 struct btrfs_device *device, *next; 485 486 mutex_lock(&uuid_mutex); 487again: 488 mutex_lock(&fs_devices->device_list_mutex); 489 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 490 if (device->in_fs_metadata) 491 continue; 492 493 if (device->bdev) { 494 close_bdev_exclusive(device->bdev, device->mode); 495 device->bdev = NULL; 496 fs_devices->open_devices--; 497 } 498 if (device->writeable) { 499 list_del_init(&device->dev_alloc_list); 500 device->writeable = 0; 501 fs_devices->rw_devices--; 502 } 503 list_del_init(&device->dev_list); 504 fs_devices->num_devices--; 505 kfree(device->name); 506 kfree(device); 507 } 508 mutex_unlock(&fs_devices->device_list_mutex); 509 510 if (fs_devices->seed) { 511 fs_devices = fs_devices->seed; 512 goto again; 513 } 514 515 mutex_unlock(&uuid_mutex); 516 return 0; 517} 518 519static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 520{ 521 struct btrfs_device *device; 522 523 if (--fs_devices->opened > 0) 524 return 0; 525 526 list_for_each_entry(device, &fs_devices->devices, dev_list) { 527 if (device->bdev) { 528 close_bdev_exclusive(device->bdev, device->mode); 529 fs_devices->open_devices--; 530 } 531 if (device->writeable) { 532 list_del_init(&device->dev_alloc_list); 533 fs_devices->rw_devices--; 534 } 535 536 device->bdev = NULL; 537 device->writeable = 0; 538 device->in_fs_metadata = 0; 539 } 540 WARN_ON(fs_devices->open_devices); 541 WARN_ON(fs_devices->rw_devices); 542 fs_devices->opened = 0; 543 fs_devices->seeding = 0; 544 545 return 0; 546} 547 548int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 549{ 550 struct btrfs_fs_devices *seed_devices = NULL; 551 int ret; 552 553 mutex_lock(&uuid_mutex); 554 ret = __btrfs_close_devices(fs_devices); 555 if (!fs_devices->opened) { 556 seed_devices = fs_devices->seed; 557 fs_devices->seed = NULL; 558 } 559 mutex_unlock(&uuid_mutex); 560 561 while (seed_devices) { 562 fs_devices = seed_devices; 563 seed_devices = fs_devices->seed; 564 __btrfs_close_devices(fs_devices); 565 free_fs_devices(fs_devices); 566 } 567 return ret; 568} 569 570static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 571 fmode_t flags, void *holder) 572{ 573 struct block_device *bdev; 574 struct list_head *head = &fs_devices->devices; 575 struct btrfs_device *device; 576 struct block_device *latest_bdev = NULL; 577 struct buffer_head *bh; 578 struct btrfs_super_block *disk_super; 579 u64 latest_devid = 0; 580 u64 latest_transid = 0; 581 u64 devid; 582 int seeding = 1; 583 int ret = 0; 584 585 list_for_each_entry(device, head, dev_list) { 586 if (device->bdev) 587 continue; 588 if (!device->name) 589 continue; 590 591 bdev = open_bdev_exclusive(device->name, flags, holder); 592 if (IS_ERR(bdev)) { 593 printk(KERN_INFO "open %s failed\n", device->name); 594 goto error; 595 } 596 set_blocksize(bdev, 4096); 597 598 bh = btrfs_read_dev_super(bdev); 599 if (!bh) 600 goto error_close; 601 602 disk_super = (struct btrfs_super_block *)bh->b_data; 603 devid = btrfs_stack_device_id(&disk_super->dev_item); 604 if (devid != device->devid) 605 goto error_brelse; 606 607 if (memcmp(device->uuid, disk_super->dev_item.uuid, 608 BTRFS_UUID_SIZE)) 609 goto error_brelse; 610 611 device->generation = btrfs_super_generation(disk_super); 612 if (!latest_transid || device->generation > latest_transid) { 613 latest_devid = devid; 614 latest_transid = device->generation; 615 latest_bdev = bdev; 616 } 617 618 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 619 device->writeable = 0; 620 } else { 621 device->writeable = !bdev_read_only(bdev); 622 seeding = 0; 623 } 624 625 device->bdev = bdev; 626 device->in_fs_metadata = 0; 627 device->mode = flags; 628 629 if (!blk_queue_nonrot(bdev_get_queue(bdev))) 630 fs_devices->rotating = 1; 631 632 fs_devices->open_devices++; 633 if (device->writeable) { 634 fs_devices->rw_devices++; 635 list_add(&device->dev_alloc_list, 636 &fs_devices->alloc_list); 637 } 638 continue; 639 640error_brelse: 641 brelse(bh); 642error_close: 643 close_bdev_exclusive(bdev, FMODE_READ); 644error: 645 continue; 646 } 647 if (fs_devices->open_devices == 0) { 648 ret = -EIO; 649 goto out; 650 } 651 fs_devices->seeding = seeding; 652 fs_devices->opened = 1; 653 fs_devices->latest_bdev = latest_bdev; 654 fs_devices->latest_devid = latest_devid; 655 fs_devices->latest_trans = latest_transid; 656 fs_devices->total_rw_bytes = 0; 657out: 658 return ret; 659} 660 661int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 662 fmode_t flags, void *holder) 663{ 664 int ret; 665 666 mutex_lock(&uuid_mutex); 667 if (fs_devices->opened) { 668 fs_devices->opened++; 669 ret = 0; 670 } else { 671 ret = __btrfs_open_devices(fs_devices, flags, holder); 672 } 673 mutex_unlock(&uuid_mutex); 674 return ret; 675} 676 677int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, 678 struct btrfs_fs_devices **fs_devices_ret) 679{ 680 struct btrfs_super_block *disk_super; 681 struct block_device *bdev; 682 struct buffer_head *bh; 683 int ret; 684 u64 devid; 685 u64 transid; 686 687 mutex_lock(&uuid_mutex); 688 689 bdev = open_bdev_exclusive(path, flags, holder); 690 691 if (IS_ERR(bdev)) { 692 ret = PTR_ERR(bdev); 693 goto error; 694 } 695 696 ret = set_blocksize(bdev, 4096); 697 if (ret) 698 goto error_close; 699 bh = btrfs_read_dev_super(bdev); 700 if (!bh) { 701 ret = -EIO; 702 goto error_close; 703 } 704 disk_super = (struct btrfs_super_block *)bh->b_data; 705 devid = btrfs_stack_device_id(&disk_super->dev_item); 706 transid = btrfs_super_generation(disk_super); 707 if (disk_super->label[0]) 708 printk(KERN_INFO "device label %s ", disk_super->label); 709 else { 710 printk(KERN_INFO "device fsid %llx-%llx ", 711 *(unsigned long long *)disk_super->fsid, 712 *(unsigned long long *)(disk_super->fsid + 8)); 713 } 714 printk(KERN_CONT "devid %llu transid %llu %s\n", 715 (unsigned long long)devid, (unsigned long long)transid, path); 716 ret = device_list_add(path, disk_super, devid, fs_devices_ret); 717 718 brelse(bh); 719error_close: 720 close_bdev_exclusive(bdev, flags); 721error: 722 mutex_unlock(&uuid_mutex); 723 return ret; 724} 725 726/* 727 * this uses a pretty simple search, the expectation is that it is 728 * called very infrequently and that a given device has a small number 729 * of extents 730 */ 731int find_free_dev_extent(struct btrfs_trans_handle *trans, 732 struct btrfs_device *device, u64 num_bytes, 733 u64 *start, u64 *max_avail) 734{ 735 struct btrfs_key key; 736 struct btrfs_root *root = device->dev_root; 737 struct btrfs_dev_extent *dev_extent = NULL; 738 struct btrfs_path *path; 739 u64 hole_size = 0; 740 u64 last_byte = 0; 741 u64 search_start = 0; 742 u64 search_end = device->total_bytes; 743 int ret; 744 int slot = 0; 745 int start_found; 746 struct extent_buffer *l; 747 748 path = btrfs_alloc_path(); 749 if (!path) 750 return -ENOMEM; 751 path->reada = 2; 752 start_found = 0; 753 754 755 /* we don't want to overwrite the superblock on the drive, 756 * so we make sure to start at an offset of at least 1MB 757 */ 758 search_start = max((u64)1024 * 1024, search_start); 759 760 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes) 761 search_start = max(root->fs_info->alloc_start, search_start); 762 763 key.objectid = device->devid; 764 key.offset = search_start; 765 key.type = BTRFS_DEV_EXTENT_KEY; 766 ret = btrfs_search_slot(trans, root, &key, path, 0, 0); 767 if (ret < 0) 768 goto error; 769 if (ret > 0) { 770 ret = btrfs_previous_item(root, path, key.objectid, key.type); 771 if (ret < 0) 772 goto error; 773 if (ret > 0) 774 start_found = 1; 775 } 776 l = path->nodes[0]; 777 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 778 while (1) { 779 l = path->nodes[0]; 780 slot = path->slots[0]; 781 if (slot >= btrfs_header_nritems(l)) { 782 ret = btrfs_next_leaf(root, path); 783 if (ret == 0) 784 continue; 785 if (ret < 0) 786 goto error; 787no_more_items: 788 if (!start_found) { 789 if (search_start >= search_end) { 790 ret = -ENOSPC; 791 goto error; 792 } 793 *start = search_start; 794 start_found = 1; 795 goto check_pending; 796 } 797 *start = last_byte > search_start ? 798 last_byte : search_start; 799 if (search_end <= *start) { 800 ret = -ENOSPC; 801 goto error; 802 } 803 goto check_pending; 804 } 805 btrfs_item_key_to_cpu(l, &key, slot); 806 807 if (key.objectid < device->devid) 808 goto next; 809 810 if (key.objectid > device->devid) 811 goto no_more_items; 812 813 if (key.offset >= search_start && key.offset > last_byte && 814 start_found) { 815 if (last_byte < search_start) 816 last_byte = search_start; 817 hole_size = key.offset - last_byte; 818 819 if (hole_size > *max_avail) 820 *max_avail = hole_size; 821 822 if (key.offset > last_byte && 823 hole_size >= num_bytes) { 824 *start = last_byte; 825 goto check_pending; 826 } 827 } 828 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) 829 goto next; 830 831 start_found = 1; 832 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 833 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent); 834next: 835 path->slots[0]++; 836 cond_resched(); 837 } 838check_pending: 839 /* we have to make sure we didn't find an extent that has already 840 * been allocated by the map tree or the original allocation 841 */ 842 BUG_ON(*start < search_start); 843 844 if (*start + num_bytes > search_end) { 845 ret = -ENOSPC; 846 goto error; 847 } 848 /* check for pending inserts here */ 849 ret = 0; 850 851error: 852 btrfs_free_path(path); 853 return ret; 854} 855 856static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 857 struct btrfs_device *device, 858 u64 start) 859{ 860 int ret; 861 struct btrfs_path *path; 862 struct btrfs_root *root = device->dev_root; 863 struct btrfs_key key; 864 struct btrfs_key found_key; 865 struct extent_buffer *leaf = NULL; 866 struct btrfs_dev_extent *extent = NULL; 867 868 path = btrfs_alloc_path(); 869 if (!path) 870 return -ENOMEM; 871 872 key.objectid = device->devid; 873 key.offset = start; 874 key.type = BTRFS_DEV_EXTENT_KEY; 875 876 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 877 if (ret > 0) { 878 ret = btrfs_previous_item(root, path, key.objectid, 879 BTRFS_DEV_EXTENT_KEY); 880 BUG_ON(ret); 881 leaf = path->nodes[0]; 882 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 883 extent = btrfs_item_ptr(leaf, path->slots[0], 884 struct btrfs_dev_extent); 885 BUG_ON(found_key.offset > start || found_key.offset + 886 btrfs_dev_extent_length(leaf, extent) < start); 887 ret = 0; 888 } else if (ret == 0) { 889 leaf = path->nodes[0]; 890 extent = btrfs_item_ptr(leaf, path->slots[0], 891 struct btrfs_dev_extent); 892 } 893 BUG_ON(ret); 894 895 if (device->bytes_used > 0) 896 device->bytes_used -= btrfs_dev_extent_length(leaf, extent); 897 ret = btrfs_del_item(trans, root, path); 898 BUG_ON(ret); 899 900 btrfs_free_path(path); 901 return ret; 902} 903 904int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, 905 struct btrfs_device *device, 906 u64 chunk_tree, u64 chunk_objectid, 907 u64 chunk_offset, u64 start, u64 num_bytes) 908{ 909 int ret; 910 struct btrfs_path *path; 911 struct btrfs_root *root = device->dev_root; 912 struct btrfs_dev_extent *extent; 913 struct extent_buffer *leaf; 914 struct btrfs_key key; 915 916 WARN_ON(!device->in_fs_metadata); 917 path = btrfs_alloc_path(); 918 if (!path) 919 return -ENOMEM; 920 921 key.objectid = device->devid; 922 key.offset = start; 923 key.type = BTRFS_DEV_EXTENT_KEY; 924 ret = btrfs_insert_empty_item(trans, root, path, &key, 925 sizeof(*extent)); 926 BUG_ON(ret); 927 928 leaf = path->nodes[0]; 929 extent = btrfs_item_ptr(leaf, path->slots[0], 930 struct btrfs_dev_extent); 931 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); 932 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); 933 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); 934 935 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, 936 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), 937 BTRFS_UUID_SIZE); 938 939 btrfs_set_dev_extent_length(leaf, extent, num_bytes); 940 btrfs_mark_buffer_dirty(leaf); 941 btrfs_free_path(path); 942 return ret; 943} 944 945static noinline int find_next_chunk(struct btrfs_root *root, 946 u64 objectid, u64 *offset) 947{ 948 struct btrfs_path *path; 949 int ret; 950 struct btrfs_key key; 951 struct btrfs_chunk *chunk; 952 struct btrfs_key found_key; 953 954 path = btrfs_alloc_path(); 955 BUG_ON(!path); 956 957 key.objectid = objectid; 958 key.offset = (u64)-1; 959 key.type = BTRFS_CHUNK_ITEM_KEY; 960 961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 962 if (ret < 0) 963 goto error; 964 965 BUG_ON(ret == 0); 966 967 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); 968 if (ret) { 969 *offset = 0; 970 } else { 971 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 972 path->slots[0]); 973 if (found_key.objectid != objectid) 974 *offset = 0; 975 else { 976 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], 977 struct btrfs_chunk); 978 *offset = found_key.offset + 979 btrfs_chunk_length(path->nodes[0], chunk); 980 } 981 } 982 ret = 0; 983error: 984 btrfs_free_path(path); 985 return ret; 986} 987 988static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) 989{ 990 int ret; 991 struct btrfs_key key; 992 struct btrfs_key found_key; 993 struct btrfs_path *path; 994 995 root = root->fs_info->chunk_root; 996 997 path = btrfs_alloc_path(); 998 if (!path) 999 return -ENOMEM; 1000 1001 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1002 key.type = BTRFS_DEV_ITEM_KEY; 1003 key.offset = (u64)-1; 1004 1005 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 1006 if (ret < 0) 1007 goto error; 1008 1009 BUG_ON(ret == 0); 1010 1011 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, 1012 BTRFS_DEV_ITEM_KEY); 1013 if (ret) { 1014 *objectid = 1; 1015 } else { 1016 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1017 path->slots[0]); 1018 *objectid = found_key.offset + 1; 1019 } 1020 ret = 0; 1021error: 1022 btrfs_free_path(path); 1023 return ret; 1024} 1025 1026/* 1027 * the device information is stored in the chunk root 1028 * the btrfs_device struct should be fully filled in 1029 */ 1030int btrfs_add_device(struct btrfs_trans_handle *trans, 1031 struct btrfs_root *root, 1032 struct btrfs_device *device) 1033{ 1034 int ret; 1035 struct btrfs_path *path; 1036 struct btrfs_dev_item *dev_item; 1037 struct extent_buffer *leaf; 1038 struct btrfs_key key; 1039 unsigned long ptr; 1040 1041 root = root->fs_info->chunk_root; 1042 1043 path = btrfs_alloc_path(); 1044 if (!path) 1045 return -ENOMEM; 1046 1047 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1048 key.type = BTRFS_DEV_ITEM_KEY; 1049 key.offset = device->devid; 1050 1051 ret = btrfs_insert_empty_item(trans, root, path, &key, 1052 sizeof(*dev_item)); 1053 if (ret) 1054 goto out; 1055 1056 leaf = path->nodes[0]; 1057 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1058 1059 btrfs_set_device_id(leaf, dev_item, device->devid); 1060 btrfs_set_device_generation(leaf, dev_item, 0); 1061 btrfs_set_device_type(leaf, dev_item, device->type); 1062 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1063 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1064 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1065 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); 1066 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); 1067 btrfs_set_device_group(leaf, dev_item, 0); 1068 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1069 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1070 btrfs_set_device_start_offset(leaf, dev_item, 0); 1071 1072 ptr = (unsigned long)btrfs_device_uuid(dev_item); 1073 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1074 ptr = (unsigned long)btrfs_device_fsid(dev_item); 1075 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); 1076 btrfs_mark_buffer_dirty(leaf); 1077 1078 ret = 0; 1079out: 1080 btrfs_free_path(path); 1081 return ret; 1082} 1083 1084static int btrfs_rm_dev_item(struct btrfs_root *root, 1085 struct btrfs_device *device) 1086{ 1087 int ret; 1088 struct btrfs_path *path; 1089 struct btrfs_key key; 1090 struct btrfs_trans_handle *trans; 1091 1092 root = root->fs_info->chunk_root; 1093 1094 path = btrfs_alloc_path(); 1095 if (!path) 1096 return -ENOMEM; 1097 1098 trans = btrfs_start_transaction(root, 0); 1099 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1100 key.type = BTRFS_DEV_ITEM_KEY; 1101 key.offset = device->devid; 1102 lock_chunks(root); 1103 1104 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1105 if (ret < 0) 1106 goto out; 1107 1108 if (ret > 0) { 1109 ret = -ENOENT; 1110 goto out; 1111 } 1112 1113 ret = btrfs_del_item(trans, root, path); 1114 if (ret) 1115 goto out; 1116out: 1117 btrfs_free_path(path); 1118 unlock_chunks(root); 1119 btrfs_commit_transaction(trans, root); 1120 return ret; 1121} 1122 1123int btrfs_rm_device(struct btrfs_root *root, char *device_path) 1124{ 1125 struct btrfs_device *device; 1126 struct btrfs_device *next_device; 1127 struct block_device *bdev; 1128 struct buffer_head *bh = NULL; 1129 struct btrfs_super_block *disk_super; 1130 u64 all_avail; 1131 u64 devid; 1132 u64 num_devices; 1133 u8 *dev_uuid; 1134 int ret = 0; 1135 1136 mutex_lock(&uuid_mutex); 1137 mutex_lock(&root->fs_info->volume_mutex); 1138 1139 all_avail = root->fs_info->avail_data_alloc_bits | 1140 root->fs_info->avail_system_alloc_bits | 1141 root->fs_info->avail_metadata_alloc_bits; 1142 1143 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && 1144 root->fs_info->fs_devices->num_devices <= 4) { 1145 printk(KERN_ERR "btrfs: unable to go below four devices " 1146 "on raid10\n"); 1147 ret = -EINVAL; 1148 goto out; 1149 } 1150 1151 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && 1152 root->fs_info->fs_devices->num_devices <= 2) { 1153 printk(KERN_ERR "btrfs: unable to go below two " 1154 "devices on raid1\n"); 1155 ret = -EINVAL; 1156 goto out; 1157 } 1158 1159 if (strcmp(device_path, "missing") == 0) { 1160 struct list_head *devices; 1161 struct btrfs_device *tmp; 1162 1163 device = NULL; 1164 devices = &root->fs_info->fs_devices->devices; 1165 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 1166 list_for_each_entry(tmp, devices, dev_list) { 1167 if (tmp->in_fs_metadata && !tmp->bdev) { 1168 device = tmp; 1169 break; 1170 } 1171 } 1172 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1173 bdev = NULL; 1174 bh = NULL; 1175 disk_super = NULL; 1176 if (!device) { 1177 printk(KERN_ERR "btrfs: no missing devices found to " 1178 "remove\n"); 1179 goto out; 1180 } 1181 } else { 1182 bdev = open_bdev_exclusive(device_path, FMODE_READ, 1183 root->fs_info->bdev_holder); 1184 if (IS_ERR(bdev)) { 1185 ret = PTR_ERR(bdev); 1186 goto out; 1187 } 1188 1189 set_blocksize(bdev, 4096); 1190 bh = btrfs_read_dev_super(bdev); 1191 if (!bh) { 1192 ret = -EIO; 1193 goto error_close; 1194 } 1195 disk_super = (struct btrfs_super_block *)bh->b_data; 1196 devid = btrfs_stack_device_id(&disk_super->dev_item); 1197 dev_uuid = disk_super->dev_item.uuid; 1198 device = btrfs_find_device(root, devid, dev_uuid, 1199 disk_super->fsid); 1200 if (!device) { 1201 ret = -ENOENT; 1202 goto error_brelse; 1203 } 1204 } 1205 1206 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { 1207 printk(KERN_ERR "btrfs: unable to remove the only writeable " 1208 "device\n"); 1209 ret = -EINVAL; 1210 goto error_brelse; 1211 } 1212 1213 if (device->writeable) { 1214 list_del_init(&device->dev_alloc_list); 1215 root->fs_info->fs_devices->rw_devices--; 1216 } 1217 1218 ret = btrfs_shrink_device(device, 0); 1219 if (ret) 1220 goto error_brelse; 1221 1222 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); 1223 if (ret) 1224 goto error_brelse; 1225 1226 device->in_fs_metadata = 0; 1227 1228 /* 1229 * the device list mutex makes sure that we don't change 1230 * the device list while someone else is writing out all 1231 * the device supers. 1232 */ 1233 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 1234 list_del_init(&device->dev_list); 1235 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1236 1237 device->fs_devices->num_devices--; 1238 1239 next_device = list_entry(root->fs_info->fs_devices->devices.next, 1240 struct btrfs_device, dev_list); 1241 if (device->bdev == root->fs_info->sb->s_bdev) 1242 root->fs_info->sb->s_bdev = next_device->bdev; 1243 if (device->bdev == root->fs_info->fs_devices->latest_bdev) 1244 root->fs_info->fs_devices->latest_bdev = next_device->bdev; 1245 1246 if (device->bdev) { 1247 close_bdev_exclusive(device->bdev, device->mode); 1248 device->bdev = NULL; 1249 device->fs_devices->open_devices--; 1250 } 1251 1252 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1; 1253 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices); 1254 1255 if (device->fs_devices->open_devices == 0) { 1256 struct btrfs_fs_devices *fs_devices; 1257 fs_devices = root->fs_info->fs_devices; 1258 while (fs_devices) { 1259 if (fs_devices->seed == device->fs_devices) 1260 break; 1261 fs_devices = fs_devices->seed; 1262 } 1263 fs_devices->seed = device->fs_devices->seed; 1264 device->fs_devices->seed = NULL; 1265 __btrfs_close_devices(device->fs_devices); 1266 free_fs_devices(device->fs_devices); 1267 } 1268 1269 /* 1270 * at this point, the device is zero sized. We want to 1271 * remove it from the devices list and zero out the old super 1272 */ 1273 if (device->writeable) { 1274 /* make sure this device isn't detected as part of 1275 * the FS anymore 1276 */ 1277 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 1278 set_buffer_dirty(bh); 1279 sync_dirty_buffer(bh); 1280 } 1281 1282 kfree(device->name); 1283 kfree(device); 1284 ret = 0; 1285 1286error_brelse: 1287 brelse(bh); 1288error_close: 1289 if (bdev) 1290 close_bdev_exclusive(bdev, FMODE_READ); 1291out: 1292 mutex_unlock(&root->fs_info->volume_mutex); 1293 mutex_unlock(&uuid_mutex); 1294 return ret; 1295} 1296 1297/* 1298 * does all the dirty work required for changing file system's UUID. 1299 */ 1300static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans, 1301 struct btrfs_root *root) 1302{ 1303 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 1304 struct btrfs_fs_devices *old_devices; 1305 struct btrfs_fs_devices *seed_devices; 1306 struct btrfs_super_block *disk_super = &root->fs_info->super_copy; 1307 struct btrfs_device *device; 1308 u64 super_flags; 1309 1310 BUG_ON(!mutex_is_locked(&uuid_mutex)); 1311 if (!fs_devices->seeding) 1312 return -EINVAL; 1313 1314 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); 1315 if (!seed_devices) 1316 return -ENOMEM; 1317 1318 old_devices = clone_fs_devices(fs_devices); 1319 if (IS_ERR(old_devices)) { 1320 kfree(seed_devices); 1321 return PTR_ERR(old_devices); 1322 } 1323 1324 list_add(&old_devices->list, &fs_uuids); 1325 1326 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 1327 seed_devices->opened = 1; 1328 INIT_LIST_HEAD(&seed_devices->devices); 1329 INIT_LIST_HEAD(&seed_devices->alloc_list); 1330 mutex_init(&seed_devices->device_list_mutex); 1331 list_splice_init(&fs_devices->devices, &seed_devices->devices); 1332 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); 1333 list_for_each_entry(device, &seed_devices->devices, dev_list) { 1334 device->fs_devices = seed_devices; 1335 } 1336 1337 fs_devices->seeding = 0; 1338 fs_devices->num_devices = 0; 1339 fs_devices->open_devices = 0; 1340 fs_devices->seed = seed_devices; 1341 1342 generate_random_uuid(fs_devices->fsid); 1343 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 1344 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 1345 super_flags = btrfs_super_flags(disk_super) & 1346 ~BTRFS_SUPER_FLAG_SEEDING; 1347 btrfs_set_super_flags(disk_super, super_flags); 1348 1349 return 0; 1350} 1351 1352/* 1353 * strore the expected generation for seed devices in device items. 1354 */ 1355static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, 1356 struct btrfs_root *root) 1357{ 1358 struct btrfs_path *path; 1359 struct extent_buffer *leaf; 1360 struct btrfs_dev_item *dev_item; 1361 struct btrfs_device *device; 1362 struct btrfs_key key; 1363 u8 fs_uuid[BTRFS_UUID_SIZE]; 1364 u8 dev_uuid[BTRFS_UUID_SIZE]; 1365 u64 devid; 1366 int ret; 1367 1368 path = btrfs_alloc_path(); 1369 if (!path) 1370 return -ENOMEM; 1371 1372 root = root->fs_info->chunk_root; 1373 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1374 key.offset = 0; 1375 key.type = BTRFS_DEV_ITEM_KEY; 1376 1377 while (1) { 1378 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 1379 if (ret < 0) 1380 goto error; 1381 1382 leaf = path->nodes[0]; 1383next_slot: 1384 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 1385 ret = btrfs_next_leaf(root, path); 1386 if (ret > 0) 1387 break; 1388 if (ret < 0) 1389 goto error; 1390 leaf = path->nodes[0]; 1391 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 1392 btrfs_release_path(root, path); 1393 continue; 1394 } 1395 1396 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 1397 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 1398 key.type != BTRFS_DEV_ITEM_KEY) 1399 break; 1400 1401 dev_item = btrfs_item_ptr(leaf, path->slots[0], 1402 struct btrfs_dev_item); 1403 devid = btrfs_device_id(leaf, dev_item); 1404 read_extent_buffer(leaf, dev_uuid, 1405 (unsigned long)btrfs_device_uuid(dev_item), 1406 BTRFS_UUID_SIZE); 1407 read_extent_buffer(leaf, fs_uuid, 1408 (unsigned long)btrfs_device_fsid(dev_item), 1409 BTRFS_UUID_SIZE); 1410 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); 1411 BUG_ON(!device); 1412 1413 if (device->fs_devices->seeding) { 1414 btrfs_set_device_generation(leaf, dev_item, 1415 device->generation); 1416 btrfs_mark_buffer_dirty(leaf); 1417 } 1418 1419 path->slots[0]++; 1420 goto next_slot; 1421 } 1422 ret = 0; 1423error: 1424 btrfs_free_path(path); 1425 return ret; 1426} 1427 1428int btrfs_init_new_device(struct btrfs_root *root, char *device_path) 1429{ 1430 struct btrfs_trans_handle *trans; 1431 struct btrfs_device *device; 1432 struct block_device *bdev; 1433 struct list_head *devices; 1434 struct super_block *sb = root->fs_info->sb; 1435 u64 total_bytes; 1436 int seeding_dev = 0; 1437 int ret = 0; 1438 1439 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) 1440 return -EINVAL; 1441 1442 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder); 1443 if (IS_ERR(bdev)) 1444 return PTR_ERR(bdev); 1445 1446 if (root->fs_info->fs_devices->seeding) { 1447 seeding_dev = 1; 1448 down_write(&sb->s_umount); 1449 mutex_lock(&uuid_mutex); 1450 } 1451 1452 filemap_write_and_wait(bdev->bd_inode->i_mapping); 1453 mutex_lock(&root->fs_info->volume_mutex); 1454 1455 devices = &root->fs_info->fs_devices->devices; 1456 /* 1457 * we have the volume lock, so we don't need the extra 1458 * device list mutex while reading the list here. 1459 */ 1460 list_for_each_entry(device, devices, dev_list) { 1461 if (device->bdev == bdev) { 1462 ret = -EEXIST; 1463 goto error; 1464 } 1465 } 1466 1467 device = kzalloc(sizeof(*device), GFP_NOFS); 1468 if (!device) { 1469 /* we can safely leave the fs_devices entry around */ 1470 ret = -ENOMEM; 1471 goto error; 1472 } 1473 1474 device->name = kstrdup(device_path, GFP_NOFS); 1475 if (!device->name) { 1476 kfree(device); 1477 ret = -ENOMEM; 1478 goto error; 1479 } 1480 1481 ret = find_next_devid(root, &device->devid); 1482 if (ret) { 1483 kfree(device); 1484 goto error; 1485 } 1486 1487 trans = btrfs_start_transaction(root, 0); 1488 lock_chunks(root); 1489 1490 device->barriers = 1; 1491 device->writeable = 1; 1492 device->work.func = pending_bios_fn; 1493 generate_random_uuid(device->uuid); 1494 spin_lock_init(&device->io_lock); 1495 device->generation = trans->transid; 1496 device->io_width = root->sectorsize; 1497 device->io_align = root->sectorsize; 1498 device->sector_size = root->sectorsize; 1499 device->total_bytes = i_size_read(bdev->bd_inode); 1500 device->disk_total_bytes = device->total_bytes; 1501 device->dev_root = root->fs_info->dev_root; 1502 device->bdev = bdev; 1503 device->in_fs_metadata = 1; 1504 device->mode = 0; 1505 set_blocksize(device->bdev, 4096); 1506 1507 if (seeding_dev) { 1508 sb->s_flags &= ~MS_RDONLY; 1509 ret = btrfs_prepare_sprout(trans, root); 1510 BUG_ON(ret); 1511 } 1512 1513 device->fs_devices = root->fs_info->fs_devices; 1514 1515 /* 1516 * we don't want write_supers to jump in here with our device 1517 * half setup 1518 */ 1519 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 1520 list_add(&device->dev_list, &root->fs_info->fs_devices->devices); 1521 list_add(&device->dev_alloc_list, 1522 &root->fs_info->fs_devices->alloc_list); 1523 root->fs_info->fs_devices->num_devices++; 1524 root->fs_info->fs_devices->open_devices++; 1525 root->fs_info->fs_devices->rw_devices++; 1526 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; 1527 1528 if (!blk_queue_nonrot(bdev_get_queue(bdev))) 1529 root->fs_info->fs_devices->rotating = 1; 1530 1531 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); 1532 btrfs_set_super_total_bytes(&root->fs_info->super_copy, 1533 total_bytes + device->total_bytes); 1534 1535 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy); 1536 btrfs_set_super_num_devices(&root->fs_info->super_copy, 1537 total_bytes + 1); 1538 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1539 1540 if (seeding_dev) { 1541 ret = init_first_rw_device(trans, root, device); 1542 BUG_ON(ret); 1543 ret = btrfs_finish_sprout(trans, root); 1544 BUG_ON(ret); 1545 } else { 1546 ret = btrfs_add_device(trans, root, device); 1547 } 1548 1549 /* 1550 * we've got more storage, clear any full flags on the space 1551 * infos 1552 */ 1553 btrfs_clear_space_info_full(root->fs_info); 1554 1555 unlock_chunks(root); 1556 btrfs_commit_transaction(trans, root); 1557 1558 if (seeding_dev) { 1559 mutex_unlock(&uuid_mutex); 1560 up_write(&sb->s_umount); 1561 1562 ret = btrfs_relocate_sys_chunks(root); 1563 BUG_ON(ret); 1564 } 1565out: 1566 mutex_unlock(&root->fs_info->volume_mutex); 1567 return ret; 1568error: 1569 close_bdev_exclusive(bdev, 0); 1570 if (seeding_dev) { 1571 mutex_unlock(&uuid_mutex); 1572 up_write(&sb->s_umount); 1573 } 1574 goto out; 1575} 1576 1577static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 1578 struct btrfs_device *device) 1579{ 1580 int ret; 1581 struct btrfs_path *path; 1582 struct btrfs_root *root; 1583 struct btrfs_dev_item *dev_item; 1584 struct extent_buffer *leaf; 1585 struct btrfs_key key; 1586 1587 root = device->dev_root->fs_info->chunk_root; 1588 1589 path = btrfs_alloc_path(); 1590 if (!path) 1591 return -ENOMEM; 1592 1593 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1594 key.type = BTRFS_DEV_ITEM_KEY; 1595 key.offset = device->devid; 1596 1597 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 1598 if (ret < 0) 1599 goto out; 1600 1601 if (ret > 0) { 1602 ret = -ENOENT; 1603 goto out; 1604 } 1605 1606 leaf = path->nodes[0]; 1607 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1608 1609 btrfs_set_device_id(leaf, dev_item, device->devid); 1610 btrfs_set_device_type(leaf, dev_item, device->type); 1611 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1612 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1613 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1614 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); 1615 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); 1616 btrfs_mark_buffer_dirty(leaf); 1617 1618out: 1619 btrfs_free_path(path); 1620 return ret; 1621} 1622 1623static int __btrfs_grow_device(struct btrfs_trans_handle *trans, 1624 struct btrfs_device *device, u64 new_size) 1625{ 1626 struct btrfs_super_block *super_copy = 1627 &device->dev_root->fs_info->super_copy; 1628 u64 old_total = btrfs_super_total_bytes(super_copy); 1629 u64 diff = new_size - device->total_bytes; 1630 1631 if (!device->writeable) 1632 return -EACCES; 1633 if (new_size <= device->total_bytes) 1634 return -EINVAL; 1635 1636 btrfs_set_super_total_bytes(super_copy, old_total + diff); 1637 device->fs_devices->total_rw_bytes += diff; 1638 1639 device->total_bytes = new_size; 1640 device->disk_total_bytes = new_size; 1641 btrfs_clear_space_info_full(device->dev_root->fs_info); 1642 1643 return btrfs_update_device(trans, device); 1644} 1645 1646int btrfs_grow_device(struct btrfs_trans_handle *trans, 1647 struct btrfs_device *device, u64 new_size) 1648{ 1649 int ret; 1650 lock_chunks(device->dev_root); 1651 ret = __btrfs_grow_device(trans, device, new_size); 1652 unlock_chunks(device->dev_root); 1653 return ret; 1654} 1655 1656static int btrfs_free_chunk(struct btrfs_trans_handle *trans, 1657 struct btrfs_root *root, 1658 u64 chunk_tree, u64 chunk_objectid, 1659 u64 chunk_offset) 1660{ 1661 int ret; 1662 struct btrfs_path *path; 1663 struct btrfs_key key; 1664 1665 root = root->fs_info->chunk_root; 1666 path = btrfs_alloc_path(); 1667 if (!path) 1668 return -ENOMEM; 1669 1670 key.objectid = chunk_objectid; 1671 key.offset = chunk_offset; 1672 key.type = BTRFS_CHUNK_ITEM_KEY; 1673 1674 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1675 BUG_ON(ret); 1676 1677 ret = btrfs_del_item(trans, root, path); 1678 BUG_ON(ret); 1679 1680 btrfs_free_path(path); 1681 return 0; 1682} 1683 1684static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 1685 chunk_offset) 1686{ 1687 struct btrfs_super_block *super_copy = &root->fs_info->super_copy; 1688 struct btrfs_disk_key *disk_key; 1689 struct btrfs_chunk *chunk; 1690 u8 *ptr; 1691 int ret = 0; 1692 u32 num_stripes; 1693 u32 array_size; 1694 u32 len = 0; 1695 u32 cur; 1696 struct btrfs_key key; 1697 1698 array_size = btrfs_super_sys_array_size(super_copy); 1699 1700 ptr = super_copy->sys_chunk_array; 1701 cur = 0; 1702 1703 while (cur < array_size) { 1704 disk_key = (struct btrfs_disk_key *)ptr; 1705 btrfs_disk_key_to_cpu(&key, disk_key); 1706 1707 len = sizeof(*disk_key); 1708 1709 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 1710 chunk = (struct btrfs_chunk *)(ptr + len); 1711 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 1712 len += btrfs_chunk_item_size(num_stripes); 1713 } else { 1714 ret = -EIO; 1715 break; 1716 } 1717 if (key.objectid == chunk_objectid && 1718 key.offset == chunk_offset) { 1719 memmove(ptr, ptr + len, array_size - (cur + len)); 1720 array_size -= len; 1721 btrfs_set_super_sys_array_size(super_copy, array_size); 1722 } else { 1723 ptr += len; 1724 cur += len; 1725 } 1726 } 1727 return ret; 1728} 1729 1730static int btrfs_relocate_chunk(struct btrfs_root *root, 1731 u64 chunk_tree, u64 chunk_objectid, 1732 u64 chunk_offset) 1733{ 1734 struct extent_map_tree *em_tree; 1735 struct btrfs_root *extent_root; 1736 struct btrfs_trans_handle *trans; 1737 struct extent_map *em; 1738 struct map_lookup *map; 1739 int ret; 1740 int i; 1741 1742 root = root->fs_info->chunk_root; 1743 extent_root = root->fs_info->extent_root; 1744 em_tree = &root->fs_info->mapping_tree.map_tree; 1745 1746 ret = btrfs_can_relocate(extent_root, chunk_offset); 1747 if (ret) 1748 return -ENOSPC; 1749 1750 /* step one, relocate all the extents inside this chunk */ 1751 ret = btrfs_relocate_block_group(extent_root, chunk_offset); 1752 if (ret) 1753 return ret; 1754 1755 trans = btrfs_start_transaction(root, 0); 1756 BUG_ON(!trans); 1757 1758 lock_chunks(root); 1759 1760 /* 1761 * step two, delete the device extents and the 1762 * chunk tree entries 1763 */ 1764 read_lock(&em_tree->lock); 1765 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1766 read_unlock(&em_tree->lock); 1767 1768 BUG_ON(em->start > chunk_offset || 1769 em->start + em->len < chunk_offset); 1770 map = (struct map_lookup *)em->bdev; 1771 1772 for (i = 0; i < map->num_stripes; i++) { 1773 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, 1774 map->stripes[i].physical); 1775 BUG_ON(ret); 1776 1777 if (map->stripes[i].dev) { 1778 ret = btrfs_update_device(trans, map->stripes[i].dev); 1779 BUG_ON(ret); 1780 } 1781 } 1782 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, 1783 chunk_offset); 1784 1785 BUG_ON(ret); 1786 1787 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 1788 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); 1789 BUG_ON(ret); 1790 } 1791 1792 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); 1793 BUG_ON(ret); 1794 1795 write_lock(&em_tree->lock); 1796 remove_extent_mapping(em_tree, em); 1797 write_unlock(&em_tree->lock); 1798 1799 kfree(map); 1800 em->bdev = NULL; 1801 1802 /* once for the tree */ 1803 free_extent_map(em); 1804 /* once for us */ 1805 free_extent_map(em); 1806 1807 unlock_chunks(root); 1808 btrfs_end_transaction(trans, root); 1809 return 0; 1810} 1811 1812static int btrfs_relocate_sys_chunks(struct btrfs_root *root) 1813{ 1814 struct btrfs_root *chunk_root = root->fs_info->chunk_root; 1815 struct btrfs_path *path; 1816 struct extent_buffer *leaf; 1817 struct btrfs_chunk *chunk; 1818 struct btrfs_key key; 1819 struct btrfs_key found_key; 1820 u64 chunk_tree = chunk_root->root_key.objectid; 1821 u64 chunk_type; 1822 bool retried = false; 1823 int failed = 0; 1824 int ret; 1825 1826 path = btrfs_alloc_path(); 1827 if (!path) 1828 return -ENOMEM; 1829 1830again: 1831 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 1832 key.offset = (u64)-1; 1833 key.type = BTRFS_CHUNK_ITEM_KEY; 1834 1835 while (1) { 1836 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 1837 if (ret < 0) 1838 goto error; 1839 BUG_ON(ret == 0); 1840 1841 ret = btrfs_previous_item(chunk_root, path, key.objectid, 1842 key.type); 1843 if (ret < 0) 1844 goto error; 1845 if (ret > 0) 1846 break; 1847 1848 leaf = path->nodes[0]; 1849 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1850 1851 chunk = btrfs_item_ptr(leaf, path->slots[0], 1852 struct btrfs_chunk); 1853 chunk_type = btrfs_chunk_type(leaf, chunk); 1854 btrfs_release_path(chunk_root, path); 1855 1856 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 1857 ret = btrfs_relocate_chunk(chunk_root, chunk_tree, 1858 found_key.objectid, 1859 found_key.offset); 1860 if (ret == -ENOSPC) 1861 failed++; 1862 else if (ret) 1863 BUG(); 1864 } 1865 1866 if (found_key.offset == 0) 1867 break; 1868 key.offset = found_key.offset - 1; 1869 } 1870 ret = 0; 1871 if (failed && !retried) { 1872 failed = 0; 1873 retried = true; 1874 goto again; 1875 } else if (failed && retried) { 1876 WARN_ON(1); 1877 ret = -ENOSPC; 1878 } 1879error: 1880 btrfs_free_path(path); 1881 return ret; 1882} 1883 1884static u64 div_factor(u64 num, int factor) 1885{ 1886 if (factor == 10) 1887 return num; 1888 num *= factor; 1889 do_div(num, 10); 1890 return num; 1891} 1892 1893int btrfs_balance(struct btrfs_root *dev_root) 1894{ 1895 int ret; 1896 struct list_head *devices = &dev_root->fs_info->fs_devices->devices; 1897 struct btrfs_device *device; 1898 u64 old_size; 1899 u64 size_to_free; 1900 struct btrfs_path *path; 1901 struct btrfs_key key; 1902 struct btrfs_chunk *chunk; 1903 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root; 1904 struct btrfs_trans_handle *trans; 1905 struct btrfs_key found_key; 1906 1907 if (dev_root->fs_info->sb->s_flags & MS_RDONLY) 1908 return -EROFS; 1909 1910 mutex_lock(&dev_root->fs_info->volume_mutex); 1911 dev_root = dev_root->fs_info->dev_root; 1912 1913 /* step one make some room on all the devices */ 1914 list_for_each_entry(device, devices, dev_list) { 1915 old_size = device->total_bytes; 1916 size_to_free = div_factor(old_size, 1); 1917 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); 1918 if (!device->writeable || 1919 device->total_bytes - device->bytes_used > size_to_free) 1920 continue; 1921 1922 ret = btrfs_shrink_device(device, old_size - size_to_free); 1923 if (ret == -ENOSPC) 1924 break; 1925 BUG_ON(ret); 1926 1927 trans = btrfs_start_transaction(dev_root, 0); 1928 BUG_ON(!trans); 1929 1930 ret = btrfs_grow_device(trans, device, old_size); 1931 BUG_ON(ret); 1932 1933 btrfs_end_transaction(trans, dev_root); 1934 } 1935 1936 /* step two, relocate all the chunks */ 1937 path = btrfs_alloc_path(); 1938 BUG_ON(!path); 1939 1940 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 1941 key.offset = (u64)-1; 1942 key.type = BTRFS_CHUNK_ITEM_KEY; 1943 1944 while (1) { 1945 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 1946 if (ret < 0) 1947 goto error; 1948 1949 /* 1950 * this shouldn't happen, it means the last relocate 1951 * failed 1952 */ 1953 if (ret == 0) 1954 break; 1955 1956 ret = btrfs_previous_item(chunk_root, path, 0, 1957 BTRFS_CHUNK_ITEM_KEY); 1958 if (ret) 1959 break; 1960 1961 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1962 path->slots[0]); 1963 if (found_key.objectid != key.objectid) 1964 break; 1965 1966 chunk = btrfs_item_ptr(path->nodes[0], 1967 path->slots[0], 1968 struct btrfs_chunk); 1969 /* chunk zero is special */ 1970 if (found_key.offset == 0) 1971 break; 1972 1973 btrfs_release_path(chunk_root, path); 1974 ret = btrfs_relocate_chunk(chunk_root, 1975 chunk_root->root_key.objectid, 1976 found_key.objectid, 1977 found_key.offset); 1978 BUG_ON(ret && ret != -ENOSPC); 1979 key.offset = found_key.offset - 1; 1980 } 1981 ret = 0; 1982error: 1983 btrfs_free_path(path); 1984 mutex_unlock(&dev_root->fs_info->volume_mutex); 1985 return ret; 1986} 1987 1988/* 1989 * shrinking a device means finding all of the device extents past 1990 * the new size, and then following the back refs to the chunks. 1991 * The chunk relocation code actually frees the device extent 1992 */ 1993int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 1994{ 1995 struct btrfs_trans_handle *trans; 1996 struct btrfs_root *root = device->dev_root; 1997 struct btrfs_dev_extent *dev_extent = NULL; 1998 struct btrfs_path *path; 1999 u64 length; 2000 u64 chunk_tree; 2001 u64 chunk_objectid; 2002 u64 chunk_offset; 2003 int ret; 2004 int slot; 2005 int failed = 0; 2006 bool retried = false; 2007 struct extent_buffer *l; 2008 struct btrfs_key key; 2009 struct btrfs_super_block *super_copy = &root->fs_info->super_copy; 2010 u64 old_total = btrfs_super_total_bytes(super_copy); 2011 u64 old_size = device->total_bytes; 2012 u64 diff = device->total_bytes - new_size; 2013 2014 if (new_size >= device->total_bytes) 2015 return -EINVAL; 2016 2017 path = btrfs_alloc_path(); 2018 if (!path) 2019 return -ENOMEM; 2020 2021 path->reada = 2; 2022 2023 lock_chunks(root); 2024 2025 device->total_bytes = new_size; 2026 if (device->writeable) 2027 device->fs_devices->total_rw_bytes -= diff; 2028 unlock_chunks(root); 2029 2030again: 2031 key.objectid = device->devid; 2032 key.offset = (u64)-1; 2033 key.type = BTRFS_DEV_EXTENT_KEY; 2034 2035 while (1) { 2036 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2037 if (ret < 0) 2038 goto done; 2039 2040 ret = btrfs_previous_item(root, path, 0, key.type); 2041 if (ret < 0) 2042 goto done; 2043 if (ret) { 2044 ret = 0; 2045 btrfs_release_path(root, path); 2046 break; 2047 } 2048 2049 l = path->nodes[0]; 2050 slot = path->slots[0]; 2051 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 2052 2053 if (key.objectid != device->devid) { 2054 btrfs_release_path(root, path); 2055 break; 2056 } 2057 2058 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 2059 length = btrfs_dev_extent_length(l, dev_extent); 2060 2061 if (key.offset + length <= new_size) { 2062 btrfs_release_path(root, path); 2063 break; 2064 } 2065 2066 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); 2067 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); 2068 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 2069 btrfs_release_path(root, path); 2070 2071 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, 2072 chunk_offset); 2073 if (ret && ret != -ENOSPC) 2074 goto done; 2075 if (ret == -ENOSPC) 2076 failed++; 2077 key.offset -= 1; 2078 } 2079 2080 if (failed && !retried) { 2081 failed = 0; 2082 retried = true; 2083 goto again; 2084 } else if (failed && retried) { 2085 ret = -ENOSPC; 2086 lock_chunks(root); 2087 2088 device->total_bytes = old_size; 2089 if (device->writeable) 2090 device->fs_devices->total_rw_bytes += diff; 2091 unlock_chunks(root); 2092 goto done; 2093 } 2094 2095 /* Shrinking succeeded, else we would be at "done". */ 2096 trans = btrfs_start_transaction(root, 0); 2097 lock_chunks(root); 2098 2099 device->disk_total_bytes = new_size; 2100 /* Now btrfs_update_device() will change the on-disk size. */ 2101 ret = btrfs_update_device(trans, device); 2102 if (ret) { 2103 unlock_chunks(root); 2104 btrfs_end_transaction(trans, root); 2105 goto done; 2106 } 2107 WARN_ON(diff > old_total); 2108 btrfs_set_super_total_bytes(super_copy, old_total - diff); 2109 unlock_chunks(root); 2110 btrfs_end_transaction(trans, root); 2111done: 2112 btrfs_free_path(path); 2113 return ret; 2114} 2115 2116static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans, 2117 struct btrfs_root *root, 2118 struct btrfs_key *key, 2119 struct btrfs_chunk *chunk, int item_size) 2120{ 2121 struct btrfs_super_block *super_copy = &root->fs_info->super_copy; 2122 struct btrfs_disk_key disk_key; 2123 u32 array_size; 2124 u8 *ptr; 2125 2126 array_size = btrfs_super_sys_array_size(super_copy); 2127 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 2128 return -EFBIG; 2129 2130 ptr = super_copy->sys_chunk_array + array_size; 2131 btrfs_cpu_key_to_disk(&disk_key, key); 2132 memcpy(ptr, &disk_key, sizeof(disk_key)); 2133 ptr += sizeof(disk_key); 2134 memcpy(ptr, chunk, item_size); 2135 item_size += sizeof(disk_key); 2136 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 2137 return 0; 2138} 2139 2140static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size, 2141 int num_stripes, int sub_stripes) 2142{ 2143 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP)) 2144 return calc_size; 2145 else if (type & BTRFS_BLOCK_GROUP_RAID10) 2146 return calc_size * (num_stripes / sub_stripes); 2147 else 2148 return calc_size * num_stripes; 2149} 2150 2151static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, 2152 struct btrfs_root *extent_root, 2153 struct map_lookup **map_ret, 2154 u64 *num_bytes, u64 *stripe_size, 2155 u64 start, u64 type) 2156{ 2157 struct btrfs_fs_info *info = extent_root->fs_info; 2158 struct btrfs_device *device = NULL; 2159 struct btrfs_fs_devices *fs_devices = info->fs_devices; 2160 struct list_head *cur; 2161 struct map_lookup *map = NULL; 2162 struct extent_map_tree *em_tree; 2163 struct extent_map *em; 2164 struct list_head private_devs; 2165 int min_stripe_size = 1 * 1024 * 1024; 2166 u64 calc_size = 1024 * 1024 * 1024; 2167 u64 max_chunk_size = calc_size; 2168 u64 min_free; 2169 u64 avail; 2170 u64 max_avail = 0; 2171 u64 dev_offset; 2172 int num_stripes = 1; 2173 int min_stripes = 1; 2174 int sub_stripes = 0; 2175 int looped = 0; 2176 int ret; 2177 int index; 2178 int stripe_len = 64 * 1024; 2179 2180 if ((type & BTRFS_BLOCK_GROUP_RAID1) && 2181 (type & BTRFS_BLOCK_GROUP_DUP)) { 2182 WARN_ON(1); 2183 type &= ~BTRFS_BLOCK_GROUP_DUP; 2184 } 2185 if (list_empty(&fs_devices->alloc_list)) 2186 return -ENOSPC; 2187 2188 if (type & (BTRFS_BLOCK_GROUP_RAID0)) { 2189 num_stripes = fs_devices->rw_devices; 2190 min_stripes = 2; 2191 } 2192 if (type & (BTRFS_BLOCK_GROUP_DUP)) { 2193 num_stripes = 2; 2194 min_stripes = 2; 2195 } 2196 if (type & (BTRFS_BLOCK_GROUP_RAID1)) { 2197 if (fs_devices->rw_devices < 2) 2198 return -ENOSPC; 2199 num_stripes = 2; 2200 min_stripes = 2; 2201 } 2202 if (type & (BTRFS_BLOCK_GROUP_RAID10)) { 2203 num_stripes = fs_devices->rw_devices; 2204 if (num_stripes < 4) 2205 return -ENOSPC; 2206 num_stripes &= ~(u32)1; 2207 sub_stripes = 2; 2208 min_stripes = 4; 2209 } 2210 2211 if (type & BTRFS_BLOCK_GROUP_DATA) { 2212 max_chunk_size = 10 * calc_size; 2213 min_stripe_size = 64 * 1024 * 1024; 2214 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 2215 max_chunk_size = 256 * 1024 * 1024; 2216 min_stripe_size = 32 * 1024 * 1024; 2217 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 2218 calc_size = 8 * 1024 * 1024; 2219 max_chunk_size = calc_size * 2; 2220 min_stripe_size = 1 * 1024 * 1024; 2221 } 2222 2223 /* we don't want a chunk larger than 10% of writeable space */ 2224 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 2225 max_chunk_size); 2226 2227again: 2228 max_avail = 0; 2229 if (!map || map->num_stripes != num_stripes) { 2230 kfree(map); 2231 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 2232 if (!map) 2233 return -ENOMEM; 2234 map->num_stripes = num_stripes; 2235 } 2236 2237 if (calc_size * num_stripes > max_chunk_size) { 2238 calc_size = max_chunk_size; 2239 do_div(calc_size, num_stripes); 2240 do_div(calc_size, stripe_len); 2241 calc_size *= stripe_len; 2242 } 2243 2244 /* we don't want tiny stripes */ 2245 if (!looped) 2246 calc_size = max_t(u64, min_stripe_size, calc_size); 2247 2248 /* 2249 * we're about to do_div by the stripe_len so lets make sure 2250 * we end up with something bigger than a stripe 2251 */ 2252 calc_size = max_t(u64, calc_size, stripe_len * 4); 2253 2254 do_div(calc_size, stripe_len); 2255 calc_size *= stripe_len; 2256 2257 cur = fs_devices->alloc_list.next; 2258 index = 0; 2259 2260 if (type & BTRFS_BLOCK_GROUP_DUP) 2261 min_free = calc_size * 2; 2262 else 2263 min_free = calc_size; 2264 2265 /* 2266 * we add 1MB because we never use the first 1MB of the device, unless 2267 * we've looped, then we are likely allocating the maximum amount of 2268 * space left already 2269 */ 2270 if (!looped) 2271 min_free += 1024 * 1024; 2272 2273 INIT_LIST_HEAD(&private_devs); 2274 while (index < num_stripes) { 2275 device = list_entry(cur, struct btrfs_device, dev_alloc_list); 2276 BUG_ON(!device->writeable); 2277 if (device->total_bytes > device->bytes_used) 2278 avail = device->total_bytes - device->bytes_used; 2279 else 2280 avail = 0; 2281 cur = cur->next; 2282 2283 if (device->in_fs_metadata && avail >= min_free) { 2284 ret = find_free_dev_extent(trans, device, 2285 min_free, &dev_offset, 2286 &max_avail); 2287 if (ret == 0) { 2288 list_move_tail(&device->dev_alloc_list, 2289 &private_devs); 2290 map->stripes[index].dev = device; 2291 map->stripes[index].physical = dev_offset; 2292 index++; 2293 if (type & BTRFS_BLOCK_GROUP_DUP) { 2294 map->stripes[index].dev = device; 2295 map->stripes[index].physical = 2296 dev_offset + calc_size; 2297 index++; 2298 } 2299 } 2300 } else if (device->in_fs_metadata && avail > max_avail) 2301 max_avail = avail; 2302 if (cur == &fs_devices->alloc_list) 2303 break; 2304 } 2305 list_splice(&private_devs, &fs_devices->alloc_list); 2306 if (index < num_stripes) { 2307 if (index >= min_stripes) { 2308 num_stripes = index; 2309 if (type & (BTRFS_BLOCK_GROUP_RAID10)) { 2310 num_stripes /= sub_stripes; 2311 num_stripes *= sub_stripes; 2312 } 2313 looped = 1; 2314 goto again; 2315 } 2316 if (!looped && max_avail > 0) { 2317 looped = 1; 2318 calc_size = max_avail; 2319 goto again; 2320 } 2321 kfree(map); 2322 return -ENOSPC; 2323 } 2324 map->sector_size = extent_root->sectorsize; 2325 map->stripe_len = stripe_len; 2326 map->io_align = stripe_len; 2327 map->io_width = stripe_len; 2328 map->type = type; 2329 map->num_stripes = num_stripes; 2330 map->sub_stripes = sub_stripes; 2331 2332 *map_ret = map; 2333 *stripe_size = calc_size; 2334 *num_bytes = chunk_bytes_by_type(type, calc_size, 2335 num_stripes, sub_stripes); 2336 2337 em = alloc_extent_map(GFP_NOFS); 2338 if (!em) { 2339 kfree(map); 2340 return -ENOMEM; 2341 } 2342 em->bdev = (struct block_device *)map; 2343 em->start = start; 2344 em->len = *num_bytes; 2345 em->block_start = 0; 2346 em->block_len = em->len; 2347 2348 em_tree = &extent_root->fs_info->mapping_tree.map_tree; 2349 write_lock(&em_tree->lock); 2350 ret = add_extent_mapping(em_tree, em); 2351 write_unlock(&em_tree->lock); 2352 BUG_ON(ret); 2353 free_extent_map(em); 2354 2355 ret = btrfs_make_block_group(trans, extent_root, 0, type, 2356 BTRFS_FIRST_CHUNK_TREE_OBJECTID, 2357 start, *num_bytes); 2358 BUG_ON(ret); 2359 2360 index = 0; 2361 while (index < map->num_stripes) { 2362 device = map->stripes[index].dev; 2363 dev_offset = map->stripes[index].physical; 2364 2365 ret = btrfs_alloc_dev_extent(trans, device, 2366 info->chunk_root->root_key.objectid, 2367 BTRFS_FIRST_CHUNK_TREE_OBJECTID, 2368 start, dev_offset, calc_size); 2369 BUG_ON(ret); 2370 index++; 2371 } 2372 2373 return 0; 2374} 2375 2376static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, 2377 struct btrfs_root *extent_root, 2378 struct map_lookup *map, u64 chunk_offset, 2379 u64 chunk_size, u64 stripe_size) 2380{ 2381 u64 dev_offset; 2382 struct btrfs_key key; 2383 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; 2384 struct btrfs_device *device; 2385 struct btrfs_chunk *chunk; 2386 struct btrfs_stripe *stripe; 2387 size_t item_size = btrfs_chunk_item_size(map->num_stripes); 2388 int index = 0; 2389 int ret; 2390 2391 chunk = kzalloc(item_size, GFP_NOFS); 2392 if (!chunk) 2393 return -ENOMEM; 2394 2395 index = 0; 2396 while (index < map->num_stripes) { 2397 device = map->stripes[index].dev; 2398 device->bytes_used += stripe_size; 2399 ret = btrfs_update_device(trans, device); 2400 BUG_ON(ret); 2401 index++; 2402 } 2403 2404 index = 0; 2405 stripe = &chunk->stripe; 2406 while (index < map->num_stripes) { 2407 device = map->stripes[index].dev; 2408 dev_offset = map->stripes[index].physical; 2409 2410 btrfs_set_stack_stripe_devid(stripe, device->devid); 2411 btrfs_set_stack_stripe_offset(stripe, dev_offset); 2412 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 2413 stripe++; 2414 index++; 2415 } 2416 2417 btrfs_set_stack_chunk_length(chunk, chunk_size); 2418 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); 2419 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 2420 btrfs_set_stack_chunk_type(chunk, map->type); 2421 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 2422 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 2423 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 2424 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); 2425 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 2426 2427 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2428 key.type = BTRFS_CHUNK_ITEM_KEY; 2429 key.offset = chunk_offset; 2430 2431 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 2432 BUG_ON(ret); 2433 2434 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 2435 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk, 2436 item_size); 2437 BUG_ON(ret); 2438 } 2439 kfree(chunk); 2440 return 0; 2441} 2442 2443/* 2444 * Chunk allocation falls into two parts. The first part does works 2445 * that make the new allocated chunk useable, but not do any operation 2446 * that modifies the chunk tree. The second part does the works that 2447 * require modifying the chunk tree. This division is important for the 2448 * bootstrap process of adding storage to a seed btrfs. 2449 */ 2450int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, 2451 struct btrfs_root *extent_root, u64 type) 2452{ 2453 u64 chunk_offset; 2454 u64 chunk_size; 2455 u64 stripe_size; 2456 struct map_lookup *map; 2457 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; 2458 int ret; 2459 2460 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, 2461 &chunk_offset); 2462 if (ret) 2463 return ret; 2464 2465 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, 2466 &stripe_size, chunk_offset, type); 2467 if (ret) 2468 return ret; 2469 2470 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, 2471 chunk_size, stripe_size); 2472 BUG_ON(ret); 2473 return 0; 2474} 2475 2476static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, 2477 struct btrfs_root *root, 2478 struct btrfs_device *device) 2479{ 2480 u64 chunk_offset; 2481 u64 sys_chunk_offset; 2482 u64 chunk_size; 2483 u64 sys_chunk_size; 2484 u64 stripe_size; 2485 u64 sys_stripe_size; 2486 u64 alloc_profile; 2487 struct map_lookup *map; 2488 struct map_lookup *sys_map; 2489 struct btrfs_fs_info *fs_info = root->fs_info; 2490 struct btrfs_root *extent_root = fs_info->extent_root; 2491 int ret; 2492 2493 ret = find_next_chunk(fs_info->chunk_root, 2494 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); 2495 BUG_ON(ret); 2496 2497 alloc_profile = BTRFS_BLOCK_GROUP_METADATA | 2498 (fs_info->metadata_alloc_profile & 2499 fs_info->avail_metadata_alloc_bits); 2500 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); 2501 2502 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, 2503 &stripe_size, chunk_offset, alloc_profile); 2504 BUG_ON(ret); 2505 2506 sys_chunk_offset = chunk_offset + chunk_size; 2507 2508 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | 2509 (fs_info->system_alloc_profile & 2510 fs_info->avail_system_alloc_bits); 2511 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); 2512 2513 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, 2514 &sys_chunk_size, &sys_stripe_size, 2515 sys_chunk_offset, alloc_profile); 2516 BUG_ON(ret); 2517 2518 ret = btrfs_add_device(trans, fs_info->chunk_root, device); 2519 BUG_ON(ret); 2520 2521 /* 2522 * Modifying chunk tree needs allocating new blocks from both 2523 * system block group and metadata block group. So we only can 2524 * do operations require modifying the chunk tree after both 2525 * block groups were created. 2526 */ 2527 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, 2528 chunk_size, stripe_size); 2529 BUG_ON(ret); 2530 2531 ret = __finish_chunk_alloc(trans, extent_root, sys_map, 2532 sys_chunk_offset, sys_chunk_size, 2533 sys_stripe_size); 2534 BUG_ON(ret); 2535 return 0; 2536} 2537 2538int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) 2539{ 2540 struct extent_map *em; 2541 struct map_lookup *map; 2542 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; 2543 int readonly = 0; 2544 int i; 2545 2546 read_lock(&map_tree->map_tree.lock); 2547 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); 2548 read_unlock(&map_tree->map_tree.lock); 2549 if (!em) 2550 return 1; 2551 2552 if (btrfs_test_opt(root, DEGRADED)) { 2553 free_extent_map(em); 2554 return 0; 2555 } 2556 2557 map = (struct map_lookup *)em->bdev; 2558 for (i = 0; i < map->num_stripes; i++) { 2559 if (!map->stripes[i].dev->writeable) { 2560 readonly = 1; 2561 break; 2562 } 2563 } 2564 free_extent_map(em); 2565 return readonly; 2566} 2567 2568void btrfs_mapping_init(struct btrfs_mapping_tree *tree) 2569{ 2570 extent_map_tree_init(&tree->map_tree, GFP_NOFS); 2571} 2572 2573void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) 2574{ 2575 struct extent_map *em; 2576 2577 while (1) { 2578 write_lock(&tree->map_tree.lock); 2579 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); 2580 if (em) 2581 remove_extent_mapping(&tree->map_tree, em); 2582 write_unlock(&tree->map_tree.lock); 2583 if (!em) 2584 break; 2585 kfree(em->bdev); 2586 /* once for us */ 2587 free_extent_map(em); 2588 /* once for the tree */ 2589 free_extent_map(em); 2590 } 2591} 2592 2593int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) 2594{ 2595 struct extent_map *em; 2596 struct map_lookup *map; 2597 struct extent_map_tree *em_tree = &map_tree->map_tree; 2598 int ret; 2599 2600 read_lock(&em_tree->lock); 2601 em = lookup_extent_mapping(em_tree, logical, len); 2602 read_unlock(&em_tree->lock); 2603 BUG_ON(!em); 2604 2605 BUG_ON(em->start > logical || em->start + em->len < logical); 2606 map = (struct map_lookup *)em->bdev; 2607 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) 2608 ret = map->num_stripes; 2609 else if (map->type & BTRFS_BLOCK_GROUP_RAID10) 2610 ret = map->sub_stripes; 2611 else 2612 ret = 1; 2613 free_extent_map(em); 2614 return ret; 2615} 2616 2617static int find_live_mirror(struct map_lookup *map, int first, int num, 2618 int optimal) 2619{ 2620 int i; 2621 if (map->stripes[optimal].dev->bdev) 2622 return optimal; 2623 for (i = first; i < first + num; i++) { 2624 if (map->stripes[i].dev->bdev) 2625 return i; 2626 } 2627 /* we couldn't find one that doesn't fail. Just return something 2628 * and the io error handling code will clean up eventually 2629 */ 2630 return optimal; 2631} 2632 2633static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, 2634 u64 logical, u64 *length, 2635 struct btrfs_multi_bio **multi_ret, 2636 int mirror_num, struct page *unplug_page) 2637{ 2638 struct extent_map *em; 2639 struct map_lookup *map; 2640 struct extent_map_tree *em_tree = &map_tree->map_tree; 2641 u64 offset; 2642 u64 stripe_offset; 2643 u64 stripe_nr; 2644 int stripes_allocated = 8; 2645 int stripes_required = 1; 2646 int stripe_index; 2647 int i; 2648 int num_stripes; 2649 int max_errors = 0; 2650 struct btrfs_multi_bio *multi = NULL; 2651 2652 if (multi_ret && !(rw & REQ_WRITE)) 2653 stripes_allocated = 1; 2654again: 2655 if (multi_ret) { 2656 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated), 2657 GFP_NOFS); 2658 if (!multi) 2659 return -ENOMEM; 2660 2661 atomic_set(&multi->error, 0); 2662 } 2663 2664 read_lock(&em_tree->lock); 2665 em = lookup_extent_mapping(em_tree, logical, *length); 2666 read_unlock(&em_tree->lock); 2667 2668 if (!em && unplug_page) { 2669 kfree(multi); 2670 return 0; 2671 } 2672 2673 if (!em) { 2674 printk(KERN_CRIT "unable to find logical %llu len %llu\n", 2675 (unsigned long long)logical, 2676 (unsigned long long)*length); 2677 BUG(); 2678 } 2679 2680 BUG_ON(em->start > logical || em->start + em->len < logical); 2681 map = (struct map_lookup *)em->bdev; 2682 offset = logical - em->start; 2683 2684 if (mirror_num > map->num_stripes) 2685 mirror_num = 0; 2686 2687 /* if our multi bio struct is too small, back off and try again */ 2688 if (rw & REQ_WRITE) { 2689 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | 2690 BTRFS_BLOCK_GROUP_DUP)) { 2691 stripes_required = map->num_stripes; 2692 max_errors = 1; 2693 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 2694 stripes_required = map->sub_stripes; 2695 max_errors = 1; 2696 } 2697 } 2698 if (multi_ret && (rw & REQ_WRITE) && 2699 stripes_allocated < stripes_required) { 2700 stripes_allocated = map->num_stripes; 2701 free_extent_map(em); 2702 kfree(multi); 2703 goto again; 2704 } 2705 stripe_nr = offset; 2706 /* 2707 * stripe_nr counts the total number of stripes we have to stride 2708 * to get to this block 2709 */ 2710 do_div(stripe_nr, map->stripe_len); 2711 2712 stripe_offset = stripe_nr * map->stripe_len; 2713 BUG_ON(offset < stripe_offset); 2714 2715 /* stripe_offset is the offset of this block in its stripe*/ 2716 stripe_offset = offset - stripe_offset; 2717 2718 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | 2719 BTRFS_BLOCK_GROUP_RAID10 | 2720 BTRFS_BLOCK_GROUP_DUP)) { 2721 /* we limit the length of each bio to what fits in a stripe */ 2722 *length = min_t(u64, em->len - offset, 2723 map->stripe_len - stripe_offset); 2724 } else { 2725 *length = em->len - offset; 2726 } 2727 2728 if (!multi_ret && !unplug_page) 2729 goto out; 2730 2731 num_stripes = 1; 2732 stripe_index = 0; 2733 if (map->type & BTRFS_BLOCK_GROUP_RAID1) { 2734 if (unplug_page || (rw & REQ_WRITE)) 2735 num_stripes = map->num_stripes; 2736 else if (mirror_num) 2737 stripe_index = mirror_num - 1; 2738 else { 2739 stripe_index = find_live_mirror(map, 0, 2740 map->num_stripes, 2741 current->pid % map->num_stripes); 2742 } 2743 2744 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 2745 if (rw & REQ_WRITE) 2746 num_stripes = map->num_stripes; 2747 else if (mirror_num) 2748 stripe_index = mirror_num - 1; 2749 2750 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 2751 int factor = map->num_stripes / map->sub_stripes; 2752 2753 stripe_index = do_div(stripe_nr, factor); 2754 stripe_index *= map->sub_stripes; 2755 2756 if (unplug_page || (rw & REQ_WRITE)) 2757 num_stripes = map->sub_stripes; 2758 else if (mirror_num) 2759 stripe_index += mirror_num - 1; 2760 else { 2761 stripe_index = find_live_mirror(map, stripe_index, 2762 map->sub_stripes, stripe_index + 2763 current->pid % map->sub_stripes); 2764 } 2765 } else { 2766 /* 2767 * after this do_div call, stripe_nr is the number of stripes 2768 * on this device we have to walk to find the data, and 2769 * stripe_index is the number of our device in the stripe array 2770 */ 2771 stripe_index = do_div(stripe_nr, map->num_stripes); 2772 } 2773 BUG_ON(stripe_index >= map->num_stripes); 2774 2775 for (i = 0; i < num_stripes; i++) { 2776 if (unplug_page) { 2777 struct btrfs_device *device; 2778 struct backing_dev_info *bdi; 2779 2780 device = map->stripes[stripe_index].dev; 2781 if (device->bdev) { 2782 bdi = blk_get_backing_dev_info(device->bdev); 2783 if (bdi->unplug_io_fn) 2784 bdi->unplug_io_fn(bdi, unplug_page); 2785 } 2786 } else { 2787 multi->stripes[i].physical = 2788 map->stripes[stripe_index].physical + 2789 stripe_offset + stripe_nr * map->stripe_len; 2790 multi->stripes[i].dev = map->stripes[stripe_index].dev; 2791 } 2792 stripe_index++; 2793 } 2794 if (multi_ret) { 2795 *multi_ret = multi; 2796 multi->num_stripes = num_stripes; 2797 multi->max_errors = max_errors; 2798 } 2799out: 2800 free_extent_map(em); 2801 return 0; 2802} 2803 2804int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, 2805 u64 logical, u64 *length, 2806 struct btrfs_multi_bio **multi_ret, int mirror_num) 2807{ 2808 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret, 2809 mirror_num, NULL); 2810} 2811 2812int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, 2813 u64 chunk_start, u64 physical, u64 devid, 2814 u64 **logical, int *naddrs, int *stripe_len) 2815{ 2816 struct extent_map_tree *em_tree = &map_tree->map_tree; 2817 struct extent_map *em; 2818 struct map_lookup *map; 2819 u64 *buf; 2820 u64 bytenr; 2821 u64 length; 2822 u64 stripe_nr; 2823 int i, j, nr = 0; 2824 2825 read_lock(&em_tree->lock); 2826 em = lookup_extent_mapping(em_tree, chunk_start, 1); 2827 read_unlock(&em_tree->lock); 2828 2829 BUG_ON(!em || em->start != chunk_start); 2830 map = (struct map_lookup *)em->bdev; 2831 2832 length = em->len; 2833 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 2834 do_div(length, map->num_stripes / map->sub_stripes); 2835 else if (map->type & BTRFS_BLOCK_GROUP_RAID0) 2836 do_div(length, map->num_stripes); 2837 2838 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); 2839 BUG_ON(!buf); 2840 2841 for (i = 0; i < map->num_stripes; i++) { 2842 if (devid && map->stripes[i].dev->devid != devid) 2843 continue; 2844 if (map->stripes[i].physical > physical || 2845 map->stripes[i].physical + length <= physical) 2846 continue; 2847 2848 stripe_nr = physical - map->stripes[i].physical; 2849 do_div(stripe_nr, map->stripe_len); 2850 2851 if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 2852 stripe_nr = stripe_nr * map->num_stripes + i; 2853 do_div(stripe_nr, map->sub_stripes); 2854 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 2855 stripe_nr = stripe_nr * map->num_stripes + i; 2856 } 2857 bytenr = chunk_start + stripe_nr * map->stripe_len; 2858 WARN_ON(nr >= map->num_stripes); 2859 for (j = 0; j < nr; j++) { 2860 if (buf[j] == bytenr) 2861 break; 2862 } 2863 if (j == nr) { 2864 WARN_ON(nr >= map->num_stripes); 2865 buf[nr++] = bytenr; 2866 } 2867 } 2868 2869 *logical = buf; 2870 *naddrs = nr; 2871 *stripe_len = map->stripe_len; 2872 2873 free_extent_map(em); 2874 return 0; 2875} 2876 2877int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree, 2878 u64 logical, struct page *page) 2879{ 2880 u64 length = PAGE_CACHE_SIZE; 2881 return __btrfs_map_block(map_tree, READ, logical, &length, 2882 NULL, 0, page); 2883} 2884 2885static void end_bio_multi_stripe(struct bio *bio, int err) 2886{ 2887 struct btrfs_multi_bio *multi = bio->bi_private; 2888 int is_orig_bio = 0; 2889 2890 if (err) 2891 atomic_inc(&multi->error); 2892 2893 if (bio == multi->orig_bio) 2894 is_orig_bio = 1; 2895 2896 if (atomic_dec_and_test(&multi->stripes_pending)) { 2897 if (!is_orig_bio) { 2898 bio_put(bio); 2899 bio = multi->orig_bio; 2900 } 2901 bio->bi_private = multi->private; 2902 bio->bi_end_io = multi->end_io; 2903 /* only send an error to the higher layers if it is 2904 * beyond the tolerance of the multi-bio 2905 */ 2906 if (atomic_read(&multi->error) > multi->max_errors) { 2907 err = -EIO; 2908 } else if (err) { 2909 /* 2910 * this bio is actually up to date, we didn't 2911 * go over the max number of errors 2912 */ 2913 set_bit(BIO_UPTODATE, &bio->bi_flags); 2914 err = 0; 2915 } 2916 kfree(multi); 2917 2918 bio_endio(bio, err); 2919 } else if (!is_orig_bio) { 2920 bio_put(bio); 2921 } 2922} 2923 2924struct async_sched { 2925 struct bio *bio; 2926 int rw; 2927 struct btrfs_fs_info *info; 2928 struct btrfs_work work; 2929}; 2930 2931/* 2932 * see run_scheduled_bios for a description of why bios are collected for 2933 * async submit. 2934 * 2935 * This will add one bio to the pending list for a device and make sure 2936 * the work struct is scheduled. 2937 */ 2938static noinline int schedule_bio(struct btrfs_root *root, 2939 struct btrfs_device *device, 2940 int rw, struct bio *bio) 2941{ 2942 int should_queue = 1; 2943 struct btrfs_pending_bios *pending_bios; 2944 2945 /* don't bother with additional async steps for reads, right now */ 2946 if (!(rw & REQ_WRITE)) { 2947 bio_get(bio); 2948 submit_bio(rw, bio); 2949 bio_put(bio); 2950 return 0; 2951 } 2952 2953 /* 2954 * nr_async_bios allows us to reliably return congestion to the 2955 * higher layers. Otherwise, the async bio makes it appear we have 2956 * made progress against dirty pages when we've really just put it 2957 * on a queue for later 2958 */ 2959 atomic_inc(&root->fs_info->nr_async_bios); 2960 WARN_ON(bio->bi_next); 2961 bio->bi_next = NULL; 2962 bio->bi_rw |= rw; 2963 2964 spin_lock(&device->io_lock); 2965 if (bio->bi_rw & REQ_SYNC) 2966 pending_bios = &device->pending_sync_bios; 2967 else 2968 pending_bios = &device->pending_bios; 2969 2970 if (pending_bios->tail) 2971 pending_bios->tail->bi_next = bio; 2972 2973 pending_bios->tail = bio; 2974 if (!pending_bios->head) 2975 pending_bios->head = bio; 2976 if (device->running_pending) 2977 should_queue = 0; 2978 2979 spin_unlock(&device->io_lock); 2980 2981 if (should_queue) 2982 btrfs_queue_worker(&root->fs_info->submit_workers, 2983 &device->work); 2984 return 0; 2985} 2986 2987int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, 2988 int mirror_num, int async_submit) 2989{ 2990 struct btrfs_mapping_tree *map_tree; 2991 struct btrfs_device *dev; 2992 struct bio *first_bio = bio; 2993 u64 logical = (u64)bio->bi_sector << 9; 2994 u64 length = 0; 2995 u64 map_length; 2996 struct btrfs_multi_bio *multi = NULL; 2997 int ret; 2998 int dev_nr = 0; 2999 int total_devs = 1; 3000 3001 length = bio->bi_size; 3002 map_tree = &root->fs_info->mapping_tree; 3003 map_length = length; 3004 3005 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi, 3006 mirror_num); 3007 BUG_ON(ret); 3008 3009 total_devs = multi->num_stripes; 3010 if (map_length < length) { 3011 printk(KERN_CRIT "mapping failed logical %llu bio len %llu " 3012 "len %llu\n", (unsigned long long)logical, 3013 (unsigned long long)length, 3014 (unsigned long long)map_length); 3015 BUG(); 3016 } 3017 multi->end_io = first_bio->bi_end_io; 3018 multi->private = first_bio->bi_private; 3019 multi->orig_bio = first_bio; 3020 atomic_set(&multi->stripes_pending, multi->num_stripes); 3021 3022 while (dev_nr < total_devs) { 3023 if (total_devs > 1) { 3024 if (dev_nr < total_devs - 1) { 3025 bio = bio_clone(first_bio, GFP_NOFS); 3026 BUG_ON(!bio); 3027 } else { 3028 bio = first_bio; 3029 } 3030 bio->bi_private = multi; 3031 bio->bi_end_io = end_bio_multi_stripe; 3032 } 3033 bio->bi_sector = multi->stripes[dev_nr].physical >> 9; 3034 dev = multi->stripes[dev_nr].dev; 3035 BUG_ON(rw == WRITE && !dev->writeable); 3036 if (dev && dev->bdev) { 3037 bio->bi_bdev = dev->bdev; 3038 if (async_submit) 3039 schedule_bio(root, dev, rw, bio); 3040 else 3041 submit_bio(rw, bio); 3042 } else { 3043 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; 3044 bio->bi_sector = logical >> 9; 3045 bio_endio(bio, -EIO); 3046 } 3047 dev_nr++; 3048 } 3049 if (total_devs == 1) 3050 kfree(multi); 3051 return 0; 3052} 3053 3054struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, 3055 u8 *uuid, u8 *fsid) 3056{ 3057 struct btrfs_device *device; 3058 struct btrfs_fs_devices *cur_devices; 3059 3060 cur_devices = root->fs_info->fs_devices; 3061 while (cur_devices) { 3062 if (!fsid || 3063 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { 3064 device = __find_device(&cur_devices->devices, 3065 devid, uuid); 3066 if (device) 3067 return device; 3068 } 3069 cur_devices = cur_devices->seed; 3070 } 3071 return NULL; 3072} 3073 3074static struct btrfs_device *add_missing_dev(struct btrfs_root *root, 3075 u64 devid, u8 *dev_uuid) 3076{ 3077 struct btrfs_device *device; 3078 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 3079 3080 device = kzalloc(sizeof(*device), GFP_NOFS); 3081 if (!device) 3082 return NULL; 3083 list_add(&device->dev_list, 3084 &fs_devices->devices); 3085 device->barriers = 1; 3086 device->dev_root = root->fs_info->dev_root; 3087 device->devid = devid; 3088 device->work.func = pending_bios_fn; 3089 device->fs_devices = fs_devices; 3090 fs_devices->num_devices++; 3091 spin_lock_init(&device->io_lock); 3092 INIT_LIST_HEAD(&device->dev_alloc_list); 3093 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); 3094 return device; 3095} 3096 3097static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, 3098 struct extent_buffer *leaf, 3099 struct btrfs_chunk *chunk) 3100{ 3101 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; 3102 struct map_lookup *map; 3103 struct extent_map *em; 3104 u64 logical; 3105 u64 length; 3106 u64 devid; 3107 u8 uuid[BTRFS_UUID_SIZE]; 3108 int num_stripes; 3109 int ret; 3110 int i; 3111 3112 logical = key->offset; 3113 length = btrfs_chunk_length(leaf, chunk); 3114 3115 read_lock(&map_tree->map_tree.lock); 3116 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); 3117 read_unlock(&map_tree->map_tree.lock); 3118 3119 /* already mapped? */ 3120 if (em && em->start <= logical && em->start + em->len > logical) { 3121 free_extent_map(em); 3122 return 0; 3123 } else if (em) { 3124 free_extent_map(em); 3125 } 3126 3127 em = alloc_extent_map(GFP_NOFS); 3128 if (!em) 3129 return -ENOMEM; 3130 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3131 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 3132 if (!map) { 3133 free_extent_map(em); 3134 return -ENOMEM; 3135 } 3136 3137 em->bdev = (struct block_device *)map; 3138 em->start = logical; 3139 em->len = length; 3140 em->block_start = 0; 3141 em->block_len = em->len; 3142 3143 map->num_stripes = num_stripes; 3144 map->io_width = btrfs_chunk_io_width(leaf, chunk); 3145 map->io_align = btrfs_chunk_io_align(leaf, chunk); 3146 map->sector_size = btrfs_chunk_sector_size(leaf, chunk); 3147 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 3148 map->type = btrfs_chunk_type(leaf, chunk); 3149 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 3150 for (i = 0; i < num_stripes; i++) { 3151 map->stripes[i].physical = 3152 btrfs_stripe_offset_nr(leaf, chunk, i); 3153 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 3154 read_extent_buffer(leaf, uuid, (unsigned long) 3155 btrfs_stripe_dev_uuid_nr(chunk, i), 3156 BTRFS_UUID_SIZE); 3157 map->stripes[i].dev = btrfs_find_device(root, devid, uuid, 3158 NULL); 3159 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { 3160 kfree(map); 3161 free_extent_map(em); 3162 return -EIO; 3163 } 3164 if (!map->stripes[i].dev) { 3165 map->stripes[i].dev = 3166 add_missing_dev(root, devid, uuid); 3167 if (!map->stripes[i].dev) { 3168 kfree(map); 3169 free_extent_map(em); 3170 return -EIO; 3171 } 3172 } 3173 map->stripes[i].dev->in_fs_metadata = 1; 3174 } 3175 3176 write_lock(&map_tree->map_tree.lock); 3177 ret = add_extent_mapping(&map_tree->map_tree, em); 3178 write_unlock(&map_tree->map_tree.lock); 3179 BUG_ON(ret); 3180 free_extent_map(em); 3181 3182 return 0; 3183} 3184 3185static int fill_device_from_item(struct extent_buffer *leaf, 3186 struct btrfs_dev_item *dev_item, 3187 struct btrfs_device *device) 3188{ 3189 unsigned long ptr; 3190 3191 device->devid = btrfs_device_id(leaf, dev_item); 3192 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 3193 device->total_bytes = device->disk_total_bytes; 3194 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 3195 device->type = btrfs_device_type(leaf, dev_item); 3196 device->io_align = btrfs_device_io_align(leaf, dev_item); 3197 device->io_width = btrfs_device_io_width(leaf, dev_item); 3198 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 3199 3200 ptr = (unsigned long)btrfs_device_uuid(dev_item); 3201 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 3202 3203 return 0; 3204} 3205 3206static int open_seed_devices(struct btrfs_root *root, u8 *fsid) 3207{ 3208 struct btrfs_fs_devices *fs_devices; 3209 int ret; 3210 3211 mutex_lock(&uuid_mutex); 3212 3213 fs_devices = root->fs_info->fs_devices->seed; 3214 while (fs_devices) { 3215 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { 3216 ret = 0; 3217 goto out; 3218 } 3219 fs_devices = fs_devices->seed; 3220 } 3221 3222 fs_devices = find_fsid(fsid); 3223 if (!fs_devices) { 3224 ret = -ENOENT; 3225 goto out; 3226 } 3227 3228 fs_devices = clone_fs_devices(fs_devices); 3229 if (IS_ERR(fs_devices)) { 3230 ret = PTR_ERR(fs_devices); 3231 goto out; 3232 } 3233 3234 ret = __btrfs_open_devices(fs_devices, FMODE_READ, 3235 root->fs_info->bdev_holder); 3236 if (ret) 3237 goto out; 3238 3239 if (!fs_devices->seeding) { 3240 __btrfs_close_devices(fs_devices); 3241 free_fs_devices(fs_devices); 3242 ret = -EINVAL; 3243 goto out; 3244 } 3245 3246 fs_devices->seed = root->fs_info->fs_devices->seed; 3247 root->fs_info->fs_devices->seed = fs_devices; 3248out: 3249 mutex_unlock(&uuid_mutex); 3250 return ret; 3251} 3252 3253static int read_one_dev(struct btrfs_root *root, 3254 struct extent_buffer *leaf, 3255 struct btrfs_dev_item *dev_item) 3256{ 3257 struct btrfs_device *device; 3258 u64 devid; 3259 int ret; 3260 u8 fs_uuid[BTRFS_UUID_SIZE]; 3261 u8 dev_uuid[BTRFS_UUID_SIZE]; 3262 3263 devid = btrfs_device_id(leaf, dev_item); 3264 read_extent_buffer(leaf, dev_uuid, 3265 (unsigned long)btrfs_device_uuid(dev_item), 3266 BTRFS_UUID_SIZE); 3267 read_extent_buffer(leaf, fs_uuid, 3268 (unsigned long)btrfs_device_fsid(dev_item), 3269 BTRFS_UUID_SIZE); 3270 3271 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { 3272 ret = open_seed_devices(root, fs_uuid); 3273 if (ret && !btrfs_test_opt(root, DEGRADED)) 3274 return ret; 3275 } 3276 3277 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); 3278 if (!device || !device->bdev) { 3279 if (!btrfs_test_opt(root, DEGRADED)) 3280 return -EIO; 3281 3282 if (!device) { 3283 printk(KERN_WARNING "warning devid %llu missing\n", 3284 (unsigned long long)devid); 3285 device = add_missing_dev(root, devid, dev_uuid); 3286 if (!device) 3287 return -ENOMEM; 3288 } 3289 } 3290 3291 if (device->fs_devices != root->fs_info->fs_devices) { 3292 BUG_ON(device->writeable); 3293 if (device->generation != 3294 btrfs_device_generation(leaf, dev_item)) 3295 return -EINVAL; 3296 } 3297 3298 fill_device_from_item(leaf, dev_item, device); 3299 device->dev_root = root->fs_info->dev_root; 3300 device->in_fs_metadata = 1; 3301 if (device->writeable) 3302 device->fs_devices->total_rw_bytes += device->total_bytes; 3303 ret = 0; 3304 return ret; 3305} 3306 3307int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf) 3308{ 3309 struct btrfs_dev_item *dev_item; 3310 3311 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block, 3312 dev_item); 3313 return read_one_dev(root, buf, dev_item); 3314} 3315 3316int btrfs_read_sys_array(struct btrfs_root *root) 3317{ 3318 struct btrfs_super_block *super_copy = &root->fs_info->super_copy; 3319 struct extent_buffer *sb; 3320 struct btrfs_disk_key *disk_key; 3321 struct btrfs_chunk *chunk; 3322 u8 *ptr; 3323 unsigned long sb_ptr; 3324 int ret = 0; 3325 u32 num_stripes; 3326 u32 array_size; 3327 u32 len = 0; 3328 u32 cur; 3329 struct btrfs_key key; 3330 3331 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, 3332 BTRFS_SUPER_INFO_SIZE); 3333 if (!sb) 3334 return -ENOMEM; 3335 btrfs_set_buffer_uptodate(sb); 3336 btrfs_set_buffer_lockdep_class(sb, 0); 3337 3338 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 3339 array_size = btrfs_super_sys_array_size(super_copy); 3340 3341 ptr = super_copy->sys_chunk_array; 3342 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); 3343 cur = 0; 3344 3345 while (cur < array_size) { 3346 disk_key = (struct btrfs_disk_key *)ptr; 3347 btrfs_disk_key_to_cpu(&key, disk_key); 3348 3349 len = sizeof(*disk_key); ptr += len; 3350 sb_ptr += len; 3351 cur += len; 3352 3353 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 3354 chunk = (struct btrfs_chunk *)sb_ptr; 3355 ret = read_one_chunk(root, &key, sb, chunk); 3356 if (ret) 3357 break; 3358 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 3359 len = btrfs_chunk_item_size(num_stripes); 3360 } else { 3361 ret = -EIO; 3362 break; 3363 } 3364 ptr += len; 3365 sb_ptr += len; 3366 cur += len; 3367 } 3368 free_extent_buffer(sb); 3369 return ret; 3370} 3371 3372int btrfs_read_chunk_tree(struct btrfs_root *root) 3373{ 3374 struct btrfs_path *path; 3375 struct extent_buffer *leaf; 3376 struct btrfs_key key; 3377 struct btrfs_key found_key; 3378 int ret; 3379 int slot; 3380 3381 root = root->fs_info->chunk_root; 3382 3383 path = btrfs_alloc_path(); 3384 if (!path) 3385 return -ENOMEM; 3386 3387 /* first we search for all of the device items, and then we 3388 * read in all of the chunk items. This way we can create chunk 3389 * mappings that reference all of the devices that are afound 3390 */ 3391 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 3392 key.offset = 0; 3393 key.type = 0; 3394again: 3395 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 3396 if (ret < 0) 3397 goto error; 3398 while (1) { 3399 leaf = path->nodes[0]; 3400 slot = path->slots[0]; 3401 if (slot >= btrfs_header_nritems(leaf)) { 3402 ret = btrfs_next_leaf(root, path); 3403 if (ret == 0) 3404 continue; 3405 if (ret < 0) 3406 goto error; 3407 break; 3408 } 3409 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3410 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { 3411 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) 3412 break; 3413 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 3414 struct btrfs_dev_item *dev_item; 3415 dev_item = btrfs_item_ptr(leaf, slot, 3416 struct btrfs_dev_item); 3417 ret = read_one_dev(root, leaf, dev_item); 3418 if (ret) 3419 goto error; 3420 } 3421 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 3422 struct btrfs_chunk *chunk; 3423 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3424 ret = read_one_chunk(root, &found_key, leaf, chunk); 3425 if (ret) 3426 goto error; 3427 } 3428 path->slots[0]++; 3429 } 3430 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { 3431 key.objectid = 0; 3432 btrfs_release_path(root, path); 3433 goto again; 3434 } 3435 ret = 0; 3436error: 3437 btrfs_free_path(path); 3438 return ret; 3439} 3440